1 @c Copyright (C) 1988, 1989, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
2 @c 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010, 2011, 2012
3 @c Free Software Foundation, Inc.
4 @c This is part of the GCC manual.
5 @c For copying conditions, see the file gcc.texi.
12 @c man begin COPYRIGHT
13 Copyright @copyright{} 1988, 1989, 1992, 1993, 1994, 1995, 1996, 1997, 1998,
14 1999, 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010, 2011,
16 Free Software Foundation, Inc.
18 Permission is granted to copy, distribute and/or modify this document
19 under the terms of the GNU Free Documentation License, Version 1.3 or
20 any later version published by the Free Software Foundation; with the
21 Invariant Sections being ``GNU General Public License'' and ``Funding
22 Free Software'', the Front-Cover texts being (a) (see below), and with
23 the Back-Cover Texts being (b) (see below). A copy of the license is
24 included in the gfdl(7) man page.
26 (a) The FSF's Front-Cover Text is:
30 (b) The FSF's Back-Cover Text is:
32 You have freedom to copy and modify this GNU Manual, like GNU
33 software. Copies published by the Free Software Foundation raise
34 funds for GNU development.
36 @c Set file name and title for the man page.
38 @settitle GNU project C and C++ compiler
40 gcc [@option{-c}|@option{-S}|@option{-E}] [@option{-std=}@var{standard}]
41 [@option{-g}] [@option{-pg}] [@option{-O}@var{level}]
42 [@option{-W}@var{warn}@dots{}] [@option{-pedantic}]
43 [@option{-I}@var{dir}@dots{}] [@option{-L}@var{dir}@dots{}]
44 [@option{-D}@var{macro}[=@var{defn}]@dots{}] [@option{-U}@var{macro}]
45 [@option{-f}@var{option}@dots{}] [@option{-m}@var{machine-option}@dots{}]
46 [@option{-o} @var{outfile}] [@@@var{file}] @var{infile}@dots{}
48 Only the most useful options are listed here; see below for the
49 remainder. @samp{g++} accepts mostly the same options as @samp{gcc}.
52 gpl(7), gfdl(7), fsf-funding(7),
53 cpp(1), gcov(1), as(1), ld(1), gdb(1), adb(1), dbx(1), sdb(1)
54 and the Info entries for @file{gcc}, @file{cpp}, @file{as},
55 @file{ld}, @file{binutils} and @file{gdb}.
58 For instructions on reporting bugs, see
62 See the Info entry for @command{gcc}, or
63 @w{@uref{http://gcc.gnu.org/onlinedocs/gcc/Contributors.html}},
64 for contributors to GCC@.
69 @chapter GCC Command Options
70 @cindex GCC command options
71 @cindex command options
72 @cindex options, GCC command
74 @c man begin DESCRIPTION
75 When you invoke GCC, it normally does preprocessing, compilation,
76 assembly and linking. The ``overall options'' allow you to stop this
77 process at an intermediate stage. For example, the @option{-c} option
78 says not to run the linker. Then the output consists of object files
79 output by the assembler.
81 Other options are passed on to one stage of processing. Some options
82 control the preprocessor and others the compiler itself. Yet other
83 options control the assembler and linker; most of these are not
84 documented here, since you rarely need to use any of them.
86 @cindex C compilation options
87 Most of the command-line options that you can use with GCC are useful
88 for C programs; when an option is only useful with another language
89 (usually C++), the explanation says so explicitly. If the description
90 for a particular option does not mention a source language, you can use
91 that option with all supported languages.
93 @cindex C++ compilation options
94 @xref{Invoking G++,,Compiling C++ Programs}, for a summary of special
95 options for compiling C++ programs.
97 @cindex grouping options
98 @cindex options, grouping
99 The @command{gcc} program accepts options and file names as operands. Many
100 options have multi-letter names; therefore multiple single-letter options
101 may @emph{not} be grouped: @option{-dv} is very different from @w{@samp{-d
104 @cindex order of options
105 @cindex options, order
106 You can mix options and other arguments. For the most part, the order
107 you use doesn't matter. Order does matter when you use several
108 options of the same kind; for example, if you specify @option{-L} more
109 than once, the directories are searched in the order specified. Also,
110 the placement of the @option{-l} option is significant.
112 Many options have long names starting with @samp{-f} or with
113 @samp{-W}---for example,
114 @option{-fmove-loop-invariants}, @option{-Wformat} and so on. Most of
115 these have both positive and negative forms; the negative form of
116 @option{-ffoo} would be @option{-fno-foo}. This manual documents
117 only one of these two forms, whichever one is not the default.
121 @xref{Option Index}, for an index to GCC's options.
124 * Option Summary:: Brief list of all options, without explanations.
125 * Overall Options:: Controlling the kind of output:
126 an executable, object files, assembler files,
127 or preprocessed source.
128 * Invoking G++:: Compiling C++ programs.
129 * C Dialect Options:: Controlling the variant of C language compiled.
130 * C++ Dialect Options:: Variations on C++.
131 * Objective-C and Objective-C++ Dialect Options:: Variations on Objective-C
133 * Language Independent Options:: Controlling how diagnostics should be
135 * Warning Options:: How picky should the compiler be?
136 * Debugging Options:: Symbol tables, measurements, and debugging dumps.
137 * Optimize Options:: How much optimization?
138 * Preprocessor Options:: Controlling header files and macro definitions.
139 Also, getting dependency information for Make.
140 * Assembler Options:: Passing options to the assembler.
141 * Link Options:: Specifying libraries and so on.
142 * Directory Options:: Where to find header files and libraries.
143 Where to find the compiler executable files.
144 * Spec Files:: How to pass switches to sub-processes.
145 * Target Options:: Running a cross-compiler, or an old version of GCC.
146 * Submodel Options:: Specifying minor hardware or convention variations,
147 such as 68010 vs 68020.
148 * Code Gen Options:: Specifying conventions for function calls, data layout
150 * Environment Variables:: Env vars that affect GCC.
151 * Precompiled Headers:: Compiling a header once, and using it many times.
157 @section Option Summary
159 Here is a summary of all the options, grouped by type. Explanations are
160 in the following sections.
163 @item Overall Options
164 @xref{Overall Options,,Options Controlling the Kind of Output}.
165 @gccoptlist{-c -S -E -o @var{file} -no-canonical-prefixes @gol
166 -pipe -pass-exit-codes @gol
167 -x @var{language} -v -### --help@r{[}=@var{class}@r{[},@dots{}@r{]]} --target-help @gol
168 --version -wrapper @@@var{file} -fplugin=@var{file} -fplugin-arg-@var{name}=@var{arg} @gol
169 -fdump-ada-spec@r{[}-slim@r{]} -fdump-go-spec=@var{file}}
171 @item C Language Options
172 @xref{C Dialect Options,,Options Controlling C Dialect}.
173 @gccoptlist{-ansi -std=@var{standard} -fgnu89-inline @gol
174 -aux-info @var{filename} -fallow-parameterless-variadic-functions @gol
175 -fno-asm -fno-builtin -fno-builtin-@var{function} @gol
176 -fhosted -ffreestanding -fopenmp -fms-extensions -fplan9-extensions @gol
177 -trigraphs -no-integrated-cpp -traditional -traditional-cpp @gol
178 -fallow-single-precision -fcond-mismatch -flax-vector-conversions @gol
179 -fsigned-bitfields -fsigned-char @gol
180 -funsigned-bitfields -funsigned-char}
182 @item C++ Language Options
183 @xref{C++ Dialect Options,,Options Controlling C++ Dialect}.
184 @gccoptlist{-fabi-version=@var{n} -fno-access-control -fcheck-new @gol
185 -fconserve-space -fconstexpr-depth=@var{n} -ffriend-injection @gol
186 -fno-elide-constructors @gol
187 -fno-enforce-eh-specs @gol
188 -ffor-scope -fno-for-scope -fno-gnu-keywords @gol
189 -fno-implicit-templates @gol
190 -fno-implicit-inline-templates @gol
191 -fno-implement-inlines -fms-extensions @gol
192 -fno-nonansi-builtins -fnothrow-opt -fno-operator-names @gol
193 -fno-optional-diags -fpermissive @gol
194 -fno-pretty-templates @gol
195 -frepo -fno-rtti -fstats -ftemplate-depth=@var{n} @gol
196 -fno-threadsafe-statics -fuse-cxa-atexit -fno-weak -nostdinc++ @gol
197 -fno-default-inline -fvisibility-inlines-hidden @gol
198 -fvisibility-ms-compat @gol
199 -Wabi -Wconversion-null -Wctor-dtor-privacy @gol
200 -Wdelete-non-virtual-dtor -Wnarrowing -Wnoexcept @gol
201 -Wnon-virtual-dtor -Wreorder @gol
202 -Weffc++ -Wstrict-null-sentinel @gol
203 -Wno-non-template-friend -Wold-style-cast @gol
204 -Woverloaded-virtual -Wno-pmf-conversions @gol
207 @item Objective-C and Objective-C++ Language Options
208 @xref{Objective-C and Objective-C++ Dialect Options,,Options Controlling
209 Objective-C and Objective-C++ Dialects}.
210 @gccoptlist{-fconstant-string-class=@var{class-name} @gol
211 -fgnu-runtime -fnext-runtime @gol
212 -fno-nil-receivers @gol
213 -fobjc-abi-version=@var{n} @gol
214 -fobjc-call-cxx-cdtors @gol
215 -fobjc-direct-dispatch @gol
216 -fobjc-exceptions @gol
219 -fobjc-std=objc1 @gol
220 -freplace-objc-classes @gol
223 -Wassign-intercept @gol
224 -Wno-protocol -Wselector @gol
225 -Wstrict-selector-match @gol
226 -Wundeclared-selector}
228 @item Language Independent Options
229 @xref{Language Independent Options,,Options to Control Diagnostic Messages Formatting}.
230 @gccoptlist{-fmessage-length=@var{n} @gol
231 -fdiagnostics-show-location=@r{[}once@r{|}every-line@r{]} @gol
232 -fno-diagnostics-show-option}
234 @item Warning Options
235 @xref{Warning Options,,Options to Request or Suppress Warnings}.
236 @gccoptlist{-fsyntax-only -fmax-errors=@var{n} -pedantic @gol
237 -pedantic-errors @gol
238 -w -Wextra -Wall -Waddress -Waggregate-return -Warray-bounds @gol
239 -Wno-attributes -Wno-builtin-macro-redefined @gol
240 -Wc++-compat -Wc++11-compat -Wcast-align -Wcast-qual @gol
241 -Wchar-subscripts -Wclobbered -Wcomment @gol
242 -Wconversion -Wcoverage-mismatch -Wno-cpp -Wno-deprecated @gol
243 -Wno-deprecated-declarations -Wdisabled-optimization @gol
244 -Wno-div-by-zero -Wdouble-promotion -Wempty-body -Wenum-compare @gol
245 -Wno-endif-labels -Werror -Werror=* @gol
246 -Wfatal-errors -Wfloat-equal -Wformat -Wformat=2 @gol
247 -Wno-format-contains-nul -Wno-format-extra-args -Wformat-nonliteral @gol
248 -Wformat-security -Wformat-y2k @gol
249 -Wframe-larger-than=@var{len} -Wno-free-nonheap-object -Wjump-misses-init @gol
250 -Wignored-qualifiers @gol
251 -Wimplicit -Wimplicit-function-declaration -Wimplicit-int @gol
252 -Winit-self -Winline -Wmaybe-uninitialized @gol
253 -Wno-int-to-pointer-cast -Wno-invalid-offsetof @gol
254 -Winvalid-pch -Wlarger-than=@var{len} -Wunsafe-loop-optimizations @gol
255 -Wlogical-op -Wlong-long @gol
256 -Wmain -Wmaybe-uninitialized -Wmissing-braces -Wmissing-field-initializers @gol
257 -Wmissing-format-attribute -Wmissing-include-dirs @gol
259 -Wno-multichar -Wnonnull -Wno-overflow @gol
260 -Woverlength-strings -Wpacked -Wpacked-bitfield-compat -Wpadded @gol
261 -Wparentheses -Wpedantic-ms-format -Wno-pedantic-ms-format @gol
262 -Wpointer-arith -Wno-pointer-to-int-cast @gol
263 -Wredundant-decls @gol
264 -Wreturn-type -Wsequence-point -Wshadow @gol
265 -Wsign-compare -Wsign-conversion -Wstack-protector @gol
266 -Wstack-usage=@var{len} -Wstrict-aliasing -Wstrict-aliasing=n @gol
267 -Wstrict-overflow -Wstrict-overflow=@var{n} @gol
268 -Wsuggest-attribute=@r{[}pure@r{|}const@r{|}noreturn@r{]} @gol
269 -Wswitch -Wswitch-default -Wswitch-enum -Wsync-nand @gol
270 -Wsystem-headers -Wtrampolines -Wtrigraphs -Wtype-limits -Wundef @gol
271 -Wuninitialized -Wunknown-pragmas -Wno-pragmas @gol
272 -Wunsuffixed-float-constants -Wunused -Wunused-function @gol
273 -Wunused-label -Wunused-local-typedefs -Wunused-parameter @gol
274 -Wno-unused-result -Wunused-value @gol -Wunused-variable @gol
275 -Wunused-but-set-parameter -Wunused-but-set-variable @gol
276 -Wvariadic-macros -Wvector-operation-performance -Wvla
277 -Wvolatile-register-var -Wwrite-strings -Wzero-as-null-pointer-constant}
279 @item C and Objective-C-only Warning Options
280 @gccoptlist{-Wbad-function-cast -Wmissing-declarations @gol
281 -Wmissing-parameter-type -Wmissing-prototypes -Wnested-externs @gol
282 -Wold-style-declaration -Wold-style-definition @gol
283 -Wstrict-prototypes -Wtraditional -Wtraditional-conversion @gol
284 -Wdeclaration-after-statement -Wpointer-sign}
286 @item Debugging Options
287 @xref{Debugging Options,,Options for Debugging Your Program or GCC}.
288 @gccoptlist{-d@var{letters} -dumpspecs -dumpmachine -dumpversion @gol
289 -fdbg-cnt-list -fdbg-cnt=@var{counter-value-list} @gol
290 -fdisable-ipa-@var{pass_name} @gol
291 -fdisable-rtl-@var{pass_name} @gol
292 -fdisable-rtl-@var{pass-name}=@var{range-list} @gol
293 -fdisable-tree-@var{pass_name} @gol
294 -fdisable-tree-@var{pass-name}=@var{range-list} @gol
295 -fdump-noaddr -fdump-unnumbered -fdump-unnumbered-links @gol
296 -fdump-translation-unit@r{[}-@var{n}@r{]} @gol
297 -fdump-class-hierarchy@r{[}-@var{n}@r{]} @gol
298 -fdump-ipa-all -fdump-ipa-cgraph -fdump-ipa-inline @gol
300 -fdump-statistics @gol
302 -fdump-tree-original@r{[}-@var{n}@r{]} @gol
303 -fdump-tree-optimized@r{[}-@var{n}@r{]} @gol
304 -fdump-tree-cfg -fdump-tree-vcg -fdump-tree-alias @gol
306 -fdump-tree-ssa@r{[}-@var{n}@r{]} -fdump-tree-pre@r{[}-@var{n}@r{]} @gol
307 -fdump-tree-ccp@r{[}-@var{n}@r{]} -fdump-tree-dce@r{[}-@var{n}@r{]} @gol
308 -fdump-tree-gimple@r{[}-raw@r{]} -fdump-tree-mudflap@r{[}-@var{n}@r{]} @gol
309 -fdump-tree-dom@r{[}-@var{n}@r{]} @gol
310 -fdump-tree-dse@r{[}-@var{n}@r{]} @gol
311 -fdump-tree-phiprop@r{[}-@var{n}@r{]} @gol
312 -fdump-tree-phiopt@r{[}-@var{n}@r{]} @gol
313 -fdump-tree-forwprop@r{[}-@var{n}@r{]} @gol
314 -fdump-tree-copyrename@r{[}-@var{n}@r{]} @gol
315 -fdump-tree-nrv -fdump-tree-vect @gol
316 -fdump-tree-sink @gol
317 -fdump-tree-sra@r{[}-@var{n}@r{]} @gol
318 -fdump-tree-forwprop@r{[}-@var{n}@r{]} @gol
319 -fdump-tree-fre@r{[}-@var{n}@r{]} @gol
320 -fdump-tree-vrp@r{[}-@var{n}@r{]} @gol
321 -ftree-vectorizer-verbose=@var{n} @gol
322 -fdump-tree-storeccp@r{[}-@var{n}@r{]} @gol
323 -fdump-final-insns=@var{file} @gol
324 -fcompare-debug@r{[}=@var{opts}@r{]} -fcompare-debug-second @gol
325 -feliminate-dwarf2-dups -feliminate-unused-debug-types @gol
326 -feliminate-unused-debug-symbols -femit-class-debug-always @gol
327 -fenable-@var{kind}-@var{pass} @gol
328 -fenable-@var{kind}-@var{pass}=@var{range-list} @gol
329 -fdebug-types-section @gol
330 -fmem-report -fpre-ipa-mem-report -fpost-ipa-mem-report -fprofile-arcs @gol
331 -frandom-seed=@var{string} -fsched-verbose=@var{n} @gol
332 -fsel-sched-verbose -fsel-sched-dump-cfg -fsel-sched-pipelining-verbose @gol
333 -fstack-usage -ftest-coverage -ftime-report -fvar-tracking @gol
334 -fvar-tracking-assignments -fvar-tracking-assignments-toggle @gol
335 -g -g@var{level} -gtoggle -gcoff -gdwarf-@var{version} @gol
336 -ggdb -grecord-gcc-switches -gno-record-gcc-switches @gol
337 -gstabs -gstabs+ -gstrict-dwarf -gno-strict-dwarf @gol
338 -gvms -gxcoff -gxcoff+ @gol
339 -fno-merge-debug-strings -fno-dwarf2-cfi-asm @gol
340 -fdebug-prefix-map=@var{old}=@var{new} @gol
341 -femit-struct-debug-baseonly -femit-struct-debug-reduced @gol
342 -femit-struct-debug-detailed@r{[}=@var{spec-list}@r{]} @gol
343 -p -pg -print-file-name=@var{library} -print-libgcc-file-name @gol
344 -print-multi-directory -print-multi-lib -print-multi-os-directory @gol
345 -print-prog-name=@var{program} -print-search-dirs -Q @gol
346 -print-sysroot -print-sysroot-headers-suffix @gol
347 -save-temps -save-temps=cwd -save-temps=obj -time@r{[}=@var{file}@r{]}}
349 @item Optimization Options
350 @xref{Optimize Options,,Options that Control Optimization}.
351 @gccoptlist{-falign-functions[=@var{n}] -falign-jumps[=@var{n}] @gol
352 -falign-labels[=@var{n}] -falign-loops[=@var{n}] -fassociative-math @gol
353 -fauto-inc-dec -fbranch-probabilities -fbranch-target-load-optimize @gol
354 -fbranch-target-load-optimize2 -fbtr-bb-exclusive -fcaller-saves @gol
355 -fcheck-data-deps -fcombine-stack-adjustments -fconserve-stack @gol
356 -fcompare-elim -fcprop-registers -fcrossjumping @gol
357 -fcse-follow-jumps -fcse-skip-blocks -fcx-fortran-rules @gol
358 -fcx-limited-range @gol
359 -fdata-sections -fdce -fdelayed-branch @gol
360 -fdelete-null-pointer-checks -fdevirtualize -fdse @gol
361 -fearly-inlining -fipa-sra -fexpensive-optimizations -ffat-lto-objects @gol
362 -ffast-math -ffinite-math-only -ffloat-store -fexcess-precision=@var{style} @gol
363 -fforward-propagate -ffp-contract=@var{style} -ffunction-sections @gol
364 -fgcse -fgcse-after-reload -fgcse-las -fgcse-lm -fgraphite-identity @gol
365 -fgcse-sm -fif-conversion -fif-conversion2 -findirect-inlining @gol
366 -finline-functions -finline-functions-called-once -finline-limit=@var{n} @gol
367 -finline-small-functions -fipa-cp -fipa-cp-clone -fipa-matrix-reorg @gol
368 -fipa-pta -fipa-profile -fipa-pure-const -fipa-reference @gol
369 -fira-algorithm=@var{algorithm} @gol
370 -fira-region=@var{region} @gol
371 -fira-loop-pressure -fno-ira-share-save-slots @gol
372 -fno-ira-share-spill-slots -fira-verbose=@var{n} @gol
373 -fivopts -fkeep-inline-functions -fkeep-static-consts @gol
374 -floop-block -floop-flatten -floop-interchange -floop-strip-mine @gol
375 -floop-parallelize-all -flto -flto-compression-level @gol
376 -flto-partition=@var{alg} -flto-report -fmerge-all-constants @gol
377 -fmerge-constants -fmodulo-sched -fmodulo-sched-allow-regmoves @gol
378 -fmove-loop-invariants fmudflap -fmudflapir -fmudflapth -fno-branch-count-reg @gol
379 -fno-default-inline @gol
380 -fno-defer-pop -fno-function-cse -fno-guess-branch-probability @gol
381 -fno-inline -fno-math-errno -fno-peephole -fno-peephole2 @gol
382 -fno-sched-interblock -fno-sched-spec -fno-signed-zeros @gol
383 -fno-toplevel-reorder -fno-trapping-math -fno-zero-initialized-in-bss @gol
384 -fomit-frame-pointer -foptimize-register-move -foptimize-sibling-calls @gol
385 -fpartial-inlining -fpeel-loops -fpredictive-commoning @gol
386 -fprefetch-loop-arrays @gol
387 -fprofile-correction -fprofile-dir=@var{path} -fprofile-generate @gol
388 -fprofile-generate=@var{path} @gol
389 -fprofile-use -fprofile-use=@var{path} -fprofile-values @gol
390 -freciprocal-math -free -fregmove -frename-registers -freorder-blocks @gol
391 -freorder-blocks-and-partition -freorder-functions @gol
392 -frerun-cse-after-loop -freschedule-modulo-scheduled-loops @gol
393 -frounding-math -fsched2-use-superblocks -fsched-pressure @gol
394 -fsched-spec-load -fsched-spec-load-dangerous @gol
395 -fsched-stalled-insns-dep[=@var{n}] -fsched-stalled-insns[=@var{n}] @gol
396 -fsched-group-heuristic -fsched-critical-path-heuristic @gol
397 -fsched-spec-insn-heuristic -fsched-rank-heuristic @gol
398 -fsched-last-insn-heuristic -fsched-dep-count-heuristic @gol
399 -fschedule-insns -fschedule-insns2 -fsection-anchors @gol
400 -fselective-scheduling -fselective-scheduling2 @gol
401 -fsel-sched-pipelining -fsel-sched-pipelining-outer-loops @gol
402 -fshrink-wrap -fsignaling-nans -fsingle-precision-constant @gol
403 -fsplit-ivs-in-unroller -fsplit-wide-types -fstack-protector @gol
404 -fstack-protector-all -fstrict-aliasing -fstrict-overflow @gol
405 -fthread-jumps -ftracer -ftree-bit-ccp @gol
406 -ftree-builtin-call-dce -ftree-ccp -ftree-ch -ftree-copy-prop @gol
407 -ftree-copyrename -ftree-dce -ftree-dominator-opts -ftree-dse @gol
408 -ftree-forwprop -ftree-fre -ftree-loop-if-convert @gol
409 -ftree-loop-if-convert-stores -ftree-loop-im @gol
410 -ftree-phiprop -ftree-loop-distribution -ftree-loop-distribute-patterns @gol
411 -ftree-loop-ivcanon -ftree-loop-linear -ftree-loop-optimize @gol
412 -ftree-parallelize-loops=@var{n} -ftree-pre -ftree-pta -ftree-reassoc @gol
413 -ftree-sink -ftree-sra -ftree-switch-conversion -ftree-tail-merge @gol
414 -ftree-ter -ftree-vect-loop-version -ftree-vectorize -ftree-vrp @gol
415 -funit-at-a-time -funroll-all-loops -funroll-loops @gol
416 -funsafe-loop-optimizations -funsafe-math-optimizations -funswitch-loops @gol
417 -fvariable-expansion-in-unroller -fvect-cost-model -fvpt -fweb @gol
418 -fwhole-program -fwpa -fuse-linker-plugin @gol
419 --param @var{name}=@var{value}
420 -O -O0 -O1 -O2 -O3 -Os -Ofast}
422 @item Preprocessor Options
423 @xref{Preprocessor Options,,Options Controlling the Preprocessor}.
424 @gccoptlist{-A@var{question}=@var{answer} @gol
425 -A-@var{question}@r{[}=@var{answer}@r{]} @gol
426 -C -dD -dI -dM -dN @gol
427 -D@var{macro}@r{[}=@var{defn}@r{]} -E -H @gol
428 -idirafter @var{dir} @gol
429 -include @var{file} -imacros @var{file} @gol
430 -iprefix @var{file} -iwithprefix @var{dir} @gol
431 -iwithprefixbefore @var{dir} -isystem @var{dir} @gol
432 -imultilib @var{dir} -isysroot @var{dir} @gol
433 -M -MM -MF -MG -MP -MQ -MT -nostdinc @gol
434 -P -fdebug-cpp -ftrack-macro-expansion -fworking-directory @gol
435 -remap -trigraphs -undef -U@var{macro} @gol
436 -Wp,@var{option} -Xpreprocessor @var{option}}
438 @item Assembler Option
439 @xref{Assembler Options,,Passing Options to the Assembler}.
440 @gccoptlist{-Wa,@var{option} -Xassembler @var{option}}
443 @xref{Link Options,,Options for Linking}.
444 @gccoptlist{@var{object-file-name} -l@var{library} @gol
445 -nostartfiles -nodefaultlibs -nostdlib -pie -rdynamic @gol
446 -s -static -static-libgcc -static-libstdc++ -shared @gol
447 -shared-libgcc -symbolic @gol
448 -T @var{script} -Wl,@var{option} -Xlinker @var{option} @gol
451 @item Directory Options
452 @xref{Directory Options,,Options for Directory Search}.
453 @gccoptlist{-B@var{prefix} -I@var{dir} -iplugindir=@var{dir} @gol
454 -iquote@var{dir} -L@var{dir} -specs=@var{file} -I- @gol
457 @item Machine Dependent Options
458 @xref{Submodel Options,,Hardware Models and Configurations}.
459 @c This list is ordered alphanumerically by subsection name.
460 @c Try and put the significant identifier (CPU or system) first,
461 @c so users have a clue at guessing where the ones they want will be.
463 @emph{Adapteva Epiphany Options}
464 @gccoptlist{-mhalf-reg-file -mprefer-short-insn-regs @gol
465 -mbranch-cost=@var{num} -mcmove -mnops=@var{num} -msoft-cmpsf @gol
466 -msplit-lohi -mpost-inc -mpost-modify -mstack-offset=@var{num} @gol
467 -mround-nearest -mlong-calls -mshort-calls -msmall16 @gol
468 -mfp-mode=@var{mode} -mvect-double -max-vect-align=@var{num} @gol
469 -msplit-vecmove-early -m1reg-@var{reg}}
472 @gccoptlist{-mapcs-frame -mno-apcs-frame @gol
473 -mabi=@var{name} @gol
474 -mapcs-stack-check -mno-apcs-stack-check @gol
475 -mapcs-float -mno-apcs-float @gol
476 -mapcs-reentrant -mno-apcs-reentrant @gol
477 -msched-prolog -mno-sched-prolog @gol
478 -mlittle-endian -mbig-endian -mwords-little-endian @gol
479 -mfloat-abi=@var{name} -mfpe @gol
480 -mfp16-format=@var{name}
481 -mthumb-interwork -mno-thumb-interwork @gol
482 -mcpu=@var{name} -march=@var{name} -mfpu=@var{name} @gol
483 -mstructure-size-boundary=@var{n} @gol
484 -mabort-on-noreturn @gol
485 -mlong-calls -mno-long-calls @gol
486 -msingle-pic-base -mno-single-pic-base @gol
487 -mpic-register=@var{reg} @gol
488 -mnop-fun-dllimport @gol
489 -mcirrus-fix-invalid-insns -mno-cirrus-fix-invalid-insns @gol
490 -mpoke-function-name @gol
492 -mtpcs-frame -mtpcs-leaf-frame @gol
493 -mcaller-super-interworking -mcallee-super-interworking @gol
494 -mtp=@var{name} -mtls-dialect=@var{dialect} @gol
495 -mword-relocations @gol
496 -mfix-cortex-m3-ldrd @gol
500 @gccoptlist{-mmcu=@var{mcu} -maccumulate-args -mbranch-cost=@var{cost} @gol
501 -mcall-prologues -mint8 -mno-interrupts -mrelax -mshort-calls @gol
502 -mstrict-X -mtiny-stack}
504 @emph{Blackfin Options}
505 @gccoptlist{-mcpu=@var{cpu}@r{[}-@var{sirevision}@r{]} @gol
506 -msim -momit-leaf-frame-pointer -mno-omit-leaf-frame-pointer @gol
507 -mspecld-anomaly -mno-specld-anomaly -mcsync-anomaly -mno-csync-anomaly @gol
508 -mlow-64k -mno-low64k -mstack-check-l1 -mid-shared-library @gol
509 -mno-id-shared-library -mshared-library-id=@var{n} @gol
510 -mleaf-id-shared-library -mno-leaf-id-shared-library @gol
511 -msep-data -mno-sep-data -mlong-calls -mno-long-calls @gol
512 -mfast-fp -minline-plt -mmulticore -mcorea -mcoreb -msdram @gol
516 @gccoptlist{-mbig-endian -mlittle-endian -march=@var{cpu} @gol
517 -msim -msdata=@var{sdata-type}}
520 @gccoptlist{-mcpu=@var{cpu} -march=@var{cpu} -mtune=@var{cpu} @gol
521 -mmax-stack-frame=@var{n} -melinux-stacksize=@var{n} @gol
522 -metrax4 -metrax100 -mpdebug -mcc-init -mno-side-effects @gol
523 -mstack-align -mdata-align -mconst-align @gol
524 -m32-bit -m16-bit -m8-bit -mno-prologue-epilogue -mno-gotplt @gol
525 -melf -maout -melinux -mlinux -sim -sim2 @gol
526 -mmul-bug-workaround -mno-mul-bug-workaround}
529 @gccoptlist{-mmac @gol
530 -mcr16cplus -mcr16c @gol
531 -msim -mint32 -mbit-ops
532 -mdata-model=@var{model}}
534 @emph{Darwin Options}
535 @gccoptlist{-all_load -allowable_client -arch -arch_errors_fatal @gol
536 -arch_only -bind_at_load -bundle -bundle_loader @gol
537 -client_name -compatibility_version -current_version @gol
539 -dependency-file -dylib_file -dylinker_install_name @gol
540 -dynamic -dynamiclib -exported_symbols_list @gol
541 -filelist -flat_namespace -force_cpusubtype_ALL @gol
542 -force_flat_namespace -headerpad_max_install_names @gol
544 -image_base -init -install_name -keep_private_externs @gol
545 -multi_module -multiply_defined -multiply_defined_unused @gol
546 -noall_load -no_dead_strip_inits_and_terms @gol
547 -nofixprebinding -nomultidefs -noprebind -noseglinkedit @gol
548 -pagezero_size -prebind -prebind_all_twolevel_modules @gol
549 -private_bundle -read_only_relocs -sectalign @gol
550 -sectobjectsymbols -whyload -seg1addr @gol
551 -sectcreate -sectobjectsymbols -sectorder @gol
552 -segaddr -segs_read_only_addr -segs_read_write_addr @gol
553 -seg_addr_table -seg_addr_table_filename -seglinkedit @gol
554 -segprot -segs_read_only_addr -segs_read_write_addr @gol
555 -single_module -static -sub_library -sub_umbrella @gol
556 -twolevel_namespace -umbrella -undefined @gol
557 -unexported_symbols_list -weak_reference_mismatches @gol
558 -whatsloaded -F -gused -gfull -mmacosx-version-min=@var{version} @gol
559 -mkernel -mone-byte-bool}
561 @emph{DEC Alpha Options}
562 @gccoptlist{-mno-fp-regs -msoft-float -malpha-as -mgas @gol
563 -mieee -mieee-with-inexact -mieee-conformant @gol
564 -mfp-trap-mode=@var{mode} -mfp-rounding-mode=@var{mode} @gol
565 -mtrap-precision=@var{mode} -mbuild-constants @gol
566 -mcpu=@var{cpu-type} -mtune=@var{cpu-type} @gol
567 -mbwx -mmax -mfix -mcix @gol
568 -mfloat-vax -mfloat-ieee @gol
569 -mexplicit-relocs -msmall-data -mlarge-data @gol
570 -msmall-text -mlarge-text @gol
571 -mmemory-latency=@var{time}}
573 @emph{DEC Alpha/VMS Options}
574 @gccoptlist{-mvms-return-codes -mdebug-main=@var{prefix} -mmalloc64}
577 @gccoptlist{-msmall-model -mno-lsim}
580 @gccoptlist{-mgpr-32 -mgpr-64 -mfpr-32 -mfpr-64 @gol
581 -mhard-float -msoft-float @gol
582 -malloc-cc -mfixed-cc -mdword -mno-dword @gol
583 -mdouble -mno-double @gol
584 -mmedia -mno-media -mmuladd -mno-muladd @gol
585 -mfdpic -minline-plt -mgprel-ro -multilib-library-pic @gol
586 -mlinked-fp -mlong-calls -malign-labels @gol
587 -mlibrary-pic -macc-4 -macc-8 @gol
588 -mpack -mno-pack -mno-eflags -mcond-move -mno-cond-move @gol
589 -moptimize-membar -mno-optimize-membar @gol
590 -mscc -mno-scc -mcond-exec -mno-cond-exec @gol
591 -mvliw-branch -mno-vliw-branch @gol
592 -mmulti-cond-exec -mno-multi-cond-exec -mnested-cond-exec @gol
593 -mno-nested-cond-exec -mtomcat-stats @gol
597 @emph{GNU/Linux Options}
598 @gccoptlist{-mglibc -muclibc -mbionic -mandroid @gol
599 -tno-android-cc -tno-android-ld}
601 @emph{H8/300 Options}
602 @gccoptlist{-mrelax -mh -ms -mn -mint32 -malign-300}
605 @gccoptlist{-march=@var{architecture-type} @gol
606 -mbig-switch -mdisable-fpregs -mdisable-indexing @gol
607 -mfast-indirect-calls -mgas -mgnu-ld -mhp-ld @gol
608 -mfixed-range=@var{register-range} @gol
609 -mjump-in-delay -mlinker-opt -mlong-calls @gol
610 -mlong-load-store -mno-big-switch -mno-disable-fpregs @gol
611 -mno-disable-indexing -mno-fast-indirect-calls -mno-gas @gol
612 -mno-jump-in-delay -mno-long-load-store @gol
613 -mno-portable-runtime -mno-soft-float @gol
614 -mno-space-regs -msoft-float -mpa-risc-1-0 @gol
615 -mpa-risc-1-1 -mpa-risc-2-0 -mportable-runtime @gol
616 -mschedule=@var{cpu-type} -mspace-regs -msio -mwsio @gol
617 -munix=@var{unix-std} -nolibdld -static -threads}
619 @emph{i386 and x86-64 Options}
620 @gccoptlist{-mtune=@var{cpu-type} -march=@var{cpu-type} @gol
621 -mfpmath=@var{unit} @gol
622 -masm=@var{dialect} -mno-fancy-math-387 @gol
623 -mno-fp-ret-in-387 -msoft-float @gol
624 -mno-wide-multiply -mrtd -malign-double @gol
625 -mpreferred-stack-boundary=@var{num} @gol
626 -mincoming-stack-boundary=@var{num} @gol
627 -mcld -mcx16 -msahf -mmovbe -mcrc32 @gol
628 -mrecip -mrecip=@var{opt} @gol
629 -mvzeroupper -mprefer-avx128 @gol
630 -mmmx -msse -msse2 -msse3 -mssse3 -msse4.1 -msse4.2 -msse4 -mavx @gol
631 -mavx2 -maes -mpclmul -mfsgsbase -mrdrnd -mf16c -mfma @gol
632 -msse4a -m3dnow -mpopcnt -mabm -mbmi -mtbm -mfma4 -mxop -mlzcnt @gol
633 -mbmi2 -mlwp -mthreads -mno-align-stringops -minline-all-stringops @gol
634 -minline-stringops-dynamically -mstringop-strategy=@var{alg} @gol
635 -mpush-args -maccumulate-outgoing-args -m128bit-long-double @gol
636 -m96bit-long-double -mregparm=@var{num} -msseregparm @gol
637 -mveclibabi=@var{type} -mvect8-ret-in-mem @gol
638 -mpc32 -mpc64 -mpc80 -mstackrealign @gol
639 -momit-leaf-frame-pointer -mno-red-zone -mno-tls-direct-seg-refs @gol
640 -mcmodel=@var{code-model} -mabi=@var{name} @gol
641 -m32 -m64 -mx32 -mlarge-data-threshold=@var{num} @gol
642 -msse2avx -mfentry -m8bit-idiv @gol
643 -mavx256-split-unaligned-load -mavx256-split-unaligned-store}
645 @emph{i386 and x86-64 Windows Options}
646 @gccoptlist{-mconsole -mcygwin -mno-cygwin -mdll @gol
647 -mnop-fun-dllimport -mthread @gol
648 -municode -mwin32 -mwindows -fno-set-stack-executable}
651 @gccoptlist{-mbig-endian -mlittle-endian -mgnu-as -mgnu-ld -mno-pic @gol
652 -mvolatile-asm-stop -mregister-names -msdata -mno-sdata @gol
653 -mconstant-gp -mauto-pic -mfused-madd @gol
654 -minline-float-divide-min-latency @gol
655 -minline-float-divide-max-throughput @gol
656 -mno-inline-float-divide @gol
657 -minline-int-divide-min-latency @gol
658 -minline-int-divide-max-throughput @gol
659 -mno-inline-int-divide @gol
660 -minline-sqrt-min-latency -minline-sqrt-max-throughput @gol
661 -mno-inline-sqrt @gol
662 -mdwarf2-asm -mearly-stop-bits @gol
663 -mfixed-range=@var{register-range} -mtls-size=@var{tls-size} @gol
664 -mtune=@var{cpu-type} -milp32 -mlp64 @gol
665 -msched-br-data-spec -msched-ar-data-spec -msched-control-spec @gol
666 -msched-br-in-data-spec -msched-ar-in-data-spec -msched-in-control-spec @gol
667 -msched-spec-ldc -msched-spec-control-ldc @gol
668 -msched-prefer-non-data-spec-insns -msched-prefer-non-control-spec-insns @gol
669 -msched-stop-bits-after-every-cycle -msched-count-spec-in-critical-path @gol
670 -msel-sched-dont-check-control-spec -msched-fp-mem-deps-zero-cost @gol
671 -msched-max-memory-insns-hard-limit -msched-max-memory-insns=@var{max-insns}}
673 @emph{IA-64/VMS Options}
674 @gccoptlist{-mvms-return-codes -mdebug-main=@var{prefix} -mmalloc64}
677 @gccoptlist{-mbarrel-shift-enabled -mdivide-enabled -mmultiply-enabled @gol
678 -msign-extend-enabled -muser-enabled}
680 @emph{M32R/D Options}
681 @gccoptlist{-m32r2 -m32rx -m32r @gol
683 -malign-loops -mno-align-loops @gol
684 -missue-rate=@var{number} @gol
685 -mbranch-cost=@var{number} @gol
686 -mmodel=@var{code-size-model-type} @gol
687 -msdata=@var{sdata-type} @gol
688 -mno-flush-func -mflush-func=@var{name} @gol
689 -mno-flush-trap -mflush-trap=@var{number} @gol
693 @gccoptlist{-mcpu=@var{cpu} -msim -memregs=@var{number}}
695 @emph{M680x0 Options}
696 @gccoptlist{-march=@var{arch} -mcpu=@var{cpu} -mtune=@var{tune}
697 -m68000 -m68020 -m68020-40 -m68020-60 -m68030 -m68040 @gol
698 -m68060 -mcpu32 -m5200 -m5206e -m528x -m5307 -m5407 @gol
699 -mcfv4e -mbitfield -mno-bitfield -mc68000 -mc68020 @gol
700 -mnobitfield -mrtd -mno-rtd -mdiv -mno-div -mshort @gol
701 -mno-short -mhard-float -m68881 -msoft-float -mpcrel @gol
702 -malign-int -mstrict-align -msep-data -mno-sep-data @gol
703 -mshared-library-id=n -mid-shared-library -mno-id-shared-library @gol
707 @gccoptlist{-mhardlit -mno-hardlit -mdiv -mno-div -mrelax-immediates @gol
708 -mno-relax-immediates -mwide-bitfields -mno-wide-bitfields @gol
709 -m4byte-functions -mno-4byte-functions -mcallgraph-data @gol
710 -mno-callgraph-data -mslow-bytes -mno-slow-bytes -mno-lsim @gol
711 -mlittle-endian -mbig-endian -m210 -m340 -mstack-increment}
714 @gccoptlist{-mabsdiff -mall-opts -maverage -mbased=@var{n} -mbitops @gol
715 -mc=@var{n} -mclip -mconfig=@var{name} -mcop -mcop32 -mcop64 -mivc2 @gol
716 -mdc -mdiv -meb -mel -mio-volatile -ml -mleadz -mm -mminmax @gol
717 -mmult -mno-opts -mrepeat -ms -msatur -msdram -msim -msimnovec -mtf @gol
720 @emph{MicroBlaze Options}
721 @gccoptlist{-msoft-float -mhard-float -msmall-divides -mcpu=@var{cpu} @gol
722 -mmemcpy -mxl-soft-mul -mxl-soft-div -mxl-barrel-shift @gol
723 -mxl-pattern-compare -mxl-stack-check -mxl-gp-opt -mno-clearbss @gol
724 -mxl-multiply-high -mxl-float-convert -mxl-float-sqrt @gol
725 -mxl-mode-@var{app-model}}
728 @gccoptlist{-EL -EB -march=@var{arch} -mtune=@var{arch} @gol
729 -mips1 -mips2 -mips3 -mips4 -mips32 -mips32r2 @gol
730 -mips64 -mips64r2 @gol
731 -mips16 -mno-mips16 -mflip-mips16 @gol
732 -minterlink-mips16 -mno-interlink-mips16 @gol
733 -mabi=@var{abi} -mabicalls -mno-abicalls @gol
734 -mshared -mno-shared -mplt -mno-plt -mxgot -mno-xgot @gol
735 -mgp32 -mgp64 -mfp32 -mfp64 -mhard-float -msoft-float @gol
736 -msingle-float -mdouble-float -mdsp -mno-dsp -mdspr2 -mno-dspr2 @gol
737 -mfpu=@var{fpu-type} @gol
738 -msmartmips -mno-smartmips @gol
739 -mpaired-single -mno-paired-single -mdmx -mno-mdmx @gol
740 -mips3d -mno-mips3d -mmt -mno-mt -mllsc -mno-llsc @gol
741 -mlong64 -mlong32 -msym32 -mno-sym32 @gol
742 -G@var{num} -mlocal-sdata -mno-local-sdata @gol
743 -mextern-sdata -mno-extern-sdata -mgpopt -mno-gopt @gol
744 -membedded-data -mno-embedded-data @gol
745 -muninit-const-in-rodata -mno-uninit-const-in-rodata @gol
746 -mcode-readable=@var{setting} @gol
747 -msplit-addresses -mno-split-addresses @gol
748 -mexplicit-relocs -mno-explicit-relocs @gol
749 -mcheck-zero-division -mno-check-zero-division @gol
750 -mdivide-traps -mdivide-breaks @gol
751 -mmemcpy -mno-memcpy -mlong-calls -mno-long-calls @gol
752 -mmad -mno-mad -mfused-madd -mno-fused-madd -nocpp @gol
753 -mfix-24k -mno-fix-24k @gol
754 -mfix-r4000 -mno-fix-r4000 -mfix-r4400 -mno-fix-r4400 @gol
755 -mfix-r10000 -mno-fix-r10000 -mfix-vr4120 -mno-fix-vr4120 @gol
756 -mfix-vr4130 -mno-fix-vr4130 -mfix-sb1 -mno-fix-sb1 @gol
757 -mflush-func=@var{func} -mno-flush-func @gol
758 -mbranch-cost=@var{num} -mbranch-likely -mno-branch-likely @gol
759 -mfp-exceptions -mno-fp-exceptions @gol
760 -mvr4130-align -mno-vr4130-align -msynci -mno-synci @gol
761 -mrelax-pic-calls -mno-relax-pic-calls -mmcount-ra-address}
764 @gccoptlist{-mlibfuncs -mno-libfuncs -mepsilon -mno-epsilon -mabi=gnu @gol
765 -mabi=mmixware -mzero-extend -mknuthdiv -mtoplevel-symbols @gol
766 -melf -mbranch-predict -mno-branch-predict -mbase-addresses @gol
767 -mno-base-addresses -msingle-exit -mno-single-exit}
769 @emph{MN10300 Options}
770 @gccoptlist{-mmult-bug -mno-mult-bug @gol
771 -mno-am33 -mam33 -mam33-2 -mam34 @gol
772 -mtune=@var{cpu-type} @gol
773 -mreturn-pointer-on-d0 @gol
774 -mno-crt0 -mrelax -mliw -msetlb}
776 @emph{PDP-11 Options}
777 @gccoptlist{-mfpu -msoft-float -mac0 -mno-ac0 -m40 -m45 -m10 @gol
778 -mbcopy -mbcopy-builtin -mint32 -mno-int16 @gol
779 -mint16 -mno-int32 -mfloat32 -mno-float64 @gol
780 -mfloat64 -mno-float32 -mabshi -mno-abshi @gol
781 -mbranch-expensive -mbranch-cheap @gol
782 -munix-asm -mdec-asm}
784 @emph{picoChip Options}
785 @gccoptlist{-mae=@var{ae_type} -mvliw-lookahead=@var{N} @gol
786 -msymbol-as-address -mno-inefficient-warnings}
788 @emph{PowerPC Options}
789 See RS/6000 and PowerPC Options.
792 @gccoptlist{-msim -mmul=none -mmul=g13 -mmul=rl78}
794 @emph{RS/6000 and PowerPC Options}
795 @gccoptlist{-mcpu=@var{cpu-type} @gol
796 -mtune=@var{cpu-type} @gol
797 -mcmodel=@var{code-model} @gol
798 -mpower -mno-power -mpower2 -mno-power2 @gol
799 -mpowerpc -mpowerpc64 -mno-powerpc @gol
800 -maltivec -mno-altivec @gol
801 -mpowerpc-gpopt -mno-powerpc-gpopt @gol
802 -mpowerpc-gfxopt -mno-powerpc-gfxopt @gol
803 -mmfcrf -mno-mfcrf -mpopcntb -mno-popcntb -mpopcntd -mno-popcntd @gol
804 -mfprnd -mno-fprnd @gol
805 -mcmpb -mno-cmpb -mmfpgpr -mno-mfpgpr -mhard-dfp -mno-hard-dfp @gol
806 -mnew-mnemonics -mold-mnemonics @gol
807 -mfull-toc -mminimal-toc -mno-fp-in-toc -mno-sum-in-toc @gol
808 -m64 -m32 -mxl-compat -mno-xl-compat -mpe @gol
809 -malign-power -malign-natural @gol
810 -msoft-float -mhard-float -mmultiple -mno-multiple @gol
811 -msingle-float -mdouble-float -msimple-fpu @gol
812 -mstring -mno-string -mupdate -mno-update @gol
813 -mavoid-indexed-addresses -mno-avoid-indexed-addresses @gol
814 -mfused-madd -mno-fused-madd -mbit-align -mno-bit-align @gol
815 -mstrict-align -mno-strict-align -mrelocatable @gol
816 -mno-relocatable -mrelocatable-lib -mno-relocatable-lib @gol
817 -mtoc -mno-toc -mlittle -mlittle-endian -mbig -mbig-endian @gol
818 -mdynamic-no-pic -maltivec -mswdiv -msingle-pic-base @gol
819 -mprioritize-restricted-insns=@var{priority} @gol
820 -msched-costly-dep=@var{dependence_type} @gol
821 -minsert-sched-nops=@var{scheme} @gol
822 -mcall-sysv -mcall-netbsd @gol
823 -maix-struct-return -msvr4-struct-return @gol
824 -mabi=@var{abi-type} -msecure-plt -mbss-plt @gol
825 -mblock-move-inline-limit=@var{num} @gol
826 -misel -mno-isel @gol
827 -misel=yes -misel=no @gol
829 -mspe=yes -mspe=no @gol
831 -mgen-cell-microcode -mwarn-cell-microcode @gol
832 -mvrsave -mno-vrsave @gol
833 -mmulhw -mno-mulhw @gol
834 -mdlmzb -mno-dlmzb @gol
835 -mfloat-gprs=yes -mfloat-gprs=no -mfloat-gprs=single -mfloat-gprs=double @gol
836 -mprototype -mno-prototype @gol
837 -msim -mmvme -mads -myellowknife -memb -msdata @gol
838 -msdata=@var{opt} -mvxworks -G @var{num} -pthread @gol
839 -mrecip -mrecip=@var{opt} -mno-recip -mrecip-precision @gol
840 -mno-recip-precision @gol
841 -mveclibabi=@var{type} -mfriz -mno-friz @gol
842 -mpointers-to-nested-functions -mno-pointers-to-nested-functions @gol
843 -msave-toc-indirect -mno-save-toc-indirect}
846 @gccoptlist{-m64bit-doubles -m32bit-doubles -fpu -nofpu@gol
848 -mbig-endian-data -mlittle-endian-data @gol
851 -mas100-syntax -mno-as100-syntax@gol
853 -mmax-constant-size=@gol
856 -msave-acc-in-interrupts}
858 @emph{S/390 and zSeries Options}
859 @gccoptlist{-mtune=@var{cpu-type} -march=@var{cpu-type} @gol
860 -mhard-float -msoft-float -mhard-dfp -mno-hard-dfp @gol
861 -mlong-double-64 -mlong-double-128 @gol
862 -mbackchain -mno-backchain -mpacked-stack -mno-packed-stack @gol
863 -msmall-exec -mno-small-exec -mmvcle -mno-mvcle @gol
864 -m64 -m31 -mdebug -mno-debug -mesa -mzarch @gol
865 -mtpf-trace -mno-tpf-trace -mfused-madd -mno-fused-madd @gol
866 -mwarn-framesize -mwarn-dynamicstack -mstack-size -mstack-guard}
869 @gccoptlist{-meb -mel @gol
873 -mscore5 -mscore5u -mscore7 -mscore7d}
876 @gccoptlist{-m1 -m2 -m2e @gol
877 -m2a-nofpu -m2a-single-only -m2a-single -m2a @gol
879 -m4-nofpu -m4-single-only -m4-single -m4 @gol
880 -m4a-nofpu -m4a-single-only -m4a-single -m4a -m4al @gol
881 -m5-64media -m5-64media-nofpu @gol
882 -m5-32media -m5-32media-nofpu @gol
883 -m5-compact -m5-compact-nofpu @gol
884 -mb -ml -mdalign -mrelax @gol
885 -mbigtable -mfmovd -mhitachi -mrenesas -mno-renesas -mnomacsave @gol
886 -mieee -mno-ieee -mbitops -misize -minline-ic_invalidate -mpadstruct @gol
887 -mspace -mprefergot -musermode -multcost=@var{number} -mdiv=@var{strategy} @gol
888 -mdivsi3_libfunc=@var{name} -mfixed-range=@var{register-range} @gol
889 -madjust-unroll -mindexed-addressing -mgettrcost=@var{number} -mpt-fixed @gol
890 -maccumulate-outgoing-args -minvalid-symbols -msoft-atomic @gol
891 -mbranch-cost=@var{num} -mcbranchdi -mcmpeqdi -mfused-madd -mpretend-cmove}
893 @emph{Solaris 2 Options}
894 @gccoptlist{-mimpure-text -mno-impure-text @gol
898 @gccoptlist{-mcpu=@var{cpu-type} @gol
899 -mtune=@var{cpu-type} @gol
900 -mcmodel=@var{code-model} @gol
901 -mmemory-model=@var{mem-model} @gol
902 -m32 -m64 -mapp-regs -mno-app-regs @gol
903 -mfaster-structs -mno-faster-structs -mflat -mno-flat @gol
904 -mfpu -mno-fpu -mhard-float -msoft-float @gol
905 -mhard-quad-float -msoft-quad-float @gol
907 -mstack-bias -mno-stack-bias @gol
908 -munaligned-doubles -mno-unaligned-doubles @gol
909 -mv8plus -mno-v8plus -mvis -mno-vis @gol
910 -mvis2 -mno-vis2 -mvis3 -mno-vis3 @gol
911 -mfmaf -mno-fmaf -mpopc -mno-popc @gol
915 @gccoptlist{-mwarn-reloc -merror-reloc @gol
916 -msafe-dma -munsafe-dma @gol
918 -msmall-mem -mlarge-mem -mstdmain @gol
919 -mfixed-range=@var{register-range} @gol
921 -maddress-space-conversion -mno-address-space-conversion @gol
922 -mcache-size=@var{cache-size} @gol
923 -matomic-updates -mno-atomic-updates}
925 @emph{System V Options}
926 @gccoptlist{-Qy -Qn -YP,@var{paths} -Ym,@var{dir}}
928 @emph{TILE-Gx Options}
929 @gccoptlist{-mcpu=@var{cpu} -m32 -m64}
931 @emph{TILEPro Options}
932 @gccoptlist{-mcpu=@var{cpu} -m32}
935 @gccoptlist{-mlong-calls -mno-long-calls -mep -mno-ep @gol
936 -mprolog-function -mno-prolog-function -mspace @gol
937 -mtda=@var{n} -msda=@var{n} -mzda=@var{n} @gol
938 -mapp-regs -mno-app-regs @gol
939 -mdisable-callt -mno-disable-callt @gol
942 -mv850e1 -mv850es @gol
947 @gccoptlist{-mg -mgnu -munix}
949 @emph{VxWorks Options}
950 @gccoptlist{-mrtp -non-static -Bstatic -Bdynamic @gol
951 -Xbind-lazy -Xbind-now}
953 @emph{x86-64 Options}
954 See i386 and x86-64 Options.
956 @emph{Xstormy16 Options}
959 @emph{Xtensa Options}
960 @gccoptlist{-mconst16 -mno-const16 @gol
961 -mfused-madd -mno-fused-madd @gol
963 -mserialize-volatile -mno-serialize-volatile @gol
964 -mtext-section-literals -mno-text-section-literals @gol
965 -mtarget-align -mno-target-align @gol
966 -mlongcalls -mno-longcalls}
968 @emph{zSeries Options}
969 See S/390 and zSeries Options.
971 @item Code Generation Options
972 @xref{Code Gen Options,,Options for Code Generation Conventions}.
973 @gccoptlist{-fcall-saved-@var{reg} -fcall-used-@var{reg} @gol
974 -ffixed-@var{reg} -fexceptions @gol
975 -fnon-call-exceptions -funwind-tables @gol
976 -fasynchronous-unwind-tables @gol
977 -finhibit-size-directive -finstrument-functions @gol
978 -finstrument-functions-exclude-function-list=@var{sym},@var{sym},@dots{} @gol
979 -finstrument-functions-exclude-file-list=@var{file},@var{file},@dots{} @gol
980 -fno-common -fno-ident @gol
981 -fpcc-struct-return -fpic -fPIC -fpie -fPIE @gol
982 -fno-jump-tables @gol
983 -frecord-gcc-switches @gol
984 -freg-struct-return -fshort-enums @gol
985 -fshort-double -fshort-wchar @gol
986 -fverbose-asm -fpack-struct[=@var{n}] -fstack-check @gol
987 -fstack-limit-register=@var{reg} -fstack-limit-symbol=@var{sym} @gol
988 -fno-stack-limit -fsplit-stack @gol
989 -fleading-underscore -ftls-model=@var{model} @gol
990 -ftrapv -fwrapv -fbounds-check @gol
991 -fvisibility -fstrict-volatile-bitfields}
995 * Overall Options:: Controlling the kind of output:
996 an executable, object files, assembler files,
997 or preprocessed source.
998 * C Dialect Options:: Controlling the variant of C language compiled.
999 * C++ Dialect Options:: Variations on C++.
1000 * Objective-C and Objective-C++ Dialect Options:: Variations on Objective-C
1002 * Language Independent Options:: Controlling how diagnostics should be
1004 * Warning Options:: How picky should the compiler be?
1005 * Debugging Options:: Symbol tables, measurements, and debugging dumps.
1006 * Optimize Options:: How much optimization?
1007 * Preprocessor Options:: Controlling header files and macro definitions.
1008 Also, getting dependency information for Make.
1009 * Assembler Options:: Passing options to the assembler.
1010 * Link Options:: Specifying libraries and so on.
1011 * Directory Options:: Where to find header files and libraries.
1012 Where to find the compiler executable files.
1013 * Spec Files:: How to pass switches to sub-processes.
1014 * Target Options:: Running a cross-compiler, or an old version of GCC.
1017 @node Overall Options
1018 @section Options Controlling the Kind of Output
1020 Compilation can involve up to four stages: preprocessing, compilation
1021 proper, assembly and linking, always in that order. GCC is capable of
1022 preprocessing and compiling several files either into several
1023 assembler input files, or into one assembler input file; then each
1024 assembler input file produces an object file, and linking combines all
1025 the object files (those newly compiled, and those specified as input)
1026 into an executable file.
1028 @cindex file name suffix
1029 For any given input file, the file name suffix determines what kind of
1030 compilation is done:
1034 C source code that must be preprocessed.
1037 C source code that should not be preprocessed.
1040 C++ source code that should not be preprocessed.
1043 Objective-C source code. Note that you must link with the @file{libobjc}
1044 library to make an Objective-C program work.
1047 Objective-C source code that should not be preprocessed.
1051 Objective-C++ source code. Note that you must link with the @file{libobjc}
1052 library to make an Objective-C++ program work. Note that @samp{.M} refers
1053 to a literal capital M@.
1055 @item @var{file}.mii
1056 Objective-C++ source code that should not be preprocessed.
1059 C, C++, Objective-C or Objective-C++ header file to be turned into a
1060 precompiled header (default), or C, C++ header file to be turned into an
1061 Ada spec (via the @option{-fdump-ada-spec} switch).
1064 @itemx @var{file}.cp
1065 @itemx @var{file}.cxx
1066 @itemx @var{file}.cpp
1067 @itemx @var{file}.CPP
1068 @itemx @var{file}.c++
1070 C++ source code that must be preprocessed. Note that in @samp{.cxx},
1071 the last two letters must both be literally @samp{x}. Likewise,
1072 @samp{.C} refers to a literal capital C@.
1076 Objective-C++ source code that must be preprocessed.
1078 @item @var{file}.mii
1079 Objective-C++ source code that should not be preprocessed.
1083 @itemx @var{file}.hp
1084 @itemx @var{file}.hxx
1085 @itemx @var{file}.hpp
1086 @itemx @var{file}.HPP
1087 @itemx @var{file}.h++
1088 @itemx @var{file}.tcc
1089 C++ header file to be turned into a precompiled header or Ada spec.
1092 @itemx @var{file}.for
1093 @itemx @var{file}.ftn
1094 Fixed form Fortran source code that should not be preprocessed.
1097 @itemx @var{file}.FOR
1098 @itemx @var{file}.fpp
1099 @itemx @var{file}.FPP
1100 @itemx @var{file}.FTN
1101 Fixed form Fortran source code that must be preprocessed (with the traditional
1104 @item @var{file}.f90
1105 @itemx @var{file}.f95
1106 @itemx @var{file}.f03
1107 @itemx @var{file}.f08
1108 Free form Fortran source code that should not be preprocessed.
1110 @item @var{file}.F90
1111 @itemx @var{file}.F95
1112 @itemx @var{file}.F03
1113 @itemx @var{file}.F08
1114 Free form Fortran source code that must be preprocessed (with the
1115 traditional preprocessor).
1120 @c FIXME: Descriptions of Java file types.
1126 @item @var{file}.ads
1127 Ada source code file that contains a library unit declaration (a
1128 declaration of a package, subprogram, or generic, or a generic
1129 instantiation), or a library unit renaming declaration (a package,
1130 generic, or subprogram renaming declaration). Such files are also
1133 @item @var{file}.adb
1134 Ada source code file containing a library unit body (a subprogram or
1135 package body). Such files are also called @dfn{bodies}.
1137 @c GCC also knows about some suffixes for languages not yet included:
1148 @itemx @var{file}.sx
1149 Assembler code that must be preprocessed.
1152 An object file to be fed straight into linking.
1153 Any file name with no recognized suffix is treated this way.
1157 You can specify the input language explicitly with the @option{-x} option:
1160 @item -x @var{language}
1161 Specify explicitly the @var{language} for the following input files
1162 (rather than letting the compiler choose a default based on the file
1163 name suffix). This option applies to all following input files until
1164 the next @option{-x} option. Possible values for @var{language} are:
1166 c c-header cpp-output
1167 c++ c++-header c++-cpp-output
1168 objective-c objective-c-header objective-c-cpp-output
1169 objective-c++ objective-c++-header objective-c++-cpp-output
1170 assembler assembler-with-cpp
1172 f77 f77-cpp-input f95 f95-cpp-input
1178 Turn off any specification of a language, so that subsequent files are
1179 handled according to their file name suffixes (as they are if @option{-x}
1180 has not been used at all).
1182 @item -pass-exit-codes
1183 @opindex pass-exit-codes
1184 Normally the @command{gcc} program will exit with the code of 1 if any
1185 phase of the compiler returns a non-success return code. If you specify
1186 @option{-pass-exit-codes}, the @command{gcc} program will instead return with
1187 numerically highest error produced by any phase that returned an error
1188 indication. The C, C++, and Fortran frontends return 4, if an internal
1189 compiler error is encountered.
1192 If you only want some of the stages of compilation, you can use
1193 @option{-x} (or filename suffixes) to tell @command{gcc} where to start, and
1194 one of the options @option{-c}, @option{-S}, or @option{-E} to say where
1195 @command{gcc} is to stop. Note that some combinations (for example,
1196 @samp{-x cpp-output -E}) instruct @command{gcc} to do nothing at all.
1201 Compile or assemble the source files, but do not link. The linking
1202 stage simply is not done. The ultimate output is in the form of an
1203 object file for each source file.
1205 By default, the object file name for a source file is made by replacing
1206 the suffix @samp{.c}, @samp{.i}, @samp{.s}, etc., with @samp{.o}.
1208 Unrecognized input files, not requiring compilation or assembly, are
1213 Stop after the stage of compilation proper; do not assemble. The output
1214 is in the form of an assembler code file for each non-assembler input
1217 By default, the assembler file name for a source file is made by
1218 replacing the suffix @samp{.c}, @samp{.i}, etc., with @samp{.s}.
1220 Input files that don't require compilation are ignored.
1224 Stop after the preprocessing stage; do not run the compiler proper. The
1225 output is in the form of preprocessed source code, which is sent to the
1228 Input files that don't require preprocessing are ignored.
1230 @cindex output file option
1233 Place output in file @var{file}. This applies regardless to whatever
1234 sort of output is being produced, whether it be an executable file,
1235 an object file, an assembler file or preprocessed C code.
1237 If @option{-o} is not specified, the default is to put an executable
1238 file in @file{a.out}, the object file for
1239 @file{@var{source}.@var{suffix}} in @file{@var{source}.o}, its
1240 assembler file in @file{@var{source}.s}, a precompiled header file in
1241 @file{@var{source}.@var{suffix}.gch}, and all preprocessed C source on
1246 Print (on standard error output) the commands executed to run the stages
1247 of compilation. Also print the version number of the compiler driver
1248 program and of the preprocessor and the compiler proper.
1252 Like @option{-v} except the commands are not executed and arguments
1253 are quoted unless they contain only alphanumeric characters or @code{./-_}.
1254 This is useful for shell scripts to capture the driver-generated command lines.
1258 Use pipes rather than temporary files for communication between the
1259 various stages of compilation. This fails to work on some systems where
1260 the assembler is unable to read from a pipe; but the GNU assembler has
1265 Print (on the standard output) a description of the command-line options
1266 understood by @command{gcc}. If the @option{-v} option is also specified
1267 then @option{--help} will also be passed on to the various processes
1268 invoked by @command{gcc}, so that they can display the command-line options
1269 they accept. If the @option{-Wextra} option has also been specified
1270 (prior to the @option{--help} option), then command-line options that
1271 have no documentation associated with them will also be displayed.
1274 @opindex target-help
1275 Print (on the standard output) a description of target-specific command-line
1276 options for each tool. For some targets extra target-specific
1277 information may also be printed.
1279 @item --help=@{@var{class}@r{|[}^@r{]}@var{qualifier}@}@r{[},@dots{}@r{]}
1280 Print (on the standard output) a description of the command-line
1281 options understood by the compiler that fit into all specified classes
1282 and qualifiers. These are the supported classes:
1285 @item @samp{optimizers}
1286 This will display all of the optimization options supported by the
1289 @item @samp{warnings}
1290 This will display all of the options controlling warning messages
1291 produced by the compiler.
1294 This will display target-specific options. Unlike the
1295 @option{--target-help} option however, target-specific options of the
1296 linker and assembler will not be displayed. This is because those
1297 tools do not currently support the extended @option{--help=} syntax.
1300 This will display the values recognized by the @option{--param}
1303 @item @var{language}
1304 This will display the options supported for @var{language}, where
1305 @var{language} is the name of one of the languages supported in this
1309 This will display the options that are common to all languages.
1312 These are the supported qualifiers:
1315 @item @samp{undocumented}
1316 Display only those options that are undocumented.
1319 Display options taking an argument that appears after an equal
1320 sign in the same continuous piece of text, such as:
1321 @samp{--help=target}.
1323 @item @samp{separate}
1324 Display options taking an argument that appears as a separate word
1325 following the original option, such as: @samp{-o output-file}.
1328 Thus for example to display all the undocumented target-specific
1329 switches supported by the compiler the following can be used:
1332 --help=target,undocumented
1335 The sense of a qualifier can be inverted by prefixing it with the
1336 @samp{^} character, so for example to display all binary warning
1337 options (i.e., ones that are either on or off and that do not take an
1338 argument) that have a description, use:
1341 --help=warnings,^joined,^undocumented
1344 The argument to @option{--help=} should not consist solely of inverted
1347 Combining several classes is possible, although this usually
1348 restricts the output by so much that there is nothing to display. One
1349 case where it does work however is when one of the classes is
1350 @var{target}. So for example to display all the target-specific
1351 optimization options the following can be used:
1354 --help=target,optimizers
1357 The @option{--help=} option can be repeated on the command line. Each
1358 successive use will display its requested class of options, skipping
1359 those that have already been displayed.
1361 If the @option{-Q} option appears on the command line before the
1362 @option{--help=} option, then the descriptive text displayed by
1363 @option{--help=} is changed. Instead of describing the displayed
1364 options, an indication is given as to whether the option is enabled,
1365 disabled or set to a specific value (assuming that the compiler
1366 knows this at the point where the @option{--help=} option is used).
1368 Here is a truncated example from the ARM port of @command{gcc}:
1371 % gcc -Q -mabi=2 --help=target -c
1372 The following options are target specific:
1374 -mabort-on-noreturn [disabled]
1378 The output is sensitive to the effects of previous command-line
1379 options, so for example it is possible to find out which optimizations
1380 are enabled at @option{-O2} by using:
1383 -Q -O2 --help=optimizers
1386 Alternatively you can discover which binary optimizations are enabled
1387 by @option{-O3} by using:
1390 gcc -c -Q -O3 --help=optimizers > /tmp/O3-opts
1391 gcc -c -Q -O2 --help=optimizers > /tmp/O2-opts
1392 diff /tmp/O2-opts /tmp/O3-opts | grep enabled
1395 @item -no-canonical-prefixes
1396 @opindex no-canonical-prefixes
1397 Do not expand any symbolic links, resolve references to @samp{/../}
1398 or @samp{/./}, or make the path absolute when generating a relative
1403 Display the version number and copyrights of the invoked GCC@.
1407 Invoke all subcommands under a wrapper program. The name of the
1408 wrapper program and its parameters are passed as a comma separated
1412 gcc -c t.c -wrapper gdb,--args
1415 This will invoke all subprograms of @command{gcc} under
1416 @samp{gdb --args}, thus the invocation of @command{cc1} will be
1417 @samp{gdb --args cc1 @dots{}}.
1419 @item -fplugin=@var{name}.so
1420 Load the plugin code in file @var{name}.so, assumed to be a
1421 shared object to be dlopen'd by the compiler. The base name of
1422 the shared object file is used to identify the plugin for the
1423 purposes of argument parsing (See
1424 @option{-fplugin-arg-@var{name}-@var{key}=@var{value}} below).
1425 Each plugin should define the callback functions specified in the
1428 @item -fplugin-arg-@var{name}-@var{key}=@var{value}
1429 Define an argument called @var{key} with a value of @var{value}
1430 for the plugin called @var{name}.
1432 @item -fdump-ada-spec@r{[}-slim@r{]}
1433 For C and C++ source and include files, generate corresponding Ada
1434 specs. @xref{Generating Ada Bindings for C and C++ headers,,, gnat_ugn,
1435 GNAT User's Guide}, which provides detailed documentation on this feature.
1437 @item -fdump-go-spec=@var{file}
1438 For input files in any language, generate corresponding Go
1439 declarations in @var{file}. This generates Go @code{const},
1440 @code{type}, @code{var}, and @code{func} declarations which may be a
1441 useful way to start writing a Go interface to code written in some
1444 @include @value{srcdir}/../libiberty/at-file.texi
1448 @section Compiling C++ Programs
1450 @cindex suffixes for C++ source
1451 @cindex C++ source file suffixes
1452 C++ source files conventionally use one of the suffixes @samp{.C},
1453 @samp{.cc}, @samp{.cpp}, @samp{.CPP}, @samp{.c++}, @samp{.cp}, or
1454 @samp{.cxx}; C++ header files often use @samp{.hh}, @samp{.hpp},
1455 @samp{.H}, or (for shared template code) @samp{.tcc}; and
1456 preprocessed C++ files use the suffix @samp{.ii}. GCC recognizes
1457 files with these names and compiles them as C++ programs even if you
1458 call the compiler the same way as for compiling C programs (usually
1459 with the name @command{gcc}).
1463 However, the use of @command{gcc} does not add the C++ library.
1464 @command{g++} is a program that calls GCC and treats @samp{.c},
1465 @samp{.h} and @samp{.i} files as C++ source files instead of C source
1466 files unless @option{-x} is used, and automatically specifies linking
1467 against the C++ library. This program is also useful when
1468 precompiling a C header file with a @samp{.h} extension for use in C++
1469 compilations. On many systems, @command{g++} is also installed with
1470 the name @command{c++}.
1472 @cindex invoking @command{g++}
1473 When you compile C++ programs, you may specify many of the same
1474 command-line options that you use for compiling programs in any
1475 language; or command-line options meaningful for C and related
1476 languages; or options that are meaningful only for C++ programs.
1477 @xref{C Dialect Options,,Options Controlling C Dialect}, for
1478 explanations of options for languages related to C@.
1479 @xref{C++ Dialect Options,,Options Controlling C++ Dialect}, for
1480 explanations of options that are meaningful only for C++ programs.
1482 @node C Dialect Options
1483 @section Options Controlling C Dialect
1484 @cindex dialect options
1485 @cindex language dialect options
1486 @cindex options, dialect
1488 The following options control the dialect of C (or languages derived
1489 from C, such as C++, Objective-C and Objective-C++) that the compiler
1493 @cindex ANSI support
1497 In C mode, this is equivalent to @samp{-std=c90}. In C++ mode, it is
1498 equivalent to @samp{-std=c++98}.
1500 This turns off certain features of GCC that are incompatible with ISO
1501 C90 (when compiling C code), or of standard C++ (when compiling C++ code),
1502 such as the @code{asm} and @code{typeof} keywords, and
1503 predefined macros such as @code{unix} and @code{vax} that identify the
1504 type of system you are using. It also enables the undesirable and
1505 rarely used ISO trigraph feature. For the C compiler,
1506 it disables recognition of C++ style @samp{//} comments as well as
1507 the @code{inline} keyword.
1509 The alternate keywords @code{__asm__}, @code{__extension__},
1510 @code{__inline__} and @code{__typeof__} continue to work despite
1511 @option{-ansi}. You would not want to use them in an ISO C program, of
1512 course, but it is useful to put them in header files that might be included
1513 in compilations done with @option{-ansi}. Alternate predefined macros
1514 such as @code{__unix__} and @code{__vax__} are also available, with or
1515 without @option{-ansi}.
1517 The @option{-ansi} option does not cause non-ISO programs to be
1518 rejected gratuitously. For that, @option{-pedantic} is required in
1519 addition to @option{-ansi}. @xref{Warning Options}.
1521 The macro @code{__STRICT_ANSI__} is predefined when the @option{-ansi}
1522 option is used. Some header files may notice this macro and refrain
1523 from declaring certain functions or defining certain macros that the
1524 ISO standard doesn't call for; this is to avoid interfering with any
1525 programs that might use these names for other things.
1527 Functions that would normally be built in but do not have semantics
1528 defined by ISO C (such as @code{alloca} and @code{ffs}) are not built-in
1529 functions when @option{-ansi} is used. @xref{Other Builtins,,Other
1530 built-in functions provided by GCC}, for details of the functions
1535 Determine the language standard. @xref{Standards,,Language Standards
1536 Supported by GCC}, for details of these standard versions. This option
1537 is currently only supported when compiling C or C++.
1539 The compiler can accept several base standards, such as @samp{c90} or
1540 @samp{c++98}, and GNU dialects of those standards, such as
1541 @samp{gnu90} or @samp{gnu++98}. By specifying a base standard, the
1542 compiler will accept all programs following that standard and those
1543 using GNU extensions that do not contradict it. For example,
1544 @samp{-std=c90} turns off certain features of GCC that are
1545 incompatible with ISO C90, such as the @code{asm} and @code{typeof}
1546 keywords, but not other GNU extensions that do not have a meaning in
1547 ISO C90, such as omitting the middle term of a @code{?:}
1548 expression. On the other hand, by specifying a GNU dialect of a
1549 standard, all features the compiler support are enabled, even when
1550 those features change the meaning of the base standard and some
1551 strict-conforming programs may be rejected. The particular standard
1552 is used by @option{-pedantic} to identify which features are GNU
1553 extensions given that version of the standard. For example
1554 @samp{-std=gnu90 -pedantic} would warn about C++ style @samp{//}
1555 comments, while @samp{-std=gnu99 -pedantic} would not.
1557 A value for this option must be provided; possible values are
1563 Support all ISO C90 programs (certain GNU extensions that conflict
1564 with ISO C90 are disabled). Same as @option{-ansi} for C code.
1566 @item iso9899:199409
1567 ISO C90 as modified in amendment 1.
1573 ISO C99. Note that this standard is not yet fully supported; see
1574 @w{@uref{http://gcc.gnu.org/gcc-4.7/c99status.html}} for more information. The
1575 names @samp{c9x} and @samp{iso9899:199x} are deprecated.
1580 ISO C11, the 2011 revision of the ISO C standard.
1581 Support is incomplete and experimental. The name @samp{c1x} is
1586 GNU dialect of ISO C90 (including some C99 features). This
1587 is the default for C code.
1591 GNU dialect of ISO C99. When ISO C99 is fully implemented in GCC,
1592 this will become the default. The name @samp{gnu9x} is deprecated.
1596 GNU dialect of ISO C11. Support is incomplete and experimental. The
1597 name @samp{gnu1x} is deprecated.
1600 The 1998 ISO C++ standard plus amendments. Same as @option{-ansi} for
1604 GNU dialect of @option{-std=c++98}. This is the default for
1608 The 2011 ISO C++ standard plus amendments. Support for C++11 is still
1609 experimental, and may change in incompatible ways in future releases.
1612 GNU dialect of @option{-std=c++11}. Support for C++11 is still
1613 experimental, and may change in incompatible ways in future releases.
1616 @item -fgnu89-inline
1617 @opindex fgnu89-inline
1618 The option @option{-fgnu89-inline} tells GCC to use the traditional
1619 GNU semantics for @code{inline} functions when in C99 mode.
1620 @xref{Inline,,An Inline Function is As Fast As a Macro}. This option
1621 is accepted and ignored by GCC versions 4.1.3 up to but not including
1622 4.3. In GCC versions 4.3 and later it changes the behavior of GCC in
1623 C99 mode. Using this option is roughly equivalent to adding the
1624 @code{gnu_inline} function attribute to all inline functions
1625 (@pxref{Function Attributes}).
1627 The option @option{-fno-gnu89-inline} explicitly tells GCC to use the
1628 C99 semantics for @code{inline} when in C99 or gnu99 mode (i.e., it
1629 specifies the default behavior). This option was first supported in
1630 GCC 4.3. This option is not supported in @option{-std=c90} or
1631 @option{-std=gnu90} mode.
1633 The preprocessor macros @code{__GNUC_GNU_INLINE__} and
1634 @code{__GNUC_STDC_INLINE__} may be used to check which semantics are
1635 in effect for @code{inline} functions. @xref{Common Predefined
1636 Macros,,,cpp,The C Preprocessor}.
1638 @item -aux-info @var{filename}
1640 Output to the given filename prototyped declarations for all functions
1641 declared and/or defined in a translation unit, including those in header
1642 files. This option is silently ignored in any language other than C@.
1644 Besides declarations, the file indicates, in comments, the origin of
1645 each declaration (source file and line), whether the declaration was
1646 implicit, prototyped or unprototyped (@samp{I}, @samp{N} for new or
1647 @samp{O} for old, respectively, in the first character after the line
1648 number and the colon), and whether it came from a declaration or a
1649 definition (@samp{C} or @samp{F}, respectively, in the following
1650 character). In the case of function definitions, a K&R-style list of
1651 arguments followed by their declarations is also provided, inside
1652 comments, after the declaration.
1654 @item -fallow-parameterless-variadic-functions
1655 Accept variadic functions without named parameters.
1657 Although it is possible to define such a function, this is not very
1658 useful as it is not possible to read the arguments. This is only
1659 supported for C as this construct is allowed by C++.
1663 Do not recognize @code{asm}, @code{inline} or @code{typeof} as a
1664 keyword, so that code can use these words as identifiers. You can use
1665 the keywords @code{__asm__}, @code{__inline__} and @code{__typeof__}
1666 instead. @option{-ansi} implies @option{-fno-asm}.
1668 In C++, this switch only affects the @code{typeof} keyword, since
1669 @code{asm} and @code{inline} are standard keywords. You may want to
1670 use the @option{-fno-gnu-keywords} flag instead, which has the same
1671 effect. In C99 mode (@option{-std=c99} or @option{-std=gnu99}), this
1672 switch only affects the @code{asm} and @code{typeof} keywords, since
1673 @code{inline} is a standard keyword in ISO C99.
1676 @itemx -fno-builtin-@var{function}
1677 @opindex fno-builtin
1678 @cindex built-in functions
1679 Don't recognize built-in functions that do not begin with
1680 @samp{__builtin_} as prefix. @xref{Other Builtins,,Other built-in
1681 functions provided by GCC}, for details of the functions affected,
1682 including those which are not built-in functions when @option{-ansi} or
1683 @option{-std} options for strict ISO C conformance are used because they
1684 do not have an ISO standard meaning.
1686 GCC normally generates special code to handle certain built-in functions
1687 more efficiently; for instance, calls to @code{alloca} may become single
1688 instructions which adjust the stack directly, and calls to @code{memcpy}
1689 may become inline copy loops. The resulting code is often both smaller
1690 and faster, but since the function calls no longer appear as such, you
1691 cannot set a breakpoint on those calls, nor can you change the behavior
1692 of the functions by linking with a different library. In addition,
1693 when a function is recognized as a built-in function, GCC may use
1694 information about that function to warn about problems with calls to
1695 that function, or to generate more efficient code, even if the
1696 resulting code still contains calls to that function. For example,
1697 warnings are given with @option{-Wformat} for bad calls to
1698 @code{printf}, when @code{printf} is built in, and @code{strlen} is
1699 known not to modify global memory.
1701 With the @option{-fno-builtin-@var{function}} option
1702 only the built-in function @var{function} is
1703 disabled. @var{function} must not begin with @samp{__builtin_}. If a
1704 function is named that is not built-in in this version of GCC, this
1705 option is ignored. There is no corresponding
1706 @option{-fbuiltin-@var{function}} option; if you wish to enable
1707 built-in functions selectively when using @option{-fno-builtin} or
1708 @option{-ffreestanding}, you may define macros such as:
1711 #define abs(n) __builtin_abs ((n))
1712 #define strcpy(d, s) __builtin_strcpy ((d), (s))
1717 @cindex hosted environment
1719 Assert that compilation takes place in a hosted environment. This implies
1720 @option{-fbuiltin}. A hosted environment is one in which the
1721 entire standard library is available, and in which @code{main} has a return
1722 type of @code{int}. Examples are nearly everything except a kernel.
1723 This is equivalent to @option{-fno-freestanding}.
1725 @item -ffreestanding
1726 @opindex ffreestanding
1727 @cindex hosted environment
1729 Assert that compilation takes place in a freestanding environment. This
1730 implies @option{-fno-builtin}. A freestanding environment
1731 is one in which the standard library may not exist, and program startup may
1732 not necessarily be at @code{main}. The most obvious example is an OS kernel.
1733 This is equivalent to @option{-fno-hosted}.
1735 @xref{Standards,,Language Standards Supported by GCC}, for details of
1736 freestanding and hosted environments.
1740 @cindex OpenMP parallel
1741 Enable handling of OpenMP directives @code{#pragma omp} in C/C++ and
1742 @code{!$omp} in Fortran. When @option{-fopenmp} is specified, the
1743 compiler generates parallel code according to the OpenMP Application
1744 Program Interface v3.0 @w{@uref{http://www.openmp.org/}}. This option
1745 implies @option{-pthread}, and thus is only supported on targets that
1746 have support for @option{-pthread}.
1750 When the option @option{-fgnu-tm} is specified, the compiler will
1751 generate code for the Linux variant of Intel's current Transactional
1752 Memory ABI specification document (Revision 1.1, May 6 2009). This is
1753 an experimental feature whose interface may change in future versions
1754 of GCC, as the official specification changes. Please note that not
1755 all architectures are supported for this feature.
1757 For more information on GCC's support for transactional memory,
1758 @xref{Enabling libitm,,The GNU Transactional Memory Library,libitm,GNU
1759 Transactional Memory Library}.
1761 Note that the transactional memory feature is not supported with
1762 non-call exceptions (@option{-fnon-call-exceptions}).
1764 @item -fms-extensions
1765 @opindex fms-extensions
1766 Accept some non-standard constructs used in Microsoft header files.
1768 In C++ code, this allows member names in structures to be similar
1769 to previous types declarations.
1778 Some cases of unnamed fields in structures and unions are only
1779 accepted with this option. @xref{Unnamed Fields,,Unnamed struct/union
1780 fields within structs/unions}, for details.
1782 @item -fplan9-extensions
1783 Accept some non-standard constructs used in Plan 9 code.
1785 This enables @option{-fms-extensions}, permits passing pointers to
1786 structures with anonymous fields to functions that expect pointers to
1787 elements of the type of the field, and permits referring to anonymous
1788 fields declared using a typedef. @xref{Unnamed Fields,,Unnamed
1789 struct/union fields within structs/unions}, for details. This is only
1790 supported for C, not C++.
1794 Support ISO C trigraphs. The @option{-ansi} option (and @option{-std}
1795 options for strict ISO C conformance) implies @option{-trigraphs}.
1797 @item -no-integrated-cpp
1798 @opindex no-integrated-cpp
1799 Performs a compilation in two passes: preprocessing and compiling. This
1800 option allows a user supplied "cc1", "cc1plus", or "cc1obj" via the
1801 @option{-B} option. The user supplied compilation step can then add in
1802 an additional preprocessing step after normal preprocessing but before
1803 compiling. The default is to use the integrated cpp (internal cpp)
1805 The semantics of this option will change if "cc1", "cc1plus", and
1806 "cc1obj" are merged.
1808 @cindex traditional C language
1809 @cindex C language, traditional
1811 @itemx -traditional-cpp
1812 @opindex traditional-cpp
1813 @opindex traditional
1814 Formerly, these options caused GCC to attempt to emulate a pre-standard
1815 C compiler. They are now only supported with the @option{-E} switch.
1816 The preprocessor continues to support a pre-standard mode. See the GNU
1817 CPP manual for details.
1819 @item -fcond-mismatch
1820 @opindex fcond-mismatch
1821 Allow conditional expressions with mismatched types in the second and
1822 third arguments. The value of such an expression is void. This option
1823 is not supported for C++.
1825 @item -flax-vector-conversions
1826 @opindex flax-vector-conversions
1827 Allow implicit conversions between vectors with differing numbers of
1828 elements and/or incompatible element types. This option should not be
1831 @item -funsigned-char
1832 @opindex funsigned-char
1833 Let the type @code{char} be unsigned, like @code{unsigned char}.
1835 Each kind of machine has a default for what @code{char} should
1836 be. It is either like @code{unsigned char} by default or like
1837 @code{signed char} by default.
1839 Ideally, a portable program should always use @code{signed char} or
1840 @code{unsigned char} when it depends on the signedness of an object.
1841 But many programs have been written to use plain @code{char} and
1842 expect it to be signed, or expect it to be unsigned, depending on the
1843 machines they were written for. This option, and its inverse, let you
1844 make such a program work with the opposite default.
1846 The type @code{char} is always a distinct type from each of
1847 @code{signed char} or @code{unsigned char}, even though its behavior
1848 is always just like one of those two.
1851 @opindex fsigned-char
1852 Let the type @code{char} be signed, like @code{signed char}.
1854 Note that this is equivalent to @option{-fno-unsigned-char}, which is
1855 the negative form of @option{-funsigned-char}. Likewise, the option
1856 @option{-fno-signed-char} is equivalent to @option{-funsigned-char}.
1858 @item -fsigned-bitfields
1859 @itemx -funsigned-bitfields
1860 @itemx -fno-signed-bitfields
1861 @itemx -fno-unsigned-bitfields
1862 @opindex fsigned-bitfields
1863 @opindex funsigned-bitfields
1864 @opindex fno-signed-bitfields
1865 @opindex fno-unsigned-bitfields
1866 These options control whether a bit-field is signed or unsigned, when the
1867 declaration does not use either @code{signed} or @code{unsigned}. By
1868 default, such a bit-field is signed, because this is consistent: the
1869 basic integer types such as @code{int} are signed types.
1872 @node C++ Dialect Options
1873 @section Options Controlling C++ Dialect
1875 @cindex compiler options, C++
1876 @cindex C++ options, command-line
1877 @cindex options, C++
1878 This section describes the command-line options that are only meaningful
1879 for C++ programs; but you can also use most of the GNU compiler options
1880 regardless of what language your program is in. For example, you
1881 might compile a file @code{firstClass.C} like this:
1884 g++ -g -frepo -O -c firstClass.C
1888 In this example, only @option{-frepo} is an option meant
1889 only for C++ programs; you can use the other options with any
1890 language supported by GCC@.
1892 Here is a list of options that are @emph{only} for compiling C++ programs:
1896 @item -fabi-version=@var{n}
1897 @opindex fabi-version
1898 Use version @var{n} of the C++ ABI@. Version 2 is the version of the
1899 C++ ABI that first appeared in G++ 3.4. Version 1 is the version of
1900 the C++ ABI that first appeared in G++ 3.2. Version 0 will always be
1901 the version that conforms most closely to the C++ ABI specification.
1902 Therefore, the ABI obtained using version 0 will change as ABI bugs
1905 The default is version 2.
1907 Version 3 corrects an error in mangling a constant address as a
1910 Version 4, which first appeared in G++ 4.5, implements a standard
1911 mangling for vector types.
1913 Version 5, which first appeared in G++ 4.6, corrects the mangling of
1914 attribute const/volatile on function pointer types, decltype of a
1915 plain decl, and use of a function parameter in the declaration of
1918 Version 6, which first appeared in G++ 4.7, corrects the promotion
1919 behavior of C++11 scoped enums and the mangling of template argument
1920 packs, const/static_cast, prefix ++ and --, and a class scope function
1921 used as a template argument.
1923 See also @option{-Wabi}.
1925 @item -fno-access-control
1926 @opindex fno-access-control
1927 Turn off all access checking. This switch is mainly useful for working
1928 around bugs in the access control code.
1932 Check that the pointer returned by @code{operator new} is non-null
1933 before attempting to modify the storage allocated. This check is
1934 normally unnecessary because the C++ standard specifies that
1935 @code{operator new} will only return @code{0} if it is declared
1936 @samp{throw()}, in which case the compiler will always check the
1937 return value even without this option. In all other cases, when
1938 @code{operator new} has a non-empty exception specification, memory
1939 exhaustion is signalled by throwing @code{std::bad_alloc}. See also
1940 @samp{new (nothrow)}.
1942 @item -fconserve-space
1943 @opindex fconserve-space
1944 Put uninitialized or run-time-initialized global variables into the
1945 common segment, as C does. This saves space in the executable at the
1946 cost of not diagnosing duplicate definitions. If you compile with this
1947 flag and your program mysteriously crashes after @code{main()} has
1948 completed, you may have an object that is being destroyed twice because
1949 two definitions were merged.
1951 This option is no longer useful on most targets, now that support has
1952 been added for putting variables into BSS without making them common.
1954 @item -fconstexpr-depth=@var{n}
1955 @opindex fconstexpr-depth
1956 Set the maximum nested evaluation depth for C++11 constexpr functions
1957 to @var{n}. A limit is needed to detect endless recursion during
1958 constant expression evaluation. The minimum specified by the standard
1961 @item -fdeduce-init-list
1962 @opindex fdeduce-init-list
1963 Enable deduction of a template type parameter as
1964 std::initializer_list from a brace-enclosed initializer list, i.e.
1967 template <class T> auto forward(T t) -> decltype (realfn (t))
1974 forward(@{1,2@}); // call forward<std::initializer_list<int>>
1978 This deduction was implemented as a possible extension to the
1979 originally proposed semantics for the C++11 standard, but was not part
1980 of the final standard, so it is disabled by default. This option is
1981 deprecated, and may be removed in a future version of G++.
1983 @item -ffriend-injection
1984 @opindex ffriend-injection
1985 Inject friend functions into the enclosing namespace, so that they are
1986 visible outside the scope of the class in which they are declared.
1987 Friend functions were documented to work this way in the old Annotated
1988 C++ Reference Manual, and versions of G++ before 4.1 always worked
1989 that way. However, in ISO C++ a friend function that is not declared
1990 in an enclosing scope can only be found using argument dependent
1991 lookup. This option causes friends to be injected as they were in
1994 This option is for compatibility, and may be removed in a future
1997 @item -fno-elide-constructors
1998 @opindex fno-elide-constructors
1999 The C++ standard allows an implementation to omit creating a temporary
2000 that is only used to initialize another object of the same type.
2001 Specifying this option disables that optimization, and forces G++ to
2002 call the copy constructor in all cases.
2004 @item -fno-enforce-eh-specs
2005 @opindex fno-enforce-eh-specs
2006 Don't generate code to check for violation of exception specifications
2007 at run time. This option violates the C++ standard, but may be useful
2008 for reducing code size in production builds, much like defining
2009 @samp{NDEBUG}. This does not give user code permission to throw
2010 exceptions in violation of the exception specifications; the compiler
2011 will still optimize based on the specifications, so throwing an
2012 unexpected exception will result in undefined behavior.
2015 @itemx -fno-for-scope
2017 @opindex fno-for-scope
2018 If @option{-ffor-scope} is specified, the scope of variables declared in
2019 a @i{for-init-statement} is limited to the @samp{for} loop itself,
2020 as specified by the C++ standard.
2021 If @option{-fno-for-scope} is specified, the scope of variables declared in
2022 a @i{for-init-statement} extends to the end of the enclosing scope,
2023 as was the case in old versions of G++, and other (traditional)
2024 implementations of C++.
2026 The default if neither flag is given to follow the standard,
2027 but to allow and give a warning for old-style code that would
2028 otherwise be invalid, or have different behavior.
2030 @item -fno-gnu-keywords
2031 @opindex fno-gnu-keywords
2032 Do not recognize @code{typeof} as a keyword, so that code can use this
2033 word as an identifier. You can use the keyword @code{__typeof__} instead.
2034 @option{-ansi} implies @option{-fno-gnu-keywords}.
2036 @item -fno-implicit-templates
2037 @opindex fno-implicit-templates
2038 Never emit code for non-inline templates that are instantiated
2039 implicitly (i.e.@: by use); only emit code for explicit instantiations.
2040 @xref{Template Instantiation}, for more information.
2042 @item -fno-implicit-inline-templates
2043 @opindex fno-implicit-inline-templates
2044 Don't emit code for implicit instantiations of inline templates, either.
2045 The default is to handle inlines differently so that compiles with and
2046 without optimization will need the same set of explicit instantiations.
2048 @item -fno-implement-inlines
2049 @opindex fno-implement-inlines
2050 To save space, do not emit out-of-line copies of inline functions
2051 controlled by @samp{#pragma implementation}. This will cause linker
2052 errors if these functions are not inlined everywhere they are called.
2054 @item -fms-extensions
2055 @opindex fms-extensions
2056 Disable pedantic warnings about constructs used in MFC, such as implicit
2057 int and getting a pointer to member function via non-standard syntax.
2059 @item -fno-nonansi-builtins
2060 @opindex fno-nonansi-builtins
2061 Disable built-in declarations of functions that are not mandated by
2062 ANSI/ISO C@. These include @code{ffs}, @code{alloca}, @code{_exit},
2063 @code{index}, @code{bzero}, @code{conjf}, and other related functions.
2066 @opindex fnothrow-opt
2067 Treat a @code{throw()} exception specification as though it were a
2068 @code{noexcept} specification to reduce or eliminate the text size
2069 overhead relative to a function with no exception specification. If
2070 the function has local variables of types with non-trivial
2071 destructors, the exception specification will actually make the
2072 function smaller because the EH cleanups for those variables can be
2073 optimized away. The semantic effect is that an exception thrown out of
2074 a function with such an exception specification will result in a call
2075 to @code{terminate} rather than @code{unexpected}.
2077 @item -fno-operator-names
2078 @opindex fno-operator-names
2079 Do not treat the operator name keywords @code{and}, @code{bitand},
2080 @code{bitor}, @code{compl}, @code{not}, @code{or} and @code{xor} as
2081 synonyms as keywords.
2083 @item -fno-optional-diags
2084 @opindex fno-optional-diags
2085 Disable diagnostics that the standard says a compiler does not need to
2086 issue. Currently, the only such diagnostic issued by G++ is the one for
2087 a name having multiple meanings within a class.
2090 @opindex fpermissive
2091 Downgrade some diagnostics about nonconformant code from errors to
2092 warnings. Thus, using @option{-fpermissive} will allow some
2093 nonconforming code to compile.
2095 @item -fno-pretty-templates
2096 @opindex fno-pretty-templates
2097 When an error message refers to a specialization of a function
2098 template, the compiler will normally print the signature of the
2099 template followed by the template arguments and any typedefs or
2100 typenames in the signature (e.g. @code{void f(T) [with T = int]}
2101 rather than @code{void f(int)}) so that it's clear which template is
2102 involved. When an error message refers to a specialization of a class
2103 template, the compiler will omit any template arguments that match
2104 the default template arguments for that template. If either of these
2105 behaviors make it harder to understand the error message rather than
2106 easier, using @option{-fno-pretty-templates} will disable them.
2110 Enable automatic template instantiation at link time. This option also
2111 implies @option{-fno-implicit-templates}. @xref{Template
2112 Instantiation}, for more information.
2116 Disable generation of information about every class with virtual
2117 functions for use by the C++ run-time type identification features
2118 (@samp{dynamic_cast} and @samp{typeid}). If you don't use those parts
2119 of the language, you can save some space by using this flag. Note that
2120 exception handling uses the same information, but it will generate it as
2121 needed. The @samp{dynamic_cast} operator can still be used for casts that
2122 do not require run-time type information, i.e.@: casts to @code{void *} or to
2123 unambiguous base classes.
2127 Emit statistics about front-end processing at the end of the compilation.
2128 This information is generally only useful to the G++ development team.
2130 @item -fstrict-enums
2131 @opindex fstrict-enums
2132 Allow the compiler to optimize using the assumption that a value of
2133 enumerated type can only be one of the values of the enumeration (as
2134 defined in the C++ standard; basically, a value that can be
2135 represented in the minimum number of bits needed to represent all the
2136 enumerators). This assumption may not be valid if the program uses a
2137 cast to convert an arbitrary integer value to the enumerated type.
2139 @item -ftemplate-depth=@var{n}
2140 @opindex ftemplate-depth
2141 Set the maximum instantiation depth for template classes to @var{n}.
2142 A limit on the template instantiation depth is needed to detect
2143 endless recursions during template class instantiation. ANSI/ISO C++
2144 conforming programs must not rely on a maximum depth greater than 17
2145 (changed to 1024 in C++11). The default value is 900, as the compiler
2146 can run out of stack space before hitting 1024 in some situations.
2148 @item -fno-threadsafe-statics
2149 @opindex fno-threadsafe-statics
2150 Do not emit the extra code to use the routines specified in the C++
2151 ABI for thread-safe initialization of local statics. You can use this
2152 option to reduce code size slightly in code that doesn't need to be
2155 @item -fuse-cxa-atexit
2156 @opindex fuse-cxa-atexit
2157 Register destructors for objects with static storage duration with the
2158 @code{__cxa_atexit} function rather than the @code{atexit} function.
2159 This option is required for fully standards-compliant handling of static
2160 destructors, but will only work if your C library supports
2161 @code{__cxa_atexit}.
2163 @item -fno-use-cxa-get-exception-ptr
2164 @opindex fno-use-cxa-get-exception-ptr
2165 Don't use the @code{__cxa_get_exception_ptr} runtime routine. This
2166 will cause @code{std::uncaught_exception} to be incorrect, but is necessary
2167 if the runtime routine is not available.
2169 @item -fvisibility-inlines-hidden
2170 @opindex fvisibility-inlines-hidden
2171 This switch declares that the user does not attempt to compare
2172 pointers to inline functions or methods where the addresses of the two functions
2173 were taken in different shared objects.
2175 The effect of this is that GCC may, effectively, mark inline methods with
2176 @code{__attribute__ ((visibility ("hidden")))} so that they do not
2177 appear in the export table of a DSO and do not require a PLT indirection
2178 when used within the DSO@. Enabling this option can have a dramatic effect
2179 on load and link times of a DSO as it massively reduces the size of the
2180 dynamic export table when the library makes heavy use of templates.
2182 The behavior of this switch is not quite the same as marking the
2183 methods as hidden directly, because it does not affect static variables
2184 local to the function or cause the compiler to deduce that
2185 the function is defined in only one shared object.
2187 You may mark a method as having a visibility explicitly to negate the
2188 effect of the switch for that method. For example, if you do want to
2189 compare pointers to a particular inline method, you might mark it as
2190 having default visibility. Marking the enclosing class with explicit
2191 visibility will have no effect.
2193 Explicitly instantiated inline methods are unaffected by this option
2194 as their linkage might otherwise cross a shared library boundary.
2195 @xref{Template Instantiation}.
2197 @item -fvisibility-ms-compat
2198 @opindex fvisibility-ms-compat
2199 This flag attempts to use visibility settings to make GCC's C++
2200 linkage model compatible with that of Microsoft Visual Studio.
2202 The flag makes these changes to GCC's linkage model:
2206 It sets the default visibility to @code{hidden}, like
2207 @option{-fvisibility=hidden}.
2210 Types, but not their members, are not hidden by default.
2213 The One Definition Rule is relaxed for types without explicit
2214 visibility specifications that are defined in more than one different
2215 shared object: those declarations are permitted if they would have
2216 been permitted when this option was not used.
2219 In new code it is better to use @option{-fvisibility=hidden} and
2220 export those classes that are intended to be externally visible.
2221 Unfortunately it is possible for code to rely, perhaps accidentally,
2222 on the Visual Studio behavior.
2224 Among the consequences of these changes are that static data members
2225 of the same type with the same name but defined in different shared
2226 objects will be different, so changing one will not change the other;
2227 and that pointers to function members defined in different shared
2228 objects may not compare equal. When this flag is given, it is a
2229 violation of the ODR to define types with the same name differently.
2233 Do not use weak symbol support, even if it is provided by the linker.
2234 By default, G++ will use weak symbols if they are available. This
2235 option exists only for testing, and should not be used by end-users;
2236 it will result in inferior code and has no benefits. This option may
2237 be removed in a future release of G++.
2241 Do not search for header files in the standard directories specific to
2242 C++, but do still search the other standard directories. (This option
2243 is used when building the C++ library.)
2246 In addition, these optimization, warning, and code generation options
2247 have meanings only for C++ programs:
2250 @item -fno-default-inline
2251 @opindex fno-default-inline
2252 Do not assume @samp{inline} for functions defined inside a class scope.
2253 @xref{Optimize Options,,Options That Control Optimization}. Note that these
2254 functions will have linkage like inline functions; they just won't be
2257 @item -Wabi @r{(C, Objective-C, C++ and Objective-C++ only)}
2260 Warn when G++ generates code that is probably not compatible with the
2261 vendor-neutral C++ ABI@. Although an effort has been made to warn about
2262 all such cases, there are probably some cases that are not warned about,
2263 even though G++ is generating incompatible code. There may also be
2264 cases where warnings are emitted even though the code that is generated
2267 You should rewrite your code to avoid these warnings if you are
2268 concerned about the fact that code generated by G++ may not be binary
2269 compatible with code generated by other compilers.
2271 The known incompatibilities in @option{-fabi-version=2} (the default) include:
2276 A template with a non-type template parameter of reference type is
2277 mangled incorrectly:
2280 template <int &> struct S @{@};
2284 This is fixed in @option{-fabi-version=3}.
2287 SIMD vector types declared using @code{__attribute ((vector_size))} are
2288 mangled in a non-standard way that does not allow for overloading of
2289 functions taking vectors of different sizes.
2291 The mangling is changed in @option{-fabi-version=4}.
2294 The known incompatibilities in @option{-fabi-version=1} include:
2299 Incorrect handling of tail-padding for bit-fields. G++ may attempt to
2300 pack data into the same byte as a base class. For example:
2303 struct A @{ virtual void f(); int f1 : 1; @};
2304 struct B : public A @{ int f2 : 1; @};
2308 In this case, G++ will place @code{B::f2} into the same byte
2309 as@code{A::f1}; other compilers will not. You can avoid this problem
2310 by explicitly padding @code{A} so that its size is a multiple of the
2311 byte size on your platform; that will cause G++ and other compilers to
2312 layout @code{B} identically.
2315 Incorrect handling of tail-padding for virtual bases. G++ does not use
2316 tail padding when laying out virtual bases. For example:
2319 struct A @{ virtual void f(); char c1; @};
2320 struct B @{ B(); char c2; @};
2321 struct C : public A, public virtual B @{@};
2325 In this case, G++ will not place @code{B} into the tail-padding for
2326 @code{A}; other compilers will. You can avoid this problem by
2327 explicitly padding @code{A} so that its size is a multiple of its
2328 alignment (ignoring virtual base classes); that will cause G++ and other
2329 compilers to layout @code{C} identically.
2332 Incorrect handling of bit-fields with declared widths greater than that
2333 of their underlying types, when the bit-fields appear in a union. For
2337 union U @{ int i : 4096; @};
2341 Assuming that an @code{int} does not have 4096 bits, G++ will make the
2342 union too small by the number of bits in an @code{int}.
2345 Empty classes can be placed at incorrect offsets. For example:
2355 struct C : public B, public A @{@};
2359 G++ will place the @code{A} base class of @code{C} at a nonzero offset;
2360 it should be placed at offset zero. G++ mistakenly believes that the
2361 @code{A} data member of @code{B} is already at offset zero.
2364 Names of template functions whose types involve @code{typename} or
2365 template template parameters can be mangled incorrectly.
2368 template <typename Q>
2369 void f(typename Q::X) @{@}
2371 template <template <typename> class Q>
2372 void f(typename Q<int>::X) @{@}
2376 Instantiations of these templates may be mangled incorrectly.
2380 It also warns psABI related changes. The known psABI changes at this
2386 For SYSV/x86-64, when passing union with long double, it is changed to
2387 pass in memory as specified in psABI. For example:
2397 @code{union U} will always be passed in memory.
2401 @item -Wctor-dtor-privacy @r{(C++ and Objective-C++ only)}
2402 @opindex Wctor-dtor-privacy
2403 @opindex Wno-ctor-dtor-privacy
2404 Warn when a class seems unusable because all the constructors or
2405 destructors in that class are private, and it has neither friends nor
2406 public static member functions.
2408 @item -Wdelete-non-virtual-dtor @r{(C++ and Objective-C++ only)}
2409 @opindex Wdelete-non-virtual-dtor
2410 @opindex Wno-delete-non-virtual-dtor
2411 Warn when @samp{delete} is used to destroy an instance of a class that
2412 has virtual functions and non-virtual destructor. It is unsafe to delete
2413 an instance of a derived class through a pointer to a base class if the
2414 base class does not have a virtual destructor. This warning is enabled
2417 @item -Wnarrowing @r{(C++ and Objective-C++ only)}
2419 @opindex Wno-narrowing
2420 Warn when a narrowing conversion prohibited by C++11 occurs within
2424 int i = @{ 2.2 @}; // error: narrowing from double to int
2427 This flag is included in @option{-Wall} and @option{-Wc++11-compat}.
2429 With -std=c++11, @option{-Wno-narrowing} suppresses the diagnostic
2430 required by the standard. Note that this does not affect the meaning
2431 of well-formed code; narrowing conversions are still considered
2432 ill-formed in SFINAE context.
2434 @item -Wnoexcept @r{(C++ and Objective-C++ only)}
2436 @opindex Wno-noexcept
2437 Warn when a noexcept-expression evaluates to false because of a call
2438 to a function that does not have a non-throwing exception
2439 specification (i.e. @samp{throw()} or @samp{noexcept}) but is known by
2440 the compiler to never throw an exception.
2442 @item -Wnon-virtual-dtor @r{(C++ and Objective-C++ only)}
2443 @opindex Wnon-virtual-dtor
2444 @opindex Wno-non-virtual-dtor
2445 Warn when a class has virtual functions and accessible non-virtual
2446 destructor, in which case it would be possible but unsafe to delete
2447 an instance of a derived class through a pointer to the base class.
2448 This warning is also enabled if @option{-Weffc++} is specified.
2450 @item -Wreorder @r{(C++ and Objective-C++ only)}
2452 @opindex Wno-reorder
2453 @cindex reordering, warning
2454 @cindex warning for reordering of member initializers
2455 Warn when the order of member initializers given in the code does not
2456 match the order in which they must be executed. For instance:
2462 A(): j (0), i (1) @{ @}
2466 The compiler will rearrange the member initializers for @samp{i}
2467 and @samp{j} to match the declaration order of the members, emitting
2468 a warning to that effect. This warning is enabled by @option{-Wall}.
2471 The following @option{-W@dots{}} options are not affected by @option{-Wall}.
2474 @item -Weffc++ @r{(C++ and Objective-C++ only)}
2477 Warn about violations of the following style guidelines from Scott Meyers'
2478 @cite{Effective C++, Second Edition} book:
2482 Item 11: Define a copy constructor and an assignment operator for classes
2483 with dynamically allocated memory.
2486 Item 12: Prefer initialization to assignment in constructors.
2489 Item 14: Make destructors virtual in base classes.
2492 Item 15: Have @code{operator=} return a reference to @code{*this}.
2495 Item 23: Don't try to return a reference when you must return an object.
2499 Also warn about violations of the following style guidelines from
2500 Scott Meyers' @cite{More Effective C++} book:
2504 Item 6: Distinguish between prefix and postfix forms of increment and
2505 decrement operators.
2508 Item 7: Never overload @code{&&}, @code{||}, or @code{,}.
2512 When selecting this option, be aware that the standard library
2513 headers do not obey all of these guidelines; use @samp{grep -v}
2514 to filter out those warnings.
2516 @item -Wstrict-null-sentinel @r{(C++ and Objective-C++ only)}
2517 @opindex Wstrict-null-sentinel
2518 @opindex Wno-strict-null-sentinel
2519 Warn also about the use of an uncasted @code{NULL} as sentinel. When
2520 compiling only with GCC this is a valid sentinel, as @code{NULL} is defined
2521 to @code{__null}. Although it is a null pointer constant not a null pointer,
2522 it is guaranteed to be of the same size as a pointer. But this use is
2523 not portable across different compilers.
2525 @item -Wno-non-template-friend @r{(C++ and Objective-C++ only)}
2526 @opindex Wno-non-template-friend
2527 @opindex Wnon-template-friend
2528 Disable warnings when non-templatized friend functions are declared
2529 within a template. Since the advent of explicit template specification
2530 support in G++, if the name of the friend is an unqualified-id (i.e.,
2531 @samp{friend foo(int)}), the C++ language specification demands that the
2532 friend declare or define an ordinary, nontemplate function. (Section
2533 14.5.3). Before G++ implemented explicit specification, unqualified-ids
2534 could be interpreted as a particular specialization of a templatized
2535 function. Because this non-conforming behavior is no longer the default
2536 behavior for G++, @option{-Wnon-template-friend} allows the compiler to
2537 check existing code for potential trouble spots and is on by default.
2538 This new compiler behavior can be turned off with
2539 @option{-Wno-non-template-friend}, which keeps the conformant compiler code
2540 but disables the helpful warning.
2542 @item -Wold-style-cast @r{(C++ and Objective-C++ only)}
2543 @opindex Wold-style-cast
2544 @opindex Wno-old-style-cast
2545 Warn if an old-style (C-style) cast to a non-void type is used within
2546 a C++ program. The new-style casts (@samp{dynamic_cast},
2547 @samp{static_cast}, @samp{reinterpret_cast}, and @samp{const_cast}) are
2548 less vulnerable to unintended effects and much easier to search for.
2550 @item -Woverloaded-virtual @r{(C++ and Objective-C++ only)}
2551 @opindex Woverloaded-virtual
2552 @opindex Wno-overloaded-virtual
2553 @cindex overloaded virtual function, warning
2554 @cindex warning for overloaded virtual function
2555 Warn when a function declaration hides virtual functions from a
2556 base class. For example, in:
2563 struct B: public A @{
2568 the @code{A} class version of @code{f} is hidden in @code{B}, and code
2576 will fail to compile.
2578 @item -Wno-pmf-conversions @r{(C++ and Objective-C++ only)}
2579 @opindex Wno-pmf-conversions
2580 @opindex Wpmf-conversions
2581 Disable the diagnostic for converting a bound pointer to member function
2584 @item -Wsign-promo @r{(C++ and Objective-C++ only)}
2585 @opindex Wsign-promo
2586 @opindex Wno-sign-promo
2587 Warn when overload resolution chooses a promotion from unsigned or
2588 enumerated type to a signed type, over a conversion to an unsigned type of
2589 the same size. Previous versions of G++ would try to preserve
2590 unsignedness, but the standard mandates the current behavior.
2595 A& operator = (int);
2605 In this example, G++ will synthesize a default @samp{A& operator =
2606 (const A&);}, while cfront will use the user-defined @samp{operator =}.
2609 @node Objective-C and Objective-C++ Dialect Options
2610 @section Options Controlling Objective-C and Objective-C++ Dialects
2612 @cindex compiler options, Objective-C and Objective-C++
2613 @cindex Objective-C and Objective-C++ options, command-line
2614 @cindex options, Objective-C and Objective-C++
2615 (NOTE: This manual does not describe the Objective-C and Objective-C++
2616 languages themselves. @xref{Standards,,Language Standards
2617 Supported by GCC}, for references.)
2619 This section describes the command-line options that are only meaningful
2620 for Objective-C and Objective-C++ programs, but you can also use most of
2621 the language-independent GNU compiler options.
2622 For example, you might compile a file @code{some_class.m} like this:
2625 gcc -g -fgnu-runtime -O -c some_class.m
2629 In this example, @option{-fgnu-runtime} is an option meant only for
2630 Objective-C and Objective-C++ programs; you can use the other options with
2631 any language supported by GCC@.
2633 Note that since Objective-C is an extension of the C language, Objective-C
2634 compilations may also use options specific to the C front-end (e.g.,
2635 @option{-Wtraditional}). Similarly, Objective-C++ compilations may use
2636 C++-specific options (e.g., @option{-Wabi}).
2638 Here is a list of options that are @emph{only} for compiling Objective-C
2639 and Objective-C++ programs:
2642 @item -fconstant-string-class=@var{class-name}
2643 @opindex fconstant-string-class
2644 Use @var{class-name} as the name of the class to instantiate for each
2645 literal string specified with the syntax @code{@@"@dots{}"}. The default
2646 class name is @code{NXConstantString} if the GNU runtime is being used, and
2647 @code{NSConstantString} if the NeXT runtime is being used (see below). The
2648 @option{-fconstant-cfstrings} option, if also present, will override the
2649 @option{-fconstant-string-class} setting and cause @code{@@"@dots{}"} literals
2650 to be laid out as constant CoreFoundation strings.
2653 @opindex fgnu-runtime
2654 Generate object code compatible with the standard GNU Objective-C
2655 runtime. This is the default for most types of systems.
2657 @item -fnext-runtime
2658 @opindex fnext-runtime
2659 Generate output compatible with the NeXT runtime. This is the default
2660 for NeXT-based systems, including Darwin and Mac OS X@. The macro
2661 @code{__NEXT_RUNTIME__} is predefined if (and only if) this option is
2664 @item -fno-nil-receivers
2665 @opindex fno-nil-receivers
2666 Assume that all Objective-C message dispatches (@code{[receiver
2667 message:arg]}) in this translation unit ensure that the receiver is
2668 not @code{nil}. This allows for more efficient entry points in the
2669 runtime to be used. This option is only available in conjunction with
2670 the NeXT runtime and ABI version 0 or 1.
2672 @item -fobjc-abi-version=@var{n}
2673 @opindex fobjc-abi-version
2674 Use version @var{n} of the Objective-C ABI for the selected runtime.
2675 This option is currently supported only for the NeXT runtime. In that
2676 case, Version 0 is the traditional (32-bit) ABI without support for
2677 properties and other Objective-C 2.0 additions. Version 1 is the
2678 traditional (32-bit) ABI with support for properties and other
2679 Objective-C 2.0 additions. Version 2 is the modern (64-bit) ABI. If
2680 nothing is specified, the default is Version 0 on 32-bit target
2681 machines, and Version 2 on 64-bit target machines.
2683 @item -fobjc-call-cxx-cdtors
2684 @opindex fobjc-call-cxx-cdtors
2685 For each Objective-C class, check if any of its instance variables is a
2686 C++ object with a non-trivial default constructor. If so, synthesize a
2687 special @code{- (id) .cxx_construct} instance method which will run
2688 non-trivial default constructors on any such instance variables, in order,
2689 and then return @code{self}. Similarly, check if any instance variable
2690 is a C++ object with a non-trivial destructor, and if so, synthesize a
2691 special @code{- (void) .cxx_destruct} method which will run
2692 all such default destructors, in reverse order.
2694 The @code{- (id) .cxx_construct} and @code{- (void) .cxx_destruct}
2695 methods thusly generated will only operate on instance variables
2696 declared in the current Objective-C class, and not those inherited
2697 from superclasses. It is the responsibility of the Objective-C
2698 runtime to invoke all such methods in an object's inheritance
2699 hierarchy. The @code{- (id) .cxx_construct} methods will be invoked
2700 by the runtime immediately after a new object instance is allocated;
2701 the @code{- (void) .cxx_destruct} methods will be invoked immediately
2702 before the runtime deallocates an object instance.
2704 As of this writing, only the NeXT runtime on Mac OS X 10.4 and later has
2705 support for invoking the @code{- (id) .cxx_construct} and
2706 @code{- (void) .cxx_destruct} methods.
2708 @item -fobjc-direct-dispatch
2709 @opindex fobjc-direct-dispatch
2710 Allow fast jumps to the message dispatcher. On Darwin this is
2711 accomplished via the comm page.
2713 @item -fobjc-exceptions
2714 @opindex fobjc-exceptions
2715 Enable syntactic support for structured exception handling in
2716 Objective-C, similar to what is offered by C++ and Java. This option
2717 is required to use the Objective-C keywords @code{@@try},
2718 @code{@@throw}, @code{@@catch}, @code{@@finally} and
2719 @code{@@synchronized}. This option is available with both the GNU
2720 runtime and the NeXT runtime (but not available in conjunction with
2721 the NeXT runtime on Mac OS X 10.2 and earlier).
2725 Enable garbage collection (GC) in Objective-C and Objective-C++
2726 programs. This option is only available with the NeXT runtime; the
2727 GNU runtime has a different garbage collection implementation that
2728 does not require special compiler flags.
2730 @item -fobjc-nilcheck
2731 @opindex fobjc-nilcheck
2732 For the NeXT runtime with version 2 of the ABI, check for a nil
2733 receiver in method invocations before doing the actual method call.
2734 This is the default and can be disabled using
2735 @option{-fno-objc-nilcheck}. Class methods and super calls are never
2736 checked for nil in this way no matter what this flag is set to.
2737 Currently this flag does nothing when the GNU runtime, or an older
2738 version of the NeXT runtime ABI, is used.
2740 @item -fobjc-std=objc1
2742 Conform to the language syntax of Objective-C 1.0, the language
2743 recognized by GCC 4.0. This only affects the Objective-C additions to
2744 the C/C++ language; it does not affect conformance to C/C++ standards,
2745 which is controlled by the separate C/C++ dialect option flags. When
2746 this option is used with the Objective-C or Objective-C++ compiler,
2747 any Objective-C syntax that is not recognized by GCC 4.0 is rejected.
2748 This is useful if you need to make sure that your Objective-C code can
2749 be compiled with older versions of GCC.
2751 @item -freplace-objc-classes
2752 @opindex freplace-objc-classes
2753 Emit a special marker instructing @command{ld(1)} not to statically link in
2754 the resulting object file, and allow @command{dyld(1)} to load it in at
2755 run time instead. This is used in conjunction with the Fix-and-Continue
2756 debugging mode, where the object file in question may be recompiled and
2757 dynamically reloaded in the course of program execution, without the need
2758 to restart the program itself. Currently, Fix-and-Continue functionality
2759 is only available in conjunction with the NeXT runtime on Mac OS X 10.3
2764 When compiling for the NeXT runtime, the compiler ordinarily replaces calls
2765 to @code{objc_getClass("@dots{}")} (when the name of the class is known at
2766 compile time) with static class references that get initialized at load time,
2767 which improves run-time performance. Specifying the @option{-fzero-link} flag
2768 suppresses this behavior and causes calls to @code{objc_getClass("@dots{}")}
2769 to be retained. This is useful in Zero-Link debugging mode, since it allows
2770 for individual class implementations to be modified during program execution.
2771 The GNU runtime currently always retains calls to @code{objc_get_class("@dots{}")}
2772 regardless of command-line options.
2776 Dump interface declarations for all classes seen in the source file to a
2777 file named @file{@var{sourcename}.decl}.
2779 @item -Wassign-intercept @r{(Objective-C and Objective-C++ only)}
2780 @opindex Wassign-intercept
2781 @opindex Wno-assign-intercept
2782 Warn whenever an Objective-C assignment is being intercepted by the
2785 @item -Wno-protocol @r{(Objective-C and Objective-C++ only)}
2786 @opindex Wno-protocol
2788 If a class is declared to implement a protocol, a warning is issued for
2789 every method in the protocol that is not implemented by the class. The
2790 default behavior is to issue a warning for every method not explicitly
2791 implemented in the class, even if a method implementation is inherited
2792 from the superclass. If you use the @option{-Wno-protocol} option, then
2793 methods inherited from the superclass are considered to be implemented,
2794 and no warning is issued for them.
2796 @item -Wselector @r{(Objective-C and Objective-C++ only)}
2798 @opindex Wno-selector
2799 Warn if multiple methods of different types for the same selector are
2800 found during compilation. The check is performed on the list of methods
2801 in the final stage of compilation. Additionally, a check is performed
2802 for each selector appearing in a @code{@@selector(@dots{})}
2803 expression, and a corresponding method for that selector has been found
2804 during compilation. Because these checks scan the method table only at
2805 the end of compilation, these warnings are not produced if the final
2806 stage of compilation is not reached, for example because an error is
2807 found during compilation, or because the @option{-fsyntax-only} option is
2810 @item -Wstrict-selector-match @r{(Objective-C and Objective-C++ only)}
2811 @opindex Wstrict-selector-match
2812 @opindex Wno-strict-selector-match
2813 Warn if multiple methods with differing argument and/or return types are
2814 found for a given selector when attempting to send a message using this
2815 selector to a receiver of type @code{id} or @code{Class}. When this flag
2816 is off (which is the default behavior), the compiler will omit such warnings
2817 if any differences found are confined to types that share the same size
2820 @item -Wundeclared-selector @r{(Objective-C and Objective-C++ only)}
2821 @opindex Wundeclared-selector
2822 @opindex Wno-undeclared-selector
2823 Warn if a @code{@@selector(@dots{})} expression referring to an
2824 undeclared selector is found. A selector is considered undeclared if no
2825 method with that name has been declared before the
2826 @code{@@selector(@dots{})} expression, either explicitly in an
2827 @code{@@interface} or @code{@@protocol} declaration, or implicitly in
2828 an @code{@@implementation} section. This option always performs its
2829 checks as soon as a @code{@@selector(@dots{})} expression is found,
2830 while @option{-Wselector} only performs its checks in the final stage of
2831 compilation. This also enforces the coding style convention
2832 that methods and selectors must be declared before being used.
2834 @item -print-objc-runtime-info
2835 @opindex print-objc-runtime-info
2836 Generate C header describing the largest structure that is passed by
2841 @node Language Independent Options
2842 @section Options to Control Diagnostic Messages Formatting
2843 @cindex options to control diagnostics formatting
2844 @cindex diagnostic messages
2845 @cindex message formatting
2847 Traditionally, diagnostic messages have been formatted irrespective of
2848 the output device's aspect (e.g.@: its width, @dots{}). The options described
2849 below can be used to control the diagnostic messages formatting
2850 algorithm, e.g.@: how many characters per line, how often source location
2851 information should be reported. Right now, only the C++ front end can
2852 honor these options. However it is expected, in the near future, that
2853 the remaining front ends would be able to digest them correctly.
2856 @item -fmessage-length=@var{n}
2857 @opindex fmessage-length
2858 Try to format error messages so that they fit on lines of about @var{n}
2859 characters. The default is 72 characters for @command{g++} and 0 for the rest of
2860 the front ends supported by GCC@. If @var{n} is zero, then no
2861 line-wrapping will be done; each error message will appear on a single
2864 @opindex fdiagnostics-show-location
2865 @item -fdiagnostics-show-location=once
2866 Only meaningful in line-wrapping mode. Instructs the diagnostic messages
2867 reporter to emit @emph{once} source location information; that is, in
2868 case the message is too long to fit on a single physical line and has to
2869 be wrapped, the source location won't be emitted (as prefix) again,
2870 over and over, in subsequent continuation lines. This is the default
2873 @item -fdiagnostics-show-location=every-line
2874 Only meaningful in line-wrapping mode. Instructs the diagnostic
2875 messages reporter to emit the same source location information (as
2876 prefix) for physical lines that result from the process of breaking
2877 a message which is too long to fit on a single line.
2879 @item -fno-diagnostics-show-option
2880 @opindex fno-diagnostics-show-option
2881 @opindex fdiagnostics-show-option
2882 By default, each diagnostic emitted includes text indicating the
2883 command-line option that directly controls the diagnostic (if such an
2884 option is known to the diagnostic machinery). Specifying the
2885 @option{-fno-diagnostics-show-option} flag suppresses that behavior.
2889 @node Warning Options
2890 @section Options to Request or Suppress Warnings
2891 @cindex options to control warnings
2892 @cindex warning messages
2893 @cindex messages, warning
2894 @cindex suppressing warnings
2896 Warnings are diagnostic messages that report constructions that
2897 are not inherently erroneous but that are risky or suggest there
2898 may have been an error.
2900 The following language-independent options do not enable specific
2901 warnings but control the kinds of diagnostics produced by GCC.
2904 @cindex syntax checking
2906 @opindex fsyntax-only
2907 Check the code for syntax errors, but don't do anything beyond that.
2909 @item -fmax-errors=@var{n}
2910 @opindex fmax-errors
2911 Limits the maximum number of error messages to @var{n}, at which point
2912 GCC bails out rather than attempting to continue processing the source
2913 code. If @var{n} is 0 (the default), there is no limit on the number
2914 of error messages produced. If @option{-Wfatal-errors} is also
2915 specified, then @option{-Wfatal-errors} takes precedence over this
2920 Inhibit all warning messages.
2925 Make all warnings into errors.
2930 Make the specified warning into an error. The specifier for a warning
2931 is appended, for example @option{-Werror=switch} turns the warnings
2932 controlled by @option{-Wswitch} into errors. This switch takes a
2933 negative form, to be used to negate @option{-Werror} for specific
2934 warnings, for example @option{-Wno-error=switch} makes
2935 @option{-Wswitch} warnings not be errors, even when @option{-Werror}
2938 The warning message for each controllable warning includes the
2939 option that controls the warning. That option can then be used with
2940 @option{-Werror=} and @option{-Wno-error=} as described above.
2941 (Printing of the option in the warning message can be disabled using the
2942 @option{-fno-diagnostics-show-option} flag.)
2944 Note that specifying @option{-Werror=}@var{foo} automatically implies
2945 @option{-W}@var{foo}. However, @option{-Wno-error=}@var{foo} does not
2948 @item -Wfatal-errors
2949 @opindex Wfatal-errors
2950 @opindex Wno-fatal-errors
2951 This option causes the compiler to abort compilation on the first error
2952 occurred rather than trying to keep going and printing further error
2957 You can request many specific warnings with options beginning
2958 @samp{-W}, for example @option{-Wimplicit} to request warnings on
2959 implicit declarations. Each of these specific warning options also
2960 has a negative form beginning @samp{-Wno-} to turn off warnings; for
2961 example, @option{-Wno-implicit}. This manual lists only one of the
2962 two forms, whichever is not the default. For further,
2963 language-specific options also refer to @ref{C++ Dialect Options} and
2964 @ref{Objective-C and Objective-C++ Dialect Options}.
2966 When an unrecognized warning option is requested (e.g.,
2967 @option{-Wunknown-warning}), GCC will emit a diagnostic stating
2968 that the option is not recognized. However, if the @option{-Wno-} form
2969 is used, the behavior is slightly different: No diagnostic will be
2970 produced for @option{-Wno-unknown-warning} unless other diagnostics
2971 are being produced. This allows the use of new @option{-Wno-} options
2972 with old compilers, but if something goes wrong, the compiler will
2973 warn that an unrecognized option was used.
2978 Issue all the warnings demanded by strict ISO C and ISO C++;
2979 reject all programs that use forbidden extensions, and some other
2980 programs that do not follow ISO C and ISO C++. For ISO C, follows the
2981 version of the ISO C standard specified by any @option{-std} option used.
2983 Valid ISO C and ISO C++ programs should compile properly with or without
2984 this option (though a rare few will require @option{-ansi} or a
2985 @option{-std} option specifying the required version of ISO C)@. However,
2986 without this option, certain GNU extensions and traditional C and C++
2987 features are supported as well. With this option, they are rejected.
2989 @option{-pedantic} does not cause warning messages for use of the
2990 alternate keywords whose names begin and end with @samp{__}. Pedantic
2991 warnings are also disabled in the expression that follows
2992 @code{__extension__}. However, only system header files should use
2993 these escape routes; application programs should avoid them.
2994 @xref{Alternate Keywords}.
2996 Some users try to use @option{-pedantic} to check programs for strict ISO
2997 C conformance. They soon find that it does not do quite what they want:
2998 it finds some non-ISO practices, but not all---only those for which
2999 ISO C @emph{requires} a diagnostic, and some others for which
3000 diagnostics have been added.
3002 A feature to report any failure to conform to ISO C might be useful in
3003 some instances, but would require considerable additional work and would
3004 be quite different from @option{-pedantic}. We don't have plans to
3005 support such a feature in the near future.
3007 Where the standard specified with @option{-std} represents a GNU
3008 extended dialect of C, such as @samp{gnu90} or @samp{gnu99}, there is a
3009 corresponding @dfn{base standard}, the version of ISO C on which the GNU
3010 extended dialect is based. Warnings from @option{-pedantic} are given
3011 where they are required by the base standard. (It would not make sense
3012 for such warnings to be given only for features not in the specified GNU
3013 C dialect, since by definition the GNU dialects of C include all
3014 features the compiler supports with the given option, and there would be
3015 nothing to warn about.)
3017 @item -pedantic-errors
3018 @opindex pedantic-errors
3019 Like @option{-pedantic}, except that errors are produced rather than
3025 This enables all the warnings about constructions that some users
3026 consider questionable, and that are easy to avoid (or modify to
3027 prevent the warning), even in conjunction with macros. This also
3028 enables some language-specific warnings described in @ref{C++ Dialect
3029 Options} and @ref{Objective-C and Objective-C++ Dialect Options}.
3031 @option{-Wall} turns on the following warning flags:
3033 @gccoptlist{-Waddress @gol
3034 -Warray-bounds @r{(only with} @option{-O2}@r{)} @gol
3036 -Wchar-subscripts @gol
3037 -Wenum-compare @r{(in C/Objc; this is on by default in C++)} @gol
3038 -Wimplicit-int @r{(C and Objective-C only)} @gol
3039 -Wimplicit-function-declaration @r{(C and Objective-C only)} @gol
3042 -Wmain @r{(only for C/ObjC and unless} @option{-ffreestanding}@r{)} @gol
3043 -Wmaybe-uninitialized @gol
3044 -Wmissing-braces @gol
3050 -Wsequence-point @gol
3051 -Wsign-compare @r{(only in C++)} @gol
3052 -Wstrict-aliasing @gol
3053 -Wstrict-overflow=1 @gol
3056 -Wuninitialized @gol
3057 -Wunknown-pragmas @gol
3058 -Wunused-function @gol
3061 -Wunused-variable @gol
3062 -Wvolatile-register-var @gol
3065 Note that some warning flags are not implied by @option{-Wall}. Some of
3066 them warn about constructions that users generally do not consider
3067 questionable, but which occasionally you might wish to check for;
3068 others warn about constructions that are necessary or hard to avoid in
3069 some cases, and there is no simple way to modify the code to suppress
3070 the warning. Some of them are enabled by @option{-Wextra} but many of
3071 them must be enabled individually.
3077 This enables some extra warning flags that are not enabled by
3078 @option{-Wall}. (This option used to be called @option{-W}. The older
3079 name is still supported, but the newer name is more descriptive.)
3081 @gccoptlist{-Wclobbered @gol
3083 -Wignored-qualifiers @gol
3084 -Wmissing-field-initializers @gol
3085 -Wmissing-parameter-type @r{(C only)} @gol
3086 -Wold-style-declaration @r{(C only)} @gol
3087 -Woverride-init @gol
3090 -Wuninitialized @gol
3091 -Wunused-parameter @r{(only with} @option{-Wunused} @r{or} @option{-Wall}@r{)} @gol
3092 -Wunused-but-set-parameter @r{(only with} @option{-Wunused} @r{or} @option{-Wall}@r{)} @gol
3095 The option @option{-Wextra} also prints warning messages for the
3101 A pointer is compared against integer zero with @samp{<}, @samp{<=},
3102 @samp{>}, or @samp{>=}.
3105 (C++ only) An enumerator and a non-enumerator both appear in a
3106 conditional expression.
3109 (C++ only) Ambiguous virtual bases.
3112 (C++ only) Subscripting an array that has been declared @samp{register}.
3115 (C++ only) Taking the address of a variable that has been declared
3119 (C++ only) A base class is not initialized in a derived class' copy
3124 @item -Wchar-subscripts
3125 @opindex Wchar-subscripts
3126 @opindex Wno-char-subscripts
3127 Warn if an array subscript has type @code{char}. This is a common cause
3128 of error, as programmers often forget that this type is signed on some
3130 This warning is enabled by @option{-Wall}.
3134 @opindex Wno-comment
3135 Warn whenever a comment-start sequence @samp{/*} appears in a @samp{/*}
3136 comment, or whenever a Backslash-Newline appears in a @samp{//} comment.
3137 This warning is enabled by @option{-Wall}.
3139 @item -Wno-coverage-mismatch
3140 @opindex Wno-coverage-mismatch
3141 Warn if feedback profiles do not match when using the
3142 @option{-fprofile-use} option.
3143 If a source file was changed between @option{-fprofile-gen} and
3144 @option{-fprofile-use}, the files with the profile feedback can fail
3145 to match the source file and GCC cannot use the profile feedback
3146 information. By default, this warning is enabled and is treated as an
3147 error. @option{-Wno-coverage-mismatch} can be used to disable the
3148 warning or @option{-Wno-error=coverage-mismatch} can be used to
3149 disable the error. Disabling the error for this warning can result in
3150 poorly optimized code and is useful only in the
3151 case of very minor changes such as bug fixes to an existing code-base.
3152 Completely disabling the warning is not recommended.
3155 @r{(C, Objective-C, C++, Objective-C++ and Fortran only)}
3157 Suppress warning messages emitted by @code{#warning} directives.
3159 @item -Wdouble-promotion @r{(C, C++, Objective-C and Objective-C++ only)}
3160 @opindex Wdouble-promotion
3161 @opindex Wno-double-promotion
3162 Give a warning when a value of type @code{float} is implicitly
3163 promoted to @code{double}. CPUs with a 32-bit ``single-precision''
3164 floating-point unit implement @code{float} in hardware, but emulate
3165 @code{double} in software. On such a machine, doing computations
3166 using @code{double} values is much more expensive because of the
3167 overhead required for software emulation.
3169 It is easy to accidentally do computations with @code{double} because
3170 floating-point literals are implicitly of type @code{double}. For
3174 float area(float radius)
3176 return 3.14159 * radius * radius;
3180 the compiler will perform the entire computation with @code{double}
3181 because the floating-point literal is a @code{double}.
3186 @opindex ffreestanding
3187 @opindex fno-builtin
3188 Check calls to @code{printf} and @code{scanf}, etc., to make sure that
3189 the arguments supplied have types appropriate to the format string
3190 specified, and that the conversions specified in the format string make
3191 sense. This includes standard functions, and others specified by format
3192 attributes (@pxref{Function Attributes}), in the @code{printf},
3193 @code{scanf}, @code{strftime} and @code{strfmon} (an X/Open extension,
3194 not in the C standard) families (or other target-specific families).
3195 Which functions are checked without format attributes having been
3196 specified depends on the standard version selected, and such checks of
3197 functions without the attribute specified are disabled by
3198 @option{-ffreestanding} or @option{-fno-builtin}.
3200 The formats are checked against the format features supported by GNU
3201 libc version 2.2. These include all ISO C90 and C99 features, as well
3202 as features from the Single Unix Specification and some BSD and GNU
3203 extensions. Other library implementations may not support all these
3204 features; GCC does not support warning about features that go beyond a
3205 particular library's limitations. However, if @option{-pedantic} is used
3206 with @option{-Wformat}, warnings will be given about format features not
3207 in the selected standard version (but not for @code{strfmon} formats,
3208 since those are not in any version of the C standard). @xref{C Dialect
3209 Options,,Options Controlling C Dialect}.
3211 Since @option{-Wformat} also checks for null format arguments for
3212 several functions, @option{-Wformat} also implies @option{-Wnonnull}.
3214 @option{-Wformat} is included in @option{-Wall}. For more control over some
3215 aspects of format checking, the options @option{-Wformat-y2k},
3216 @option{-Wno-format-extra-args}, @option{-Wno-format-zero-length},
3217 @option{-Wformat-nonliteral}, @option{-Wformat-security}, and
3218 @option{-Wformat=2} are available, but are not included in @option{-Wall}.
3221 @opindex Wformat-y2k
3222 @opindex Wno-format-y2k
3223 If @option{-Wformat} is specified, also warn about @code{strftime}
3224 formats that may yield only a two-digit year.
3226 @item -Wno-format-contains-nul
3227 @opindex Wno-format-contains-nul
3228 @opindex Wformat-contains-nul
3229 If @option{-Wformat} is specified, do not warn about format strings that
3232 @item -Wno-format-extra-args
3233 @opindex Wno-format-extra-args
3234 @opindex Wformat-extra-args
3235 If @option{-Wformat} is specified, do not warn about excess arguments to a
3236 @code{printf} or @code{scanf} format function. The C standard specifies
3237 that such arguments are ignored.
3239 Where the unused arguments lie between used arguments that are
3240 specified with @samp{$} operand number specifications, normally
3241 warnings are still given, since the implementation could not know what
3242 type to pass to @code{va_arg} to skip the unused arguments. However,
3243 in the case of @code{scanf} formats, this option will suppress the
3244 warning if the unused arguments are all pointers, since the Single
3245 Unix Specification says that such unused arguments are allowed.
3247 @item -Wno-format-zero-length
3248 @opindex Wno-format-zero-length
3249 @opindex Wformat-zero-length
3250 If @option{-Wformat} is specified, do not warn about zero-length formats.
3251 The C standard specifies that zero-length formats are allowed.
3253 @item -Wformat-nonliteral
3254 @opindex Wformat-nonliteral
3255 @opindex Wno-format-nonliteral
3256 If @option{-Wformat} is specified, also warn if the format string is not a
3257 string literal and so cannot be checked, unless the format function
3258 takes its format arguments as a @code{va_list}.
3260 @item -Wformat-security
3261 @opindex Wformat-security
3262 @opindex Wno-format-security
3263 If @option{-Wformat} is specified, also warn about uses of format
3264 functions that represent possible security problems. At present, this
3265 warns about calls to @code{printf} and @code{scanf} functions where the
3266 format string is not a string literal and there are no format arguments,
3267 as in @code{printf (foo);}. This may be a security hole if the format
3268 string came from untrusted input and contains @samp{%n}. (This is
3269 currently a subset of what @option{-Wformat-nonliteral} warns about, but
3270 in future warnings may be added to @option{-Wformat-security} that are not
3271 included in @option{-Wformat-nonliteral}.)
3275 @opindex Wno-format=2
3276 Enable @option{-Wformat} plus format checks not included in
3277 @option{-Wformat}. Currently equivalent to @samp{-Wformat
3278 -Wformat-nonliteral -Wformat-security -Wformat-y2k}.
3282 @opindex Wno-nonnull
3283 Warn about passing a null pointer for arguments marked as
3284 requiring a non-null value by the @code{nonnull} function attribute.
3286 @option{-Wnonnull} is included in @option{-Wall} and @option{-Wformat}. It
3287 can be disabled with the @option{-Wno-nonnull} option.
3289 @item -Winit-self @r{(C, C++, Objective-C and Objective-C++ only)}
3291 @opindex Wno-init-self
3292 Warn about uninitialized variables that are initialized with themselves.
3293 Note this option can only be used with the @option{-Wuninitialized} option.
3295 For example, GCC will warn about @code{i} being uninitialized in the
3296 following snippet only when @option{-Winit-self} has been specified:
3307 @item -Wimplicit-int @r{(C and Objective-C only)}
3308 @opindex Wimplicit-int
3309 @opindex Wno-implicit-int
3310 Warn when a declaration does not specify a type.
3311 This warning is enabled by @option{-Wall}.
3313 @item -Wimplicit-function-declaration @r{(C and Objective-C only)}
3314 @opindex Wimplicit-function-declaration
3315 @opindex Wno-implicit-function-declaration
3316 Give a warning whenever a function is used before being declared. In
3317 C99 mode (@option{-std=c99} or @option{-std=gnu99}), this warning is
3318 enabled by default and it is made into an error by
3319 @option{-pedantic-errors}. This warning is also enabled by
3322 @item -Wimplicit @r{(C and Objective-C only)}
3324 @opindex Wno-implicit
3325 Same as @option{-Wimplicit-int} and @option{-Wimplicit-function-declaration}.
3326 This warning is enabled by @option{-Wall}.
3328 @item -Wignored-qualifiers @r{(C and C++ only)}
3329 @opindex Wignored-qualifiers
3330 @opindex Wno-ignored-qualifiers
3331 Warn if the return type of a function has a type qualifier
3332 such as @code{const}. For ISO C such a type qualifier has no effect,
3333 since the value returned by a function is not an lvalue.
3334 For C++, the warning is only emitted for scalar types or @code{void}.
3335 ISO C prohibits qualified @code{void} return types on function
3336 definitions, so such return types always receive a warning
3337 even without this option.
3339 This warning is also enabled by @option{-Wextra}.
3344 Warn if the type of @samp{main} is suspicious. @samp{main} should be
3345 a function with external linkage, returning int, taking either zero
3346 arguments, two, or three arguments of appropriate types. This warning
3347 is enabled by default in C++ and is enabled by either @option{-Wall}
3348 or @option{-pedantic}.
3350 @item -Wmissing-braces
3351 @opindex Wmissing-braces
3352 @opindex Wno-missing-braces
3353 Warn if an aggregate or union initializer is not fully bracketed. In
3354 the following example, the initializer for @samp{a} is not fully
3355 bracketed, but that for @samp{b} is fully bracketed.
3358 int a[2][2] = @{ 0, 1, 2, 3 @};
3359 int b[2][2] = @{ @{ 0, 1 @}, @{ 2, 3 @} @};
3362 This warning is enabled by @option{-Wall}.
3364 @item -Wmissing-include-dirs @r{(C, C++, Objective-C and Objective-C++ only)}
3365 @opindex Wmissing-include-dirs
3366 @opindex Wno-missing-include-dirs
3367 Warn if a user-supplied include directory does not exist.
3370 @opindex Wparentheses
3371 @opindex Wno-parentheses
3372 Warn if parentheses are omitted in certain contexts, such
3373 as when there is an assignment in a context where a truth value
3374 is expected, or when operators are nested whose precedence people
3375 often get confused about.
3377 Also warn if a comparison like @samp{x<=y<=z} appears; this is
3378 equivalent to @samp{(x<=y ? 1 : 0) <= z}, which is a different
3379 interpretation from that of ordinary mathematical notation.
3381 Also warn about constructions where there may be confusion to which
3382 @code{if} statement an @code{else} branch belongs. Here is an example of
3397 In C/C++, every @code{else} branch belongs to the innermost possible
3398 @code{if} statement, which in this example is @code{if (b)}. This is
3399 often not what the programmer expected, as illustrated in the above
3400 example by indentation the programmer chose. When there is the
3401 potential for this confusion, GCC will issue a warning when this flag
3402 is specified. To eliminate the warning, add explicit braces around
3403 the innermost @code{if} statement so there is no way the @code{else}
3404 could belong to the enclosing @code{if}. The resulting code would
3421 Also warn for dangerous uses of the
3422 ?: with omitted middle operand GNU extension. When the condition
3423 in the ?: operator is a boolean expression the omitted value will
3424 be always 1. Often the user expects it to be a value computed
3425 inside the conditional expression instead.
3427 This warning is enabled by @option{-Wall}.
3429 @item -Wsequence-point
3430 @opindex Wsequence-point
3431 @opindex Wno-sequence-point
3432 Warn about code that may have undefined semantics because of violations
3433 of sequence point rules in the C and C++ standards.
3435 The C and C++ standards defines the order in which expressions in a C/C++
3436 program are evaluated in terms of @dfn{sequence points}, which represent
3437 a partial ordering between the execution of parts of the program: those
3438 executed before the sequence point, and those executed after it. These
3439 occur after the evaluation of a full expression (one which is not part
3440 of a larger expression), after the evaluation of the first operand of a
3441 @code{&&}, @code{||}, @code{? :} or @code{,} (comma) operator, before a
3442 function is called (but after the evaluation of its arguments and the
3443 expression denoting the called function), and in certain other places.
3444 Other than as expressed by the sequence point rules, the order of
3445 evaluation of subexpressions of an expression is not specified. All
3446 these rules describe only a partial order rather than a total order,
3447 since, for example, if two functions are called within one expression
3448 with no sequence point between them, the order in which the functions
3449 are called is not specified. However, the standards committee have
3450 ruled that function calls do not overlap.
3452 It is not specified when between sequence points modifications to the
3453 values of objects take effect. Programs whose behavior depends on this
3454 have undefined behavior; the C and C++ standards specify that ``Between
3455 the previous and next sequence point an object shall have its stored
3456 value modified at most once by the evaluation of an expression.
3457 Furthermore, the prior value shall be read only to determine the value
3458 to be stored.''. If a program breaks these rules, the results on any
3459 particular implementation are entirely unpredictable.
3461 Examples of code with undefined behavior are @code{a = a++;}, @code{a[n]
3462 = b[n++]} and @code{a[i++] = i;}. Some more complicated cases are not
3463 diagnosed by this option, and it may give an occasional false positive
3464 result, but in general it has been found fairly effective at detecting
3465 this sort of problem in programs.
3467 The standard is worded confusingly, therefore there is some debate
3468 over the precise meaning of the sequence point rules in subtle cases.
3469 Links to discussions of the problem, including proposed formal
3470 definitions, may be found on the GCC readings page, at
3471 @uref{http://gcc.gnu.org/@/readings.html}.
3473 This warning is enabled by @option{-Wall} for C and C++.
3476 @opindex Wreturn-type
3477 @opindex Wno-return-type
3478 Warn whenever a function is defined with a return-type that defaults
3479 to @code{int}. Also warn about any @code{return} statement with no
3480 return-value in a function whose return-type is not @code{void}
3481 (falling off the end of the function body is considered returning
3482 without a value), and about a @code{return} statement with an
3483 expression in a function whose return-type is @code{void}.
3485 For C++, a function without return type always produces a diagnostic
3486 message, even when @option{-Wno-return-type} is specified. The only
3487 exceptions are @samp{main} and functions defined in system headers.
3489 This warning is enabled by @option{-Wall}.
3494 Warn whenever a @code{switch} statement has an index of enumerated type
3495 and lacks a @code{case} for one or more of the named codes of that
3496 enumeration. (The presence of a @code{default} label prevents this
3497 warning.) @code{case} labels outside the enumeration range also
3498 provoke warnings when this option is used (even if there is a
3499 @code{default} label).
3500 This warning is enabled by @option{-Wall}.
3502 @item -Wswitch-default
3503 @opindex Wswitch-default
3504 @opindex Wno-switch-default
3505 Warn whenever a @code{switch} statement does not have a @code{default}
3509 @opindex Wswitch-enum
3510 @opindex Wno-switch-enum
3511 Warn whenever a @code{switch} statement has an index of enumerated type
3512 and lacks a @code{case} for one or more of the named codes of that
3513 enumeration. @code{case} labels outside the enumeration range also
3514 provoke warnings when this option is used. The only difference
3515 between @option{-Wswitch} and this option is that this option gives a
3516 warning about an omitted enumeration code even if there is a
3517 @code{default} label.
3519 @item -Wsync-nand @r{(C and C++ only)}
3521 @opindex Wno-sync-nand
3522 Warn when @code{__sync_fetch_and_nand} and @code{__sync_nand_and_fetch}
3523 built-in functions are used. These functions changed semantics in GCC 4.4.
3527 @opindex Wno-trigraphs
3528 Warn if any trigraphs are encountered that might change the meaning of
3529 the program (trigraphs within comments are not warned about).
3530 This warning is enabled by @option{-Wall}.
3532 @item -Wunused-but-set-parameter
3533 @opindex Wunused-but-set-parameter
3534 @opindex Wno-unused-but-set-parameter
3535 Warn whenever a function parameter is assigned to, but otherwise unused
3536 (aside from its declaration).
3538 To suppress this warning use the @samp{unused} attribute
3539 (@pxref{Variable Attributes}).
3541 This warning is also enabled by @option{-Wunused} together with
3544 @item -Wunused-but-set-variable
3545 @opindex Wunused-but-set-variable
3546 @opindex Wno-unused-but-set-variable
3547 Warn whenever a local variable is assigned to, but otherwise unused
3548 (aside from its declaration).
3549 This warning is enabled by @option{-Wall}.
3551 To suppress this warning use the @samp{unused} attribute
3552 (@pxref{Variable Attributes}).
3554 This warning is also enabled by @option{-Wunused}, which is enabled
3557 @item -Wunused-function
3558 @opindex Wunused-function
3559 @opindex Wno-unused-function
3560 Warn whenever a static function is declared but not defined or a
3561 non-inline static function is unused.
3562 This warning is enabled by @option{-Wall}.
3564 @item -Wunused-label
3565 @opindex Wunused-label
3566 @opindex Wno-unused-label
3567 Warn whenever a label is declared but not used.
3568 This warning is enabled by @option{-Wall}.
3570 To suppress this warning use the @samp{unused} attribute
3571 (@pxref{Variable Attributes}).
3573 @item -Wunused-local-typedefs @r{(C, Objective-C, C++ and Objective-C++ only)}
3574 @opindex Wunused-local-typedefs
3575 Warn when a typedef locally defined in a function is not used.
3577 @item -Wunused-parameter
3578 @opindex Wunused-parameter
3579 @opindex Wno-unused-parameter
3580 Warn whenever a function parameter is unused aside from its declaration.
3582 To suppress this warning use the @samp{unused} attribute
3583 (@pxref{Variable Attributes}).
3585 @item -Wno-unused-result
3586 @opindex Wunused-result
3587 @opindex Wno-unused-result
3588 Do not warn if a caller of a function marked with attribute
3589 @code{warn_unused_result} (@pxref{Function Attributes}) does not use
3590 its return value. The default is @option{-Wunused-result}.
3592 @item -Wunused-variable
3593 @opindex Wunused-variable
3594 @opindex Wno-unused-variable
3595 Warn whenever a local variable or non-constant static variable is unused
3596 aside from its declaration.
3597 This warning is enabled by @option{-Wall}.
3599 To suppress this warning use the @samp{unused} attribute
3600 (@pxref{Variable Attributes}).
3602 @item -Wunused-value
3603 @opindex Wunused-value
3604 @opindex Wno-unused-value
3605 Warn whenever a statement computes a result that is explicitly not
3606 used. To suppress this warning cast the unused expression to
3607 @samp{void}. This includes an expression-statement or the left-hand
3608 side of a comma expression that contains no side effects. For example,
3609 an expression such as @samp{x[i,j]} will cause a warning, while
3610 @samp{x[(void)i,j]} will not.
3612 This warning is enabled by @option{-Wall}.
3617 All the above @option{-Wunused} options combined.
3619 In order to get a warning about an unused function parameter, you must
3620 either specify @samp{-Wextra -Wunused} (note that @samp{-Wall} implies
3621 @samp{-Wunused}), or separately specify @option{-Wunused-parameter}.
3623 @item -Wuninitialized
3624 @opindex Wuninitialized
3625 @opindex Wno-uninitialized
3626 Warn if an automatic variable is used without first being initialized
3627 or if a variable may be clobbered by a @code{setjmp} call. In C++,
3628 warn if a non-static reference or non-static @samp{const} member
3629 appears in a class without constructors.
3631 If you want to warn about code that uses the uninitialized value of the
3632 variable in its own initializer, use the @option{-Winit-self} option.
3634 These warnings occur for individual uninitialized or clobbered
3635 elements of structure, union or array variables as well as for
3636 variables that are uninitialized or clobbered as a whole. They do
3637 not occur for variables or elements declared @code{volatile}. Because
3638 these warnings depend on optimization, the exact variables or elements
3639 for which there are warnings will depend on the precise optimization
3640 options and version of GCC used.
3642 Note that there may be no warning about a variable that is used only
3643 to compute a value that itself is never used, because such
3644 computations may be deleted by data flow analysis before the warnings
3647 @item -Wmaybe-uninitialized
3648 @opindex Wmaybe-uninitialized
3649 @opindex Wno-maybe-uninitialized
3650 For an automatic variable, if there exists a path from the function
3651 entry to a use of the variable that is initialized, but there exist
3652 some other paths the variable is not initialized, the compiler will
3653 emit a warning if it can not prove the uninitialized paths do not
3654 happen at run time. These warnings are made optional because GCC is
3655 not smart enough to see all the reasons why the code might be correct
3656 despite appearing to have an error. Here is one example of how
3677 If the value of @code{y} is always 1, 2 or 3, then @code{x} is
3678 always initialized, but GCC doesn't know this. To suppress the
3679 warning, the user needs to provide a default case with assert(0) or
3682 @cindex @code{longjmp} warnings
3683 This option also warns when a non-volatile automatic variable might be
3684 changed by a call to @code{longjmp}. These warnings as well are possible
3685 only in optimizing compilation.
3687 The compiler sees only the calls to @code{setjmp}. It cannot know
3688 where @code{longjmp} will be called; in fact, a signal handler could
3689 call it at any point in the code. As a result, you may get a warning
3690 even when there is in fact no problem because @code{longjmp} cannot
3691 in fact be called at the place that would cause a problem.
3693 Some spurious warnings can be avoided if you declare all the functions
3694 you use that never return as @code{noreturn}. @xref{Function
3697 This warning is enabled by @option{-Wall} or @option{-Wextra}.
3699 @item -Wunknown-pragmas
3700 @opindex Wunknown-pragmas
3701 @opindex Wno-unknown-pragmas
3702 @cindex warning for unknown pragmas
3703 @cindex unknown pragmas, warning
3704 @cindex pragmas, warning of unknown
3705 Warn when a @code{#pragma} directive is encountered that is not understood by
3706 GCC@. If this command-line option is used, warnings will even be issued
3707 for unknown pragmas in system header files. This is not the case if
3708 the warnings were only enabled by the @option{-Wall} command-line option.
3711 @opindex Wno-pragmas
3713 Do not warn about misuses of pragmas, such as incorrect parameters,
3714 invalid syntax, or conflicts between pragmas. See also
3715 @samp{-Wunknown-pragmas}.
3717 @item -Wstrict-aliasing
3718 @opindex Wstrict-aliasing
3719 @opindex Wno-strict-aliasing
3720 This option is only active when @option{-fstrict-aliasing} is active.
3721 It warns about code that might break the strict aliasing rules that the
3722 compiler is using for optimization. The warning does not catch all
3723 cases, but does attempt to catch the more common pitfalls. It is
3724 included in @option{-Wall}.
3725 It is equivalent to @option{-Wstrict-aliasing=3}
3727 @item -Wstrict-aliasing=n
3728 @opindex Wstrict-aliasing=n
3729 @opindex Wno-strict-aliasing=n
3730 This option is only active when @option{-fstrict-aliasing} is active.
3731 It warns about code that might break the strict aliasing rules that the
3732 compiler is using for optimization.
3733 Higher levels correspond to higher accuracy (fewer false positives).
3734 Higher levels also correspond to more effort, similar to the way -O works.
3735 @option{-Wstrict-aliasing} is equivalent to @option{-Wstrict-aliasing=n},
3738 Level 1: Most aggressive, quick, least accurate.
3739 Possibly useful when higher levels
3740 do not warn but -fstrict-aliasing still breaks the code, as it has very few
3741 false negatives. However, it has many false positives.
3742 Warns for all pointer conversions between possibly incompatible types,
3743 even if never dereferenced. Runs in the front end only.
3745 Level 2: Aggressive, quick, not too precise.
3746 May still have many false positives (not as many as level 1 though),
3747 and few false negatives (but possibly more than level 1).
3748 Unlike level 1, it only warns when an address is taken. Warns about
3749 incomplete types. Runs in the front end only.
3751 Level 3 (default for @option{-Wstrict-aliasing}):
3752 Should have very few false positives and few false
3753 negatives. Slightly slower than levels 1 or 2 when optimization is enabled.
3754 Takes care of the common pun+dereference pattern in the front end:
3755 @code{*(int*)&some_float}.
3756 If optimization is enabled, it also runs in the back end, where it deals
3757 with multiple statement cases using flow-sensitive points-to information.
3758 Only warns when the converted pointer is dereferenced.
3759 Does not warn about incomplete types.
3761 @item -Wstrict-overflow
3762 @itemx -Wstrict-overflow=@var{n}
3763 @opindex Wstrict-overflow
3764 @opindex Wno-strict-overflow
3765 This option is only active when @option{-fstrict-overflow} is active.
3766 It warns about cases where the compiler optimizes based on the
3767 assumption that signed overflow does not occur. Note that it does not
3768 warn about all cases where the code might overflow: it only warns
3769 about cases where the compiler implements some optimization. Thus
3770 this warning depends on the optimization level.
3772 An optimization that assumes that signed overflow does not occur is
3773 perfectly safe if the values of the variables involved are such that
3774 overflow never does, in fact, occur. Therefore this warning can
3775 easily give a false positive: a warning about code that is not
3776 actually a problem. To help focus on important issues, several
3777 warning levels are defined. No warnings are issued for the use of
3778 undefined signed overflow when estimating how many iterations a loop
3779 will require, in particular when determining whether a loop will be
3783 @item -Wstrict-overflow=1
3784 Warn about cases that are both questionable and easy to avoid. For
3785 example: @code{x + 1 > x}; with @option{-fstrict-overflow}, the
3786 compiler will simplify this to @code{1}. This level of
3787 @option{-Wstrict-overflow} is enabled by @option{-Wall}; higher levels
3788 are not, and must be explicitly requested.
3790 @item -Wstrict-overflow=2
3791 Also warn about other cases where a comparison is simplified to a
3792 constant. For example: @code{abs (x) >= 0}. This can only be
3793 simplified when @option{-fstrict-overflow} is in effect, because
3794 @code{abs (INT_MIN)} overflows to @code{INT_MIN}, which is less than
3795 zero. @option{-Wstrict-overflow} (with no level) is the same as
3796 @option{-Wstrict-overflow=2}.
3798 @item -Wstrict-overflow=3
3799 Also warn about other cases where a comparison is simplified. For
3800 example: @code{x + 1 > 1} will be simplified to @code{x > 0}.
3802 @item -Wstrict-overflow=4
3803 Also warn about other simplifications not covered by the above cases.
3804 For example: @code{(x * 10) / 5} will be simplified to @code{x * 2}.
3806 @item -Wstrict-overflow=5
3807 Also warn about cases where the compiler reduces the magnitude of a
3808 constant involved in a comparison. For example: @code{x + 2 > y} will
3809 be simplified to @code{x + 1 >= y}. This is reported only at the
3810 highest warning level because this simplification applies to many
3811 comparisons, so this warning level will give a very large number of
3815 @item -Wsuggest-attribute=@r{[}pure@r{|}const@r{|}noreturn@r{]}
3816 @opindex Wsuggest-attribute=
3817 @opindex Wno-suggest-attribute=
3818 Warn for cases where adding an attribute may be beneficial. The
3819 attributes currently supported are listed below.
3822 @item -Wsuggest-attribute=pure
3823 @itemx -Wsuggest-attribute=const
3824 @itemx -Wsuggest-attribute=noreturn
3825 @opindex Wsuggest-attribute=pure
3826 @opindex Wno-suggest-attribute=pure
3827 @opindex Wsuggest-attribute=const
3828 @opindex Wno-suggest-attribute=const
3829 @opindex Wsuggest-attribute=noreturn
3830 @opindex Wno-suggest-attribute=noreturn
3832 Warn about functions that might be candidates for attributes
3833 @code{pure}, @code{const} or @code{noreturn}. The compiler only warns for
3834 functions visible in other compilation units or (in the case of @code{pure} and
3835 @code{const}) if it cannot prove that the function returns normally. A function
3836 returns normally if it doesn't contain an infinite loop nor returns abnormally
3837 by throwing, calling @code{abort()} or trapping. This analysis requires option
3838 @option{-fipa-pure-const}, which is enabled by default at @option{-O} and
3839 higher. Higher optimization levels improve the accuracy of the analysis.
3842 @item -Warray-bounds
3843 @opindex Wno-array-bounds
3844 @opindex Warray-bounds
3845 This option is only active when @option{-ftree-vrp} is active
3846 (default for @option{-O2} and above). It warns about subscripts to arrays
3847 that are always out of bounds. This warning is enabled by @option{-Wall}.
3849 @item -Wno-div-by-zero
3850 @opindex Wno-div-by-zero
3851 @opindex Wdiv-by-zero
3852 Do not warn about compile-time integer division by zero. Floating-point
3853 division by zero is not warned about, as it can be a legitimate way of
3854 obtaining infinities and NaNs.
3856 @item -Wsystem-headers
3857 @opindex Wsystem-headers
3858 @opindex Wno-system-headers
3859 @cindex warnings from system headers
3860 @cindex system headers, warnings from
3861 Print warning messages for constructs found in system header files.
3862 Warnings from system headers are normally suppressed, on the assumption
3863 that they usually do not indicate real problems and would only make the
3864 compiler output harder to read. Using this command-line option tells
3865 GCC to emit warnings from system headers as if they occurred in user
3866 code. However, note that using @option{-Wall} in conjunction with this
3867 option will @emph{not} warn about unknown pragmas in system
3868 headers---for that, @option{-Wunknown-pragmas} must also be used.
3871 @opindex Wtrampolines
3872 @opindex Wno-trampolines
3873 Warn about trampolines generated for pointers to nested functions.
3875 A trampoline is a small piece of data or code that is created at run
3876 time on the stack when the address of a nested function is taken, and
3877 is used to call the nested function indirectly. For some targets, it
3878 is made up of data only and thus requires no special treatment. But,
3879 for most targets, it is made up of code and thus requires the stack
3880 to be made executable in order for the program to work properly.
3883 @opindex Wfloat-equal
3884 @opindex Wno-float-equal
3885 Warn if floating-point values are used in equality comparisons.
3887 The idea behind this is that sometimes it is convenient (for the
3888 programmer) to consider floating-point values as approximations to
3889 infinitely precise real numbers. If you are doing this, then you need
3890 to compute (by analyzing the code, or in some other way) the maximum or
3891 likely maximum error that the computation introduces, and allow for it
3892 when performing comparisons (and when producing output, but that's a
3893 different problem). In particular, instead of testing for equality, you
3894 would check to see whether the two values have ranges that overlap; and
3895 this is done with the relational operators, so equality comparisons are
3898 @item -Wtraditional @r{(C and Objective-C only)}
3899 @opindex Wtraditional
3900 @opindex Wno-traditional
3901 Warn about certain constructs that behave differently in traditional and
3902 ISO C@. Also warn about ISO C constructs that have no traditional C
3903 equivalent, and/or problematic constructs that should be avoided.
3907 Macro parameters that appear within string literals in the macro body.
3908 In traditional C macro replacement takes place within string literals,
3909 but does not in ISO C@.
3912 In traditional C, some preprocessor directives did not exist.
3913 Traditional preprocessors would only consider a line to be a directive
3914 if the @samp{#} appeared in column 1 on the line. Therefore
3915 @option{-Wtraditional} warns about directives that traditional C
3916 understands but would ignore because the @samp{#} does not appear as the
3917 first character on the line. It also suggests you hide directives like
3918 @samp{#pragma} not understood by traditional C by indenting them. Some
3919 traditional implementations would not recognize @samp{#elif}, so it
3920 suggests avoiding it altogether.
3923 A function-like macro that appears without arguments.
3926 The unary plus operator.
3929 The @samp{U} integer constant suffix, or the @samp{F} or @samp{L} floating-point
3930 constant suffixes. (Traditional C does support the @samp{L} suffix on integer
3931 constants.) Note, these suffixes appear in macros defined in the system
3932 headers of most modern systems, e.g.@: the @samp{_MIN}/@samp{_MAX} macros in @code{<limits.h>}.
3933 Use of these macros in user code might normally lead to spurious
3934 warnings, however GCC's integrated preprocessor has enough context to
3935 avoid warning in these cases.
3938 A function declared external in one block and then used after the end of
3942 A @code{switch} statement has an operand of type @code{long}.
3945 A non-@code{static} function declaration follows a @code{static} one.
3946 This construct is not accepted by some traditional C compilers.
3949 The ISO type of an integer constant has a different width or
3950 signedness from its traditional type. This warning is only issued if
3951 the base of the constant is ten. I.e.@: hexadecimal or octal values, which
3952 typically represent bit patterns, are not warned about.
3955 Usage of ISO string concatenation is detected.
3958 Initialization of automatic aggregates.
3961 Identifier conflicts with labels. Traditional C lacks a separate
3962 namespace for labels.
3965 Initialization of unions. If the initializer is zero, the warning is
3966 omitted. This is done under the assumption that the zero initializer in
3967 user code appears conditioned on e.g.@: @code{__STDC__} to avoid missing
3968 initializer warnings and relies on default initialization to zero in the
3972 Conversions by prototypes between fixed/floating-point values and vice
3973 versa. The absence of these prototypes when compiling with traditional
3974 C would cause serious problems. This is a subset of the possible
3975 conversion warnings, for the full set use @option{-Wtraditional-conversion}.
3978 Use of ISO C style function definitions. This warning intentionally is
3979 @emph{not} issued for prototype declarations or variadic functions
3980 because these ISO C features will appear in your code when using
3981 libiberty's traditional C compatibility macros, @code{PARAMS} and
3982 @code{VPARAMS}. This warning is also bypassed for nested functions
3983 because that feature is already a GCC extension and thus not relevant to
3984 traditional C compatibility.
3987 @item -Wtraditional-conversion @r{(C and Objective-C only)}
3988 @opindex Wtraditional-conversion
3989 @opindex Wno-traditional-conversion
3990 Warn if a prototype causes a type conversion that is different from what
3991 would happen to the same argument in the absence of a prototype. This
3992 includes conversions of fixed point to floating and vice versa, and
3993 conversions changing the width or signedness of a fixed-point argument
3994 except when the same as the default promotion.
3996 @item -Wdeclaration-after-statement @r{(C and Objective-C only)}
3997 @opindex Wdeclaration-after-statement
3998 @opindex Wno-declaration-after-statement
3999 Warn when a declaration is found after a statement in a block. This
4000 construct, known from C++, was introduced with ISO C99 and is by default
4001 allowed in GCC@. It is not supported by ISO C90 and was not supported by
4002 GCC versions before GCC 3.0. @xref{Mixed Declarations}.
4007 Warn if an undefined identifier is evaluated in an @samp{#if} directive.
4009 @item -Wno-endif-labels
4010 @opindex Wno-endif-labels
4011 @opindex Wendif-labels
4012 Do not warn whenever an @samp{#else} or an @samp{#endif} are followed by text.
4017 Warn whenever a local variable or type declaration shadows another variable,
4018 parameter, type, or class member (in C++), or whenever a built-in function
4019 is shadowed. Note that in C++, the compiler will not warn if a local variable
4020 shadows a struct/class/enum, but will warn if it shadows an explicit typedef.
4022 @item -Wlarger-than=@var{len}
4023 @opindex Wlarger-than=@var{len}
4024 @opindex Wlarger-than-@var{len}
4025 Warn whenever an object of larger than @var{len} bytes is defined.
4027 @item -Wframe-larger-than=@var{len}
4028 @opindex Wframe-larger-than
4029 Warn if the size of a function frame is larger than @var{len} bytes.
4030 The computation done to determine the stack frame size is approximate
4031 and not conservative.
4032 The actual requirements may be somewhat greater than @var{len}
4033 even if you do not get a warning. In addition, any space allocated
4034 via @code{alloca}, variable-length arrays, or related constructs
4035 is not included by the compiler when determining
4036 whether or not to issue a warning.
4038 @item -Wno-free-nonheap-object
4039 @opindex Wno-free-nonheap-object
4040 @opindex Wfree-nonheap-object
4041 Do not warn when attempting to free an object that was not allocated
4044 @item -Wstack-usage=@var{len}
4045 @opindex Wstack-usage
4046 Warn if the stack usage of a function might be larger than @var{len} bytes.
4047 The computation done to determine the stack usage is conservative.
4048 Any space allocated via @code{alloca}, variable-length arrays, or related
4049 constructs is included by the compiler when determining whether or not to
4052 The message is in keeping with the output of @option{-fstack-usage}.
4056 If the stack usage is fully static but exceeds the specified amount, it's:
4059 warning: stack usage is 1120 bytes
4062 If the stack usage is (partly) dynamic but bounded, it's:
4065 warning: stack usage might be 1648 bytes
4068 If the stack usage is (partly) dynamic and not bounded, it's:
4071 warning: stack usage might be unbounded
4075 @item -Wunsafe-loop-optimizations
4076 @opindex Wunsafe-loop-optimizations
4077 @opindex Wno-unsafe-loop-optimizations
4078 Warn if the loop cannot be optimized because the compiler could not
4079 assume anything on the bounds of the loop indices. With
4080 @option{-funsafe-loop-optimizations} warn if the compiler made
4083 @item -Wno-pedantic-ms-format @r{(MinGW targets only)}
4084 @opindex Wno-pedantic-ms-format
4085 @opindex Wpedantic-ms-format
4086 Disables the warnings about non-ISO @code{printf} / @code{scanf} format
4087 width specifiers @code{I32}, @code{I64}, and @code{I} used on Windows targets
4088 depending on the MS runtime, when you are using the options @option{-Wformat}
4089 and @option{-pedantic} without gnu-extensions.
4091 @item -Wpointer-arith
4092 @opindex Wpointer-arith
4093 @opindex Wno-pointer-arith
4094 Warn about anything that depends on the ``size of'' a function type or
4095 of @code{void}. GNU C assigns these types a size of 1, for
4096 convenience in calculations with @code{void *} pointers and pointers
4097 to functions. In C++, warn also when an arithmetic operation involves
4098 @code{NULL}. This warning is also enabled by @option{-pedantic}.
4101 @opindex Wtype-limits
4102 @opindex Wno-type-limits
4103 Warn if a comparison is always true or always false due to the limited
4104 range of the data type, but do not warn for constant expressions. For
4105 example, warn if an unsigned variable is compared against zero with
4106 @samp{<} or @samp{>=}. This warning is also enabled by
4109 @item -Wbad-function-cast @r{(C and Objective-C only)}
4110 @opindex Wbad-function-cast
4111 @opindex Wno-bad-function-cast
4112 Warn whenever a function call is cast to a non-matching type.
4113 For example, warn if @code{int malloc()} is cast to @code{anything *}.
4115 @item -Wc++-compat @r{(C and Objective-C only)}
4116 Warn about ISO C constructs that are outside of the common subset of
4117 ISO C and ISO C++, e.g.@: request for implicit conversion from
4118 @code{void *} to a pointer to non-@code{void} type.
4120 @item -Wc++11-compat @r{(C++ and Objective-C++ only)}
4121 Warn about C++ constructs whose meaning differs between ISO C++ 1998
4122 and ISO C++ 2011, e.g., identifiers in ISO C++ 1998 that are keywords
4123 in ISO C++ 2011. This warning turns on @option{-Wnarrowing} and is
4124 enabled by @option{-Wall}.
4128 @opindex Wno-cast-qual
4129 Warn whenever a pointer is cast so as to remove a type qualifier from
4130 the target type. For example, warn if a @code{const char *} is cast
4131 to an ordinary @code{char *}.
4133 Also warn when making a cast that introduces a type qualifier in an
4134 unsafe way. For example, casting @code{char **} to @code{const char **}
4135 is unsafe, as in this example:
4138 /* p is char ** value. */
4139 const char **q = (const char **) p;
4140 /* Assignment of readonly string to const char * is OK. */
4142 /* Now char** pointer points to read-only memory. */
4147 @opindex Wcast-align
4148 @opindex Wno-cast-align
4149 Warn whenever a pointer is cast such that the required alignment of the
4150 target is increased. For example, warn if a @code{char *} is cast to
4151 an @code{int *} on machines where integers can only be accessed at
4152 two- or four-byte boundaries.
4154 @item -Wwrite-strings
4155 @opindex Wwrite-strings
4156 @opindex Wno-write-strings
4157 When compiling C, give string constants the type @code{const
4158 char[@var{length}]} so that copying the address of one into a
4159 non-@code{const} @code{char *} pointer will get a warning. These
4160 warnings will help you find at compile time code that can try to write
4161 into a string constant, but only if you have been very careful about
4162 using @code{const} in declarations and prototypes. Otherwise, it will
4163 just be a nuisance. This is why we did not make @option{-Wall} request
4166 When compiling C++, warn about the deprecated conversion from string
4167 literals to @code{char *}. This warning is enabled by default for C++
4172 @opindex Wno-clobbered
4173 Warn for variables that might be changed by @samp{longjmp} or
4174 @samp{vfork}. This warning is also enabled by @option{-Wextra}.
4177 @opindex Wconversion
4178 @opindex Wno-conversion
4179 Warn for implicit conversions that may alter a value. This includes
4180 conversions between real and integer, like @code{abs (x)} when
4181 @code{x} is @code{double}; conversions between signed and unsigned,
4182 like @code{unsigned ui = -1}; and conversions to smaller types, like
4183 @code{sqrtf (M_PI)}. Do not warn for explicit casts like @code{abs
4184 ((int) x)} and @code{ui = (unsigned) -1}, or if the value is not
4185 changed by the conversion like in @code{abs (2.0)}. Warnings about
4186 conversions between signed and unsigned integers can be disabled by
4187 using @option{-Wno-sign-conversion}.
4189 For C++, also warn for confusing overload resolution for user-defined
4190 conversions; and conversions that will never use a type conversion
4191 operator: conversions to @code{void}, the same type, a base class or a
4192 reference to them. Warnings about conversions between signed and
4193 unsigned integers are disabled by default in C++ unless
4194 @option{-Wsign-conversion} is explicitly enabled.
4196 @item -Wno-conversion-null @r{(C++ and Objective-C++ only)}
4197 @opindex Wconversion-null
4198 @opindex Wno-conversion-null
4199 Do not warn for conversions between @code{NULL} and non-pointer
4200 types. @option{-Wconversion-null} is enabled by default.
4202 @item -Wzero-as-null-pointer-constant @r{(C++ and Objective-C++ only)}
4203 @opindex Wzero-as-null-pointer-constant
4204 @opindex Wno-zero-as-null-pointer-constant
4205 Warn when a literal '0' is used as null pointer constant. This can
4206 be useful to facilitate the conversion to @code{nullptr} in C++11.
4209 @opindex Wempty-body
4210 @opindex Wno-empty-body
4211 Warn if an empty body occurs in an @samp{if}, @samp{else} or @samp{do
4212 while} statement. This warning is also enabled by @option{-Wextra}.
4214 @item -Wenum-compare
4215 @opindex Wenum-compare
4216 @opindex Wno-enum-compare
4217 Warn about a comparison between values of different enumerated types.
4218 In C++ enumeral mismatches in conditional expressions are also
4219 diagnosed and the warning is enabled by default. In C this warning is
4220 enabled by @option{-Wall}.
4222 @item -Wjump-misses-init @r{(C, Objective-C only)}
4223 @opindex Wjump-misses-init
4224 @opindex Wno-jump-misses-init
4225 Warn if a @code{goto} statement or a @code{switch} statement jumps
4226 forward across the initialization of a variable, or jumps backward to a
4227 label after the variable has been initialized. This only warns about
4228 variables that are initialized when they are declared. This warning is
4229 only supported for C and Objective-C; in C++ this sort of branch is an
4232 @option{-Wjump-misses-init} is included in @option{-Wc++-compat}. It
4233 can be disabled with the @option{-Wno-jump-misses-init} option.
4235 @item -Wsign-compare
4236 @opindex Wsign-compare
4237 @opindex Wno-sign-compare
4238 @cindex warning for comparison of signed and unsigned values
4239 @cindex comparison of signed and unsigned values, warning
4240 @cindex signed and unsigned values, comparison warning
4241 Warn when a comparison between signed and unsigned values could produce
4242 an incorrect result when the signed value is converted to unsigned.
4243 This warning is also enabled by @option{-Wextra}; to get the other warnings
4244 of @option{-Wextra} without this warning, use @samp{-Wextra -Wno-sign-compare}.
4246 @item -Wsign-conversion
4247 @opindex Wsign-conversion
4248 @opindex Wno-sign-conversion
4249 Warn for implicit conversions that may change the sign of an integer
4250 value, like assigning a signed integer expression to an unsigned
4251 integer variable. An explicit cast silences the warning. In C, this
4252 option is enabled also by @option{-Wconversion}.
4256 @opindex Wno-address
4257 Warn about suspicious uses of memory addresses. These include using
4258 the address of a function in a conditional expression, such as
4259 @code{void func(void); if (func)}, and comparisons against the memory
4260 address of a string literal, such as @code{if (x == "abc")}. Such
4261 uses typically indicate a programmer error: the address of a function
4262 always evaluates to true, so their use in a conditional usually
4263 indicate that the programmer forgot the parentheses in a function
4264 call; and comparisons against string literals result in unspecified
4265 behavior and are not portable in C, so they usually indicate that the
4266 programmer intended to use @code{strcmp}. This warning is enabled by
4270 @opindex Wlogical-op
4271 @opindex Wno-logical-op
4272 Warn about suspicious uses of logical operators in expressions.
4273 This includes using logical operators in contexts where a
4274 bit-wise operator is likely to be expected.
4276 @item -Waggregate-return
4277 @opindex Waggregate-return
4278 @opindex Wno-aggregate-return
4279 Warn if any functions that return structures or unions are defined or
4280 called. (In languages where you can return an array, this also elicits
4283 @item -Wno-attributes
4284 @opindex Wno-attributes
4285 @opindex Wattributes
4286 Do not warn if an unexpected @code{__attribute__} is used, such as
4287 unrecognized attributes, function attributes applied to variables,
4288 etc. This will not stop errors for incorrect use of supported
4291 @item -Wno-builtin-macro-redefined
4292 @opindex Wno-builtin-macro-redefined
4293 @opindex Wbuiltin-macro-redefined
4294 Do not warn if certain built-in macros are redefined. This suppresses
4295 warnings for redefinition of @code{__TIMESTAMP__}, @code{__TIME__},
4296 @code{__DATE__}, @code{__FILE__}, and @code{__BASE_FILE__}.
4298 @item -Wstrict-prototypes @r{(C and Objective-C only)}
4299 @opindex Wstrict-prototypes
4300 @opindex Wno-strict-prototypes
4301 Warn if a function is declared or defined without specifying the
4302 argument types. (An old-style function definition is permitted without
4303 a warning if preceded by a declaration that specifies the argument
4306 @item -Wold-style-declaration @r{(C and Objective-C only)}
4307 @opindex Wold-style-declaration
4308 @opindex Wno-old-style-declaration
4309 Warn for obsolescent usages, according to the C Standard, in a
4310 declaration. For example, warn if storage-class specifiers like
4311 @code{static} are not the first things in a declaration. This warning
4312 is also enabled by @option{-Wextra}.
4314 @item -Wold-style-definition @r{(C and Objective-C only)}
4315 @opindex Wold-style-definition
4316 @opindex Wno-old-style-definition
4317 Warn if an old-style function definition is used. A warning is given
4318 even if there is a previous prototype.
4320 @item -Wmissing-parameter-type @r{(C and Objective-C only)}
4321 @opindex Wmissing-parameter-type
4322 @opindex Wno-missing-parameter-type
4323 A function parameter is declared without a type specifier in K&R-style
4330 This warning is also enabled by @option{-Wextra}.
4332 @item -Wmissing-prototypes @r{(C and Objective-C only)}
4333 @opindex Wmissing-prototypes
4334 @opindex Wno-missing-prototypes
4335 Warn if a global function is defined without a previous prototype
4336 declaration. This warning is issued even if the definition itself
4337 provides a prototype. The aim is to detect global functions that
4338 are not declared in header files.
4340 @item -Wmissing-declarations
4341 @opindex Wmissing-declarations
4342 @opindex Wno-missing-declarations
4343 Warn if a global function is defined without a previous declaration.
4344 Do so even if the definition itself provides a prototype.
4345 Use this option to detect global functions that are not declared in
4346 header files. In C++, no warnings are issued for function templates,
4347 or for inline functions, or for functions in anonymous namespaces.
4349 @item -Wmissing-field-initializers
4350 @opindex Wmissing-field-initializers
4351 @opindex Wno-missing-field-initializers
4355 Warn if a structure's initializer has some fields missing. For
4356 example, the following code would cause such a warning, because
4357 @code{x.h} is implicitly zero:
4360 struct s @{ int f, g, h; @};
4361 struct s x = @{ 3, 4 @};
4364 This option does not warn about designated initializers, so the following
4365 modification would not trigger a warning:
4368 struct s @{ int f, g, h; @};
4369 struct s x = @{ .f = 3, .g = 4 @};
4372 This warning is included in @option{-Wextra}. To get other @option{-Wextra}
4373 warnings without this one, use @samp{-Wextra -Wno-missing-field-initializers}.
4375 @item -Wmissing-format-attribute
4376 @opindex Wmissing-format-attribute
4377 @opindex Wno-missing-format-attribute
4380 Warn about function pointers that might be candidates for @code{format}
4381 attributes. Note these are only possible candidates, not absolute ones.
4382 GCC will guess that function pointers with @code{format} attributes that
4383 are used in assignment, initialization, parameter passing or return
4384 statements should have a corresponding @code{format} attribute in the
4385 resulting type. I.e.@: the left-hand side of the assignment or
4386 initialization, the type of the parameter variable, or the return type
4387 of the containing function respectively should also have a @code{format}
4388 attribute to avoid the warning.
4390 GCC will also warn about function definitions that might be
4391 candidates for @code{format} attributes. Again, these are only
4392 possible candidates. GCC will guess that @code{format} attributes
4393 might be appropriate for any function that calls a function like
4394 @code{vprintf} or @code{vscanf}, but this might not always be the
4395 case, and some functions for which @code{format} attributes are
4396 appropriate may not be detected.
4398 @item -Wno-multichar
4399 @opindex Wno-multichar
4401 Do not warn if a multicharacter constant (@samp{'FOOF'}) is used.
4402 Usually they indicate a typo in the user's code, as they have
4403 implementation-defined values, and should not be used in portable code.
4405 @item -Wnormalized=<none|id|nfc|nfkc>
4406 @opindex Wnormalized=
4409 @cindex character set, input normalization
4410 In ISO C and ISO C++, two identifiers are different if they are
4411 different sequences of characters. However, sometimes when characters
4412 outside the basic ASCII character set are used, you can have two
4413 different character sequences that look the same. To avoid confusion,
4414 the ISO 10646 standard sets out some @dfn{normalization rules} which
4415 when applied ensure that two sequences that look the same are turned into
4416 the same sequence. GCC can warn you if you are using identifiers that
4417 have not been normalized; this option controls that warning.
4419 There are four levels of warning supported by GCC. The default is
4420 @option{-Wnormalized=nfc}, which warns about any identifier that is
4421 not in the ISO 10646 ``C'' normalized form, @dfn{NFC}. NFC is the
4422 recommended form for most uses.
4424 Unfortunately, there are some characters allowed in identifiers by
4425 ISO C and ISO C++ that, when turned into NFC, are not allowed in
4426 identifiers. That is, there's no way to use these symbols in portable
4427 ISO C or C++ and have all your identifiers in NFC@.
4428 @option{-Wnormalized=id} suppresses the warning for these characters.
4429 It is hoped that future versions of the standards involved will correct
4430 this, which is why this option is not the default.
4432 You can switch the warning off for all characters by writing
4433 @option{-Wnormalized=none}. You would only want to do this if you
4434 were using some other normalization scheme (like ``D''), because
4435 otherwise you can easily create bugs that are literally impossible to see.
4437 Some characters in ISO 10646 have distinct meanings but look identical
4438 in some fonts or display methodologies, especially once formatting has
4439 been applied. For instance @code{\u207F}, ``SUPERSCRIPT LATIN SMALL
4440 LETTER N'', will display just like a regular @code{n} that has been
4441 placed in a superscript. ISO 10646 defines the @dfn{NFKC}
4442 normalization scheme to convert all these into a standard form as
4443 well, and GCC will warn if your code is not in NFKC if you use
4444 @option{-Wnormalized=nfkc}. This warning is comparable to warning
4445 about every identifier that contains the letter O because it might be
4446 confused with the digit 0, and so is not the default, but may be
4447 useful as a local coding convention if the programming environment is
4448 unable to be fixed to display these characters distinctly.
4450 @item -Wno-deprecated
4451 @opindex Wno-deprecated
4452 @opindex Wdeprecated
4453 Do not warn about usage of deprecated features. @xref{Deprecated Features}.
4455 @item -Wno-deprecated-declarations
4456 @opindex Wno-deprecated-declarations
4457 @opindex Wdeprecated-declarations
4458 Do not warn about uses of functions (@pxref{Function Attributes}),
4459 variables (@pxref{Variable Attributes}), and types (@pxref{Type
4460 Attributes}) marked as deprecated by using the @code{deprecated}
4464 @opindex Wno-overflow
4466 Do not warn about compile-time overflow in constant expressions.
4468 @item -Woverride-init @r{(C and Objective-C only)}
4469 @opindex Woverride-init
4470 @opindex Wno-override-init
4474 Warn if an initialized field without side effects is overridden when
4475 using designated initializers (@pxref{Designated Inits, , Designated
4478 This warning is included in @option{-Wextra}. To get other
4479 @option{-Wextra} warnings without this one, use @samp{-Wextra
4480 -Wno-override-init}.
4485 Warn if a structure is given the packed attribute, but the packed
4486 attribute has no effect on the layout or size of the structure.
4487 Such structures may be mis-aligned for little benefit. For
4488 instance, in this code, the variable @code{f.x} in @code{struct bar}
4489 will be misaligned even though @code{struct bar} does not itself
4490 have the packed attribute:
4497 @} __attribute__((packed));
4505 @item -Wpacked-bitfield-compat
4506 @opindex Wpacked-bitfield-compat
4507 @opindex Wno-packed-bitfield-compat
4508 The 4.1, 4.2 and 4.3 series of GCC ignore the @code{packed} attribute
4509 on bit-fields of type @code{char}. This has been fixed in GCC 4.4 but
4510 the change can lead to differences in the structure layout. GCC
4511 informs you when the offset of such a field has changed in GCC 4.4.
4512 For example there is no longer a 4-bit padding between field @code{a}
4513 and @code{b} in this structure:
4520 @} __attribute__ ((packed));
4523 This warning is enabled by default. Use
4524 @option{-Wno-packed-bitfield-compat} to disable this warning.
4529 Warn if padding is included in a structure, either to align an element
4530 of the structure or to align the whole structure. Sometimes when this
4531 happens it is possible to rearrange the fields of the structure to
4532 reduce the padding and so make the structure smaller.
4534 @item -Wredundant-decls
4535 @opindex Wredundant-decls
4536 @opindex Wno-redundant-decls
4537 Warn if anything is declared more than once in the same scope, even in
4538 cases where multiple declaration is valid and changes nothing.
4540 @item -Wnested-externs @r{(C and Objective-C only)}
4541 @opindex Wnested-externs
4542 @opindex Wno-nested-externs
4543 Warn if an @code{extern} declaration is encountered within a function.
4548 Warn if a function can not be inlined and it was declared as inline.
4549 Even with this option, the compiler will not warn about failures to
4550 inline functions declared in system headers.
4552 The compiler uses a variety of heuristics to determine whether or not
4553 to inline a function. For example, the compiler takes into account
4554 the size of the function being inlined and the amount of inlining
4555 that has already been done in the current function. Therefore,
4556 seemingly insignificant changes in the source program can cause the
4557 warnings produced by @option{-Winline} to appear or disappear.
4559 @item -Wno-invalid-offsetof @r{(C++ and Objective-C++ only)}
4560 @opindex Wno-invalid-offsetof
4561 @opindex Winvalid-offsetof
4562 Suppress warnings from applying the @samp{offsetof} macro to a non-POD
4563 type. According to the 1998 ISO C++ standard, applying @samp{offsetof}
4564 to a non-POD type is undefined. In existing C++ implementations,
4565 however, @samp{offsetof} typically gives meaningful results even when
4566 applied to certain kinds of non-POD types. (Such as a simple
4567 @samp{struct} that fails to be a POD type only by virtue of having a
4568 constructor.) This flag is for users who are aware that they are
4569 writing nonportable code and who have deliberately chosen to ignore the
4572 The restrictions on @samp{offsetof} may be relaxed in a future version
4573 of the C++ standard.
4575 @item -Wno-int-to-pointer-cast
4576 @opindex Wno-int-to-pointer-cast
4577 @opindex Wint-to-pointer-cast
4578 Suppress warnings from casts to pointer type of an integer of a
4579 different size. In C++, casting to a pointer type of smaller size is
4580 an error. @option{Wint-to-pointer-cast} is enabled by default.
4583 @item -Wno-pointer-to-int-cast @r{(C and Objective-C only)}
4584 @opindex Wno-pointer-to-int-cast
4585 @opindex Wpointer-to-int-cast
4586 Suppress warnings from casts from a pointer to an integer type of a
4590 @opindex Winvalid-pch
4591 @opindex Wno-invalid-pch
4592 Warn if a precompiled header (@pxref{Precompiled Headers}) is found in
4593 the search path but can't be used.
4597 @opindex Wno-long-long
4598 Warn if @samp{long long} type is used. This is enabled by either
4599 @option{-pedantic} or @option{-Wtraditional} in ISO C90 and C++98
4600 modes. To inhibit the warning messages, use @option{-Wno-long-long}.
4602 @item -Wvariadic-macros
4603 @opindex Wvariadic-macros
4604 @opindex Wno-variadic-macros
4605 Warn if variadic macros are used in pedantic ISO C90 mode, or the GNU
4606 alternate syntax when in pedantic ISO C99 mode. This is default.
4607 To inhibit the warning messages, use @option{-Wno-variadic-macros}.
4609 @item -Wvector-operation-performance
4610 @opindex Wvector-operation-performance
4611 @opindex Wno-vector-operation-performance
4612 Warn if vector operation is not implemented via SIMD capabilities of the
4613 architecture. Mainly useful for the performance tuning.
4614 Vector operation can be implemented @code{piecewise}, which means that the
4615 scalar operation is performed on every vector element;
4616 @code{in parallel}, which means that the vector operation is implemented
4617 using scalars of wider type, which normally is more performance efficient;
4618 and @code{as a single scalar}, which means that vector fits into a
4624 Warn if variable length array is used in the code.
4625 @option{-Wno-vla} will prevent the @option{-pedantic} warning of
4626 the variable length array.
4628 @item -Wvolatile-register-var
4629 @opindex Wvolatile-register-var
4630 @opindex Wno-volatile-register-var
4631 Warn if a register variable is declared volatile. The volatile
4632 modifier does not inhibit all optimizations that may eliminate reads
4633 and/or writes to register variables. This warning is enabled by
4636 @item -Wdisabled-optimization
4637 @opindex Wdisabled-optimization
4638 @opindex Wno-disabled-optimization
4639 Warn if a requested optimization pass is disabled. This warning does
4640 not generally indicate that there is anything wrong with your code; it
4641 merely indicates that GCC's optimizers were unable to handle the code
4642 effectively. Often, the problem is that your code is too big or too
4643 complex; GCC will refuse to optimize programs when the optimization
4644 itself is likely to take inordinate amounts of time.
4646 @item -Wpointer-sign @r{(C and Objective-C only)}
4647 @opindex Wpointer-sign
4648 @opindex Wno-pointer-sign
4649 Warn for pointer argument passing or assignment with different signedness.
4650 This option is only supported for C and Objective-C@. It is implied by
4651 @option{-Wall} and by @option{-pedantic}, which can be disabled with
4652 @option{-Wno-pointer-sign}.
4654 @item -Wstack-protector
4655 @opindex Wstack-protector
4656 @opindex Wno-stack-protector
4657 This option is only active when @option{-fstack-protector} is active. It
4658 warns about functions that will not be protected against stack smashing.
4661 @opindex Wno-mudflap
4662 Suppress warnings about constructs that cannot be instrumented by
4665 @item -Woverlength-strings
4666 @opindex Woverlength-strings
4667 @opindex Wno-overlength-strings
4668 Warn about string constants that are longer than the ``minimum
4669 maximum'' length specified in the C standard. Modern compilers
4670 generally allow string constants that are much longer than the
4671 standard's minimum limit, but very portable programs should avoid
4672 using longer strings.
4674 The limit applies @emph{after} string constant concatenation, and does
4675 not count the trailing NUL@. In C90, the limit was 509 characters; in
4676 C99, it was raised to 4095. C++98 does not specify a normative
4677 minimum maximum, so we do not diagnose overlength strings in C++@.
4679 This option is implied by @option{-pedantic}, and can be disabled with
4680 @option{-Wno-overlength-strings}.
4682 @item -Wunsuffixed-float-constants @r{(C and Objective-C only)}
4683 @opindex Wunsuffixed-float-constants
4685 GCC will issue a warning for any floating constant that does not have
4686 a suffix. When used together with @option{-Wsystem-headers} it will
4687 warn about such constants in system header files. This can be useful
4688 when preparing code to use with the @code{FLOAT_CONST_DECIMAL64} pragma
4689 from the decimal floating-point extension to C99.
4692 @node Debugging Options
4693 @section Options for Debugging Your Program or GCC
4694 @cindex options, debugging
4695 @cindex debugging information options
4697 GCC has various special options that are used for debugging
4698 either your program or GCC:
4703 Produce debugging information in the operating system's native format
4704 (stabs, COFF, XCOFF, or DWARF 2)@. GDB can work with this debugging
4707 On most systems that use stabs format, @option{-g} enables use of extra
4708 debugging information that only GDB can use; this extra information
4709 makes debugging work better in GDB but will probably make other debuggers
4711 refuse to read the program. If you want to control for certain whether
4712 to generate the extra information, use @option{-gstabs+}, @option{-gstabs},
4713 @option{-gxcoff+}, @option{-gxcoff}, or @option{-gvms} (see below).
4715 GCC allows you to use @option{-g} with
4716 @option{-O}. The shortcuts taken by optimized code may occasionally
4717 produce surprising results: some variables you declared may not exist
4718 at all; flow of control may briefly move where you did not expect it;
4719 some statements may not be executed because they compute constant
4720 results or their values were already at hand; some statements may
4721 execute in different places because they were moved out of loops.
4723 Nevertheless it proves possible to debug optimized output. This makes
4724 it reasonable to use the optimizer for programs that might have bugs.
4726 The following options are useful when GCC is generated with the
4727 capability for more than one debugging format.
4731 Produce debugging information for use by GDB@. This means to use the
4732 most expressive format available (DWARF 2, stabs, or the native format
4733 if neither of those are supported), including GDB extensions if at all
4738 Produce debugging information in stabs format (if that is supported),
4739 without GDB extensions. This is the format used by DBX on most BSD
4740 systems. On MIPS, Alpha and System V Release 4 systems this option
4741 produces stabs debugging output that is not understood by DBX or SDB@.
4742 On System V Release 4 systems this option requires the GNU assembler.
4744 @item -feliminate-unused-debug-symbols
4745 @opindex feliminate-unused-debug-symbols
4746 Produce debugging information in stabs format (if that is supported),
4747 for only symbols that are actually used.
4749 @item -femit-class-debug-always
4750 Instead of emitting debugging information for a C++ class in only one
4751 object file, emit it in all object files using the class. This option
4752 should be used only with debuggers that are unable to handle the way GCC
4753 normally emits debugging information for classes because using this
4754 option will increase the size of debugging information by as much as a
4757 @item -fno-debug-types-section
4758 @opindex fno-debug-types-section
4759 @opindex fdebug-types-section
4760 By default when using DWARF v4 or higher type DIEs will be put into
4761 their own .debug_types section instead of making them part of the
4762 .debug_info section. It is more efficient to put them in a separate
4763 comdat sections since the linker will then be able to remove duplicates.
4764 But not all DWARF consumers support .debug_types sections yet.
4768 Produce debugging information in stabs format (if that is supported),
4769 using GNU extensions understood only by the GNU debugger (GDB)@. The
4770 use of these extensions is likely to make other debuggers crash or
4771 refuse to read the program.
4775 Produce debugging information in COFF format (if that is supported).
4776 This is the format used by SDB on most System V systems prior to
4781 Produce debugging information in XCOFF format (if that is supported).
4782 This is the format used by the DBX debugger on IBM RS/6000 systems.
4786 Produce debugging information in XCOFF format (if that is supported),
4787 using GNU extensions understood only by the GNU debugger (GDB)@. The
4788 use of these extensions is likely to make other debuggers crash or
4789 refuse to read the program, and may cause assemblers other than the GNU
4790 assembler (GAS) to fail with an error.
4792 @item -gdwarf-@var{version}
4793 @opindex gdwarf-@var{version}
4794 Produce debugging information in DWARF format (if that is
4795 supported). This is the format used by DBX on IRIX 6. The value
4796 of @var{version} may be either 2, 3 or 4; the default version is 2.
4798 Note that with DWARF version 2 some ports require, and will always
4799 use, some non-conflicting DWARF 3 extensions in the unwind tables.
4801 Version 4 may require GDB 7.0 and @option{-fvar-tracking-assignments}
4802 for maximum benefit.
4804 @item -grecord-gcc-switches
4805 @opindex grecord-gcc-switches
4806 This switch causes the command-line options used to invoke the
4807 compiler that may affect code generation to be appended to the
4808 DW_AT_producer attribute in DWARF debugging information. The options
4809 are concatenated with spaces separating them from each other and from
4810 the compiler version. See also @option{-frecord-gcc-switches} for another
4811 way of storing compiler options into the object file.
4813 @item -gno-record-gcc-switches
4814 @opindex gno-record-gcc-switches
4815 Disallow appending command-line options to the DW_AT_producer attribute
4816 in DWARF debugging information. This is the default.
4818 @item -gstrict-dwarf
4819 @opindex gstrict-dwarf
4820 Disallow using extensions of later DWARF standard version than selected
4821 with @option{-gdwarf-@var{version}}. On most targets using non-conflicting
4822 DWARF extensions from later standard versions is allowed.
4824 @item -gno-strict-dwarf
4825 @opindex gno-strict-dwarf
4826 Allow using extensions of later DWARF standard version than selected with
4827 @option{-gdwarf-@var{version}}.
4831 Produce debugging information in VMS debug format (if that is
4832 supported). This is the format used by DEBUG on VMS systems.
4835 @itemx -ggdb@var{level}
4836 @itemx -gstabs@var{level}
4837 @itemx -gcoff@var{level}
4838 @itemx -gxcoff@var{level}
4839 @itemx -gvms@var{level}
4840 Request debugging information and also use @var{level} to specify how
4841 much information. The default level is 2.
4843 Level 0 produces no debug information at all. Thus, @option{-g0} negates
4846 Level 1 produces minimal information, enough for making backtraces in
4847 parts of the program that you don't plan to debug. This includes
4848 descriptions of functions and external variables, but no information
4849 about local variables and no line numbers.
4851 Level 3 includes extra information, such as all the macro definitions
4852 present in the program. Some debuggers support macro expansion when
4853 you use @option{-g3}.
4855 @option{-gdwarf-2} does not accept a concatenated debug level, because
4856 GCC used to support an option @option{-gdwarf} that meant to generate
4857 debug information in version 1 of the DWARF format (which is very
4858 different from version 2), and it would have been too confusing. That
4859 debug format is long obsolete, but the option cannot be changed now.
4860 Instead use an additional @option{-g@var{level}} option to change the
4861 debug level for DWARF.
4865 Turn off generation of debug info, if leaving out this option would have
4866 generated it, or turn it on at level 2 otherwise. The position of this
4867 argument in the command line does not matter, it takes effect after all
4868 other options are processed, and it does so only once, no matter how
4869 many times it is given. This is mainly intended to be used with
4870 @option{-fcompare-debug}.
4872 @item -fdump-final-insns@r{[}=@var{file}@r{]}
4873 @opindex fdump-final-insns
4874 Dump the final internal representation (RTL) to @var{file}. If the
4875 optional argument is omitted (or if @var{file} is @code{.}), the name
4876 of the dump file will be determined by appending @code{.gkd} to the
4877 compilation output file name.
4879 @item -fcompare-debug@r{[}=@var{opts}@r{]}
4880 @opindex fcompare-debug
4881 @opindex fno-compare-debug
4882 If no error occurs during compilation, run the compiler a second time,
4883 adding @var{opts} and @option{-fcompare-debug-second} to the arguments
4884 passed to the second compilation. Dump the final internal
4885 representation in both compilations, and print an error if they differ.
4887 If the equal sign is omitted, the default @option{-gtoggle} is used.
4889 The environment variable @env{GCC_COMPARE_DEBUG}, if defined, non-empty
4890 and nonzero, implicitly enables @option{-fcompare-debug}. If
4891 @env{GCC_COMPARE_DEBUG} is defined to a string starting with a dash,
4892 then it is used for @var{opts}, otherwise the default @option{-gtoggle}
4895 @option{-fcompare-debug=}, with the equal sign but without @var{opts},
4896 is equivalent to @option{-fno-compare-debug}, which disables the dumping
4897 of the final representation and the second compilation, preventing even
4898 @env{GCC_COMPARE_DEBUG} from taking effect.
4900 To verify full coverage during @option{-fcompare-debug} testing, set
4901 @env{GCC_COMPARE_DEBUG} to say @samp{-fcompare-debug-not-overridden},
4902 which GCC will reject as an invalid option in any actual compilation
4903 (rather than preprocessing, assembly or linking). To get just a
4904 warning, setting @env{GCC_COMPARE_DEBUG} to @samp{-w%n-fcompare-debug
4905 not overridden} will do.
4907 @item -fcompare-debug-second
4908 @opindex fcompare-debug-second
4909 This option is implicitly passed to the compiler for the second
4910 compilation requested by @option{-fcompare-debug}, along with options to
4911 silence warnings, and omitting other options that would cause
4912 side-effect compiler outputs to files or to the standard output. Dump
4913 files and preserved temporary files are renamed so as to contain the
4914 @code{.gk} additional extension during the second compilation, to avoid
4915 overwriting those generated by the first.
4917 When this option is passed to the compiler driver, it causes the
4918 @emph{first} compilation to be skipped, which makes it useful for little
4919 other than debugging the compiler proper.
4921 @item -feliminate-dwarf2-dups
4922 @opindex feliminate-dwarf2-dups
4923 Compress DWARF2 debugging information by eliminating duplicated
4924 information about each symbol. This option only makes sense when
4925 generating DWARF2 debugging information with @option{-gdwarf-2}.
4927 @item -femit-struct-debug-baseonly
4928 Emit debug information for struct-like types
4929 only when the base name of the compilation source file
4930 matches the base name of file in which the struct was defined.
4932 This option substantially reduces the size of debugging information,
4933 but at significant potential loss in type information to the debugger.
4934 See @option{-femit-struct-debug-reduced} for a less aggressive option.
4935 See @option{-femit-struct-debug-detailed} for more detailed control.
4937 This option works only with DWARF 2.
4939 @item -femit-struct-debug-reduced
4940 Emit debug information for struct-like types
4941 only when the base name of the compilation source file
4942 matches the base name of file in which the type was defined,
4943 unless the struct is a template or defined in a system header.
4945 This option significantly reduces the size of debugging information,
4946 with some potential loss in type information to the debugger.
4947 See @option{-femit-struct-debug-baseonly} for a more aggressive option.
4948 See @option{-femit-struct-debug-detailed} for more detailed control.
4950 This option works only with DWARF 2.
4952 @item -femit-struct-debug-detailed@r{[}=@var{spec-list}@r{]}
4953 Specify the struct-like types
4954 for which the compiler will generate debug information.
4955 The intent is to reduce duplicate struct debug information
4956 between different object files within the same program.
4958 This option is a detailed version of
4959 @option{-femit-struct-debug-reduced} and @option{-femit-struct-debug-baseonly},
4960 which will serve for most needs.
4962 A specification has the syntax@*
4963 [@samp{dir:}|@samp{ind:}][@samp{ord:}|@samp{gen:}](@samp{any}|@samp{sys}|@samp{base}|@samp{none})
4965 The optional first word limits the specification to
4966 structs that are used directly (@samp{dir:}) or used indirectly (@samp{ind:}).
4967 A struct type is used directly when it is the type of a variable, member.
4968 Indirect uses arise through pointers to structs.
4969 That is, when use of an incomplete struct would be legal, the use is indirect.
4971 @samp{struct one direct; struct two * indirect;}.
4973 The optional second word limits the specification to
4974 ordinary structs (@samp{ord:}) or generic structs (@samp{gen:}).
4975 Generic structs are a bit complicated to explain.
4976 For C++, these are non-explicit specializations of template classes,
4977 or non-template classes within the above.
4978 Other programming languages have generics,
4979 but @samp{-femit-struct-debug-detailed} does not yet implement them.
4981 The third word specifies the source files for those
4982 structs for which the compiler will emit debug information.
4983 The values @samp{none} and @samp{any} have the normal meaning.
4984 The value @samp{base} means that
4985 the base of name of the file in which the type declaration appears
4986 must match the base of the name of the main compilation file.
4987 In practice, this means that
4988 types declared in @file{foo.c} and @file{foo.h} will have debug information,
4989 but types declared in other header will not.
4990 The value @samp{sys} means those types satisfying @samp{base}
4991 or declared in system or compiler headers.
4993 You may need to experiment to determine the best settings for your application.
4995 The default is @samp{-femit-struct-debug-detailed=all}.
4997 This option works only with DWARF 2.
4999 @item -fno-merge-debug-strings
5000 @opindex fmerge-debug-strings
5001 @opindex fno-merge-debug-strings
5002 Direct the linker to not merge together strings in the debugging
5003 information that are identical in different object files. Merging is
5004 not supported by all assemblers or linkers. Merging decreases the size
5005 of the debug information in the output file at the cost of increasing
5006 link processing time. Merging is enabled by default.
5008 @item -fdebug-prefix-map=@var{old}=@var{new}
5009 @opindex fdebug-prefix-map
5010 When compiling files in directory @file{@var{old}}, record debugging
5011 information describing them as in @file{@var{new}} instead.
5013 @item -fno-dwarf2-cfi-asm
5014 @opindex fdwarf2-cfi-asm
5015 @opindex fno-dwarf2-cfi-asm
5016 Emit DWARF 2 unwind info as compiler generated @code{.eh_frame} section
5017 instead of using GAS @code{.cfi_*} directives.
5019 @cindex @command{prof}
5022 Generate extra code to write profile information suitable for the
5023 analysis program @command{prof}. You must use this option when compiling
5024 the source files you want data about, and you must also use it when
5027 @cindex @command{gprof}
5030 Generate extra code to write profile information suitable for the
5031 analysis program @command{gprof}. You must use this option when compiling
5032 the source files you want data about, and you must also use it when
5037 Makes the compiler print out each function name as it is compiled, and
5038 print some statistics about each pass when it finishes.
5041 @opindex ftime-report
5042 Makes the compiler print some statistics about the time consumed by each
5043 pass when it finishes.
5046 @opindex fmem-report
5047 Makes the compiler print some statistics about permanent memory
5048 allocation when it finishes.
5050 @item -fpre-ipa-mem-report
5051 @opindex fpre-ipa-mem-report
5052 @item -fpost-ipa-mem-report
5053 @opindex fpost-ipa-mem-report
5054 Makes the compiler print some statistics about permanent memory
5055 allocation before or after interprocedural optimization.
5058 @opindex fstack-usage
5059 Makes the compiler output stack usage information for the program, on a
5060 per-function basis. The filename for the dump is made by appending
5061 @file{.su} to the @var{auxname}. @var{auxname} is generated from the name of
5062 the output file, if explicitly specified and it is not an executable,
5063 otherwise it is the basename of the source file. An entry is made up
5068 The name of the function.
5072 One or more qualifiers: @code{static}, @code{dynamic}, @code{bounded}.
5075 The qualifier @code{static} means that the function manipulates the stack
5076 statically: a fixed number of bytes are allocated for the frame on function
5077 entry and released on function exit; no stack adjustments are otherwise made
5078 in the function. The second field is this fixed number of bytes.
5080 The qualifier @code{dynamic} means that the function manipulates the stack
5081 dynamically: in addition to the static allocation described above, stack
5082 adjustments are made in the body of the function, for example to push/pop
5083 arguments around function calls. If the qualifier @code{bounded} is also
5084 present, the amount of these adjustments is bounded at compile time and
5085 the second field is an upper bound of the total amount of stack used by
5086 the function. If it is not present, the amount of these adjustments is
5087 not bounded at compile time and the second field only represents the
5090 @item -fprofile-arcs
5091 @opindex fprofile-arcs
5092 Add code so that program flow @dfn{arcs} are instrumented. During
5093 execution the program records how many times each branch and call is
5094 executed and how many times it is taken or returns. When the compiled
5095 program exits it saves this data to a file called
5096 @file{@var{auxname}.gcda} for each source file. The data may be used for
5097 profile-directed optimizations (@option{-fbranch-probabilities}), or for
5098 test coverage analysis (@option{-ftest-coverage}). Each object file's
5099 @var{auxname} is generated from the name of the output file, if
5100 explicitly specified and it is not the final executable, otherwise it is
5101 the basename of the source file. In both cases any suffix is removed
5102 (e.g.@: @file{foo.gcda} for input file @file{dir/foo.c}, or
5103 @file{dir/foo.gcda} for output file specified as @option{-o dir/foo.o}).
5104 @xref{Cross-profiling}.
5106 @cindex @command{gcov}
5110 This option is used to compile and link code instrumented for coverage
5111 analysis. The option is a synonym for @option{-fprofile-arcs}
5112 @option{-ftest-coverage} (when compiling) and @option{-lgcov} (when
5113 linking). See the documentation for those options for more details.
5118 Compile the source files with @option{-fprofile-arcs} plus optimization
5119 and code generation options. For test coverage analysis, use the
5120 additional @option{-ftest-coverage} option. You do not need to profile
5121 every source file in a program.
5124 Link your object files with @option{-lgcov} or @option{-fprofile-arcs}
5125 (the latter implies the former).
5128 Run the program on a representative workload to generate the arc profile
5129 information. This may be repeated any number of times. You can run
5130 concurrent instances of your program, and provided that the file system
5131 supports locking, the data files will be correctly updated. Also
5132 @code{fork} calls are detected and correctly handled (double counting
5136 For profile-directed optimizations, compile the source files again with
5137 the same optimization and code generation options plus
5138 @option{-fbranch-probabilities} (@pxref{Optimize Options,,Options that
5139 Control Optimization}).
5142 For test coverage analysis, use @command{gcov} to produce human readable
5143 information from the @file{.gcno} and @file{.gcda} files. Refer to the
5144 @command{gcov} documentation for further information.
5148 With @option{-fprofile-arcs}, for each function of your program GCC
5149 creates a program flow graph, then finds a spanning tree for the graph.
5150 Only arcs that are not on the spanning tree have to be instrumented: the
5151 compiler adds code to count the number of times that these arcs are
5152 executed. When an arc is the only exit or only entrance to a block, the
5153 instrumentation code can be added to the block; otherwise, a new basic
5154 block must be created to hold the instrumentation code.
5157 @item -ftest-coverage
5158 @opindex ftest-coverage
5159 Produce a notes file that the @command{gcov} code-coverage utility
5160 (@pxref{Gcov,, @command{gcov}---a Test Coverage Program}) can use to
5161 show program coverage. Each source file's note file is called
5162 @file{@var{auxname}.gcno}. Refer to the @option{-fprofile-arcs} option
5163 above for a description of @var{auxname} and instructions on how to
5164 generate test coverage data. Coverage data will match the source files
5165 more closely, if you do not optimize.
5167 @item -fdbg-cnt-list
5168 @opindex fdbg-cnt-list
5169 Print the name and the counter upper bound for all debug counters.
5172 @item -fdbg-cnt=@var{counter-value-list}
5174 Set the internal debug counter upper bound. @var{counter-value-list}
5175 is a comma-separated list of @var{name}:@var{value} pairs
5176 which sets the upper bound of each debug counter @var{name} to @var{value}.
5177 All debug counters have the initial upper bound of @var{UINT_MAX},
5178 thus dbg_cnt() returns true always unless the upper bound is set by this option.
5179 e.g. With -fdbg-cnt=dce:10,tail_call:0
5180 dbg_cnt(dce) will return true only for first 10 invocations
5182 @item -fenable-@var{kind}-@var{pass}
5183 @itemx -fdisable-@var{kind}-@var{pass}=@var{range-list}
5187 This is a set of debugging options that are used to explicitly disable/enable
5188 optimization passes. For compiler users, regular options for enabling/disabling
5189 passes should be used instead.
5193 @item -fdisable-ipa-@var{pass}
5194 Disable ipa pass @var{pass}. @var{pass} is the pass name. If the same pass is
5195 statically invoked in the compiler multiple times, the pass name should be
5196 appended with a sequential number starting from 1.
5198 @item -fdisable-rtl-@var{pass}
5199 @item -fdisable-rtl-@var{pass}=@var{range-list}
5200 Disable rtl pass @var{pass}. @var{pass} is the pass name. If the same pass is
5201 statically invoked in the compiler multiple times, the pass name should be
5202 appended with a sequential number starting from 1. @var{range-list} is a comma
5203 seperated list of function ranges or assembler names. Each range is a number
5204 pair seperated by a colon. The range is inclusive in both ends. If the range
5205 is trivial, the number pair can be simplified as a single number. If the
5206 function's cgraph node's @var{uid} is falling within one of the specified ranges,
5207 the @var{pass} is disabled for that function. The @var{uid} is shown in the
5208 function header of a dump file, and the pass names can be dumped by using
5209 option @option{-fdump-passes}.
5211 @item -fdisable-tree-@var{pass}
5212 @item -fdisable-tree-@var{pass}=@var{range-list}
5213 Disable tree pass @var{pass}. See @option{-fdisable-rtl} for the description of
5216 @item -fenable-ipa-@var{pass}
5217 Enable ipa pass @var{pass}. @var{pass} is the pass name. If the same pass is
5218 statically invoked in the compiler multiple times, the pass name should be
5219 appended with a sequential number starting from 1.
5221 @item -fenable-rtl-@var{pass}
5222 @item -fenable-rtl-@var{pass}=@var{range-list}
5223 Enable rtl pass @var{pass}. See @option{-fdisable-rtl} for option argument
5224 description and examples.
5226 @item -fenable-tree-@var{pass}
5227 @item -fenable-tree-@var{pass}=@var{range-list}
5228 Enable tree pass @var{pass}. See @option{-fdisable-rtl} for the description
5229 of option arguments.
5233 # disable ccp1 for all functions
5235 # disable complete unroll for function whose cgraph node uid is 1
5236 -fenable-tree-cunroll=1
5237 # disable gcse2 for functions at the following ranges [1,1],
5238 # [300,400], and [400,1000]
5239 # disable gcse2 for functions foo and foo2
5240 -fdisable-rtl-gcse2=foo,foo2
5241 # disable early inlining
5242 -fdisable-tree-einline
5243 # disable ipa inlining
5244 -fdisable-ipa-inline
5245 # enable tree full unroll
5246 -fenable-tree-unroll
5252 @item -d@var{letters}
5253 @itemx -fdump-rtl-@var{pass}
5255 Says to make debugging dumps during compilation at times specified by
5256 @var{letters}. This is used for debugging the RTL-based passes of the
5257 compiler. The file names for most of the dumps are made by appending
5258 a pass number and a word to the @var{dumpname}, and the files are
5259 created in the directory of the output file. Note that the pass
5260 number is computed statically as passes get registered into the pass
5261 manager. Thus the numbering is not related to the dynamic order of
5262 execution of passes. In particular, a pass installed by a plugin
5263 could have a number over 200 even if it executed quite early.
5264 @var{dumpname} is generated from the name of the output file, if
5265 explicitly specified and it is not an executable, otherwise it is the
5266 basename of the source file. These switches may have different effects
5267 when @option{-E} is used for preprocessing.
5269 Debug dumps can be enabled with a @option{-fdump-rtl} switch or some
5270 @option{-d} option @var{letters}. Here are the possible
5271 letters for use in @var{pass} and @var{letters}, and their meanings:
5275 @item -fdump-rtl-alignments
5276 @opindex fdump-rtl-alignments
5277 Dump after branch alignments have been computed.
5279 @item -fdump-rtl-asmcons
5280 @opindex fdump-rtl-asmcons
5281 Dump after fixing rtl statements that have unsatisfied in/out constraints.
5283 @item -fdump-rtl-auto_inc_dec
5284 @opindex fdump-rtl-auto_inc_dec
5285 Dump after auto-inc-dec discovery. This pass is only run on
5286 architectures that have auto inc or auto dec instructions.
5288 @item -fdump-rtl-barriers
5289 @opindex fdump-rtl-barriers
5290 Dump after cleaning up the barrier instructions.
5292 @item -fdump-rtl-bbpart
5293 @opindex fdump-rtl-bbpart
5294 Dump after partitioning hot and cold basic blocks.
5296 @item -fdump-rtl-bbro
5297 @opindex fdump-rtl-bbro
5298 Dump after block reordering.
5300 @item -fdump-rtl-btl1
5301 @itemx -fdump-rtl-btl2
5302 @opindex fdump-rtl-btl2
5303 @opindex fdump-rtl-btl2
5304 @option{-fdump-rtl-btl1} and @option{-fdump-rtl-btl2} enable dumping
5305 after the two branch
5306 target load optimization passes.
5308 @item -fdump-rtl-bypass
5309 @opindex fdump-rtl-bypass
5310 Dump after jump bypassing and control flow optimizations.
5312 @item -fdump-rtl-combine
5313 @opindex fdump-rtl-combine
5314 Dump after the RTL instruction combination pass.
5316 @item -fdump-rtl-compgotos
5317 @opindex fdump-rtl-compgotos
5318 Dump after duplicating the computed gotos.
5320 @item -fdump-rtl-ce1
5321 @itemx -fdump-rtl-ce2
5322 @itemx -fdump-rtl-ce3
5323 @opindex fdump-rtl-ce1
5324 @opindex fdump-rtl-ce2
5325 @opindex fdump-rtl-ce3
5326 @option{-fdump-rtl-ce1}, @option{-fdump-rtl-ce2}, and
5327 @option{-fdump-rtl-ce3} enable dumping after the three
5328 if conversion passes.
5330 @item -fdump-rtl-cprop_hardreg
5331 @opindex fdump-rtl-cprop_hardreg
5332 Dump after hard register copy propagation.
5334 @item -fdump-rtl-csa
5335 @opindex fdump-rtl-csa
5336 Dump after combining stack adjustments.
5338 @item -fdump-rtl-cse1
5339 @itemx -fdump-rtl-cse2
5340 @opindex fdump-rtl-cse1
5341 @opindex fdump-rtl-cse2
5342 @option{-fdump-rtl-cse1} and @option{-fdump-rtl-cse2} enable dumping after
5343 the two common sub-expression elimination passes.
5345 @item -fdump-rtl-dce
5346 @opindex fdump-rtl-dce
5347 Dump after the standalone dead code elimination passes.
5349 @item -fdump-rtl-dbr
5350 @opindex fdump-rtl-dbr
5351 Dump after delayed branch scheduling.
5353 @item -fdump-rtl-dce1
5354 @itemx -fdump-rtl-dce2
5355 @opindex fdump-rtl-dce1
5356 @opindex fdump-rtl-dce2
5357 @option{-fdump-rtl-dce1} and @option{-fdump-rtl-dce2} enable dumping after
5358 the two dead store elimination passes.
5361 @opindex fdump-rtl-eh
5362 Dump after finalization of EH handling code.
5364 @item -fdump-rtl-eh_ranges
5365 @opindex fdump-rtl-eh_ranges
5366 Dump after conversion of EH handling range regions.
5368 @item -fdump-rtl-expand
5369 @opindex fdump-rtl-expand
5370 Dump after RTL generation.
5372 @item -fdump-rtl-fwprop1
5373 @itemx -fdump-rtl-fwprop2
5374 @opindex fdump-rtl-fwprop1
5375 @opindex fdump-rtl-fwprop2
5376 @option{-fdump-rtl-fwprop1} and @option{-fdump-rtl-fwprop2} enable
5377 dumping after the two forward propagation passes.
5379 @item -fdump-rtl-gcse1
5380 @itemx -fdump-rtl-gcse2
5381 @opindex fdump-rtl-gcse1
5382 @opindex fdump-rtl-gcse2
5383 @option{-fdump-rtl-gcse1} and @option{-fdump-rtl-gcse2} enable dumping
5384 after global common subexpression elimination.
5386 @item -fdump-rtl-init-regs
5387 @opindex fdump-rtl-init-regs
5388 Dump after the initialization of the registers.
5390 @item -fdump-rtl-initvals
5391 @opindex fdump-rtl-initvals
5392 Dump after the computation of the initial value sets.
5394 @item -fdump-rtl-into_cfglayout
5395 @opindex fdump-rtl-into_cfglayout
5396 Dump after converting to cfglayout mode.
5398 @item -fdump-rtl-ira
5399 @opindex fdump-rtl-ira
5400 Dump after iterated register allocation.
5402 @item -fdump-rtl-jump
5403 @opindex fdump-rtl-jump
5404 Dump after the second jump optimization.
5406 @item -fdump-rtl-loop2
5407 @opindex fdump-rtl-loop2
5408 @option{-fdump-rtl-loop2} enables dumping after the rtl
5409 loop optimization passes.
5411 @item -fdump-rtl-mach
5412 @opindex fdump-rtl-mach
5413 Dump after performing the machine dependent reorganization pass, if that
5416 @item -fdump-rtl-mode_sw
5417 @opindex fdump-rtl-mode_sw
5418 Dump after removing redundant mode switches.
5420 @item -fdump-rtl-rnreg
5421 @opindex fdump-rtl-rnreg
5422 Dump after register renumbering.
5424 @item -fdump-rtl-outof_cfglayout
5425 @opindex fdump-rtl-outof_cfglayout
5426 Dump after converting from cfglayout mode.
5428 @item -fdump-rtl-peephole2
5429 @opindex fdump-rtl-peephole2
5430 Dump after the peephole pass.
5432 @item -fdump-rtl-postreload
5433 @opindex fdump-rtl-postreload
5434 Dump after post-reload optimizations.
5436 @item -fdump-rtl-pro_and_epilogue
5437 @opindex fdump-rtl-pro_and_epilogue
5438 Dump after generating the function prologues and epilogues.
5440 @item -fdump-rtl-regmove
5441 @opindex fdump-rtl-regmove
5442 Dump after the register move pass.
5444 @item -fdump-rtl-sched1
5445 @itemx -fdump-rtl-sched2
5446 @opindex fdump-rtl-sched1
5447 @opindex fdump-rtl-sched2
5448 @option{-fdump-rtl-sched1} and @option{-fdump-rtl-sched2} enable dumping
5449 after the basic block scheduling passes.
5451 @item -fdump-rtl-see
5452 @opindex fdump-rtl-see
5453 Dump after sign extension elimination.
5455 @item -fdump-rtl-seqabstr
5456 @opindex fdump-rtl-seqabstr
5457 Dump after common sequence discovery.
5459 @item -fdump-rtl-shorten
5460 @opindex fdump-rtl-shorten
5461 Dump after shortening branches.
5463 @item -fdump-rtl-sibling
5464 @opindex fdump-rtl-sibling
5465 Dump after sibling call optimizations.
5467 @item -fdump-rtl-split1
5468 @itemx -fdump-rtl-split2
5469 @itemx -fdump-rtl-split3
5470 @itemx -fdump-rtl-split4
5471 @itemx -fdump-rtl-split5
5472 @opindex fdump-rtl-split1
5473 @opindex fdump-rtl-split2
5474 @opindex fdump-rtl-split3
5475 @opindex fdump-rtl-split4
5476 @opindex fdump-rtl-split5
5477 @option{-fdump-rtl-split1}, @option{-fdump-rtl-split2},
5478 @option{-fdump-rtl-split3}, @option{-fdump-rtl-split4} and
5479 @option{-fdump-rtl-split5} enable dumping after five rounds of
5480 instruction splitting.
5482 @item -fdump-rtl-sms
5483 @opindex fdump-rtl-sms
5484 Dump after modulo scheduling. This pass is only run on some
5487 @item -fdump-rtl-stack
5488 @opindex fdump-rtl-stack
5489 Dump after conversion from GCC's "flat register file" registers to the
5490 x87's stack-like registers. This pass is only run on x86 variants.
5492 @item -fdump-rtl-subreg1
5493 @itemx -fdump-rtl-subreg2
5494 @opindex fdump-rtl-subreg1
5495 @opindex fdump-rtl-subreg2
5496 @option{-fdump-rtl-subreg1} and @option{-fdump-rtl-subreg2} enable dumping after
5497 the two subreg expansion passes.
5499 @item -fdump-rtl-unshare
5500 @opindex fdump-rtl-unshare
5501 Dump after all rtl has been unshared.
5503 @item -fdump-rtl-vartrack
5504 @opindex fdump-rtl-vartrack
5505 Dump after variable tracking.
5507 @item -fdump-rtl-vregs
5508 @opindex fdump-rtl-vregs
5509 Dump after converting virtual registers to hard registers.
5511 @item -fdump-rtl-web
5512 @opindex fdump-rtl-web
5513 Dump after live range splitting.
5515 @item -fdump-rtl-regclass
5516 @itemx -fdump-rtl-subregs_of_mode_init
5517 @itemx -fdump-rtl-subregs_of_mode_finish
5518 @itemx -fdump-rtl-dfinit
5519 @itemx -fdump-rtl-dfinish
5520 @opindex fdump-rtl-regclass
5521 @opindex fdump-rtl-subregs_of_mode_init
5522 @opindex fdump-rtl-subregs_of_mode_finish
5523 @opindex fdump-rtl-dfinit
5524 @opindex fdump-rtl-dfinish
5525 These dumps are defined but always produce empty files.
5528 @itemx -fdump-rtl-all
5530 @opindex fdump-rtl-all
5531 Produce all the dumps listed above.
5535 Annotate the assembler output with miscellaneous debugging information.
5539 Dump all macro definitions, at the end of preprocessing, in addition to
5544 Produce a core dump whenever an error occurs.
5548 Annotate the assembler output with a comment indicating which
5549 pattern and alternative was used. The length of each instruction is
5554 Dump the RTL in the assembler output as a comment before each instruction.
5555 Also turns on @option{-dp} annotation.
5559 For each of the other indicated dump files (@option{-fdump-rtl-@var{pass}}),
5560 dump a representation of the control flow graph suitable for viewing with VCG
5561 to @file{@var{file}.@var{pass}.vcg}.
5565 Just generate RTL for a function instead of compiling it. Usually used
5566 with @option{-fdump-rtl-expand}.
5570 @opindex fdump-noaddr
5571 When doing debugging dumps, suppress address output. This makes it more
5572 feasible to use diff on debugging dumps for compiler invocations with
5573 different compiler binaries and/or different
5574 text / bss / data / heap / stack / dso start locations.
5576 @item -fdump-unnumbered
5577 @opindex fdump-unnumbered
5578 When doing debugging dumps, suppress instruction numbers and address output.
5579 This makes it more feasible to use diff on debugging dumps for compiler
5580 invocations with different options, in particular with and without
5583 @item -fdump-unnumbered-links
5584 @opindex fdump-unnumbered-links
5585 When doing debugging dumps (see @option{-d} option above), suppress
5586 instruction numbers for the links to the previous and next instructions
5589 @item -fdump-translation-unit @r{(C++ only)}
5590 @itemx -fdump-translation-unit-@var{options} @r{(C++ only)}
5591 @opindex fdump-translation-unit
5592 Dump a representation of the tree structure for the entire translation
5593 unit to a file. The file name is made by appending @file{.tu} to the
5594 source file name, and the file is created in the same directory as the
5595 output file. If the @samp{-@var{options}} form is used, @var{options}
5596 controls the details of the dump as described for the
5597 @option{-fdump-tree} options.
5599 @item -fdump-class-hierarchy @r{(C++ only)}
5600 @itemx -fdump-class-hierarchy-@var{options} @r{(C++ only)}
5601 @opindex fdump-class-hierarchy
5602 Dump a representation of each class's hierarchy and virtual function
5603 table layout to a file. The file name is made by appending
5604 @file{.class} to the source file name, and the file is created in the
5605 same directory as the output file. If the @samp{-@var{options}} form
5606 is used, @var{options} controls the details of the dump as described
5607 for the @option{-fdump-tree} options.
5609 @item -fdump-ipa-@var{switch}
5611 Control the dumping at various stages of inter-procedural analysis
5612 language tree to a file. The file name is generated by appending a
5613 switch specific suffix to the source file name, and the file is created
5614 in the same directory as the output file. The following dumps are
5619 Enables all inter-procedural analysis dumps.
5622 Dumps information about call-graph optimization, unused function removal,
5623 and inlining decisions.
5626 Dump after function inlining.
5631 @opindex fdump-passes
5632 Dump the list of optimization passes that are turned on and off by
5633 the current command-line options.
5635 @item -fdump-statistics-@var{option}
5636 @opindex fdump-statistics
5637 Enable and control dumping of pass statistics in a separate file. The
5638 file name is generated by appending a suffix ending in
5639 @samp{.statistics} to the source file name, and the file is created in
5640 the same directory as the output file. If the @samp{-@var{option}}
5641 form is used, @samp{-stats} will cause counters to be summed over the
5642 whole compilation unit while @samp{-details} will dump every event as
5643 the passes generate them. The default with no option is to sum
5644 counters for each function compiled.
5646 @item -fdump-tree-@var{switch}
5647 @itemx -fdump-tree-@var{switch}-@var{options}
5649 Control the dumping at various stages of processing the intermediate
5650 language tree to a file. The file name is generated by appending a
5651 switch specific suffix to the source file name, and the file is
5652 created in the same directory as the output file. If the
5653 @samp{-@var{options}} form is used, @var{options} is a list of
5654 @samp{-} separated options which control the details of the dump. Not
5655 all options are applicable to all dumps; those that are not
5656 meaningful will be ignored. The following options are available
5660 Print the address of each node. Usually this is not meaningful as it
5661 changes according to the environment and source file. Its primary use
5662 is for tying up a dump file with a debug environment.
5664 If @code{DECL_ASSEMBLER_NAME} has been set for a given decl, use that
5665 in the dump instead of @code{DECL_NAME}. Its primary use is ease of
5666 use working backward from mangled names in the assembly file.
5668 Inhibit dumping of members of a scope or body of a function merely
5669 because that scope has been reached. Only dump such items when they
5670 are directly reachable by some other path. When dumping pretty-printed
5671 trees, this option inhibits dumping the bodies of control structures.
5673 Print a raw representation of the tree. By default, trees are
5674 pretty-printed into a C-like representation.
5676 Enable more detailed dumps (not honored by every dump option).
5678 Enable dumping various statistics about the pass (not honored by every dump
5681 Enable showing basic block boundaries (disabled in raw dumps).
5683 Enable showing virtual operands for every statement.
5685 Enable showing line numbers for statements.
5687 Enable showing the unique ID (@code{DECL_UID}) for each variable.
5689 Enable showing the tree dump for each statement.
5691 Enable showing the EH region number holding each statement.
5693 Enable showing scalar evolution analysis details.
5695 Turn on all options, except @option{raw}, @option{slim}, @option{verbose}
5696 and @option{lineno}.
5699 The following tree dumps are possible:
5703 @opindex fdump-tree-original
5704 Dump before any tree based optimization, to @file{@var{file}.original}.
5707 @opindex fdump-tree-optimized
5708 Dump after all tree based optimization, to @file{@var{file}.optimized}.
5711 @opindex fdump-tree-gimple
5712 Dump each function before and after the gimplification pass to a file. The
5713 file name is made by appending @file{.gimple} to the source file name.
5716 @opindex fdump-tree-cfg
5717 Dump the control flow graph of each function to a file. The file name is
5718 made by appending @file{.cfg} to the source file name.
5721 @opindex fdump-tree-vcg
5722 Dump the control flow graph of each function to a file in VCG format. The
5723 file name is made by appending @file{.vcg} to the source file name. Note
5724 that if the file contains more than one function, the generated file cannot
5725 be used directly by VCG@. You will need to cut and paste each function's
5726 graph into its own separate file first.
5729 @opindex fdump-tree-ch
5730 Dump each function after copying loop headers. The file name is made by
5731 appending @file{.ch} to the source file name.
5734 @opindex fdump-tree-ssa
5735 Dump SSA related information to a file. The file name is made by appending
5736 @file{.ssa} to the source file name.
5739 @opindex fdump-tree-alias
5740 Dump aliasing information for each function. The file name is made by
5741 appending @file{.alias} to the source file name.
5744 @opindex fdump-tree-ccp
5745 Dump each function after CCP@. The file name is made by appending
5746 @file{.ccp} to the source file name.
5749 @opindex fdump-tree-storeccp
5750 Dump each function after STORE-CCP@. The file name is made by appending
5751 @file{.storeccp} to the source file name.
5754 @opindex fdump-tree-pre
5755 Dump trees after partial redundancy elimination. The file name is made
5756 by appending @file{.pre} to the source file name.
5759 @opindex fdump-tree-fre
5760 Dump trees after full redundancy elimination. The file name is made
5761 by appending @file{.fre} to the source file name.
5764 @opindex fdump-tree-copyprop
5765 Dump trees after copy propagation. The file name is made
5766 by appending @file{.copyprop} to the source file name.
5768 @item store_copyprop
5769 @opindex fdump-tree-store_copyprop
5770 Dump trees after store copy-propagation. The file name is made
5771 by appending @file{.store_copyprop} to the source file name.
5774 @opindex fdump-tree-dce
5775 Dump each function after dead code elimination. The file name is made by
5776 appending @file{.dce} to the source file name.
5779 @opindex fdump-tree-mudflap
5780 Dump each function after adding mudflap instrumentation. The file name is
5781 made by appending @file{.mudflap} to the source file name.
5784 @opindex fdump-tree-sra
5785 Dump each function after performing scalar replacement of aggregates. The
5786 file name is made by appending @file{.sra} to the source file name.
5789 @opindex fdump-tree-sink
5790 Dump each function after performing code sinking. The file name is made
5791 by appending @file{.sink} to the source file name.
5794 @opindex fdump-tree-dom
5795 Dump each function after applying dominator tree optimizations. The file
5796 name is made by appending @file{.dom} to the source file name.
5799 @opindex fdump-tree-dse
5800 Dump each function after applying dead store elimination. The file
5801 name is made by appending @file{.dse} to the source file name.
5804 @opindex fdump-tree-phiopt
5805 Dump each function after optimizing PHI nodes into straightline code. The file
5806 name is made by appending @file{.phiopt} to the source file name.
5809 @opindex fdump-tree-forwprop
5810 Dump each function after forward propagating single use variables. The file
5811 name is made by appending @file{.forwprop} to the source file name.
5814 @opindex fdump-tree-copyrename
5815 Dump each function after applying the copy rename optimization. The file
5816 name is made by appending @file{.copyrename} to the source file name.
5819 @opindex fdump-tree-nrv
5820 Dump each function after applying the named return value optimization on
5821 generic trees. The file name is made by appending @file{.nrv} to the source
5825 @opindex fdump-tree-vect
5826 Dump each function after applying vectorization of loops. The file name is
5827 made by appending @file{.vect} to the source file name.
5830 @opindex fdump-tree-slp
5831 Dump each function after applying vectorization of basic blocks. The file name
5832 is made by appending @file{.slp} to the source file name.
5835 @opindex fdump-tree-vrp
5836 Dump each function after Value Range Propagation (VRP). The file name
5837 is made by appending @file{.vrp} to the source file name.
5840 @opindex fdump-tree-all
5841 Enable all the available tree dumps with the flags provided in this option.
5844 @item -ftree-vectorizer-verbose=@var{n}
5845 @opindex ftree-vectorizer-verbose
5846 This option controls the amount of debugging output the vectorizer prints.
5847 This information is written to standard error, unless
5848 @option{-fdump-tree-all} or @option{-fdump-tree-vect} is specified,
5849 in which case it is output to the usual dump listing file, @file{.vect}.
5850 For @var{n}=0 no diagnostic information is reported.
5851 If @var{n}=1 the vectorizer reports each loop that got vectorized,
5852 and the total number of loops that got vectorized.
5853 If @var{n}=2 the vectorizer also reports non-vectorized loops that passed
5854 the first analysis phase (vect_analyze_loop_form) - i.e.@: countable,
5855 inner-most, single-bb, single-entry/exit loops. This is the same verbosity
5856 level that @option{-fdump-tree-vect-stats} uses.
5857 Higher verbosity levels mean either more information dumped for each
5858 reported loop, or same amount of information reported for more loops:
5859 if @var{n}=3, vectorizer cost model information is reported.
5860 If @var{n}=4, alignment related information is added to the reports.
5861 If @var{n}=5, data-references related information (e.g.@: memory dependences,
5862 memory access-patterns) is added to the reports.
5863 If @var{n}=6, the vectorizer reports also non-vectorized inner-most loops
5864 that did not pass the first analysis phase (i.e., may not be countable, or
5865 may have complicated control-flow).
5866 If @var{n}=7, the vectorizer reports also non-vectorized nested loops.
5867 If @var{n}=8, SLP related information is added to the reports.
5868 For @var{n}=9, all the information the vectorizer generates during its
5869 analysis and transformation is reported. This is the same verbosity level
5870 that @option{-fdump-tree-vect-details} uses.
5872 @item -frandom-seed=@var{string}
5873 @opindex frandom-seed
5874 This option provides a seed that GCC uses when it would otherwise use
5875 random numbers. It is used to generate certain symbol names
5876 that have to be different in every compiled file. It is also used to
5877 place unique stamps in coverage data files and the object files that
5878 produce them. You can use the @option{-frandom-seed} option to produce
5879 reproducibly identical object files.
5881 The @var{string} should be different for every file you compile.
5883 @item -fsched-verbose=@var{n}
5884 @opindex fsched-verbose
5885 On targets that use instruction scheduling, this option controls the
5886 amount of debugging output the scheduler prints. This information is
5887 written to standard error, unless @option{-fdump-rtl-sched1} or
5888 @option{-fdump-rtl-sched2} is specified, in which case it is output
5889 to the usual dump listing file, @file{.sched1} or @file{.sched2}
5890 respectively. However for @var{n} greater than nine, the output is
5891 always printed to standard error.
5893 For @var{n} greater than zero, @option{-fsched-verbose} outputs the
5894 same information as @option{-fdump-rtl-sched1} and @option{-fdump-rtl-sched2}.
5895 For @var{n} greater than one, it also output basic block probabilities,
5896 detailed ready list information and unit/insn info. For @var{n} greater
5897 than two, it includes RTL at abort point, control-flow and regions info.
5898 And for @var{n} over four, @option{-fsched-verbose} also includes
5902 @itemx -save-temps=cwd
5904 Store the usual ``temporary'' intermediate files permanently; place them
5905 in the current directory and name them based on the source file. Thus,
5906 compiling @file{foo.c} with @samp{-c -save-temps} would produce files
5907 @file{foo.i} and @file{foo.s}, as well as @file{foo.o}. This creates a
5908 preprocessed @file{foo.i} output file even though the compiler now
5909 normally uses an integrated preprocessor.
5911 When used in combination with the @option{-x} command-line option,
5912 @option{-save-temps} is sensible enough to avoid over writing an
5913 input source file with the same extension as an intermediate file.
5914 The corresponding intermediate file may be obtained by renaming the
5915 source file before using @option{-save-temps}.
5917 If you invoke GCC in parallel, compiling several different source
5918 files that share a common base name in different subdirectories or the
5919 same source file compiled for multiple output destinations, it is
5920 likely that the different parallel compilers will interfere with each
5921 other, and overwrite the temporary files. For instance:
5924 gcc -save-temps -o outdir1/foo.o indir1/foo.c&
5925 gcc -save-temps -o outdir2/foo.o indir2/foo.c&
5928 may result in @file{foo.i} and @file{foo.o} being written to
5929 simultaneously by both compilers.
5931 @item -save-temps=obj
5932 @opindex save-temps=obj
5933 Store the usual ``temporary'' intermediate files permanently. If the
5934 @option{-o} option is used, the temporary files are based on the
5935 object file. If the @option{-o} option is not used, the
5936 @option{-save-temps=obj} switch behaves like @option{-save-temps}.
5941 gcc -save-temps=obj -c foo.c
5942 gcc -save-temps=obj -c bar.c -o dir/xbar.o
5943 gcc -save-temps=obj foobar.c -o dir2/yfoobar
5946 would create @file{foo.i}, @file{foo.s}, @file{dir/xbar.i},
5947 @file{dir/xbar.s}, @file{dir2/yfoobar.i}, @file{dir2/yfoobar.s}, and
5948 @file{dir2/yfoobar.o}.
5950 @item -time@r{[}=@var{file}@r{]}
5952 Report the CPU time taken by each subprocess in the compilation
5953 sequence. For C source files, this is the compiler proper and assembler
5954 (plus the linker if linking is done).
5956 Without the specification of an output file, the output looks like this:
5963 The first number on each line is the ``user time'', that is time spent
5964 executing the program itself. The second number is ``system time'',
5965 time spent executing operating system routines on behalf of the program.
5966 Both numbers are in seconds.
5968 With the specification of an output file, the output is appended to the
5969 named file, and it looks like this:
5972 0.12 0.01 cc1 @var{options}
5973 0.00 0.01 as @var{options}
5976 The ``user time'' and the ``system time'' are moved before the program
5977 name, and the options passed to the program are displayed, so that one
5978 can later tell what file was being compiled, and with which options.
5980 @item -fvar-tracking
5981 @opindex fvar-tracking
5982 Run variable tracking pass. It computes where variables are stored at each
5983 position in code. Better debugging information is then generated
5984 (if the debugging information format supports this information).
5986 It is enabled by default when compiling with optimization (@option{-Os},
5987 @option{-O}, @option{-O2}, @dots{}), debugging information (@option{-g}) and
5988 the debug info format supports it.
5990 @item -fvar-tracking-assignments
5991 @opindex fvar-tracking-assignments
5992 @opindex fno-var-tracking-assignments
5993 Annotate assignments to user variables early in the compilation and
5994 attempt to carry the annotations over throughout the compilation all the
5995 way to the end, in an attempt to improve debug information while
5996 optimizing. Use of @option{-gdwarf-4} is recommended along with it.
5998 It can be enabled even if var-tracking is disabled, in which case
5999 annotations will be created and maintained, but discarded at the end.
6001 @item -fvar-tracking-assignments-toggle
6002 @opindex fvar-tracking-assignments-toggle
6003 @opindex fno-var-tracking-assignments-toggle
6004 Toggle @option{-fvar-tracking-assignments}, in the same way that
6005 @option{-gtoggle} toggles @option{-g}.
6007 @item -print-file-name=@var{library}
6008 @opindex print-file-name
6009 Print the full absolute name of the library file @var{library} that
6010 would be used when linking---and don't do anything else. With this
6011 option, GCC does not compile or link anything; it just prints the
6014 @item -print-multi-directory
6015 @opindex print-multi-directory
6016 Print the directory name corresponding to the multilib selected by any
6017 other switches present in the command line. This directory is supposed
6018 to exist in @env{GCC_EXEC_PREFIX}.
6020 @item -print-multi-lib
6021 @opindex print-multi-lib
6022 Print the mapping from multilib directory names to compiler switches
6023 that enable them. The directory name is separated from the switches by
6024 @samp{;}, and each switch starts with an @samp{@@} instead of the
6025 @samp{-}, without spaces between multiple switches. This is supposed to
6026 ease shell-processing.
6028 @item -print-multi-os-directory
6029 @opindex print-multi-os-directory
6030 Print the path to OS libraries for the selected
6031 multilib, relative to some @file{lib} subdirectory. If OS libraries are
6032 present in the @file{lib} subdirectory and no multilibs are used, this is
6033 usually just @file{.}, if OS libraries are present in @file{lib@var{suffix}}
6034 sibling directories this prints e.g.@: @file{../lib64}, @file{../lib} or
6035 @file{../lib32}, or if OS libraries are present in @file{lib/@var{subdir}}
6036 subdirectories it prints e.g.@: @file{amd64}, @file{sparcv9} or @file{ev6}.
6038 @item -print-multiarch
6039 @opindex print-multiarch
6040 Print the path to OS libraries for the selected multiarch,
6041 relative to some @file{lib} subdirectory.
6043 @item -print-prog-name=@var{program}
6044 @opindex print-prog-name
6045 Like @option{-print-file-name}, but searches for a program such as @samp{cpp}.
6047 @item -print-libgcc-file-name
6048 @opindex print-libgcc-file-name
6049 Same as @option{-print-file-name=libgcc.a}.
6051 This is useful when you use @option{-nostdlib} or @option{-nodefaultlibs}
6052 but you do want to link with @file{libgcc.a}. You can do
6055 gcc -nostdlib @var{files}@dots{} `gcc -print-libgcc-file-name`
6058 @item -print-search-dirs
6059 @opindex print-search-dirs
6060 Print the name of the configured installation directory and a list of
6061 program and library directories @command{gcc} will search---and don't do anything else.
6063 This is useful when @command{gcc} prints the error message
6064 @samp{installation problem, cannot exec cpp0: No such file or directory}.
6065 To resolve this you either need to put @file{cpp0} and the other compiler
6066 components where @command{gcc} expects to find them, or you can set the environment
6067 variable @env{GCC_EXEC_PREFIX} to the directory where you installed them.
6068 Don't forget the trailing @samp{/}.
6069 @xref{Environment Variables}.
6071 @item -print-sysroot
6072 @opindex print-sysroot
6073 Print the target sysroot directory that will be used during
6074 compilation. This is the target sysroot specified either at configure
6075 time or using the @option{--sysroot} option, possibly with an extra
6076 suffix that depends on compilation options. If no target sysroot is
6077 specified, the option prints nothing.
6079 @item -print-sysroot-headers-suffix
6080 @opindex print-sysroot-headers-suffix
6081 Print the suffix added to the target sysroot when searching for
6082 headers, or give an error if the compiler is not configured with such
6083 a suffix---and don't do anything else.
6086 @opindex dumpmachine
6087 Print the compiler's target machine (for example,
6088 @samp{i686-pc-linux-gnu})---and don't do anything else.
6091 @opindex dumpversion
6092 Print the compiler version (for example, @samp{3.0})---and don't do
6097 Print the compiler's built-in specs---and don't do anything else. (This
6098 is used when GCC itself is being built.) @xref{Spec Files}.
6100 @item -feliminate-unused-debug-types
6101 @opindex feliminate-unused-debug-types
6102 Normally, when producing DWARF2 output, GCC will emit debugging
6103 information for all types declared in a compilation
6104 unit, regardless of whether or not they are actually used
6105 in that compilation unit. Sometimes this is useful, such as
6106 if, in the debugger, you want to cast a value to a type that is
6107 not actually used in your program (but is declared). More often,
6108 however, this results in a significant amount of wasted space.
6109 With this option, GCC will avoid producing debug symbol output
6110 for types that are nowhere used in the source file being compiled.
6113 @node Optimize Options
6114 @section Options That Control Optimization
6115 @cindex optimize options
6116 @cindex options, optimization
6118 These options control various sorts of optimizations.
6120 Without any optimization option, the compiler's goal is to reduce the
6121 cost of compilation and to make debugging produce the expected
6122 results. Statements are independent: if you stop the program with a
6123 breakpoint between statements, you can then assign a new value to any
6124 variable or change the program counter to any other statement in the
6125 function and get exactly the results you would expect from the source
6128 Turning on optimization flags makes the compiler attempt to improve
6129 the performance and/or code size at the expense of compilation time
6130 and possibly the ability to debug the program.
6132 The compiler performs optimization based on the knowledge it has of the
6133 program. Compiling multiple files at once to a single output file mode allows
6134 the compiler to use information gained from all of the files when compiling
6137 Not all optimizations are controlled directly by a flag. Only
6138 optimizations that have a flag are listed in this section.
6140 Most optimizations are only enabled if an @option{-O} level is set on
6141 the command line. Otherwise they are disabled, even if individual
6142 optimization flags are specified.
6144 Depending on the target and how GCC was configured, a slightly different
6145 set of optimizations may be enabled at each @option{-O} level than
6146 those listed here. You can invoke GCC with @samp{-Q --help=optimizers}
6147 to find out the exact set of optimizations that are enabled at each level.
6148 @xref{Overall Options}, for examples.
6155 Optimize. Optimizing compilation takes somewhat more time, and a lot
6156 more memory for a large function.
6158 With @option{-O}, the compiler tries to reduce code size and execution
6159 time, without performing any optimizations that take a great deal of
6162 @option{-O} turns on the following optimization flags:
6166 -fcprop-registers @gol
6169 -fdelayed-branch @gol
6171 -fguess-branch-probability @gol
6172 -fif-conversion2 @gol
6173 -fif-conversion @gol
6174 -fipa-pure-const @gol
6176 -fipa-reference @gol
6178 -fsplit-wide-types @gol
6180 -ftree-builtin-call-dce @gol
6183 -ftree-copyrename @gol
6185 -ftree-dominator-opts @gol
6187 -ftree-forwprop @gol
6195 @option{-O} also turns on @option{-fomit-frame-pointer} on machines
6196 where doing so does not interfere with debugging.
6200 Optimize even more. GCC performs nearly all supported optimizations
6201 that do not involve a space-speed tradeoff.
6202 As compared to @option{-O}, this option increases both compilation time
6203 and the performance of the generated code.
6205 @option{-O2} turns on all optimization flags specified by @option{-O}. It
6206 also turns on the following optimization flags:
6207 @gccoptlist{-fthread-jumps @gol
6208 -falign-functions -falign-jumps @gol
6209 -falign-loops -falign-labels @gol
6212 -fcse-follow-jumps -fcse-skip-blocks @gol
6213 -fdelete-null-pointer-checks @gol
6215 -fexpensive-optimizations @gol
6216 -fgcse -fgcse-lm @gol
6217 -finline-small-functions @gol
6218 -findirect-inlining @gol
6220 -foptimize-sibling-calls @gol
6221 -fpartial-inlining @gol
6224 -freorder-blocks -freorder-functions @gol
6225 -frerun-cse-after-loop @gol
6226 -fsched-interblock -fsched-spec @gol
6227 -fschedule-insns -fschedule-insns2 @gol
6228 -fstrict-aliasing -fstrict-overflow @gol
6229 -ftree-switch-conversion -ftree-tail-merge @gol
6233 Please note the warning under @option{-fgcse} about
6234 invoking @option{-O2} on programs that use computed gotos.
6238 Optimize yet more. @option{-O3} turns on all optimizations specified
6239 by @option{-O2} and also turns on the @option{-finline-functions},
6240 @option{-funswitch-loops}, @option{-fpredictive-commoning},
6241 @option{-fgcse-after-reload}, @option{-ftree-vectorize} and
6242 @option{-fipa-cp-clone} options.
6246 Reduce compilation time and make debugging produce the expected
6247 results. This is the default.
6251 Optimize for size. @option{-Os} enables all @option{-O2} optimizations that
6252 do not typically increase code size. It also performs further
6253 optimizations designed to reduce code size.
6255 @option{-Os} disables the following optimization flags:
6256 @gccoptlist{-falign-functions -falign-jumps -falign-loops @gol
6257 -falign-labels -freorder-blocks -freorder-blocks-and-partition @gol
6258 -fprefetch-loop-arrays -ftree-vect-loop-version}
6262 Disregard strict standards compliance. @option{-Ofast} enables all
6263 @option{-O3} optimizations. It also enables optimizations that are not
6264 valid for all standard compliant programs.
6265 It turns on @option{-ffast-math} and the Fortran-specific
6266 @option{-fno-protect-parens} and @option{-fstack-arrays}.
6268 If you use multiple @option{-O} options, with or without level numbers,
6269 the last such option is the one that is effective.
6272 Options of the form @option{-f@var{flag}} specify machine-independent
6273 flags. Most flags have both positive and negative forms; the negative
6274 form of @option{-ffoo} would be @option{-fno-foo}. In the table
6275 below, only one of the forms is listed---the one you typically will
6276 use. You can figure out the other form by either removing @samp{no-}
6279 The following options control specific optimizations. They are either
6280 activated by @option{-O} options or are related to ones that are. You
6281 can use the following flags in the rare cases when ``fine-tuning'' of
6282 optimizations to be performed is desired.
6285 @item -fno-default-inline
6286 @opindex fno-default-inline
6287 Do not make member functions inline by default merely because they are
6288 defined inside the class scope (C++ only). Otherwise, when you specify
6289 @w{@option{-O}}, member functions defined inside class scope are compiled
6290 inline by default; i.e., you don't need to add @samp{inline} in front of
6291 the member function name.
6293 @item -fno-defer-pop
6294 @opindex fno-defer-pop
6295 Always pop the arguments to each function call as soon as that function
6296 returns. For machines that must pop arguments after a function call,
6297 the compiler normally lets arguments accumulate on the stack for several
6298 function calls and pops them all at once.
6300 Disabled at levels @option{-O}, @option{-O2}, @option{-O3}, @option{-Os}.
6302 @item -fforward-propagate
6303 @opindex fforward-propagate
6304 Perform a forward propagation pass on RTL@. The pass tries to combine two
6305 instructions and checks if the result can be simplified. If loop unrolling
6306 is active, two passes are performed and the second is scheduled after
6309 This option is enabled by default at optimization levels @option{-O},
6310 @option{-O2}, @option{-O3}, @option{-Os}.
6312 @item -ffp-contract=@var{style}
6313 @opindex ffp-contract
6314 @option{-ffp-contract=off} disables floating-point expression contraction.
6315 @option{-ffp-contract=fast} enables floating-point expression contraction
6316 such as forming of fused multiply-add operations if the target has
6317 native support for them.
6318 @option{-ffp-contract=on} enables floating-point expression contraction
6319 if allowed by the language standard. This is currently not implemented
6320 and treated equal to @option{-ffp-contract=off}.
6322 The default is @option{-ffp-contract=fast}.
6324 @item -fomit-frame-pointer
6325 @opindex fomit-frame-pointer
6326 Don't keep the frame pointer in a register for functions that
6327 don't need one. This avoids the instructions to save, set up and
6328 restore frame pointers; it also makes an extra register available
6329 in many functions. @strong{It also makes debugging impossible on
6332 On some machines, such as the VAX, this flag has no effect, because
6333 the standard calling sequence automatically handles the frame pointer
6334 and nothing is saved by pretending it doesn't exist. The
6335 machine-description macro @code{FRAME_POINTER_REQUIRED} controls
6336 whether a target machine supports this flag. @xref{Registers,,Register
6337 Usage, gccint, GNU Compiler Collection (GCC) Internals}.
6339 Starting with GCC version 4.6, the default setting (when not optimizing for
6340 size) for 32-bit Linux x86 and 32-bit Darwin x86 targets has been changed to
6341 @option{-fomit-frame-pointer}. The default can be reverted to
6342 @option{-fno-omit-frame-pointer} by configuring GCC with the
6343 @option{--enable-frame-pointer} configure option.
6345 Enabled at levels @option{-O}, @option{-O2}, @option{-O3}, @option{-Os}.
6347 @item -foptimize-sibling-calls
6348 @opindex foptimize-sibling-calls
6349 Optimize sibling and tail recursive calls.
6351 Enabled at levels @option{-O2}, @option{-O3}, @option{-Os}.
6355 Do not expand any functions inline apart from those marked with
6356 the @code{always_inline} attribute. This is the default when not
6359 Single functions can be exempted from inlining by marking them
6360 with the @code{noinline} attribute.
6362 @item -finline-small-functions
6363 @opindex finline-small-functions
6364 Integrate functions into their callers when their body is smaller than expected
6365 function call code (so overall size of program gets smaller). The compiler
6366 heuristically decides which functions are simple enough to be worth integrating
6367 in this way. This inlining applies to all functions, even those not declared
6370 Enabled at level @option{-O2}.
6372 @item -findirect-inlining
6373 @opindex findirect-inlining
6374 Inline also indirect calls that are discovered to be known at compile
6375 time thanks to previous inlining. This option has any effect only
6376 when inlining itself is turned on by the @option{-finline-functions}
6377 or @option{-finline-small-functions} options.
6379 Enabled at level @option{-O2}.
6381 @item -finline-functions
6382 @opindex finline-functions
6383 Consider all functions for inlining, even if they are not declared inline.
6384 The compiler heuristically decides which functions are worth integrating
6387 If all calls to a given function are integrated, and the function is
6388 declared @code{static}, then the function is normally not output as
6389 assembler code in its own right.
6391 Enabled at level @option{-O3}.
6393 @item -finline-functions-called-once
6394 @opindex finline-functions-called-once
6395 Consider all @code{static} functions called once for inlining into their
6396 caller even if they are not marked @code{inline}. If a call to a given
6397 function is integrated, then the function is not output as assembler code
6400 Enabled at levels @option{-O1}, @option{-O2}, @option{-O3} and @option{-Os}.
6402 @item -fearly-inlining
6403 @opindex fearly-inlining
6404 Inline functions marked by @code{always_inline} and functions whose body seems
6405 smaller than the function call overhead early before doing
6406 @option{-fprofile-generate} instrumentation and real inlining pass. Doing so
6407 makes profiling significantly cheaper and usually inlining faster on programs
6408 having large chains of nested wrapper functions.
6414 Perform interprocedural scalar replacement of aggregates, removal of
6415 unused parameters and replacement of parameters passed by reference
6416 by parameters passed by value.
6418 Enabled at levels @option{-O2}, @option{-O3} and @option{-Os}.
6420 @item -finline-limit=@var{n}
6421 @opindex finline-limit
6422 By default, GCC limits the size of functions that can be inlined. This flag
6423 allows coarse control of this limit. @var{n} is the size of functions that
6424 can be inlined in number of pseudo instructions.
6426 Inlining is actually controlled by a number of parameters, which may be
6427 specified individually by using @option{--param @var{name}=@var{value}}.
6428 The @option{-finline-limit=@var{n}} option sets some of these parameters
6432 @item max-inline-insns-single
6433 is set to @var{n}/2.
6434 @item max-inline-insns-auto
6435 is set to @var{n}/2.
6438 See below for a documentation of the individual
6439 parameters controlling inlining and for the defaults of these parameters.
6441 @emph{Note:} there may be no value to @option{-finline-limit} that results
6442 in default behavior.
6444 @emph{Note:} pseudo instruction represents, in this particular context, an
6445 abstract measurement of function's size. In no way does it represent a count
6446 of assembly instructions and as such its exact meaning might change from one
6447 release to an another.
6449 @item -fno-keep-inline-dllexport
6450 @opindex -fno-keep-inline-dllexport
6451 This is a more fine-grained version of @option{-fkeep-inline-functions},
6452 which applies only to functions that are declared using the @code{dllexport}
6453 attribute or declspec (@xref{Function Attributes,,Declaring Attributes of
6456 @item -fkeep-inline-functions
6457 @opindex fkeep-inline-functions
6458 In C, emit @code{static} functions that are declared @code{inline}
6459 into the object file, even if the function has been inlined into all
6460 of its callers. This switch does not affect functions using the
6461 @code{extern inline} extension in GNU C90@. In C++, emit any and all
6462 inline functions into the object file.
6464 @item -fkeep-static-consts
6465 @opindex fkeep-static-consts
6466 Emit variables declared @code{static const} when optimization isn't turned
6467 on, even if the variables aren't referenced.
6469 GCC enables this option by default. If you want to force the compiler to
6470 check if the variable was referenced, regardless of whether or not
6471 optimization is turned on, use the @option{-fno-keep-static-consts} option.
6473 @item -fmerge-constants
6474 @opindex fmerge-constants
6475 Attempt to merge identical constants (string constants and floating-point
6476 constants) across compilation units.
6478 This option is the default for optimized compilation if the assembler and
6479 linker support it. Use @option{-fno-merge-constants} to inhibit this
6482 Enabled at levels @option{-O}, @option{-O2}, @option{-O3}, @option{-Os}.
6484 @item -fmerge-all-constants
6485 @opindex fmerge-all-constants
6486 Attempt to merge identical constants and identical variables.
6488 This option implies @option{-fmerge-constants}. In addition to
6489 @option{-fmerge-constants} this considers e.g.@: even constant initialized
6490 arrays or initialized constant variables with integral or floating-point
6491 types. Languages like C or C++ require each variable, including multiple
6492 instances of the same variable in recursive calls, to have distinct locations,
6493 so using this option will result in non-conforming
6496 @item -fmodulo-sched
6497 @opindex fmodulo-sched
6498 Perform swing modulo scheduling immediately before the first scheduling
6499 pass. This pass looks at innermost loops and reorders their
6500 instructions by overlapping different iterations.
6502 @item -fmodulo-sched-allow-regmoves
6503 @opindex fmodulo-sched-allow-regmoves
6504 Perform more aggressive SMS based modulo scheduling with register moves
6505 allowed. By setting this flag certain anti-dependences edges will be
6506 deleted which will trigger the generation of reg-moves based on the
6507 life-range analysis. This option is effective only with
6508 @option{-fmodulo-sched} enabled.
6510 @item -fno-branch-count-reg
6511 @opindex fno-branch-count-reg
6512 Do not use ``decrement and branch'' instructions on a count register,
6513 but instead generate a sequence of instructions that decrement a
6514 register, compare it against zero, then branch based upon the result.
6515 This option is only meaningful on architectures that support such
6516 instructions, which include x86, PowerPC, IA-64 and S/390.
6518 The default is @option{-fbranch-count-reg}.
6520 @item -fno-function-cse
6521 @opindex fno-function-cse
6522 Do not put function addresses in registers; make each instruction that
6523 calls a constant function contain the function's address explicitly.
6525 This option results in less efficient code, but some strange hacks
6526 that alter the assembler output may be confused by the optimizations
6527 performed when this option is not used.
6529 The default is @option{-ffunction-cse}
6531 @item -fno-zero-initialized-in-bss
6532 @opindex fno-zero-initialized-in-bss
6533 If the target supports a BSS section, GCC by default puts variables that
6534 are initialized to zero into BSS@. This can save space in the resulting
6537 This option turns off this behavior because some programs explicitly
6538 rely on variables going to the data section. E.g., so that the
6539 resulting executable can find the beginning of that section and/or make
6540 assumptions based on that.
6542 The default is @option{-fzero-initialized-in-bss}.
6544 @item -fmudflap -fmudflapth -fmudflapir
6548 @cindex bounds checking
6550 For front-ends that support it (C and C++), instrument all risky
6551 pointer/array dereferencing operations, some standard library
6552 string/heap functions, and some other associated constructs with
6553 range/validity tests. Modules so instrumented should be immune to
6554 buffer overflows, invalid heap use, and some other classes of C/C++
6555 programming errors. The instrumentation relies on a separate runtime
6556 library (@file{libmudflap}), which will be linked into a program if
6557 @option{-fmudflap} is given at link time. Run-time behavior of the
6558 instrumented program is controlled by the @env{MUDFLAP_OPTIONS}
6559 environment variable. See @code{env MUDFLAP_OPTIONS=-help a.out}
6562 Use @option{-fmudflapth} instead of @option{-fmudflap} to compile and to
6563 link if your program is multi-threaded. Use @option{-fmudflapir}, in
6564 addition to @option{-fmudflap} or @option{-fmudflapth}, if
6565 instrumentation should ignore pointer reads. This produces less
6566 instrumentation (and therefore faster execution) and still provides
6567 some protection against outright memory corrupting writes, but allows
6568 erroneously read data to propagate within a program.
6570 @item -fthread-jumps
6571 @opindex fthread-jumps
6572 Perform optimizations where we check to see if a jump branches to a
6573 location where another comparison subsumed by the first is found. If
6574 so, the first branch is redirected to either the destination of the
6575 second branch or a point immediately following it, depending on whether
6576 the condition is known to be true or false.
6578 Enabled at levels @option{-O2}, @option{-O3}, @option{-Os}.
6580 @item -fsplit-wide-types
6581 @opindex fsplit-wide-types
6582 When using a type that occupies multiple registers, such as @code{long
6583 long} on a 32-bit system, split the registers apart and allocate them
6584 independently. This normally generates better code for those types,
6585 but may make debugging more difficult.
6587 Enabled at levels @option{-O}, @option{-O2}, @option{-O3},
6590 @item -fcse-follow-jumps
6591 @opindex fcse-follow-jumps
6592 In common subexpression elimination (CSE), scan through jump instructions
6593 when the target of the jump is not reached by any other path. For
6594 example, when CSE encounters an @code{if} statement with an
6595 @code{else} clause, CSE will follow the jump when the condition
6598 Enabled at levels @option{-O2}, @option{-O3}, @option{-Os}.
6600 @item -fcse-skip-blocks
6601 @opindex fcse-skip-blocks
6602 This is similar to @option{-fcse-follow-jumps}, but causes CSE to
6603 follow jumps that conditionally skip over blocks. When CSE
6604 encounters a simple @code{if} statement with no else clause,
6605 @option{-fcse-skip-blocks} causes CSE to follow the jump around the
6606 body of the @code{if}.
6608 Enabled at levels @option{-O2}, @option{-O3}, @option{-Os}.
6610 @item -frerun-cse-after-loop
6611 @opindex frerun-cse-after-loop
6612 Re-run common subexpression elimination after loop optimizations has been
6615 Enabled at levels @option{-O2}, @option{-O3}, @option{-Os}.
6619 Perform a global common subexpression elimination pass.
6620 This pass also performs global constant and copy propagation.
6622 @emph{Note:} When compiling a program using computed gotos, a GCC
6623 extension, you may get better run-time performance if you disable
6624 the global common subexpression elimination pass by adding
6625 @option{-fno-gcse} to the command line.
6627 Enabled at levels @option{-O2}, @option{-O3}, @option{-Os}.
6631 When @option{-fgcse-lm} is enabled, global common subexpression elimination will
6632 attempt to move loads that are only killed by stores into themselves. This
6633 allows a loop containing a load/store sequence to be changed to a load outside
6634 the loop, and a copy/store within the loop.
6636 Enabled by default when gcse is enabled.
6640 When @option{-fgcse-sm} is enabled, a store motion pass is run after
6641 global common subexpression elimination. This pass will attempt to move
6642 stores out of loops. When used in conjunction with @option{-fgcse-lm},
6643 loops containing a load/store sequence can be changed to a load before
6644 the loop and a store after the loop.
6646 Not enabled at any optimization level.
6650 When @option{-fgcse-las} is enabled, the global common subexpression
6651 elimination pass eliminates redundant loads that come after stores to the
6652 same memory location (both partial and full redundancies).
6654 Not enabled at any optimization level.
6656 @item -fgcse-after-reload
6657 @opindex fgcse-after-reload
6658 When @option{-fgcse-after-reload} is enabled, a redundant load elimination
6659 pass is performed after reload. The purpose of this pass is to cleanup
6662 @item -funsafe-loop-optimizations
6663 @opindex funsafe-loop-optimizations
6664 If given, the loop optimizer will assume that loop indices do not
6665 overflow, and that the loops with nontrivial exit condition are not
6666 infinite. This enables a wider range of loop optimizations even if
6667 the loop optimizer itself cannot prove that these assumptions are valid.
6668 Using @option{-Wunsafe-loop-optimizations}, the compiler will warn you
6669 if it finds this kind of loop.
6671 @item -fcrossjumping
6672 @opindex fcrossjumping
6673 Perform cross-jumping transformation. This transformation unifies equivalent code and save code size. The
6674 resulting code may or may not perform better than without cross-jumping.
6676 Enabled at levels @option{-O2}, @option{-O3}, @option{-Os}.
6678 @item -fauto-inc-dec
6679 @opindex fauto-inc-dec
6680 Combine increments or decrements of addresses with memory accesses.
6681 This pass is always skipped on architectures that do not have
6682 instructions to support this. Enabled by default at @option{-O} and
6683 higher on architectures that support this.
6687 Perform dead code elimination (DCE) on RTL@.
6688 Enabled by default at @option{-O} and higher.
6692 Perform dead store elimination (DSE) on RTL@.
6693 Enabled by default at @option{-O} and higher.
6695 @item -fif-conversion
6696 @opindex fif-conversion
6697 Attempt to transform conditional jumps into branch-less equivalents. This
6698 include use of conditional moves, min, max, set flags and abs instructions, and
6699 some tricks doable by standard arithmetics. The use of conditional execution
6700 on chips where it is available is controlled by @code{if-conversion2}.
6702 Enabled at levels @option{-O}, @option{-O2}, @option{-O3}, @option{-Os}.
6704 @item -fif-conversion2
6705 @opindex fif-conversion2
6706 Use conditional execution (where available) to transform conditional jumps into
6707 branch-less equivalents.
6709 Enabled at levels @option{-O}, @option{-O2}, @option{-O3}, @option{-Os}.
6711 @item -fdelete-null-pointer-checks
6712 @opindex fdelete-null-pointer-checks
6713 Assume that programs cannot safely dereference null pointers, and that
6714 no code or data element resides there. This enables simple constant
6715 folding optimizations at all optimization levels. In addition, other
6716 optimization passes in GCC use this flag to control global dataflow
6717 analyses that eliminate useless checks for null pointers; these assume
6718 that if a pointer is checked after it has already been dereferenced,
6721 Note however that in some environments this assumption is not true.
6722 Use @option{-fno-delete-null-pointer-checks} to disable this optimization
6723 for programs that depend on that behavior.
6725 Some targets, especially embedded ones, disable this option at all levels.
6726 Otherwise it is enabled at all levels: @option{-O0}, @option{-O1},
6727 @option{-O2}, @option{-O3}, @option{-Os}. Passes that use the information
6728 are enabled independently at different optimization levels.
6730 @item -fdevirtualize
6731 @opindex fdevirtualize
6732 Attempt to convert calls to virtual functions to direct calls. This
6733 is done both within a procedure and interprocedurally as part of
6734 indirect inlining (@code{-findirect-inlining}) and interprocedural constant
6735 propagation (@option{-fipa-cp}).
6736 Enabled at levels @option{-O2}, @option{-O3}, @option{-Os}.
6738 @item -fexpensive-optimizations
6739 @opindex fexpensive-optimizations
6740 Perform a number of minor optimizations that are relatively expensive.
6742 Enabled at levels @option{-O2}, @option{-O3}, @option{-Os}.
6746 Attempt to remove redundant extension instructions. This is especially
6747 helpful for the x86-64 architecture which implicitly zero-extends in 64-bit
6748 registers after writing to their lower 32-bit half.
6750 Enabled for x86 at levels @option{-O2}, @option{-O3}.
6752 @item -foptimize-register-move
6754 @opindex foptimize-register-move
6756 Attempt to reassign register numbers in move instructions and as
6757 operands of other simple instructions in order to maximize the amount of
6758 register tying. This is especially helpful on machines with two-operand
6761 Note @option{-fregmove} and @option{-foptimize-register-move} are the same
6764 Enabled at levels @option{-O2}, @option{-O3}, @option{-Os}.
6766 @item -fira-algorithm=@var{algorithm}
6767 Use the specified coloring algorithm for the integrated register
6768 allocator. The @var{algorithm} argument can be @samp{priority}, which
6769 specifies Chow's priority coloring, or @samp{CB}, which specifies
6770 Chaitin-Briggs coloring. Chaitin-Briggs coloring is not implemented
6771 for all architectures, but for those targets that do support it, it is
6772 the default because it generates better code.
6774 @item -fira-region=@var{region}
6775 Use specified regions for the integrated register allocator. The
6776 @var{region} argument should be one of the following:
6781 Use all loops as register allocation regions.
6782 This can give the best results for machines with a small and/or
6783 irregular register set.
6786 Use all loops except for loops with small register pressure
6787 as the regions. This value usually gives
6788 the best results in most cases and for most architectures,
6789 and is enabled by default when compiling with optimization for speed
6790 (@option{-O}, @option{-O2}, @dots{}).
6793 Use all functions as a single region.
6794 This typically results in the smallest code size, and is enabled by default for
6795 @option{-Os} or @option{-O0}.
6799 @item -fira-loop-pressure
6800 @opindex fira-loop-pressure
6801 Use IRA to evaluate register pressure in loops for decisions to move
6802 loop invariants. This option usually results in generation
6803 of faster and smaller code on machines with large register files (>= 32
6804 registers), but it can slow the compiler down.
6806 This option is enabled at level @option{-O3} for some targets.
6808 @item -fno-ira-share-save-slots
6809 @opindex fno-ira-share-save-slots
6810 Disable sharing of stack slots used for saving call-used hard
6811 registers living through a call. Each hard register gets a
6812 separate stack slot, and as a result function stack frames are
6815 @item -fno-ira-share-spill-slots
6816 @opindex fno-ira-share-spill-slots
6817 Disable sharing of stack slots allocated for pseudo-registers. Each
6818 pseudo-register that does not get a hard register gets a separate
6819 stack slot, and as a result function stack frames are larger.
6821 @item -fira-verbose=@var{n}
6822 @opindex fira-verbose
6823 Control the verbosity of the dump file for the integrated register allocator.
6824 The default value is 5. If the value @var{n} is greater or equal to 10,
6825 the dump output is sent to stderr using the same format as @var{n} minus 10.
6827 @item -fdelayed-branch
6828 @opindex fdelayed-branch
6829 If supported for the target machine, attempt to reorder instructions
6830 to exploit instruction slots available after delayed branch
6833 Enabled at levels @option{-O}, @option{-O2}, @option{-O3}, @option{-Os}.
6835 @item -fschedule-insns
6836 @opindex fschedule-insns
6837 If supported for the target machine, attempt to reorder instructions to
6838 eliminate execution stalls due to required data being unavailable. This
6839 helps machines that have slow floating point or memory load instructions
6840 by allowing other instructions to be issued until the result of the load
6841 or floating-point instruction is required.
6843 Enabled at levels @option{-O2}, @option{-O3}.
6845 @item -fschedule-insns2
6846 @opindex fschedule-insns2
6847 Similar to @option{-fschedule-insns}, but requests an additional pass of
6848 instruction scheduling after register allocation has been done. This is
6849 especially useful on machines with a relatively small number of
6850 registers and where memory load instructions take more than one cycle.
6852 Enabled at levels @option{-O2}, @option{-O3}, @option{-Os}.
6854 @item -fno-sched-interblock
6855 @opindex fno-sched-interblock
6856 Don't schedule instructions across basic blocks. This is normally
6857 enabled by default when scheduling before register allocation, i.e.@:
6858 with @option{-fschedule-insns} or at @option{-O2} or higher.
6860 @item -fno-sched-spec
6861 @opindex fno-sched-spec
6862 Don't allow speculative motion of non-load instructions. This is normally
6863 enabled by default when scheduling before register allocation, i.e.@:
6864 with @option{-fschedule-insns} or at @option{-O2} or higher.
6866 @item -fsched-pressure
6867 @opindex fsched-pressure
6868 Enable register pressure sensitive insn scheduling before the register
6869 allocation. This only makes sense when scheduling before register
6870 allocation is enabled, i.e.@: with @option{-fschedule-insns} or at
6871 @option{-O2} or higher. Usage of this option can improve the
6872 generated code and decrease its size by preventing register pressure
6873 increase above the number of available hard registers and as a
6874 consequence register spills in the register allocation.
6876 @item -fsched-spec-load
6877 @opindex fsched-spec-load
6878 Allow speculative motion of some load instructions. This only makes
6879 sense when scheduling before register allocation, i.e.@: with
6880 @option{-fschedule-insns} or at @option{-O2} or higher.
6882 @item -fsched-spec-load-dangerous
6883 @opindex fsched-spec-load-dangerous
6884 Allow speculative motion of more load instructions. This only makes
6885 sense when scheduling before register allocation, i.e.@: with
6886 @option{-fschedule-insns} or at @option{-O2} or higher.
6888 @item -fsched-stalled-insns
6889 @itemx -fsched-stalled-insns=@var{n}
6890 @opindex fsched-stalled-insns
6891 Define how many insns (if any) can be moved prematurely from the queue
6892 of stalled insns into the ready list, during the second scheduling pass.
6893 @option{-fno-sched-stalled-insns} means that no insns will be moved
6894 prematurely, @option{-fsched-stalled-insns=0} means there is no limit
6895 on how many queued insns can be moved prematurely.
6896 @option{-fsched-stalled-insns} without a value is equivalent to
6897 @option{-fsched-stalled-insns=1}.
6899 @item -fsched-stalled-insns-dep
6900 @itemx -fsched-stalled-insns-dep=@var{n}
6901 @opindex fsched-stalled-insns-dep
6902 Define how many insn groups (cycles) will be examined for a dependency
6903 on a stalled insn that is candidate for premature removal from the queue
6904 of stalled insns. This has an effect only during the second scheduling pass,
6905 and only if @option{-fsched-stalled-insns} is used.
6906 @option{-fno-sched-stalled-insns-dep} is equivalent to
6907 @option{-fsched-stalled-insns-dep=0}.
6908 @option{-fsched-stalled-insns-dep} without a value is equivalent to
6909 @option{-fsched-stalled-insns-dep=1}.
6911 @item -fsched2-use-superblocks
6912 @opindex fsched2-use-superblocks
6913 When scheduling after register allocation, do use superblock scheduling
6914 algorithm. Superblock scheduling allows motion across basic block boundaries
6915 resulting on faster schedules. This option is experimental, as not all machine
6916 descriptions used by GCC model the CPU closely enough to avoid unreliable
6917 results from the algorithm.
6919 This only makes sense when scheduling after register allocation, i.e.@: with
6920 @option{-fschedule-insns2} or at @option{-O2} or higher.
6922 @item -fsched-group-heuristic
6923 @opindex fsched-group-heuristic
6924 Enable the group heuristic in the scheduler. This heuristic favors
6925 the instruction that belongs to a schedule group. This is enabled
6926 by default when scheduling is enabled, i.e.@: with @option{-fschedule-insns}
6927 or @option{-fschedule-insns2} or at @option{-O2} or higher.
6929 @item -fsched-critical-path-heuristic
6930 @opindex fsched-critical-path-heuristic
6931 Enable the critical-path heuristic in the scheduler. This heuristic favors
6932 instructions on the critical path. This is enabled by default when
6933 scheduling is enabled, i.e.@: with @option{-fschedule-insns}
6934 or @option{-fschedule-insns2} or at @option{-O2} or higher.
6936 @item -fsched-spec-insn-heuristic
6937 @opindex fsched-spec-insn-heuristic
6938 Enable the speculative instruction heuristic in the scheduler. This
6939 heuristic favors speculative instructions with greater dependency weakness.
6940 This is enabled by default when scheduling is enabled, i.e.@:
6941 with @option{-fschedule-insns} or @option{-fschedule-insns2}
6942 or at @option{-O2} or higher.
6944 @item -fsched-rank-heuristic
6945 @opindex fsched-rank-heuristic
6946 Enable the rank heuristic in the scheduler. This heuristic favors
6947 the instruction belonging to a basic block with greater size or frequency.
6948 This is enabled by default when scheduling is enabled, i.e.@:
6949 with @option{-fschedule-insns} or @option{-fschedule-insns2} or
6950 at @option{-O2} or higher.
6952 @item -fsched-last-insn-heuristic
6953 @opindex fsched-last-insn-heuristic
6954 Enable the last-instruction heuristic in the scheduler. This heuristic
6955 favors the instruction that is less dependent on the last instruction
6956 scheduled. This is enabled by default when scheduling is enabled,
6957 i.e.@: with @option{-fschedule-insns} or @option{-fschedule-insns2} or
6958 at @option{-O2} or higher.
6960 @item -fsched-dep-count-heuristic
6961 @opindex fsched-dep-count-heuristic
6962 Enable the dependent-count heuristic in the scheduler. This heuristic
6963 favors the instruction that has more instructions depending on it.
6964 This is enabled by default when scheduling is enabled, i.e.@:
6965 with @option{-fschedule-insns} or @option{-fschedule-insns2} or
6966 at @option{-O2} or higher.
6968 @item -freschedule-modulo-scheduled-loops
6969 @opindex freschedule-modulo-scheduled-loops
6970 The modulo scheduling comes before the traditional scheduling, if a loop
6971 was modulo scheduled we may want to prevent the later scheduling passes
6972 from changing its schedule, we use this option to control that.
6974 @item -fselective-scheduling
6975 @opindex fselective-scheduling
6976 Schedule instructions using selective scheduling algorithm. Selective
6977 scheduling runs instead of the first scheduler pass.
6979 @item -fselective-scheduling2
6980 @opindex fselective-scheduling2
6981 Schedule instructions using selective scheduling algorithm. Selective
6982 scheduling runs instead of the second scheduler pass.
6984 @item -fsel-sched-pipelining
6985 @opindex fsel-sched-pipelining
6986 Enable software pipelining of innermost loops during selective scheduling.
6987 This option has no effect until one of @option{-fselective-scheduling} or
6988 @option{-fselective-scheduling2} is turned on.
6990 @item -fsel-sched-pipelining-outer-loops
6991 @opindex fsel-sched-pipelining-outer-loops
6992 When pipelining loops during selective scheduling, also pipeline outer loops.
6993 This option has no effect until @option{-fsel-sched-pipelining} is turned on.
6996 @opindex fshrink-wrap
6997 Emit function prologues only before parts of the function that need it,
6998 rather than at the top of the function. This flag is enabled by default at
6999 @option{-O} and higher.
7001 @item -fcaller-saves
7002 @opindex fcaller-saves
7003 Enable values to be allocated in registers that will be clobbered by
7004 function calls, by emitting extra instructions to save and restore the
7005 registers around such calls. Such allocation is done only when it
7006 seems to result in better code than would otherwise be produced.
7008 This option is always enabled by default on certain machines, usually
7009 those which have no call-preserved registers to use instead.
7011 Enabled at levels @option{-O2}, @option{-O3}, @option{-Os}.
7013 @item -fcombine-stack-adjustments
7014 @opindex fcombine-stack-adjustments
7015 Tracks stack adjustments (pushes and pops) and stack memory references
7016 and then tries to find ways to combine them.
7018 Enabled by default at @option{-O1} and higher.
7020 @item -fconserve-stack
7021 @opindex fconserve-stack
7022 Attempt to minimize stack usage. The compiler will attempt to use less
7023 stack space, even if that makes the program slower. This option
7024 implies setting the @option{large-stack-frame} parameter to 100
7025 and the @option{large-stack-frame-growth} parameter to 400.
7027 @item -ftree-reassoc
7028 @opindex ftree-reassoc
7029 Perform reassociation on trees. This flag is enabled by default
7030 at @option{-O} and higher.
7034 Perform partial redundancy elimination (PRE) on trees. This flag is
7035 enabled by default at @option{-O2} and @option{-O3}.
7037 @item -ftree-forwprop
7038 @opindex ftree-forwprop
7039 Perform forward propagation on trees. This flag is enabled by default
7040 at @option{-O} and higher.
7044 Perform full redundancy elimination (FRE) on trees. The difference
7045 between FRE and PRE is that FRE only considers expressions
7046 that are computed on all paths leading to the redundant computation.
7047 This analysis is faster than PRE, though it exposes fewer redundancies.
7048 This flag is enabled by default at @option{-O} and higher.
7050 @item -ftree-phiprop
7051 @opindex ftree-phiprop
7052 Perform hoisting of loads from conditional pointers on trees. This
7053 pass is enabled by default at @option{-O} and higher.
7055 @item -ftree-copy-prop
7056 @opindex ftree-copy-prop
7057 Perform copy propagation on trees. This pass eliminates unnecessary
7058 copy operations. This flag is enabled by default at @option{-O} and
7061 @item -fipa-pure-const
7062 @opindex fipa-pure-const
7063 Discover which functions are pure or constant.
7064 Enabled by default at @option{-O} and higher.
7066 @item -fipa-reference
7067 @opindex fipa-reference
7068 Discover which static variables do not escape cannot escape the
7070 Enabled by default at @option{-O} and higher.
7074 Perform interprocedural pointer analysis and interprocedural modification
7075 and reference analysis. This option can cause excessive memory and
7076 compile-time usage on large compilation units. It is not enabled by
7077 default at any optimization level.
7080 @opindex fipa-profile
7081 Perform interprocedural profile propagation. The functions called only from
7082 cold functions are marked as cold. Also functions executed once (such as
7083 @code{cold}, @code{noreturn}, static constructors or destructors) are identified. Cold
7084 functions and loop less parts of functions executed once are then optimized for
7086 Enabled by default at @option{-O} and higher.
7090 Perform interprocedural constant propagation.
7091 This optimization analyzes the program to determine when values passed
7092 to functions are constants and then optimizes accordingly.
7093 This optimization can substantially increase performance
7094 if the application has constants passed to functions.
7095 This flag is enabled by default at @option{-O2}, @option{-Os} and @option{-O3}.
7097 @item -fipa-cp-clone
7098 @opindex fipa-cp-clone
7099 Perform function cloning to make interprocedural constant propagation stronger.
7100 When enabled, interprocedural constant propagation will perform function cloning
7101 when externally visible function can be called with constant arguments.
7102 Because this optimization can create multiple copies of functions,
7103 it may significantly increase code size
7104 (see @option{--param ipcp-unit-growth=@var{value}}).
7105 This flag is enabled by default at @option{-O3}.
7107 @item -fipa-matrix-reorg
7108 @opindex fipa-matrix-reorg
7109 Perform matrix flattening and transposing.
7110 Matrix flattening tries to replace an @math{m}-dimensional matrix
7111 with its equivalent @math{n}-dimensional matrix, where @math{n < m}.
7112 This reduces the level of indirection needed for accessing the elements
7113 of the matrix. The second optimization is matrix transposing, which
7114 attempts to change the order of the matrix's dimensions in order to
7115 improve cache locality.
7116 Both optimizations need the @option{-fwhole-program} flag.
7117 Transposing is enabled only if profiling information is available.
7121 Perform forward store motion on trees. This flag is
7122 enabled by default at @option{-O} and higher.
7124 @item -ftree-bit-ccp
7125 @opindex ftree-bit-ccp
7126 Perform sparse conditional bit constant propagation on trees and propagate
7127 pointer alignment information.
7128 This pass only operates on local scalar variables and is enabled by default
7129 at @option{-O} and higher. It requires that @option{-ftree-ccp} is enabled.
7133 Perform sparse conditional constant propagation (CCP) on trees. This
7134 pass only operates on local scalar variables and is enabled by default
7135 at @option{-O} and higher.
7137 @item -ftree-switch-conversion
7138 Perform conversion of simple initializations in a switch to
7139 initializations from a scalar array. This flag is enabled by default
7140 at @option{-O2} and higher.
7142 @item -ftree-tail-merge
7143 Look for identical code sequences. When found, replace one with a jump to the
7144 other. This optimization is known as tail merging or cross jumping. This flag
7145 is enabled by default at @option{-O2} and higher. The compilation time
7147 be limited using @option{max-tail-merge-comparisons} parameter and
7148 @option{max-tail-merge-iterations} parameter.
7152 Perform dead code elimination (DCE) on trees. This flag is enabled by
7153 default at @option{-O} and higher.
7155 @item -ftree-builtin-call-dce
7156 @opindex ftree-builtin-call-dce
7157 Perform conditional dead code elimination (DCE) for calls to builtin functions
7158 that may set @code{errno} but are otherwise side-effect free. This flag is
7159 enabled by default at @option{-O2} and higher if @option{-Os} is not also
7162 @item -ftree-dominator-opts
7163 @opindex ftree-dominator-opts
7164 Perform a variety of simple scalar cleanups (constant/copy
7165 propagation, redundancy elimination, range propagation and expression
7166 simplification) based on a dominator tree traversal. This also
7167 performs jump threading (to reduce jumps to jumps). This flag is
7168 enabled by default at @option{-O} and higher.
7172 Perform dead store elimination (DSE) on trees. A dead store is a store into
7173 a memory location that is later overwritten by another store without
7174 any intervening loads. In this case the earlier store can be deleted. This
7175 flag is enabled by default at @option{-O} and higher.
7179 Perform loop header copying on trees. This is beneficial since it increases
7180 effectiveness of code motion optimizations. It also saves one jump. This flag
7181 is enabled by default at @option{-O} and higher. It is not enabled
7182 for @option{-Os}, since it usually increases code size.
7184 @item -ftree-loop-optimize
7185 @opindex ftree-loop-optimize
7186 Perform loop optimizations on trees. This flag is enabled by default
7187 at @option{-O} and higher.
7189 @item -ftree-loop-linear
7190 @opindex ftree-loop-linear
7191 Perform loop interchange transformations on tree. Same as
7192 @option{-floop-interchange}. To use this code transformation, GCC has
7193 to be configured with @option{--with-ppl} and @option{--with-cloog} to
7194 enable the Graphite loop transformation infrastructure.
7196 @item -floop-interchange
7197 @opindex floop-interchange
7198 Perform loop interchange transformations on loops. Interchanging two
7199 nested loops switches the inner and outer loops. For example, given a
7204 A(J, I) = A(J, I) * C
7208 loop interchange will transform the loop as if the user had written:
7212 A(J, I) = A(J, I) * C
7216 which can be beneficial when @code{N} is larger than the caches,
7217 because in Fortran, the elements of an array are stored in memory
7218 contiguously by column, and the original loop iterates over rows,
7219 potentially creating at each access a cache miss. This optimization
7220 applies to all the languages supported by GCC and is not limited to
7221 Fortran. To use this code transformation, GCC has to be configured
7222 with @option{--with-ppl} and @option{--with-cloog} to enable the
7223 Graphite loop transformation infrastructure.
7225 @item -floop-strip-mine
7226 @opindex floop-strip-mine
7227 Perform loop strip mining transformations on loops. Strip mining
7228 splits a loop into two nested loops. The outer loop has strides
7229 equal to the strip size and the inner loop has strides of the
7230 original loop within a strip. The strip length can be changed
7231 using the @option{loop-block-tile-size} parameter. For example,
7238 loop strip mining will transform the loop as if the user had written:
7241 DO I = II, min (II + 50, N)
7246 This optimization applies to all the languages supported by GCC and is
7247 not limited to Fortran. To use this code transformation, GCC has to
7248 be configured with @option{--with-ppl} and @option{--with-cloog} to
7249 enable the Graphite loop transformation infrastructure.
7252 @opindex floop-block
7253 Perform loop blocking transformations on loops. Blocking strip mines
7254 each loop in the loop nest such that the memory accesses of the
7255 element loops fit inside caches. The strip length can be changed
7256 using the @option{loop-block-tile-size} parameter. For example, given
7261 A(J, I) = B(I) + C(J)
7265 loop blocking will transform the loop as if the user had written:
7269 DO I = II, min (II + 50, N)
7270 DO J = JJ, min (JJ + 50, M)
7271 A(J, I) = B(I) + C(J)
7277 which can be beneficial when @code{M} is larger than the caches,
7278 because the innermost loop will iterate over a smaller amount of data
7279 which can be kept in the caches. This optimization applies to all the
7280 languages supported by GCC and is not limited to Fortran. To use this
7281 code transformation, GCC has to be configured with @option{--with-ppl}
7282 and @option{--with-cloog} to enable the Graphite loop transformation
7285 @item -fgraphite-identity
7286 @opindex fgraphite-identity
7287 Enable the identity transformation for graphite. For every SCoP we generate
7288 the polyhedral representation and transform it back to gimple. Using
7289 @option{-fgraphite-identity} we can check the costs or benefits of the
7290 GIMPLE -> GRAPHITE -> GIMPLE transformation. Some minimal optimizations
7291 are also performed by the code generator CLooG, like index splitting and
7292 dead code elimination in loops.
7294 @item -floop-flatten
7295 @opindex floop-flatten
7296 Removes the loop nesting structure: transforms the loop nest into a
7297 single loop. This transformation can be useful as an enablement
7298 transform for vectorization and parallelization. This feature
7300 To use this code transformation, GCC has to be configured
7301 with @option{--with-ppl} and @option{--with-cloog} to enable the
7302 Graphite loop transformation infrastructure.
7304 @item -floop-parallelize-all
7305 @opindex floop-parallelize-all
7306 Use the Graphite data dependence analysis to identify loops that can
7307 be parallelized. Parallelize all the loops that can be analyzed to
7308 not contain loop carried dependences without checking that it is
7309 profitable to parallelize the loops.
7311 @item -fcheck-data-deps
7312 @opindex fcheck-data-deps
7313 Compare the results of several data dependence analyzers. This option
7314 is used for debugging the data dependence analyzers.
7316 @item -ftree-loop-if-convert
7317 Attempt to transform conditional jumps in the innermost loops to
7318 branch-less equivalents. The intent is to remove control-flow from
7319 the innermost loops in order to improve the ability of the
7320 vectorization pass to handle these loops. This is enabled by default
7321 if vectorization is enabled.
7323 @item -ftree-loop-if-convert-stores
7324 Attempt to also if-convert conditional jumps containing memory writes.
7325 This transformation can be unsafe for multi-threaded programs as it
7326 transforms conditional memory writes into unconditional memory writes.
7329 for (i = 0; i < N; i++)
7333 would be transformed to
7335 for (i = 0; i < N; i++)
7336 A[i] = cond ? expr : A[i];
7338 potentially producing data races.
7340 @item -ftree-loop-distribution
7341 Perform loop distribution. This flag can improve cache performance on
7342 big loop bodies and allow further loop optimizations, like
7343 parallelization or vectorization, to take place. For example, the loop
7360 @item -ftree-loop-distribute-patterns
7361 Perform loop distribution of patterns that can be code generated with
7362 calls to a library. This flag is enabled by default at @option{-O3}.
7364 This pass distributes the initialization loops and generates a call to
7365 memset zero. For example, the loop
7381 and the initialization loop is transformed into a call to memset zero.
7383 @item -ftree-loop-im
7384 @opindex ftree-loop-im
7385 Perform loop invariant motion on trees. This pass moves only invariants that
7386 would be hard to handle at RTL level (function calls, operations that expand to
7387 nontrivial sequences of insns). With @option{-funswitch-loops} it also moves
7388 operands of conditions that are invariant out of the loop, so that we can use
7389 just trivial invariantness analysis in loop unswitching. The pass also includes
7392 @item -ftree-loop-ivcanon
7393 @opindex ftree-loop-ivcanon
7394 Create a canonical counter for number of iterations in loops for which
7395 determining number of iterations requires complicated analysis. Later
7396 optimizations then may determine the number easily. Useful especially
7397 in connection with unrolling.
7401 Perform induction variable optimizations (strength reduction, induction
7402 variable merging and induction variable elimination) on trees.
7404 @item -ftree-parallelize-loops=n
7405 @opindex ftree-parallelize-loops
7406 Parallelize loops, i.e., split their iteration space to run in n threads.
7407 This is only possible for loops whose iterations are independent
7408 and can be arbitrarily reordered. The optimization is only
7409 profitable on multiprocessor machines, for loops that are CPU-intensive,
7410 rather than constrained e.g.@: by memory bandwidth. This option
7411 implies @option{-pthread}, and thus is only supported on targets
7412 that have support for @option{-pthread}.
7416 Perform function-local points-to analysis on trees. This flag is
7417 enabled by default at @option{-O} and higher.
7421 Perform scalar replacement of aggregates. This pass replaces structure
7422 references with scalars to prevent committing structures to memory too
7423 early. This flag is enabled by default at @option{-O} and higher.
7425 @item -ftree-copyrename
7426 @opindex ftree-copyrename
7427 Perform copy renaming on trees. This pass attempts to rename compiler
7428 temporaries to other variables at copy locations, usually resulting in
7429 variable names which more closely resemble the original variables. This flag
7430 is enabled by default at @option{-O} and higher.
7434 Perform temporary expression replacement during the SSA->normal phase. Single
7435 use/single def temporaries are replaced at their use location with their
7436 defining expression. This results in non-GIMPLE code, but gives the expanders
7437 much more complex trees to work on resulting in better RTL generation. This is
7438 enabled by default at @option{-O} and higher.
7440 @item -ftree-vectorize
7441 @opindex ftree-vectorize
7442 Perform loop vectorization on trees. This flag is enabled by default at
7445 @item -ftree-slp-vectorize
7446 @opindex ftree-slp-vectorize
7447 Perform basic block vectorization on trees. This flag is enabled by default at
7448 @option{-O3} and when @option{-ftree-vectorize} is enabled.
7450 @item -ftree-vect-loop-version
7451 @opindex ftree-vect-loop-version
7452 Perform loop versioning when doing loop vectorization on trees. When a loop
7453 appears to be vectorizable except that data alignment or data dependence cannot
7454 be determined at compile time, then vectorized and non-vectorized versions of
7455 the loop are generated along with run-time checks for alignment or dependence
7456 to control which version is executed. This option is enabled by default
7457 except at level @option{-Os} where it is disabled.
7459 @item -fvect-cost-model
7460 @opindex fvect-cost-model
7461 Enable cost model for vectorization.
7465 Perform Value Range Propagation on trees. This is similar to the
7466 constant propagation pass, but instead of values, ranges of values are
7467 propagated. This allows the optimizers to remove unnecessary range
7468 checks like array bound checks and null pointer checks. This is
7469 enabled by default at @option{-O2} and higher. Null pointer check
7470 elimination is only done if @option{-fdelete-null-pointer-checks} is
7475 Perform tail duplication to enlarge superblock size. This transformation
7476 simplifies the control flow of the function allowing other optimizations to do
7479 @item -funroll-loops
7480 @opindex funroll-loops
7481 Unroll loops whose number of iterations can be determined at compile
7482 time or upon entry to the loop. @option{-funroll-loops} implies
7483 @option{-frerun-cse-after-loop}. This option makes code larger,
7484 and may or may not make it run faster.
7486 @item -funroll-all-loops
7487 @opindex funroll-all-loops
7488 Unroll all loops, even if their number of iterations is uncertain when
7489 the loop is entered. This usually makes programs run more slowly.
7490 @option{-funroll-all-loops} implies the same options as
7491 @option{-funroll-loops},
7493 @item -fsplit-ivs-in-unroller
7494 @opindex fsplit-ivs-in-unroller
7495 Enables expressing of values of induction variables in later iterations
7496 of the unrolled loop using the value in the first iteration. This breaks
7497 long dependency chains, thus improving efficiency of the scheduling passes.
7499 Combination of @option{-fweb} and CSE is often sufficient to obtain the
7500 same effect. However in cases the loop body is more complicated than
7501 a single basic block, this is not reliable. It also does not work at all
7502 on some of the architectures due to restrictions in the CSE pass.
7504 This optimization is enabled by default.
7506 @item -fvariable-expansion-in-unroller
7507 @opindex fvariable-expansion-in-unroller
7508 With this option, the compiler will create multiple copies of some
7509 local variables when unrolling a loop which can result in superior code.
7511 @item -fpartial-inlining
7512 @opindex fpartial-inlining
7513 Inline parts of functions. This option has any effect only
7514 when inlining itself is turned on by the @option{-finline-functions}
7515 or @option{-finline-small-functions} options.
7517 Enabled at level @option{-O2}.
7519 @item -fpredictive-commoning
7520 @opindex fpredictive-commoning
7521 Perform predictive commoning optimization, i.e., reusing computations
7522 (especially memory loads and stores) performed in previous
7523 iterations of loops.
7525 This option is enabled at level @option{-O3}.
7527 @item -fprefetch-loop-arrays
7528 @opindex fprefetch-loop-arrays
7529 If supported by the target machine, generate instructions to prefetch
7530 memory to improve the performance of loops that access large arrays.
7532 This option may generate better or worse code; results are highly
7533 dependent on the structure of loops within the source code.
7535 Disabled at level @option{-Os}.
7538 @itemx -fno-peephole2
7539 @opindex fno-peephole
7540 @opindex fno-peephole2
7541 Disable any machine-specific peephole optimizations. The difference
7542 between @option{-fno-peephole} and @option{-fno-peephole2} is in how they
7543 are implemented in the compiler; some targets use one, some use the
7544 other, a few use both.
7546 @option{-fpeephole} is enabled by default.
7547 @option{-fpeephole2} enabled at levels @option{-O2}, @option{-O3}, @option{-Os}.
7549 @item -fno-guess-branch-probability
7550 @opindex fno-guess-branch-probability
7551 Do not guess branch probabilities using heuristics.
7553 GCC will use heuristics to guess branch probabilities if they are
7554 not provided by profiling feedback (@option{-fprofile-arcs}). These
7555 heuristics are based on the control flow graph. If some branch probabilities
7556 are specified by @samp{__builtin_expect}, then the heuristics will be
7557 used to guess branch probabilities for the rest of the control flow graph,
7558 taking the @samp{__builtin_expect} info into account. The interactions
7559 between the heuristics and @samp{__builtin_expect} can be complex, and in
7560 some cases, it may be useful to disable the heuristics so that the effects
7561 of @samp{__builtin_expect} are easier to understand.
7563 The default is @option{-fguess-branch-probability} at levels
7564 @option{-O}, @option{-O2}, @option{-O3}, @option{-Os}.
7566 @item -freorder-blocks
7567 @opindex freorder-blocks
7568 Reorder basic blocks in the compiled function in order to reduce number of
7569 taken branches and improve code locality.
7571 Enabled at levels @option{-O2}, @option{-O3}.
7573 @item -freorder-blocks-and-partition
7574 @opindex freorder-blocks-and-partition
7575 In addition to reordering basic blocks in the compiled function, in order
7576 to reduce number of taken branches, partitions hot and cold basic blocks
7577 into separate sections of the assembly and .o files, to improve
7578 paging and cache locality performance.
7580 This optimization is automatically turned off in the presence of
7581 exception handling, for linkonce sections, for functions with a user-defined
7582 section attribute and on any architecture that does not support named
7585 @item -freorder-functions
7586 @opindex freorder-functions
7587 Reorder functions in the object file in order to
7588 improve code locality. This is implemented by using special
7589 subsections @code{.text.hot} for most frequently executed functions and
7590 @code{.text.unlikely} for unlikely executed functions. Reordering is done by
7591 the linker so object file format must support named sections and linker must
7592 place them in a reasonable way.
7594 Also profile feedback must be available in to make this option effective. See
7595 @option{-fprofile-arcs} for details.
7597 Enabled at levels @option{-O2}, @option{-O3}, @option{-Os}.
7599 @item -fstrict-aliasing
7600 @opindex fstrict-aliasing
7601 Allow the compiler to assume the strictest aliasing rules applicable to
7602 the language being compiled. For C (and C++), this activates
7603 optimizations based on the type of expressions. In particular, an
7604 object of one type is assumed never to reside at the same address as an
7605 object of a different type, unless the types are almost the same. For
7606 example, an @code{unsigned int} can alias an @code{int}, but not a
7607 @code{void*} or a @code{double}. A character type may alias any other
7610 @anchor{Type-punning}Pay special attention to code like this:
7623 The practice of reading from a different union member than the one most
7624 recently written to (called ``type-punning'') is common. Even with
7625 @option{-fstrict-aliasing}, type-punning is allowed, provided the memory
7626 is accessed through the union type. So, the code above will work as
7627 expected. @xref{Structures unions enumerations and bit-fields
7628 implementation}. However, this code might not:
7639 Similarly, access by taking the address, casting the resulting pointer
7640 and dereferencing the result has undefined behavior, even if the cast
7641 uses a union type, e.g.:
7645 return ((union a_union *) &d)->i;
7649 The @option{-fstrict-aliasing} option is enabled at levels
7650 @option{-O2}, @option{-O3}, @option{-Os}.
7652 @item -fstrict-overflow
7653 @opindex fstrict-overflow
7654 Allow the compiler to assume strict signed overflow rules, depending
7655 on the language being compiled. For C (and C++) this means that
7656 overflow when doing arithmetic with signed numbers is undefined, which
7657 means that the compiler may assume that it will not happen. This
7658 permits various optimizations. For example, the compiler will assume
7659 that an expression like @code{i + 10 > i} will always be true for
7660 signed @code{i}. This assumption is only valid if signed overflow is
7661 undefined, as the expression is false if @code{i + 10} overflows when
7662 using twos complement arithmetic. When this option is in effect any
7663 attempt to determine whether an operation on signed numbers will
7664 overflow must be written carefully to not actually involve overflow.
7666 This option also allows the compiler to assume strict pointer
7667 semantics: given a pointer to an object, if adding an offset to that
7668 pointer does not produce a pointer to the same object, the addition is
7669 undefined. This permits the compiler to conclude that @code{p + u >
7670 p} is always true for a pointer @code{p} and unsigned integer
7671 @code{u}. This assumption is only valid because pointer wraparound is
7672 undefined, as the expression is false if @code{p + u} overflows using
7673 twos complement arithmetic.
7675 See also the @option{-fwrapv} option. Using @option{-fwrapv} means
7676 that integer signed overflow is fully defined: it wraps. When
7677 @option{-fwrapv} is used, there is no difference between
7678 @option{-fstrict-overflow} and @option{-fno-strict-overflow} for
7679 integers. With @option{-fwrapv} certain types of overflow are
7680 permitted. For example, if the compiler gets an overflow when doing
7681 arithmetic on constants, the overflowed value can still be used with
7682 @option{-fwrapv}, but not otherwise.
7684 The @option{-fstrict-overflow} option is enabled at levels
7685 @option{-O2}, @option{-O3}, @option{-Os}.
7687 @item -falign-functions
7688 @itemx -falign-functions=@var{n}
7689 @opindex falign-functions
7690 Align the start of functions to the next power-of-two greater than
7691 @var{n}, skipping up to @var{n} bytes. For instance,
7692 @option{-falign-functions=32} aligns functions to the next 32-byte
7693 boundary, but @option{-falign-functions=24} would align to the next
7694 32-byte boundary only if this can be done by skipping 23 bytes or less.
7696 @option{-fno-align-functions} and @option{-falign-functions=1} are
7697 equivalent and mean that functions will not be aligned.
7699 Some assemblers only support this flag when @var{n} is a power of two;
7700 in that case, it is rounded up.
7702 If @var{n} is not specified or is zero, use a machine-dependent default.
7704 Enabled at levels @option{-O2}, @option{-O3}.
7706 @item -falign-labels
7707 @itemx -falign-labels=@var{n}
7708 @opindex falign-labels
7709 Align all branch targets to a power-of-two boundary, skipping up to
7710 @var{n} bytes like @option{-falign-functions}. This option can easily
7711 make code slower, because it must insert dummy operations for when the
7712 branch target is reached in the usual flow of the code.
7714 @option{-fno-align-labels} and @option{-falign-labels=1} are
7715 equivalent and mean that labels will not be aligned.
7717 If @option{-falign-loops} or @option{-falign-jumps} are applicable and
7718 are greater than this value, then their values are used instead.
7720 If @var{n} is not specified or is zero, use a machine-dependent default
7721 which is very likely to be @samp{1}, meaning no alignment.
7723 Enabled at levels @option{-O2}, @option{-O3}.
7726 @itemx -falign-loops=@var{n}
7727 @opindex falign-loops
7728 Align loops to a power-of-two boundary, skipping up to @var{n} bytes
7729 like @option{-falign-functions}. The hope is that the loop will be
7730 executed many times, which will make up for any execution of the dummy
7733 @option{-fno-align-loops} and @option{-falign-loops=1} are
7734 equivalent and mean that loops will not be aligned.
7736 If @var{n} is not specified or is zero, use a machine-dependent default.
7738 Enabled at levels @option{-O2}, @option{-O3}.
7741 @itemx -falign-jumps=@var{n}
7742 @opindex falign-jumps
7743 Align branch targets to a power-of-two boundary, for branch targets
7744 where the targets can only be reached by jumping, skipping up to @var{n}
7745 bytes like @option{-falign-functions}. In this case, no dummy operations
7748 @option{-fno-align-jumps} and @option{-falign-jumps=1} are
7749 equivalent and mean that loops will not be aligned.
7751 If @var{n} is not specified or is zero, use a machine-dependent default.
7753 Enabled at levels @option{-O2}, @option{-O3}.
7755 @item -funit-at-a-time
7756 @opindex funit-at-a-time
7757 This option is left for compatibility reasons. @option{-funit-at-a-time}
7758 has no effect, while @option{-fno-unit-at-a-time} implies
7759 @option{-fno-toplevel-reorder} and @option{-fno-section-anchors}.
7763 @item -fno-toplevel-reorder
7764 @opindex fno-toplevel-reorder
7765 Do not reorder top-level functions, variables, and @code{asm}
7766 statements. Output them in the same order that they appear in the
7767 input file. When this option is used, unreferenced static variables
7768 will not be removed. This option is intended to support existing code
7769 that relies on a particular ordering. For new code, it is better to
7772 Enabled at level @option{-O0}. When disabled explicitly, it also implies
7773 @option{-fno-section-anchors}, which is otherwise enabled at @option{-O0} on some
7778 Constructs webs as commonly used for register allocation purposes and assign
7779 each web individual pseudo register. This allows the register allocation pass
7780 to operate on pseudos directly, but also strengthens several other optimization
7781 passes, such as CSE, loop optimizer and trivial dead code remover. It can,
7782 however, make debugging impossible, since variables will no longer stay in a
7785 Enabled by default with @option{-funroll-loops}.
7787 @item -fwhole-program
7788 @opindex fwhole-program
7789 Assume that the current compilation unit represents the whole program being
7790 compiled. All public functions and variables with the exception of @code{main}
7791 and those merged by attribute @code{externally_visible} become static functions
7792 and in effect are optimized more aggressively by interprocedural optimizers. If @command{gold} is used as the linker plugin, @code{externally_visible} attributes are automatically added to functions (not variable yet due to a current @command{gold} issue) that are accessed outside of LTO objects according to resolution file produced by @command{gold}. For other linkers that cannot generate resolution file, explicit @code{externally_visible} attributes are still necessary.
7793 While this option is equivalent to proper use of the @code{static} keyword for
7794 programs consisting of a single file, in combination with option
7795 @option{-flto} this flag can be used to
7796 compile many smaller scale programs since the functions and variables become
7797 local for the whole combined compilation unit, not for the single source file
7800 This option implies @option{-fwhole-file} for Fortran programs.
7802 @item -flto[=@var{n}]
7804 This option runs the standard link-time optimizer. When invoked
7805 with source code, it generates GIMPLE (one of GCC's internal
7806 representations) and writes it to special ELF sections in the object
7807 file. When the object files are linked together, all the function
7808 bodies are read from these ELF sections and instantiated as if they
7809 had been part of the same translation unit.
7811 To use the link-time optimizer, @option{-flto} needs to be specified at
7812 compile time and during the final link. For example:
7815 gcc -c -O2 -flto foo.c
7816 gcc -c -O2 -flto bar.c
7817 gcc -o myprog -flto -O2 foo.o bar.o
7820 The first two invocations to GCC save a bytecode representation
7821 of GIMPLE into special ELF sections inside @file{foo.o} and
7822 @file{bar.o}. The final invocation reads the GIMPLE bytecode from
7823 @file{foo.o} and @file{bar.o}, merges the two files into a single
7824 internal image, and compiles the result as usual. Since both
7825 @file{foo.o} and @file{bar.o} are merged into a single image, this
7826 causes all the interprocedural analyses and optimizations in GCC to
7827 work across the two files as if they were a single one. This means,
7828 for example, that the inliner is able to inline functions in
7829 @file{bar.o} into functions in @file{foo.o} and vice-versa.
7831 Another (simpler) way to enable link-time optimization is:
7834 gcc -o myprog -flto -O2 foo.c bar.c
7837 The above generates bytecode for @file{foo.c} and @file{bar.c},
7838 merges them together into a single GIMPLE representation and optimizes
7839 them as usual to produce @file{myprog}.
7841 The only important thing to keep in mind is that to enable link-time
7842 optimizations the @option{-flto} flag needs to be passed to both the
7843 compile and the link commands.
7845 To make whole program optimization effective, it is necessary to make
7846 certain whole program assumptions. The compiler needs to know
7847 what functions and variables can be accessed by libraries and runtime
7848 outside of the link-time optimized unit. When supported by the linker,
7849 the linker plugin (see @option{-fuse-linker-plugin}) passes information
7850 to the compiler about used and externally visible symbols. When
7851 the linker plugin is not available, @option{-fwhole-program} should be
7852 used to allow the compiler to make these assumptions, which leads
7853 to more aggressive optimization decisions.
7855 Note that when a file is compiled with @option{-flto}, the generated
7856 object file is larger than a regular object file because it
7857 contains GIMPLE bytecodes and the usual final code. This means that
7858 object files with LTO information can be linked as normal object
7859 files; if @option{-flto} is not passed to the linker, no
7860 interprocedural optimizations are applied.
7862 Additionally, the optimization flags used to compile individual files
7863 are not necessarily related to those used at link time. For instance,
7866 gcc -c -O0 -flto foo.c
7867 gcc -c -O0 -flto bar.c
7868 gcc -o myprog -flto -O3 foo.o bar.o
7871 This produces individual object files with unoptimized assembler
7872 code, but the resulting binary @file{myprog} is optimized at
7873 @option{-O3}. If, instead, the final binary is generated without
7874 @option{-flto}, then @file{myprog} is not optimized.
7876 When producing the final binary with @option{-flto}, GCC only
7877 applies link-time optimizations to those files that contain bytecode.
7878 Therefore, you can mix and match object files and libraries with
7879 GIMPLE bytecodes and final object code. GCC automatically selects
7880 which files to optimize in LTO mode and which files to link without
7883 There are some code generation flags preserved by GCC when
7884 generating bytecodes, as they need to be used during the final link
7885 stage. Currently, the following options are saved into the GIMPLE
7886 bytecode files: @option{-fPIC}, @option{-fcommon} and all the
7887 @option{-m} target flags.
7889 At link time, these options are read in and reapplied. Note that the
7890 current implementation makes no attempt to recognize conflicting
7891 values for these options. If different files have conflicting option
7892 values (e.g., one file is compiled with @option{-fPIC} and another
7893 isn't), the compiler simply uses the last value read from the
7894 bytecode files. It is recommended, then, that you compile all the files
7895 participating in the same link with the same options.
7897 If LTO encounters objects with C linkage declared with incompatible
7898 types in separate translation units to be linked together (undefined
7899 behavior according to ISO C99 6.2.7), a non-fatal diagnostic may be
7900 issued. The behavior is still undefined at run time.
7902 Another feature of LTO is that it is possible to apply interprocedural
7903 optimizations on files written in different languages. This requires
7904 support in the language front end. Currently, the C, C++ and
7905 Fortran front ends are capable of emitting GIMPLE bytecodes, so
7906 something like this should work:
7911 gfortran -c -flto baz.f90
7912 g++ -o myprog -flto -O3 foo.o bar.o baz.o -lgfortran
7915 Notice that the final link is done with @command{g++} to get the C++
7916 runtime libraries and @option{-lgfortran} is added to get the Fortran
7917 runtime libraries. In general, when mixing languages in LTO mode, you
7918 should use the same link command options as when mixing languages in a
7919 regular (non-LTO) compilation; all you need to add is @option{-flto} to
7920 all the compile and link commands.
7922 If object files containing GIMPLE bytecode are stored in a library archive, say
7923 @file{libfoo.a}, it is possible to extract and use them in an LTO link if you
7924 are using a linker with plugin support. To enable this feature, use
7925 the flag @option{-fuse-linker-plugin} at link time:
7928 gcc -o myprog -O2 -flto -fuse-linker-plugin a.o b.o -lfoo
7931 With the linker plugin enabled, the linker extracts the needed
7932 GIMPLE files from @file{libfoo.a} and passes them on to the running GCC
7933 to make them part of the aggregated GIMPLE image to be optimized.
7935 If you are not using a linker with plugin support and/or do not
7936 enable the linker plugin, then the objects inside @file{libfoo.a}
7937 are extracted and linked as usual, but they do not participate
7938 in the LTO optimization process.
7940 Link-time optimizations do not require the presence of the whole program to
7941 operate. If the program does not require any symbols to be exported, it is
7942 possible to combine @option{-flto} and @option{-fwhole-program} to allow
7943 the interprocedural optimizers to use more aggressive assumptions which may
7944 lead to improved optimization opportunities.
7945 Use of @option{-fwhole-program} is not needed when linker plugin is
7946 active (see @option{-fuse-linker-plugin}).
7948 The current implementation of LTO makes no
7949 attempt to generate bytecode that is portable between different
7950 types of hosts. The bytecode files are versioned and there is a
7951 strict version check, so bytecode files generated in one version of
7952 GCC will not work with an older/newer version of GCC.
7954 Link-time optimization does not work well with generation of debugging
7955 information. Combining @option{-flto} with
7956 @option{-g} is currently experimental and expected to produce wrong
7959 If you specify the optional @var{n}, the optimization and code
7960 generation done at link time is executed in parallel using @var{n}
7961 parallel jobs by utilizing an installed @command{make} program. The
7962 environment variable @env{MAKE} may be used to override the program
7963 used. The default value for @var{n} is 1.
7965 You can also specify @option{-flto=jobserver} to use GNU make's
7966 job server mode to determine the number of parallel jobs. This
7967 is useful when the Makefile calling GCC is already executing in parallel.
7968 You must prepend a @samp{+} to the command recipe in the parent Makefile
7969 for this to work. This option likely only works if @env{MAKE} is
7972 This option is disabled by default
7974 @item -flto-partition=@var{alg}
7975 @opindex flto-partition
7976 Specify the partitioning algorithm used by the link-time optimizer.
7977 The value is either @code{1to1} to specify a partitioning mirroring
7978 the original source files or @code{balanced} to specify partitioning
7979 into equally sized chunks (whenever possible). Specifying @code{none}
7980 as an algorithm disables partitioning and streaming completely. The
7981 default value is @code{balanced}.
7983 @item -flto-compression-level=@var{n}
7984 This option specifies the level of compression used for intermediate
7985 language written to LTO object files, and is only meaningful in
7986 conjunction with LTO mode (@option{-flto}). Valid
7987 values are 0 (no compression) to 9 (maximum compression). Values
7988 outside this range are clamped to either 0 or 9. If the option is not
7989 given, a default balanced compression setting is used.
7992 Prints a report with internal details on the workings of the link-time
7993 optimizer. The contents of this report vary from version to version.
7994 It is meant to be useful to GCC developers when processing object
7995 files in LTO mode (via @option{-flto}).
7997 Disabled by default.
7999 @item -fuse-linker-plugin
8000 Enables the use of a linker plugin during link-time optimization. This
8001 option relies on plugin support in the linker, which is available in gold
8002 or in GNU ld 2.21 or newer.
8004 This option enables the extraction of object files with GIMPLE bytecode out
8005 of library archives. This improves the quality of optimization by exposing
8006 more code to the link-time optimizer. This information specifies what
8007 symbols can be accessed externally (by non-LTO object or during dynamic
8008 linking). Resulting code quality improvements on binaries (and shared
8009 libraries that use hidden visibility) are similar to @code{-fwhole-program}.
8010 See @option{-flto} for a description of the effect of this flag and how to
8013 This option is enabled by default when LTO support in GCC is enabled
8014 and GCC was configured for use with
8015 a linker supporting plugins (GNU ld 2.21 or newer or gold).
8017 @item -ffat-lto-objects
8018 @opindex ffat-lto-objects
8019 Fat LTO objects are object files that contain both the intermediate language
8020 and the object code. This makes them usable for both LTO linking and normal
8021 linking. This option is effective only when compiling with @option{-flto}
8022 and is ignored at link time.
8024 @option{-fno-fat-lto-objects} improves compilation time over plain LTO, but
8025 requires the complete toolchain to be aware of LTO. It requires a linker with
8026 linker plugin support for basic functionality. Additionally, nm, ar and ranlib
8027 need to support linker plugins to allow a full-featured build environment
8028 (capable of building static libraries etc).
8030 The default is @option{-ffat-lto-objects} but this default is intended to
8031 change in future releases when linker plugin enabled environments become more
8034 @item -fcompare-elim
8035 @opindex fcompare-elim
8036 After register allocation and post-register allocation instruction splitting,
8037 identify arithmetic instructions that compute processor flags similar to a
8038 comparison operation based on that arithmetic. If possible, eliminate the
8039 explicit comparison operation.
8041 This pass only applies to certain targets that cannot explicitly represent
8042 the comparison operation before register allocation is complete.
8044 Enabled at levels @option{-O}, @option{-O2}, @option{-O3}, @option{-Os}.
8046 @item -fcprop-registers
8047 @opindex fcprop-registers
8048 After register allocation and post-register allocation instruction splitting,
8049 we perform a copy-propagation pass to try to reduce scheduling dependencies
8050 and occasionally eliminate the copy.
8052 Enabled at levels @option{-O}, @option{-O2}, @option{-O3}, @option{-Os}.
8054 @item -fprofile-correction
8055 @opindex fprofile-correction
8056 Profiles collected using an instrumented binary for multi-threaded programs may
8057 be inconsistent due to missed counter updates. When this option is specified,
8058 GCC will use heuristics to correct or smooth out such inconsistencies. By
8059 default, GCC will emit an error message when an inconsistent profile is detected.
8061 @item -fprofile-dir=@var{path}
8062 @opindex fprofile-dir
8064 Set the directory to search for the profile data files in to @var{path}.
8065 This option affects only the profile data generated by
8066 @option{-fprofile-generate}, @option{-ftest-coverage}, @option{-fprofile-arcs}
8067 and used by @option{-fprofile-use} and @option{-fbranch-probabilities}
8068 and its related options. Both absolute and relative paths can be used.
8069 By default, GCC will use the current directory as @var{path}, thus the
8070 profile data file will appear in the same directory as the object file.
8072 @item -fprofile-generate
8073 @itemx -fprofile-generate=@var{path}
8074 @opindex fprofile-generate
8076 Enable options usually used for instrumenting application to produce
8077 profile useful for later recompilation with profile feedback based
8078 optimization. You must use @option{-fprofile-generate} both when
8079 compiling and when linking your program.
8081 The following options are enabled: @code{-fprofile-arcs}, @code{-fprofile-values}, @code{-fvpt}.
8083 If @var{path} is specified, GCC will look at the @var{path} to find
8084 the profile feedback data files. See @option{-fprofile-dir}.
8087 @itemx -fprofile-use=@var{path}
8088 @opindex fprofile-use
8089 Enable profile feedback directed optimizations, and optimizations
8090 generally profitable only with profile feedback available.
8092 The following options are enabled: @code{-fbranch-probabilities}, @code{-fvpt},
8093 @code{-funroll-loops}, @code{-fpeel-loops}, @code{-ftracer}
8095 By default, GCC emits an error message if the feedback profiles do not
8096 match the source code. This error can be turned into a warning by using
8097 @option{-Wcoverage-mismatch}. Note this may result in poorly optimized
8100 If @var{path} is specified, GCC will look at the @var{path} to find
8101 the profile feedback data files. See @option{-fprofile-dir}.
8104 The following options control compiler behavior regarding floating-point
8105 arithmetic. These options trade off between speed and
8106 correctness. All must be specifically enabled.
8110 @opindex ffloat-store
8111 Do not store floating-point variables in registers, and inhibit other
8112 options that might change whether a floating-point value is taken from a
8115 @cindex floating-point precision
8116 This option prevents undesirable excess precision on machines such as
8117 the 68000 where the floating registers (of the 68881) keep more
8118 precision than a @code{double} is supposed to have. Similarly for the
8119 x86 architecture. For most programs, the excess precision does only
8120 good, but a few programs rely on the precise definition of IEEE floating
8121 point. Use @option{-ffloat-store} for such programs, after modifying
8122 them to store all pertinent intermediate computations into variables.
8124 @item -fexcess-precision=@var{style}
8125 @opindex fexcess-precision
8126 This option allows further control over excess precision on machines
8127 where floating-point registers have more precision than the IEEE
8128 @code{float} and @code{double} types and the processor does not
8129 support operations rounding to those types. By default,
8130 @option{-fexcess-precision=fast} is in effect; this means that
8131 operations are carried out in the precision of the registers and that
8132 it is unpredictable when rounding to the types specified in the source
8133 code takes place. When compiling C, if
8134 @option{-fexcess-precision=standard} is specified then excess
8135 precision will follow the rules specified in ISO C99; in particular,
8136 both casts and assignments cause values to be rounded to their
8137 semantic types (whereas @option{-ffloat-store} only affects
8138 assignments). This option is enabled by default for C if a strict
8139 conformance option such as @option{-std=c99} is used.
8142 @option{-fexcess-precision=standard} is not implemented for languages
8143 other than C, and has no effect if
8144 @option{-funsafe-math-optimizations} or @option{-ffast-math} is
8145 specified. On the x86, it also has no effect if @option{-mfpmath=sse}
8146 or @option{-mfpmath=sse+387} is specified; in the former case, IEEE
8147 semantics apply without excess precision, and in the latter, rounding
8152 Sets @option{-fno-math-errno}, @option{-funsafe-math-optimizations},
8153 @option{-ffinite-math-only}, @option{-fno-rounding-math},
8154 @option{-fno-signaling-nans} and @option{-fcx-limited-range}.
8156 This option causes the preprocessor macro @code{__FAST_MATH__} to be defined.
8158 This option is not turned on by any @option{-O} option besides
8159 @option{-Ofast} since it can result in incorrect output for programs
8160 that depend on an exact implementation of IEEE or ISO rules/specifications
8161 for math functions. It may, however, yield faster code for programs
8162 that do not require the guarantees of these specifications.
8164 @item -fno-math-errno
8165 @opindex fno-math-errno
8166 Do not set ERRNO after calling math functions that are executed
8167 with a single instruction, e.g., sqrt. A program that relies on
8168 IEEE exceptions for math error handling may want to use this flag
8169 for speed while maintaining IEEE arithmetic compatibility.
8171 This option is not turned on by any @option{-O} option since
8172 it can result in incorrect output for programs that depend on
8173 an exact implementation of IEEE or ISO rules/specifications for
8174 math functions. It may, however, yield faster code for programs
8175 that do not require the guarantees of these specifications.
8177 The default is @option{-fmath-errno}.
8179 On Darwin systems, the math library never sets @code{errno}. There is
8180 therefore no reason for the compiler to consider the possibility that
8181 it might, and @option{-fno-math-errno} is the default.
8183 @item -funsafe-math-optimizations
8184 @opindex funsafe-math-optimizations
8186 Allow optimizations for floating-point arithmetic that (a) assume
8187 that arguments and results are valid and (b) may violate IEEE or
8188 ANSI standards. When used at link-time, it may include libraries
8189 or startup files that change the default FPU control word or other
8190 similar optimizations.
8192 This option is not turned on by any @option{-O} option since
8193 it can result in incorrect output for programs that depend on
8194 an exact implementation of IEEE or ISO rules/specifications for
8195 math functions. It may, however, yield faster code for programs
8196 that do not require the guarantees of these specifications.
8197 Enables @option{-fno-signed-zeros}, @option{-fno-trapping-math},
8198 @option{-fassociative-math} and @option{-freciprocal-math}.
8200 The default is @option{-fno-unsafe-math-optimizations}.
8202 @item -fassociative-math
8203 @opindex fassociative-math
8205 Allow re-association of operands in series of floating-point operations.
8206 This violates the ISO C and C++ language standard by possibly changing
8207 computation result. NOTE: re-ordering may change the sign of zero as
8208 well as ignore NaNs and inhibit or create underflow or overflow (and
8209 thus cannot be used on code that relies on rounding behavior like
8210 @code{(x + 2**52) - 2**52}. May also reorder floating-point comparisons
8211 and thus may not be used when ordered comparisons are required.
8212 This option requires that both @option{-fno-signed-zeros} and
8213 @option{-fno-trapping-math} be in effect. Moreover, it doesn't make
8214 much sense with @option{-frounding-math}. For Fortran the option
8215 is automatically enabled when both @option{-fno-signed-zeros} and
8216 @option{-fno-trapping-math} are in effect.
8218 The default is @option{-fno-associative-math}.
8220 @item -freciprocal-math
8221 @opindex freciprocal-math
8223 Allow the reciprocal of a value to be used instead of dividing by
8224 the value if this enables optimizations. For example @code{x / y}
8225 can be replaced with @code{x * (1/y)}, which is useful if @code{(1/y)}
8226 is subject to common subexpression elimination. Note that this loses
8227 precision and increases the number of flops operating on the value.
8229 The default is @option{-fno-reciprocal-math}.
8231 @item -ffinite-math-only
8232 @opindex ffinite-math-only
8233 Allow optimizations for floating-point arithmetic that assume
8234 that arguments and results are not NaNs or +-Infs.
8236 This option is not turned on by any @option{-O} option since
8237 it can result in incorrect output for programs that depend on
8238 an exact implementation of IEEE or ISO rules/specifications for
8239 math functions. It may, however, yield faster code for programs
8240 that do not require the guarantees of these specifications.
8242 The default is @option{-fno-finite-math-only}.
8244 @item -fno-signed-zeros
8245 @opindex fno-signed-zeros
8246 Allow optimizations for floating-point arithmetic that ignore the
8247 signedness of zero. IEEE arithmetic specifies the behavior of
8248 distinct +0.0 and @minus{}0.0 values, which then prohibits simplification
8249 of expressions such as x+0.0 or 0.0*x (even with @option{-ffinite-math-only}).
8250 This option implies that the sign of a zero result isn't significant.
8252 The default is @option{-fsigned-zeros}.
8254 @item -fno-trapping-math
8255 @opindex fno-trapping-math
8256 Compile code assuming that floating-point operations cannot generate
8257 user-visible traps. These traps include division by zero, overflow,
8258 underflow, inexact result and invalid operation. This option requires
8259 that @option{-fno-signaling-nans} be in effect. Setting this option may
8260 allow faster code if one relies on ``non-stop'' IEEE arithmetic, for example.
8262 This option should never be turned on by any @option{-O} option since
8263 it can result in incorrect output for programs that depend on
8264 an exact implementation of IEEE or ISO rules/specifications for
8267 The default is @option{-ftrapping-math}.
8269 @item -frounding-math
8270 @opindex frounding-math
8271 Disable transformations and optimizations that assume default floating-point
8272 rounding behavior. This is round-to-zero for all floating point
8273 to integer conversions, and round-to-nearest for all other arithmetic
8274 truncations. This option should be specified for programs that change
8275 the FP rounding mode dynamically, or that may be executed with a
8276 non-default rounding mode. This option disables constant folding of
8277 floating-point expressions at compile time (which may be affected by
8278 rounding mode) and arithmetic transformations that are unsafe in the
8279 presence of sign-dependent rounding modes.
8281 The default is @option{-fno-rounding-math}.
8283 This option is experimental and does not currently guarantee to
8284 disable all GCC optimizations that are affected by rounding mode.
8285 Future versions of GCC may provide finer control of this setting
8286 using C99's @code{FENV_ACCESS} pragma. This command-line option
8287 will be used to specify the default state for @code{FENV_ACCESS}.
8289 @item -fsignaling-nans
8290 @opindex fsignaling-nans
8291 Compile code assuming that IEEE signaling NaNs may generate user-visible
8292 traps during floating-point operations. Setting this option disables
8293 optimizations that may change the number of exceptions visible with
8294 signaling NaNs. This option implies @option{-ftrapping-math}.
8296 This option causes the preprocessor macro @code{__SUPPORT_SNAN__} to
8299 The default is @option{-fno-signaling-nans}.
8301 This option is experimental and does not currently guarantee to
8302 disable all GCC optimizations that affect signaling NaN behavior.
8304 @item -fsingle-precision-constant
8305 @opindex fsingle-precision-constant
8306 Treat floating-point constants as single precision instead of
8307 implicitly converting them to double-precision constants.
8309 @item -fcx-limited-range
8310 @opindex fcx-limited-range
8311 When enabled, this option states that a range reduction step is not
8312 needed when performing complex division. Also, there is no checking
8313 whether the result of a complex multiplication or division is @code{NaN
8314 + I*NaN}, with an attempt to rescue the situation in that case. The
8315 default is @option{-fno-cx-limited-range}, but is enabled by
8316 @option{-ffast-math}.
8318 This option controls the default setting of the ISO C99
8319 @code{CX_LIMITED_RANGE} pragma. Nevertheless, the option applies to
8322 @item -fcx-fortran-rules
8323 @opindex fcx-fortran-rules
8324 Complex multiplication and division follow Fortran rules. Range
8325 reduction is done as part of complex division, but there is no checking
8326 whether the result of a complex multiplication or division is @code{NaN
8327 + I*NaN}, with an attempt to rescue the situation in that case.
8329 The default is @option{-fno-cx-fortran-rules}.
8333 The following options control optimizations that may improve
8334 performance, but are not enabled by any @option{-O} options. This
8335 section includes experimental options that may produce broken code.
8338 @item -fbranch-probabilities
8339 @opindex fbranch-probabilities
8340 After running a program compiled with @option{-fprofile-arcs}
8341 (@pxref{Debugging Options,, Options for Debugging Your Program or
8342 @command{gcc}}), you can compile it a second time using
8343 @option{-fbranch-probabilities}, to improve optimizations based on
8344 the number of times each branch was taken. When the program
8345 compiled with @option{-fprofile-arcs} exits it saves arc execution
8346 counts to a file called @file{@var{sourcename}.gcda} for each source
8347 file. The information in this data file is very dependent on the
8348 structure of the generated code, so you must use the same source code
8349 and the same optimization options for both compilations.
8351 With @option{-fbranch-probabilities}, GCC puts a
8352 @samp{REG_BR_PROB} note on each @samp{JUMP_INSN} and @samp{CALL_INSN}.
8353 These can be used to improve optimization. Currently, they are only
8354 used in one place: in @file{reorg.c}, instead of guessing which path a
8355 branch is most likely to take, the @samp{REG_BR_PROB} values are used to
8356 exactly determine which path is taken more often.
8358 @item -fprofile-values
8359 @opindex fprofile-values
8360 If combined with @option{-fprofile-arcs}, it adds code so that some
8361 data about values of expressions in the program is gathered.
8363 With @option{-fbranch-probabilities}, it reads back the data gathered
8364 from profiling values of expressions for usage in optimizations.
8366 Enabled with @option{-fprofile-generate} and @option{-fprofile-use}.
8370 If combined with @option{-fprofile-arcs}, it instructs the compiler to add
8371 a code to gather information about values of expressions.
8373 With @option{-fbranch-probabilities}, it reads back the data gathered
8374 and actually performs the optimizations based on them.
8375 Currently the optimizations include specialization of division operation
8376 using the knowledge about the value of the denominator.
8378 @item -frename-registers
8379 @opindex frename-registers
8380 Attempt to avoid false dependencies in scheduled code by making use
8381 of registers left over after register allocation. This optimization
8382 will most benefit processors with lots of registers. Depending on the
8383 debug information format adopted by the target, however, it can
8384 make debugging impossible, since variables will no longer stay in
8385 a ``home register''.
8387 Enabled by default with @option{-funroll-loops} and @option{-fpeel-loops}.
8391 Perform tail duplication to enlarge superblock size. This transformation
8392 simplifies the control flow of the function allowing other optimizations to do
8395 Enabled with @option{-fprofile-use}.
8397 @item -funroll-loops
8398 @opindex funroll-loops
8399 Unroll loops whose number of iterations can be determined at compile time or
8400 upon entry to the loop. @option{-funroll-loops} implies
8401 @option{-frerun-cse-after-loop}, @option{-fweb} and @option{-frename-registers}.
8402 It also turns on complete loop peeling (i.e.@: complete removal of loops with
8403 small constant number of iterations). This option makes code larger, and may
8404 or may not make it run faster.
8406 Enabled with @option{-fprofile-use}.
8408 @item -funroll-all-loops
8409 @opindex funroll-all-loops
8410 Unroll all loops, even if their number of iterations is uncertain when
8411 the loop is entered. This usually makes programs run more slowly.
8412 @option{-funroll-all-loops} implies the same options as
8413 @option{-funroll-loops}.
8416 @opindex fpeel-loops
8417 Peels loops for which there is enough information that they do not
8418 roll much (from profile feedback). It also turns on complete loop peeling
8419 (i.e.@: complete removal of loops with small constant number of iterations).
8421 Enabled with @option{-fprofile-use}.
8423 @item -fmove-loop-invariants
8424 @opindex fmove-loop-invariants
8425 Enables the loop invariant motion pass in the RTL loop optimizer. Enabled
8426 at level @option{-O1}
8428 @item -funswitch-loops
8429 @opindex funswitch-loops
8430 Move branches with loop invariant conditions out of the loop, with duplicates
8431 of the loop on both branches (modified according to result of the condition).
8433 @item -ffunction-sections
8434 @itemx -fdata-sections
8435 @opindex ffunction-sections
8436 @opindex fdata-sections
8437 Place each function or data item into its own section in the output
8438 file if the target supports arbitrary sections. The name of the
8439 function or the name of the data item determines the section's name
8442 Use these options on systems where the linker can perform optimizations
8443 to improve locality of reference in the instruction space. Most systems
8444 using the ELF object format and SPARC processors running Solaris 2 have
8445 linkers with such optimizations. AIX may have these optimizations in
8448 Only use these options when there are significant benefits from doing
8449 so. When you specify these options, the assembler and linker will
8450 create larger object and executable files and will also be slower.
8451 You will not be able to use @code{gprof} on all systems if you
8452 specify this option and you may have problems with debugging if
8453 you specify both this option and @option{-g}.
8455 @item -fbranch-target-load-optimize
8456 @opindex fbranch-target-load-optimize
8457 Perform branch target register load optimization before prologue / epilogue
8459 The use of target registers can typically be exposed only during reload,
8460 thus hoisting loads out of loops and doing inter-block scheduling needs
8461 a separate optimization pass.
8463 @item -fbranch-target-load-optimize2
8464 @opindex fbranch-target-load-optimize2
8465 Perform branch target register load optimization after prologue / epilogue
8468 @item -fbtr-bb-exclusive
8469 @opindex fbtr-bb-exclusive
8470 When performing branch target register load optimization, don't reuse
8471 branch target registers in within any basic block.
8473 @item -fstack-protector
8474 @opindex fstack-protector
8475 Emit extra code to check for buffer overflows, such as stack smashing
8476 attacks. This is done by adding a guard variable to functions with
8477 vulnerable objects. This includes functions that call alloca, and
8478 functions with buffers larger than 8 bytes. The guards are initialized
8479 when a function is entered and then checked when the function exits.
8480 If a guard check fails, an error message is printed and the program exits.
8482 @item -fstack-protector-all
8483 @opindex fstack-protector-all
8484 Like @option{-fstack-protector} except that all functions are protected.
8486 @item -fsection-anchors
8487 @opindex fsection-anchors
8488 Try to reduce the number of symbolic address calculations by using
8489 shared ``anchor'' symbols to address nearby objects. This transformation
8490 can help to reduce the number of GOT entries and GOT accesses on some
8493 For example, the implementation of the following function @code{foo}:
8497 int foo (void) @{ return a + b + c; @}
8500 would usually calculate the addresses of all three variables, but if you
8501 compile it with @option{-fsection-anchors}, it will access the variables
8502 from a common anchor point instead. The effect is similar to the
8503 following pseudocode (which isn't valid C):
8508 register int *xr = &x;
8509 return xr[&a - &x] + xr[&b - &x] + xr[&c - &x];
8513 Not all targets support this option.
8515 @item --param @var{name}=@var{value}
8517 In some places, GCC uses various constants to control the amount of
8518 optimization that is done. For example, GCC will not inline functions
8519 that contain more than a certain number of instructions. You can
8520 control some of these constants on the command line using the
8521 @option{--param} option.
8523 The names of specific parameters, and the meaning of the values, are
8524 tied to the internals of the compiler, and are subject to change
8525 without notice in future releases.
8527 In each case, the @var{value} is an integer. The allowable choices for
8528 @var{name} are given in the following table:
8531 @item predictable-branch-outcome
8532 When branch is predicted to be taken with probability lower than this threshold
8533 (in percent), then it is considered well predictable. The default is 10.
8535 @item max-crossjump-edges
8536 The maximum number of incoming edges to consider for crossjumping.
8537 The algorithm used by @option{-fcrossjumping} is @math{O(N^2)} in
8538 the number of edges incoming to each block. Increasing values mean
8539 more aggressive optimization, making the compilation time increase with
8540 probably small improvement in executable size.
8542 @item min-crossjump-insns
8543 The minimum number of instructions that must be matched at the end
8544 of two blocks before crossjumping will be performed on them. This
8545 value is ignored in the case where all instructions in the block being
8546 crossjumped from are matched. The default value is 5.
8548 @item max-grow-copy-bb-insns
8549 The maximum code size expansion factor when copying basic blocks
8550 instead of jumping. The expansion is relative to a jump instruction.
8551 The default value is 8.
8553 @item max-goto-duplication-insns
8554 The maximum number of instructions to duplicate to a block that jumps
8555 to a computed goto. To avoid @math{O(N^2)} behavior in a number of
8556 passes, GCC factors computed gotos early in the compilation process,
8557 and unfactors them as late as possible. Only computed jumps at the
8558 end of a basic blocks with no more than max-goto-duplication-insns are
8559 unfactored. The default value is 8.
8561 @item max-delay-slot-insn-search
8562 The maximum number of instructions to consider when looking for an
8563 instruction to fill a delay slot. If more than this arbitrary number of
8564 instructions is searched, the time savings from filling the delay slot
8565 will be minimal so stop searching. Increasing values mean more
8566 aggressive optimization, making the compilation time increase with probably
8567 small improvement in execution time.
8569 @item max-delay-slot-live-search
8570 When trying to fill delay slots, the maximum number of instructions to
8571 consider when searching for a block with valid live register
8572 information. Increasing this arbitrarily chosen value means more
8573 aggressive optimization, increasing the compilation time. This parameter
8574 should be removed when the delay slot code is rewritten to maintain the
8577 @item max-gcse-memory
8578 The approximate maximum amount of memory that will be allocated in
8579 order to perform the global common subexpression elimination
8580 optimization. If more memory than specified is required, the
8581 optimization will not be done.
8583 @item max-gcse-insertion-ratio
8584 If the ratio of expression insertions to deletions is larger than this value
8585 for any expression, then RTL PRE will insert or remove the expression and thus
8586 leave partially redundant computations in the instruction stream. The default value is 20.
8588 @item max-pending-list-length
8589 The maximum number of pending dependencies scheduling will allow
8590 before flushing the current state and starting over. Large functions
8591 with few branches or calls can create excessively large lists which
8592 needlessly consume memory and resources.
8594 @item max-modulo-backtrack-attempts
8595 The maximum number of backtrack attempts the scheduler should make
8596 when modulo scheduling a loop. Larger values can exponentially increase
8599 @item max-inline-insns-single
8600 Several parameters control the tree inliner used in gcc.
8601 This number sets the maximum number of instructions (counted in GCC's
8602 internal representation) in a single function that the tree inliner
8603 will consider for inlining. This only affects functions declared
8604 inline and methods implemented in a class declaration (C++).
8605 The default value is 400.
8607 @item max-inline-insns-auto
8608 When you use @option{-finline-functions} (included in @option{-O3}),
8609 a lot of functions that would otherwise not be considered for inlining
8610 by the compiler will be investigated. To those functions, a different
8611 (more restrictive) limit compared to functions declared inline can
8613 The default value is 40.
8615 @item large-function-insns
8616 The limit specifying really large functions. For functions larger than this
8617 limit after inlining, inlining is constrained by
8618 @option{--param large-function-growth}. This parameter is useful primarily
8619 to avoid extreme compilation time caused by non-linear algorithms used by the
8621 The default value is 2700.
8623 @item large-function-growth
8624 Specifies maximal growth of large function caused by inlining in percents.
8625 The default value is 100 which limits large function growth to 2.0 times
8628 @item large-unit-insns
8629 The limit specifying large translation unit. Growth caused by inlining of
8630 units larger than this limit is limited by @option{--param inline-unit-growth}.
8631 For small units this might be too tight (consider unit consisting of function A
8632 that is inline and B that just calls A three time. If B is small relative to
8633 A, the growth of unit is 300\% and yet such inlining is very sane. For very
8634 large units consisting of small inlineable functions however the overall unit
8635 growth limit is needed to avoid exponential explosion of code size. Thus for
8636 smaller units, the size is increased to @option{--param large-unit-insns}
8637 before applying @option{--param inline-unit-growth}. The default is 10000
8639 @item inline-unit-growth
8640 Specifies maximal overall growth of the compilation unit caused by inlining.
8641 The default value is 30 which limits unit growth to 1.3 times the original
8644 @item ipcp-unit-growth
8645 Specifies maximal overall growth of the compilation unit caused by
8646 interprocedural constant propagation. The default value is 10 which limits
8647 unit growth to 1.1 times the original size.
8649 @item large-stack-frame
8650 The limit specifying large stack frames. While inlining the algorithm is trying
8651 to not grow past this limit too much. Default value is 256 bytes.
8653 @item large-stack-frame-growth
8654 Specifies maximal growth of large stack frames caused by inlining in percents.
8655 The default value is 1000 which limits large stack frame growth to 11 times
8658 @item max-inline-insns-recursive
8659 @itemx max-inline-insns-recursive-auto
8660 Specifies maximum number of instructions out-of-line copy of self recursive inline
8661 function can grow into by performing recursive inlining.
8663 For functions declared inline @option{--param max-inline-insns-recursive} is
8664 taken into account. For function not declared inline, recursive inlining
8665 happens only when @option{-finline-functions} (included in @option{-O3}) is
8666 enabled and @option{--param max-inline-insns-recursive-auto} is used. The
8667 default value is 450.
8669 @item max-inline-recursive-depth
8670 @itemx max-inline-recursive-depth-auto
8671 Specifies maximum recursion depth used by the recursive inlining.
8673 For functions declared inline @option{--param max-inline-recursive-depth} is
8674 taken into account. For function not declared inline, recursive inlining
8675 happens only when @option{-finline-functions} (included in @option{-O3}) is
8676 enabled and @option{--param max-inline-recursive-depth-auto} is used. The
8679 @item min-inline-recursive-probability
8680 Recursive inlining is profitable only for function having deep recursion
8681 in average and can hurt for function having little recursion depth by
8682 increasing the prologue size or complexity of function body to other
8685 When profile feedback is available (see @option{-fprofile-generate}) the actual
8686 recursion depth can be guessed from probability that function will recurse via
8687 given call expression. This parameter limits inlining only to call expression
8688 whose probability exceeds given threshold (in percents). The default value is
8691 @item early-inlining-insns
8692 Specify growth that early inliner can make. In effect it increases amount of
8693 inlining for code having large abstraction penalty. The default value is 10.
8695 @item max-early-inliner-iterations
8696 @itemx max-early-inliner-iterations
8697 Limit of iterations of early inliner. This basically bounds number of nested
8698 indirect calls early inliner can resolve. Deeper chains are still handled by
8701 @item comdat-sharing-probability
8702 @itemx comdat-sharing-probability
8703 Probability (in percent) that C++ inline function with comdat visibility
8704 will be shared across multiple compilation units. The default value is 20.
8706 @item min-vect-loop-bound
8707 The minimum number of iterations under which a loop will not get vectorized
8708 when @option{-ftree-vectorize} is used. The number of iterations after
8709 vectorization needs to be greater than the value specified by this option
8710 to allow vectorization. The default value is 0.
8712 @item gcse-cost-distance-ratio
8713 Scaling factor in calculation of maximum distance an expression
8714 can be moved by GCSE optimizations. This is currently supported only in the
8715 code hoisting pass. The bigger the ratio, the more aggressive code hoisting
8716 will be with simple expressions, i.e., the expressions that have cost
8717 less than @option{gcse-unrestricted-cost}. Specifying 0 will disable
8718 hoisting of simple expressions. The default value is 10.
8720 @item gcse-unrestricted-cost
8721 Cost, roughly measured as the cost of a single typical machine
8722 instruction, at which GCSE optimizations will not constrain
8723 the distance an expression can travel. This is currently
8724 supported only in the code hoisting pass. The lesser the cost,
8725 the more aggressive code hoisting will be. Specifying 0 will
8726 allow all expressions to travel unrestricted distances.
8727 The default value is 3.
8729 @item max-hoist-depth
8730 The depth of search in the dominator tree for expressions to hoist.
8731 This is used to avoid quadratic behavior in hoisting algorithm.
8732 The value of 0 will avoid limiting the search, but may slow down compilation
8733 of huge functions. The default value is 30.
8735 @item max-tail-merge-comparisons
8736 The maximum amount of similar bbs to compare a bb with. This is used to
8737 avoid quadratic behavior in tree tail merging. The default value is 10.
8739 @item max-tail-merge-iterations
8740 The maximum amount of iterations of the pass over the function. This is used to
8741 limit compilation time in tree tail merging. The default value is 2.
8743 @item max-unrolled-insns
8744 The maximum number of instructions that a loop should have if that loop
8745 is unrolled, and if the loop is unrolled, it determines how many times
8746 the loop code is unrolled.
8748 @item max-average-unrolled-insns
8749 The maximum number of instructions biased by probabilities of their execution
8750 that a loop should have if that loop is unrolled, and if the loop is unrolled,
8751 it determines how many times the loop code is unrolled.
8753 @item max-unroll-times
8754 The maximum number of unrollings of a single loop.
8756 @item max-peeled-insns
8757 The maximum number of instructions that a loop should have if that loop
8758 is peeled, and if the loop is peeled, it determines how many times
8759 the loop code is peeled.
8761 @item max-peel-times
8762 The maximum number of peelings of a single loop.
8764 @item max-completely-peeled-insns
8765 The maximum number of insns of a completely peeled loop.
8767 @item max-completely-peel-times
8768 The maximum number of iterations of a loop to be suitable for complete peeling.
8770 @item max-completely-peel-loop-nest-depth
8771 The maximum depth of a loop nest suitable for complete peeling.
8773 @item max-unswitch-insns
8774 The maximum number of insns of an unswitched loop.
8776 @item max-unswitch-level
8777 The maximum number of branches unswitched in a single loop.
8780 The minimum cost of an expensive expression in the loop invariant motion.
8782 @item iv-consider-all-candidates-bound
8783 Bound on number of candidates for induction variables below that
8784 all candidates are considered for each use in induction variable
8785 optimizations. Only the most relevant candidates are considered
8786 if there are more candidates, to avoid quadratic time complexity.
8788 @item iv-max-considered-uses
8789 The induction variable optimizations give up on loops that contain more
8790 induction variable uses.
8792 @item iv-always-prune-cand-set-bound
8793 If number of candidates in the set is smaller than this value,
8794 we always try to remove unnecessary ivs from the set during its
8795 optimization when a new iv is added to the set.
8797 @item scev-max-expr-size
8798 Bound on size of expressions used in the scalar evolutions analyzer.
8799 Large expressions slow the analyzer.
8801 @item scev-max-expr-complexity
8802 Bound on the complexity of the expressions in the scalar evolutions analyzer.
8803 Complex expressions slow the analyzer.
8805 @item omega-max-vars
8806 The maximum number of variables in an Omega constraint system.
8807 The default value is 128.
8809 @item omega-max-geqs
8810 The maximum number of inequalities in an Omega constraint system.
8811 The default value is 256.
8814 The maximum number of equalities in an Omega constraint system.
8815 The default value is 128.
8817 @item omega-max-wild-cards
8818 The maximum number of wildcard variables that the Omega solver will
8819 be able to insert. The default value is 18.
8821 @item omega-hash-table-size
8822 The size of the hash table in the Omega solver. The default value is
8825 @item omega-max-keys
8826 The maximal number of keys used by the Omega solver. The default
8829 @item omega-eliminate-redundant-constraints
8830 When set to 1, use expensive methods to eliminate all redundant
8831 constraints. The default value is 0.
8833 @item vect-max-version-for-alignment-checks
8834 The maximum number of run-time checks that can be performed when
8835 doing loop versioning for alignment in the vectorizer. See option
8836 ftree-vect-loop-version for more information.
8838 @item vect-max-version-for-alias-checks
8839 The maximum number of run-time checks that can be performed when
8840 doing loop versioning for alias in the vectorizer. See option
8841 ftree-vect-loop-version for more information.
8843 @item max-iterations-to-track
8845 The maximum number of iterations of a loop the brute force algorithm
8846 for analysis of # of iterations of the loop tries to evaluate.
8848 @item hot-bb-count-fraction
8849 Select fraction of the maximal count of repetitions of basic block in program
8850 given basic block needs to have to be considered hot.
8852 @item hot-bb-frequency-fraction
8853 Select fraction of the entry block frequency of executions of basic block in
8854 function given basic block needs to have to be considered hot.
8856 @item max-predicted-iterations
8857 The maximum number of loop iterations we predict statically. This is useful
8858 in cases where function contain single loop with known bound and other loop
8859 with unknown. We predict the known number of iterations correctly, while
8860 the unknown number of iterations average to roughly 10. This means that the
8861 loop without bounds would appear artificially cold relative to the other one.
8863 @item align-threshold
8865 Select fraction of the maximal frequency of executions of basic block in
8866 function given basic block will get aligned.
8868 @item align-loop-iterations
8870 A loop expected to iterate at lest the selected number of iterations will get
8873 @item tracer-dynamic-coverage
8874 @itemx tracer-dynamic-coverage-feedback
8876 This value is used to limit superblock formation once the given percentage of
8877 executed instructions is covered. This limits unnecessary code size
8880 The @option{tracer-dynamic-coverage-feedback} is used only when profile
8881 feedback is available. The real profiles (as opposed to statically estimated
8882 ones) are much less balanced allowing the threshold to be larger value.
8884 @item tracer-max-code-growth
8885 Stop tail duplication once code growth has reached given percentage. This is
8886 rather hokey argument, as most of the duplicates will be eliminated later in
8887 cross jumping, so it may be set to much higher values than is the desired code
8890 @item tracer-min-branch-ratio
8892 Stop reverse growth when the reverse probability of best edge is less than this
8893 threshold (in percent).
8895 @item tracer-min-branch-ratio
8896 @itemx tracer-min-branch-ratio-feedback
8898 Stop forward growth if the best edge do have probability lower than this
8901 Similarly to @option{tracer-dynamic-coverage} two values are present, one for
8902 compilation for profile feedback and one for compilation without. The value
8903 for compilation with profile feedback needs to be more conservative (higher) in
8904 order to make tracer effective.
8906 @item max-cse-path-length
8908 Maximum number of basic blocks on path that cse considers. The default is 10.
8911 The maximum instructions CSE process before flushing. The default is 1000.
8913 @item ggc-min-expand
8915 GCC uses a garbage collector to manage its own memory allocation. This
8916 parameter specifies the minimum percentage by which the garbage
8917 collector's heap should be allowed to expand between collections.
8918 Tuning this may improve compilation speed; it has no effect on code
8921 The default is 30% + 70% * (RAM/1GB) with an upper bound of 100% when
8922 RAM >= 1GB@. If @code{getrlimit} is available, the notion of "RAM" is
8923 the smallest of actual RAM and @code{RLIMIT_DATA} or @code{RLIMIT_AS}. If
8924 GCC is not able to calculate RAM on a particular platform, the lower
8925 bound of 30% is used. Setting this parameter and
8926 @option{ggc-min-heapsize} to zero causes a full collection to occur at
8927 every opportunity. This is extremely slow, but can be useful for
8930 @item ggc-min-heapsize
8932 Minimum size of the garbage collector's heap before it begins bothering
8933 to collect garbage. The first collection occurs after the heap expands
8934 by @option{ggc-min-expand}% beyond @option{ggc-min-heapsize}. Again,
8935 tuning this may improve compilation speed, and has no effect on code
8938 The default is the smaller of RAM/8, RLIMIT_RSS, or a limit that
8939 tries to ensure that RLIMIT_DATA or RLIMIT_AS are not exceeded, but
8940 with a lower bound of 4096 (four megabytes) and an upper bound of
8941 131072 (128 megabytes). If GCC is not able to calculate RAM on a
8942 particular platform, the lower bound is used. Setting this parameter
8943 very large effectively disables garbage collection. Setting this
8944 parameter and @option{ggc-min-expand} to zero causes a full collection
8945 to occur at every opportunity.
8947 @item max-reload-search-insns
8948 The maximum number of instruction reload should look backward for equivalent
8949 register. Increasing values mean more aggressive optimization, making the
8950 compilation time increase with probably slightly better performance.
8951 The default value is 100.
8953 @item max-cselib-memory-locations
8954 The maximum number of memory locations cselib should take into account.
8955 Increasing values mean more aggressive optimization, making the compilation time
8956 increase with probably slightly better performance. The default value is 500.
8958 @item reorder-blocks-duplicate
8959 @itemx reorder-blocks-duplicate-feedback
8961 Used by basic block reordering pass to decide whether to use unconditional
8962 branch or duplicate the code on its destination. Code is duplicated when its
8963 estimated size is smaller than this value multiplied by the estimated size of
8964 unconditional jump in the hot spots of the program.
8966 The @option{reorder-block-duplicate-feedback} is used only when profile
8967 feedback is available and may be set to higher values than
8968 @option{reorder-block-duplicate} since information about the hot spots is more
8971 @item max-sched-ready-insns
8972 The maximum number of instructions ready to be issued the scheduler should
8973 consider at any given time during the first scheduling pass. Increasing
8974 values mean more thorough searches, making the compilation time increase
8975 with probably little benefit. The default value is 100.
8977 @item max-sched-region-blocks
8978 The maximum number of blocks in a region to be considered for
8979 interblock scheduling. The default value is 10.
8981 @item max-pipeline-region-blocks
8982 The maximum number of blocks in a region to be considered for
8983 pipelining in the selective scheduler. The default value is 15.
8985 @item max-sched-region-insns
8986 The maximum number of insns in a region to be considered for
8987 interblock scheduling. The default value is 100.
8989 @item max-pipeline-region-insns
8990 The maximum number of insns in a region to be considered for
8991 pipelining in the selective scheduler. The default value is 200.
8994 The minimum probability (in percents) of reaching a source block
8995 for interblock speculative scheduling. The default value is 40.
8997 @item max-sched-extend-regions-iters
8998 The maximum number of iterations through CFG to extend regions.
8999 0 - disable region extension,
9000 N - do at most N iterations.
9001 The default value is 0.
9003 @item max-sched-insn-conflict-delay
9004 The maximum conflict delay for an insn to be considered for speculative motion.
9005 The default value is 3.
9007 @item sched-spec-prob-cutoff
9008 The minimal probability of speculation success (in percents), so that
9009 speculative insn will be scheduled.
9010 The default value is 40.
9012 @item sched-mem-true-dep-cost
9013 Minimal distance (in CPU cycles) between store and load targeting same
9014 memory locations. The default value is 1.
9016 @item selsched-max-lookahead
9017 The maximum size of the lookahead window of selective scheduling. It is a
9018 depth of search for available instructions.
9019 The default value is 50.
9021 @item selsched-max-sched-times
9022 The maximum number of times that an instruction will be scheduled during
9023 selective scheduling. This is the limit on the number of iterations
9024 through which the instruction may be pipelined. The default value is 2.
9026 @item selsched-max-insns-to-rename
9027 The maximum number of best instructions in the ready list that are considered
9028 for renaming in the selective scheduler. The default value is 2.
9031 The minimum value of stage count that swing modulo scheduler will
9032 generate. The default value is 2.
9034 @item max-last-value-rtl
9035 The maximum size measured as number of RTLs that can be recorded in an expression
9036 in combiner for a pseudo register as last known value of that register. The default
9039 @item integer-share-limit
9040 Small integer constants can use a shared data structure, reducing the
9041 compiler's memory usage and increasing its speed. This sets the maximum
9042 value of a shared integer constant. The default value is 256.
9044 @item min-virtual-mappings
9045 Specifies the minimum number of virtual mappings in the incremental
9046 SSA updater that should be registered to trigger the virtual mappings
9047 heuristic defined by virtual-mappings-ratio. The default value is
9050 @item virtual-mappings-ratio
9051 If the number of virtual mappings is virtual-mappings-ratio bigger
9052 than the number of virtual symbols to be updated, then the incremental
9053 SSA updater switches to a full update for those symbols. The default
9056 @item ssp-buffer-size
9057 The minimum size of buffers (i.e.@: arrays) that will receive stack smashing
9058 protection when @option{-fstack-protection} is used.
9060 @item max-jump-thread-duplication-stmts
9061 Maximum number of statements allowed in a block that needs to be
9062 duplicated when threading jumps.
9064 @item max-fields-for-field-sensitive
9065 Maximum number of fields in a structure we will treat in
9066 a field sensitive manner during pointer analysis. The default is zero
9067 for -O0, and -O1 and 100 for -Os, -O2, and -O3.
9069 @item prefetch-latency
9070 Estimate on average number of instructions that are executed before
9071 prefetch finishes. The distance we prefetch ahead is proportional
9072 to this constant. Increasing this number may also lead to less
9073 streams being prefetched (see @option{simultaneous-prefetches}).
9075 @item simultaneous-prefetches
9076 Maximum number of prefetches that can run at the same time.
9078 @item l1-cache-line-size
9079 The size of cache line in L1 cache, in bytes.
9082 The size of L1 cache, in kilobytes.
9085 The size of L2 cache, in kilobytes.
9087 @item min-insn-to-prefetch-ratio
9088 The minimum ratio between the number of instructions and the
9089 number of prefetches to enable prefetching in a loop.
9091 @item prefetch-min-insn-to-mem-ratio
9092 The minimum ratio between the number of instructions and the
9093 number of memory references to enable prefetching in a loop.
9095 @item use-canonical-types
9096 Whether the compiler should use the ``canonical'' type system. By
9097 default, this should always be 1, which uses a more efficient internal
9098 mechanism for comparing types in C++ and Objective-C++. However, if
9099 bugs in the canonical type system are causing compilation failures,
9100 set this value to 0 to disable canonical types.
9102 @item switch-conversion-max-branch-ratio
9103 Switch initialization conversion will refuse to create arrays that are
9104 bigger than @option{switch-conversion-max-branch-ratio} times the number of
9105 branches in the switch.
9107 @item max-partial-antic-length
9108 Maximum length of the partial antic set computed during the tree
9109 partial redundancy elimination optimization (@option{-ftree-pre}) when
9110 optimizing at @option{-O3} and above. For some sorts of source code
9111 the enhanced partial redundancy elimination optimization can run away,
9112 consuming all of the memory available on the host machine. This
9113 parameter sets a limit on the length of the sets that are computed,
9114 which prevents the runaway behavior. Setting a value of 0 for
9115 this parameter will allow an unlimited set length.
9117 @item sccvn-max-scc-size
9118 Maximum size of a strongly connected component (SCC) during SCCVN
9119 processing. If this limit is hit, SCCVN processing for the whole
9120 function will not be done and optimizations depending on it will
9121 be disabled. The default maximum SCC size is 10000.
9123 @item ira-max-loops-num
9124 IRA uses regional register allocation by default. If a function
9125 contains more loops than the number given by this parameter, only at most
9126 the given number of the most frequently-executed loops form regions
9127 for regional register allocation. The default value of the
9130 @item ira-max-conflict-table-size
9131 Although IRA uses a sophisticated algorithm to compress the conflict
9132 table, the table can still require excessive amounts of memory for
9133 huge functions. If the conflict table for a function could be more
9134 than the size in MB given by this parameter, the register allocator
9135 instead uses a faster, simpler, and lower-quality
9136 algorithm that does not require building a pseudo-register conflict table.
9137 The default value of the parameter is 2000.
9139 @item ira-loop-reserved-regs
9140 IRA can be used to evaluate more accurate register pressure in loops
9141 for decisions to move loop invariants (see @option{-O3}). The number
9142 of available registers reserved for some other purposes is given
9143 by this parameter. The default value of the parameter is 2, which is
9144 the minimal number of registers needed by typical instructions.
9145 This value is the best found from numerous experiments.
9147 @item loop-invariant-max-bbs-in-loop
9148 Loop invariant motion can be very expensive, both in compilation time and
9149 in amount of needed compile-time memory, with very large loops. Loops
9150 with more basic blocks than this parameter won't have loop invariant
9151 motion optimization performed on them. The default value of the
9152 parameter is 1000 for -O1 and 10000 for -O2 and above.
9154 @item loop-max-datarefs-for-datadeps
9155 Building data dapendencies is expensive for very large loops. This
9156 parameter limits the number of data references in loops that are
9157 considered for data dependence analysis. These large loops will not
9158 be handled then by the optimizations using loop data dependencies.
9159 The default value is 1000.
9161 @item max-vartrack-size
9162 Sets a maximum number of hash table slots to use during variable
9163 tracking dataflow analysis of any function. If this limit is exceeded
9164 with variable tracking at assignments enabled, analysis for that
9165 function is retried without it, after removing all debug insns from
9166 the function. If the limit is exceeded even without debug insns, var
9167 tracking analysis is completely disabled for the function. Setting
9168 the parameter to zero makes it unlimited.
9170 @item max-vartrack-expr-depth
9171 Sets a maximum number of recursion levels when attempting to map
9172 variable names or debug temporaries to value expressions. This trades
9173 compilation time for more complete debug information. If this is set too
9174 low, value expressions that are available and could be represented in
9175 debug information may end up not being used; setting this higher may
9176 enable the compiler to find more complex debug expressions, but compile
9177 time and memory use may grow. The default is 12.
9179 @item min-nondebug-insn-uid
9180 Use uids starting at this parameter for nondebug insns. The range below
9181 the parameter is reserved exclusively for debug insns created by
9182 @option{-fvar-tracking-assignments}, but debug insns may get
9183 (non-overlapping) uids above it if the reserved range is exhausted.
9185 @item ipa-sra-ptr-growth-factor
9186 IPA-SRA will replace a pointer to an aggregate with one or more new
9187 parameters only when their cumulative size is less or equal to
9188 @option{ipa-sra-ptr-growth-factor} times the size of the original
9191 @item tm-max-aggregate-size
9192 When making copies of thread-local variables in a transaction, this
9193 parameter specifies the size in bytes after which variables will be
9194 saved with the logging functions as opposed to save/restore code
9195 sequence pairs. This option only applies when using
9198 @item graphite-max-nb-scop-params
9199 To avoid exponential effects in the Graphite loop transforms, the
9200 number of parameters in a Static Control Part (SCoP) is bounded. The
9201 default value is 10 parameters. A variable whose value is unknown at
9202 compilation time and defined outside a SCoP is a parameter of the SCoP.
9204 @item graphite-max-bbs-per-function
9205 To avoid exponential effects in the detection of SCoPs, the size of
9206 the functions analyzed by Graphite is bounded. The default value is
9209 @item loop-block-tile-size
9210 Loop blocking or strip mining transforms, enabled with
9211 @option{-floop-block} or @option{-floop-strip-mine}, strip mine each
9212 loop in the loop nest by a given number of iterations. The strip
9213 length can be changed using the @option{loop-block-tile-size}
9214 parameter. The default value is 51 iterations.
9216 @item ipa-cp-value-list-size
9217 IPA-CP attempts to track all possible values and types passed to a function's
9218 parameter in order to propagate them and perform devirtualization.
9219 @option{ipa-cp-value-list-size} is the maximum number of values and types it
9220 stores per one formal parameter of a function.
9222 @item lto-partitions
9223 Specify desired number of partitions produced during WHOPR compilation.
9224 The number of partitions should exceed the number of CPUs used for compilation.
9225 The default value is 32.
9227 @item lto-minpartition
9228 Size of minimal partition for WHOPR (in estimated instructions).
9229 This prevents expenses of splitting very small programs into too many
9232 @item cxx-max-namespaces-for-diagnostic-help
9233 The maximum number of namespaces to consult for suggestions when C++
9234 name lookup fails for an identifier. The default is 1000.
9236 @item sink-frequency-threshold
9237 The maximum relative execution frequency (in percents) of the target block
9238 relative to a statement's original block to allow statement sinking of a
9239 statement. Larger numbers result in more aggressive statement sinking.
9240 The default value is 75. A small positive adjustment is applied for
9241 statements with memory operands as those are even more profitable so sink.
9243 @item max-stores-to-sink
9244 The maximum number of conditional stores paires that can be sunk. Set to 0
9245 if either vectorization (@option{-ftree-vectorize}) or if-conversion
9246 (@option{-ftree-loop-if-convert}) is disabled. The default is 2.
9248 @item allow-load-data-races
9249 Allow optimizers to introduce new data races on loads.
9250 Set to 1 to allow, otherwise to 0. This option is enabled by default
9251 unless implicitly set by the @option{-fmemory-model=} option.
9253 @item allow-store-data-races
9254 Allow optimizers to introduce new data races on stores.
9255 Set to 1 to allow, otherwise to 0. This option is enabled by default
9256 unless implicitly set by the @option{-fmemory-model=} option.
9258 @item allow-packed-load-data-races
9259 Allow optimizers to introduce new data races on packed data loads.
9260 Set to 1 to allow, otherwise to 0. This option is enabled by default
9261 unless implicitly set by the @option{-fmemory-model=} option.
9263 @item allow-packed-store-data-races
9264 Allow optimizers to introduce new data races on packed data stores.
9265 Set to 1 to allow, otherwise to 0. This option is enabled by default
9266 unless implicitly set by the @option{-fmemory-model=} option.
9268 @item case-values-threshold
9269 The smallest number of different values for which it is best to use a
9270 jump-table instead of a tree of conditional branches. If the value is
9271 0, use the default for the machine. The default is 0.
9273 @item tree-reassoc-width
9274 Set the maximum number of instructions executed in parallel in
9275 reassociated tree. This parameter overrides target dependent
9276 heuristics used by default if has non zero value.
9281 @node Preprocessor Options
9282 @section Options Controlling the Preprocessor
9283 @cindex preprocessor options
9284 @cindex options, preprocessor
9286 These options control the C preprocessor, which is run on each C source
9287 file before actual compilation.
9289 If you use the @option{-E} option, nothing is done except preprocessing.
9290 Some of these options make sense only together with @option{-E} because
9291 they cause the preprocessor output to be unsuitable for actual
9295 @item -Wp,@var{option}
9297 You can use @option{-Wp,@var{option}} to bypass the compiler driver
9298 and pass @var{option} directly through to the preprocessor. If
9299 @var{option} contains commas, it is split into multiple options at the
9300 commas. However, many options are modified, translated or interpreted
9301 by the compiler driver before being passed to the preprocessor, and
9302 @option{-Wp} forcibly bypasses this phase. The preprocessor's direct
9303 interface is undocumented and subject to change, so whenever possible
9304 you should avoid using @option{-Wp} and let the driver handle the
9307 @item -Xpreprocessor @var{option}
9308 @opindex Xpreprocessor
9309 Pass @var{option} as an option to the preprocessor. You can use this to
9310 supply system-specific preprocessor options that GCC does not know how to
9313 If you want to pass an option that takes an argument, you must use
9314 @option{-Xpreprocessor} twice, once for the option and once for the argument.
9317 @include cppopts.texi
9319 @node Assembler Options
9320 @section Passing Options to the Assembler
9322 @c prevent bad page break with this line
9323 You can pass options to the assembler.
9326 @item -Wa,@var{option}
9328 Pass @var{option} as an option to the assembler. If @var{option}
9329 contains commas, it is split into multiple options at the commas.
9331 @item -Xassembler @var{option}
9333 Pass @var{option} as an option to the assembler. You can use this to
9334 supply system-specific assembler options that GCC does not know how to
9337 If you want to pass an option that takes an argument, you must use
9338 @option{-Xassembler} twice, once for the option and once for the argument.
9343 @section Options for Linking
9344 @cindex link options
9345 @cindex options, linking
9347 These options come into play when the compiler links object files into
9348 an executable output file. They are meaningless if the compiler is
9349 not doing a link step.
9353 @item @var{object-file-name}
9354 A file name that does not end in a special recognized suffix is
9355 considered to name an object file or library. (Object files are
9356 distinguished from libraries by the linker according to the file
9357 contents.) If linking is done, these object files are used as input
9366 If any of these options is used, then the linker is not run, and
9367 object file names should not be used as arguments. @xref{Overall
9371 @item -l@var{library}
9372 @itemx -l @var{library}
9374 Search the library named @var{library} when linking. (The second
9375 alternative with the library as a separate argument is only for
9376 POSIX compliance and is not recommended.)
9378 It makes a difference where in the command you write this option; the
9379 linker searches and processes libraries and object files in the order they
9380 are specified. Thus, @samp{foo.o -lz bar.o} searches library @samp{z}
9381 after file @file{foo.o} but before @file{bar.o}. If @file{bar.o} refers
9382 to functions in @samp{z}, those functions may not be loaded.
9384 The linker searches a standard list of directories for the library,
9385 which is actually a file named @file{lib@var{library}.a}. The linker
9386 then uses this file as if it had been specified precisely by name.
9388 The directories searched include several standard system directories
9389 plus any that you specify with @option{-L}.
9391 Normally the files found this way are library files---archive files
9392 whose members are object files. The linker handles an archive file by
9393 scanning through it for members which define symbols that have so far
9394 been referenced but not defined. But if the file that is found is an
9395 ordinary object file, it is linked in the usual fashion. The only
9396 difference between using an @option{-l} option and specifying a file name
9397 is that @option{-l} surrounds @var{library} with @samp{lib} and @samp{.a}
9398 and searches several directories.
9402 You need this special case of the @option{-l} option in order to
9403 link an Objective-C or Objective-C++ program.
9406 @opindex nostartfiles
9407 Do not use the standard system startup files when linking.
9408 The standard system libraries are used normally, unless @option{-nostdlib}
9409 or @option{-nodefaultlibs} is used.
9411 @item -nodefaultlibs
9412 @opindex nodefaultlibs
9413 Do not use the standard system libraries when linking.
9414 Only the libraries you specify will be passed to the linker, options
9415 specifying linkage of the system libraries, such as @code{-static-libgcc}
9416 or @code{-shared-libgcc}, will be ignored.
9417 The standard startup files are used normally, unless @option{-nostartfiles}
9418 is used. The compiler may generate calls to @code{memcmp},
9419 @code{memset}, @code{memcpy} and @code{memmove}.
9420 These entries are usually resolved by entries in
9421 libc. These entry points should be supplied through some other
9422 mechanism when this option is specified.
9426 Do not use the standard system startup files or libraries when linking.
9427 No startup files and only the libraries you specify will be passed to
9428 the linker, options specifying linkage of the system libraries, such as
9429 @code{-static-libgcc} or @code{-shared-libgcc}, will be ignored.
9430 The compiler may generate calls to @code{memcmp}, @code{memset},
9431 @code{memcpy} and @code{memmove}.
9432 These entries are usually resolved by entries in
9433 libc. These entry points should be supplied through some other
9434 mechanism when this option is specified.
9436 @cindex @option{-lgcc}, use with @option{-nostdlib}
9437 @cindex @option{-nostdlib} and unresolved references
9438 @cindex unresolved references and @option{-nostdlib}
9439 @cindex @option{-lgcc}, use with @option{-nodefaultlibs}
9440 @cindex @option{-nodefaultlibs} and unresolved references
9441 @cindex unresolved references and @option{-nodefaultlibs}
9442 One of the standard libraries bypassed by @option{-nostdlib} and
9443 @option{-nodefaultlibs} is @file{libgcc.a}, a library of internal subroutines
9444 which GCC uses to overcome shortcomings of particular machines, or special
9445 needs for some languages.
9446 (@xref{Interface,,Interfacing to GCC Output,gccint,GNU Compiler
9447 Collection (GCC) Internals},
9448 for more discussion of @file{libgcc.a}.)
9449 In most cases, you need @file{libgcc.a} even when you want to avoid
9450 other standard libraries. In other words, when you specify @option{-nostdlib}
9451 or @option{-nodefaultlibs} you should usually specify @option{-lgcc} as well.
9452 This ensures that you have no unresolved references to internal GCC
9453 library subroutines. (For example, @samp{__main}, used to ensure C++
9454 constructors will be called; @pxref{Collect2,,@code{collect2}, gccint,
9455 GNU Compiler Collection (GCC) Internals}.)
9459 Produce a position independent executable on targets that support it.
9460 For predictable results, you must also specify the same set of options
9461 that were used to generate code (@option{-fpie}, @option{-fPIE},
9462 or model suboptions) when you specify this option.
9466 Pass the flag @option{-export-dynamic} to the ELF linker, on targets
9467 that support it. This instructs the linker to add all symbols, not
9468 only used ones, to the dynamic symbol table. This option is needed
9469 for some uses of @code{dlopen} or to allow obtaining backtraces
9470 from within a program.
9474 Remove all symbol table and relocation information from the executable.
9478 On systems that support dynamic linking, this prevents linking with the shared
9479 libraries. On other systems, this option has no effect.
9483 Produce a shared object which can then be linked with other objects to
9484 form an executable. Not all systems support this option. For predictable
9485 results, you must also specify the same set of options that were used to
9486 generate code (@option{-fpic}, @option{-fPIC}, or model suboptions)
9487 when you specify this option.@footnote{On some systems, @samp{gcc -shared}
9488 needs to build supplementary stub code for constructors to work. On
9489 multi-libbed systems, @samp{gcc -shared} must select the correct support
9490 libraries to link against. Failing to supply the correct flags may lead
9491 to subtle defects. Supplying them in cases where they are not necessary
9494 @item -shared-libgcc
9495 @itemx -static-libgcc
9496 @opindex shared-libgcc
9497 @opindex static-libgcc
9498 On systems that provide @file{libgcc} as a shared library, these options
9499 force the use of either the shared or static version respectively.
9500 If no shared version of @file{libgcc} was built when the compiler was
9501 configured, these options have no effect.
9503 There are several situations in which an application should use the
9504 shared @file{libgcc} instead of the static version. The most common
9505 of these is when the application wishes to throw and catch exceptions
9506 across different shared libraries. In that case, each of the libraries
9507 as well as the application itself should use the shared @file{libgcc}.
9509 Therefore, the G++ and GCJ drivers automatically add
9510 @option{-shared-libgcc} whenever you build a shared library or a main
9511 executable, because C++ and Java programs typically use exceptions, so
9512 this is the right thing to do.
9514 If, instead, you use the GCC driver to create shared libraries, you may
9515 find that they will not always be linked with the shared @file{libgcc}.
9516 If GCC finds, at its configuration time, that you have a non-GNU linker
9517 or a GNU linker that does not support option @option{--eh-frame-hdr},
9518 it will link the shared version of @file{libgcc} into shared libraries
9519 by default. Otherwise, it will take advantage of the linker and optimize
9520 away the linking with the shared version of @file{libgcc}, linking with
9521 the static version of libgcc by default. This allows exceptions to
9522 propagate through such shared libraries, without incurring relocation
9523 costs at library load time.
9525 However, if a library or main executable is supposed to throw or catch
9526 exceptions, you must link it using the G++ or GCJ driver, as appropriate
9527 for the languages used in the program, or using the option
9528 @option{-shared-libgcc}, such that it is linked with the shared
9531 @item -static-libstdc++
9532 When the @command{g++} program is used to link a C++ program, it will
9533 normally automatically link against @option{libstdc++}. If
9534 @file{libstdc++} is available as a shared library, and the
9535 @option{-static} option is not used, then this will link against the
9536 shared version of @file{libstdc++}. That is normally fine. However, it
9537 is sometimes useful to freeze the version of @file{libstdc++} used by
9538 the program without going all the way to a fully static link. The
9539 @option{-static-libstdc++} option directs the @command{g++} driver to
9540 link @file{libstdc++} statically, without necessarily linking other
9541 libraries statically.
9545 Bind references to global symbols when building a shared object. Warn
9546 about any unresolved references (unless overridden by the link editor
9547 option @samp{-Xlinker -z -Xlinker defs}). Only a few systems support
9550 @item -T @var{script}
9552 @cindex linker script
9553 Use @var{script} as the linker script. This option is supported by most
9554 systems using the GNU linker. On some targets, such as bare-board
9555 targets without an operating system, the @option{-T} option may be required
9556 when linking to avoid references to undefined symbols.
9558 @item -Xlinker @var{option}
9560 Pass @var{option} as an option to the linker. You can use this to
9561 supply system-specific linker options that GCC does not recognize.
9563 If you want to pass an option that takes a separate argument, you must use
9564 @option{-Xlinker} twice, once for the option and once for the argument.
9565 For example, to pass @option{-assert definitions}, you must write
9566 @samp{-Xlinker -assert -Xlinker definitions}. It does not work to write
9567 @option{-Xlinker "-assert definitions"}, because this passes the entire
9568 string as a single argument, which is not what the linker expects.
9570 When using the GNU linker, it is usually more convenient to pass
9571 arguments to linker options using the @option{@var{option}=@var{value}}
9572 syntax than as separate arguments. For example, you can specify
9573 @samp{-Xlinker -Map=output.map} rather than
9574 @samp{-Xlinker -Map -Xlinker output.map}. Other linkers may not support
9575 this syntax for command-line options.
9577 @item -Wl,@var{option}
9579 Pass @var{option} as an option to the linker. If @var{option} contains
9580 commas, it is split into multiple options at the commas. You can use this
9581 syntax to pass an argument to the option.
9582 For example, @samp{-Wl,-Map,output.map} passes @samp{-Map output.map} to the
9583 linker. When using the GNU linker, you can also get the same effect with
9584 @samp{-Wl,-Map=output.map}.
9586 @item -u @var{symbol}
9588 Pretend the symbol @var{symbol} is undefined, to force linking of
9589 library modules to define it. You can use @option{-u} multiple times with
9590 different symbols to force loading of additional library modules.
9593 @node Directory Options
9594 @section Options for Directory Search
9595 @cindex directory options
9596 @cindex options, directory search
9599 These options specify directories to search for header files, for
9600 libraries and for parts of the compiler:
9605 Add the directory @var{dir} to the head of the list of directories to be
9606 searched for header files. This can be used to override a system header
9607 file, substituting your own version, since these directories are
9608 searched before the system header file directories. However, you should
9609 not use this option to add directories that contain vendor-supplied
9610 system header files (use @option{-isystem} for that). If you use more than
9611 one @option{-I} option, the directories are scanned in left-to-right
9612 order; the standard system directories come after.
9614 If a standard system include directory, or a directory specified with
9615 @option{-isystem}, is also specified with @option{-I}, the @option{-I}
9616 option will be ignored. The directory will still be searched but as a
9617 system directory at its normal position in the system include chain.
9618 This is to ensure that GCC's procedure to fix buggy system headers and
9619 the ordering for the include_next directive are not inadvertently changed.
9620 If you really need to change the search order for system directories,
9621 use the @option{-nostdinc} and/or @option{-isystem} options.
9623 @item -iplugindir=@var{dir}
9624 Set the directory to search for plugins that are passed
9625 by @option{-fplugin=@var{name}} instead of
9626 @option{-fplugin=@var{path}/@var{name}.so}. This option is not meant
9627 to be used by the user, but only passed by the driver.
9629 @item -iquote@var{dir}
9631 Add the directory @var{dir} to the head of the list of directories to
9632 be searched for header files only for the case of @samp{#include
9633 "@var{file}"}; they are not searched for @samp{#include <@var{file}>},
9634 otherwise just like @option{-I}.
9638 Add directory @var{dir} to the list of directories to be searched
9641 @item -B@var{prefix}
9643 This option specifies where to find the executables, libraries,
9644 include files, and data files of the compiler itself.
9646 The compiler driver program runs one or more of the subprograms
9647 @file{cpp}, @file{cc1}, @file{as} and @file{ld}. It tries
9648 @var{prefix} as a prefix for each program it tries to run, both with and
9649 without @samp{@var{machine}/@var{version}/} (@pxref{Target Options}).
9651 For each subprogram to be run, the compiler driver first tries the
9652 @option{-B} prefix, if any. If that name is not found, or if @option{-B}
9653 was not specified, the driver tries two standard prefixes,
9654 @file{/usr/lib/gcc/} and @file{/usr/local/lib/gcc/}. If neither of
9655 those results in a file name that is found, the unmodified program
9656 name is searched for using the directories specified in your
9657 @env{PATH} environment variable.
9659 The compiler will check to see if the path provided by the @option{-B}
9660 refers to a directory, and if necessary it will add a directory
9661 separator character at the end of the path.
9663 @option{-B} prefixes that effectively specify directory names also apply
9664 to libraries in the linker, because the compiler translates these
9665 options into @option{-L} options for the linker. They also apply to
9666 includes files in the preprocessor, because the compiler translates these
9667 options into @option{-isystem} options for the preprocessor. In this case,
9668 the compiler appends @samp{include} to the prefix.
9670 The runtime support file @file{libgcc.a} can also be searched for using
9671 the @option{-B} prefix, if needed. If it is not found there, the two
9672 standard prefixes above are tried, and that is all. The file is left
9673 out of the link if it is not found by those means.
9675 Another way to specify a prefix much like the @option{-B} prefix is to use
9676 the environment variable @env{GCC_EXEC_PREFIX}. @xref{Environment
9679 As a special kludge, if the path provided by @option{-B} is
9680 @file{[dir/]stage@var{N}/}, where @var{N} is a number in the range 0 to
9681 9, then it will be replaced by @file{[dir/]include}. This is to help
9682 with boot-strapping the compiler.
9684 @item -specs=@var{file}
9686 Process @var{file} after the compiler reads in the standard @file{specs}
9687 file, in order to override the defaults which the @file{gcc} driver
9688 program uses when determining what switches to pass to @file{cc1},
9689 @file{cc1plus}, @file{as}, @file{ld}, etc. More than one
9690 @option{-specs=@var{file}} can be specified on the command line, and they
9691 are processed in order, from left to right.
9693 @item --sysroot=@var{dir}
9695 Use @var{dir} as the logical root directory for headers and libraries.
9696 For example, if the compiler would normally search for headers in
9697 @file{/usr/include} and libraries in @file{/usr/lib}, it will instead
9698 search @file{@var{dir}/usr/include} and @file{@var{dir}/usr/lib}.
9700 If you use both this option and the @option{-isysroot} option, then
9701 the @option{--sysroot} option will apply to libraries, but the
9702 @option{-isysroot} option will apply to header files.
9704 The GNU linker (beginning with version 2.16) has the necessary support
9705 for this option. If your linker does not support this option, the
9706 header file aspect of @option{--sysroot} will still work, but the
9707 library aspect will not.
9711 This option has been deprecated. Please use @option{-iquote} instead for
9712 @option{-I} directories before the @option{-I-} and remove the @option{-I-}.
9713 Any directories you specify with @option{-I} options before the @option{-I-}
9714 option are searched only for the case of @samp{#include "@var{file}"};
9715 they are not searched for @samp{#include <@var{file}>}.
9717 If additional directories are specified with @option{-I} options after
9718 the @option{-I-}, these directories are searched for all @samp{#include}
9719 directives. (Ordinarily @emph{all} @option{-I} directories are used
9722 In addition, the @option{-I-} option inhibits the use of the current
9723 directory (where the current input file came from) as the first search
9724 directory for @samp{#include "@var{file}"}. There is no way to
9725 override this effect of @option{-I-}. With @option{-I.} you can specify
9726 searching the directory that was current when the compiler was
9727 invoked. That is not exactly the same as what the preprocessor does
9728 by default, but it is often satisfactory.
9730 @option{-I-} does not inhibit the use of the standard system directories
9731 for header files. Thus, @option{-I-} and @option{-nostdinc} are
9738 @section Specifying subprocesses and the switches to pass to them
9741 @command{gcc} is a driver program. It performs its job by invoking a
9742 sequence of other programs to do the work of compiling, assembling and
9743 linking. GCC interprets its command-line parameters and uses these to
9744 deduce which programs it should invoke, and which command-line options
9745 it ought to place on their command lines. This behavior is controlled
9746 by @dfn{spec strings}. In most cases there is one spec string for each
9747 program that GCC can invoke, but a few programs have multiple spec
9748 strings to control their behavior. The spec strings built into GCC can
9749 be overridden by using the @option{-specs=} command-line switch to specify
9752 @dfn{Spec files} are plaintext files that are used to construct spec
9753 strings. They consist of a sequence of directives separated by blank
9754 lines. The type of directive is determined by the first non-whitespace
9755 character on the line, which can be one of the following:
9758 @item %@var{command}
9759 Issues a @var{command} to the spec file processor. The commands that can
9763 @item %include <@var{file}>
9764 @cindex @code{%include}
9765 Search for @var{file} and insert its text at the current point in the
9768 @item %include_noerr <@var{file}>
9769 @cindex @code{%include_noerr}
9770 Just like @samp{%include}, but do not generate an error message if the include
9771 file cannot be found.
9773 @item %rename @var{old_name} @var{new_name}
9774 @cindex @code{%rename}
9775 Rename the spec string @var{old_name} to @var{new_name}.
9779 @item *[@var{spec_name}]:
9780 This tells the compiler to create, override or delete the named spec
9781 string. All lines after this directive up to the next directive or
9782 blank line are considered to be the text for the spec string. If this
9783 results in an empty string then the spec will be deleted. (Or, if the
9784 spec did not exist, then nothing will happen.) Otherwise, if the spec
9785 does not currently exist a new spec will be created. If the spec does
9786 exist then its contents will be overridden by the text of this
9787 directive, unless the first character of that text is the @samp{+}
9788 character, in which case the text will be appended to the spec.
9790 @item [@var{suffix}]:
9791 Creates a new @samp{[@var{suffix}] spec} pair. All lines after this directive
9792 and up to the next directive or blank line are considered to make up the
9793 spec string for the indicated suffix. When the compiler encounters an
9794 input file with the named suffix, it will processes the spec string in
9795 order to work out how to compile that file. For example:
9802 This says that any input file whose name ends in @samp{.ZZ} should be
9803 passed to the program @samp{z-compile}, which should be invoked with the
9804 command-line switch @option{-input} and with the result of performing the
9805 @samp{%i} substitution. (See below.)
9807 As an alternative to providing a spec string, the text that follows a
9808 suffix directive can be one of the following:
9811 @item @@@var{language}
9812 This says that the suffix is an alias for a known @var{language}. This is
9813 similar to using the @option{-x} command-line switch to GCC to specify a
9814 language explicitly. For example:
9821 Says that .ZZ files are, in fact, C++ source files.
9824 This causes an error messages saying:
9827 @var{name} compiler not installed on this system.
9831 GCC already has an extensive list of suffixes built into it.
9832 This directive will add an entry to the end of the list of suffixes, but
9833 since the list is searched from the end backwards, it is effectively
9834 possible to override earlier entries using this technique.
9838 GCC has the following spec strings built into it. Spec files can
9839 override these strings or create their own. Note that individual
9840 targets can also add their own spec strings to this list.
9843 asm Options to pass to the assembler
9844 asm_final Options to pass to the assembler post-processor
9845 cpp Options to pass to the C preprocessor
9846 cc1 Options to pass to the C compiler
9847 cc1plus Options to pass to the C++ compiler
9848 endfile Object files to include at the end of the link
9849 link Options to pass to the linker
9850 lib Libraries to include on the command line to the linker
9851 libgcc Decides which GCC support library to pass to the linker
9852 linker Sets the name of the linker
9853 predefines Defines to be passed to the C preprocessor
9854 signed_char Defines to pass to CPP to say whether @code{char} is signed
9856 startfile Object files to include at the start of the link
9859 Here is a small example of a spec file:
9865 --start-group -lgcc -lc -leval1 --end-group %(old_lib)
9868 This example renames the spec called @samp{lib} to @samp{old_lib} and
9869 then overrides the previous definition of @samp{lib} with a new one.
9870 The new definition adds in some extra command-line options before
9871 including the text of the old definition.
9873 @dfn{Spec strings} are a list of command-line options to be passed to their
9874 corresponding program. In addition, the spec strings can contain
9875 @samp{%}-prefixed sequences to substitute variable text or to
9876 conditionally insert text into the command line. Using these constructs
9877 it is possible to generate quite complex command lines.
9879 Here is a table of all defined @samp{%}-sequences for spec
9880 strings. Note that spaces are not generated automatically around the
9881 results of expanding these sequences. Therefore you can concatenate them
9882 together or combine them with constant text in a single argument.
9886 Substitute one @samp{%} into the program name or argument.
9889 Substitute the name of the input file being processed.
9892 Substitute the basename of the input file being processed.
9893 This is the substring up to (and not including) the last period
9894 and not including the directory.
9897 This is the same as @samp{%b}, but include the file suffix (text after
9901 Marks the argument containing or following the @samp{%d} as a
9902 temporary file name, so that that file will be deleted if GCC exits
9903 successfully. Unlike @samp{%g}, this contributes no text to the
9906 @item %g@var{suffix}
9907 Substitute a file name that has suffix @var{suffix} and is chosen
9908 once per compilation, and mark the argument in the same way as
9909 @samp{%d}. To reduce exposure to denial-of-service attacks, the file
9910 name is now chosen in a way that is hard to predict even when previously
9911 chosen file names are known. For example, @samp{%g.s @dots{} %g.o @dots{} %g.s}
9912 might turn into @samp{ccUVUUAU.s ccXYAXZ12.o ccUVUUAU.s}. @var{suffix} matches
9913 the regexp @samp{[.A-Za-z]*} or the special string @samp{%O}, which is
9914 treated exactly as if @samp{%O} had been preprocessed. Previously, @samp{%g}
9915 was simply substituted with a file name chosen once per compilation,
9916 without regard to any appended suffix (which was therefore treated
9917 just like ordinary text), making such attacks more likely to succeed.
9919 @item %u@var{suffix}
9920 Like @samp{%g}, but generates a new temporary file name even if
9921 @samp{%u@var{suffix}} was already seen.
9923 @item %U@var{suffix}
9924 Substitutes the last file name generated with @samp{%u@var{suffix}}, generating a
9925 new one if there is no such last file name. In the absence of any
9926 @samp{%u@var{suffix}}, this is just like @samp{%g@var{suffix}}, except they don't share
9927 the same suffix @emph{space}, so @samp{%g.s @dots{} %U.s @dots{} %g.s @dots{} %U.s}
9928 would involve the generation of two distinct file names, one
9929 for each @samp{%g.s} and another for each @samp{%U.s}. Previously, @samp{%U} was
9930 simply substituted with a file name chosen for the previous @samp{%u},
9931 without regard to any appended suffix.
9933 @item %j@var{suffix}
9934 Substitutes the name of the @code{HOST_BIT_BUCKET}, if any, and if it is
9935 writable, and if save-temps is off; otherwise, substitute the name
9936 of a temporary file, just like @samp{%u}. This temporary file is not
9937 meant for communication between processes, but rather as a junk
9940 @item %|@var{suffix}
9941 @itemx %m@var{suffix}
9942 Like @samp{%g}, except if @option{-pipe} is in effect. In that case
9943 @samp{%|} substitutes a single dash and @samp{%m} substitutes nothing at
9944 all. These are the two most common ways to instruct a program that it
9945 should read from standard input or write to standard output. If you
9946 need something more elaborate you can use an @samp{%@{pipe:@code{X}@}}
9947 construct: see for example @file{f/lang-specs.h}.
9949 @item %.@var{SUFFIX}
9950 Substitutes @var{.SUFFIX} for the suffixes of a matched switch's args
9951 when it is subsequently output with @samp{%*}. @var{SUFFIX} is
9952 terminated by the next space or %.
9955 Marks the argument containing or following the @samp{%w} as the
9956 designated output file of this compilation. This puts the argument
9957 into the sequence of arguments that @samp{%o} will substitute later.
9960 Substitutes the names of all the output files, with spaces
9961 automatically placed around them. You should write spaces
9962 around the @samp{%o} as well or the results are undefined.
9963 @samp{%o} is for use in the specs for running the linker.
9964 Input files whose names have no recognized suffix are not compiled
9965 at all, but they are included among the output files, so they will
9969 Substitutes the suffix for object files. Note that this is
9970 handled specially when it immediately follows @samp{%g, %u, or %U},
9971 because of the need for those to form complete file names. The
9972 handling is such that @samp{%O} is treated exactly as if it had already
9973 been substituted, except that @samp{%g, %u, and %U} do not currently
9974 support additional @var{suffix} characters following @samp{%O} as they would
9975 following, for example, @samp{.o}.
9978 Substitutes the standard macro predefinitions for the
9979 current target machine. Use this when running @code{cpp}.
9982 Like @samp{%p}, but puts @samp{__} before and after the name of each
9983 predefined macro, except for macros that start with @samp{__} or with
9984 @samp{_@var{L}}, where @var{L} is an uppercase letter. This is for ISO
9988 Substitute any of @option{-iprefix} (made from @env{GCC_EXEC_PREFIX}),
9989 @option{-isysroot} (made from @env{TARGET_SYSTEM_ROOT}),
9990 @option{-isystem} (made from @env{COMPILER_PATH} and @option{-B} options)
9991 and @option{-imultilib} as necessary.
9994 Current argument is the name of a library or startup file of some sort.
9995 Search for that file in a standard list of directories and substitute
9996 the full name found. The current working directory is included in the
9997 list of directories scanned.
10000 Current argument is the name of a linker script. Search for that file
10001 in the current list of directories to scan for libraries. If the file
10002 is located insert a @option{--script} option into the command line
10003 followed by the full path name found. If the file is not found then
10004 generate an error message. Note: the current working directory is not
10008 Print @var{str} as an error message. @var{str} is terminated by a newline.
10009 Use this when inconsistent options are detected.
10011 @item %(@var{name})
10012 Substitute the contents of spec string @var{name} at this point.
10014 @item %x@{@var{option}@}
10015 Accumulate an option for @samp{%X}.
10018 Output the accumulated linker options specified by @option{-Wl} or a @samp{%x}
10022 Output the accumulated assembler options specified by @option{-Wa}.
10025 Output the accumulated preprocessor options specified by @option{-Wp}.
10028 Process the @code{asm} spec. This is used to compute the
10029 switches to be passed to the assembler.
10032 Process the @code{asm_final} spec. This is a spec string for
10033 passing switches to an assembler post-processor, if such a program is
10037 Process the @code{link} spec. This is the spec for computing the
10038 command line passed to the linker. Typically it will make use of the
10039 @samp{%L %G %S %D and %E} sequences.
10042 Dump out a @option{-L} option for each directory that GCC believes might
10043 contain startup files. If the target supports multilibs then the
10044 current multilib directory will be prepended to each of these paths.
10047 Process the @code{lib} spec. This is a spec string for deciding which
10048 libraries should be included on the command line to the linker.
10051 Process the @code{libgcc} spec. This is a spec string for deciding
10052 which GCC support library should be included on the command line to the linker.
10055 Process the @code{startfile} spec. This is a spec for deciding which
10056 object files should be the first ones passed to the linker. Typically
10057 this might be a file named @file{crt0.o}.
10060 Process the @code{endfile} spec. This is a spec string that specifies
10061 the last object files that will be passed to the linker.
10064 Process the @code{cpp} spec. This is used to construct the arguments
10065 to be passed to the C preprocessor.
10068 Process the @code{cc1} spec. This is used to construct the options to be
10069 passed to the actual C compiler (@samp{cc1}).
10072 Process the @code{cc1plus} spec. This is used to construct the options to be
10073 passed to the actual C++ compiler (@samp{cc1plus}).
10076 Substitute the variable part of a matched option. See below.
10077 Note that each comma in the substituted string is replaced by
10081 Remove all occurrences of @code{-S} from the command line. Note---this
10082 command is position dependent. @samp{%} commands in the spec string
10083 before this one will see @code{-S}, @samp{%} commands in the spec string
10084 after this one will not.
10086 @item %:@var{function}(@var{args})
10087 Call the named function @var{function}, passing it @var{args}.
10088 @var{args} is first processed as a nested spec string, then split
10089 into an argument vector in the usual fashion. The function returns
10090 a string which is processed as if it had appeared literally as part
10091 of the current spec.
10093 The following built-in spec functions are provided:
10096 @item @code{getenv}
10097 The @code{getenv} spec function takes two arguments: an environment
10098 variable name and a string. If the environment variable is not
10099 defined, a fatal error is issued. Otherwise, the return value is the
10100 value of the environment variable concatenated with the string. For
10101 example, if @env{TOPDIR} is defined as @file{/path/to/top}, then:
10104 %:getenv(TOPDIR /include)
10107 expands to @file{/path/to/top/include}.
10109 @item @code{if-exists}
10110 The @code{if-exists} spec function takes one argument, an absolute
10111 pathname to a file. If the file exists, @code{if-exists} returns the
10112 pathname. Here is a small example of its usage:
10116 crt0%O%s %:if-exists(crti%O%s) crtbegin%O%s
10119 @item @code{if-exists-else}
10120 The @code{if-exists-else} spec function is similar to the @code{if-exists}
10121 spec function, except that it takes two arguments. The first argument is
10122 an absolute pathname to a file. If the file exists, @code{if-exists-else}
10123 returns the pathname. If it does not exist, it returns the second argument.
10124 This way, @code{if-exists-else} can be used to select one file or another,
10125 based on the existence of the first. Here is a small example of its usage:
10129 crt0%O%s %:if-exists(crti%O%s) \
10130 %:if-exists-else(crtbeginT%O%s crtbegin%O%s)
10133 @item @code{replace-outfile}
10134 The @code{replace-outfile} spec function takes two arguments. It looks for the
10135 first argument in the outfiles array and replaces it with the second argument. Here
10136 is a small example of its usage:
10139 %@{fgnu-runtime:%:replace-outfile(-lobjc -lobjc-gnu)@}
10142 @item @code{remove-outfile}
10143 The @code{remove-outfile} spec function takes one argument. It looks for the
10144 first argument in the outfiles array and removes it. Here is a small example
10148 %:remove-outfile(-lm)
10151 @item @code{pass-through-libs}
10152 The @code{pass-through-libs} spec function takes any number of arguments. It
10153 finds any @option{-l} options and any non-options ending in ".a" (which it
10154 assumes are the names of linker input library archive files) and returns a
10155 result containing all the found arguments each prepended by
10156 @option{-plugin-opt=-pass-through=} and joined by spaces. This list is
10157 intended to be passed to the LTO linker plugin.
10160 %:pass-through-libs(%G %L %G)
10163 @item @code{print-asm-header}
10164 The @code{print-asm-header} function takes no arguments and simply
10165 prints a banner like:
10171 Use "-Wa,OPTION" to pass "OPTION" to the assembler.
10174 It is used to separate compiler options from assembler options
10175 in the @option{--target-help} output.
10178 @item %@{@code{S}@}
10179 Substitutes the @code{-S} switch, if that switch was given to GCC@.
10180 If that switch was not specified, this substitutes nothing. Note that
10181 the leading dash is omitted when specifying this option, and it is
10182 automatically inserted if the substitution is performed. Thus the spec
10183 string @samp{%@{foo@}} would match the command-line option @option{-foo}
10184 and would output the command-line option @option{-foo}.
10186 @item %W@{@code{S}@}
10187 Like %@{@code{S}@} but mark last argument supplied within as a file to be
10188 deleted on failure.
10190 @item %@{@code{S}*@}
10191 Substitutes all the switches specified to GCC whose names start
10192 with @code{-S}, but which also take an argument. This is used for
10193 switches like @option{-o}, @option{-D}, @option{-I}, etc.
10194 GCC considers @option{-o foo} as being
10195 one switch whose names starts with @samp{o}. %@{o*@} would substitute this
10196 text, including the space. Thus two arguments would be generated.
10198 @item %@{@code{S}*&@code{T}*@}
10199 Like %@{@code{S}*@}, but preserve order of @code{S} and @code{T} options
10200 (the order of @code{S} and @code{T} in the spec is not significant).
10201 There can be any number of ampersand-separated variables; for each the
10202 wild card is optional. Useful for CPP as @samp{%@{D*&U*&A*@}}.
10204 @item %@{@code{S}:@code{X}@}
10205 Substitutes @code{X}, if the @samp{-S} switch was given to GCC@.
10207 @item %@{!@code{S}:@code{X}@}
10208 Substitutes @code{X}, if the @samp{-S} switch was @emph{not} given to GCC@.
10210 @item %@{@code{S}*:@code{X}@}
10211 Substitutes @code{X} if one or more switches whose names start with
10212 @code{-S} are specified to GCC@. Normally @code{X} is substituted only
10213 once, no matter how many such switches appeared. However, if @code{%*}
10214 appears somewhere in @code{X}, then @code{X} will be substituted once
10215 for each matching switch, with the @code{%*} replaced by the part of
10216 that switch that matched the @code{*}.
10218 @item %@{.@code{S}:@code{X}@}
10219 Substitutes @code{X}, if processing a file with suffix @code{S}.
10221 @item %@{!.@code{S}:@code{X}@}
10222 Substitutes @code{X}, if @emph{not} processing a file with suffix @code{S}.
10224 @item %@{,@code{S}:@code{X}@}
10225 Substitutes @code{X}, if processing a file for language @code{S}.
10227 @item %@{!,@code{S}:@code{X}@}
10228 Substitutes @code{X}, if not processing a file for language @code{S}.
10230 @item %@{@code{S}|@code{P}:@code{X}@}
10231 Substitutes @code{X} if either @code{-S} or @code{-P} was given to
10232 GCC@. This may be combined with @samp{!}, @samp{.}, @samp{,}, and
10233 @code{*} sequences as well, although they have a stronger binding than
10234 the @samp{|}. If @code{%*} appears in @code{X}, all of the
10235 alternatives must be starred, and only the first matching alternative
10238 For example, a spec string like this:
10241 %@{.c:-foo@} %@{!.c:-bar@} %@{.c|d:-baz@} %@{!.c|d:-boggle@}
10244 will output the following command-line options from the following input
10245 command-line options:
10250 -d fred.c -foo -baz -boggle
10251 -d jim.d -bar -baz -boggle
10254 @item %@{S:X; T:Y; :D@}
10256 If @code{S} was given to GCC, substitutes @code{X}; else if @code{T} was
10257 given to GCC, substitutes @code{Y}; else substitutes @code{D}. There can
10258 be as many clauses as you need. This may be combined with @code{.},
10259 @code{,}, @code{!}, @code{|}, and @code{*} as needed.
10264 The conditional text @code{X} in a %@{@code{S}:@code{X}@} or similar
10265 construct may contain other nested @samp{%} constructs or spaces, or
10266 even newlines. They are processed as usual, as described above.
10267 Trailing white space in @code{X} is ignored. White space may also
10268 appear anywhere on the left side of the colon in these constructs,
10269 except between @code{.} or @code{*} and the corresponding word.
10271 The @option{-O}, @option{-f}, @option{-m}, and @option{-W} switches are
10272 handled specifically in these constructs. If another value of
10273 @option{-O} or the negated form of a @option{-f}, @option{-m}, or
10274 @option{-W} switch is found later in the command line, the earlier
10275 switch value is ignored, except with @{@code{S}*@} where @code{S} is
10276 just one letter, which passes all matching options.
10278 The character @samp{|} at the beginning of the predicate text is used to
10279 indicate that a command should be piped to the following command, but
10280 only if @option{-pipe} is specified.
10282 It is built into GCC which switches take arguments and which do not.
10283 (You might think it would be useful to generalize this to allow each
10284 compiler's spec to say which switches take arguments. But this cannot
10285 be done in a consistent fashion. GCC cannot even decide which input
10286 files have been specified without knowing which switches take arguments,
10287 and it must know which input files to compile in order to tell which
10290 GCC also knows implicitly that arguments starting in @option{-l} are to be
10291 treated as compiler output files, and passed to the linker in their
10292 proper position among the other output files.
10294 @c man begin OPTIONS
10296 @node Target Options
10297 @section Specifying Target Machine and Compiler Version
10298 @cindex target options
10299 @cindex cross compiling
10300 @cindex specifying machine version
10301 @cindex specifying compiler version and target machine
10302 @cindex compiler version, specifying
10303 @cindex target machine, specifying
10305 The usual way to run GCC is to run the executable called @command{gcc}, or
10306 @command{@var{machine}-gcc} when cross-compiling, or
10307 @command{@var{machine}-gcc-@var{version}} to run a version other than the
10308 one that was installed last.
10310 @node Submodel Options
10311 @section Hardware Models and Configurations
10312 @cindex submodel options
10313 @cindex specifying hardware config
10314 @cindex hardware models and configurations, specifying
10315 @cindex machine dependent options
10317 Each target machine types can have its own
10318 special options, starting with @samp{-m}, to choose among various
10319 hardware models or configurations---for example, 68010 vs 68020,
10320 floating coprocessor or none. A single installed version of the
10321 compiler can compile for any model or configuration, according to the
10324 Some configurations of the compiler also support additional special
10325 options, usually for compatibility with other compilers on the same
10328 @c This list is ordered alphanumerically by subsection name.
10329 @c It should be the same order and spelling as these options are listed
10330 @c in Machine Dependent Options
10333 * Adapteva Epiphany Options::
10336 * Blackfin Options::
10341 * DEC Alpha Options::
10342 * DEC Alpha/VMS Options::
10345 * GNU/Linux Options::
10348 * i386 and x86-64 Options::
10349 * i386 and x86-64 Windows Options::
10351 * IA-64/VMS Options::
10358 * MicroBlaze Options::
10361 * MN10300 Options::
10363 * picoChip Options::
10364 * PowerPC Options::
10366 * RS/6000 and PowerPC Options::
10368 * S/390 and zSeries Options::
10371 * Solaris 2 Options::
10374 * System V Options::
10375 * TILE-Gx Options::
10376 * TILEPro Options::
10379 * VxWorks Options::
10381 * Xstormy16 Options::
10383 * zSeries Options::
10386 @node Adapteva Epiphany Options
10387 @subsection Adapteva Epiphany Options
10389 These @samp{-m} options are defined for Adapteva Epiphany:
10392 @item -mhalf-reg-file
10393 @opindex mhalf-reg-file
10394 Don't allocate any register in the range @code{r32}@dots{}@code{r63}.
10395 That allows code to run on hardware variants that lack these registers.
10397 @item -mprefer-short-insn-regs
10398 @opindex mprefer-short-insn-regs
10399 Preferrentially allocate registers that allow short instruction generation.
10400 This can result in increasesd instruction count, so if this reduces or
10401 increases code size might vary from case to case.
10403 @item -mbranch-cost=@var{num}
10404 @opindex mbranch-cost
10405 Set the cost of branches to roughly @var{num} ``simple'' instructions.
10406 This cost is only a heuristic and is not guaranteed to produce
10407 consistent results across releases.
10411 Enable the generation of conditional moves.
10413 @item -mnops=@var{num}
10415 Emit @var{num} nops before every other generated instruction.
10417 @item -mno-soft-cmpsf
10418 @opindex mno-soft-cmpsf
10419 For single-precision floating-point comparisons, emit an fsub instruction
10420 and test the flags. This is faster than a software comparison, but can
10421 get incorrect results in the presence of NaNs, or when two different small
10422 numbers are compared such that their difference is calculated as zero.
10423 The default is @option{-msoft-cmpsf}, which uses slower, but IEEE-compliant,
10424 software comparisons.
10426 @item -mstack-offset=@var{num}
10427 @opindex mstack-offset
10428 Set the offset between the top of the stack and the stack pointer.
10429 E.g., a value of 8 means that the eight bytes in the range sp+0@dots{}sp+7
10430 can be used by leaf functions without stack allocation.
10431 Values other than @samp{8} or @samp{16} are untested and unlikely to work.
10432 Note also that this option changes the ABI, compiling a program with a
10433 different stack offset than the libraries have been compiled with
10434 will generally not work.
10435 This option can be useful if you want to evaluate if a different stack
10436 offset would give you better code, but to actually use a different stack
10437 offset to build working programs, it is recommended to configure the
10438 toolchain with the appropriate @samp{--with-stack-offset=@var{num}} option.
10440 @item -mno-round-nearest
10441 @opindex mno-round-nearest
10442 Make the scheduler assume that the rounding mode has been set to
10443 truncating. The default is @option{-mround-nearest}.
10446 @opindex mlong-calls
10447 If not otherwise specified by an attribute, assume all calls might be beyond
10448 the offset range of the b / bl instructions, and therefore load the
10449 function address into a register before performing a (otherwise direct) call.
10450 This is the default.
10452 @item -mshort-calls
10453 @opindex short-calls
10454 If not otherwise specified by an attribute, assume all direct calls are
10455 in the range of the b / bl instructions, so use these instructions
10456 for direct calls. The default is @option{-mlong-calls}.
10460 Assume addresses can be loaded as 16-bit unsigned values. This does not
10461 apply to function addresses for which @option{-mlong-calls} semantics
10464 @item -mfp-mode=@var{mode}
10466 Set the prevailing mode of the floating-point unit.
10467 This determines the floating-point mode that is provided and expected
10468 at function call and return time. Making this mode match the mode you
10469 predominantly need at function start can make your programs smaller and
10470 faster by avoiding unnecessary mode switches.
10472 @var{mode} can be set to one the following values:
10476 Any mode at function entry is valid, and retained or restored when
10477 the function returns, and when it calls other functions.
10478 This mode is useful for compiling libraries or other compilation units
10479 you might want to incorporate into different programs with different
10480 prevailing FPU modes, and the convenience of being able to use a single
10481 object file outweighs the size and speed overhead for any extra
10482 mode switching that might be needed, compared with what would be needed
10483 with a more specific choice of prevailing FPU mode.
10486 This is the mode used for floating-point calculations with
10487 truncating (i.e.@: round towards zero) rounding mode. That includes
10488 conversion from floating point to integer.
10490 @item round-nearest
10491 This is the mode used for floating-point calculations with
10492 round-to-nearest-or-even rounding mode.
10495 This is the mode used to perform integer calculations in the FPU, e.g.@:
10496 integer multiply, or integer multiply-and-accumulate.
10499 The default is @option{-mfp-mode=caller}
10501 @item -mnosplit-lohi
10502 @opindex mnosplit-lohi
10504 @opindex mno-postinc
10505 @item -mno-postmodify
10506 @opindex mno-postmodify
10507 Code generation tweaks that disable, respectively, splitting of 32-bit
10508 loads, generation of post-increment addresses, and generation of
10509 post-modify addresses. The defaults are @option{msplit-lohi},
10510 @option{-mpost-inc}, and @option{-mpost-modify}.
10512 @item -mnovect-double
10513 @opindex mno-vect-double
10514 Change the preferred SIMD mode to SImode. The default is
10515 @option{-mvect-double}, which uses DImode as preferred SIMD mode.
10517 @item -max-vect-align=@var{num}
10518 @opindex max-vect-align
10519 The maximum alignment for SIMD vector mode types.
10520 @var{num} may be 4 or 8. The default is 8.
10521 Note that this is an ABI change, even though many library function
10522 interfaces will be unaffected, if they don't use SIMD vector modes
10523 in places where they affect size and/or alignment of relevant types.
10525 @item -msplit-vecmove-early
10526 @opindex msplit-vecmove-early
10527 Split vector moves into single word moves before reload. In theory this
10528 could give better register allocation, but so far the reverse seems to be
10529 generally the case.
10531 @item -m1reg-@var{reg}
10533 Specify a register to hold the constant @minus{}1, which makes loading small negative
10534 constants and certain bitmasks faster.
10535 Allowable values for reg are r43 and r63, which specify to use that register
10536 as a fixed register, and none, which means that no register is used for this
10537 purpose. The default is @option{-m1reg-none}.
10542 @subsection ARM Options
10543 @cindex ARM options
10545 These @samp{-m} options are defined for Advanced RISC Machines (ARM)
10549 @item -mabi=@var{name}
10551 Generate code for the specified ABI@. Permissible values are: @samp{apcs-gnu},
10552 @samp{atpcs}, @samp{aapcs}, @samp{aapcs-linux} and @samp{iwmmxt}.
10555 @opindex mapcs-frame
10556 Generate a stack frame that is compliant with the ARM Procedure Call
10557 Standard for all functions, even if this is not strictly necessary for
10558 correct execution of the code. Specifying @option{-fomit-frame-pointer}
10559 with this option will cause the stack frames not to be generated for
10560 leaf functions. The default is @option{-mno-apcs-frame}.
10564 This is a synonym for @option{-mapcs-frame}.
10567 @c not currently implemented
10568 @item -mapcs-stack-check
10569 @opindex mapcs-stack-check
10570 Generate code to check the amount of stack space available upon entry to
10571 every function (that actually uses some stack space). If there is
10572 insufficient space available then either the function
10573 @samp{__rt_stkovf_split_small} or @samp{__rt_stkovf_split_big} will be
10574 called, depending upon the amount of stack space required. The runtime
10575 system is required to provide these functions. The default is
10576 @option{-mno-apcs-stack-check}, since this produces smaller code.
10578 @c not currently implemented
10580 @opindex mapcs-float
10581 Pass floating-point arguments using the floating-point registers. This is
10582 one of the variants of the APCS@. This option is recommended if the
10583 target hardware has a floating-point unit or if a lot of floating-point
10584 arithmetic is going to be performed by the code. The default is
10585 @option{-mno-apcs-float}, since integer only code is slightly increased in
10586 size if @option{-mapcs-float} is used.
10588 @c not currently implemented
10589 @item -mapcs-reentrant
10590 @opindex mapcs-reentrant
10591 Generate reentrant, position independent code. The default is
10592 @option{-mno-apcs-reentrant}.
10595 @item -mthumb-interwork
10596 @opindex mthumb-interwork
10597 Generate code that supports calling between the ARM and Thumb
10598 instruction sets. Without this option, on pre-v5 architectures, the
10599 two instruction sets cannot be reliably used inside one program. The
10600 default is @option{-mno-thumb-interwork}, since slightly larger code
10601 is generated when @option{-mthumb-interwork} is specified. In AAPCS
10602 configurations this option is meaningless.
10604 @item -mno-sched-prolog
10605 @opindex mno-sched-prolog
10606 Prevent the reordering of instructions in the function prologue, or the
10607 merging of those instruction with the instructions in the function's
10608 body. This means that all functions will start with a recognizable set
10609 of instructions (or in fact one of a choice from a small set of
10610 different function prologues), and this information can be used to
10611 locate the start if functions inside an executable piece of code. The
10612 default is @option{-msched-prolog}.
10614 @item -mfloat-abi=@var{name}
10615 @opindex mfloat-abi
10616 Specifies which floating-point ABI to use. Permissible values
10617 are: @samp{soft}, @samp{softfp} and @samp{hard}.
10619 Specifying @samp{soft} causes GCC to generate output containing
10620 library calls for floating-point operations.
10621 @samp{softfp} allows the generation of code using hardware floating-point
10622 instructions, but still uses the soft-float calling conventions.
10623 @samp{hard} allows generation of floating-point instructions
10624 and uses FPU-specific calling conventions.
10626 The default depends on the specific target configuration. Note that
10627 the hard-float and soft-float ABIs are not link-compatible; you must
10628 compile your entire program with the same ABI, and link with a
10629 compatible set of libraries.
10631 @item -mlittle-endian
10632 @opindex mlittle-endian
10633 Generate code for a processor running in little-endian mode. This is
10634 the default for all standard configurations.
10637 @opindex mbig-endian
10638 Generate code for a processor running in big-endian mode; the default is
10639 to compile code for a little-endian processor.
10641 @item -mwords-little-endian
10642 @opindex mwords-little-endian
10643 This option only applies when generating code for big-endian processors.
10644 Generate code for a little-endian word order but a big-endian byte
10645 order. That is, a byte order of the form @samp{32107654}. Note: this
10646 option should only be used if you require compatibility with code for
10647 big-endian ARM processors generated by versions of the compiler prior to
10648 2.8. This option is now deprecated.
10650 @item -mcpu=@var{name}
10652 This specifies the name of the target ARM processor. GCC uses this name
10653 to determine what kind of instructions it can emit when generating
10654 assembly code. Permissible names are: @samp{arm2}, @samp{arm250},
10655 @samp{arm3}, @samp{arm6}, @samp{arm60}, @samp{arm600}, @samp{arm610},
10656 @samp{arm620}, @samp{arm7}, @samp{arm7m}, @samp{arm7d}, @samp{arm7dm},
10657 @samp{arm7di}, @samp{arm7dmi}, @samp{arm70}, @samp{arm700},
10658 @samp{arm700i}, @samp{arm710}, @samp{arm710c}, @samp{arm7100},
10660 @samp{arm7500}, @samp{arm7500fe}, @samp{arm7tdmi}, @samp{arm7tdmi-s},
10661 @samp{arm710t}, @samp{arm720t}, @samp{arm740t},
10662 @samp{strongarm}, @samp{strongarm110}, @samp{strongarm1100},
10663 @samp{strongarm1110},
10664 @samp{arm8}, @samp{arm810}, @samp{arm9}, @samp{arm9e}, @samp{arm920},
10665 @samp{arm920t}, @samp{arm922t}, @samp{arm946e-s}, @samp{arm966e-s},
10666 @samp{arm968e-s}, @samp{arm926ej-s}, @samp{arm940t}, @samp{arm9tdmi},
10667 @samp{arm10tdmi}, @samp{arm1020t}, @samp{arm1026ej-s},
10668 @samp{arm10e}, @samp{arm1020e}, @samp{arm1022e},
10669 @samp{arm1136j-s}, @samp{arm1136jf-s}, @samp{mpcore}, @samp{mpcorenovfp},
10670 @samp{arm1156t2-s}, @samp{arm1156t2f-s}, @samp{arm1176jz-s}, @samp{arm1176jzf-s},
10671 @samp{cortex-a5}, @samp{cortex-a7}, @samp{cortex-a8}, @samp{cortex-a9},
10672 @samp{cortex-a15}, @samp{cortex-r4}, @samp{cortex-r4f}, @samp{cortex-r5},
10673 @samp{cortex-m4}, @samp{cortex-m3},
10676 @samp{xscale}, @samp{iwmmxt}, @samp{iwmmxt2}, @samp{ep9312},
10677 @samp{fa526}, @samp{fa626},
10678 @samp{fa606te}, @samp{fa626te}, @samp{fmp626}, @samp{fa726te}.
10681 @option{-mcpu=generic-@var{arch}} is also permissible, and is
10682 equivalent to @option{-march=@var{arch} -mtune=generic-@var{arch}}.
10683 See @option{-mtune} for more information.
10685 @option{-mcpu=native} causes the compiler to auto-detect the CPU
10686 of the build computer. At present, this feature is only supported on
10687 Linux, and not all architectures are recognized. If the auto-detect is
10688 unsuccessful the option has no effect.
10690 @item -mtune=@var{name}
10692 This option is very similar to the @option{-mcpu=} option, except that
10693 instead of specifying the actual target processor type, and hence
10694 restricting which instructions can be used, it specifies that GCC should
10695 tune the performance of the code as if the target were of the type
10696 specified in this option, but still choosing the instructions that it
10697 will generate based on the CPU specified by a @option{-mcpu=} option.
10698 For some ARM implementations better performance can be obtained by using
10701 @option{-mtune=generic-@var{arch}} specifies that GCC should tune the
10702 performance for a blend of processors within architecture @var{arch}.
10703 The aim is to generate code that run well on the current most popular
10704 processors, balancing between optimizations that benefit some CPUs in the
10705 range, and avoiding performance pitfalls of other CPUs. The effects of
10706 this option may change in future GCC versions as CPU models come and go.
10708 @option{-mtune=native} causes the compiler to auto-detect the CPU
10709 of the build computer. At present, this feature is only supported on
10710 Linux, and not all architectures are recognized. If the auto-detect is
10711 unsuccessful the option has no effect.
10713 @item -march=@var{name}
10715 This specifies the name of the target ARM architecture. GCC uses this
10716 name to determine what kind of instructions it can emit when generating
10717 assembly code. This option can be used in conjunction with or instead
10718 of the @option{-mcpu=} option. Permissible names are: @samp{armv2},
10719 @samp{armv2a}, @samp{armv3}, @samp{armv3m}, @samp{armv4}, @samp{armv4t},
10720 @samp{armv5}, @samp{armv5t}, @samp{armv5e}, @samp{armv5te},
10721 @samp{armv6}, @samp{armv6j},
10722 @samp{armv6t2}, @samp{armv6z}, @samp{armv6zk}, @samp{armv6-m},
10723 @samp{armv7}, @samp{armv7-a}, @samp{armv7-r}, @samp{armv7-m},
10724 @samp{iwmmxt}, @samp{iwmmxt2}, @samp{ep9312}.
10726 @option{-march=native} causes the compiler to auto-detect the architecture
10727 of the build computer. At present, this feature is only supported on
10728 Linux, and not all architectures are recognized. If the auto-detect is
10729 unsuccessful the option has no effect.
10731 @item -mfpu=@var{name}
10732 @itemx -mfpe=@var{number}
10733 @itemx -mfp=@var{number}
10737 This specifies what floating-point hardware (or hardware emulation) is
10738 available on the target. Permissible names are: @samp{fpa}, @samp{fpe2},
10739 @samp{fpe3}, @samp{maverick}, @samp{vfp}, @samp{vfpv3}, @samp{vfpv3-fp16},
10740 @samp{vfpv3-d16}, @samp{vfpv3-d16-fp16}, @samp{vfpv3xd}, @samp{vfpv3xd-fp16},
10741 @samp{neon}, @samp{neon-fp16}, @samp{vfpv4}, @samp{vfpv4-d16},
10742 @samp{fpv4-sp-d16} and @samp{neon-vfpv4}.
10743 @option{-mfp} and @option{-mfpe} are synonyms for
10744 @option{-mfpu}=@samp{fpe}@var{number}, for compatibility with older versions
10747 If @option{-msoft-float} is specified this specifies the format of
10748 floating-point values.
10750 If the selected floating-point hardware includes the NEON extension
10751 (e.g. @option{-mfpu}=@samp{neon}), note that floating-point
10752 operations will not be used by GCC's auto-vectorization pass unless
10753 @option{-funsafe-math-optimizations} is also specified. This is
10754 because NEON hardware does not fully implement the IEEE 754 standard for
10755 floating-point arithmetic (in particular denormal values are treated as
10756 zero), so the use of NEON instructions may lead to a loss of precision.
10758 @item -mfp16-format=@var{name}
10759 @opindex mfp16-format
10760 Specify the format of the @code{__fp16} half-precision floating-point type.
10761 Permissible names are @samp{none}, @samp{ieee}, and @samp{alternative};
10762 the default is @samp{none}, in which case the @code{__fp16} type is not
10763 defined. @xref{Half-Precision}, for more information.
10765 @item -mstructure-size-boundary=@var{n}
10766 @opindex mstructure-size-boundary
10767 The size of all structures and unions will be rounded up to a multiple
10768 of the number of bits set by this option. Permissible values are 8, 32
10769 and 64. The default value varies for different toolchains. For the COFF
10770 targeted toolchain the default value is 8. A value of 64 is only allowed
10771 if the underlying ABI supports it.
10773 Specifying the larger number can produce faster, more efficient code, but
10774 can also increase the size of the program. Different values are potentially
10775 incompatible. Code compiled with one value cannot necessarily expect to
10776 work with code or libraries compiled with another value, if they exchange
10777 information using structures or unions.
10779 @item -mabort-on-noreturn
10780 @opindex mabort-on-noreturn
10781 Generate a call to the function @code{abort} at the end of a
10782 @code{noreturn} function. It will be executed if the function tries to
10786 @itemx -mno-long-calls
10787 @opindex mlong-calls
10788 @opindex mno-long-calls
10789 Tells the compiler to perform function calls by first loading the
10790 address of the function into a register and then performing a subroutine
10791 call on this register. This switch is needed if the target function
10792 will lie outside of the 64 megabyte addressing range of the offset based
10793 version of subroutine call instruction.
10795 Even if this switch is enabled, not all function calls will be turned
10796 into long calls. The heuristic is that static functions, functions
10797 that have the @samp{short-call} attribute, functions that are inside
10798 the scope of a @samp{#pragma no_long_calls} directive and functions whose
10799 definitions have already been compiled within the current compilation
10800 unit, will not be turned into long calls. The exception to this rule is
10801 that weak function definitions, functions with the @samp{long-call}
10802 attribute or the @samp{section} attribute, and functions that are within
10803 the scope of a @samp{#pragma long_calls} directive, will always be
10804 turned into long calls.
10806 This feature is not enabled by default. Specifying
10807 @option{-mno-long-calls} will restore the default behavior, as will
10808 placing the function calls within the scope of a @samp{#pragma
10809 long_calls_off} directive. Note these switches have no effect on how
10810 the compiler generates code to handle function calls via function
10813 @item -msingle-pic-base
10814 @opindex msingle-pic-base
10815 Treat the register used for PIC addressing as read-only, rather than
10816 loading it in the prologue for each function. The runtime system is
10817 responsible for initializing this register with an appropriate value
10818 before execution begins.
10820 @item -mpic-register=@var{reg}
10821 @opindex mpic-register
10822 Specify the register to be used for PIC addressing. The default is R10
10823 unless stack-checking is enabled, when R9 is used.
10825 @item -mcirrus-fix-invalid-insns
10826 @opindex mcirrus-fix-invalid-insns
10827 @opindex mno-cirrus-fix-invalid-insns
10828 Insert NOPs into the instruction stream to in order to work around
10829 problems with invalid Maverick instruction combinations. This option
10830 is only valid if the @option{-mcpu=ep9312} option has been used to
10831 enable generation of instructions for the Cirrus Maverick floating-point
10832 co-processor. This option is not enabled by default, since the
10833 problem is only present in older Maverick implementations. The default
10834 can be re-enabled by use of the @option{-mno-cirrus-fix-invalid-insns}
10837 @item -mpoke-function-name
10838 @opindex mpoke-function-name
10839 Write the name of each function into the text section, directly
10840 preceding the function prologue. The generated code is similar to this:
10844 .ascii "arm_poke_function_name", 0
10847 .word 0xff000000 + (t1 - t0)
10848 arm_poke_function_name
10850 stmfd sp!, @{fp, ip, lr, pc@}
10854 When performing a stack backtrace, code can inspect the value of
10855 @code{pc} stored at @code{fp + 0}. If the trace function then looks at
10856 location @code{pc - 12} and the top 8 bits are set, then we know that
10857 there is a function name embedded immediately preceding this location
10858 and has length @code{((pc[-3]) & 0xff000000)}.
10865 Select between generating code that executes in ARM and Thumb
10866 states. The default for most configurations is to generate code
10867 that executes in ARM state, but the default can be changed by
10868 configuring GCC with the @option{--with-mode=}@var{state}
10872 @opindex mtpcs-frame
10873 Generate a stack frame that is compliant with the Thumb Procedure Call
10874 Standard for all non-leaf functions. (A leaf function is one that does
10875 not call any other functions.) The default is @option{-mno-tpcs-frame}.
10877 @item -mtpcs-leaf-frame
10878 @opindex mtpcs-leaf-frame
10879 Generate a stack frame that is compliant with the Thumb Procedure Call
10880 Standard for all leaf functions. (A leaf function is one that does
10881 not call any other functions.) The default is @option{-mno-apcs-leaf-frame}.
10883 @item -mcallee-super-interworking
10884 @opindex mcallee-super-interworking
10885 Gives all externally visible functions in the file being compiled an ARM
10886 instruction set header which switches to Thumb mode before executing the
10887 rest of the function. This allows these functions to be called from
10888 non-interworking code. This option is not valid in AAPCS configurations
10889 because interworking is enabled by default.
10891 @item -mcaller-super-interworking
10892 @opindex mcaller-super-interworking
10893 Allows calls via function pointers (including virtual functions) to
10894 execute correctly regardless of whether the target code has been
10895 compiled for interworking or not. There is a small overhead in the cost
10896 of executing a function pointer if this option is enabled. This option
10897 is not valid in AAPCS configurations because interworking is enabled
10900 @item -mtp=@var{name}
10902 Specify the access model for the thread local storage pointer. The valid
10903 models are @option{soft}, which generates calls to @code{__aeabi_read_tp},
10904 @option{cp15}, which fetches the thread pointer from @code{cp15} directly
10905 (supported in the arm6k architecture), and @option{auto}, which uses the
10906 best available method for the selected processor. The default setting is
10909 @item -mtls-dialect=@var{dialect}
10910 @opindex mtls-dialect
10911 Specify the dialect to use for accessing thread local storage. Two
10912 dialects are supported --- @option{gnu} and @option{gnu2}. The
10913 @option{gnu} dialect selects the original GNU scheme for supporting
10914 local and global dynamic TLS models. The @option{gnu2} dialect
10915 selects the GNU descriptor scheme, which provides better performance
10916 for shared libraries. The GNU descriptor scheme is compatible with
10917 the original scheme, but does require new assembler, linker and
10918 library support. Initial and local exec TLS models are unaffected by
10919 this option and always use the original scheme.
10921 @item -mword-relocations
10922 @opindex mword-relocations
10923 Only generate absolute relocations on word-sized values (i.e. R_ARM_ABS32).
10924 This is enabled by default on targets (uClinux, SymbianOS) where the runtime
10925 loader imposes this restriction, and when @option{-fpic} or @option{-fPIC}
10928 @item -mfix-cortex-m3-ldrd
10929 @opindex mfix-cortex-m3-ldrd
10930 Some Cortex-M3 cores can cause data corruption when @code{ldrd} instructions
10931 with overlapping destination and base registers are used. This option avoids
10932 generating these instructions. This option is enabled by default when
10933 @option{-mcpu=cortex-m3} is specified.
10935 @item -munaligned-access
10936 @itemx -mno-unaligned-access
10937 @opindex munaligned-access
10938 @opindex mno-unaligned-access
10939 Enables (or disables) reading and writing of 16- and 32- bit values
10940 from addresses that are not 16- or 32- bit aligned. By default
10941 unaligned access is disabled for all pre-ARMv6 and all ARMv6-M
10942 architectures, and enabled for all other architectures. If unaligned
10943 access is not enabled then words in packed data structures will be
10944 accessed a byte at a time.
10946 The ARM attribute @code{Tag_CPU_unaligned_access} will be set in the
10947 generated object file to either true or false, depending upon the
10948 setting of this option. If unaligned access is enabled then the
10949 preprocessor symbol @code{__ARM_FEATURE_UNALIGNED} will also be
10955 @subsection AVR Options
10956 @cindex AVR Options
10959 @item -mmcu=@var{mcu}
10961 Specify Atmel AVR instruction set architectures (ISA) or MCU type.
10963 The default for this option is@tie{}@code{avr2}.
10965 GCC supports the following AVR devices and ISAs:
10970 ``Classic'' devices with up to 8@tie{}KiB of program memory.
10971 @*@var{mcu}@tie{}= @code{attiny22}, @code{attiny26}, @code{at90c8534},
10972 @code{at90s2313}, @code{at90s2323}, @code{at90s2333},
10973 @code{at90s2343}, @code{at90s4414}, @code{at90s4433},
10974 @code{at90s4434}, @code{at90s8515}, @code{at90s8535}.
10977 ``Classic'' devices with up to 8@tie{}KiB of program memory and with
10978 the @code{MOVW} instruction.
10979 @*@var{mcu}@tie{}= @code{ata6289}, @code{attiny13}, @code{attiny13a},
10980 @code{attiny2313}, @code{attiny2313a}, @code{attiny24},
10981 @code{attiny24a}, @code{attiny25}, @code{attiny261},
10982 @code{attiny261a}, @code{attiny43u}, @code{attiny4313},
10983 @code{attiny44}, @code{attiny44a}, @code{attiny45}, @code{attiny461},
10984 @code{attiny461a}, @code{attiny48}, @code{attiny84}, @code{attiny84a},
10985 @code{attiny85}, @code{attiny861}, @code{attiny861a}, @code{attiny87},
10986 @code{attiny88}, @code{at86rf401}.
10989 ``Classic'' devices with 16@tie{}KiB up to 64@tie{}KiB of program memory.
10990 @*@var{mcu}@tie{}= @code{at43usb355}, @code{at76c711}.
10993 ``Classic'' devices with 128@tie{}KiB of program memory.
10994 @*@var{mcu}@tie{}= @code{atmega103}, @code{at43usb320}.
10997 ``Classic'' devices with 16@tie{}KiB up to 64@tie{}KiB of program
10998 memory and with the @code{MOVW} instruction.
10999 @*@var{mcu}@tie{}= @code{atmega16u2}, @code{atmega32u2},
11000 @code{atmega8u2}, @code{attiny167}, @code{at90usb162},
11004 ``Enhanced'' devices with up to 8@tie{}KiB of program memory.
11005 @*@var{mcu}@tie{}= @code{atmega48}, @code{atmega48a},
11006 @code{atmega48p}, @code{atmega8}, @code{atmega8hva},
11007 @code{atmega8515}, @code{atmega8535}, @code{atmega88},
11008 @code{atmega88a}, @code{atmega88p}, @code{atmega88pa},
11009 @code{at90pwm1}, @code{at90pwm2}, @code{at90pwm2b}, @code{at90pwm3},
11010 @code{at90pwm3b}, @code{at90pwm81}.
11013 ``Enhanced'' devices with 16@tie{}KiB up to 64@tie{}KiB of program memory.
11014 @*@var{mcu}@tie{}= @code{atmega16}, @code{atmega16a},
11015 @code{atmega16hva}, @code{atmega16hva2}, @code{atmega16hvb},
11016 @code{atmega16m1}, @code{atmega16u4}, @code{atmega161},
11017 @code{atmega162}, @code{atmega163}, @code{atmega164a},
11018 @code{atmega164p}, @code{atmega165}, @code{atmega165a},
11019 @code{atmega165p}, @code{atmega168}, @code{atmega168a},
11020 @code{atmega168p}, @code{atmega169}, @code{atmega169a},
11021 @code{atmega169p}, @code{atmega169pa}, @code{atmega32},
11022 @code{atmega32c1}, @code{atmega32hvb}, @code{atmega32m1},
11023 @code{atmega32u4}, @code{atmega32u6}, @code{atmega323},
11024 @code{atmega324a}, @code{atmega324p}, @code{atmega324pa},
11025 @code{atmega325}, @code{atmega325a}, @code{atmega325p},
11026 @code{atmega3250}, @code{atmega3250a}, @code{atmega3250p},
11027 @code{atmega328}, @code{atmega328p}, @code{atmega329},
11028 @code{atmega329a}, @code{atmega329p}, @code{atmega329pa},
11029 @code{atmega3290}, @code{atmega3290a}, @code{atmega3290p},
11030 @code{atmega406}, @code{atmega64}, @code{atmega64c1},
11031 @code{atmega64hve}, @code{atmega64m1}, @code{atmega640},
11032 @code{atmega644}, @code{atmega644a}, @code{atmega644p},
11033 @code{atmega644pa}, @code{atmega645}, @code{atmega645a},
11034 @code{atmega645p}, @code{atmega6450}, @code{atmega6450a},
11035 @code{atmega6450p}, @code{atmega649}, @code{atmega649a},
11036 @code{atmega649p}, @code{atmega6490}, @code{at90can32},
11037 @code{at90can64}, @code{at90pwm216}, @code{at90pwm316},
11038 @code{at90scr100}, @code{at90usb646}, @code{at90usb647}, @code{at94k},
11042 ``Enhanced'' devices with 128@tie{}KiB of program memory.
11043 @*@var{mcu}@tie{}= @code{atmega128}, @code{atmega128rfa1},
11044 @code{atmega1280}, @code{atmega1281}, @code{atmega1284p},
11045 @code{at90can128}, @code{at90usb1286}, @code{at90usb1287}.
11048 ``Enhanced'' devices with 3-byte PC, i.e.@: with more than
11049 128@tie{}KiB of program memory.
11050 @*@var{mcu}@tie{}= @code{atmega2560}, @code{atmega2561}.
11053 ``XMEGA'' devices with more than 8@tie{}KiB and up to 64@tie{}KiB of
11055 @*@var{mcu}@tie{}= @code{atxmega16a4}, @code{atxmega16d4},
11056 @code{atxmega16x1}, @code{atxmega32a4}, @code{atxmega32d4},
11057 @code{atxmega32x1}.
11060 ``XMEGA'' devices with more than 64@tie{}KiB and up to 128@tie{}KiB of
11062 @*@var{mcu}@tie{}= @code{atxmega64a3}, @code{atxmega64d3}.
11065 ``XMEGA'' devices with more than 64@tie{}KiB and up to 128@tie{}KiB of
11066 program memory and more than 64@tie{}KiB of RAM.
11067 @*@var{mcu}@tie{}= @code{atxmega64a1}, @code{atxmega64a1u}.
11070 ``XMEGA'' devices with more than 128@tie{}KiB of program memory.
11071 @*@var{mcu}@tie{}= @code{atxmega128a3}, @code{atxmega128d3},
11072 @code{atxmega192a3}, @code{atxmega192d3}, @code{atxmega256a3},
11073 @code{atxmega256a3b}, @code{atxmega256a3bu}, @code{atxmega256d3}.
11076 ``XMEGA'' devices with more than 128@tie{}KiB of program memory and
11077 more than 64@tie{}KiB of RAM.
11078 @*@var{mcu}@tie{}= @code{atxmega128a1}, @code{atxmega128a1u}.
11081 This ISA is implemented by the minimal AVR core and supported for
11083 @*@var{mcu}@tie{}= @code{attiny11}, @code{attiny12}, @code{attiny15},
11084 @code{attiny28}, @code{at90s1200}.
11088 @item -maccumulate-args
11089 @opindex maccumulate-args
11090 Accumulate outgoing function arguments and acquire/release the needed
11091 stack space for outgoing function arguments once in function
11092 prologue/epilogue. Without this option, outgoing arguments are pushed
11093 before calling a function and popped afterwards.
11095 Popping the arguments after the function call can be expensive on
11096 AVR so that accumulating the stack space might lead to smaller
11097 executables because arguments need not to be removed from the
11098 stack after such a function call.
11100 This option can lead to reduced code size for functions that perform
11101 several calls to functions that get their arguments on the stack like
11102 calls to printf-like functions.
11104 @item -mbranch-cost=@var{cost}
11105 @opindex mbranch-cost
11106 Set the branch costs for conditional branch instructions to
11107 @var{cost}. Reasonable values for @var{cost} are small, non-negative
11108 integers. The default branch cost is 0.
11110 @item -mcall-prologues
11111 @opindex mcall-prologues
11112 Functions prologues/epilogues are expanded as calls to appropriate
11113 subroutines. Code size is smaller.
11117 Assume @code{int} to be 8-bit integer. This affects the sizes of all types: a
11118 @code{char} is 1 byte, an @code{int} is 1 byte, a @code{long} is 2 bytes,
11119 and @code{long long} is 4 bytes. Please note that this option does not
11120 conform to the C standards, but it results in smaller code
11123 @item -mno-interrupts
11124 @opindex mno-interrupts
11125 Generated code is not compatible with hardware interrupts.
11126 Code size is smaller.
11130 Try to replace @code{CALL} resp.@: @code{JMP} instruction by the shorter
11131 @code{RCALL} resp.@: @code{RJMP} instruction if applicable.
11132 Setting @code{-mrelax} just adds the @code{--relax} option to the
11133 linker command line when the linker is called.
11135 Jump relaxing is performed by the linker because jump offsets are not
11136 known before code is located. Therefore, the assembler code generated by the
11137 compiler is the same, but the instructions in the executable may
11138 differ from instructions in the assembler code.
11140 Relaxing must be turned on if linker stubs are needed, see the
11141 section on @code{EIND} and linker stubs below.
11143 @item -mshort-calls
11144 @opindex mshort-calls
11145 This option has been deprecated and will be removed in GCC 4.8.
11146 See @code{-mrelax} for a replacement.
11148 Use @code{RCALL}/@code{RJMP} instructions even on devices with
11149 16@tie{}KiB or more of program memory, i.e.@: on devices that
11150 have the @code{CALL} and @code{JMP} instructions.
11154 Treat the stack pointer register as an 8-bit register,
11155 i.e.@: assume the high byte of the stack pointer is zero.
11156 In general, you don't need to set this option by hand.
11158 This option is used internally by the compiler to select and
11159 build multilibs for architectures @code{avr2} and @code{avr25}.
11160 These architectures mix devices with and without @code{SPH}.
11161 For any setting other than @code{-mmcu=avr2} or @code{-mmcu=avr25}
11162 the compiler driver will add or remove this option from the compiler
11163 proper's command line, because the compiler then knows if the device
11164 or architecture has an 8-bit stack pointer and thus no @code{SPH}
11169 Use address register @code{X} in a way proposed by the hardware. This means
11170 that @code{X} is only used in indirect, post-increment or
11171 pre-decrement addressing.
11173 Without this option, the @code{X} register may be used in the same way
11174 as @code{Y} or @code{Z} which then is emulated by additional
11176 For example, loading a value with @code{X+const} addressing with a
11177 small non-negative @code{const < 64} to a register @var{Rn} is
11181 adiw r26, const ; X += const
11182 ld @var{Rn}, X ; @var{Rn} = *X
11183 sbiw r26, const ; X -= const
11187 @opindex mtiny-stack
11188 Only change the lower 8@tie{}bits of the stack pointer.
11191 @subsubsection @code{EIND} and Devices with more than 128 Ki Bytes of Flash
11192 @cindex @code{EIND}
11193 Pointers in the implementation are 16@tie{}bits wide.
11194 The address of a function or label is represented as word address so
11195 that indirect jumps and calls can target any code address in the
11196 range of 64@tie{}Ki words.
11198 In order to facilitate indirect jump on devices with more than 128@tie{}Ki
11199 bytes of program memory space, there is a special function register called
11200 @code{EIND} that serves as most significant part of the target address
11201 when @code{EICALL} or @code{EIJMP} instructions are used.
11203 Indirect jumps and calls on these devices are handled as follows by
11204 the compiler and are subject to some limitations:
11209 The compiler never sets @code{EIND}.
11212 The compiler uses @code{EIND} implicitely in @code{EICALL}/@code{EIJMP}
11213 instructions or might read @code{EIND} directly in order to emulate an
11214 indirect call/jump by means of a @code{RET} instruction.
11217 The compiler assumes that @code{EIND} never changes during the startup
11218 code or during the application. In particular, @code{EIND} is not
11219 saved/restored in function or interrupt service routine
11223 For indirect calls to functions and computed goto, the linker
11224 generates @emph{stubs}. Stubs are jump pads sometimes also called
11225 @emph{trampolines}. Thus, the indirect call/jump jumps to such a stub.
11226 The stub contains a direct jump to the desired address.
11229 Linker relaxation must be turned on so that the linker will generate
11230 the stubs correctly an all situaltion. See the compiler option
11231 @code{-mrelax} and the linler option @code{--relax}.
11232 There are corner cases where the linker is supposed to generate stubs
11233 but aborts without relaxation and without a helpful error message.
11236 The default linker script is arranged for code with @code{EIND = 0}.
11237 If code is supposed to work for a setup with @code{EIND != 0}, a custom
11238 linker script has to be used in order to place the sections whose
11239 name start with @code{.trampolines} into the segment where @code{EIND}
11243 The startup code from libgcc never sets @code{EIND}.
11244 Notice that startup code is a blend of code from libgcc and AVR-LibC.
11245 For the impact of AVR-LibC on @code{EIND}, see the
11246 @w{@uref{http://nongnu.org/avr-libc/user-manual,AVR-LibC user manual}}.
11249 It is legitimate for user-specific startup code to set up @code{EIND}
11250 early, for example by means of initialization code located in
11251 section @code{.init3}. Such code runs prior to general startup code
11252 that initializes RAM and calls constructors, but after the bit
11253 of startup code from AVR-LibC that sets @code{EIND} to the segment
11254 where the vector table is located.
11256 #include <avr/io.h>
11259 __attribute__((section(".init3"),naked,used,no_instrument_function))
11260 init3_set_eind (void)
11262 __asm volatile ("ldi r24,pm_hh8(__trampolines_start)\n\t"
11263 "out %i0,r24" :: "n" (&EIND) : "r24","memory");
11268 The @code{__trampolines_start} symbol is defined in the linker script.
11271 Stubs are generated automatically by the linker if
11272 the following two conditions are met:
11275 @item The address of a label is taken by means of the @code{gs} modifier
11276 (short for @emph{generate stubs}) like so:
11278 LDI r24, lo8(gs(@var{func}))
11279 LDI r25, hi8(gs(@var{func}))
11281 @item The final location of that label is in a code segment
11282 @emph{outside} the segment where the stubs are located.
11286 The compiler emits such @code{gs} modifiers for code labels in the
11287 following situations:
11289 @item Taking address of a function or code label.
11290 @item Computed goto.
11291 @item If prologue-save function is used, see @option{-mcall-prologues}
11292 command-line option.
11293 @item Switch/case dispatch tables. If you do not want such dispatch
11294 tables you can specify the @option{-fno-jump-tables} command-line option.
11295 @item C and C++ constructors/destructors called during startup/shutdown.
11296 @item If the tools hit a @code{gs()} modifier explained above.
11300 Jumping to non-symbolic addresses like so is @emph{not} supported:
11305 /* Call function at word address 0x2 */
11306 return ((int(*)(void)) 0x2)();
11310 Instead, a stub has to be set up, i.e.@: the function has to be called
11311 through a symbol (@code{func_4} in the example):
11316 extern int func_4 (void);
11318 /* Call function at byte address 0x4 */
11323 and the application be linked with @code{-Wl,--defsym,func_4=0x4}.
11324 Alternatively, @code{func_4} can be defined in the linker script.
11327 @subsubsection Handling of the @code{RAMPD}, @code{RAMPX}, @code{RAMPY} and @code{RAMPZ} Special Function Registers
11328 @cindex @code{RAMPD}
11329 @cindex @code{RAMPX}
11330 @cindex @code{RAMPY}
11331 @cindex @code{RAMPZ}
11332 Some AVR devices support memories larger than the 64@tie{}KiB range
11333 that can be accessed with 16-bit pointers. To access memory locations
11334 outside this 64@tie{}KiB range, the contentent of a @code{RAMP}
11335 register is used as high part of the address:
11336 The @code{X}, @code{Y}, @code{Z} address register is concatenated
11337 with the @code{RAMPX}, @code{RAMPY}, @code{RAMPZ} special function
11338 register, respectively, to get a wide address. Similarly,
11339 @code{RAMPD} is used together with direct addressing.
11343 The startup code initializes the @code{RAMP} special function
11344 registers with zero.
11347 If a @ref{AVR Named Address Spaces,named address space} other than
11348 generic or @code{__flash} is used, then @code{RAMPZ} is set
11349 as needed before the operation.
11352 If the device supports RAM larger than 64@tie{KiB} and the compiler
11353 needs to change @code{RAMPZ} to accomplish an operation, @code{RAMPZ}
11354 is reset to zero after the operation.
11357 If the device comes with a specific @code{RAMP} register, the ISR
11358 prologue/epilogue saves/restores that SFR and initializes it with
11359 zero in case the ISR code might (implicitly) use it.
11362 RAM larger than 64@tie{KiB} is not supported by GCC for AVR targets.
11363 If you use inline assembler to read from locations outside the
11364 16-bit address range and change one of the @code{RAMP} registers,
11365 you must reset it to zero after the access.
11369 @subsubsection AVR Built-in Macros
11371 GCC defines several built-in macros so that the user code can test
11372 for the presence or absence of features. Almost any of the following
11373 built-in macros are deduced from device capabilities and thus
11374 triggered by the @code{-mmcu=} command-line option.
11376 For even more AVR-specific built-in macros see
11377 @ref{AVR Named Address Spaces} and @ref{AVR Built-in Functions}.
11382 Build-in macro that resolves to a decimal number that identifies the
11383 architecture and depends on the @code{-mmcu=@var{mcu}} option.
11384 Possible values are:
11386 @code{2}, @code{25}, @code{3}, @code{31}, @code{35},
11387 @code{4}, @code{5}, @code{51}, @code{6}, @code{102}, @code{104},
11388 @code{105}, @code{106}, @code{107}
11390 for @var{mcu}=@code{avr2}, @code{avr25}, @code{avr3},
11391 @code{avr31}, @code{avr35}, @code{avr4}, @code{avr5}, @code{avr51},
11392 @code{avr6}, @code{avrxmega2}, @code{avrxmega4}, @code{avrxmega5},
11393 @code{avrxmega6}, @code{avrxmega7}, respectively.
11394 If @var{mcu} specifies a device, this built-in macro is set
11395 accordingly. For example, with @code{-mmcu=atmega8} the macro will be
11396 defined to @code{4}.
11398 @item __AVR_@var{Device}__
11399 Setting @code{-mmcu=@var{device}} defines this built-in macro which reflects
11400 the device's name. For example, @code{-mmcu=atmega8} defines the
11401 built-in macro @code{__AVR_ATmega8__}, @code{-mmcu=attiny261a} defines
11402 @code{__AVR_ATtiny261A__}, etc.
11404 The built-in macros' names follow
11405 the scheme @code{__AVR_@var{Device}__} where @var{Device} is
11406 the device name as from the AVR user manual. The difference between
11407 @var{Device} in the built-in macro and @var{device} in
11408 @code{-mmcu=@var{device}} is that the latter is always lowercase.
11410 If @var{device} is not a device but only a core architecture like
11411 @code{avr51}, this macro will not be defined.
11413 @item __AVR_XMEGA__
11414 The device/architecture belongs to the XMEGA family of devices.
11416 @item __AVR_HAVE_ELPM__
11417 The device has the the @code{ELPM} instruction.
11419 @item __AVR_HAVE_ELPMX__
11420 The device has the @code{ELPM R@var{n},Z} and @code{ELPM
11421 R@var{n},Z+} instructions.
11423 @item __AVR_HAVE_MOVW__
11424 The device has the @code{MOVW} instruction to perform 16-bit
11425 register-register moves.
11427 @item __AVR_HAVE_LPMX__
11428 The device has the @code{LPM R@var{n},Z} and
11429 @code{LPM R@var{n},Z+} instructions.
11431 @item __AVR_HAVE_MUL__
11432 The device has a hardware multiplier.
11434 @item __AVR_HAVE_JMP_CALL__
11435 The device has the @code{JMP} and @code{CALL} instructions.
11436 This is the case for devices with at least 16@tie{}KiB of program
11437 memory and if @code{-mshort-calls} is not set.
11439 @item __AVR_HAVE_EIJMP_EICALL__
11440 @item __AVR_3_BYTE_PC__
11441 The device has the @code{EIJMP} and @code{EICALL} instructions.
11442 This is the case for devices with more than 128@tie{}KiB of program memory.
11443 This also means that the program counter
11444 (PC) is 3@tie{}bytes wide.
11446 @item __AVR_2_BYTE_PC__
11447 The program counter (PC) is 2@tie{}bytes wide. This is the case for devices
11448 with up to 128@tie{}KiB of program memory.
11450 @item __AVR_HAVE_8BIT_SP__
11451 @item __AVR_HAVE_16BIT_SP__
11452 The stack pointer (SP) register is treated as 8-bit respectively
11453 16-bit register by the compiler.
11454 The definition of these macros is affected by @code{-mtiny-stack}.
11456 @item __AVR_HAVE_SPH__
11458 The device has the SPH (high part of stack pointer) special function
11459 register or has an 8-bit stack pointer, respectively.
11460 The definition of these macros is affected by @code{-mmcu=} and
11461 in the cases of @code{-mmcu=avr2} and @code{-mmcu=avr25} also
11464 @item __AVR_HAVE_RAMPD__
11465 @item __AVR_HAVE_RAMPX__
11466 @item __AVR_HAVE_RAMPY__
11467 @item __AVR_HAVE_RAMPZ__
11468 The device has the @code{RAMPD}, @code{RAMPX}, @code{RAMPY},
11469 @code{RAMPZ} special function register, respectively.
11471 @item __NO_INTERRUPTS__
11472 This macro reflects the @code{-mno-interrupts} command line option.
11474 @item __AVR_ERRATA_SKIP__
11475 @item __AVR_ERRATA_SKIP_JMP_CALL__
11476 Some AVR devices (AT90S8515, ATmega103) must not skip 32-bit
11477 instructions because of a hardware erratum. Skip instructions are
11478 @code{SBRS}, @code{SBRC}, @code{SBIS}, @code{SBIC} and @code{CPSE}.
11479 The second macro is only defined if @code{__AVR_HAVE_JMP_CALL__} is also
11482 @item __AVR_SFR_OFFSET__=@var{offset}
11483 Instructions that can address I/O special function registers directly
11484 like @code{IN}, @code{OUT}, @code{SBI}, etc.@: may use a different
11485 address as if addressed by an instruction to access RAM like @code{LD}
11486 or @code{STS}. This offset depends on the device architecture and has
11487 to be subtracted from the RAM address in order to get the
11488 respective I/O@tie{}address.
11490 @item __WITH_AVRLIBC__
11491 The compiler is configured to be used together with AVR-Libc.
11492 See the @code{--with-avrlibc} configure option.
11496 @node Blackfin Options
11497 @subsection Blackfin Options
11498 @cindex Blackfin Options
11501 @item -mcpu=@var{cpu}@r{[}-@var{sirevision}@r{]}
11503 Specifies the name of the target Blackfin processor. Currently, @var{cpu}
11504 can be one of @samp{bf512}, @samp{bf514}, @samp{bf516}, @samp{bf518},
11505 @samp{bf522}, @samp{bf523}, @samp{bf524}, @samp{bf525}, @samp{bf526},
11506 @samp{bf527}, @samp{bf531}, @samp{bf532}, @samp{bf533},
11507 @samp{bf534}, @samp{bf536}, @samp{bf537}, @samp{bf538}, @samp{bf539},
11508 @samp{bf542}, @samp{bf544}, @samp{bf547}, @samp{bf548}, @samp{bf549},
11509 @samp{bf542m}, @samp{bf544m}, @samp{bf547m}, @samp{bf548m}, @samp{bf549m},
11510 @samp{bf561}, @samp{bf592}.
11511 The optional @var{sirevision} specifies the silicon revision of the target
11512 Blackfin processor. Any workarounds available for the targeted silicon revision
11513 will be enabled. If @var{sirevision} is @samp{none}, no workarounds are enabled.
11514 If @var{sirevision} is @samp{any}, all workarounds for the targeted processor
11515 will be enabled. The @code{__SILICON_REVISION__} macro is defined to two
11516 hexadecimal digits representing the major and minor numbers in the silicon
11517 revision. If @var{sirevision} is @samp{none}, the @code{__SILICON_REVISION__}
11518 is not defined. If @var{sirevision} is @samp{any}, the
11519 @code{__SILICON_REVISION__} is defined to be @code{0xffff}.
11520 If this optional @var{sirevision} is not used, GCC assumes the latest known
11521 silicon revision of the targeted Blackfin processor.
11523 Support for @samp{bf561} is incomplete. For @samp{bf561},
11524 Only the processor macro is defined.
11525 Without this option, @samp{bf532} is used as the processor by default.
11526 The corresponding predefined processor macros for @var{cpu} is to
11527 be defined. And for @samp{bfin-elf} toolchain, this causes the hardware BSP
11528 provided by libgloss to be linked in if @option{-msim} is not given.
11532 Specifies that the program will be run on the simulator. This causes
11533 the simulator BSP provided by libgloss to be linked in. This option
11534 has effect only for @samp{bfin-elf} toolchain.
11535 Certain other options, such as @option{-mid-shared-library} and
11536 @option{-mfdpic}, imply @option{-msim}.
11538 @item -momit-leaf-frame-pointer
11539 @opindex momit-leaf-frame-pointer
11540 Don't keep the frame pointer in a register for leaf functions. This
11541 avoids the instructions to save, set up and restore frame pointers and
11542 makes an extra register available in leaf functions. The option
11543 @option{-fomit-frame-pointer} removes the frame pointer for all functions,
11544 which might make debugging harder.
11546 @item -mspecld-anomaly
11547 @opindex mspecld-anomaly
11548 When enabled, the compiler will ensure that the generated code does not
11549 contain speculative loads after jump instructions. If this option is used,
11550 @code{__WORKAROUND_SPECULATIVE_LOADS} is defined.
11552 @item -mno-specld-anomaly
11553 @opindex mno-specld-anomaly
11554 Don't generate extra code to prevent speculative loads from occurring.
11556 @item -mcsync-anomaly
11557 @opindex mcsync-anomaly
11558 When enabled, the compiler will ensure that the generated code does not
11559 contain CSYNC or SSYNC instructions too soon after conditional branches.
11560 If this option is used, @code{__WORKAROUND_SPECULATIVE_SYNCS} is defined.
11562 @item -mno-csync-anomaly
11563 @opindex mno-csync-anomaly
11564 Don't generate extra code to prevent CSYNC or SSYNC instructions from
11565 occurring too soon after a conditional branch.
11569 When enabled, the compiler is free to take advantage of the knowledge that
11570 the entire program fits into the low 64k of memory.
11573 @opindex mno-low-64k
11574 Assume that the program is arbitrarily large. This is the default.
11576 @item -mstack-check-l1
11577 @opindex mstack-check-l1
11578 Do stack checking using information placed into L1 scratchpad memory by the
11581 @item -mid-shared-library
11582 @opindex mid-shared-library
11583 Generate code that supports shared libraries via the library ID method.
11584 This allows for execute in place and shared libraries in an environment
11585 without virtual memory management. This option implies @option{-fPIC}.
11586 With a @samp{bfin-elf} target, this option implies @option{-msim}.
11588 @item -mno-id-shared-library
11589 @opindex mno-id-shared-library
11590 Generate code that doesn't assume ID based shared libraries are being used.
11591 This is the default.
11593 @item -mleaf-id-shared-library
11594 @opindex mleaf-id-shared-library
11595 Generate code that supports shared libraries via the library ID method,
11596 but assumes that this library or executable won't link against any other
11597 ID shared libraries. That allows the compiler to use faster code for jumps
11600 @item -mno-leaf-id-shared-library
11601 @opindex mno-leaf-id-shared-library
11602 Do not assume that the code being compiled won't link against any ID shared
11603 libraries. Slower code will be generated for jump and call insns.
11605 @item -mshared-library-id=n
11606 @opindex mshared-library-id
11607 Specified the identification number of the ID based shared library being
11608 compiled. Specifying a value of 0 will generate more compact code, specifying
11609 other values will force the allocation of that number to the current
11610 library but is no more space or time efficient than omitting this option.
11614 Generate code that allows the data segment to be located in a different
11615 area of memory from the text segment. This allows for execute in place in
11616 an environment without virtual memory management by eliminating relocations
11617 against the text section.
11619 @item -mno-sep-data
11620 @opindex mno-sep-data
11621 Generate code that assumes that the data segment follows the text segment.
11622 This is the default.
11625 @itemx -mno-long-calls
11626 @opindex mlong-calls
11627 @opindex mno-long-calls
11628 Tells the compiler to perform function calls by first loading the
11629 address of the function into a register and then performing a subroutine
11630 call on this register. This switch is needed if the target function
11631 lies outside of the 24-bit addressing range of the offset-based
11632 version of subroutine call instruction.
11634 This feature is not enabled by default. Specifying
11635 @option{-mno-long-calls} will restore the default behavior. Note these
11636 switches have no effect on how the compiler generates code to handle
11637 function calls via function pointers.
11641 Link with the fast floating-point library. This library relaxes some of
11642 the IEEE floating-point standard's rules for checking inputs against
11643 Not-a-Number (NAN), in the interest of performance.
11646 @opindex minline-plt
11647 Enable inlining of PLT entries in function calls to functions that are
11648 not known to bind locally. It has no effect without @option{-mfdpic}.
11651 @opindex mmulticore
11652 Build standalone application for multicore Blackfin processor. Proper
11653 start files and link scripts will be used to support multicore.
11654 This option defines @code{__BFIN_MULTICORE}. It can only be used with
11655 @option{-mcpu=bf561@r{[}-@var{sirevision}@r{]}}. It can be used with
11656 @option{-mcorea} or @option{-mcoreb}. If it's used without
11657 @option{-mcorea} or @option{-mcoreb}, single application/dual core
11658 programming model is used. In this model, the main function of Core B
11659 should be named as coreb_main. If it's used with @option{-mcorea} or
11660 @option{-mcoreb}, one application per core programming model is used.
11661 If this option is not used, single core application programming
11666 Build standalone application for Core A of BF561 when using
11667 one application per core programming model. Proper start files
11668 and link scripts will be used to support Core A. This option
11669 defines @code{__BFIN_COREA}. It must be used with @option{-mmulticore}.
11673 Build standalone application for Core B of BF561 when using
11674 one application per core programming model. Proper start files
11675 and link scripts will be used to support Core B. This option
11676 defines @code{__BFIN_COREB}. When this option is used, coreb_main
11677 should be used instead of main. It must be used with
11678 @option{-mmulticore}.
11682 Build standalone application for SDRAM. Proper start files and
11683 link scripts will be used to put the application into SDRAM.
11684 Loader should initialize SDRAM before loading the application
11685 into SDRAM. This option defines @code{__BFIN_SDRAM}.
11689 Assume that ICPLBs are enabled at run time. This has an effect on certain
11690 anomaly workarounds. For Linux targets, the default is to assume ICPLBs
11691 are enabled; for standalone applications the default is off.
11695 @subsection C6X Options
11696 @cindex C6X Options
11699 @item -march=@var{name}
11701 This specifies the name of the target architecture. GCC uses this
11702 name to determine what kind of instructions it can emit when generating
11703 assembly code. Permissible names are: @samp{c62x},
11704 @samp{c64x}, @samp{c64x+}, @samp{c67x}, @samp{c67x+}, @samp{c674x}.
11707 @opindex mbig-endian
11708 Generate code for a big-endian target.
11710 @item -mlittle-endian
11711 @opindex mlittle-endian
11712 Generate code for a little-endian target. This is the default.
11716 Choose startup files and linker script suitable for the simulator.
11718 @item -msdata=default
11719 @opindex msdata=default
11720 Put small global and static data in the @samp{.neardata} section,
11721 which is pointed to by register @code{B14}. Put small uninitialized
11722 global and static data in the @samp{.bss} section, which is adjacent
11723 to the @samp{.neardata} section. Put small read-only data into the
11724 @samp{.rodata} section. The corresponding sections used for large
11725 pieces of data are @samp{.fardata}, @samp{.far} and @samp{.const}.
11728 @opindex msdata=all
11729 Put all data, not just small objets, into the sections reserved for
11730 small data, and use addressing relative to the @code{B14} register to
11734 @opindex msdata=none
11735 Make no use of the sections reserved for small data, and use absolute
11736 addresses to access all data. Put all initialized global and static
11737 data in the @samp{.fardata} section, and all uninitialized data in the
11738 @samp{.far} section. Put all constant data into the @samp{.const}
11743 @subsection CRIS Options
11744 @cindex CRIS Options
11746 These options are defined specifically for the CRIS ports.
11749 @item -march=@var{architecture-type}
11750 @itemx -mcpu=@var{architecture-type}
11753 Generate code for the specified architecture. The choices for
11754 @var{architecture-type} are @samp{v3}, @samp{v8} and @samp{v10} for
11755 respectively ETRAX@w{ }4, ETRAX@w{ }100, and ETRAX@w{ }100@w{ }LX@.
11756 Default is @samp{v0} except for cris-axis-linux-gnu, where the default is
11759 @item -mtune=@var{architecture-type}
11761 Tune to @var{architecture-type} everything applicable about the generated
11762 code, except for the ABI and the set of available instructions. The
11763 choices for @var{architecture-type} are the same as for
11764 @option{-march=@var{architecture-type}}.
11766 @item -mmax-stack-frame=@var{n}
11767 @opindex mmax-stack-frame
11768 Warn when the stack frame of a function exceeds @var{n} bytes.
11774 The options @option{-metrax4} and @option{-metrax100} are synonyms for
11775 @option{-march=v3} and @option{-march=v8} respectively.
11777 @item -mmul-bug-workaround
11778 @itemx -mno-mul-bug-workaround
11779 @opindex mmul-bug-workaround
11780 @opindex mno-mul-bug-workaround
11781 Work around a bug in the @code{muls} and @code{mulu} instructions for CPU
11782 models where it applies. This option is active by default.
11786 Enable CRIS-specific verbose debug-related information in the assembly
11787 code. This option also has the effect to turn off the @samp{#NO_APP}
11788 formatted-code indicator to the assembler at the beginning of the
11793 Do not use condition-code results from previous instruction; always emit
11794 compare and test instructions before use of condition codes.
11796 @item -mno-side-effects
11797 @opindex mno-side-effects
11798 Do not emit instructions with side-effects in addressing modes other than
11801 @item -mstack-align
11802 @itemx -mno-stack-align
11803 @itemx -mdata-align
11804 @itemx -mno-data-align
11805 @itemx -mconst-align
11806 @itemx -mno-const-align
11807 @opindex mstack-align
11808 @opindex mno-stack-align
11809 @opindex mdata-align
11810 @opindex mno-data-align
11811 @opindex mconst-align
11812 @opindex mno-const-align
11813 These options (no-options) arranges (eliminate arrangements) for the
11814 stack-frame, individual data and constants to be aligned for the maximum
11815 single data access size for the chosen CPU model. The default is to
11816 arrange for 32-bit alignment. ABI details such as structure layout are
11817 not affected by these options.
11825 Similar to the stack- data- and const-align options above, these options
11826 arrange for stack-frame, writable data and constants to all be 32-bit,
11827 16-bit or 8-bit aligned. The default is 32-bit alignment.
11829 @item -mno-prologue-epilogue
11830 @itemx -mprologue-epilogue
11831 @opindex mno-prologue-epilogue
11832 @opindex mprologue-epilogue
11833 With @option{-mno-prologue-epilogue}, the normal function prologue and
11834 epilogue which set up the stack frame are omitted and no return
11835 instructions or return sequences are generated in the code. Use this
11836 option only together with visual inspection of the compiled code: no
11837 warnings or errors are generated when call-saved registers must be saved,
11838 or storage for local variable needs to be allocated.
11842 @opindex mno-gotplt
11844 With @option{-fpic} and @option{-fPIC}, don't generate (do generate)
11845 instruction sequences that load addresses for functions from the PLT part
11846 of the GOT rather than (traditional on other architectures) calls to the
11847 PLT@. The default is @option{-mgotplt}.
11851 Legacy no-op option only recognized with the cris-axis-elf and
11852 cris-axis-linux-gnu targets.
11856 Legacy no-op option only recognized with the cris-axis-linux-gnu target.
11860 This option, recognized for the cris-axis-elf arranges
11861 to link with input-output functions from a simulator library. Code,
11862 initialized data and zero-initialized data are allocated consecutively.
11866 Like @option{-sim}, but pass linker options to locate initialized data at
11867 0x40000000 and zero-initialized data at 0x80000000.
11871 @subsection CR16 Options
11872 @cindex CR16 Options
11874 These options are defined specifically for the CR16 ports.
11880 Enable the use of multiply-accumulate instructions. Disabled by default.
11884 @opindex mcr16cplus
11886 Generate code for CR16C or CR16C+ architecture. CR16C+ architecture
11891 Links the library libsim.a which is in compatible with simulator. Applicable
11892 to elf compiler only.
11896 Choose integer type as 32-bit wide.
11900 Generates sbit/cbit instructions for bit manipulations.
11902 @item -mdata-model=@var{model}
11903 @opindex mdata-model
11904 Choose a data model. The choices for @var{model} are @samp{near},
11905 @samp{far} or @samp{medium}. @samp{medium} is default.
11906 However, @samp{far} is not valid when -mcr16c option is chosen as
11907 CR16C architecture does not support far data model.
11910 @node Darwin Options
11911 @subsection Darwin Options
11912 @cindex Darwin options
11914 These options are defined for all architectures running the Darwin operating
11917 FSF GCC on Darwin does not create ``fat'' object files; it will create
11918 an object file for the single architecture that it was built to
11919 target. Apple's GCC on Darwin does create ``fat'' files if multiple
11920 @option{-arch} options are used; it does so by running the compiler or
11921 linker multiple times and joining the results together with
11924 The subtype of the file created (like @samp{ppc7400} or @samp{ppc970} or
11925 @samp{i686}) is determined by the flags that specify the ISA
11926 that GCC is targetting, like @option{-mcpu} or @option{-march}. The
11927 @option{-force_cpusubtype_ALL} option can be used to override this.
11929 The Darwin tools vary in their behavior when presented with an ISA
11930 mismatch. The assembler, @file{as}, will only permit instructions to
11931 be used that are valid for the subtype of the file it is generating,
11932 so you cannot put 64-bit instructions in a @samp{ppc750} object file.
11933 The linker for shared libraries, @file{/usr/bin/libtool}, will fail
11934 and print an error if asked to create a shared library with a less
11935 restrictive subtype than its input files (for instance, trying to put
11936 a @samp{ppc970} object file in a @samp{ppc7400} library). The linker
11937 for executables, @file{ld}, will quietly give the executable the most
11938 restrictive subtype of any of its input files.
11943 Add the framework directory @var{dir} to the head of the list of
11944 directories to be searched for header files. These directories are
11945 interleaved with those specified by @option{-I} options and are
11946 scanned in a left-to-right order.
11948 A framework directory is a directory with frameworks in it. A
11949 framework is a directory with a @samp{"Headers"} and/or
11950 @samp{"PrivateHeaders"} directory contained directly in it that ends
11951 in @samp{".framework"}. The name of a framework is the name of this
11952 directory excluding the @samp{".framework"}. Headers associated with
11953 the framework are found in one of those two directories, with
11954 @samp{"Headers"} being searched first. A subframework is a framework
11955 directory that is in a framework's @samp{"Frameworks"} directory.
11956 Includes of subframework headers can only appear in a header of a
11957 framework that contains the subframework, or in a sibling subframework
11958 header. Two subframeworks are siblings if they occur in the same
11959 framework. A subframework should not have the same name as a
11960 framework, a warning will be issued if this is violated. Currently a
11961 subframework cannot have subframeworks, in the future, the mechanism
11962 may be extended to support this. The standard frameworks can be found
11963 in @samp{"/System/Library/Frameworks"} and
11964 @samp{"/Library/Frameworks"}. An example include looks like
11965 @code{#include <Framework/header.h>}, where @samp{Framework} denotes
11966 the name of the framework and header.h is found in the
11967 @samp{"PrivateHeaders"} or @samp{"Headers"} directory.
11969 @item -iframework@var{dir}
11970 @opindex iframework
11971 Like @option{-F} except the directory is a treated as a system
11972 directory. The main difference between this @option{-iframework} and
11973 @option{-F} is that with @option{-iframework} the compiler does not
11974 warn about constructs contained within header files found via
11975 @var{dir}. This option is valid only for the C family of languages.
11979 Emit debugging information for symbols that are used. For STABS
11980 debugging format, this enables @option{-feliminate-unused-debug-symbols}.
11981 This is by default ON@.
11985 Emit debugging information for all symbols and types.
11987 @item -mmacosx-version-min=@var{version}
11988 The earliest version of MacOS X that this executable will run on
11989 is @var{version}. Typical values of @var{version} include @code{10.1},
11990 @code{10.2}, and @code{10.3.9}.
11992 If the compiler was built to use the system's headers by default,
11993 then the default for this option is the system version on which the
11994 compiler is running, otherwise the default is to make choices that
11995 are compatible with as many systems and code bases as possible.
11999 Enable kernel development mode. The @option{-mkernel} option sets
12000 @option{-static}, @option{-fno-common}, @option{-fno-cxa-atexit},
12001 @option{-fno-exceptions}, @option{-fno-non-call-exceptions},
12002 @option{-fapple-kext}, @option{-fno-weak} and @option{-fno-rtti} where
12003 applicable. This mode also sets @option{-mno-altivec},
12004 @option{-msoft-float}, @option{-fno-builtin} and
12005 @option{-mlong-branch} for PowerPC targets.
12007 @item -mone-byte-bool
12008 @opindex mone-byte-bool
12009 Override the defaults for @samp{bool} so that @samp{sizeof(bool)==1}.
12010 By default @samp{sizeof(bool)} is @samp{4} when compiling for
12011 Darwin/PowerPC and @samp{1} when compiling for Darwin/x86, so this
12012 option has no effect on x86.
12014 @strong{Warning:} The @option{-mone-byte-bool} switch causes GCC
12015 to generate code that is not binary compatible with code generated
12016 without that switch. Using this switch may require recompiling all
12017 other modules in a program, including system libraries. Use this
12018 switch to conform to a non-default data model.
12020 @item -mfix-and-continue
12021 @itemx -ffix-and-continue
12022 @itemx -findirect-data
12023 @opindex mfix-and-continue
12024 @opindex ffix-and-continue
12025 @opindex findirect-data
12026 Generate code suitable for fast turn around development. Needed to
12027 enable gdb to dynamically load @code{.o} files into already running
12028 programs. @option{-findirect-data} and @option{-ffix-and-continue}
12029 are provided for backwards compatibility.
12033 Loads all members of static archive libraries.
12034 See man ld(1) for more information.
12036 @item -arch_errors_fatal
12037 @opindex arch_errors_fatal
12038 Cause the errors having to do with files that have the wrong architecture
12041 @item -bind_at_load
12042 @opindex bind_at_load
12043 Causes the output file to be marked such that the dynamic linker will
12044 bind all undefined references when the file is loaded or launched.
12048 Produce a Mach-o bundle format file.
12049 See man ld(1) for more information.
12051 @item -bundle_loader @var{executable}
12052 @opindex bundle_loader
12053 This option specifies the @var{executable} that will be loading the build
12054 output file being linked. See man ld(1) for more information.
12057 @opindex dynamiclib
12058 When passed this option, GCC will produce a dynamic library instead of
12059 an executable when linking, using the Darwin @file{libtool} command.
12061 @item -force_cpusubtype_ALL
12062 @opindex force_cpusubtype_ALL
12063 This causes GCC's output file to have the @var{ALL} subtype, instead of
12064 one controlled by the @option{-mcpu} or @option{-march} option.
12066 @item -allowable_client @var{client_name}
12067 @itemx -client_name
12068 @itemx -compatibility_version
12069 @itemx -current_version
12071 @itemx -dependency-file
12073 @itemx -dylinker_install_name
12075 @itemx -exported_symbols_list
12078 @itemx -flat_namespace
12079 @itemx -force_flat_namespace
12080 @itemx -headerpad_max_install_names
12083 @itemx -install_name
12084 @itemx -keep_private_externs
12085 @itemx -multi_module
12086 @itemx -multiply_defined
12087 @itemx -multiply_defined_unused
12090 @itemx -no_dead_strip_inits_and_terms
12091 @itemx -nofixprebinding
12092 @itemx -nomultidefs
12094 @itemx -noseglinkedit
12095 @itemx -pagezero_size
12097 @itemx -prebind_all_twolevel_modules
12098 @itemx -private_bundle
12100 @itemx -read_only_relocs
12102 @itemx -sectobjectsymbols
12106 @itemx -sectobjectsymbols
12109 @itemx -segs_read_only_addr
12111 @itemx -segs_read_write_addr
12112 @itemx -seg_addr_table
12113 @itemx -seg_addr_table_filename
12114 @itemx -seglinkedit
12116 @itemx -segs_read_only_addr
12117 @itemx -segs_read_write_addr
12118 @itemx -single_module
12120 @itemx -sub_library
12122 @itemx -sub_umbrella
12123 @itemx -twolevel_namespace
12126 @itemx -unexported_symbols_list
12127 @itemx -weak_reference_mismatches
12128 @itemx -whatsloaded
12129 @opindex allowable_client
12130 @opindex client_name
12131 @opindex compatibility_version
12132 @opindex current_version
12133 @opindex dead_strip
12134 @opindex dependency-file
12135 @opindex dylib_file
12136 @opindex dylinker_install_name
12138 @opindex exported_symbols_list
12140 @opindex flat_namespace
12141 @opindex force_flat_namespace
12142 @opindex headerpad_max_install_names
12143 @opindex image_base
12145 @opindex install_name
12146 @opindex keep_private_externs
12147 @opindex multi_module
12148 @opindex multiply_defined
12149 @opindex multiply_defined_unused
12150 @opindex noall_load
12151 @opindex no_dead_strip_inits_and_terms
12152 @opindex nofixprebinding
12153 @opindex nomultidefs
12155 @opindex noseglinkedit
12156 @opindex pagezero_size
12158 @opindex prebind_all_twolevel_modules
12159 @opindex private_bundle
12160 @opindex read_only_relocs
12162 @opindex sectobjectsymbols
12165 @opindex sectcreate
12166 @opindex sectobjectsymbols
12169 @opindex segs_read_only_addr
12170 @opindex segs_read_write_addr
12171 @opindex seg_addr_table
12172 @opindex seg_addr_table_filename
12173 @opindex seglinkedit
12175 @opindex segs_read_only_addr
12176 @opindex segs_read_write_addr
12177 @opindex single_module
12179 @opindex sub_library
12180 @opindex sub_umbrella
12181 @opindex twolevel_namespace
12184 @opindex unexported_symbols_list
12185 @opindex weak_reference_mismatches
12186 @opindex whatsloaded
12187 These options are passed to the Darwin linker. The Darwin linker man page
12188 describes them in detail.
12191 @node DEC Alpha Options
12192 @subsection DEC Alpha Options
12194 These @samp{-m} options are defined for the DEC Alpha implementations:
12197 @item -mno-soft-float
12198 @itemx -msoft-float
12199 @opindex mno-soft-float
12200 @opindex msoft-float
12201 Use (do not use) the hardware floating-point instructions for
12202 floating-point operations. When @option{-msoft-float} is specified,
12203 functions in @file{libgcc.a} will be used to perform floating-point
12204 operations. Unless they are replaced by routines that emulate the
12205 floating-point operations, or compiled in such a way as to call such
12206 emulations routines, these routines will issue floating-point
12207 operations. If you are compiling for an Alpha without floating-point
12208 operations, you must ensure that the library is built so as not to call
12211 Note that Alpha implementations without floating-point operations are
12212 required to have floating-point registers.
12215 @itemx -mno-fp-regs
12217 @opindex mno-fp-regs
12218 Generate code that uses (does not use) the floating-point register set.
12219 @option{-mno-fp-regs} implies @option{-msoft-float}. If the floating-point
12220 register set is not used, floating-point operands are passed in integer
12221 registers as if they were integers and floating-point results are passed
12222 in @code{$0} instead of @code{$f0}. This is a non-standard calling sequence,
12223 so any function with a floating-point argument or return value called by code
12224 compiled with @option{-mno-fp-regs} must also be compiled with that
12227 A typical use of this option is building a kernel that does not use,
12228 and hence need not save and restore, any floating-point registers.
12232 The Alpha architecture implements floating-point hardware optimized for
12233 maximum performance. It is mostly compliant with the IEEE floating-point
12234 standard. However, for full compliance, software assistance is
12235 required. This option generates code fully IEEE-compliant code
12236 @emph{except} that the @var{inexact-flag} is not maintained (see below).
12237 If this option is turned on, the preprocessor macro @code{_IEEE_FP} is
12238 defined during compilation. The resulting code is less efficient but is
12239 able to correctly support denormalized numbers and exceptional IEEE
12240 values such as not-a-number and plus/minus infinity. Other Alpha
12241 compilers call this option @option{-ieee_with_no_inexact}.
12243 @item -mieee-with-inexact
12244 @opindex mieee-with-inexact
12245 This is like @option{-mieee} except the generated code also maintains
12246 the IEEE @var{inexact-flag}. Turning on this option causes the
12247 generated code to implement fully-compliant IEEE math. In addition to
12248 @code{_IEEE_FP}, @code{_IEEE_FP_EXACT} is defined as a preprocessor
12249 macro. On some Alpha implementations the resulting code may execute
12250 significantly slower than the code generated by default. Since there is
12251 very little code that depends on the @var{inexact-flag}, you should
12252 normally not specify this option. Other Alpha compilers call this
12253 option @option{-ieee_with_inexact}.
12255 @item -mfp-trap-mode=@var{trap-mode}
12256 @opindex mfp-trap-mode
12257 This option controls what floating-point related traps are enabled.
12258 Other Alpha compilers call this option @option{-fptm @var{trap-mode}}.
12259 The trap mode can be set to one of four values:
12263 This is the default (normal) setting. The only traps that are enabled
12264 are the ones that cannot be disabled in software (e.g., division by zero
12268 In addition to the traps enabled by @samp{n}, underflow traps are enabled
12272 Like @samp{u}, but the instructions are marked to be safe for software
12273 completion (see Alpha architecture manual for details).
12276 Like @samp{su}, but inexact traps are enabled as well.
12279 @item -mfp-rounding-mode=@var{rounding-mode}
12280 @opindex mfp-rounding-mode
12281 Selects the IEEE rounding mode. Other Alpha compilers call this option
12282 @option{-fprm @var{rounding-mode}}. The @var{rounding-mode} can be one
12287 Normal IEEE rounding mode. Floating-point numbers are rounded towards
12288 the nearest machine number or towards the even machine number in case
12292 Round towards minus infinity.
12295 Chopped rounding mode. Floating-point numbers are rounded towards zero.
12298 Dynamic rounding mode. A field in the floating-point control register
12299 (@var{fpcr}, see Alpha architecture reference manual) controls the
12300 rounding mode in effect. The C library initializes this register for
12301 rounding towards plus infinity. Thus, unless your program modifies the
12302 @var{fpcr}, @samp{d} corresponds to round towards plus infinity.
12305 @item -mtrap-precision=@var{trap-precision}
12306 @opindex mtrap-precision
12307 In the Alpha architecture, floating-point traps are imprecise. This
12308 means without software assistance it is impossible to recover from a
12309 floating trap and program execution normally needs to be terminated.
12310 GCC can generate code that can assist operating system trap handlers
12311 in determining the exact location that caused a floating-point trap.
12312 Depending on the requirements of an application, different levels of
12313 precisions can be selected:
12317 Program precision. This option is the default and means a trap handler
12318 can only identify which program caused a floating-point exception.
12321 Function precision. The trap handler can determine the function that
12322 caused a floating-point exception.
12325 Instruction precision. The trap handler can determine the exact
12326 instruction that caused a floating-point exception.
12329 Other Alpha compilers provide the equivalent options called
12330 @option{-scope_safe} and @option{-resumption_safe}.
12332 @item -mieee-conformant
12333 @opindex mieee-conformant
12334 This option marks the generated code as IEEE conformant. You must not
12335 use this option unless you also specify @option{-mtrap-precision=i} and either
12336 @option{-mfp-trap-mode=su} or @option{-mfp-trap-mode=sui}. Its only effect
12337 is to emit the line @samp{.eflag 48} in the function prologue of the
12338 generated assembly file. Under DEC Unix, this has the effect that
12339 IEEE-conformant math library routines will be linked in.
12341 @item -mbuild-constants
12342 @opindex mbuild-constants
12343 Normally GCC examines a 32- or 64-bit integer constant to
12344 see if it can construct it from smaller constants in two or three
12345 instructions. If it cannot, it will output the constant as a literal and
12346 generate code to load it from the data segment at run time.
12348 Use this option to require GCC to construct @emph{all} integer constants
12349 using code, even if it takes more instructions (the maximum is six).
12351 You would typically use this option to build a shared library dynamic
12352 loader. Itself a shared library, it must relocate itself in memory
12353 before it can find the variables and constants in its own data segment.
12359 Select whether to generate code to be assembled by the vendor-supplied
12360 assembler (@option{-malpha-as}) or by the GNU assembler @option{-mgas}.
12378 Indicate whether GCC should generate code to use the optional BWX,
12379 CIX, FIX and MAX instruction sets. The default is to use the instruction
12380 sets supported by the CPU type specified via @option{-mcpu=} option or that
12381 of the CPU on which GCC was built if none was specified.
12384 @itemx -mfloat-ieee
12385 @opindex mfloat-vax
12386 @opindex mfloat-ieee
12387 Generate code that uses (does not use) VAX F and G floating-point
12388 arithmetic instead of IEEE single and double precision.
12390 @item -mexplicit-relocs
12391 @itemx -mno-explicit-relocs
12392 @opindex mexplicit-relocs
12393 @opindex mno-explicit-relocs
12394 Older Alpha assemblers provided no way to generate symbol relocations
12395 except via assembler macros. Use of these macros does not allow
12396 optimal instruction scheduling. GNU binutils as of version 2.12
12397 supports a new syntax that allows the compiler to explicitly mark
12398 which relocations should apply to which instructions. This option
12399 is mostly useful for debugging, as GCC detects the capabilities of
12400 the assembler when it is built and sets the default accordingly.
12403 @itemx -mlarge-data
12404 @opindex msmall-data
12405 @opindex mlarge-data
12406 When @option{-mexplicit-relocs} is in effect, static data is
12407 accessed via @dfn{gp-relative} relocations. When @option{-msmall-data}
12408 is used, objects 8 bytes long or smaller are placed in a @dfn{small data area}
12409 (the @code{.sdata} and @code{.sbss} sections) and are accessed via
12410 16-bit relocations off of the @code{$gp} register. This limits the
12411 size of the small data area to 64KB, but allows the variables to be
12412 directly accessed via a single instruction.
12414 The default is @option{-mlarge-data}. With this option the data area
12415 is limited to just below 2GB@. Programs that require more than 2GB of
12416 data must use @code{malloc} or @code{mmap} to allocate the data in the
12417 heap instead of in the program's data segment.
12419 When generating code for shared libraries, @option{-fpic} implies
12420 @option{-msmall-data} and @option{-fPIC} implies @option{-mlarge-data}.
12423 @itemx -mlarge-text
12424 @opindex msmall-text
12425 @opindex mlarge-text
12426 When @option{-msmall-text} is used, the compiler assumes that the
12427 code of the entire program (or shared library) fits in 4MB, and is
12428 thus reachable with a branch instruction. When @option{-msmall-data}
12429 is used, the compiler can assume that all local symbols share the
12430 same @code{$gp} value, and thus reduce the number of instructions
12431 required for a function call from 4 to 1.
12433 The default is @option{-mlarge-text}.
12435 @item -mcpu=@var{cpu_type}
12437 Set the instruction set and instruction scheduling parameters for
12438 machine type @var{cpu_type}. You can specify either the @samp{EV}
12439 style name or the corresponding chip number. GCC supports scheduling
12440 parameters for the EV4, EV5 and EV6 family of processors and will
12441 choose the default values for the instruction set from the processor
12442 you specify. If you do not specify a processor type, GCC will default
12443 to the processor on which the compiler was built.
12445 Supported values for @var{cpu_type} are
12451 Schedules as an EV4 and has no instruction set extensions.
12455 Schedules as an EV5 and has no instruction set extensions.
12459 Schedules as an EV5 and supports the BWX extension.
12464 Schedules as an EV5 and supports the BWX and MAX extensions.
12468 Schedules as an EV6 and supports the BWX, FIX, and MAX extensions.
12472 Schedules as an EV6 and supports the BWX, CIX, FIX, and MAX extensions.
12475 Native toolchains also support the value @samp{native},
12476 which selects the best architecture option for the host processor.
12477 @option{-mcpu=native} has no effect if GCC does not recognize
12480 @item -mtune=@var{cpu_type}
12482 Set only the instruction scheduling parameters for machine type
12483 @var{cpu_type}. The instruction set is not changed.
12485 Native toolchains also support the value @samp{native},
12486 which selects the best architecture option for the host processor.
12487 @option{-mtune=native} has no effect if GCC does not recognize
12490 @item -mmemory-latency=@var{time}
12491 @opindex mmemory-latency
12492 Sets the latency the scheduler should assume for typical memory
12493 references as seen by the application. This number is highly
12494 dependent on the memory access patterns used by the application
12495 and the size of the external cache on the machine.
12497 Valid options for @var{time} are
12501 A decimal number representing clock cycles.
12507 The compiler contains estimates of the number of clock cycles for
12508 ``typical'' EV4 & EV5 hardware for the Level 1, 2 & 3 caches
12509 (also called Dcache, Scache, and Bcache), as well as to main memory.
12510 Note that L3 is only valid for EV5.
12515 @node DEC Alpha/VMS Options
12516 @subsection DEC Alpha/VMS Options
12518 These @samp{-m} options are defined for the DEC Alpha/VMS implementations:
12521 @item -mvms-return-codes
12522 @opindex mvms-return-codes
12523 Return VMS condition codes from main. The default is to return POSIX
12524 style condition (e.g.@: error) codes.
12526 @item -mdebug-main=@var{prefix}
12527 @opindex mdebug-main=@var{prefix}
12528 Flag the first routine whose name starts with @var{prefix} as the main
12529 routine for the debugger.
12533 Default to 64-bit memory allocation routines.
12537 @subsection FR30 Options
12538 @cindex FR30 Options
12540 These options are defined specifically for the FR30 port.
12544 @item -msmall-model
12545 @opindex msmall-model
12546 Use the small address space model. This can produce smaller code, but
12547 it does assume that all symbolic values and addresses will fit into a
12552 Assume that runtime support has been provided and so there is no need
12553 to include the simulator library (@file{libsim.a}) on the linker
12559 @subsection FRV Options
12560 @cindex FRV Options
12566 Only use the first 32 general-purpose registers.
12571 Use all 64 general-purpose registers.
12576 Use only the first 32 floating-point registers.
12581 Use all 64 floating-point registers.
12584 @opindex mhard-float
12586 Use hardware instructions for floating-point operations.
12589 @opindex msoft-float
12591 Use library routines for floating-point operations.
12596 Dynamically allocate condition code registers.
12601 Do not try to dynamically allocate condition code registers, only
12602 use @code{icc0} and @code{fcc0}.
12607 Change ABI to use double word insns.
12612 Do not use double word instructions.
12617 Use floating-point double instructions.
12620 @opindex mno-double
12622 Do not use floating-point double instructions.
12627 Use media instructions.
12632 Do not use media instructions.
12637 Use multiply and add/subtract instructions.
12640 @opindex mno-muladd
12642 Do not use multiply and add/subtract instructions.
12647 Select the FDPIC ABI, which uses function descriptors to represent
12648 pointers to functions. Without any PIC/PIE-related options, it
12649 implies @option{-fPIE}. With @option{-fpic} or @option{-fpie}, it
12650 assumes GOT entries and small data are within a 12-bit range from the
12651 GOT base address; with @option{-fPIC} or @option{-fPIE}, GOT offsets
12652 are computed with 32 bits.
12653 With a @samp{bfin-elf} target, this option implies @option{-msim}.
12656 @opindex minline-plt
12658 Enable inlining of PLT entries in function calls to functions that are
12659 not known to bind locally. It has no effect without @option{-mfdpic}.
12660 It's enabled by default if optimizing for speed and compiling for
12661 shared libraries (i.e., @option{-fPIC} or @option{-fpic}), or when an
12662 optimization option such as @option{-O3} or above is present in the
12668 Assume a large TLS segment when generating thread-local code.
12673 Do not assume a large TLS segment when generating thread-local code.
12678 Enable the use of @code{GPREL} relocations in the FDPIC ABI for data
12679 that is known to be in read-only sections. It's enabled by default,
12680 except for @option{-fpic} or @option{-fpie}: even though it may help
12681 make the global offset table smaller, it trades 1 instruction for 4.
12682 With @option{-fPIC} or @option{-fPIE}, it trades 3 instructions for 4,
12683 one of which may be shared by multiple symbols, and it avoids the need
12684 for a GOT entry for the referenced symbol, so it's more likely to be a
12685 win. If it is not, @option{-mno-gprel-ro} can be used to disable it.
12687 @item -multilib-library-pic
12688 @opindex multilib-library-pic
12690 Link with the (library, not FD) pic libraries. It's implied by
12691 @option{-mlibrary-pic}, as well as by @option{-fPIC} and
12692 @option{-fpic} without @option{-mfdpic}. You should never have to use
12696 @opindex mlinked-fp
12698 Follow the EABI requirement of always creating a frame pointer whenever
12699 a stack frame is allocated. This option is enabled by default and can
12700 be disabled with @option{-mno-linked-fp}.
12703 @opindex mlong-calls
12705 Use indirect addressing to call functions outside the current
12706 compilation unit. This allows the functions to be placed anywhere
12707 within the 32-bit address space.
12709 @item -malign-labels
12710 @opindex malign-labels
12712 Try to align labels to an 8-byte boundary by inserting nops into the
12713 previous packet. This option only has an effect when VLIW packing
12714 is enabled. It doesn't create new packets; it merely adds nops to
12717 @item -mlibrary-pic
12718 @opindex mlibrary-pic
12720 Generate position-independent EABI code.
12725 Use only the first four media accumulator registers.
12730 Use all eight media accumulator registers.
12735 Pack VLIW instructions.
12740 Do not pack VLIW instructions.
12743 @opindex mno-eflags
12745 Do not mark ABI switches in e_flags.
12748 @opindex mcond-move
12750 Enable the use of conditional-move instructions (default).
12752 This switch is mainly for debugging the compiler and will likely be removed
12753 in a future version.
12755 @item -mno-cond-move
12756 @opindex mno-cond-move
12758 Disable the use of conditional-move instructions.
12760 This switch is mainly for debugging the compiler and will likely be removed
12761 in a future version.
12766 Enable the use of conditional set instructions (default).
12768 This switch is mainly for debugging the compiler and will likely be removed
12769 in a future version.
12774 Disable the use of conditional set instructions.
12776 This switch is mainly for debugging the compiler and will likely be removed
12777 in a future version.
12780 @opindex mcond-exec
12782 Enable the use of conditional execution (default).
12784 This switch is mainly for debugging the compiler and will likely be removed
12785 in a future version.
12787 @item -mno-cond-exec
12788 @opindex mno-cond-exec
12790 Disable the use of conditional execution.
12792 This switch is mainly for debugging the compiler and will likely be removed
12793 in a future version.
12795 @item -mvliw-branch
12796 @opindex mvliw-branch
12798 Run a pass to pack branches into VLIW instructions (default).
12800 This switch is mainly for debugging the compiler and will likely be removed
12801 in a future version.
12803 @item -mno-vliw-branch
12804 @opindex mno-vliw-branch
12806 Do not run a pass to pack branches into VLIW instructions.
12808 This switch is mainly for debugging the compiler and will likely be removed
12809 in a future version.
12811 @item -mmulti-cond-exec
12812 @opindex mmulti-cond-exec
12814 Enable optimization of @code{&&} and @code{||} in conditional execution
12817 This switch is mainly for debugging the compiler and will likely be removed
12818 in a future version.
12820 @item -mno-multi-cond-exec
12821 @opindex mno-multi-cond-exec
12823 Disable optimization of @code{&&} and @code{||} in conditional execution.
12825 This switch is mainly for debugging the compiler and will likely be removed
12826 in a future version.
12828 @item -mnested-cond-exec
12829 @opindex mnested-cond-exec
12831 Enable nested conditional execution optimizations (default).
12833 This switch is mainly for debugging the compiler and will likely be removed
12834 in a future version.
12836 @item -mno-nested-cond-exec
12837 @opindex mno-nested-cond-exec
12839 Disable nested conditional execution optimizations.
12841 This switch is mainly for debugging the compiler and will likely be removed
12842 in a future version.
12844 @item -moptimize-membar
12845 @opindex moptimize-membar
12847 This switch removes redundant @code{membar} instructions from the
12848 compiler generated code. It is enabled by default.
12850 @item -mno-optimize-membar
12851 @opindex mno-optimize-membar
12853 This switch disables the automatic removal of redundant @code{membar}
12854 instructions from the generated code.
12856 @item -mtomcat-stats
12857 @opindex mtomcat-stats
12859 Cause gas to print out tomcat statistics.
12861 @item -mcpu=@var{cpu}
12864 Select the processor type for which to generate code. Possible values are
12865 @samp{frv}, @samp{fr550}, @samp{tomcat}, @samp{fr500}, @samp{fr450},
12866 @samp{fr405}, @samp{fr400}, @samp{fr300} and @samp{simple}.
12870 @node GNU/Linux Options
12871 @subsection GNU/Linux Options
12873 These @samp{-m} options are defined for GNU/Linux targets:
12878 Use the GNU C library. This is the default except
12879 on @samp{*-*-linux-*uclibc*} and @samp{*-*-linux-*android*} targets.
12883 Use uClibc C library. This is the default on
12884 @samp{*-*-linux-*uclibc*} targets.
12888 Use Bionic C library. This is the default on
12889 @samp{*-*-linux-*android*} targets.
12893 Compile code compatible with Android platform. This is the default on
12894 @samp{*-*-linux-*android*} targets.
12896 When compiling, this option enables @option{-mbionic}, @option{-fPIC},
12897 @option{-fno-exceptions} and @option{-fno-rtti} by default. When linking,
12898 this option makes the GCC driver pass Android-specific options to the linker.
12899 Finally, this option causes the preprocessor macro @code{__ANDROID__}
12902 @item -tno-android-cc
12903 @opindex tno-android-cc
12904 Disable compilation effects of @option{-mandroid}, i.e., do not enable
12905 @option{-mbionic}, @option{-fPIC}, @option{-fno-exceptions} and
12906 @option{-fno-rtti} by default.
12908 @item -tno-android-ld
12909 @opindex tno-android-ld
12910 Disable linking effects of @option{-mandroid}, i.e., pass standard Linux
12911 linking options to the linker.
12915 @node H8/300 Options
12916 @subsection H8/300 Options
12918 These @samp{-m} options are defined for the H8/300 implementations:
12923 Shorten some address references at link time, when possible; uses the
12924 linker option @option{-relax}. @xref{H8/300,, @code{ld} and the H8/300,
12925 ld, Using ld}, for a fuller description.
12929 Generate code for the H8/300H@.
12933 Generate code for the H8S@.
12937 Generate code for the H8S and H8/300H in the normal mode. This switch
12938 must be used either with @option{-mh} or @option{-ms}.
12942 Generate code for the H8S/2600. This switch must be used with @option{-ms}.
12946 Make @code{int} data 32 bits by default.
12949 @opindex malign-300
12950 On the H8/300H and H8S, use the same alignment rules as for the H8/300.
12951 The default for the H8/300H and H8S is to align longs and floats on
12953 @option{-malign-300} causes them to be aligned on 2-byte boundaries.
12954 This option has no effect on the H8/300.
12958 @subsection HPPA Options
12959 @cindex HPPA Options
12961 These @samp{-m} options are defined for the HPPA family of computers:
12964 @item -march=@var{architecture-type}
12966 Generate code for the specified architecture. The choices for
12967 @var{architecture-type} are @samp{1.0} for PA 1.0, @samp{1.1} for PA
12968 1.1, and @samp{2.0} for PA 2.0 processors. Refer to
12969 @file{/usr/lib/sched.models} on an HP-UX system to determine the proper
12970 architecture option for your machine. Code compiled for lower numbered
12971 architectures will run on higher numbered architectures, but not the
12974 @item -mpa-risc-1-0
12975 @itemx -mpa-risc-1-1
12976 @itemx -mpa-risc-2-0
12977 @opindex mpa-risc-1-0
12978 @opindex mpa-risc-1-1
12979 @opindex mpa-risc-2-0
12980 Synonyms for @option{-march=1.0}, @option{-march=1.1}, and @option{-march=2.0} respectively.
12983 @opindex mbig-switch
12984 Generate code suitable for big switch tables. Use this option only if
12985 the assembler/linker complain about out of range branches within a switch
12988 @item -mjump-in-delay
12989 @opindex mjump-in-delay
12990 Fill delay slots of function calls with unconditional jump instructions
12991 by modifying the return pointer for the function call to be the target
12992 of the conditional jump.
12994 @item -mdisable-fpregs
12995 @opindex mdisable-fpregs
12996 Prevent floating-point registers from being used in any manner. This is
12997 necessary for compiling kernels that perform lazy context switching of
12998 floating-point registers. If you use this option and attempt to perform
12999 floating-point operations, the compiler aborts.
13001 @item -mdisable-indexing
13002 @opindex mdisable-indexing
13003 Prevent the compiler from using indexing address modes. This avoids some
13004 rather obscure problems when compiling MIG generated code under MACH@.
13006 @item -mno-space-regs
13007 @opindex mno-space-regs
13008 Generate code that assumes the target has no space registers. This allows
13009 GCC to generate faster indirect calls and use unscaled index address modes.
13011 Such code is suitable for level 0 PA systems and kernels.
13013 @item -mfast-indirect-calls
13014 @opindex mfast-indirect-calls
13015 Generate code that assumes calls never cross space boundaries. This
13016 allows GCC to emit code that performs faster indirect calls.
13018 This option will not work in the presence of shared libraries or nested
13021 @item -mfixed-range=@var{register-range}
13022 @opindex mfixed-range
13023 Generate code treating the given register range as fixed registers.
13024 A fixed register is one that the register allocator can not use. This is
13025 useful when compiling kernel code. A register range is specified as
13026 two registers separated by a dash. Multiple register ranges can be
13027 specified separated by a comma.
13029 @item -mlong-load-store
13030 @opindex mlong-load-store
13031 Generate 3-instruction load and store sequences as sometimes required by
13032 the HP-UX 10 linker. This is equivalent to the @samp{+k} option to
13035 @item -mportable-runtime
13036 @opindex mportable-runtime
13037 Use the portable calling conventions proposed by HP for ELF systems.
13041 Enable the use of assembler directives only GAS understands.
13043 @item -mschedule=@var{cpu-type}
13045 Schedule code according to the constraints for the machine type
13046 @var{cpu-type}. The choices for @var{cpu-type} are @samp{700}
13047 @samp{7100}, @samp{7100LC}, @samp{7200}, @samp{7300} and @samp{8000}. Refer
13048 to @file{/usr/lib/sched.models} on an HP-UX system to determine the
13049 proper scheduling option for your machine. The default scheduling is
13053 @opindex mlinker-opt
13054 Enable the optimization pass in the HP-UX linker. Note this makes symbolic
13055 debugging impossible. It also triggers a bug in the HP-UX 8 and HP-UX 9
13056 linkers in which they give bogus error messages when linking some programs.
13059 @opindex msoft-float
13060 Generate output containing library calls for floating point.
13061 @strong{Warning:} the requisite libraries are not available for all HPPA
13062 targets. Normally the facilities of the machine's usual C compiler are
13063 used, but this cannot be done directly in cross-compilation. You must make
13064 your own arrangements to provide suitable library functions for
13067 @option{-msoft-float} changes the calling convention in the output file;
13068 therefore, it is only useful if you compile @emph{all} of a program with
13069 this option. In particular, you need to compile @file{libgcc.a}, the
13070 library that comes with GCC, with @option{-msoft-float} in order for
13075 Generate the predefine, @code{_SIO}, for server IO@. The default is
13076 @option{-mwsio}. This generates the predefines, @code{__hp9000s700},
13077 @code{__hp9000s700__} and @code{_WSIO}, for workstation IO@. These
13078 options are available under HP-UX and HI-UX@.
13082 Use GNU ld specific options. This passes @option{-shared} to ld when
13083 building a shared library. It is the default when GCC is configured,
13084 explicitly or implicitly, with the GNU linker. This option does not
13085 have any affect on which ld is called, it only changes what parameters
13086 are passed to that ld. The ld that is called is determined by the
13087 @option{--with-ld} configure option, GCC's program search path, and
13088 finally by the user's @env{PATH}. The linker used by GCC can be printed
13089 using @samp{which `gcc -print-prog-name=ld`}. This option is only available
13090 on the 64-bit HP-UX GCC, i.e.@: configured with @samp{hppa*64*-*-hpux*}.
13094 Use HP ld specific options. This passes @option{-b} to ld when building
13095 a shared library and passes @option{+Accept TypeMismatch} to ld on all
13096 links. It is the default when GCC is configured, explicitly or
13097 implicitly, with the HP linker. This option does not have any affect on
13098 which ld is called, it only changes what parameters are passed to that
13099 ld. The ld that is called is determined by the @option{--with-ld}
13100 configure option, GCC's program search path, and finally by the user's
13101 @env{PATH}. The linker used by GCC can be printed using @samp{which
13102 `gcc -print-prog-name=ld`}. This option is only available on the 64-bit
13103 HP-UX GCC, i.e.@: configured with @samp{hppa*64*-*-hpux*}.
13106 @opindex mno-long-calls
13107 Generate code that uses long call sequences. This ensures that a call
13108 is always able to reach linker generated stubs. The default is to generate
13109 long calls only when the distance from the call site to the beginning
13110 of the function or translation unit, as the case may be, exceeds a
13111 predefined limit set by the branch type being used. The limits for
13112 normal calls are 7,600,000 and 240,000 bytes, respectively for the
13113 PA 2.0 and PA 1.X architectures. Sibcalls are always limited at
13116 Distances are measured from the beginning of functions when using the
13117 @option{-ffunction-sections} option, or when using the @option{-mgas}
13118 and @option{-mno-portable-runtime} options together under HP-UX with
13121 It is normally not desirable to use this option as it will degrade
13122 performance. However, it may be useful in large applications,
13123 particularly when partial linking is used to build the application.
13125 The types of long calls used depends on the capabilities of the
13126 assembler and linker, and the type of code being generated. The
13127 impact on systems that support long absolute calls, and long pic
13128 symbol-difference or pc-relative calls should be relatively small.
13129 However, an indirect call is used on 32-bit ELF systems in pic code
13130 and it is quite long.
13132 @item -munix=@var{unix-std}
13134 Generate compiler predefines and select a startfile for the specified
13135 UNIX standard. The choices for @var{unix-std} are @samp{93}, @samp{95}
13136 and @samp{98}. @samp{93} is supported on all HP-UX versions. @samp{95}
13137 is available on HP-UX 10.10 and later. @samp{98} is available on HP-UX
13138 11.11 and later. The default values are @samp{93} for HP-UX 10.00,
13139 @samp{95} for HP-UX 10.10 though to 11.00, and @samp{98} for HP-UX 11.11
13142 @option{-munix=93} provides the same predefines as GCC 3.3 and 3.4.
13143 @option{-munix=95} provides additional predefines for @code{XOPEN_UNIX}
13144 and @code{_XOPEN_SOURCE_EXTENDED}, and the startfile @file{unix95.o}.
13145 @option{-munix=98} provides additional predefines for @code{_XOPEN_UNIX},
13146 @code{_XOPEN_SOURCE_EXTENDED}, @code{_INCLUDE__STDC_A1_SOURCE} and
13147 @code{_INCLUDE_XOPEN_SOURCE_500}, and the startfile @file{unix98.o}.
13149 It is @emph{important} to note that this option changes the interfaces
13150 for various library routines. It also affects the operational behavior
13151 of the C library. Thus, @emph{extreme} care is needed in using this
13154 Library code that is intended to operate with more than one UNIX
13155 standard must test, set and restore the variable @var{__xpg4_extended_mask}
13156 as appropriate. Most GNU software doesn't provide this capability.
13160 Suppress the generation of link options to search libdld.sl when the
13161 @option{-static} option is specified on HP-UX 10 and later.
13165 The HP-UX implementation of setlocale in libc has a dependency on
13166 libdld.sl. There isn't an archive version of libdld.sl. Thus,
13167 when the @option{-static} option is specified, special link options
13168 are needed to resolve this dependency.
13170 On HP-UX 10 and later, the GCC driver adds the necessary options to
13171 link with libdld.sl when the @option{-static} option is specified.
13172 This causes the resulting binary to be dynamic. On the 64-bit port,
13173 the linkers generate dynamic binaries by default in any case. The
13174 @option{-nolibdld} option can be used to prevent the GCC driver from
13175 adding these link options.
13179 Add support for multithreading with the @dfn{dce thread} library
13180 under HP-UX@. This option sets flags for both the preprocessor and
13184 @node i386 and x86-64 Options
13185 @subsection Intel 386 and AMD x86-64 Options
13186 @cindex i386 Options
13187 @cindex x86-64 Options
13188 @cindex Intel 386 Options
13189 @cindex AMD x86-64 Options
13191 These @samp{-m} options are defined for the i386 and x86-64 family of
13195 @item -mtune=@var{cpu-type}
13197 Tune to @var{cpu-type} everything applicable about the generated code, except
13198 for the ABI and the set of available instructions. The choices for
13199 @var{cpu-type} are:
13202 Produce code optimized for the most common IA32/@/AMD64/@/EM64T processors.
13203 If you know the CPU on which your code will run, then you should use
13204 the corresponding @option{-mtune} option instead of
13205 @option{-mtune=generic}. But, if you do not know exactly what CPU users
13206 of your application will have, then you should use this option.
13208 As new processors are deployed in the marketplace, the behavior of this
13209 option will change. Therefore, if you upgrade to a newer version of
13210 GCC, the code generated option will change to reflect the processors
13211 that were most common when that version of GCC was released.
13213 There is no @option{-march=generic} option because @option{-march}
13214 indicates the instruction set the compiler can use, and there is no
13215 generic instruction set applicable to all processors. In contrast,
13216 @option{-mtune} indicates the processor (or, in this case, collection of
13217 processors) for which the code is optimized.
13219 This selects the CPU to tune for at compilation time by determining
13220 the processor type of the compiling machine. Using @option{-mtune=native}
13221 will produce code optimized for the local machine under the constraints
13222 of the selected instruction set. Using @option{-march=native} will
13223 enable all instruction subsets supported by the local machine (hence
13224 the result might not run on different machines).
13226 Original Intel's i386 CPU@.
13228 Intel's i486 CPU@. (No scheduling is implemented for this chip.)
13229 @item i586, pentium
13230 Intel Pentium CPU with no MMX support.
13232 Intel PentiumMMX CPU based on Pentium core with MMX instruction set support.
13234 Intel PentiumPro CPU@.
13236 Same as @code{generic}, but when used as @code{march} option, PentiumPro
13237 instruction set will be used, so the code will run on all i686 family chips.
13239 Intel Pentium2 CPU based on PentiumPro core with MMX instruction set support.
13240 @item pentium3, pentium3m
13241 Intel Pentium3 CPU based on PentiumPro core with MMX and SSE instruction set
13244 Low power version of Intel Pentium3 CPU with MMX, SSE and SSE2 instruction set
13245 support. Used by Centrino notebooks.
13246 @item pentium4, pentium4m
13247 Intel Pentium4 CPU with MMX, SSE and SSE2 instruction set support.
13249 Improved version of Intel Pentium4 CPU with MMX, SSE, SSE2 and SSE3 instruction
13252 Improved version of Intel Pentium4 CPU with 64-bit extensions, MMX, SSE,
13253 SSE2 and SSE3 instruction set support.
13255 Intel Core2 CPU with 64-bit extensions, MMX, SSE, SSE2, SSE3 and SSSE3
13256 instruction set support.
13258 Intel Core i7 CPU with 64-bit extensions, MMX, SSE, SSE2, SSE3, SSSE3, SSE4.1
13259 and SSE4.2 instruction set support.
13261 Intel Core i7 CPU with 64-bit extensions, MMX, SSE, SSE2, SSE3, SSSE3,
13262 SSE4.1, SSE4.2, AVX, AES and PCLMUL instruction set support.
13264 Intel Core CPU with 64-bit extensions, MMX, SSE, SSE2, SSE3, SSSE3,
13265 SSE4.1, SSE4.2, AVX, AES, PCLMUL, FSGSBASE, RDRND and F16C instruction
13268 Intel Atom CPU with 64-bit extensions, MMX, SSE, SSE2, SSE3 and SSSE3
13269 instruction set support.
13271 AMD K6 CPU with MMX instruction set support.
13273 Improved versions of AMD K6 CPU with MMX and 3DNow!@: instruction set support.
13274 @item athlon, athlon-tbird
13275 AMD Athlon CPU with MMX, 3dNOW!, enhanced 3DNow!@: and SSE prefetch instructions
13277 @item athlon-4, athlon-xp, athlon-mp
13278 Improved AMD Athlon CPU with MMX, 3DNow!, enhanced 3DNow!@: and full SSE
13279 instruction set support.
13280 @item k8, opteron, athlon64, athlon-fx
13281 AMD K8 core based CPUs with x86-64 instruction set support. (This supersets
13282 MMX, SSE, SSE2, 3DNow!, enhanced 3DNow!@: and 64-bit instruction set extensions.)
13283 @item k8-sse3, opteron-sse3, athlon64-sse3
13284 Improved versions of k8, opteron and athlon64 with SSE3 instruction set support.
13285 @item amdfam10, barcelona
13286 AMD Family 10h core based CPUs with x86-64 instruction set support. (This
13287 supersets MMX, SSE, SSE2, SSE3, SSE4A, 3DNow!, enhanced 3DNow!, ABM and 64-bit
13288 instruction set extensions.)
13290 AMD Family 15h core based CPUs with x86-64 instruction set support. (This
13291 supersets FMA4, AVX, XOP, LWP, AES, PCL_MUL, CX16, MMX, SSE, SSE2, SSE3, SSE4A,
13292 SSSE3, SSE4.1, SSE4.2, ABM and 64-bit instruction set extensions.)
13294 AMD Family 15h core based CPUs with x86-64 instruction set support. (This
13295 supersets BMI, TBM, F16C, FMA, AVX, XOP, LWP, AES, PCL_MUL, CX16, MMX, SSE,
13296 SSE2, SSE3, SSE4A, SSSE3, SSE4.1, SSE4.2, ABM and 64-bit instruction set
13299 AMD Family 14h core based CPUs with x86-64 instruction set support. (This
13300 supersets MMX, SSE, SSE2, SSE3, SSSE3, SSE4A, CX16, ABM and 64-bit
13301 instruction set extensions.)
13303 IDT Winchip C6 CPU, dealt in same way as i486 with additional MMX instruction
13306 IDT Winchip2 CPU, dealt in same way as i486 with additional MMX and 3DNow!@:
13307 instruction set support.
13309 Via C3 CPU with MMX and 3DNow!@: instruction set support. (No scheduling is
13310 implemented for this chip.)
13312 Via C3-2 CPU with MMX and SSE instruction set support. (No scheduling is
13313 implemented for this chip.)
13315 Embedded AMD CPU with MMX and 3DNow!@: instruction set support.
13318 While picking a specific @var{cpu-type} will schedule things appropriately
13319 for that particular chip, the compiler will not generate any code that
13320 does not run on the default machine type without the @option{-march=@var{cpu-type}}
13321 option being used. For example, if GCC is configured for i686-pc-linux-gnu
13322 then @option{-mtune=pentium4} will generate code that is tuned for Pentium4
13323 but will still run on i686 machines.
13325 @item -march=@var{cpu-type}
13327 Generate instructions for the machine type @var{cpu-type}. The choices
13328 for @var{cpu-type} are the same as for @option{-mtune}. Moreover,
13329 specifying @option{-march=@var{cpu-type}} implies @option{-mtune=@var{cpu-type}}.
13331 @item -mcpu=@var{cpu-type}
13333 A deprecated synonym for @option{-mtune}.
13335 @item -mfpmath=@var{unit}
13337 Generate floating-point arithmetic for selected unit @var{unit}. The choices
13338 for @var{unit} are:
13342 Use the standard 387 floating-point coprocessor present on the majority of chips and
13343 emulated otherwise. Code compiled with this option runs almost everywhere.
13344 The temporary results are computed in 80-bit precision instead of the precision
13345 specified by the type, resulting in slightly different results compared to most
13346 of other chips. See @option{-ffloat-store} for more detailed description.
13348 This is the default choice for i386 compiler.
13351 Use scalar floating-point instructions present in the SSE instruction set.
13352 This instruction set is supported by Pentium3 and newer chips, in the AMD line
13353 by Athlon-4, Athlon-xp and Athlon-mp chips. The earlier version of SSE
13354 instruction set supports only single-precision arithmetic, thus the double and
13355 extended-precision arithmetic are still done using 387. A later version, present
13356 only in Pentium4 and the future AMD x86-64 chips, supports double-precision
13359 For the i386 compiler, you need to use @option{-march=@var{cpu-type}}, @option{-msse}
13360 or @option{-msse2} switches to enable SSE extensions and make this option
13361 effective. For the x86-64 compiler, these extensions are enabled by default.
13363 The resulting code should be considerably faster in the majority of cases and avoid
13364 the numerical instability problems of 387 code, but may break some existing
13365 code that expects temporaries to be 80 bits.
13367 This is the default choice for the x86-64 compiler.
13372 Attempt to utilize both instruction sets at once. This effectively double the
13373 amount of available registers and on chips with separate execution units for
13374 387 and SSE the execution resources too. Use this option with care, as it is
13375 still experimental, because the GCC register allocator does not model separate
13376 functional units well resulting in instable performance.
13379 @item -masm=@var{dialect}
13380 @opindex masm=@var{dialect}
13381 Output asm instructions using selected @var{dialect}. Supported
13382 choices are @samp{intel} or @samp{att} (the default one). Darwin does
13383 not support @samp{intel}.
13386 @itemx -mno-ieee-fp
13388 @opindex mno-ieee-fp
13389 Control whether or not the compiler uses IEEE floating-point
13390 comparisons. These handle correctly the case where the result of a
13391 comparison is unordered.
13394 @opindex msoft-float
13395 Generate output containing library calls for floating point.
13396 @strong{Warning:} the requisite libraries are not part of GCC@.
13397 Normally the facilities of the machine's usual C compiler are used, but
13398 this can't be done directly in cross-compilation. You must make your
13399 own arrangements to provide suitable library functions for
13402 On machines where a function returns floating-point results in the 80387
13403 register stack, some floating-point opcodes may be emitted even if
13404 @option{-msoft-float} is used.
13406 @item -mno-fp-ret-in-387
13407 @opindex mno-fp-ret-in-387
13408 Do not use the FPU registers for return values of functions.
13410 The usual calling convention has functions return values of types
13411 @code{float} and @code{double} in an FPU register, even if there
13412 is no FPU@. The idea is that the operating system should emulate
13415 The option @option{-mno-fp-ret-in-387} causes such values to be returned
13416 in ordinary CPU registers instead.
13418 @item -mno-fancy-math-387
13419 @opindex mno-fancy-math-387
13420 Some 387 emulators do not support the @code{sin}, @code{cos} and
13421 @code{sqrt} instructions for the 387. Specify this option to avoid
13422 generating those instructions. This option is the default on FreeBSD,
13423 OpenBSD and NetBSD@. This option is overridden when @option{-march}
13424 indicates that the target CPU will always have an FPU and so the
13425 instruction will not need emulation. As of revision 2.6.1, these
13426 instructions are not generated unless you also use the
13427 @option{-funsafe-math-optimizations} switch.
13429 @item -malign-double
13430 @itemx -mno-align-double
13431 @opindex malign-double
13432 @opindex mno-align-double
13433 Control whether GCC aligns @code{double}, @code{long double}, and
13434 @code{long long} variables on a two-word boundary or a one-word
13435 boundary. Aligning @code{double} variables on a two-word boundary
13436 produces code that runs somewhat faster on a @samp{Pentium} at the
13437 expense of more memory.
13439 On x86-64, @option{-malign-double} is enabled by default.
13441 @strong{Warning:} if you use the @option{-malign-double} switch,
13442 structures containing the above types will be aligned differently than
13443 the published application binary interface specifications for the 386
13444 and will not be binary compatible with structures in code compiled
13445 without that switch.
13447 @item -m96bit-long-double
13448 @itemx -m128bit-long-double
13449 @opindex m96bit-long-double
13450 @opindex m128bit-long-double
13451 These switches control the size of @code{long double} type. The i386
13452 application binary interface specifies the size to be 96 bits,
13453 so @option{-m96bit-long-double} is the default in 32-bit mode.
13455 Modern architectures (Pentium and newer) prefer @code{long double}
13456 to be aligned to an 8- or 16-byte boundary. In arrays or structures
13457 conforming to the ABI, this is not possible. So specifying
13458 @option{-m128bit-long-double} aligns @code{long double}
13459 to a 16-byte boundary by padding the @code{long double} with an additional
13462 In the x86-64 compiler, @option{-m128bit-long-double} is the default choice as
13463 its ABI specifies that @code{long double} is to be aligned on 16-byte boundary.
13465 Notice that neither of these options enable any extra precision over the x87
13466 standard of 80 bits for a @code{long double}.
13468 @strong{Warning:} if you override the default value for your target ABI, the
13469 structures and arrays containing @code{long double} variables will change
13470 their size as well as function calling convention for function taking
13471 @code{long double} will be modified. Hence they will not be binary
13472 compatible with arrays or structures in code compiled without that switch.
13474 @item -mlarge-data-threshold=@var{number}
13475 @opindex mlarge-data-threshold=@var{number}
13476 When @option{-mcmodel=medium} is specified, the data greater than
13477 @var{threshold} are placed in large data section. This value must be the
13478 same across all object linked into the binary and defaults to 65535.
13482 Use a different function-calling convention, in which functions that
13483 take a fixed number of arguments return with the @code{ret} @var{num}
13484 instruction, which pops their arguments while returning. This saves one
13485 instruction in the caller since there is no need to pop the arguments
13488 You can specify that an individual function is called with this calling
13489 sequence with the function attribute @samp{stdcall}. You can also
13490 override the @option{-mrtd} option by using the function attribute
13491 @samp{cdecl}. @xref{Function Attributes}.
13493 @strong{Warning:} this calling convention is incompatible with the one
13494 normally used on Unix, so you cannot use it if you need to call
13495 libraries compiled with the Unix compiler.
13497 Also, you must provide function prototypes for all functions that
13498 take variable numbers of arguments (including @code{printf});
13499 otherwise incorrect code will be generated for calls to those
13502 In addition, seriously incorrect code will result if you call a
13503 function with too many arguments. (Normally, extra arguments are
13504 harmlessly ignored.)
13506 @item -mregparm=@var{num}
13508 Control how many registers are used to pass integer arguments. By
13509 default, no registers are used to pass arguments, and at most 3
13510 registers can be used. You can control this behavior for a specific
13511 function by using the function attribute @samp{regparm}.
13512 @xref{Function Attributes}.
13514 @strong{Warning:} if you use this switch, and
13515 @var{num} is nonzero, then you must build all modules with the same
13516 value, including any libraries. This includes the system libraries and
13520 @opindex msseregparm
13521 Use SSE register passing conventions for float and double arguments
13522 and return values. You can control this behavior for a specific
13523 function by using the function attribute @samp{sseregparm}.
13524 @xref{Function Attributes}.
13526 @strong{Warning:} if you use this switch then you must build all
13527 modules with the same value, including any libraries. This includes
13528 the system libraries and startup modules.
13530 @item -mvect8-ret-in-mem
13531 @opindex mvect8-ret-in-mem
13532 Return 8-byte vectors in memory instead of MMX registers. This is the
13533 default on Solaris@tie{}8 and 9 and VxWorks to match the ABI of the Sun
13534 Studio compilers until version 12. Later compiler versions (starting
13535 with Studio 12 Update@tie{}1) follow the ABI used by other x86 targets, which
13536 is the default on Solaris@tie{}10 and later. @emph{Only} use this option if
13537 you need to remain compatible with existing code produced by those
13538 previous compiler versions or older versions of GCC.
13547 Set 80387 floating-point precision to 32, 64 or 80 bits. When @option{-mpc32}
13548 is specified, the significands of results of floating-point operations are
13549 rounded to 24 bits (single precision); @option{-mpc64} rounds the
13550 significands of results of floating-point operations to 53 bits (double
13551 precision) and @option{-mpc80} rounds the significands of results of
13552 floating-point operations to 64 bits (extended double precision), which is
13553 the default. When this option is used, floating-point operations in higher
13554 precisions are not available to the programmer without setting the FPU
13555 control word explicitly.
13557 Setting the rounding of floating-point operations to less than the default
13558 80 bits can speed some programs by 2% or more. Note that some mathematical
13559 libraries assume that extended-precision (80-bit) floating-point operations
13560 are enabled by default; routines in such libraries could suffer significant
13561 loss of accuracy, typically through so-called "catastrophic cancellation",
13562 when this option is used to set the precision to less than extended precision.
13564 @item -mstackrealign
13565 @opindex mstackrealign
13566 Realign the stack at entry. On the Intel x86, the @option{-mstackrealign}
13567 option will generate an alternate prologue and epilogue that realigns the
13568 run-time stack if necessary. This supports mixing legacy codes that keep
13569 a 4-byte aligned stack with modern codes that keep a 16-byte stack for
13570 SSE compatibility. See also the attribute @code{force_align_arg_pointer},
13571 applicable to individual functions.
13573 @item -mpreferred-stack-boundary=@var{num}
13574 @opindex mpreferred-stack-boundary
13575 Attempt to keep the stack boundary aligned to a 2 raised to @var{num}
13576 byte boundary. If @option{-mpreferred-stack-boundary} is not specified,
13577 the default is 4 (16 bytes or 128 bits).
13579 @item -mincoming-stack-boundary=@var{num}
13580 @opindex mincoming-stack-boundary
13581 Assume the incoming stack is aligned to a 2 raised to @var{num} byte
13582 boundary. If @option{-mincoming-stack-boundary} is not specified,
13583 the one specified by @option{-mpreferred-stack-boundary} will be used.
13585 On Pentium and PentiumPro, @code{double} and @code{long double} values
13586 should be aligned to an 8-byte boundary (see @option{-malign-double}) or
13587 suffer significant run time performance penalties. On Pentium III, the
13588 Streaming SIMD Extension (SSE) data type @code{__m128} may not work
13589 properly if it is not 16-byte aligned.
13591 To ensure proper alignment of this values on the stack, the stack boundary
13592 must be as aligned as that required by any value stored on the stack.
13593 Further, every function must be generated such that it keeps the stack
13594 aligned. Thus calling a function compiled with a higher preferred
13595 stack boundary from a function compiled with a lower preferred stack
13596 boundary will most likely misalign the stack. It is recommended that
13597 libraries that use callbacks always use the default setting.
13599 This extra alignment does consume extra stack space, and generally
13600 increases code size. Code that is sensitive to stack space usage, such
13601 as embedded systems and operating system kernels, may want to reduce the
13602 preferred alignment to @option{-mpreferred-stack-boundary=2}.
13631 @itemx -mno-fsgsbase
13667 These switches enable or disable the use of instructions in the MMX, SSE,
13668 SSE2, SSE3, SSSE3, SSE4.1, AVX, AVX2, AES, PCLMUL, FSGSBASE, RDRND, F16C,
13669 FMA, SSE4A, FMA4, XOP, LWP, ABM, BMI, BMI2, LZCNT or 3DNow!
13670 @: extended instruction sets.
13671 These extensions are also available as built-in functions: see
13672 @ref{X86 Built-in Functions}, for details of the functions enabled and
13673 disabled by these switches.
13675 To have SSE/SSE2 instructions generated automatically from floating-point
13676 code (as opposed to 387 instructions), see @option{-mfpmath=sse}.
13678 GCC depresses SSEx instructions when @option{-mavx} is used. Instead, it
13679 generates new AVX instructions or AVX equivalence for all SSEx instructions
13682 These options will enable GCC to use these extended instructions in
13683 generated code, even without @option{-mfpmath=sse}. Applications that
13684 perform run-time CPU detection must compile separate files for each
13685 supported architecture, using the appropriate flags. In particular,
13686 the file containing the CPU detection code should be compiled without
13691 This option instructs GCC to emit a @code{cld} instruction in the prologue
13692 of functions that use string instructions. String instructions depend on
13693 the DF flag to select between autoincrement or autodecrement mode. While the
13694 ABI specifies the DF flag to be cleared on function entry, some operating
13695 systems violate this specification by not clearing the DF flag in their
13696 exception dispatchers. The exception handler can be invoked with the DF flag
13697 set, which leads to wrong direction mode when string instructions are used.
13698 This option can be enabled by default on 32-bit x86 targets by configuring
13699 GCC with the @option{--enable-cld} configure option. Generation of @code{cld}
13700 instructions can be suppressed with the @option{-mno-cld} compiler option
13704 @opindex mvzeroupper
13705 This option instructs GCC to emit a @code{vzeroupper} instruction
13706 before a transfer of control flow out of the function to minimize
13707 AVX to SSE transition penalty as well as remove unnecessary zeroupper
13710 @item -mprefer-avx128
13711 @opindex mprefer-avx128
13712 This option instructs GCC to use 128-bit AVX instructions instead of
13713 256-bit AVX instructions in the auto-vectorizer.
13717 This option will enable GCC to use CMPXCHG16B instruction in generated code.
13718 CMPXCHG16B allows for atomic operations on 128-bit double quadword (or oword)
13719 data types. This is useful for high resolution counters that could be updated
13720 by multiple processors (or cores). This instruction is generated as part of
13721 atomic built-in functions: see @ref{__sync Builtins} or
13722 @ref{__atomic Builtins} for details.
13726 This option will enable GCC to use SAHF instruction in generated 64-bit code.
13727 Early Intel CPUs with Intel 64 lacked LAHF and SAHF instructions supported
13728 by AMD64 until introduction of Pentium 4 G1 step in December 2005. LAHF and
13729 SAHF are load and store instructions, respectively, for certain status flags.
13730 In 64-bit mode, SAHF instruction is used to optimize @code{fmod}, @code{drem}
13731 or @code{remainder} built-in functions: see @ref{Other Builtins} for details.
13735 This option will enable GCC to use movbe instruction to implement
13736 @code{__builtin_bswap32} and @code{__builtin_bswap64}.
13740 This option will enable built-in functions, @code{__builtin_ia32_crc32qi},
13741 @code{__builtin_ia32_crc32hi}. @code{__builtin_ia32_crc32si} and
13742 @code{__builtin_ia32_crc32di} to generate the crc32 machine instruction.
13746 This option will enable GCC to use RCPSS and RSQRTSS instructions (and their
13747 vectorized variants RCPPS and RSQRTPS) with an additional Newton-Raphson step
13748 to increase precision instead of DIVSS and SQRTSS (and their vectorized
13749 variants) for single-precision floating-point arguments. These instructions
13750 are generated only when @option{-funsafe-math-optimizations} is enabled
13751 together with @option{-finite-math-only} and @option{-fno-trapping-math}.
13752 Note that while the throughput of the sequence is higher than the throughput
13753 of the non-reciprocal instruction, the precision of the sequence can be
13754 decreased by up to 2 ulp (i.e. the inverse of 1.0 equals 0.99999994).
13756 Note that GCC implements @code{1.0f/sqrtf(@var{x})} in terms of RSQRTSS
13757 (or RSQRTPS) already with @option{-ffast-math} (or the above option
13758 combination), and doesn't need @option{-mrecip}.
13760 Also note that GCC emits the above sequence with additional Newton-Raphson step
13761 for vectorized single-float division and vectorized @code{sqrtf(@var{x})}
13762 already with @option{-ffast-math} (or the above option combination), and
13763 doesn't need @option{-mrecip}.
13765 @item -mrecip=@var{opt}
13766 @opindex mrecip=opt
13767 This option allows to control which reciprocal estimate instructions
13768 may be used. @var{opt} is a comma separated list of options, which may
13769 be preceded by a @code{!} to invert the option:
13770 @code{all}: enable all estimate instructions,
13771 @code{default}: enable the default instructions, equivalent to @option{-mrecip},
13772 @code{none}: disable all estimate instructions, equivalent to @option{-mno-recip},
13773 @code{div}: enable the approximation for scalar division,
13774 @code{vec-div}: enable the approximation for vectorized division,
13775 @code{sqrt}: enable the approximation for scalar square root,
13776 @code{vec-sqrt}: enable the approximation for vectorized square root.
13778 So for example, @option{-mrecip=all,!sqrt} would enable
13779 all of the reciprocal approximations, except for square root.
13781 @item -mveclibabi=@var{type}
13782 @opindex mveclibabi
13783 Specifies the ABI type to use for vectorizing intrinsics using an
13784 external library. Supported types are @code{svml} for the Intel short
13785 vector math library and @code{acml} for the AMD math core library style
13786 of interfacing. GCC will currently emit calls to @code{vmldExp2},
13787 @code{vmldLn2}, @code{vmldLog102}, @code{vmldLog102}, @code{vmldPow2},
13788 @code{vmldTanh2}, @code{vmldTan2}, @code{vmldAtan2}, @code{vmldAtanh2},
13789 @code{vmldCbrt2}, @code{vmldSinh2}, @code{vmldSin2}, @code{vmldAsinh2},
13790 @code{vmldAsin2}, @code{vmldCosh2}, @code{vmldCos2}, @code{vmldAcosh2},
13791 @code{vmldAcos2}, @code{vmlsExp4}, @code{vmlsLn4}, @code{vmlsLog104},
13792 @code{vmlsLog104}, @code{vmlsPow4}, @code{vmlsTanh4}, @code{vmlsTan4},
13793 @code{vmlsAtan4}, @code{vmlsAtanh4}, @code{vmlsCbrt4}, @code{vmlsSinh4},
13794 @code{vmlsSin4}, @code{vmlsAsinh4}, @code{vmlsAsin4}, @code{vmlsCosh4},
13795 @code{vmlsCos4}, @code{vmlsAcosh4} and @code{vmlsAcos4} for corresponding
13796 function type when @option{-mveclibabi=svml} is used and @code{__vrd2_sin},
13797 @code{__vrd2_cos}, @code{__vrd2_exp}, @code{__vrd2_log}, @code{__vrd2_log2},
13798 @code{__vrd2_log10}, @code{__vrs4_sinf}, @code{__vrs4_cosf},
13799 @code{__vrs4_expf}, @code{__vrs4_logf}, @code{__vrs4_log2f},
13800 @code{__vrs4_log10f} and @code{__vrs4_powf} for corresponding function type
13801 when @option{-mveclibabi=acml} is used. Both @option{-ftree-vectorize} and
13802 @option{-funsafe-math-optimizations} have to be enabled. A SVML or ACML ABI
13803 compatible library will have to be specified at link time.
13805 @item -mabi=@var{name}
13807 Generate code for the specified calling convention. Permissible values
13808 are: @samp{sysv} for the ABI used on GNU/Linux and other systems and
13809 @samp{ms} for the Microsoft ABI. The default is to use the Microsoft
13810 ABI when targeting Windows. On all other systems, the default is the
13811 SYSV ABI. You can control this behavior for a specific function by
13812 using the function attribute @samp{ms_abi}/@samp{sysv_abi}.
13813 @xref{Function Attributes}.
13815 @item -mtls-dialect=@var{type}
13816 @opindex mtls-dialect
13817 Generate code to access thread-local storage using the @samp{gnu} or
13818 @samp{gnu2} conventions. @samp{gnu} is the conservative default;
13819 @samp{gnu2} is more efficient, but it may add compile- and run-time
13820 requirements that cannot be satisfied on all systems.
13823 @itemx -mno-push-args
13824 @opindex mpush-args
13825 @opindex mno-push-args
13826 Use PUSH operations to store outgoing parameters. This method is shorter
13827 and usually equally fast as method using SUB/MOV operations and is enabled
13828 by default. In some cases disabling it may improve performance because of
13829 improved scheduling and reduced dependencies.
13831 @item -maccumulate-outgoing-args
13832 @opindex maccumulate-outgoing-args
13833 If enabled, the maximum amount of space required for outgoing arguments will be
13834 computed in the function prologue. This is faster on most modern CPUs
13835 because of reduced dependencies, improved scheduling and reduced stack usage
13836 when preferred stack boundary is not equal to 2. The drawback is a notable
13837 increase in code size. This switch implies @option{-mno-push-args}.
13841 Support thread-safe exception handling on @samp{Mingw32}. Code that relies
13842 on thread-safe exception handling must compile and link all code with the
13843 @option{-mthreads} option. When compiling, @option{-mthreads} defines
13844 @option{-D_MT}; when linking, it links in a special thread helper library
13845 @option{-lmingwthrd} which cleans up per thread exception handling data.
13847 @item -mno-align-stringops
13848 @opindex mno-align-stringops
13849 Do not align destination of inlined string operations. This switch reduces
13850 code size and improves performance in case the destination is already aligned,
13851 but GCC doesn't know about it.
13853 @item -minline-all-stringops
13854 @opindex minline-all-stringops
13855 By default GCC inlines string operations only when the destination is
13856 known to be aligned to least a 4-byte boundary.
13857 This enables more inlining, increase code
13858 size, but may improve performance of code that depends on fast memcpy, strlen
13859 and memset for short lengths.
13861 @item -minline-stringops-dynamically
13862 @opindex minline-stringops-dynamically
13863 For string operations of unknown size, use run-time checks with
13864 inline code for small blocks and a library call for large blocks.
13866 @item -mstringop-strategy=@var{alg}
13867 @opindex mstringop-strategy=@var{alg}
13868 Overwrite internal decision heuristic about particular algorithm to inline
13869 string operation with. The allowed values are @code{rep_byte},
13870 @code{rep_4byte}, @code{rep_8byte} for expanding using i386 @code{rep} prefix
13871 of specified size, @code{byte_loop}, @code{loop}, @code{unrolled_loop} for
13872 expanding inline loop, @code{libcall} for always expanding library call.
13874 @item -momit-leaf-frame-pointer
13875 @opindex momit-leaf-frame-pointer
13876 Don't keep the frame pointer in a register for leaf functions. This
13877 avoids the instructions to save, set up and restore frame pointers and
13878 makes an extra register available in leaf functions. The option
13879 @option{-fomit-frame-pointer} removes the frame pointer for all functions,
13880 which might make debugging harder.
13882 @item -mtls-direct-seg-refs
13883 @itemx -mno-tls-direct-seg-refs
13884 @opindex mtls-direct-seg-refs
13885 Controls whether TLS variables may be accessed with offsets from the
13886 TLS segment register (@code{%gs} for 32-bit, @code{%fs} for 64-bit),
13887 or whether the thread base pointer must be added. Whether or not this
13888 is legal depends on the operating system, and whether it maps the
13889 segment to cover the entire TLS area.
13891 For systems that use GNU libc, the default is on.
13894 @itemx -mno-sse2avx
13896 Specify that the assembler should encode SSE instructions with VEX
13897 prefix. The option @option{-mavx} turns this on by default.
13902 If profiling is active @option{-pg} put the profiling
13903 counter call before prologue.
13904 Note: On x86 architectures the attribute @code{ms_hook_prologue}
13905 isn't possible at the moment for @option{-mfentry} and @option{-pg}.
13908 @itemx -mno-8bit-idiv
13910 On some processors, like Intel Atom, 8-bit unsigned integer divide is
13911 much faster than 32-bit/64-bit integer divide. This option generates a
13912 run-time check. If both dividend and divisor are within range of 0
13913 to 255, 8-bit unsigned integer divide is used instead of
13914 32-bit/64-bit integer divide.
13916 @item -mavx256-split-unaligned-load
13917 @item -mavx256-split-unaligned-store
13918 @opindex avx256-split-unaligned-load
13919 @opindex avx256-split-unaligned-store
13920 Split 32-byte AVX unaligned load and store.
13924 These @samp{-m} switches are supported in addition to the above
13925 on AMD x86-64 processors in 64-bit environments.
13934 Generate code for a 32-bit or 64-bit environment.
13935 The @option{-m32} option sets int, long and pointer to 32 bits and
13936 generates code that runs on any i386 system.
13937 The @option{-m64} option sets int to 32 bits and long and pointer
13938 to 64 bits and generates code for AMD's x86-64 architecture.
13939 The @option{-mx32} option sets int, long and pointer to 32 bits and
13940 generates code for AMD's x86-64 architecture.
13941 For darwin only the @option{-m64} option turns off the @option{-fno-pic}
13942 and @option{-mdynamic-no-pic} options.
13944 @item -mno-red-zone
13945 @opindex mno-red-zone
13946 Do not use a so called red zone for x86-64 code. The red zone is mandated
13947 by the x86-64 ABI, it is a 128-byte area beyond the location of the
13948 stack pointer that will not be modified by signal or interrupt handlers
13949 and therefore can be used for temporary data without adjusting the stack
13950 pointer. The flag @option{-mno-red-zone} disables this red zone.
13952 @item -mcmodel=small
13953 @opindex mcmodel=small
13954 Generate code for the small code model: the program and its symbols must
13955 be linked in the lower 2 GB of the address space. Pointers are 64 bits.
13956 Programs can be statically or dynamically linked. This is the default
13959 @item -mcmodel=kernel
13960 @opindex mcmodel=kernel
13961 Generate code for the kernel code model. The kernel runs in the
13962 negative 2 GB of the address space.
13963 This model has to be used for Linux kernel code.
13965 @item -mcmodel=medium
13966 @opindex mcmodel=medium
13967 Generate code for the medium model: The program is linked in the lower 2
13968 GB of the address space. Small symbols are also placed there. Symbols
13969 with sizes larger than @option{-mlarge-data-threshold} are put into
13970 large data or bss sections and can be located above 2GB. Programs can
13971 be statically or dynamically linked.
13973 @item -mcmodel=large
13974 @opindex mcmodel=large
13975 Generate code for the large model: This model makes no assumptions
13976 about addresses and sizes of sections.
13979 @node i386 and x86-64 Windows Options
13980 @subsection i386 and x86-64 Windows Options
13981 @cindex i386 and x86-64 Windows Options
13983 These additional options are available for Windows targets:
13988 This option is available for Cygwin and MinGW targets. It
13989 specifies that a console application is to be generated, by
13990 instructing the linker to set the PE header subsystem type
13991 required for console applications.
13992 This is the default behavior for Cygwin and MinGW targets.
13996 This option is available for Cygwin and MinGW targets. It
13997 specifies that a DLL - a dynamic link library - is to be
13998 generated, enabling the selection of the required runtime
13999 startup object and entry point.
14001 @item -mnop-fun-dllimport
14002 @opindex mnop-fun-dllimport
14003 This option is available for Cygwin and MinGW targets. It
14004 specifies that the dllimport attribute should be ignored.
14008 This option is available for MinGW targets. It specifies
14009 that MinGW-specific thread support is to be used.
14013 This option is available for mingw-w64 targets. It specifies
14014 that the UNICODE macro is getting pre-defined and that the
14015 unicode capable runtime startup code is chosen.
14019 This option is available for Cygwin and MinGW targets. It
14020 specifies that the typical Windows pre-defined macros are to
14021 be set in the pre-processor, but does not influence the choice
14022 of runtime library/startup code.
14026 This option is available for Cygwin and MinGW targets. It
14027 specifies that a GUI application is to be generated by
14028 instructing the linker to set the PE header subsystem type
14031 @item -fno-set-stack-executable
14032 @opindex fno-set-stack-executable
14033 This option is available for MinGW targets. It specifies that
14034 the executable flag for stack used by nested functions isn't
14035 set. This is necessary for binaries running in kernel mode of
14036 Windows, as there the user32 API, which is used to set executable
14037 privileges, isn't available.
14039 @item -mpe-aligned-commons
14040 @opindex mpe-aligned-commons
14041 This option is available for Cygwin and MinGW targets. It
14042 specifies that the GNU extension to the PE file format that
14043 permits the correct alignment of COMMON variables should be
14044 used when generating code. It will be enabled by default if
14045 GCC detects that the target assembler found during configuration
14046 supports the feature.
14049 See also under @ref{i386 and x86-64 Options} for standard options.
14051 @node IA-64 Options
14052 @subsection IA-64 Options
14053 @cindex IA-64 Options
14055 These are the @samp{-m} options defined for the Intel IA-64 architecture.
14059 @opindex mbig-endian
14060 Generate code for a big-endian target. This is the default for HP-UX@.
14062 @item -mlittle-endian
14063 @opindex mlittle-endian
14064 Generate code for a little-endian target. This is the default for AIX5
14070 @opindex mno-gnu-as
14071 Generate (or don't) code for the GNU assembler. This is the default.
14072 @c Also, this is the default if the configure option @option{--with-gnu-as}
14078 @opindex mno-gnu-ld
14079 Generate (or don't) code for the GNU linker. This is the default.
14080 @c Also, this is the default if the configure option @option{--with-gnu-ld}
14085 Generate code that does not use a global pointer register. The result
14086 is not position independent code, and violates the IA-64 ABI@.
14088 @item -mvolatile-asm-stop
14089 @itemx -mno-volatile-asm-stop
14090 @opindex mvolatile-asm-stop
14091 @opindex mno-volatile-asm-stop
14092 Generate (or don't) a stop bit immediately before and after volatile asm
14095 @item -mregister-names
14096 @itemx -mno-register-names
14097 @opindex mregister-names
14098 @opindex mno-register-names
14099 Generate (or don't) @samp{in}, @samp{loc}, and @samp{out} register names for
14100 the stacked registers. This may make assembler output more readable.
14106 Disable (or enable) optimizations that use the small data section. This may
14107 be useful for working around optimizer bugs.
14109 @item -mconstant-gp
14110 @opindex mconstant-gp
14111 Generate code that uses a single constant global pointer value. This is
14112 useful when compiling kernel code.
14116 Generate code that is self-relocatable. This implies @option{-mconstant-gp}.
14117 This is useful when compiling firmware code.
14119 @item -minline-float-divide-min-latency
14120 @opindex minline-float-divide-min-latency
14121 Generate code for inline divides of floating-point values
14122 using the minimum latency algorithm.
14124 @item -minline-float-divide-max-throughput
14125 @opindex minline-float-divide-max-throughput
14126 Generate code for inline divides of floating-point values
14127 using the maximum throughput algorithm.
14129 @item -mno-inline-float-divide
14130 @opindex mno-inline-float-divide
14131 Do not generate inline code for divides of floating-point values.
14133 @item -minline-int-divide-min-latency
14134 @opindex minline-int-divide-min-latency
14135 Generate code for inline divides of integer values
14136 using the minimum latency algorithm.
14138 @item -minline-int-divide-max-throughput
14139 @opindex minline-int-divide-max-throughput
14140 Generate code for inline divides of integer values
14141 using the maximum throughput algorithm.
14143 @item -mno-inline-int-divide
14144 @opindex mno-inline-int-divide
14145 Do not generate inline code for divides of integer values.
14147 @item -minline-sqrt-min-latency
14148 @opindex minline-sqrt-min-latency
14149 Generate code for inline square roots
14150 using the minimum latency algorithm.
14152 @item -minline-sqrt-max-throughput
14153 @opindex minline-sqrt-max-throughput
14154 Generate code for inline square roots
14155 using the maximum throughput algorithm.
14157 @item -mno-inline-sqrt
14158 @opindex mno-inline-sqrt
14159 Do not generate inline code for sqrt.
14162 @itemx -mno-fused-madd
14163 @opindex mfused-madd
14164 @opindex mno-fused-madd
14165 Do (don't) generate code that uses the fused multiply/add or multiply/subtract
14166 instructions. The default is to use these instructions.
14168 @item -mno-dwarf2-asm
14169 @itemx -mdwarf2-asm
14170 @opindex mno-dwarf2-asm
14171 @opindex mdwarf2-asm
14172 Don't (or do) generate assembler code for the DWARF2 line number debugging
14173 info. This may be useful when not using the GNU assembler.
14175 @item -mearly-stop-bits
14176 @itemx -mno-early-stop-bits
14177 @opindex mearly-stop-bits
14178 @opindex mno-early-stop-bits
14179 Allow stop bits to be placed earlier than immediately preceding the
14180 instruction that triggered the stop bit. This can improve instruction
14181 scheduling, but does not always do so.
14183 @item -mfixed-range=@var{register-range}
14184 @opindex mfixed-range
14185 Generate code treating the given register range as fixed registers.
14186 A fixed register is one that the register allocator can not use. This is
14187 useful when compiling kernel code. A register range is specified as
14188 two registers separated by a dash. Multiple register ranges can be
14189 specified separated by a comma.
14191 @item -mtls-size=@var{tls-size}
14193 Specify bit size of immediate TLS offsets. Valid values are 14, 22, and
14196 @item -mtune=@var{cpu-type}
14198 Tune the instruction scheduling for a particular CPU, Valid values are
14199 itanium, itanium1, merced, itanium2, and mckinley.
14205 Generate code for a 32-bit or 64-bit environment.
14206 The 32-bit environment sets int, long and pointer to 32 bits.
14207 The 64-bit environment sets int to 32 bits and long and pointer
14208 to 64 bits. These are HP-UX specific flags.
14210 @item -mno-sched-br-data-spec
14211 @itemx -msched-br-data-spec
14212 @opindex mno-sched-br-data-spec
14213 @opindex msched-br-data-spec
14214 (Dis/En)able data speculative scheduling before reload.
14215 This will result in generation of the ld.a instructions and
14216 the corresponding check instructions (ld.c / chk.a).
14217 The default is 'disable'.
14219 @item -msched-ar-data-spec
14220 @itemx -mno-sched-ar-data-spec
14221 @opindex msched-ar-data-spec
14222 @opindex mno-sched-ar-data-spec
14223 (En/Dis)able data speculative scheduling after reload.
14224 This will result in generation of the ld.a instructions and
14225 the corresponding check instructions (ld.c / chk.a).
14226 The default is 'enable'.
14228 @item -mno-sched-control-spec
14229 @itemx -msched-control-spec
14230 @opindex mno-sched-control-spec
14231 @opindex msched-control-spec
14232 (Dis/En)able control speculative scheduling. This feature is
14233 available only during region scheduling (i.e.@: before reload).
14234 This will result in generation of the ld.s instructions and
14235 the corresponding check instructions chk.s .
14236 The default is 'disable'.
14238 @item -msched-br-in-data-spec
14239 @itemx -mno-sched-br-in-data-spec
14240 @opindex msched-br-in-data-spec
14241 @opindex mno-sched-br-in-data-spec
14242 (En/Dis)able speculative scheduling of the instructions that
14243 are dependent on the data speculative loads before reload.
14244 This is effective only with @option{-msched-br-data-spec} enabled.
14245 The default is 'enable'.
14247 @item -msched-ar-in-data-spec
14248 @itemx -mno-sched-ar-in-data-spec
14249 @opindex msched-ar-in-data-spec
14250 @opindex mno-sched-ar-in-data-spec
14251 (En/Dis)able speculative scheduling of the instructions that
14252 are dependent on the data speculative loads after reload.
14253 This is effective only with @option{-msched-ar-data-spec} enabled.
14254 The default is 'enable'.
14256 @item -msched-in-control-spec
14257 @itemx -mno-sched-in-control-spec
14258 @opindex msched-in-control-spec
14259 @opindex mno-sched-in-control-spec
14260 (En/Dis)able speculative scheduling of the instructions that
14261 are dependent on the control speculative loads.
14262 This is effective only with @option{-msched-control-spec} enabled.
14263 The default is 'enable'.
14265 @item -mno-sched-prefer-non-data-spec-insns
14266 @itemx -msched-prefer-non-data-spec-insns
14267 @opindex mno-sched-prefer-non-data-spec-insns
14268 @opindex msched-prefer-non-data-spec-insns
14269 If enabled, data speculative instructions will be chosen for schedule
14270 only if there are no other choices at the moment. This will make
14271 the use of the data speculation much more conservative.
14272 The default is 'disable'.
14274 @item -mno-sched-prefer-non-control-spec-insns
14275 @itemx -msched-prefer-non-control-spec-insns
14276 @opindex mno-sched-prefer-non-control-spec-insns
14277 @opindex msched-prefer-non-control-spec-insns
14278 If enabled, control speculative instructions will be chosen for schedule
14279 only if there are no other choices at the moment. This will make
14280 the use of the control speculation much more conservative.
14281 The default is 'disable'.
14283 @item -mno-sched-count-spec-in-critical-path
14284 @itemx -msched-count-spec-in-critical-path
14285 @opindex mno-sched-count-spec-in-critical-path
14286 @opindex msched-count-spec-in-critical-path
14287 If enabled, speculative dependencies will be considered during
14288 computation of the instructions priorities. This will make the use of the
14289 speculation a bit more conservative.
14290 The default is 'disable'.
14292 @item -msched-spec-ldc
14293 @opindex msched-spec-ldc
14294 Use a simple data speculation check. This option is on by default.
14296 @item -msched-control-spec-ldc
14297 @opindex msched-spec-ldc
14298 Use a simple check for control speculation. This option is on by default.
14300 @item -msched-stop-bits-after-every-cycle
14301 @opindex msched-stop-bits-after-every-cycle
14302 Place a stop bit after every cycle when scheduling. This option is on
14305 @item -msched-fp-mem-deps-zero-cost
14306 @opindex msched-fp-mem-deps-zero-cost
14307 Assume that floating-point stores and loads are not likely to cause a conflict
14308 when placed into the same instruction group. This option is disabled by
14311 @item -msel-sched-dont-check-control-spec
14312 @opindex msel-sched-dont-check-control-spec
14313 Generate checks for control speculation in selective scheduling.
14314 This flag is disabled by default.
14316 @item -msched-max-memory-insns=@var{max-insns}
14317 @opindex msched-max-memory-insns
14318 Limit on the number of memory insns per instruction group, giving lower
14319 priority to subsequent memory insns attempting to schedule in the same
14320 instruction group. Frequently useful to prevent cache bank conflicts.
14321 The default value is 1.
14323 @item -msched-max-memory-insns-hard-limit
14324 @opindex msched-max-memory-insns-hard-limit
14325 Disallow more than `msched-max-memory-insns' in instruction group.
14326 Otherwise, limit is `soft' meaning that we would prefer non-memory operations
14327 when limit is reached but may still schedule memory operations.
14331 @node IA-64/VMS Options
14332 @subsection IA-64/VMS Options
14334 These @samp{-m} options are defined for the IA-64/VMS implementations:
14337 @item -mvms-return-codes
14338 @opindex mvms-return-codes
14339 Return VMS condition codes from main. The default is to return POSIX
14340 style condition (e.g.@ error) codes.
14342 @item -mdebug-main=@var{prefix}
14343 @opindex mdebug-main=@var{prefix}
14344 Flag the first routine whose name starts with @var{prefix} as the main
14345 routine for the debugger.
14349 Default to 64-bit memory allocation routines.
14353 @subsection LM32 Options
14354 @cindex LM32 options
14356 These @option{-m} options are defined for the Lattice Mico32 architecture:
14359 @item -mbarrel-shift-enabled
14360 @opindex mbarrel-shift-enabled
14361 Enable barrel-shift instructions.
14363 @item -mdivide-enabled
14364 @opindex mdivide-enabled
14365 Enable divide and modulus instructions.
14367 @item -mmultiply-enabled
14368 @opindex multiply-enabled
14369 Enable multiply instructions.
14371 @item -msign-extend-enabled
14372 @opindex msign-extend-enabled
14373 Enable sign extend instructions.
14375 @item -muser-enabled
14376 @opindex muser-enabled
14377 Enable user-defined instructions.
14382 @subsection M32C Options
14383 @cindex M32C options
14386 @item -mcpu=@var{name}
14388 Select the CPU for which code is generated. @var{name} may be one of
14389 @samp{r8c} for the R8C/Tiny series, @samp{m16c} for the M16C (up to
14390 /60) series, @samp{m32cm} for the M16C/80 series, or @samp{m32c} for
14391 the M32C/80 series.
14395 Specifies that the program will be run on the simulator. This causes
14396 an alternate runtime library to be linked in which supports, for
14397 example, file I/O@. You must not use this option when generating
14398 programs that will run on real hardware; you must provide your own
14399 runtime library for whatever I/O functions are needed.
14401 @item -memregs=@var{number}
14403 Specifies the number of memory-based pseudo-registers GCC will use
14404 during code generation. These pseudo-registers will be used like real
14405 registers, so there is a tradeoff between GCC's ability to fit the
14406 code into available registers, and the performance penalty of using
14407 memory instead of registers. Note that all modules in a program must
14408 be compiled with the same value for this option. Because of that, you
14409 must not use this option with the default runtime libraries gcc
14414 @node M32R/D Options
14415 @subsection M32R/D Options
14416 @cindex M32R/D options
14418 These @option{-m} options are defined for Renesas M32R/D architectures:
14423 Generate code for the M32R/2@.
14427 Generate code for the M32R/X@.
14431 Generate code for the M32R@. This is the default.
14433 @item -mmodel=small
14434 @opindex mmodel=small
14435 Assume all objects live in the lower 16MB of memory (so that their addresses
14436 can be loaded with the @code{ld24} instruction), and assume all subroutines
14437 are reachable with the @code{bl} instruction.
14438 This is the default.
14440 The addressability of a particular object can be set with the
14441 @code{model} attribute.
14443 @item -mmodel=medium
14444 @opindex mmodel=medium
14445 Assume objects may be anywhere in the 32-bit address space (the compiler
14446 will generate @code{seth/add3} instructions to load their addresses), and
14447 assume all subroutines are reachable with the @code{bl} instruction.
14449 @item -mmodel=large
14450 @opindex mmodel=large
14451 Assume objects may be anywhere in the 32-bit address space (the compiler
14452 will generate @code{seth/add3} instructions to load their addresses), and
14453 assume subroutines may not be reachable with the @code{bl} instruction
14454 (the compiler will generate the much slower @code{seth/add3/jl}
14455 instruction sequence).
14458 @opindex msdata=none
14459 Disable use of the small data area. Variables will be put into
14460 one of @samp{.data}, @samp{bss}, or @samp{.rodata} (unless the
14461 @code{section} attribute has been specified).
14462 This is the default.
14464 The small data area consists of sections @samp{.sdata} and @samp{.sbss}.
14465 Objects may be explicitly put in the small data area with the
14466 @code{section} attribute using one of these sections.
14468 @item -msdata=sdata
14469 @opindex msdata=sdata
14470 Put small global and static data in the small data area, but do not
14471 generate special code to reference them.
14474 @opindex msdata=use
14475 Put small global and static data in the small data area, and generate
14476 special instructions to reference them.
14480 @cindex smaller data references
14481 Put global and static objects less than or equal to @var{num} bytes
14482 into the small data or bss sections instead of the normal data or bss
14483 sections. The default value of @var{num} is 8.
14484 The @option{-msdata} option must be set to one of @samp{sdata} or @samp{use}
14485 for this option to have any effect.
14487 All modules should be compiled with the same @option{-G @var{num}} value.
14488 Compiling with different values of @var{num} may or may not work; if it
14489 doesn't the linker will give an error message---incorrect code will not be
14494 Makes the M32R specific code in the compiler display some statistics
14495 that might help in debugging programs.
14497 @item -malign-loops
14498 @opindex malign-loops
14499 Align all loops to a 32-byte boundary.
14501 @item -mno-align-loops
14502 @opindex mno-align-loops
14503 Do not enforce a 32-byte alignment for loops. This is the default.
14505 @item -missue-rate=@var{number}
14506 @opindex missue-rate=@var{number}
14507 Issue @var{number} instructions per cycle. @var{number} can only be 1
14510 @item -mbranch-cost=@var{number}
14511 @opindex mbranch-cost=@var{number}
14512 @var{number} can only be 1 or 2. If it is 1 then branches will be
14513 preferred over conditional code, if it is 2, then the opposite will
14516 @item -mflush-trap=@var{number}
14517 @opindex mflush-trap=@var{number}
14518 Specifies the trap number to use to flush the cache. The default is
14519 12. Valid numbers are between 0 and 15 inclusive.
14521 @item -mno-flush-trap
14522 @opindex mno-flush-trap
14523 Specifies that the cache cannot be flushed by using a trap.
14525 @item -mflush-func=@var{name}
14526 @opindex mflush-func=@var{name}
14527 Specifies the name of the operating system function to call to flush
14528 the cache. The default is @emph{_flush_cache}, but a function call
14529 will only be used if a trap is not available.
14531 @item -mno-flush-func
14532 @opindex mno-flush-func
14533 Indicates that there is no OS function for flushing the cache.
14537 @node M680x0 Options
14538 @subsection M680x0 Options
14539 @cindex M680x0 options
14541 These are the @samp{-m} options defined for M680x0 and ColdFire processors.
14542 The default settings depend on which architecture was selected when
14543 the compiler was configured; the defaults for the most common choices
14547 @item -march=@var{arch}
14549 Generate code for a specific M680x0 or ColdFire instruction set
14550 architecture. Permissible values of @var{arch} for M680x0
14551 architectures are: @samp{68000}, @samp{68010}, @samp{68020},
14552 @samp{68030}, @samp{68040}, @samp{68060} and @samp{cpu32}. ColdFire
14553 architectures are selected according to Freescale's ISA classification
14554 and the permissible values are: @samp{isaa}, @samp{isaaplus},
14555 @samp{isab} and @samp{isac}.
14557 gcc defines a macro @samp{__mcf@var{arch}__} whenever it is generating
14558 code for a ColdFire target. The @var{arch} in this macro is one of the
14559 @option{-march} arguments given above.
14561 When used together, @option{-march} and @option{-mtune} select code
14562 that runs on a family of similar processors but that is optimized
14563 for a particular microarchitecture.
14565 @item -mcpu=@var{cpu}
14567 Generate code for a specific M680x0 or ColdFire processor.
14568 The M680x0 @var{cpu}s are: @samp{68000}, @samp{68010}, @samp{68020},
14569 @samp{68030}, @samp{68040}, @samp{68060}, @samp{68302}, @samp{68332}
14570 and @samp{cpu32}. The ColdFire @var{cpu}s are given by the table
14571 below, which also classifies the CPUs into families:
14573 @multitable @columnfractions 0.20 0.80
14574 @item @strong{Family} @tab @strong{@samp{-mcpu} arguments}
14575 @item @samp{51} @tab @samp{51} @samp{51ac} @samp{51cn} @samp{51em} @samp{51qe}
14576 @item @samp{5206} @tab @samp{5202} @samp{5204} @samp{5206}
14577 @item @samp{5206e} @tab @samp{5206e}
14578 @item @samp{5208} @tab @samp{5207} @samp{5208}
14579 @item @samp{5211a} @tab @samp{5210a} @samp{5211a}
14580 @item @samp{5213} @tab @samp{5211} @samp{5212} @samp{5213}
14581 @item @samp{5216} @tab @samp{5214} @samp{5216}
14582 @item @samp{52235} @tab @samp{52230} @samp{52231} @samp{52232} @samp{52233} @samp{52234} @samp{52235}
14583 @item @samp{5225} @tab @samp{5224} @samp{5225}
14584 @item @samp{52259} @tab @samp{52252} @samp{52254} @samp{52255} @samp{52256} @samp{52258} @samp{52259}
14585 @item @samp{5235} @tab @samp{5232} @samp{5233} @samp{5234} @samp{5235} @samp{523x}
14586 @item @samp{5249} @tab @samp{5249}
14587 @item @samp{5250} @tab @samp{5250}
14588 @item @samp{5271} @tab @samp{5270} @samp{5271}
14589 @item @samp{5272} @tab @samp{5272}
14590 @item @samp{5275} @tab @samp{5274} @samp{5275}
14591 @item @samp{5282} @tab @samp{5280} @samp{5281} @samp{5282} @samp{528x}
14592 @item @samp{53017} @tab @samp{53011} @samp{53012} @samp{53013} @samp{53014} @samp{53015} @samp{53016} @samp{53017}
14593 @item @samp{5307} @tab @samp{5307}
14594 @item @samp{5329} @tab @samp{5327} @samp{5328} @samp{5329} @samp{532x}
14595 @item @samp{5373} @tab @samp{5372} @samp{5373} @samp{537x}
14596 @item @samp{5407} @tab @samp{5407}
14597 @item @samp{5475} @tab @samp{5470} @samp{5471} @samp{5472} @samp{5473} @samp{5474} @samp{5475} @samp{547x} @samp{5480} @samp{5481} @samp{5482} @samp{5483} @samp{5484} @samp{5485}
14600 @option{-mcpu=@var{cpu}} overrides @option{-march=@var{arch}} if
14601 @var{arch} is compatible with @var{cpu}. Other combinations of
14602 @option{-mcpu} and @option{-march} are rejected.
14604 gcc defines the macro @samp{__mcf_cpu_@var{cpu}} when ColdFire target
14605 @var{cpu} is selected. It also defines @samp{__mcf_family_@var{family}},
14606 where the value of @var{family} is given by the table above.
14608 @item -mtune=@var{tune}
14610 Tune the code for a particular microarchitecture, within the
14611 constraints set by @option{-march} and @option{-mcpu}.
14612 The M680x0 microarchitectures are: @samp{68000}, @samp{68010},
14613 @samp{68020}, @samp{68030}, @samp{68040}, @samp{68060}
14614 and @samp{cpu32}. The ColdFire microarchitectures
14615 are: @samp{cfv1}, @samp{cfv2}, @samp{cfv3}, @samp{cfv4} and @samp{cfv4e}.
14617 You can also use @option{-mtune=68020-40} for code that needs
14618 to run relatively well on 68020, 68030 and 68040 targets.
14619 @option{-mtune=68020-60} is similar but includes 68060 targets
14620 as well. These two options select the same tuning decisions as
14621 @option{-m68020-40} and @option{-m68020-60} respectively.
14623 gcc defines the macros @samp{__mc@var{arch}} and @samp{__mc@var{arch}__}
14624 when tuning for 680x0 architecture @var{arch}. It also defines
14625 @samp{mc@var{arch}} unless either @option{-ansi} or a non-GNU @option{-std}
14626 option is used. If gcc is tuning for a range of architectures,
14627 as selected by @option{-mtune=68020-40} or @option{-mtune=68020-60},
14628 it defines the macros for every architecture in the range.
14630 gcc also defines the macro @samp{__m@var{uarch}__} when tuning for
14631 ColdFire microarchitecture @var{uarch}, where @var{uarch} is one
14632 of the arguments given above.
14638 Generate output for a 68000. This is the default
14639 when the compiler is configured for 68000-based systems.
14640 It is equivalent to @option{-march=68000}.
14642 Use this option for microcontrollers with a 68000 or EC000 core,
14643 including the 68008, 68302, 68306, 68307, 68322, 68328 and 68356.
14647 Generate output for a 68010. This is the default
14648 when the compiler is configured for 68010-based systems.
14649 It is equivalent to @option{-march=68010}.
14655 Generate output for a 68020. This is the default
14656 when the compiler is configured for 68020-based systems.
14657 It is equivalent to @option{-march=68020}.
14661 Generate output for a 68030. This is the default when the compiler is
14662 configured for 68030-based systems. It is equivalent to
14663 @option{-march=68030}.
14667 Generate output for a 68040. This is the default when the compiler is
14668 configured for 68040-based systems. It is equivalent to
14669 @option{-march=68040}.
14671 This option inhibits the use of 68881/68882 instructions that have to be
14672 emulated by software on the 68040. Use this option if your 68040 does not
14673 have code to emulate those instructions.
14677 Generate output for a 68060. This is the default when the compiler is
14678 configured for 68060-based systems. It is equivalent to
14679 @option{-march=68060}.
14681 This option inhibits the use of 68020 and 68881/68882 instructions that
14682 have to be emulated by software on the 68060. Use this option if your 68060
14683 does not have code to emulate those instructions.
14687 Generate output for a CPU32. This is the default
14688 when the compiler is configured for CPU32-based systems.
14689 It is equivalent to @option{-march=cpu32}.
14691 Use this option for microcontrollers with a
14692 CPU32 or CPU32+ core, including the 68330, 68331, 68332, 68333, 68334,
14693 68336, 68340, 68341, 68349 and 68360.
14697 Generate output for a 520X ColdFire CPU@. This is the default
14698 when the compiler is configured for 520X-based systems.
14699 It is equivalent to @option{-mcpu=5206}, and is now deprecated
14700 in favor of that option.
14702 Use this option for microcontroller with a 5200 core, including
14703 the MCF5202, MCF5203, MCF5204 and MCF5206.
14707 Generate output for a 5206e ColdFire CPU@. The option is now
14708 deprecated in favor of the equivalent @option{-mcpu=5206e}.
14712 Generate output for a member of the ColdFire 528X family.
14713 The option is now deprecated in favor of the equivalent
14714 @option{-mcpu=528x}.
14718 Generate output for a ColdFire 5307 CPU@. The option is now deprecated
14719 in favor of the equivalent @option{-mcpu=5307}.
14723 Generate output for a ColdFire 5407 CPU@. The option is now deprecated
14724 in favor of the equivalent @option{-mcpu=5407}.
14728 Generate output for a ColdFire V4e family CPU (e.g.@: 547x/548x).
14729 This includes use of hardware floating-point instructions.
14730 The option is equivalent to @option{-mcpu=547x}, and is now
14731 deprecated in favor of that option.
14735 Generate output for a 68040, without using any of the new instructions.
14736 This results in code that can run relatively efficiently on either a
14737 68020/68881 or a 68030 or a 68040. The generated code does use the
14738 68881 instructions that are emulated on the 68040.
14740 The option is equivalent to @option{-march=68020} @option{-mtune=68020-40}.
14744 Generate output for a 68060, without using any of the new instructions.
14745 This results in code that can run relatively efficiently on either a
14746 68020/68881 or a 68030 or a 68040. The generated code does use the
14747 68881 instructions that are emulated on the 68060.
14749 The option is equivalent to @option{-march=68020} @option{-mtune=68020-60}.
14753 @opindex mhard-float
14755 Generate floating-point instructions. This is the default for 68020
14756 and above, and for ColdFire devices that have an FPU@. It defines the
14757 macro @samp{__HAVE_68881__} on M680x0 targets and @samp{__mcffpu__}
14758 on ColdFire targets.
14761 @opindex msoft-float
14762 Do not generate floating-point instructions; use library calls instead.
14763 This is the default for 68000, 68010, and 68832 targets. It is also
14764 the default for ColdFire devices that have no FPU.
14770 Generate (do not generate) ColdFire hardware divide and remainder
14771 instructions. If @option{-march} is used without @option{-mcpu},
14772 the default is ``on'' for ColdFire architectures and ``off'' for M680x0
14773 architectures. Otherwise, the default is taken from the target CPU
14774 (either the default CPU, or the one specified by @option{-mcpu}). For
14775 example, the default is ``off'' for @option{-mcpu=5206} and ``on'' for
14776 @option{-mcpu=5206e}.
14778 gcc defines the macro @samp{__mcfhwdiv__} when this option is enabled.
14782 Consider type @code{int} to be 16 bits wide, like @code{short int}.
14783 Additionally, parameters passed on the stack are also aligned to a
14784 16-bit boundary even on targets whose API mandates promotion to 32-bit.
14788 Do not consider type @code{int} to be 16 bits wide. This is the default.
14791 @itemx -mno-bitfield
14792 @opindex mnobitfield
14793 @opindex mno-bitfield
14794 Do not use the bit-field instructions. The @option{-m68000}, @option{-mcpu32}
14795 and @option{-m5200} options imply @w{@option{-mnobitfield}}.
14799 Do use the bit-field instructions. The @option{-m68020} option implies
14800 @option{-mbitfield}. This is the default if you use a configuration
14801 designed for a 68020.
14805 Use a different function-calling convention, in which functions
14806 that take a fixed number of arguments return with the @code{rtd}
14807 instruction, which pops their arguments while returning. This
14808 saves one instruction in the caller since there is no need to pop
14809 the arguments there.
14811 This calling convention is incompatible with the one normally
14812 used on Unix, so you cannot use it if you need to call libraries
14813 compiled with the Unix compiler.
14815 Also, you must provide function prototypes for all functions that
14816 take variable numbers of arguments (including @code{printf});
14817 otherwise incorrect code will be generated for calls to those
14820 In addition, seriously incorrect code will result if you call a
14821 function with too many arguments. (Normally, extra arguments are
14822 harmlessly ignored.)
14824 The @code{rtd} instruction is supported by the 68010, 68020, 68030,
14825 68040, 68060 and CPU32 processors, but not by the 68000 or 5200.
14829 Do not use the calling conventions selected by @option{-mrtd}.
14830 This is the default.
14833 @itemx -mno-align-int
14834 @opindex malign-int
14835 @opindex mno-align-int
14836 Control whether GCC aligns @code{int}, @code{long}, @code{long long},
14837 @code{float}, @code{double}, and @code{long double} variables on a 32-bit
14838 boundary (@option{-malign-int}) or a 16-bit boundary (@option{-mno-align-int}).
14839 Aligning variables on 32-bit boundaries produces code that runs somewhat
14840 faster on processors with 32-bit busses at the expense of more memory.
14842 @strong{Warning:} if you use the @option{-malign-int} switch, GCC will
14843 align structures containing the above types differently than
14844 most published application binary interface specifications for the m68k.
14848 Use the pc-relative addressing mode of the 68000 directly, instead of
14849 using a global offset table. At present, this option implies @option{-fpic},
14850 allowing at most a 16-bit offset for pc-relative addressing. @option{-fPIC} is
14851 not presently supported with @option{-mpcrel}, though this could be supported for
14852 68020 and higher processors.
14854 @item -mno-strict-align
14855 @itemx -mstrict-align
14856 @opindex mno-strict-align
14857 @opindex mstrict-align
14858 Do not (do) assume that unaligned memory references will be handled by
14862 Generate code that allows the data segment to be located in a different
14863 area of memory from the text segment. This allows for execute in place in
14864 an environment without virtual memory management. This option implies
14867 @item -mno-sep-data
14868 Generate code that assumes that the data segment follows the text segment.
14869 This is the default.
14871 @item -mid-shared-library
14872 Generate code that supports shared libraries via the library ID method.
14873 This allows for execute in place and shared libraries in an environment
14874 without virtual memory management. This option implies @option{-fPIC}.
14876 @item -mno-id-shared-library
14877 Generate code that doesn't assume ID based shared libraries are being used.
14878 This is the default.
14880 @item -mshared-library-id=n
14881 Specified the identification number of the ID based shared library being
14882 compiled. Specifying a value of 0 will generate more compact code, specifying
14883 other values will force the allocation of that number to the current
14884 library but is no more space or time efficient than omitting this option.
14890 When generating position-independent code for ColdFire, generate code
14891 that works if the GOT has more than 8192 entries. This code is
14892 larger and slower than code generated without this option. On M680x0
14893 processors, this option is not needed; @option{-fPIC} suffices.
14895 GCC normally uses a single instruction to load values from the GOT@.
14896 While this is relatively efficient, it only works if the GOT
14897 is smaller than about 64k. Anything larger causes the linker
14898 to report an error such as:
14900 @cindex relocation truncated to fit (ColdFire)
14902 relocation truncated to fit: R_68K_GOT16O foobar
14905 If this happens, you should recompile your code with @option{-mxgot}.
14906 It should then work with very large GOTs. However, code generated with
14907 @option{-mxgot} is less efficient, since it takes 4 instructions to fetch
14908 the value of a global symbol.
14910 Note that some linkers, including newer versions of the GNU linker,
14911 can create multiple GOTs and sort GOT entries. If you have such a linker,
14912 you should only need to use @option{-mxgot} when compiling a single
14913 object file that accesses more than 8192 GOT entries. Very few do.
14915 These options have no effect unless GCC is generating
14916 position-independent code.
14920 @node MCore Options
14921 @subsection MCore Options
14922 @cindex MCore options
14924 These are the @samp{-m} options defined for the Motorola M*Core
14930 @itemx -mno-hardlit
14932 @opindex mno-hardlit
14933 Inline constants into the code stream if it can be done in two
14934 instructions or less.
14940 Use the divide instruction. (Enabled by default).
14942 @item -mrelax-immediate
14943 @itemx -mno-relax-immediate
14944 @opindex mrelax-immediate
14945 @opindex mno-relax-immediate
14946 Allow arbitrary sized immediates in bit operations.
14948 @item -mwide-bitfields
14949 @itemx -mno-wide-bitfields
14950 @opindex mwide-bitfields
14951 @opindex mno-wide-bitfields
14952 Always treat bit-fields as int-sized.
14954 @item -m4byte-functions
14955 @itemx -mno-4byte-functions
14956 @opindex m4byte-functions
14957 @opindex mno-4byte-functions
14958 Force all functions to be aligned to a 4-byte boundary.
14960 @item -mcallgraph-data
14961 @itemx -mno-callgraph-data
14962 @opindex mcallgraph-data
14963 @opindex mno-callgraph-data
14964 Emit callgraph information.
14967 @itemx -mno-slow-bytes
14968 @opindex mslow-bytes
14969 @opindex mno-slow-bytes
14970 Prefer word access when reading byte quantities.
14972 @item -mlittle-endian
14973 @itemx -mbig-endian
14974 @opindex mlittle-endian
14975 @opindex mbig-endian
14976 Generate code for a little-endian target.
14982 Generate code for the 210 processor.
14986 Assume that runtime support has been provided and so omit the
14987 simulator library (@file{libsim.a)} from the linker command line.
14989 @item -mstack-increment=@var{size}
14990 @opindex mstack-increment
14991 Set the maximum amount for a single stack increment operation. Large
14992 values can increase the speed of programs that contain functions
14993 that need a large amount of stack space, but they can also trigger a
14994 segmentation fault if the stack is extended too much. The default
15000 @subsection MeP Options
15001 @cindex MeP options
15007 Enables the @code{abs} instruction, which is the absolute difference
15008 between two registers.
15012 Enables all the optional instructions - average, multiply, divide, bit
15013 operations, leading zero, absolute difference, min/max, clip, and
15019 Enables the @code{ave} instruction, which computes the average of two
15022 @item -mbased=@var{n}
15024 Variables of size @var{n} bytes or smaller will be placed in the
15025 @code{.based} section by default. Based variables use the @code{$tp}
15026 register as a base register, and there is a 128-byte limit to the
15027 @code{.based} section.
15031 Enables the bit operation instructions - bit test (@code{btstm}), set
15032 (@code{bsetm}), clear (@code{bclrm}), invert (@code{bnotm}), and
15033 test-and-set (@code{tas}).
15035 @item -mc=@var{name}
15037 Selects which section constant data will be placed in. @var{name} may
15038 be @code{tiny}, @code{near}, or @code{far}.
15042 Enables the @code{clip} instruction. Note that @code{-mclip} is not
15043 useful unless you also provide @code{-mminmax}.
15045 @item -mconfig=@var{name}
15047 Selects one of the build-in core configurations. Each MeP chip has
15048 one or more modules in it; each module has a core CPU and a variety of
15049 coprocessors, optional instructions, and peripherals. The
15050 @code{MeP-Integrator} tool, not part of GCC, provides these
15051 configurations through this option; using this option is the same as
15052 using all the corresponding command-line options. The default
15053 configuration is @code{default}.
15057 Enables the coprocessor instructions. By default, this is a 32-bit
15058 coprocessor. Note that the coprocessor is normally enabled via the
15059 @code{-mconfig=} option.
15063 Enables the 32-bit coprocessor's instructions.
15067 Enables the 64-bit coprocessor's instructions.
15071 Enables IVC2 scheduling. IVC2 is a 64-bit VLIW coprocessor.
15075 Causes constant variables to be placed in the @code{.near} section.
15079 Enables the @code{div} and @code{divu} instructions.
15083 Generate big-endian code.
15087 Generate little-endian code.
15089 @item -mio-volatile
15090 @opindex mio-volatile
15091 Tells the compiler that any variable marked with the @code{io}
15092 attribute is to be considered volatile.
15096 Causes variables to be assigned to the @code{.far} section by default.
15100 Enables the @code{leadz} (leading zero) instruction.
15104 Causes variables to be assigned to the @code{.near} section by default.
15108 Enables the @code{min} and @code{max} instructions.
15112 Enables the multiplication and multiply-accumulate instructions.
15116 Disables all the optional instructions enabled by @code{-mall-opts}.
15120 Enables the @code{repeat} and @code{erepeat} instructions, used for
15121 low-overhead looping.
15125 Causes all variables to default to the @code{.tiny} section. Note
15126 that there is a 65536-byte limit to this section. Accesses to these
15127 variables use the @code{%gp} base register.
15131 Enables the saturation instructions. Note that the compiler does not
15132 currently generate these itself, but this option is included for
15133 compatibility with other tools, like @code{as}.
15137 Link the SDRAM-based runtime instead of the default ROM-based runtime.
15141 Link the simulator runtime libraries.
15145 Link the simulator runtime libraries, excluding built-in support
15146 for reset and exception vectors and tables.
15150 Causes all functions to default to the @code{.far} section. Without
15151 this option, functions default to the @code{.near} section.
15153 @item -mtiny=@var{n}
15155 Variables that are @var{n} bytes or smaller will be allocated to the
15156 @code{.tiny} section. These variables use the @code{$gp} base
15157 register. The default for this option is 4, but note that there's a
15158 65536-byte limit to the @code{.tiny} section.
15162 @node MicroBlaze Options
15163 @subsection MicroBlaze Options
15164 @cindex MicroBlaze Options
15169 @opindex msoft-float
15170 Use software emulation for floating point (default).
15173 @opindex mhard-float
15174 Use hardware floating-point instructions.
15178 Do not optimize block moves, use @code{memcpy}.
15180 @item -mno-clearbss
15181 @opindex mno-clearbss
15182 This option is deprecated. Use @option{-fno-zero-initialized-in-bss} instead.
15184 @item -mcpu=@var{cpu-type}
15186 Use features of and schedule code for given CPU.
15187 Supported values are in the format @samp{v@var{X}.@var{YY}.@var{Z}},
15188 where @var{X} is a major version, @var{YY} is the minor version, and
15189 @var{Z} is compatibility code. Example values are @samp{v3.00.a},
15190 @samp{v4.00.b}, @samp{v5.00.a}, @samp{v5.00.b}, @samp{v5.00.b}, @samp{v6.00.a}.
15192 @item -mxl-soft-mul
15193 @opindex mxl-soft-mul
15194 Use software multiply emulation (default).
15196 @item -mxl-soft-div
15197 @opindex mxl-soft-div
15198 Use software emulation for divides (default).
15200 @item -mxl-barrel-shift
15201 @opindex mxl-barrel-shift
15202 Use the hardware barrel shifter.
15204 @item -mxl-pattern-compare
15205 @opindex mxl-pattern-compare
15206 Use pattern compare instructions.
15208 @item -msmall-divides
15209 @opindex msmall-divides
15210 Use table lookup optimization for small signed integer divisions.
15212 @item -mxl-stack-check
15213 @opindex mxl-stack-check
15214 This option is deprecated. Use -fstack-check instead.
15217 @opindex mxl-gp-opt
15218 Use GP relative sdata/sbss sections.
15220 @item -mxl-multiply-high
15221 @opindex mxl-multiply-high
15222 Use multiply high instructions for high part of 32x32 multiply.
15224 @item -mxl-float-convert
15225 @opindex mxl-float-convert
15226 Use hardware floating-point conversion instructions.
15228 @item -mxl-float-sqrt
15229 @opindex mxl-float-sqrt
15230 Use hardware floating-point square root instruction.
15232 @item -mxl-mode-@var{app-model}
15233 Select application model @var{app-model}. Valid models are
15236 normal executable (default), uses startup code @file{crt0.o}.
15239 for use with Xilinx Microprocessor Debugger (XMD) based
15240 software intrusive debug agent called xmdstub. This uses startup file
15241 @file{crt1.o} and sets the start address of the program to be 0x800.
15244 for applications that are loaded using a bootloader.
15245 This model uses startup file @file{crt2.o} which does not contain a processor
15246 reset vector handler. This is suitable for transferring control on a
15247 processor reset to the bootloader rather than the application.
15250 for applications that do not require any of the
15251 MicroBlaze vectors. This option may be useful for applications running
15252 within a monitoring application. This model uses @file{crt3.o} as a startup file.
15255 Option @option{-xl-mode-@var{app-model}} is a deprecated alias for
15256 @option{-mxl-mode-@var{app-model}}.
15261 @subsection MIPS Options
15262 @cindex MIPS options
15268 Generate big-endian code.
15272 Generate little-endian code. This is the default for @samp{mips*el-*-*}
15275 @item -march=@var{arch}
15277 Generate code that will run on @var{arch}, which can be the name of a
15278 generic MIPS ISA, or the name of a particular processor.
15280 @samp{mips1}, @samp{mips2}, @samp{mips3}, @samp{mips4},
15281 @samp{mips32}, @samp{mips32r2}, @samp{mips64} and @samp{mips64r2}.
15282 The processor names are:
15283 @samp{4kc}, @samp{4km}, @samp{4kp}, @samp{4ksc},
15284 @samp{4kec}, @samp{4kem}, @samp{4kep}, @samp{4ksd},
15285 @samp{5kc}, @samp{5kf},
15287 @samp{24kc}, @samp{24kf2_1}, @samp{24kf1_1},
15288 @samp{24kec}, @samp{24kef2_1}, @samp{24kef1_1},
15289 @samp{34kc}, @samp{34kf2_1}, @samp{34kf1_1},
15290 @samp{74kc}, @samp{74kf2_1}, @samp{74kf1_1}, @samp{74kf3_2},
15291 @samp{1004kc}, @samp{1004kf2_1}, @samp{1004kf1_1},
15292 @samp{loongson2e}, @samp{loongson2f}, @samp{loongson3a},
15294 @samp{octeon}, @samp{octeon+}, @samp{octeon2},
15296 @samp{r2000}, @samp{r3000}, @samp{r3900}, @samp{r4000}, @samp{r4400},
15297 @samp{r4600}, @samp{r4650}, @samp{r6000}, @samp{r8000},
15298 @samp{rm7000}, @samp{rm9000},
15299 @samp{r10000}, @samp{r12000}, @samp{r14000}, @samp{r16000},
15302 @samp{vr4100}, @samp{vr4111}, @samp{vr4120}, @samp{vr4130}, @samp{vr4300},
15303 @samp{vr5000}, @samp{vr5400}, @samp{vr5500}
15305 The special value @samp{from-abi} selects the
15306 most compatible architecture for the selected ABI (that is,
15307 @samp{mips1} for 32-bit ABIs and @samp{mips3} for 64-bit ABIs)@.
15309 Native Linux/GNU and IRIX toolchains also support the value @samp{native},
15310 which selects the best architecture option for the host processor.
15311 @option{-march=native} has no effect if GCC does not recognize
15314 In processor names, a final @samp{000} can be abbreviated as @samp{k}
15315 (for example, @samp{-march=r2k}). Prefixes are optional, and
15316 @samp{vr} may be written @samp{r}.
15318 Names of the form @samp{@var{n}f2_1} refer to processors with
15319 FPUs clocked at half the rate of the core, names of the form
15320 @samp{@var{n}f1_1} refer to processors with FPUs clocked at the same
15321 rate as the core, and names of the form @samp{@var{n}f3_2} refer to
15322 processors with FPUs clocked a ratio of 3:2 with respect to the core.
15323 For compatibility reasons, @samp{@var{n}f} is accepted as a synonym
15324 for @samp{@var{n}f2_1} while @samp{@var{n}x} and @samp{@var{b}fx} are
15325 accepted as synonyms for @samp{@var{n}f1_1}.
15327 GCC defines two macros based on the value of this option. The first
15328 is @samp{_MIPS_ARCH}, which gives the name of target architecture, as
15329 a string. The second has the form @samp{_MIPS_ARCH_@var{foo}},
15330 where @var{foo} is the capitalized value of @samp{_MIPS_ARCH}@.
15331 For example, @samp{-march=r2000} will set @samp{_MIPS_ARCH}
15332 to @samp{"r2000"} and define the macro @samp{_MIPS_ARCH_R2000}.
15334 Note that the @samp{_MIPS_ARCH} macro uses the processor names given
15335 above. In other words, it will have the full prefix and will not
15336 abbreviate @samp{000} as @samp{k}. In the case of @samp{from-abi},
15337 the macro names the resolved architecture (either @samp{"mips1"} or
15338 @samp{"mips3"}). It names the default architecture when no
15339 @option{-march} option is given.
15341 @item -mtune=@var{arch}
15343 Optimize for @var{arch}. Among other things, this option controls
15344 the way instructions are scheduled, and the perceived cost of arithmetic
15345 operations. The list of @var{arch} values is the same as for
15348 When this option is not used, GCC will optimize for the processor
15349 specified by @option{-march}. By using @option{-march} and
15350 @option{-mtune} together, it is possible to generate code that will
15351 run on a family of processors, but optimize the code for one
15352 particular member of that family.
15354 @samp{-mtune} defines the macros @samp{_MIPS_TUNE} and
15355 @samp{_MIPS_TUNE_@var{foo}}, which work in the same way as the
15356 @samp{-march} ones described above.
15360 Equivalent to @samp{-march=mips1}.
15364 Equivalent to @samp{-march=mips2}.
15368 Equivalent to @samp{-march=mips3}.
15372 Equivalent to @samp{-march=mips4}.
15376 Equivalent to @samp{-march=mips32}.
15380 Equivalent to @samp{-march=mips32r2}.
15384 Equivalent to @samp{-march=mips64}.
15388 Equivalent to @samp{-march=mips64r2}.
15393 @opindex mno-mips16
15394 Generate (do not generate) MIPS16 code. If GCC is targetting a
15395 MIPS32 or MIPS64 architecture, it will make use of the MIPS16e ASE@.
15397 MIPS16 code generation can also be controlled on a per-function basis
15398 by means of @code{mips16} and @code{nomips16} attributes.
15399 @xref{Function Attributes}, for more information.
15401 @item -mflip-mips16
15402 @opindex mflip-mips16
15403 Generate MIPS16 code on alternating functions. This option is provided
15404 for regression testing of mixed MIPS16/non-MIPS16 code generation, and is
15405 not intended for ordinary use in compiling user code.
15407 @item -minterlink-mips16
15408 @itemx -mno-interlink-mips16
15409 @opindex minterlink-mips16
15410 @opindex mno-interlink-mips16
15411 Require (do not require) that non-MIPS16 code be link-compatible with
15414 For example, non-MIPS16 code cannot jump directly to MIPS16 code;
15415 it must either use a call or an indirect jump. @option{-minterlink-mips16}
15416 therefore disables direct jumps unless GCC knows that the target of the
15417 jump is not MIPS16.
15429 Generate code for the given ABI@.
15431 Note that the EABI has a 32-bit and a 64-bit variant. GCC normally
15432 generates 64-bit code when you select a 64-bit architecture, but you
15433 can use @option{-mgp32} to get 32-bit code instead.
15435 For information about the O64 ABI, see
15436 @uref{http://gcc.gnu.org/@/projects/@/mipso64-abi.html}.
15438 GCC supports a variant of the o32 ABI in which floating-point registers
15439 are 64 rather than 32 bits wide. You can select this combination with
15440 @option{-mabi=32} @option{-mfp64}. This ABI relies on the @samp{mthc1}
15441 and @samp{mfhc1} instructions and is therefore only supported for
15442 MIPS32R2 processors.
15444 The register assignments for arguments and return values remain the
15445 same, but each scalar value is passed in a single 64-bit register
15446 rather than a pair of 32-bit registers. For example, scalar
15447 floating-point values are returned in @samp{$f0} only, not a
15448 @samp{$f0}/@samp{$f1} pair. The set of call-saved registers also
15449 remains the same, but all 64 bits are saved.
15452 @itemx -mno-abicalls
15454 @opindex mno-abicalls
15455 Generate (do not generate) code that is suitable for SVR4-style
15456 dynamic objects. @option{-mabicalls} is the default for SVR4-based
15461 Generate (do not generate) code that is fully position-independent,
15462 and that can therefore be linked into shared libraries. This option
15463 only affects @option{-mabicalls}.
15465 All @option{-mabicalls} code has traditionally been position-independent,
15466 regardless of options like @option{-fPIC} and @option{-fpic}. However,
15467 as an extension, the GNU toolchain allows executables to use absolute
15468 accesses for locally-binding symbols. It can also use shorter GP
15469 initialization sequences and generate direct calls to locally-defined
15470 functions. This mode is selected by @option{-mno-shared}.
15472 @option{-mno-shared} depends on binutils 2.16 or higher and generates
15473 objects that can only be linked by the GNU linker. However, the option
15474 does not affect the ABI of the final executable; it only affects the ABI
15475 of relocatable objects. Using @option{-mno-shared} will generally make
15476 executables both smaller and quicker.
15478 @option{-mshared} is the default.
15484 Assume (do not assume) that the static and dynamic linkers
15485 support PLTs and copy relocations. This option only affects
15486 @samp{-mno-shared -mabicalls}. For the n64 ABI, this option
15487 has no effect without @samp{-msym32}.
15489 You can make @option{-mplt} the default by configuring
15490 GCC with @option{--with-mips-plt}. The default is
15491 @option{-mno-plt} otherwise.
15497 Lift (do not lift) the usual restrictions on the size of the global
15500 GCC normally uses a single instruction to load values from the GOT@.
15501 While this is relatively efficient, it will only work if the GOT
15502 is smaller than about 64k. Anything larger will cause the linker
15503 to report an error such as:
15505 @cindex relocation truncated to fit (MIPS)
15507 relocation truncated to fit: R_MIPS_GOT16 foobar
15510 If this happens, you should recompile your code with @option{-mxgot}.
15511 It should then work with very large GOTs, although it will also be
15512 less efficient, since it will take three instructions to fetch the
15513 value of a global symbol.
15515 Note that some linkers can create multiple GOTs. If you have such a
15516 linker, you should only need to use @option{-mxgot} when a single object
15517 file accesses more than 64k's worth of GOT entries. Very few do.
15519 These options have no effect unless GCC is generating position
15524 Assume that general-purpose registers are 32 bits wide.
15528 Assume that general-purpose registers are 64 bits wide.
15532 Assume that floating-point registers are 32 bits wide.
15536 Assume that floating-point registers are 64 bits wide.
15539 @opindex mhard-float
15540 Use floating-point coprocessor instructions.
15543 @opindex msoft-float
15544 Do not use floating-point coprocessor instructions. Implement
15545 floating-point calculations using library calls instead.
15547 @item -msingle-float
15548 @opindex msingle-float
15549 Assume that the floating-point coprocessor only supports single-precision
15552 @item -mdouble-float
15553 @opindex mdouble-float
15554 Assume that the floating-point coprocessor supports double-precision
15555 operations. This is the default.
15561 Use (do not use) @samp{ll}, @samp{sc}, and @samp{sync} instructions to
15562 implement atomic memory built-in functions. When neither option is
15563 specified, GCC will use the instructions if the target architecture
15566 @option{-mllsc} is useful if the runtime environment can emulate the
15567 instructions and @option{-mno-llsc} can be useful when compiling for
15568 nonstandard ISAs. You can make either option the default by
15569 configuring GCC with @option{--with-llsc} and @option{--without-llsc}
15570 respectively. @option{--with-llsc} is the default for some
15571 configurations; see the installation documentation for details.
15577 Use (do not use) revision 1 of the MIPS DSP ASE@.
15578 @xref{MIPS DSP Built-in Functions}. This option defines the
15579 preprocessor macro @samp{__mips_dsp}. It also defines
15580 @samp{__mips_dsp_rev} to 1.
15586 Use (do not use) revision 2 of the MIPS DSP ASE@.
15587 @xref{MIPS DSP Built-in Functions}. This option defines the
15588 preprocessor macros @samp{__mips_dsp} and @samp{__mips_dspr2}.
15589 It also defines @samp{__mips_dsp_rev} to 2.
15592 @itemx -mno-smartmips
15593 @opindex msmartmips
15594 @opindex mno-smartmips
15595 Use (do not use) the MIPS SmartMIPS ASE.
15597 @item -mpaired-single
15598 @itemx -mno-paired-single
15599 @opindex mpaired-single
15600 @opindex mno-paired-single
15601 Use (do not use) paired-single floating-point instructions.
15602 @xref{MIPS Paired-Single Support}. This option requires
15603 hardware floating-point support to be enabled.
15609 Use (do not use) MIPS Digital Media Extension instructions.
15610 This option can only be used when generating 64-bit code and requires
15611 hardware floating-point support to be enabled.
15616 @opindex mno-mips3d
15617 Use (do not use) the MIPS-3D ASE@. @xref{MIPS-3D Built-in Functions}.
15618 The option @option{-mips3d} implies @option{-mpaired-single}.
15624 Use (do not use) MT Multithreading instructions.
15628 Force @code{long} types to be 64 bits wide. See @option{-mlong32} for
15629 an explanation of the default and the way that the pointer size is
15634 Force @code{long}, @code{int}, and pointer types to be 32 bits wide.
15636 The default size of @code{int}s, @code{long}s and pointers depends on
15637 the ABI@. All the supported ABIs use 32-bit @code{int}s. The n64 ABI
15638 uses 64-bit @code{long}s, as does the 64-bit EABI; the others use
15639 32-bit @code{long}s. Pointers are the same size as @code{long}s,
15640 or the same size as integer registers, whichever is smaller.
15646 Assume (do not assume) that all symbols have 32-bit values, regardless
15647 of the selected ABI@. This option is useful in combination with
15648 @option{-mabi=64} and @option{-mno-abicalls} because it allows GCC
15649 to generate shorter and faster references to symbolic addresses.
15653 Put definitions of externally-visible data in a small data section
15654 if that data is no bigger than @var{num} bytes. GCC can then access
15655 the data more efficiently; see @option{-mgpopt} for details.
15657 The default @option{-G} option depends on the configuration.
15659 @item -mlocal-sdata
15660 @itemx -mno-local-sdata
15661 @opindex mlocal-sdata
15662 @opindex mno-local-sdata
15663 Extend (do not extend) the @option{-G} behavior to local data too,
15664 such as to static variables in C@. @option{-mlocal-sdata} is the
15665 default for all configurations.
15667 If the linker complains that an application is using too much small data,
15668 you might want to try rebuilding the less performance-critical parts with
15669 @option{-mno-local-sdata}. You might also want to build large
15670 libraries with @option{-mno-local-sdata}, so that the libraries leave
15671 more room for the main program.
15673 @item -mextern-sdata
15674 @itemx -mno-extern-sdata
15675 @opindex mextern-sdata
15676 @opindex mno-extern-sdata
15677 Assume (do not assume) that externally-defined data will be in
15678 a small data section if that data is within the @option{-G} limit.
15679 @option{-mextern-sdata} is the default for all configurations.
15681 If you compile a module @var{Mod} with @option{-mextern-sdata} @option{-G
15682 @var{num}} @option{-mgpopt}, and @var{Mod} references a variable @var{Var}
15683 that is no bigger than @var{num} bytes, you must make sure that @var{Var}
15684 is placed in a small data section. If @var{Var} is defined by another
15685 module, you must either compile that module with a high-enough
15686 @option{-G} setting or attach a @code{section} attribute to @var{Var}'s
15687 definition. If @var{Var} is common, you must link the application
15688 with a high-enough @option{-G} setting.
15690 The easiest way of satisfying these restrictions is to compile
15691 and link every module with the same @option{-G} option. However,
15692 you may wish to build a library that supports several different
15693 small data limits. You can do this by compiling the library with
15694 the highest supported @option{-G} setting and additionally using
15695 @option{-mno-extern-sdata} to stop the library from making assumptions
15696 about externally-defined data.
15702 Use (do not use) GP-relative accesses for symbols that are known to be
15703 in a small data section; see @option{-G}, @option{-mlocal-sdata} and
15704 @option{-mextern-sdata}. @option{-mgpopt} is the default for all
15707 @option{-mno-gpopt} is useful for cases where the @code{$gp} register
15708 might not hold the value of @code{_gp}. For example, if the code is
15709 part of a library that might be used in a boot monitor, programs that
15710 call boot monitor routines will pass an unknown value in @code{$gp}.
15711 (In such situations, the boot monitor itself would usually be compiled
15712 with @option{-G0}.)
15714 @option{-mno-gpopt} implies @option{-mno-local-sdata} and
15715 @option{-mno-extern-sdata}.
15717 @item -membedded-data
15718 @itemx -mno-embedded-data
15719 @opindex membedded-data
15720 @opindex mno-embedded-data
15721 Allocate variables to the read-only data section first if possible, then
15722 next in the small data section if possible, otherwise in data. This gives
15723 slightly slower code than the default, but reduces the amount of RAM required
15724 when executing, and thus may be preferred for some embedded systems.
15726 @item -muninit-const-in-rodata
15727 @itemx -mno-uninit-const-in-rodata
15728 @opindex muninit-const-in-rodata
15729 @opindex mno-uninit-const-in-rodata
15730 Put uninitialized @code{const} variables in the read-only data section.
15731 This option is only meaningful in conjunction with @option{-membedded-data}.
15733 @item -mcode-readable=@var{setting}
15734 @opindex mcode-readable
15735 Specify whether GCC may generate code that reads from executable sections.
15736 There are three possible settings:
15739 @item -mcode-readable=yes
15740 Instructions may freely access executable sections. This is the
15743 @item -mcode-readable=pcrel
15744 MIPS16 PC-relative load instructions can access executable sections,
15745 but other instructions must not do so. This option is useful on 4KSc
15746 and 4KSd processors when the code TLBs have the Read Inhibit bit set.
15747 It is also useful on processors that can be configured to have a dual
15748 instruction/data SRAM interface and that, like the M4K, automatically
15749 redirect PC-relative loads to the instruction RAM.
15751 @item -mcode-readable=no
15752 Instructions must not access executable sections. This option can be
15753 useful on targets that are configured to have a dual instruction/data
15754 SRAM interface but that (unlike the M4K) do not automatically redirect
15755 PC-relative loads to the instruction RAM.
15758 @item -msplit-addresses
15759 @itemx -mno-split-addresses
15760 @opindex msplit-addresses
15761 @opindex mno-split-addresses
15762 Enable (disable) use of the @code{%hi()} and @code{%lo()} assembler
15763 relocation operators. This option has been superseded by
15764 @option{-mexplicit-relocs} but is retained for backwards compatibility.
15766 @item -mexplicit-relocs
15767 @itemx -mno-explicit-relocs
15768 @opindex mexplicit-relocs
15769 @opindex mno-explicit-relocs
15770 Use (do not use) assembler relocation operators when dealing with symbolic
15771 addresses. The alternative, selected by @option{-mno-explicit-relocs},
15772 is to use assembler macros instead.
15774 @option{-mexplicit-relocs} is the default if GCC was configured
15775 to use an assembler that supports relocation operators.
15777 @item -mcheck-zero-division
15778 @itemx -mno-check-zero-division
15779 @opindex mcheck-zero-division
15780 @opindex mno-check-zero-division
15781 Trap (do not trap) on integer division by zero.
15783 The default is @option{-mcheck-zero-division}.
15785 @item -mdivide-traps
15786 @itemx -mdivide-breaks
15787 @opindex mdivide-traps
15788 @opindex mdivide-breaks
15789 MIPS systems check for division by zero by generating either a
15790 conditional trap or a break instruction. Using traps results in
15791 smaller code, but is only supported on MIPS II and later. Also, some
15792 versions of the Linux kernel have a bug that prevents trap from
15793 generating the proper signal (@code{SIGFPE}). Use @option{-mdivide-traps} to
15794 allow conditional traps on architectures that support them and
15795 @option{-mdivide-breaks} to force the use of breaks.
15797 The default is usually @option{-mdivide-traps}, but this can be
15798 overridden at configure time using @option{--with-divide=breaks}.
15799 Divide-by-zero checks can be completely disabled using
15800 @option{-mno-check-zero-division}.
15805 @opindex mno-memcpy
15806 Force (do not force) the use of @code{memcpy()} for non-trivial block
15807 moves. The default is @option{-mno-memcpy}, which allows GCC to inline
15808 most constant-sized copies.
15811 @itemx -mno-long-calls
15812 @opindex mlong-calls
15813 @opindex mno-long-calls
15814 Disable (do not disable) use of the @code{jal} instruction. Calling
15815 functions using @code{jal} is more efficient but requires the caller
15816 and callee to be in the same 256 megabyte segment.
15818 This option has no effect on abicalls code. The default is
15819 @option{-mno-long-calls}.
15825 Enable (disable) use of the @code{mad}, @code{madu} and @code{mul}
15826 instructions, as provided by the R4650 ISA@.
15829 @itemx -mno-fused-madd
15830 @opindex mfused-madd
15831 @opindex mno-fused-madd
15832 Enable (disable) use of the floating-point multiply-accumulate
15833 instructions, when they are available. The default is
15834 @option{-mfused-madd}.
15836 When multiply-accumulate instructions are used, the intermediate
15837 product is calculated to infinite precision and is not subject to
15838 the FCSR Flush to Zero bit. This may be undesirable in some
15843 Tell the MIPS assembler to not run its preprocessor over user
15844 assembler files (with a @samp{.s} suffix) when assembling them.
15849 @opindex mno-fix-24k
15850 Work around the 24K E48 (lost data on stores during refill) errata.
15851 The workarounds are implemented by the assembler rather than by GCC.
15854 @itemx -mno-fix-r4000
15855 @opindex mfix-r4000
15856 @opindex mno-fix-r4000
15857 Work around certain R4000 CPU errata:
15860 A double-word or a variable shift may give an incorrect result if executed
15861 immediately after starting an integer division.
15863 A double-word or a variable shift may give an incorrect result if executed
15864 while an integer multiplication is in progress.
15866 An integer division may give an incorrect result if started in a delay slot
15867 of a taken branch or a jump.
15871 @itemx -mno-fix-r4400
15872 @opindex mfix-r4400
15873 @opindex mno-fix-r4400
15874 Work around certain R4400 CPU errata:
15877 A double-word or a variable shift may give an incorrect result if executed
15878 immediately after starting an integer division.
15882 @itemx -mno-fix-r10000
15883 @opindex mfix-r10000
15884 @opindex mno-fix-r10000
15885 Work around certain R10000 errata:
15888 @code{ll}/@code{sc} sequences may not behave atomically on revisions
15889 prior to 3.0. They may deadlock on revisions 2.6 and earlier.
15892 This option can only be used if the target architecture supports
15893 branch-likely instructions. @option{-mfix-r10000} is the default when
15894 @option{-march=r10000} is used; @option{-mno-fix-r10000} is the default
15898 @itemx -mno-fix-vr4120
15899 @opindex mfix-vr4120
15900 Work around certain VR4120 errata:
15903 @code{dmultu} does not always produce the correct result.
15905 @code{div} and @code{ddiv} do not always produce the correct result if one
15906 of the operands is negative.
15908 The workarounds for the division errata rely on special functions in
15909 @file{libgcc.a}. At present, these functions are only provided by
15910 the @code{mips64vr*-elf} configurations.
15912 Other VR4120 errata require a nop to be inserted between certain pairs of
15913 instructions. These errata are handled by the assembler, not by GCC itself.
15916 @opindex mfix-vr4130
15917 Work around the VR4130 @code{mflo}/@code{mfhi} errata. The
15918 workarounds are implemented by the assembler rather than by GCC,
15919 although GCC will avoid using @code{mflo} and @code{mfhi} if the
15920 VR4130 @code{macc}, @code{macchi}, @code{dmacc} and @code{dmacchi}
15921 instructions are available instead.
15924 @itemx -mno-fix-sb1
15926 Work around certain SB-1 CPU core errata.
15927 (This flag currently works around the SB-1 revision 2
15928 ``F1'' and ``F2'' floating-point errata.)
15930 @item -mr10k-cache-barrier=@var{setting}
15931 @opindex mr10k-cache-barrier
15932 Specify whether GCC should insert cache barriers to avoid the
15933 side-effects of speculation on R10K processors.
15935 In common with many processors, the R10K tries to predict the outcome
15936 of a conditional branch and speculatively executes instructions from
15937 the ``taken'' branch. It later aborts these instructions if the
15938 predicted outcome was wrong. However, on the R10K, even aborted
15939 instructions can have side effects.
15941 This problem only affects kernel stores and, depending on the system,
15942 kernel loads. As an example, a speculatively-executed store may load
15943 the target memory into cache and mark the cache line as dirty, even if
15944 the store itself is later aborted. If a DMA operation writes to the
15945 same area of memory before the ``dirty'' line is flushed, the cached
15946 data will overwrite the DMA-ed data. See the R10K processor manual
15947 for a full description, including other potential problems.
15949 One workaround is to insert cache barrier instructions before every memory
15950 access that might be speculatively executed and that might have side
15951 effects even if aborted. @option{-mr10k-cache-barrier=@var{setting}}
15952 controls GCC's implementation of this workaround. It assumes that
15953 aborted accesses to any byte in the following regions will not have
15958 the memory occupied by the current function's stack frame;
15961 the memory occupied by an incoming stack argument;
15964 the memory occupied by an object with a link-time-constant address.
15967 It is the kernel's responsibility to ensure that speculative
15968 accesses to these regions are indeed safe.
15970 If the input program contains a function declaration such as:
15976 then the implementation of @code{foo} must allow @code{j foo} and
15977 @code{jal foo} to be executed speculatively. GCC honors this
15978 restriction for functions it compiles itself. It expects non-GCC
15979 functions (such as hand-written assembly code) to do the same.
15981 The option has three forms:
15984 @item -mr10k-cache-barrier=load-store
15985 Insert a cache barrier before a load or store that might be
15986 speculatively executed and that might have side effects even
15989 @item -mr10k-cache-barrier=store
15990 Insert a cache barrier before a store that might be speculatively
15991 executed and that might have side effects even if aborted.
15993 @item -mr10k-cache-barrier=none
15994 Disable the insertion of cache barriers. This is the default setting.
15997 @item -mflush-func=@var{func}
15998 @itemx -mno-flush-func
15999 @opindex mflush-func
16000 Specifies the function to call to flush the I and D caches, or to not
16001 call any such function. If called, the function must take the same
16002 arguments as the common @code{_flush_func()}, that is, the address of the
16003 memory range for which the cache is being flushed, the size of the
16004 memory range, and the number 3 (to flush both caches). The default
16005 depends on the target GCC was configured for, but commonly is either
16006 @samp{_flush_func} or @samp{__cpu_flush}.
16008 @item mbranch-cost=@var{num}
16009 @opindex mbranch-cost
16010 Set the cost of branches to roughly @var{num} ``simple'' instructions.
16011 This cost is only a heuristic and is not guaranteed to produce
16012 consistent results across releases. A zero cost redundantly selects
16013 the default, which is based on the @option{-mtune} setting.
16015 @item -mbranch-likely
16016 @itemx -mno-branch-likely
16017 @opindex mbranch-likely
16018 @opindex mno-branch-likely
16019 Enable or disable use of Branch Likely instructions, regardless of the
16020 default for the selected architecture. By default, Branch Likely
16021 instructions may be generated if they are supported by the selected
16022 architecture. An exception is for the MIPS32 and MIPS64 architectures
16023 and processors that implement those architectures; for those, Branch
16024 Likely instructions will not be generated by default because the MIPS32
16025 and MIPS64 architectures specifically deprecate their use.
16027 @item -mfp-exceptions
16028 @itemx -mno-fp-exceptions
16029 @opindex mfp-exceptions
16030 Specifies whether FP exceptions are enabled. This affects how we schedule
16031 FP instructions for some processors. The default is that FP exceptions are
16034 For instance, on the SB-1, if FP exceptions are disabled, and we are emitting
16035 64-bit code, then we can use both FP pipes. Otherwise, we can only use one
16038 @item -mvr4130-align
16039 @itemx -mno-vr4130-align
16040 @opindex mvr4130-align
16041 The VR4130 pipeline is two-way superscalar, but can only issue two
16042 instructions together if the first one is 8-byte aligned. When this
16043 option is enabled, GCC will align pairs of instructions that it
16044 thinks should execute in parallel.
16046 This option only has an effect when optimizing for the VR4130.
16047 It normally makes code faster, but at the expense of making it bigger.
16048 It is enabled by default at optimization level @option{-O3}.
16053 Enable (disable) generation of @code{synci} instructions on
16054 architectures that support it. The @code{synci} instructions (if
16055 enabled) will be generated when @code{__builtin___clear_cache()} is
16058 This option defaults to @code{-mno-synci}, but the default can be
16059 overridden by configuring with @code{--with-synci}.
16061 When compiling code for single processor systems, it is generally safe
16062 to use @code{synci}. However, on many multi-core (SMP) systems, it
16063 will not invalidate the instruction caches on all cores and may lead
16064 to undefined behavior.
16066 @item -mrelax-pic-calls
16067 @itemx -mno-relax-pic-calls
16068 @opindex mrelax-pic-calls
16069 Try to turn PIC calls that are normally dispatched via register
16070 @code{$25} into direct calls. This is only possible if the linker can
16071 resolve the destination at link-time and if the destination is within
16072 range for a direct call.
16074 @option{-mrelax-pic-calls} is the default if GCC was configured to use
16075 an assembler and a linker that supports the @code{.reloc} assembly
16076 directive and @code{-mexplicit-relocs} is in effect. With
16077 @code{-mno-explicit-relocs}, this optimization can be performed by the
16078 assembler and the linker alone without help from the compiler.
16080 @item -mmcount-ra-address
16081 @itemx -mno-mcount-ra-address
16082 @opindex mmcount-ra-address
16083 @opindex mno-mcount-ra-address
16084 Emit (do not emit) code that allows @code{_mcount} to modify the
16085 calling function's return address. When enabled, this option extends
16086 the usual @code{_mcount} interface with a new @var{ra-address}
16087 parameter, which has type @code{intptr_t *} and is passed in register
16088 @code{$12}. @code{_mcount} can then modify the return address by
16089 doing both of the following:
16092 Returning the new address in register @code{$31}.
16094 Storing the new address in @code{*@var{ra-address}},
16095 if @var{ra-address} is nonnull.
16098 The default is @option{-mno-mcount-ra-address}.
16103 @subsection MMIX Options
16104 @cindex MMIX Options
16106 These options are defined for the MMIX:
16110 @itemx -mno-libfuncs
16112 @opindex mno-libfuncs
16113 Specify that intrinsic library functions are being compiled, passing all
16114 values in registers, no matter the size.
16117 @itemx -mno-epsilon
16119 @opindex mno-epsilon
16120 Generate floating-point comparison instructions that compare with respect
16121 to the @code{rE} epsilon register.
16123 @item -mabi=mmixware
16125 @opindex mabi=mmixware
16127 Generate code that passes function parameters and return values that (in
16128 the called function) are seen as registers @code{$0} and up, as opposed to
16129 the GNU ABI which uses global registers @code{$231} and up.
16131 @item -mzero-extend
16132 @itemx -mno-zero-extend
16133 @opindex mzero-extend
16134 @opindex mno-zero-extend
16135 When reading data from memory in sizes shorter than 64 bits, use (do not
16136 use) zero-extending load instructions by default, rather than
16137 sign-extending ones.
16140 @itemx -mno-knuthdiv
16142 @opindex mno-knuthdiv
16143 Make the result of a division yielding a remainder have the same sign as
16144 the divisor. With the default, @option{-mno-knuthdiv}, the sign of the
16145 remainder follows the sign of the dividend. Both methods are
16146 arithmetically valid, the latter being almost exclusively used.
16148 @item -mtoplevel-symbols
16149 @itemx -mno-toplevel-symbols
16150 @opindex mtoplevel-symbols
16151 @opindex mno-toplevel-symbols
16152 Prepend (do not prepend) a @samp{:} to all global symbols, so the assembly
16153 code can be used with the @code{PREFIX} assembly directive.
16157 Generate an executable in the ELF format, rather than the default
16158 @samp{mmo} format used by the @command{mmix} simulator.
16160 @item -mbranch-predict
16161 @itemx -mno-branch-predict
16162 @opindex mbranch-predict
16163 @opindex mno-branch-predict
16164 Use (do not use) the probable-branch instructions, when static branch
16165 prediction indicates a probable branch.
16167 @item -mbase-addresses
16168 @itemx -mno-base-addresses
16169 @opindex mbase-addresses
16170 @opindex mno-base-addresses
16171 Generate (do not generate) code that uses @emph{base addresses}. Using a
16172 base address automatically generates a request (handled by the assembler
16173 and the linker) for a constant to be set up in a global register. The
16174 register is used for one or more base address requests within the range 0
16175 to 255 from the value held in the register. The generally leads to short
16176 and fast code, but the number of different data items that can be
16177 addressed is limited. This means that a program that uses lots of static
16178 data may require @option{-mno-base-addresses}.
16180 @item -msingle-exit
16181 @itemx -mno-single-exit
16182 @opindex msingle-exit
16183 @opindex mno-single-exit
16184 Force (do not force) generated code to have a single exit point in each
16188 @node MN10300 Options
16189 @subsection MN10300 Options
16190 @cindex MN10300 options
16192 These @option{-m} options are defined for Matsushita MN10300 architectures:
16197 Generate code to avoid bugs in the multiply instructions for the MN10300
16198 processors. This is the default.
16200 @item -mno-mult-bug
16201 @opindex mno-mult-bug
16202 Do not generate code to avoid bugs in the multiply instructions for the
16203 MN10300 processors.
16207 Generate code using features specific to the AM33 processor.
16211 Do not generate code using features specific to the AM33 processor. This
16216 Generate code using features specific to the AM33/2.0 processor.
16220 Generate code using features specific to the AM34 processor.
16222 @item -mtune=@var{cpu-type}
16224 Use the timing characteristics of the indicated CPU type when
16225 scheduling instructions. This does not change the targeted processor
16226 type. The CPU type must be one of @samp{mn10300}, @samp{am33},
16227 @samp{am33-2} or @samp{am34}.
16229 @item -mreturn-pointer-on-d0
16230 @opindex mreturn-pointer-on-d0
16231 When generating a function that returns a pointer, return the pointer
16232 in both @code{a0} and @code{d0}. Otherwise, the pointer is returned
16233 only in a0, and attempts to call such functions without a prototype
16234 would result in errors. Note that this option is on by default; use
16235 @option{-mno-return-pointer-on-d0} to disable it.
16239 Do not link in the C run-time initialization object file.
16243 Indicate to the linker that it should perform a relaxation optimization pass
16244 to shorten branches, calls and absolute memory addresses. This option only
16245 has an effect when used on the command line for the final link step.
16247 This option makes symbolic debugging impossible.
16251 Allow the compiler to generate @emph{Long Instruction Word}
16252 instructions if the target is the @samp{AM33} or later. This is the
16253 default. This option defines the preprocessor macro @samp{__LIW__}.
16257 Do not allow the compiler to generate @emph{Long Instruction Word}
16258 instructions. This option defines the preprocessor macro
16263 Allow the compiler to generate the @emph{SETLB} and @emph{Lcc}
16264 instructions if the target is the @samp{AM33} or later. This is the
16265 default. This option defines the preprocessor macro @samp{__SETLB__}.
16269 Do not allow the compiler to generate @emph{SETLB} or @emph{Lcc}
16270 instructions. This option defines the preprocessor macro
16271 @samp{__NO_SETLB__}.
16275 @node PDP-11 Options
16276 @subsection PDP-11 Options
16277 @cindex PDP-11 Options
16279 These options are defined for the PDP-11:
16284 Use hardware FPP floating point. This is the default. (FIS floating
16285 point on the PDP-11/40 is not supported.)
16288 @opindex msoft-float
16289 Do not use hardware floating point.
16293 Return floating-point results in ac0 (fr0 in Unix assembler syntax).
16297 Return floating-point results in memory. This is the default.
16301 Generate code for a PDP-11/40.
16305 Generate code for a PDP-11/45. This is the default.
16309 Generate code for a PDP-11/10.
16311 @item -mbcopy-builtin
16312 @opindex mbcopy-builtin
16313 Use inline @code{movmemhi} patterns for copying memory. This is the
16318 Do not use inline @code{movmemhi} patterns for copying memory.
16324 Use 16-bit @code{int}. This is the default.
16330 Use 32-bit @code{int}.
16333 @itemx -mno-float32
16335 @opindex mno-float32
16336 Use 64-bit @code{float}. This is the default.
16339 @itemx -mno-float64
16341 @opindex mno-float64
16342 Use 32-bit @code{float}.
16346 Use @code{abshi2} pattern. This is the default.
16350 Do not use @code{abshi2} pattern.
16352 @item -mbranch-expensive
16353 @opindex mbranch-expensive
16354 Pretend that branches are expensive. This is for experimenting with
16355 code generation only.
16357 @item -mbranch-cheap
16358 @opindex mbranch-cheap
16359 Do not pretend that branches are expensive. This is the default.
16363 Use Unix assembler syntax. This is the default when configured for
16364 @samp{pdp11-*-bsd}.
16368 Use DEC assembler syntax. This is the default when configured for any
16369 PDP-11 target other than @samp{pdp11-*-bsd}.
16372 @node picoChip Options
16373 @subsection picoChip Options
16374 @cindex picoChip options
16376 These @samp{-m} options are defined for picoChip implementations:
16380 @item -mae=@var{ae_type}
16382 Set the instruction set, register set, and instruction scheduling
16383 parameters for array element type @var{ae_type}. Supported values
16384 for @var{ae_type} are @samp{ANY}, @samp{MUL}, and @samp{MAC}.
16386 @option{-mae=ANY} selects a completely generic AE type. Code
16387 generated with this option will run on any of the other AE types. The
16388 code will not be as efficient as it would be if compiled for a specific
16389 AE type, and some types of operation (e.g., multiplication) will not
16390 work properly on all types of AE.
16392 @option{-mae=MUL} selects a MUL AE type. This is the most useful AE type
16393 for compiled code, and is the default.
16395 @option{-mae=MAC} selects a DSP-style MAC AE. Code compiled with this
16396 option may suffer from poor performance of byte (char) manipulation,
16397 since the DSP AE does not provide hardware support for byte load/stores.
16399 @item -msymbol-as-address
16400 Enable the compiler to directly use a symbol name as an address in a
16401 load/store instruction, without first loading it into a
16402 register. Typically, the use of this option will generate larger
16403 programs, which run faster than when the option isn't used. However, the
16404 results vary from program to program, so it is left as a user option,
16405 rather than being permanently enabled.
16407 @item -mno-inefficient-warnings
16408 Disables warnings about the generation of inefficient code. These
16409 warnings can be generated, for example, when compiling code that
16410 performs byte-level memory operations on the MAC AE type. The MAC AE has
16411 no hardware support for byte-level memory operations, so all byte
16412 load/stores must be synthesized from word load/store operations. This is
16413 inefficient and a warning will be generated indicating to the programmer
16414 that they should rewrite the code to avoid byte operations, or to target
16415 an AE type that has the necessary hardware support. This option enables
16416 the warning to be turned off.
16420 @node PowerPC Options
16421 @subsection PowerPC Options
16422 @cindex PowerPC options
16424 These are listed under @xref{RS/6000 and PowerPC Options}.
16427 @subsection RL78 Options
16428 @cindex RL78 Options
16434 Links in additional target libraries to support operation within a
16441 Specifies the type of hardware multiplication support to be used. The
16442 default is @code{none}, which uses software multiplication functions.
16443 The @code{g13} option is for the hardware multiply/divide peripheral
16444 only on the RL78/G13 targets. The @code{rl78} option is for the
16445 standard hardware multiplication defined in the RL78 software manual.
16449 @node RS/6000 and PowerPC Options
16450 @subsection IBM RS/6000 and PowerPC Options
16451 @cindex RS/6000 and PowerPC Options
16452 @cindex IBM RS/6000 and PowerPC Options
16454 These @samp{-m} options are defined for the IBM RS/6000 and PowerPC:
16461 @itemx -mno-powerpc
16462 @itemx -mpowerpc-gpopt
16463 @itemx -mno-powerpc-gpopt
16464 @itemx -mpowerpc-gfxopt
16465 @itemx -mno-powerpc-gfxopt
16468 @itemx -mno-powerpc64
16472 @itemx -mno-popcntb
16474 @itemx -mno-popcntd
16483 @itemx -mno-hard-dfp
16487 @opindex mno-power2
16489 @opindex mno-powerpc
16490 @opindex mpowerpc-gpopt
16491 @opindex mno-powerpc-gpopt
16492 @opindex mpowerpc-gfxopt
16493 @opindex mno-powerpc-gfxopt
16494 @opindex mpowerpc64
16495 @opindex mno-powerpc64
16499 @opindex mno-popcntb
16501 @opindex mno-popcntd
16507 @opindex mno-mfpgpr
16509 @opindex mno-hard-dfp
16510 GCC supports two related instruction set architectures for the
16511 RS/6000 and PowerPC@. The @dfn{POWER} instruction set are those
16512 instructions supported by the @samp{rios} chip set used in the original
16513 RS/6000 systems and the @dfn{PowerPC} instruction set is the
16514 architecture of the Freescale MPC5xx, MPC6xx, MPC8xx microprocessors, and
16515 the IBM 4xx, 6xx, and follow-on microprocessors.
16517 Neither architecture is a subset of the other. However there is a
16518 large common subset of instructions supported by both. An MQ
16519 register is included in processors supporting the POWER architecture.
16521 You use these options to specify which instructions are available on the
16522 processor you are using. The default value of these options is
16523 determined when configuring GCC@. Specifying the
16524 @option{-mcpu=@var{cpu_type}} overrides the specification of these
16525 options. We recommend you use the @option{-mcpu=@var{cpu_type}} option
16526 rather than the options listed above.
16528 The @option{-mpower} option allows GCC to generate instructions that
16529 are found only in the POWER architecture and to use the MQ register.
16530 Specifying @option{-mpower2} implies @option{-power} and also allows GCC
16531 to generate instructions that are present in the POWER2 architecture but
16532 not the original POWER architecture.
16534 The @option{-mpowerpc} option allows GCC to generate instructions that
16535 are found only in the 32-bit subset of the PowerPC architecture.
16536 Specifying @option{-mpowerpc-gpopt} implies @option{-mpowerpc} and also allows
16537 GCC to use the optional PowerPC architecture instructions in the
16538 General Purpose group, including floating-point square root. Specifying
16539 @option{-mpowerpc-gfxopt} implies @option{-mpowerpc} and also allows GCC to
16540 use the optional PowerPC architecture instructions in the Graphics
16541 group, including floating-point select.
16543 The @option{-mmfcrf} option allows GCC to generate the move from
16544 condition register field instruction implemented on the POWER4
16545 processor and other processors that support the PowerPC V2.01
16547 The @option{-mpopcntb} option allows GCC to generate the popcount and
16548 double-precision FP reciprocal estimate instruction implemented on the
16549 POWER5 processor and other processors that support the PowerPC V2.02
16551 The @option{-mpopcntd} option allows GCC to generate the popcount
16552 instruction implemented on the POWER7 processor and other processors
16553 that support the PowerPC V2.06 architecture.
16554 The @option{-mfprnd} option allows GCC to generate the FP round to
16555 integer instructions implemented on the POWER5+ processor and other
16556 processors that support the PowerPC V2.03 architecture.
16557 The @option{-mcmpb} option allows GCC to generate the compare bytes
16558 instruction implemented on the POWER6 processor and other processors
16559 that support the PowerPC V2.05 architecture.
16560 The @option{-mmfpgpr} option allows GCC to generate the FP move to/from
16561 general-purpose register instructions implemented on the POWER6X
16562 processor and other processors that support the extended PowerPC V2.05
16564 The @option{-mhard-dfp} option allows GCC to generate the decimal
16565 floating-point instructions implemented on some POWER processors.
16567 The @option{-mpowerpc64} option allows GCC to generate the additional
16568 64-bit instructions that are found in the full PowerPC64 architecture
16569 and to treat GPRs as 64-bit, doubleword quantities. GCC defaults to
16570 @option{-mno-powerpc64}.
16572 If you specify both @option{-mno-power} and @option{-mno-powerpc}, GCC
16573 will use only the instructions in the common subset of both
16574 architectures plus some special AIX common-mode calls, and will not use
16575 the MQ register. Specifying both @option{-mpower} and @option{-mpowerpc}
16576 permits GCC to use any instruction from either architecture and to
16577 allow use of the MQ register; specify this for the Motorola MPC601.
16579 @item -mnew-mnemonics
16580 @itemx -mold-mnemonics
16581 @opindex mnew-mnemonics
16582 @opindex mold-mnemonics
16583 Select which mnemonics to use in the generated assembler code. With
16584 @option{-mnew-mnemonics}, GCC uses the assembler mnemonics defined for
16585 the PowerPC architecture. With @option{-mold-mnemonics} it uses the
16586 assembler mnemonics defined for the POWER architecture. Instructions
16587 defined in only one architecture have only one mnemonic; GCC uses that
16588 mnemonic irrespective of which of these options is specified.
16590 GCC defaults to the mnemonics appropriate for the architecture in
16591 use. Specifying @option{-mcpu=@var{cpu_type}} sometimes overrides the
16592 value of these option. Unless you are building a cross-compiler, you
16593 should normally not specify either @option{-mnew-mnemonics} or
16594 @option{-mold-mnemonics}, but should instead accept the default.
16596 @item -mcpu=@var{cpu_type}
16598 Set architecture type, register usage, choice of mnemonics, and
16599 instruction scheduling parameters for machine type @var{cpu_type}.
16600 Supported values for @var{cpu_type} are @samp{401}, @samp{403},
16601 @samp{405}, @samp{405fp}, @samp{440}, @samp{440fp}, @samp{464}, @samp{464fp},
16602 @samp{476}, @samp{476fp}, @samp{505}, @samp{601}, @samp{602}, @samp{603},
16603 @samp{603e}, @samp{604}, @samp{604e}, @samp{620}, @samp{630}, @samp{740},
16604 @samp{7400}, @samp{7450}, @samp{750}, @samp{801}, @samp{821}, @samp{823},
16605 @samp{860}, @samp{970}, @samp{8540}, @samp{a2}, @samp{e300c2},
16606 @samp{e300c3}, @samp{e500mc}, @samp{e500mc64}, @samp{ec603e}, @samp{G3},
16607 @samp{G4}, @samp{G5}, @samp{titan}, @samp{power}, @samp{power2}, @samp{power3},
16608 @samp{power4}, @samp{power5}, @samp{power5+}, @samp{power6}, @samp{power6x},
16609 @samp{power7}, @samp{common}, @samp{powerpc}, @samp{powerpc64}, @samp{rios},
16610 @samp{rios1}, @samp{rios2}, @samp{rsc}, and @samp{rs64}.
16612 @option{-mcpu=common} selects a completely generic processor. Code
16613 generated under this option will run on any POWER or PowerPC processor.
16614 GCC will use only the instructions in the common subset of both
16615 architectures, and will not use the MQ register. GCC assumes a generic
16616 processor model for scheduling purposes.
16618 @option{-mcpu=power}, @option{-mcpu=power2}, @option{-mcpu=powerpc}, and
16619 @option{-mcpu=powerpc64} specify generic POWER, POWER2, pure 32-bit
16620 PowerPC (i.e., not MPC601), and 64-bit PowerPC architecture machine
16621 types, with an appropriate, generic processor model assumed for
16622 scheduling purposes.
16624 The other options specify a specific processor. Code generated under
16625 those options will run best on that processor, and may not run at all on
16628 The @option{-mcpu} options automatically enable or disable the
16631 @gccoptlist{-maltivec -mfprnd -mhard-float -mmfcrf -mmultiple @gol
16632 -mnew-mnemonics -mpopcntb -mpopcntd -mpower -mpower2 -mpowerpc64 @gol
16633 -mpowerpc-gpopt -mpowerpc-gfxopt -msingle-float -mdouble-float @gol
16634 -msimple-fpu -mstring -mmulhw -mdlmzb -mmfpgpr -mvsx}
16636 The particular options set for any particular CPU will vary between
16637 compiler versions, depending on what setting seems to produce optimal
16638 code for that CPU; it doesn't necessarily reflect the actual hardware's
16639 capabilities. If you wish to set an individual option to a particular
16640 value, you may specify it after the @option{-mcpu} option, like
16641 @samp{-mcpu=970 -mno-altivec}.
16643 On AIX, the @option{-maltivec} and @option{-mpowerpc64} options are
16644 not enabled or disabled by the @option{-mcpu} option at present because
16645 AIX does not have full support for these options. You may still
16646 enable or disable them individually if you're sure it'll work in your
16649 @item -mtune=@var{cpu_type}
16651 Set the instruction scheduling parameters for machine type
16652 @var{cpu_type}, but do not set the architecture type, register usage, or
16653 choice of mnemonics, as @option{-mcpu=@var{cpu_type}} would. The same
16654 values for @var{cpu_type} are used for @option{-mtune} as for
16655 @option{-mcpu}. If both are specified, the code generated will use the
16656 architecture, registers, and mnemonics set by @option{-mcpu}, but the
16657 scheduling parameters set by @option{-mtune}.
16659 @item -mcmodel=small
16660 @opindex mcmodel=small
16661 Generate PowerPC64 code for the small model: The TOC is limited to
16664 @item -mcmodel=medium
16665 @opindex mcmodel=medium
16666 Generate PowerPC64 code for the medium model: The TOC and other static
16667 data may be up to a total of 4G in size.
16669 @item -mcmodel=large
16670 @opindex mcmodel=large
16671 Generate PowerPC64 code for the large model: The TOC may be up to 4G
16672 in size. Other data and code is only limited by the 64-bit address
16676 @itemx -mno-altivec
16678 @opindex mno-altivec
16679 Generate code that uses (does not use) AltiVec instructions, and also
16680 enable the use of built-in functions that allow more direct access to
16681 the AltiVec instruction set. You may also need to set
16682 @option{-mabi=altivec} to adjust the current ABI with AltiVec ABI
16688 @opindex mno-vrsave
16689 Generate VRSAVE instructions when generating AltiVec code.
16691 @item -mgen-cell-microcode
16692 @opindex mgen-cell-microcode
16693 Generate Cell microcode instructions
16695 @item -mwarn-cell-microcode
16696 @opindex mwarn-cell-microcode
16697 Warning when a Cell microcode instruction is going to emitted. An example
16698 of a Cell microcode instruction is a variable shift.
16701 @opindex msecure-plt
16702 Generate code that allows ld and ld.so to build executables and shared
16703 libraries with non-exec .plt and .got sections. This is a PowerPC
16704 32-bit SYSV ABI option.
16708 Generate code that uses a BSS .plt section that ld.so fills in, and
16709 requires .plt and .got sections that are both writable and executable.
16710 This is a PowerPC 32-bit SYSV ABI option.
16716 This switch enables or disables the generation of ISEL instructions.
16718 @item -misel=@var{yes/no}
16719 This switch has been deprecated. Use @option{-misel} and
16720 @option{-mno-isel} instead.
16726 This switch enables or disables the generation of SPE simd
16732 @opindex mno-paired
16733 This switch enables or disables the generation of PAIRED simd
16736 @item -mspe=@var{yes/no}
16737 This option has been deprecated. Use @option{-mspe} and
16738 @option{-mno-spe} instead.
16744 Generate code that uses (does not use) vector/scalar (VSX)
16745 instructions, and also enable the use of built-in functions that allow
16746 more direct access to the VSX instruction set.
16748 @item -mfloat-gprs=@var{yes/single/double/no}
16749 @itemx -mfloat-gprs
16750 @opindex mfloat-gprs
16751 This switch enables or disables the generation of floating-point
16752 operations on the general-purpose registers for architectures that
16755 The argument @var{yes} or @var{single} enables the use of
16756 single-precision floating-point operations.
16758 The argument @var{double} enables the use of single and
16759 double-precision floating-point operations.
16761 The argument @var{no} disables floating-point operations on the
16762 general-purpose registers.
16764 This option is currently only available on the MPC854x.
16770 Generate code for 32-bit or 64-bit environments of Darwin and SVR4
16771 targets (including GNU/Linux). The 32-bit environment sets int, long
16772 and pointer to 32 bits and generates code that runs on any PowerPC
16773 variant. The 64-bit environment sets int to 32 bits and long and
16774 pointer to 64 bits, and generates code for PowerPC64, as for
16775 @option{-mpowerpc64}.
16778 @itemx -mno-fp-in-toc
16779 @itemx -mno-sum-in-toc
16780 @itemx -mminimal-toc
16782 @opindex mno-fp-in-toc
16783 @opindex mno-sum-in-toc
16784 @opindex mminimal-toc
16785 Modify generation of the TOC (Table Of Contents), which is created for
16786 every executable file. The @option{-mfull-toc} option is selected by
16787 default. In that case, GCC will allocate at least one TOC entry for
16788 each unique non-automatic variable reference in your program. GCC
16789 will also place floating-point constants in the TOC@. However, only
16790 16,384 entries are available in the TOC@.
16792 If you receive a linker error message that saying you have overflowed
16793 the available TOC space, you can reduce the amount of TOC space used
16794 with the @option{-mno-fp-in-toc} and @option{-mno-sum-in-toc} options.
16795 @option{-mno-fp-in-toc} prevents GCC from putting floating-point
16796 constants in the TOC and @option{-mno-sum-in-toc} forces GCC to
16797 generate code to calculate the sum of an address and a constant at
16798 run time instead of putting that sum into the TOC@. You may specify one
16799 or both of these options. Each causes GCC to produce very slightly
16800 slower and larger code at the expense of conserving TOC space.
16802 If you still run out of space in the TOC even when you specify both of
16803 these options, specify @option{-mminimal-toc} instead. This option causes
16804 GCC to make only one TOC entry for every file. When you specify this
16805 option, GCC will produce code that is slower and larger but which
16806 uses extremely little TOC space. You may wish to use this option
16807 only on files that contain less frequently executed code.
16813 Enable 64-bit AIX ABI and calling convention: 64-bit pointers, 64-bit
16814 @code{long} type, and the infrastructure needed to support them.
16815 Specifying @option{-maix64} implies @option{-mpowerpc64} and
16816 @option{-mpowerpc}, while @option{-maix32} disables the 64-bit ABI and
16817 implies @option{-mno-powerpc64}. GCC defaults to @option{-maix32}.
16820 @itemx -mno-xl-compat
16821 @opindex mxl-compat
16822 @opindex mno-xl-compat
16823 Produce code that conforms more closely to IBM XL compiler semantics
16824 when using AIX-compatible ABI@. Pass floating-point arguments to
16825 prototyped functions beyond the register save area (RSA) on the stack
16826 in addition to argument FPRs. Do not assume that most significant
16827 double in 128-bit long double value is properly rounded when comparing
16828 values and converting to double. Use XL symbol names for long double
16831 The AIX calling convention was extended but not initially documented to
16832 handle an obscure K&R C case of calling a function that takes the
16833 address of its arguments with fewer arguments than declared. IBM XL
16834 compilers access floating-point arguments that do not fit in the
16835 RSA from the stack when a subroutine is compiled without
16836 optimization. Because always storing floating-point arguments on the
16837 stack is inefficient and rarely needed, this option is not enabled by
16838 default and only is necessary when calling subroutines compiled by IBM
16839 XL compilers without optimization.
16843 Support @dfn{IBM RS/6000 SP} @dfn{Parallel Environment} (PE)@. Link an
16844 application written to use message passing with special startup code to
16845 enable the application to run. The system must have PE installed in the
16846 standard location (@file{/usr/lpp/ppe.poe/}), or the @file{specs} file
16847 must be overridden with the @option{-specs=} option to specify the
16848 appropriate directory location. The Parallel Environment does not
16849 support threads, so the @option{-mpe} option and the @option{-pthread}
16850 option are incompatible.
16852 @item -malign-natural
16853 @itemx -malign-power
16854 @opindex malign-natural
16855 @opindex malign-power
16856 On AIX, 32-bit Darwin, and 64-bit PowerPC GNU/Linux, the option
16857 @option{-malign-natural} overrides the ABI-defined alignment of larger
16858 types, such as floating-point doubles, on their natural size-based boundary.
16859 The option @option{-malign-power} instructs GCC to follow the ABI-specified
16860 alignment rules. GCC defaults to the standard alignment defined in the ABI@.
16862 On 64-bit Darwin, natural alignment is the default, and @option{-malign-power}
16866 @itemx -mhard-float
16867 @opindex msoft-float
16868 @opindex mhard-float
16869 Generate code that does not use (uses) the floating-point register set.
16870 Software floating-point emulation is provided if you use the
16871 @option{-msoft-float} option, and pass the option to GCC when linking.
16873 @item -msingle-float
16874 @itemx -mdouble-float
16875 @opindex msingle-float
16876 @opindex mdouble-float
16877 Generate code for single- or double-precision floating-point operations.
16878 @option{-mdouble-float} implies @option{-msingle-float}.
16881 @opindex msimple-fpu
16882 Do not generate sqrt and div instructions for hardware floating-point unit.
16886 Specify type of floating-point unit. Valid values are @var{sp_lite}
16887 (equivalent to -msingle-float -msimple-fpu), @var{dp_lite} (equivalent
16888 to -mdouble-float -msimple-fpu), @var{sp_full} (equivalent to -msingle-float),
16889 and @var{dp_full} (equivalent to -mdouble-float).
16892 @opindex mxilinx-fpu
16893 Perform optimizations for the floating-point unit on Xilinx PPC 405/440.
16896 @itemx -mno-multiple
16898 @opindex mno-multiple
16899 Generate code that uses (does not use) the load multiple word
16900 instructions and the store multiple word instructions. These
16901 instructions are generated by default on POWER systems, and not
16902 generated on PowerPC systems. Do not use @option{-mmultiple} on little-endian
16903 PowerPC systems, since those instructions do not work when the
16904 processor is in little-endian mode. The exceptions are PPC740 and
16905 PPC750 which permit these instructions in little-endian mode.
16910 @opindex mno-string
16911 Generate code that uses (does not use) the load string instructions
16912 and the store string word instructions to save multiple registers and
16913 do small block moves. These instructions are generated by default on
16914 POWER systems, and not generated on PowerPC systems. Do not use
16915 @option{-mstring} on little-endian PowerPC systems, since those
16916 instructions do not work when the processor is in little-endian mode.
16917 The exceptions are PPC740 and PPC750 which permit these instructions
16918 in little-endian mode.
16923 @opindex mno-update
16924 Generate code that uses (does not use) the load or store instructions
16925 that update the base register to the address of the calculated memory
16926 location. These instructions are generated by default. If you use
16927 @option{-mno-update}, there is a small window between the time that the
16928 stack pointer is updated and the address of the previous frame is
16929 stored, which means code that walks the stack frame across interrupts or
16930 signals may get corrupted data.
16932 @item -mavoid-indexed-addresses
16933 @itemx -mno-avoid-indexed-addresses
16934 @opindex mavoid-indexed-addresses
16935 @opindex mno-avoid-indexed-addresses
16936 Generate code that tries to avoid (not avoid) the use of indexed load
16937 or store instructions. These instructions can incur a performance
16938 penalty on Power6 processors in certain situations, such as when
16939 stepping through large arrays that cross a 16M boundary. This option
16940 is enabled by default when targetting Power6 and disabled otherwise.
16943 @itemx -mno-fused-madd
16944 @opindex mfused-madd
16945 @opindex mno-fused-madd
16946 Generate code that uses (does not use) the floating-point multiply and
16947 accumulate instructions. These instructions are generated by default
16948 if hardware floating point is used. The machine-dependent
16949 @option{-mfused-madd} option is now mapped to the machine-independent
16950 @option{-ffp-contract=fast} option, and @option{-mno-fused-madd} is
16951 mapped to @option{-ffp-contract=off}.
16957 Generate code that uses (does not use) the half-word multiply and
16958 multiply-accumulate instructions on the IBM 405, 440, 464 and 476 processors.
16959 These instructions are generated by default when targetting those
16966 Generate code that uses (does not use) the string-search @samp{dlmzb}
16967 instruction on the IBM 405, 440, 464 and 476 processors. This instruction is
16968 generated by default when targetting those processors.
16970 @item -mno-bit-align
16972 @opindex mno-bit-align
16973 @opindex mbit-align
16974 On System V.4 and embedded PowerPC systems do not (do) force structures
16975 and unions that contain bit-fields to be aligned to the base type of the
16978 For example, by default a structure containing nothing but 8
16979 @code{unsigned} bit-fields of length 1 is aligned to a 4-byte
16980 boundary and has a size of 4 bytes. By using @option{-mno-bit-align},
16981 the structure is aligned to a 1-byte boundary and is 1 byte in
16984 @item -mno-strict-align
16985 @itemx -mstrict-align
16986 @opindex mno-strict-align
16987 @opindex mstrict-align
16988 On System V.4 and embedded PowerPC systems do not (do) assume that
16989 unaligned memory references will be handled by the system.
16991 @item -mrelocatable
16992 @itemx -mno-relocatable
16993 @opindex mrelocatable
16994 @opindex mno-relocatable
16995 Generate code that allows (does not allow) a static executable to be
16996 relocated to a different address at run time. A simple embedded
16997 PowerPC system loader should relocate the entire contents of
16998 @code{.got2} and 4-byte locations listed in the @code{.fixup} section,
16999 a table of 32-bit addresses generated by this option. For this to
17000 work, all objects linked together must be compiled with
17001 @option{-mrelocatable} or @option{-mrelocatable-lib}.
17002 @option{-mrelocatable} code aligns the stack to an 8-byte boundary.
17004 @item -mrelocatable-lib
17005 @itemx -mno-relocatable-lib
17006 @opindex mrelocatable-lib
17007 @opindex mno-relocatable-lib
17008 Like @option{-mrelocatable}, @option{-mrelocatable-lib} generates a
17009 @code{.fixup} section to allow static executables to be relocated at
17010 run time, but @option{-mrelocatable-lib} does not use the smaller stack
17011 alignment of @option{-mrelocatable}. Objects compiled with
17012 @option{-mrelocatable-lib} may be linked with objects compiled with
17013 any combination of the @option{-mrelocatable} options.
17019 On System V.4 and embedded PowerPC systems do not (do) assume that
17020 register 2 contains a pointer to a global area pointing to the addresses
17021 used in the program.
17024 @itemx -mlittle-endian
17026 @opindex mlittle-endian
17027 On System V.4 and embedded PowerPC systems compile code for the
17028 processor in little-endian mode. The @option{-mlittle-endian} option is
17029 the same as @option{-mlittle}.
17032 @itemx -mbig-endian
17034 @opindex mbig-endian
17035 On System V.4 and embedded PowerPC systems compile code for the
17036 processor in big-endian mode. The @option{-mbig-endian} option is
17037 the same as @option{-mbig}.
17039 @item -mdynamic-no-pic
17040 @opindex mdynamic-no-pic
17041 On Darwin and Mac OS X systems, compile code so that it is not
17042 relocatable, but that its external references are relocatable. The
17043 resulting code is suitable for applications, but not shared
17046 @item -msingle-pic-base
17047 @opindex msingle-pic-base
17048 Treat the register used for PIC addressing as read-only, rather than
17049 loading it in the prologue for each function. The runtime system is
17050 responsible for initializing this register with an appropriate value
17051 before execution begins.
17053 @item -mprioritize-restricted-insns=@var{priority}
17054 @opindex mprioritize-restricted-insns
17055 This option controls the priority that is assigned to
17056 dispatch-slot restricted instructions during the second scheduling
17057 pass. The argument @var{priority} takes the value @var{0/1/2} to assign
17058 @var{no/highest/second-highest} priority to dispatch slot restricted
17061 @item -msched-costly-dep=@var{dependence_type}
17062 @opindex msched-costly-dep
17063 This option controls which dependences are considered costly
17064 by the target during instruction scheduling. The argument
17065 @var{dependence_type} takes one of the following values:
17066 @var{no}: no dependence is costly,
17067 @var{all}: all dependences are costly,
17068 @var{true_store_to_load}: a true dependence from store to load is costly,
17069 @var{store_to_load}: any dependence from store to load is costly,
17070 @var{number}: any dependence for which latency >= @var{number} is costly.
17072 @item -minsert-sched-nops=@var{scheme}
17073 @opindex minsert-sched-nops
17074 This option controls which nop insertion scheme will be used during
17075 the second scheduling pass. The argument @var{scheme} takes one of the
17077 @var{no}: Don't insert nops.
17078 @var{pad}: Pad with nops any dispatch group that has vacant issue slots,
17079 according to the scheduler's grouping.
17080 @var{regroup_exact}: Insert nops to force costly dependent insns into
17081 separate groups. Insert exactly as many nops as needed to force an insn
17082 to a new group, according to the estimated processor grouping.
17083 @var{number}: Insert nops to force costly dependent insns into
17084 separate groups. Insert @var{number} nops to force an insn to a new group.
17087 @opindex mcall-sysv
17088 On System V.4 and embedded PowerPC systems compile code using calling
17089 conventions that adheres to the March 1995 draft of the System V
17090 Application Binary Interface, PowerPC processor supplement. This is the
17091 default unless you configured GCC using @samp{powerpc-*-eabiaix}.
17093 @item -mcall-sysv-eabi
17095 @opindex mcall-sysv-eabi
17096 @opindex mcall-eabi
17097 Specify both @option{-mcall-sysv} and @option{-meabi} options.
17099 @item -mcall-sysv-noeabi
17100 @opindex mcall-sysv-noeabi
17101 Specify both @option{-mcall-sysv} and @option{-mno-eabi} options.
17103 @item -mcall-aixdesc
17105 On System V.4 and embedded PowerPC systems compile code for the AIX
17109 @opindex mcall-linux
17110 On System V.4 and embedded PowerPC systems compile code for the
17111 Linux-based GNU system.
17113 @item -mcall-freebsd
17114 @opindex mcall-freebsd
17115 On System V.4 and embedded PowerPC systems compile code for the
17116 FreeBSD operating system.
17118 @item -mcall-netbsd
17119 @opindex mcall-netbsd
17120 On System V.4 and embedded PowerPC systems compile code for the
17121 NetBSD operating system.
17123 @item -mcall-openbsd
17124 @opindex mcall-netbsd
17125 On System V.4 and embedded PowerPC systems compile code for the
17126 OpenBSD operating system.
17128 @item -maix-struct-return
17129 @opindex maix-struct-return
17130 Return all structures in memory (as specified by the AIX ABI)@.
17132 @item -msvr4-struct-return
17133 @opindex msvr4-struct-return
17134 Return structures smaller than 8 bytes in registers (as specified by the
17137 @item -mabi=@var{abi-type}
17139 Extend the current ABI with a particular extension, or remove such extension.
17140 Valid values are @var{altivec}, @var{no-altivec}, @var{spe},
17141 @var{no-spe}, @var{ibmlongdouble}, @var{ieeelongdouble}@.
17145 Extend the current ABI with SPE ABI extensions. This does not change
17146 the default ABI, instead it adds the SPE ABI extensions to the current
17150 @opindex mabi=no-spe
17151 Disable Booke SPE ABI extensions for the current ABI@.
17153 @item -mabi=ibmlongdouble
17154 @opindex mabi=ibmlongdouble
17155 Change the current ABI to use IBM extended-precision long double.
17156 This is a PowerPC 32-bit SYSV ABI option.
17158 @item -mabi=ieeelongdouble
17159 @opindex mabi=ieeelongdouble
17160 Change the current ABI to use IEEE extended-precision long double.
17161 This is a PowerPC 32-bit Linux ABI option.
17164 @itemx -mno-prototype
17165 @opindex mprototype
17166 @opindex mno-prototype
17167 On System V.4 and embedded PowerPC systems assume that all calls to
17168 variable argument functions are properly prototyped. Otherwise, the
17169 compiler must insert an instruction before every non prototyped call to
17170 set or clear bit 6 of the condition code register (@var{CR}) to
17171 indicate whether floating-point values were passed in the floating-point
17172 registers in case the function takes variable arguments. With
17173 @option{-mprototype}, only calls to prototyped variable argument functions
17174 will set or clear the bit.
17178 On embedded PowerPC systems, assume that the startup module is called
17179 @file{sim-crt0.o} and that the standard C libraries are @file{libsim.a} and
17180 @file{libc.a}. This is the default for @samp{powerpc-*-eabisim}
17185 On embedded PowerPC systems, assume that the startup module is called
17186 @file{crt0.o} and the standard C libraries are @file{libmvme.a} and
17191 On embedded PowerPC systems, assume that the startup module is called
17192 @file{crt0.o} and the standard C libraries are @file{libads.a} and
17195 @item -myellowknife
17196 @opindex myellowknife
17197 On embedded PowerPC systems, assume that the startup module is called
17198 @file{crt0.o} and the standard C libraries are @file{libyk.a} and
17203 On System V.4 and embedded PowerPC systems, specify that you are
17204 compiling for a VxWorks system.
17208 On embedded PowerPC systems, set the @var{PPC_EMB} bit in the ELF flags
17209 header to indicate that @samp{eabi} extended relocations are used.
17215 On System V.4 and embedded PowerPC systems do (do not) adhere to the
17216 Embedded Applications Binary Interface (eabi) which is a set of
17217 modifications to the System V.4 specifications. Selecting @option{-meabi}
17218 means that the stack is aligned to an 8-byte boundary, a function
17219 @code{__eabi} is called to from @code{main} to set up the eabi
17220 environment, and the @option{-msdata} option can use both @code{r2} and
17221 @code{r13} to point to two separate small data areas. Selecting
17222 @option{-mno-eabi} means that the stack is aligned to a 16-byte boundary,
17223 do not call an initialization function from @code{main}, and the
17224 @option{-msdata} option will only use @code{r13} to point to a single
17225 small data area. The @option{-meabi} option is on by default if you
17226 configured GCC using one of the @samp{powerpc*-*-eabi*} options.
17229 @opindex msdata=eabi
17230 On System V.4 and embedded PowerPC systems, put small initialized
17231 @code{const} global and static data in the @samp{.sdata2} section, which
17232 is pointed to by register @code{r2}. Put small initialized
17233 non-@code{const} global and static data in the @samp{.sdata} section,
17234 which is pointed to by register @code{r13}. Put small uninitialized
17235 global and static data in the @samp{.sbss} section, which is adjacent to
17236 the @samp{.sdata} section. The @option{-msdata=eabi} option is
17237 incompatible with the @option{-mrelocatable} option. The
17238 @option{-msdata=eabi} option also sets the @option{-memb} option.
17241 @opindex msdata=sysv
17242 On System V.4 and embedded PowerPC systems, put small global and static
17243 data in the @samp{.sdata} section, which is pointed to by register
17244 @code{r13}. Put small uninitialized global and static data in the
17245 @samp{.sbss} section, which is adjacent to the @samp{.sdata} section.
17246 The @option{-msdata=sysv} option is incompatible with the
17247 @option{-mrelocatable} option.
17249 @item -msdata=default
17251 @opindex msdata=default
17253 On System V.4 and embedded PowerPC systems, if @option{-meabi} is used,
17254 compile code the same as @option{-msdata=eabi}, otherwise compile code the
17255 same as @option{-msdata=sysv}.
17258 @opindex msdata=data
17259 On System V.4 and embedded PowerPC systems, put small global
17260 data in the @samp{.sdata} section. Put small uninitialized global
17261 data in the @samp{.sbss} section. Do not use register @code{r13}
17262 to address small data however. This is the default behavior unless
17263 other @option{-msdata} options are used.
17267 @opindex msdata=none
17269 On embedded PowerPC systems, put all initialized global and static data
17270 in the @samp{.data} section, and all uninitialized data in the
17271 @samp{.bss} section.
17273 @item -mblock-move-inline-limit=@var{num}
17274 @opindex mblock-move-inline-limit
17275 Inline all block moves (such as calls to @code{memcpy} or structure
17276 copies) less than or equal to @var{num} bytes. The minimum value for
17277 @var{num} is 32 bytes on 32-bit targets and 64 bytes on 64-bit
17278 targets. The default value is target-specific.
17282 @cindex smaller data references (PowerPC)
17283 @cindex .sdata/.sdata2 references (PowerPC)
17284 On embedded PowerPC systems, put global and static items less than or
17285 equal to @var{num} bytes into the small data or bss sections instead of
17286 the normal data or bss section. By default, @var{num} is 8. The
17287 @option{-G @var{num}} switch is also passed to the linker.
17288 All modules should be compiled with the same @option{-G @var{num}} value.
17291 @itemx -mno-regnames
17293 @opindex mno-regnames
17294 On System V.4 and embedded PowerPC systems do (do not) emit register
17295 names in the assembly language output using symbolic forms.
17298 @itemx -mno-longcall
17300 @opindex mno-longcall
17301 By default assume that all calls are far away so that a longer more
17302 expensive calling sequence is required. This is required for calls
17303 further than 32 megabytes (33,554,432 bytes) from the current location.
17304 A short call will be generated if the compiler knows
17305 the call cannot be that far away. This setting can be overridden by
17306 the @code{shortcall} function attribute, or by @code{#pragma
17309 Some linkers are capable of detecting out-of-range calls and generating
17310 glue code on the fly. On these systems, long calls are unnecessary and
17311 generate slower code. As of this writing, the AIX linker can do this,
17312 as can the GNU linker for PowerPC/64. It is planned to add this feature
17313 to the GNU linker for 32-bit PowerPC systems as well.
17315 On Darwin/PPC systems, @code{#pragma longcall} will generate ``jbsr
17316 callee, L42'', plus a ``branch island'' (glue code). The two target
17317 addresses represent the callee and the ``branch island''. The
17318 Darwin/PPC linker will prefer the first address and generate a ``bl
17319 callee'' if the PPC ``bl'' instruction will reach the callee directly;
17320 otherwise, the linker will generate ``bl L42'' to call the ``branch
17321 island''. The ``branch island'' is appended to the body of the
17322 calling function; it computes the full 32-bit address of the callee
17325 On Mach-O (Darwin) systems, this option directs the compiler emit to
17326 the glue for every direct call, and the Darwin linker decides whether
17327 to use or discard it.
17329 In the future, we may cause GCC to ignore all longcall specifications
17330 when the linker is known to generate glue.
17332 @item -mtls-markers
17333 @itemx -mno-tls-markers
17334 @opindex mtls-markers
17335 @opindex mno-tls-markers
17336 Mark (do not mark) calls to @code{__tls_get_addr} with a relocation
17337 specifying the function argument. The relocation allows ld to
17338 reliably associate function call with argument setup instructions for
17339 TLS optimization, which in turn allows gcc to better schedule the
17344 Adds support for multithreading with the @dfn{pthreads} library.
17345 This option sets flags for both the preprocessor and linker.
17350 This option will enable GCC to use the reciprocal estimate and
17351 reciprocal square root estimate instructions with additional
17352 Newton-Raphson steps to increase precision instead of doing a divide or
17353 square root and divide for floating-point arguments. You should use
17354 the @option{-ffast-math} option when using @option{-mrecip} (or at
17355 least @option{-funsafe-math-optimizations},
17356 @option{-finite-math-only}, @option{-freciprocal-math} and
17357 @option{-fno-trapping-math}). Note that while the throughput of the
17358 sequence is generally higher than the throughput of the non-reciprocal
17359 instruction, the precision of the sequence can be decreased by up to 2
17360 ulp (i.e. the inverse of 1.0 equals 0.99999994) for reciprocal square
17363 @item -mrecip=@var{opt}
17364 @opindex mrecip=opt
17365 This option allows to control which reciprocal estimate instructions
17366 may be used. @var{opt} is a comma separated list of options, which may
17367 be preceded by a @code{!} to invert the option:
17368 @code{all}: enable all estimate instructions,
17369 @code{default}: enable the default instructions, equivalent to @option{-mrecip},
17370 @code{none}: disable all estimate instructions, equivalent to @option{-mno-recip};
17371 @code{div}: enable the reciprocal approximation instructions for both single and double precision;
17372 @code{divf}: enable the single-precision reciprocal approximation instructions;
17373 @code{divd}: enable the double-precision reciprocal approximation instructions;
17374 @code{rsqrt}: enable the reciprocal square root approximation instructions for both single and double precision;
17375 @code{rsqrtf}: enable the single-precision reciprocal square root approximation instructions;
17376 @code{rsqrtd}: enable the double-precision reciprocal square root approximation instructions;
17378 So for example, @option{-mrecip=all,!rsqrtd} would enable the
17379 all of the reciprocal estimate instructions, except for the
17380 @code{FRSQRTE}, @code{XSRSQRTEDP}, and @code{XVRSQRTEDP} instructions
17381 which handle the double-precision reciprocal square root calculations.
17383 @item -mrecip-precision
17384 @itemx -mno-recip-precision
17385 @opindex mrecip-precision
17386 Assume (do not assume) that the reciprocal estimate instructions
17387 provide higher-precision estimates than is mandated by the PowerPC
17388 ABI. Selecting @option{-mcpu=power6} or @option{-mcpu=power7}
17389 automatically selects @option{-mrecip-precision}. The double-precision
17390 square root estimate instructions are not generated by
17391 default on low-precision machines, since they do not provide an
17392 estimate that converges after three steps.
17394 @item -mveclibabi=@var{type}
17395 @opindex mveclibabi
17396 Specifies the ABI type to use for vectorizing intrinsics using an
17397 external library. The only type supported at present is @code{mass},
17398 which specifies to use IBM's Mathematical Acceleration Subsystem
17399 (MASS) libraries for vectorizing intrinsics using external libraries.
17400 GCC will currently emit calls to @code{acosd2}, @code{acosf4},
17401 @code{acoshd2}, @code{acoshf4}, @code{asind2}, @code{asinf4},
17402 @code{asinhd2}, @code{asinhf4}, @code{atan2d2}, @code{atan2f4},
17403 @code{atand2}, @code{atanf4}, @code{atanhd2}, @code{atanhf4},
17404 @code{cbrtd2}, @code{cbrtf4}, @code{cosd2}, @code{cosf4},
17405 @code{coshd2}, @code{coshf4}, @code{erfcd2}, @code{erfcf4},
17406 @code{erfd2}, @code{erff4}, @code{exp2d2}, @code{exp2f4},
17407 @code{expd2}, @code{expf4}, @code{expm1d2}, @code{expm1f4},
17408 @code{hypotd2}, @code{hypotf4}, @code{lgammad2}, @code{lgammaf4},
17409 @code{log10d2}, @code{log10f4}, @code{log1pd2}, @code{log1pf4},
17410 @code{log2d2}, @code{log2f4}, @code{logd2}, @code{logf4},
17411 @code{powd2}, @code{powf4}, @code{sind2}, @code{sinf4}, @code{sinhd2},
17412 @code{sinhf4}, @code{sqrtd2}, @code{sqrtf4}, @code{tand2},
17413 @code{tanf4}, @code{tanhd2}, and @code{tanhf4} when generating code
17414 for power7. Both @option{-ftree-vectorize} and
17415 @option{-funsafe-math-optimizations} have to be enabled. The MASS
17416 libraries will have to be specified at link time.
17421 Generate (do not generate) the @code{friz} instruction when the
17422 @option{-funsafe-math-optimizations} option is used to optimize
17423 rounding of floating-point values to 64-bit integer and back to floating
17424 point. The @code{friz} instruction does not return the same value if
17425 the floating-point number is too large to fit in an integer.
17427 @item -mpointers-to-nested-functions
17428 @itemx -mno-pointers-to-nested-functions
17429 @opindex mpointers-to-nested-functions
17430 Generate (do not generate) code to load up the static chain register
17431 (@var{r11}) when calling through a pointer on AIX and 64-bit Linux
17432 systems where a function pointer points to a 3-word descriptor giving
17433 the function address, TOC value to be loaded in register @var{r2}, and
17434 static chain value to be loaded in register @var{r11}. The
17435 @option{-mpointers-to-nested-functions} is on by default. You will
17436 not be able to call through pointers to nested functions or pointers
17437 to functions compiled in other languages that use the static chain if
17438 you use the @option{-mno-pointers-to-nested-functions}.
17440 @item -msave-toc-indirect
17441 @itemx -mno-save-toc-indirect
17442 @opindex msave-toc-indirect
17443 Generate (do not generate) code to save the TOC value in the reserved
17444 stack location in the function prologue if the function calls through
17445 a pointer on AIX and 64-bit Linux systems. If the TOC value is not
17446 saved in the prologue, it is saved just before the call through the
17447 pointer. The @option{-mno-save-toc-indirect} option is the default.
17451 @subsection RX Options
17454 These command-line options are defined for RX targets:
17457 @item -m64bit-doubles
17458 @itemx -m32bit-doubles
17459 @opindex m64bit-doubles
17460 @opindex m32bit-doubles
17461 Make the @code{double} data type be 64 bits (@option{-m64bit-doubles})
17462 or 32 bits (@option{-m32bit-doubles}) in size. The default is
17463 @option{-m32bit-doubles}. @emph{Note} RX floating-point hardware only
17464 works on 32-bit values, which is why the default is
17465 @option{-m32bit-doubles}.
17471 Enables (@option{-fpu}) or disables (@option{-nofpu}) the use of RX
17472 floating-point hardware. The default is enabled for the @var{RX600}
17473 series and disabled for the @var{RX200} series.
17475 Floating-point instructions will only be generated for 32-bit floating-point
17476 values however, so if the @option{-m64bit-doubles} option is in
17477 use then the FPU hardware will not be used for doubles.
17479 @emph{Note} If the @option{-fpu} option is enabled then
17480 @option{-funsafe-math-optimizations} is also enabled automatically.
17481 This is because the RX FPU instructions are themselves unsafe.
17483 @item -mcpu=@var{name}
17485 Selects the type of RX CPU to be targeted. Currently three types are
17486 supported, the generic @var{RX600} and @var{RX200} series hardware and
17487 the specific @var{RX610} CPU. The default is @var{RX600}.
17489 The only difference between @var{RX600} and @var{RX610} is that the
17490 @var{RX610} does not support the @code{MVTIPL} instruction.
17492 The @var{RX200} series does not have a hardware floating-point unit
17493 and so @option{-nofpu} is enabled by default when this type is
17496 @item -mbig-endian-data
17497 @itemx -mlittle-endian-data
17498 @opindex mbig-endian-data
17499 @opindex mlittle-endian-data
17500 Store data (but not code) in the big-endian format. The default is
17501 @option{-mlittle-endian-data}, i.e.@: to store data in the little-endian
17504 @item -msmall-data-limit=@var{N}
17505 @opindex msmall-data-limit
17506 Specifies the maximum size in bytes of global and static variables
17507 which can be placed into the small data area. Using the small data
17508 area can lead to smaller and faster code, but the size of area is
17509 limited and it is up to the programmer to ensure that the area does
17510 not overflow. Also when the small data area is used one of the RX's
17511 registers (usually @code{r13}) is reserved for use pointing to this
17512 area, so it is no longer available for use by the compiler. This
17513 could result in slower and/or larger code if variables which once
17514 could have been held in the reserved register are now pushed onto the
17517 Note, common variables (variables that have not been initialized) and
17518 constants are not placed into the small data area as they are assigned
17519 to other sections in the output executable.
17521 The default value is zero, which disables this feature. Note, this
17522 feature is not enabled by default with higher optimization levels
17523 (@option{-O2} etc) because of the potentially detrimental effects of
17524 reserving a register. It is up to the programmer to experiment and
17525 discover whether this feature is of benefit to their program. See the
17526 description of the @option{-mpid} option for a description of how the
17527 actual register to hold the small data area pointer is chosen.
17533 Use the simulator runtime. The default is to use the libgloss board
17536 @item -mas100-syntax
17537 @itemx -mno-as100-syntax
17538 @opindex mas100-syntax
17539 @opindex mno-as100-syntax
17540 When generating assembler output use a syntax that is compatible with
17541 Renesas's AS100 assembler. This syntax can also be handled by the GAS
17542 assembler but it has some restrictions so generating it is not the
17545 @item -mmax-constant-size=@var{N}
17546 @opindex mmax-constant-size
17547 Specifies the maximum size, in bytes, of a constant that can be used as
17548 an operand in a RX instruction. Although the RX instruction set does
17549 allow constants of up to 4 bytes in length to be used in instructions,
17550 a longer value equates to a longer instruction. Thus in some
17551 circumstances it can be beneficial to restrict the size of constants
17552 that are used in instructions. Constants that are too big are instead
17553 placed into a constant pool and referenced via register indirection.
17555 The value @var{N} can be between 0 and 4. A value of 0 (the default)
17556 or 4 means that constants of any size are allowed.
17560 Enable linker relaxation. Linker relaxation is a process whereby the
17561 linker will attempt to reduce the size of a program by finding shorter
17562 versions of various instructions. Disabled by default.
17564 @item -mint-register=@var{N}
17565 @opindex mint-register
17566 Specify the number of registers to reserve for fast interrupt handler
17567 functions. The value @var{N} can be between 0 and 4. A value of 1
17568 means that register @code{r13} will be reserved for the exclusive use
17569 of fast interrupt handlers. A value of 2 reserves @code{r13} and
17570 @code{r12}. A value of 3 reserves @code{r13}, @code{r12} and
17571 @code{r11}, and a value of 4 reserves @code{r13} through @code{r10}.
17572 A value of 0, the default, does not reserve any registers.
17574 @item -msave-acc-in-interrupts
17575 @opindex msave-acc-in-interrupts
17576 Specifies that interrupt handler functions should preserve the
17577 accumulator register. This is only necessary if normal code might use
17578 the accumulator register, for example because it performs 64-bit
17579 multiplications. The default is to ignore the accumulator as this
17580 makes the interrupt handlers faster.
17586 Enables the generation of position independent data. When enabled any
17587 access to constant data will done via an offset from a base address
17588 held in a register. This allows the location of constant data to be
17589 determined at run time without requiring the executable to be
17590 relocated, which is a benefit to embedded applications with tight
17591 memory constraints. Data that can be modified is not affected by this
17594 Note, using this feature reserves a register, usually @code{r13}, for
17595 the constant data base address. This can result in slower and/or
17596 larger code, especially in complicated functions.
17598 The actual register chosen to hold the constant data base address
17599 depends upon whether the @option{-msmall-data-limit} and/or the
17600 @option{-mint-register} command-line options are enabled. Starting
17601 with register @code{r13} and proceeding downwards, registers are
17602 allocated first to satisfy the requirements of @option{-mint-register},
17603 then @option{-mpid} and finally @option{-msmall-data-limit}. Thus it
17604 is possible for the small data area register to be @code{r8} if both
17605 @option{-mint-register=4} and @option{-mpid} are specified on the
17608 By default this feature is not enabled. The default can be restored
17609 via the @option{-mno-pid} command-line option.
17613 @emph{Note:} The generic GCC command-line option @option{-ffixed-@var{reg}}
17614 has special significance to the RX port when used with the
17615 @code{interrupt} function attribute. This attribute indicates a
17616 function intended to process fast interrupts. GCC will will ensure
17617 that it only uses the registers @code{r10}, @code{r11}, @code{r12}
17618 and/or @code{r13} and only provided that the normal use of the
17619 corresponding registers have been restricted via the
17620 @option{-ffixed-@var{reg}} or @option{-mint-register} command-line
17623 @node S/390 and zSeries Options
17624 @subsection S/390 and zSeries Options
17625 @cindex S/390 and zSeries Options
17627 These are the @samp{-m} options defined for the S/390 and zSeries architecture.
17631 @itemx -msoft-float
17632 @opindex mhard-float
17633 @opindex msoft-float
17634 Use (do not use) the hardware floating-point instructions and registers
17635 for floating-point operations. When @option{-msoft-float} is specified,
17636 functions in @file{libgcc.a} will be used to perform floating-point
17637 operations. When @option{-mhard-float} is specified, the compiler
17638 generates IEEE floating-point instructions. This is the default.
17641 @itemx -mno-hard-dfp
17643 @opindex mno-hard-dfp
17644 Use (do not use) the hardware decimal-floating-point instructions for
17645 decimal-floating-point operations. When @option{-mno-hard-dfp} is
17646 specified, functions in @file{libgcc.a} will be used to perform
17647 decimal-floating-point operations. When @option{-mhard-dfp} is
17648 specified, the compiler generates decimal-floating-point hardware
17649 instructions. This is the default for @option{-march=z9-ec} or higher.
17651 @item -mlong-double-64
17652 @itemx -mlong-double-128
17653 @opindex mlong-double-64
17654 @opindex mlong-double-128
17655 These switches control the size of @code{long double} type. A size
17656 of 64 bits makes the @code{long double} type equivalent to the @code{double}
17657 type. This is the default.
17660 @itemx -mno-backchain
17661 @opindex mbackchain
17662 @opindex mno-backchain
17663 Store (do not store) the address of the caller's frame as backchain pointer
17664 into the callee's stack frame.
17665 A backchain may be needed to allow debugging using tools that do not understand
17666 DWARF-2 call frame information.
17667 When @option{-mno-packed-stack} is in effect, the backchain pointer is stored
17668 at the bottom of the stack frame; when @option{-mpacked-stack} is in effect,
17669 the backchain is placed into the topmost word of the 96/160 byte register
17672 In general, code compiled with @option{-mbackchain} is call-compatible with
17673 code compiled with @option{-mmo-backchain}; however, use of the backchain
17674 for debugging purposes usually requires that the whole binary is built with
17675 @option{-mbackchain}. Note that the combination of @option{-mbackchain},
17676 @option{-mpacked-stack} and @option{-mhard-float} is not supported. In order
17677 to build a linux kernel use @option{-msoft-float}.
17679 The default is to not maintain the backchain.
17681 @item -mpacked-stack
17682 @itemx -mno-packed-stack
17683 @opindex mpacked-stack
17684 @opindex mno-packed-stack
17685 Use (do not use) the packed stack layout. When @option{-mno-packed-stack} is
17686 specified, the compiler uses the all fields of the 96/160 byte register save
17687 area only for their default purpose; unused fields still take up stack space.
17688 When @option{-mpacked-stack} is specified, register save slots are densely
17689 packed at the top of the register save area; unused space is reused for other
17690 purposes, allowing for more efficient use of the available stack space.
17691 However, when @option{-mbackchain} is also in effect, the topmost word of
17692 the save area is always used to store the backchain, and the return address
17693 register is always saved two words below the backchain.
17695 As long as the stack frame backchain is not used, code generated with
17696 @option{-mpacked-stack} is call-compatible with code generated with
17697 @option{-mno-packed-stack}. Note that some non-FSF releases of GCC 2.95 for
17698 S/390 or zSeries generated code that uses the stack frame backchain at run
17699 time, not just for debugging purposes. Such code is not call-compatible
17700 with code compiled with @option{-mpacked-stack}. Also, note that the
17701 combination of @option{-mbackchain},
17702 @option{-mpacked-stack} and @option{-mhard-float} is not supported. In order
17703 to build a linux kernel use @option{-msoft-float}.
17705 The default is to not use the packed stack layout.
17708 @itemx -mno-small-exec
17709 @opindex msmall-exec
17710 @opindex mno-small-exec
17711 Generate (or do not generate) code using the @code{bras} instruction
17712 to do subroutine calls.
17713 This only works reliably if the total executable size does not
17714 exceed 64k. The default is to use the @code{basr} instruction instead,
17715 which does not have this limitation.
17721 When @option{-m31} is specified, generate code compliant to the
17722 GNU/Linux for S/390 ABI@. When @option{-m64} is specified, generate
17723 code compliant to the GNU/Linux for zSeries ABI@. This allows GCC in
17724 particular to generate 64-bit instructions. For the @samp{s390}
17725 targets, the default is @option{-m31}, while the @samp{s390x}
17726 targets default to @option{-m64}.
17732 When @option{-mzarch} is specified, generate code using the
17733 instructions available on z/Architecture.
17734 When @option{-mesa} is specified, generate code using the
17735 instructions available on ESA/390. Note that @option{-mesa} is
17736 not possible with @option{-m64}.
17737 When generating code compliant to the GNU/Linux for S/390 ABI,
17738 the default is @option{-mesa}. When generating code compliant
17739 to the GNU/Linux for zSeries ABI, the default is @option{-mzarch}.
17745 Generate (or do not generate) code using the @code{mvcle} instruction
17746 to perform block moves. When @option{-mno-mvcle} is specified,
17747 use a @code{mvc} loop instead. This is the default unless optimizing for
17754 Print (or do not print) additional debug information when compiling.
17755 The default is to not print debug information.
17757 @item -march=@var{cpu-type}
17759 Generate code that will run on @var{cpu-type}, which is the name of a system
17760 representing a certain processor type. Possible values for
17761 @var{cpu-type} are @samp{g5}, @samp{g6}, @samp{z900}, @samp{z990},
17762 @samp{z9-109}, @samp{z9-ec} and @samp{z10}.
17763 When generating code using the instructions available on z/Architecture,
17764 the default is @option{-march=z900}. Otherwise, the default is
17765 @option{-march=g5}.
17767 @item -mtune=@var{cpu-type}
17769 Tune to @var{cpu-type} everything applicable about the generated code,
17770 except for the ABI and the set of available instructions.
17771 The list of @var{cpu-type} values is the same as for @option{-march}.
17772 The default is the value used for @option{-march}.
17775 @itemx -mno-tpf-trace
17776 @opindex mtpf-trace
17777 @opindex mno-tpf-trace
17778 Generate code that adds (does not add) in TPF OS specific branches to trace
17779 routines in the operating system. This option is off by default, even
17780 when compiling for the TPF OS@.
17783 @itemx -mno-fused-madd
17784 @opindex mfused-madd
17785 @opindex mno-fused-madd
17786 Generate code that uses (does not use) the floating-point multiply and
17787 accumulate instructions. These instructions are generated by default if
17788 hardware floating point is used.
17790 @item -mwarn-framesize=@var{framesize}
17791 @opindex mwarn-framesize
17792 Emit a warning if the current function exceeds the given frame size. Because
17793 this is a compile-time check it doesn't need to be a real problem when the program
17794 runs. It is intended to identify functions that most probably cause
17795 a stack overflow. It is useful to be used in an environment with limited stack
17796 size e.g.@: the linux kernel.
17798 @item -mwarn-dynamicstack
17799 @opindex mwarn-dynamicstack
17800 Emit a warning if the function calls alloca or uses dynamically
17801 sized arrays. This is generally a bad idea with a limited stack size.
17803 @item -mstack-guard=@var{stack-guard}
17804 @itemx -mstack-size=@var{stack-size}
17805 @opindex mstack-guard
17806 @opindex mstack-size
17807 If these options are provided the s390 back end emits additional instructions in
17808 the function prologue which trigger a trap if the stack size is @var{stack-guard}
17809 bytes above the @var{stack-size} (remember that the stack on s390 grows downward).
17810 If the @var{stack-guard} option is omitted the smallest power of 2 larger than
17811 the frame size of the compiled function is chosen.
17812 These options are intended to be used to help debugging stack overflow problems.
17813 The additionally emitted code causes only little overhead and hence can also be
17814 used in production like systems without greater performance degradation. The given
17815 values have to be exact powers of 2 and @var{stack-size} has to be greater than
17816 @var{stack-guard} without exceeding 64k.
17817 In order to be efficient the extra code makes the assumption that the stack starts
17818 at an address aligned to the value given by @var{stack-size}.
17819 The @var{stack-guard} option can only be used in conjunction with @var{stack-size}.
17822 @node Score Options
17823 @subsection Score Options
17824 @cindex Score Options
17826 These options are defined for Score implementations:
17831 Compile code for big-endian mode. This is the default.
17835 Compile code for little-endian mode.
17839 Disable generate bcnz instruction.
17843 Enable generate unaligned load and store instruction.
17847 Enable the use of multiply-accumulate instructions. Disabled by default.
17851 Specify the SCORE5 as the target architecture.
17855 Specify the SCORE5U of the target architecture.
17859 Specify the SCORE7 as the target architecture. This is the default.
17863 Specify the SCORE7D as the target architecture.
17867 @subsection SH Options
17869 These @samp{-m} options are defined for the SH implementations:
17874 Generate code for the SH1.
17878 Generate code for the SH2.
17881 Generate code for the SH2e.
17885 Generate code for the SH2a without FPU, or for a SH2a-FPU in such a way
17886 that the floating-point unit is not used.
17888 @item -m2a-single-only
17889 @opindex m2a-single-only
17890 Generate code for the SH2a-FPU, in such a way that no double-precision
17891 floating-point operations are used.
17894 @opindex m2a-single
17895 Generate code for the SH2a-FPU assuming the floating-point unit is in
17896 single-precision mode by default.
17900 Generate code for the SH2a-FPU assuming the floating-point unit is in
17901 double-precision mode by default.
17905 Generate code for the SH3.
17909 Generate code for the SH3e.
17913 Generate code for the SH4 without a floating-point unit.
17915 @item -m4-single-only
17916 @opindex m4-single-only
17917 Generate code for the SH4 with a floating-point unit that only
17918 supports single-precision arithmetic.
17922 Generate code for the SH4 assuming the floating-point unit is in
17923 single-precision mode by default.
17927 Generate code for the SH4.
17931 Generate code for the SH4al-dsp, or for a SH4a in such a way that the
17932 floating-point unit is not used.
17934 @item -m4a-single-only
17935 @opindex m4a-single-only
17936 Generate code for the SH4a, in such a way that no double-precision
17937 floating-point operations are used.
17940 @opindex m4a-single
17941 Generate code for the SH4a assuming the floating-point unit is in
17942 single-precision mode by default.
17946 Generate code for the SH4a.
17950 Same as @option{-m4a-nofpu}, except that it implicitly passes
17951 @option{-dsp} to the assembler. GCC doesn't generate any DSP
17952 instructions at the moment.
17956 Compile code for the processor in big-endian mode.
17960 Compile code for the processor in little-endian mode.
17964 Align doubles at 64-bit boundaries. Note that this changes the calling
17965 conventions, and thus some functions from the standard C library will
17966 not work unless you recompile it first with @option{-mdalign}.
17970 Shorten some address references at link time, when possible; uses the
17971 linker option @option{-relax}.
17975 Use 32-bit offsets in @code{switch} tables. The default is to use
17980 Enable the use of bit manipulation instructions on SH2A.
17984 Enable the use of the instruction @code{fmovd}. Check @option{-mdalign} for
17985 alignment constraints.
17989 Comply with the calling conventions defined by Renesas.
17993 Comply with the calling conventions defined by Renesas.
17997 Comply with the calling conventions defined for GCC before the Renesas
17998 conventions were available. This option is the default for all
17999 targets of the SH toolchain.
18002 @opindex mnomacsave
18003 Mark the @code{MAC} register as call-clobbered, even if
18004 @option{-mhitachi} is given.
18010 Control the IEEE compliance of floating-point comparisons, which affects the
18011 handling of cases where the result of a comparison is unordered. By default
18012 @option{-mieee} is implicitly enabled. If @option{-ffinite-math-only} is
18013 enabled @option{-mno-ieee} is implicitly set, which results in faster
18014 floating-point greater-equal and less-equal comparisons. The implcit settings
18015 can be overridden by specifying either @option{-mieee} or @option{-mno-ieee}.
18017 @item -minline-ic_invalidate
18018 @opindex minline-ic_invalidate
18019 Inline code to invalidate instruction cache entries after setting up
18020 nested function trampolines.
18021 This option has no effect if -musermode is in effect and the selected
18022 code generation option (e.g. -m4) does not allow the use of the icbi
18024 If the selected code generation option does not allow the use of the icbi
18025 instruction, and -musermode is not in effect, the inlined code will
18026 manipulate the instruction cache address array directly with an associative
18027 write. This not only requires privileged mode, but it will also
18028 fail if the cache line had been mapped via the TLB and has become unmapped.
18032 Dump instruction size and location in the assembly code.
18035 @opindex mpadstruct
18036 This option is deprecated. It pads structures to multiple of 4 bytes,
18037 which is incompatible with the SH ABI@.
18039 @item -msoft-atomic
18040 @opindex msoft-atomic
18041 Generate GNU/Linux compatible gUSA software atomic sequences for the atomic
18042 built-in functions. The generated atomic sequences require support from the
18043 interrupt / exception handling code of the system and are only suitable for
18044 single-core systems. They will not perform correctly on multi-core systems.
18045 This option is enabled by default when the target is @code{sh-*-linux*}.
18046 For details on the atomic built-in functions see @ref{__atomic Builtins}.
18050 Optimize for space instead of speed. Implied by @option{-Os}.
18053 @opindex mprefergot
18054 When generating position-independent code, emit function calls using
18055 the Global Offset Table instead of the Procedure Linkage Table.
18059 Don't generate privileged mode only code; implies -mno-inline-ic_invalidate
18060 if the inlined code would not work in user mode.
18061 This is the default when the target is @code{sh-*-linux*}.
18063 @item -multcost=@var{number}
18064 @opindex multcost=@var{number}
18065 Set the cost to assume for a multiply insn.
18067 @item -mdiv=@var{strategy}
18068 @opindex mdiv=@var{strategy}
18069 Set the division strategy to be used for integer division operations.
18070 For SHmedia @var{strategy} can be one of:
18075 Performs the operation in floating point. This has a very high latency,
18076 but needs only a few instructions, so it might be a good choice if
18077 your code has enough easily-exploitable ILP to allow the compiler to
18078 schedule the floating-point instructions together with other instructions.
18079 Division by zero causes a floating-point exception.
18082 Uses integer operations to calculate the inverse of the divisor,
18083 and then multiplies the dividend with the inverse. This strategy allows
18084 CSE and hoisting of the inverse calculation. Division by zero calculates
18085 an unspecified result, but does not trap.
18088 A variant of @samp{inv} where, if no CSE or hoisting opportunities
18089 have been found, or if the entire operation has been hoisted to the same
18090 place, the last stages of the inverse calculation are intertwined with the
18091 final multiply to reduce the overall latency, at the expense of using a few
18092 more instructions, and thus offering fewer scheduling opportunities with
18096 Calls a library function that usually implements the @samp{inv:minlat}
18098 This gives high code density for @code{m5-*media-nofpu} compilations.
18101 Uses a different entry point of the same library function, where it
18102 assumes that a pointer to a lookup table has already been set up, which
18103 exposes the pointer load to CSE and code hoisting optimizations.
18108 Use the @samp{inv} algorithm for initial
18109 code generation, but if the code stays unoptimized, revert to the @samp{call},
18110 @samp{call2}, or @samp{fp} strategies, respectively. Note that the
18111 potentially-trapping side effect of division by zero is carried by a
18112 separate instruction, so it is possible that all the integer instructions
18113 are hoisted out, but the marker for the side effect stays where it is.
18114 A recombination to floating-point operations or a call is not possible
18119 Variants of the @samp{inv:minlat} strategy. In the case
18120 that the inverse calculation is not separated from the multiply, they speed
18121 up division where the dividend fits into 20 bits (plus sign where applicable)
18122 by inserting a test to skip a number of operations in this case; this test
18123 slows down the case of larger dividends. @samp{inv20u} assumes the case of a such
18124 a small dividend to be unlikely, and @samp{inv20l} assumes it to be likely.
18128 For targets other than SHmedia @var{strategy} can be one of:
18133 Calls a library function that uses the single-step division instruction
18134 @code{div1} to perform the operation. Division by zero calculates an
18135 unspecified result and does not trap. This is the default except for SH4,
18136 SH2A and SHcompact.
18139 Calls a library function that performs the operation in double precision
18140 floating point. Division by zero causes a floating-point exception. This is
18141 the default for SHcompact with FPU. Specifying this for targets that do not
18142 have a double precision FPU will default to @code{call-div1}.
18145 Calls a library function that uses a lookup table for small divisors and
18146 the @code{div1} instruction with case distinction for larger divisors. Division
18147 by zero calculates an unspecified result and does not trap. This is the default
18148 for SH4. Specifying this for targets that do not have dynamic shift
18149 instructions will default to @code{call-div1}.
18153 When a division strategy has not been specified the default strategy will be
18154 selected based on the current target. For SH2A the default strategy is to
18155 use the @code{divs} and @code{divu} instructions instead of library function
18158 @item -maccumulate-outgoing-args
18159 @opindex maccumulate-outgoing-args
18160 Reserve space once for outgoing arguments in the function prologue rather
18161 than around each call. Generally beneficial for performance and size. Also
18162 needed for unwinding to avoid changing the stack frame around conditional code.
18164 @item -mdivsi3_libfunc=@var{name}
18165 @opindex mdivsi3_libfunc=@var{name}
18166 Set the name of the library function used for 32-bit signed division to
18167 @var{name}. This only affect the name used in the call and inv:call
18168 division strategies, and the compiler will still expect the same
18169 sets of input/output/clobbered registers as if this option was not present.
18171 @item -mfixed-range=@var{register-range}
18172 @opindex mfixed-range
18173 Generate code treating the given register range as fixed registers.
18174 A fixed register is one that the register allocator can not use. This is
18175 useful when compiling kernel code. A register range is specified as
18176 two registers separated by a dash. Multiple register ranges can be
18177 specified separated by a comma.
18179 @item -madjust-unroll
18180 @opindex madjust-unroll
18181 Throttle unrolling to avoid thrashing target registers.
18182 This option only has an effect if the gcc code base supports the
18183 TARGET_ADJUST_UNROLL_MAX target hook.
18185 @item -mindexed-addressing
18186 @opindex mindexed-addressing
18187 Enable the use of the indexed addressing mode for SHmedia32/SHcompact.
18188 This is only safe if the hardware and/or OS implement 32-bit wrap-around
18189 semantics for the indexed addressing mode. The architecture allows the
18190 implementation of processors with 64-bit MMU, which the OS could use to
18191 get 32-bit addressing, but since no current hardware implementation supports
18192 this or any other way to make the indexed addressing mode safe to use in
18193 the 32-bit ABI, the default is @option{-mno-indexed-addressing}.
18195 @item -mgettrcost=@var{number}
18196 @opindex mgettrcost=@var{number}
18197 Set the cost assumed for the gettr instruction to @var{number}.
18198 The default is 2 if @option{-mpt-fixed} is in effect, 100 otherwise.
18202 Assume pt* instructions won't trap. This will generally generate better
18203 scheduled code, but is unsafe on current hardware. The current architecture
18204 definition says that ptabs and ptrel trap when the target anded with 3 is 3.
18205 This has the unintentional effect of making it unsafe to schedule ptabs /
18206 ptrel before a branch, or hoist it out of a loop. For example,
18207 __do_global_ctors, a part of libgcc that runs constructors at program
18208 startup, calls functions in a list which is delimited by @minus{}1. With the
18209 -mpt-fixed option, the ptabs will be done before testing against @minus{}1.
18210 That means that all the constructors will be run a bit quicker, but when
18211 the loop comes to the end of the list, the program crashes because ptabs
18212 loads @minus{}1 into a target register. Since this option is unsafe for any
18213 hardware implementing the current architecture specification, the default
18214 is -mno-pt-fixed. Unless the user specifies a specific cost with
18215 @option{-mgettrcost}, -mno-pt-fixed also implies @option{-mgettrcost=100};
18216 this deters register allocation using target registers for storing
18219 @item -minvalid-symbols
18220 @opindex minvalid-symbols
18221 Assume symbols might be invalid. Ordinary function symbols generated by
18222 the compiler will always be valid to load with movi/shori/ptabs or
18223 movi/shori/ptrel, but with assembler and/or linker tricks it is possible
18224 to generate symbols that will cause ptabs / ptrel to trap.
18225 This option is only meaningful when @option{-mno-pt-fixed} is in effect.
18226 It will then prevent cross-basic-block cse, hoisting and most scheduling
18227 of symbol loads. The default is @option{-mno-invalid-symbols}.
18229 @item -mbranch-cost=@var{num}
18230 @opindex mbranch-cost=@var{num}
18231 Assume @var{num} to be the cost for a branch instruction. Higher numbers
18232 will make the compiler try to generate more branch-free code if possible.
18233 If not specified the value is selected depending on the processor type that
18234 is being compiled for.
18237 @opindex mcbranchdi
18238 Enable the @code{cbranchdi4} instruction pattern.
18242 Emit the @code{cmpeqdi_t} instruction pattern even when @option{-mcbranchdi}
18246 @opindex mfused-madd
18247 Allow the usage of the @code{fmac} instruction (floating-point
18248 multiply-accumulate) if the processor type supports it. Enabling this
18249 option might generate code that produces different numeric floating-point
18250 results compared to strict IEEE 754 arithmetic.
18252 @item -mpretend-cmove
18253 @opindex mpretend-cmove
18254 Prefer zero-displacement conditional branches for conditional move instruction
18255 patterns. This can result in faster code on the SH4 processor.
18259 @node Solaris 2 Options
18260 @subsection Solaris 2 Options
18261 @cindex Solaris 2 options
18263 These @samp{-m} options are supported on Solaris 2:
18266 @item -mimpure-text
18267 @opindex mimpure-text
18268 @option{-mimpure-text}, used in addition to @option{-shared}, tells
18269 the compiler to not pass @option{-z text} to the linker when linking a
18270 shared object. Using this option, you can link position-dependent
18271 code into a shared object.
18273 @option{-mimpure-text} suppresses the ``relocations remain against
18274 allocatable but non-writable sections'' linker error message.
18275 However, the necessary relocations will trigger copy-on-write, and the
18276 shared object is not actually shared across processes. Instead of
18277 using @option{-mimpure-text}, you should compile all source code with
18278 @option{-fpic} or @option{-fPIC}.
18282 These switches are supported in addition to the above on Solaris 2:
18287 Add support for multithreading using the POSIX threads library. This
18288 option sets flags for both the preprocessor and linker. This option does
18289 not affect the thread safety of object code produced by the compiler or
18290 that of libraries supplied with it.
18294 This is a synonym for @option{-pthreads}.
18297 @node SPARC Options
18298 @subsection SPARC Options
18299 @cindex SPARC options
18301 These @samp{-m} options are supported on the SPARC:
18304 @item -mno-app-regs
18306 @opindex mno-app-regs
18308 Specify @option{-mapp-regs} to generate output using the global registers
18309 2 through 4, which the SPARC SVR4 ABI reserves for applications. This
18312 To be fully SVR4 ABI compliant at the cost of some performance loss,
18313 specify @option{-mno-app-regs}. You should compile libraries and system
18314 software with this option.
18320 With @option{-mflat}, the compiler does not generate save/restore instructions
18321 and uses a ``flat'' or single register window model. This model is compatible
18322 with the regular register window model. The local registers and the input
18323 registers (0--5) are still treated as ``call-saved'' registers and will be
18324 saved on the stack as needed.
18326 With @option{-mno-flat} (the default), the compiler generates save/restore
18327 instructions (except for leaf functions). This is the normal operating mode.
18330 @itemx -mhard-float
18332 @opindex mhard-float
18333 Generate output containing floating-point instructions. This is the
18337 @itemx -msoft-float
18339 @opindex msoft-float
18340 Generate output containing library calls for floating point.
18341 @strong{Warning:} the requisite libraries are not available for all SPARC
18342 targets. Normally the facilities of the machine's usual C compiler are
18343 used, but this cannot be done directly in cross-compilation. You must make
18344 your own arrangements to provide suitable library functions for
18345 cross-compilation. The embedded targets @samp{sparc-*-aout} and
18346 @samp{sparclite-*-*} do provide software floating-point support.
18348 @option{-msoft-float} changes the calling convention in the output file;
18349 therefore, it is only useful if you compile @emph{all} of a program with
18350 this option. In particular, you need to compile @file{libgcc.a}, the
18351 library that comes with GCC, with @option{-msoft-float} in order for
18354 @item -mhard-quad-float
18355 @opindex mhard-quad-float
18356 Generate output containing quad-word (long double) floating-point
18359 @item -msoft-quad-float
18360 @opindex msoft-quad-float
18361 Generate output containing library calls for quad-word (long double)
18362 floating-point instructions. The functions called are those specified
18363 in the SPARC ABI@. This is the default.
18365 As of this writing, there are no SPARC implementations that have hardware
18366 support for the quad-word floating-point instructions. They all invoke
18367 a trap handler for one of these instructions, and then the trap handler
18368 emulates the effect of the instruction. Because of the trap handler overhead,
18369 this is much slower than calling the ABI library routines. Thus the
18370 @option{-msoft-quad-float} option is the default.
18372 @item -mno-unaligned-doubles
18373 @itemx -munaligned-doubles
18374 @opindex mno-unaligned-doubles
18375 @opindex munaligned-doubles
18376 Assume that doubles have 8-byte alignment. This is the default.
18378 With @option{-munaligned-doubles}, GCC assumes that doubles have 8-byte
18379 alignment only if they are contained in another type, or if they have an
18380 absolute address. Otherwise, it assumes they have 4-byte alignment.
18381 Specifying this option avoids some rare compatibility problems with code
18382 generated by other compilers. It is not the default because it results
18383 in a performance loss, especially for floating-point code.
18385 @item -mno-faster-structs
18386 @itemx -mfaster-structs
18387 @opindex mno-faster-structs
18388 @opindex mfaster-structs
18389 With @option{-mfaster-structs}, the compiler assumes that structures
18390 should have 8-byte alignment. This enables the use of pairs of
18391 @code{ldd} and @code{std} instructions for copies in structure
18392 assignment, in place of twice as many @code{ld} and @code{st} pairs.
18393 However, the use of this changed alignment directly violates the SPARC
18394 ABI@. Thus, it's intended only for use on targets where the developer
18395 acknowledges that their resulting code will not be directly in line with
18396 the rules of the ABI@.
18398 @item -mcpu=@var{cpu_type}
18400 Set the instruction set, register set, and instruction scheduling parameters
18401 for machine type @var{cpu_type}. Supported values for @var{cpu_type} are
18402 @samp{v7}, @samp{cypress}, @samp{v8}, @samp{supersparc}, @samp{hypersparc},
18403 @samp{leon}, @samp{sparclite}, @samp{f930}, @samp{f934}, @samp{sparclite86x},
18404 @samp{sparclet}, @samp{tsc701}, @samp{v9}, @samp{ultrasparc},
18405 @samp{ultrasparc3}, @samp{niagara}, @samp{niagara2}, @samp{niagara3},
18406 and @samp{niagara4}.
18408 Native Solaris and GNU/Linux toolchains also support the value @samp{native},
18409 which selects the best architecture option for the host processor.
18410 @option{-mcpu=native} has no effect if GCC does not recognize
18413 Default instruction scheduling parameters are used for values that select
18414 an architecture and not an implementation. These are @samp{v7}, @samp{v8},
18415 @samp{sparclite}, @samp{sparclet}, @samp{v9}.
18417 Here is a list of each supported architecture and their supported
18425 supersparc, hypersparc, leon
18428 f930, f934, sparclite86x
18434 ultrasparc, ultrasparc3, niagara, niagara2, niagara3, niagara4
18437 By default (unless configured otherwise), GCC generates code for the V7
18438 variant of the SPARC architecture. With @option{-mcpu=cypress}, the compiler
18439 additionally optimizes it for the Cypress CY7C602 chip, as used in the
18440 SPARCStation/SPARCServer 3xx series. This is also appropriate for the older
18441 SPARCStation 1, 2, IPX etc.
18443 With @option{-mcpu=v8}, GCC generates code for the V8 variant of the SPARC
18444 architecture. The only difference from V7 code is that the compiler emits
18445 the integer multiply and integer divide instructions which exist in SPARC-V8
18446 but not in SPARC-V7. With @option{-mcpu=supersparc}, the compiler additionally
18447 optimizes it for the SuperSPARC chip, as used in the SPARCStation 10, 1000 and
18450 With @option{-mcpu=sparclite}, GCC generates code for the SPARClite variant of
18451 the SPARC architecture. This adds the integer multiply, integer divide step
18452 and scan (@code{ffs}) instructions which exist in SPARClite but not in SPARC-V7.
18453 With @option{-mcpu=f930}, the compiler additionally optimizes it for the
18454 Fujitsu MB86930 chip, which is the original SPARClite, with no FPU@. With
18455 @option{-mcpu=f934}, the compiler additionally optimizes it for the Fujitsu
18456 MB86934 chip, which is the more recent SPARClite with FPU@.
18458 With @option{-mcpu=sparclet}, GCC generates code for the SPARClet variant of
18459 the SPARC architecture. This adds the integer multiply, multiply/accumulate,
18460 integer divide step and scan (@code{ffs}) instructions which exist in SPARClet
18461 but not in SPARC-V7. With @option{-mcpu=tsc701}, the compiler additionally
18462 optimizes it for the TEMIC SPARClet chip.
18464 With @option{-mcpu=v9}, GCC generates code for the V9 variant of the SPARC
18465 architecture. This adds 64-bit integer and floating-point move instructions,
18466 3 additional floating-point condition code registers and conditional move
18467 instructions. With @option{-mcpu=ultrasparc}, the compiler additionally
18468 optimizes it for the Sun UltraSPARC I/II/IIi chips. With
18469 @option{-mcpu=ultrasparc3}, the compiler additionally optimizes it for the
18470 Sun UltraSPARC III/III+/IIIi/IIIi+/IV/IV+ chips. With
18471 @option{-mcpu=niagara}, the compiler additionally optimizes it for
18472 Sun UltraSPARC T1 chips. With @option{-mcpu=niagara2}, the compiler
18473 additionally optimizes it for Sun UltraSPARC T2 chips. With
18474 @option{-mcpu=niagara3}, the compiler additionally optimizes it for Sun
18475 UltraSPARC T3 chips. With @option{-mcpu=niagara4}, the compiler
18476 additionally optimizes it for Sun UltraSPARC T4 chips.
18478 @item -mtune=@var{cpu_type}
18480 Set the instruction scheduling parameters for machine type
18481 @var{cpu_type}, but do not set the instruction set or register set that the
18482 option @option{-mcpu=@var{cpu_type}} would.
18484 The same values for @option{-mcpu=@var{cpu_type}} can be used for
18485 @option{-mtune=@var{cpu_type}}, but the only useful values are those
18486 that select a particular CPU implementation. Those are @samp{cypress},
18487 @samp{supersparc}, @samp{hypersparc}, @samp{leon}, @samp{f930}, @samp{f934},
18488 @samp{sparclite86x}, @samp{tsc701}, @samp{ultrasparc}, @samp{ultrasparc3},
18489 @samp{niagara}, @samp{niagara2}, @samp{niagara3} and @samp{niagara4}. With
18490 native Solaris and GNU/Linux toolchains, @samp{native} can also be used.
18495 @opindex mno-v8plus
18496 With @option{-mv8plus}, GCC generates code for the SPARC-V8+ ABI@. The
18497 difference from the V8 ABI is that the global and out registers are
18498 considered 64 bits wide. This is enabled by default on Solaris in 32-bit
18499 mode for all SPARC-V9 processors.
18505 With @option{-mvis}, GCC generates code that takes advantage of the UltraSPARC
18506 Visual Instruction Set extensions. The default is @option{-mno-vis}.
18512 With @option{-mvis2}, GCC generates code that takes advantage of
18513 version 2.0 of the UltraSPARC Visual Instruction Set extensions. The
18514 default is @option{-mvis2} when targetting a cpu that supports such
18515 instructions, such as UltraSPARC-III and later. Setting @option{-mvis2}
18516 also sets @option{-mvis}.
18522 With @option{-mvis3}, GCC generates code that takes advantage of
18523 version 3.0 of the UltraSPARC Visual Instruction Set extensions. The
18524 default is @option{-mvis3} when targetting a cpu that supports such
18525 instructions, such as niagara-3 and later. Setting @option{-mvis3}
18526 also sets @option{-mvis2} and @option{-mvis}.
18532 With @option{-mpopc}, GCC generates code that takes advantage of the UltraSPARC
18533 population count instruction. The default is @option{-mpopc}
18534 when targetting a cpu that supports such instructions, such as Niagara-2 and
18541 With @option{-mfmaf}, GCC generates code that takes advantage of the UltraSPARC
18542 Fused Multiply-Add Floating-point extensions. The default is @option{-mfmaf}
18543 when targetting a cpu that supports such instructions, such as Niagara-3 and
18547 @opindex mfix-at697f
18548 Enable the documented workaround for the single erratum of the Atmel AT697F
18549 processor (which corresponds to erratum #13 of the AT697E processor).
18552 These @samp{-m} options are supported in addition to the above
18553 on SPARC-V9 processors in 64-bit environments:
18556 @item -mlittle-endian
18557 @opindex mlittle-endian
18558 Generate code for a processor running in little-endian mode. It is only
18559 available for a few configurations and most notably not on Solaris and Linux.
18565 Generate code for a 32-bit or 64-bit environment.
18566 The 32-bit environment sets int, long and pointer to 32 bits.
18567 The 64-bit environment sets int to 32 bits and long and pointer
18570 @item -mcmodel=@var{which}
18572 Set the code model to one of
18576 The Medium/Low code model: 64-bit addresses, programs
18577 must be linked in the low 32 bits of memory. Programs can be statically
18578 or dynamically linked.
18581 The Medium/Middle code model: 64-bit addresses, programs
18582 must be linked in the low 44 bits of memory, the text and data segments must
18583 be less than 2GB in size and the data segment must be located within 2GB of
18587 The Medium/Anywhere code model: 64-bit addresses, programs
18588 may be linked anywhere in memory, the text and data segments must be less
18589 than 2GB in size and the data segment must be located within 2GB of the
18593 The Medium/Anywhere code model for embedded systems:
18594 64-bit addresses, the text and data segments must be less than 2GB in
18595 size, both starting anywhere in memory (determined at link time). The
18596 global register %g4 points to the base of the data segment. Programs
18597 are statically linked and PIC is not supported.
18600 @item -mmemory-model=@var{mem-model}
18601 @opindex mmemory-model
18602 Set the memory model in force on the processor to one of
18606 The default memory model for the processor and operating system.
18609 Relaxed Memory Order
18612 Partial Store Order
18618 Sequential Consistency
18621 These memory models are formally defined in Appendix D of the Sparc V9
18622 architecture manual, as set in the processor's @code{PSTATE.MM} field.
18625 @itemx -mno-stack-bias
18626 @opindex mstack-bias
18627 @opindex mno-stack-bias
18628 With @option{-mstack-bias}, GCC assumes that the stack pointer, and
18629 frame pointer if present, are offset by @minus{}2047 which must be added back
18630 when making stack frame references. This is the default in 64-bit mode.
18631 Otherwise, assume no such offset is present.
18635 @subsection SPU Options
18636 @cindex SPU options
18638 These @samp{-m} options are supported on the SPU:
18642 @itemx -merror-reloc
18643 @opindex mwarn-reloc
18644 @opindex merror-reloc
18646 The loader for SPU does not handle dynamic relocations. By default, GCC
18647 will give an error when it generates code that requires a dynamic
18648 relocation. @option{-mno-error-reloc} disables the error,
18649 @option{-mwarn-reloc} will generate a warning instead.
18652 @itemx -munsafe-dma
18654 @opindex munsafe-dma
18656 Instructions that initiate or test completion of DMA must not be
18657 reordered with respect to loads and stores of the memory that is being
18658 accessed. Users typically address this problem using the volatile
18659 keyword, but that can lead to inefficient code in places where the
18660 memory is known to not change. Rather than mark the memory as volatile
18661 we treat the DMA instructions as potentially effecting all memory. With
18662 @option{-munsafe-dma} users must use the volatile keyword to protect
18665 @item -mbranch-hints
18666 @opindex mbranch-hints
18668 By default, GCC will generate a branch hint instruction to avoid
18669 pipeline stalls for always taken or probably taken branches. A hint
18670 will not be generated closer than 8 instructions away from its branch.
18671 There is little reason to disable them, except for debugging purposes,
18672 or to make an object a little bit smaller.
18676 @opindex msmall-mem
18677 @opindex mlarge-mem
18679 By default, GCC generates code assuming that addresses are never larger
18680 than 18 bits. With @option{-mlarge-mem} code is generated that assumes
18681 a full 32-bit address.
18686 By default, GCC links against startup code that assumes the SPU-style
18687 main function interface (which has an unconventional parameter list).
18688 With @option{-mstdmain}, GCC will link your program against startup
18689 code that assumes a C99-style interface to @code{main}, including a
18690 local copy of @code{argv} strings.
18692 @item -mfixed-range=@var{register-range}
18693 @opindex mfixed-range
18694 Generate code treating the given register range as fixed registers.
18695 A fixed register is one that the register allocator can not use. This is
18696 useful when compiling kernel code. A register range is specified as
18697 two registers separated by a dash. Multiple register ranges can be
18698 specified separated by a comma.
18704 Compile code assuming that pointers to the PPU address space accessed
18705 via the @code{__ea} named address space qualifier are either 32 or 64
18706 bits wide. The default is 32 bits. As this is an ABI changing option,
18707 all object code in an executable must be compiled with the same setting.
18709 @item -maddress-space-conversion
18710 @itemx -mno-address-space-conversion
18711 @opindex maddress-space-conversion
18712 @opindex mno-address-space-conversion
18713 Allow/disallow treating the @code{__ea} address space as superset
18714 of the generic address space. This enables explicit type casts
18715 between @code{__ea} and generic pointer as well as implicit
18716 conversions of generic pointers to @code{__ea} pointers. The
18717 default is to allow address space pointer conversions.
18719 @item -mcache-size=@var{cache-size}
18720 @opindex mcache-size
18721 This option controls the version of libgcc that the compiler links to an
18722 executable and selects a software-managed cache for accessing variables
18723 in the @code{__ea} address space with a particular cache size. Possible
18724 options for @var{cache-size} are @samp{8}, @samp{16}, @samp{32}, @samp{64}
18725 and @samp{128}. The default cache size is 64KB.
18727 @item -matomic-updates
18728 @itemx -mno-atomic-updates
18729 @opindex matomic-updates
18730 @opindex mno-atomic-updates
18731 This option controls the version of libgcc that the compiler links to an
18732 executable and selects whether atomic updates to the software-managed
18733 cache of PPU-side variables are used. If you use atomic updates, changes
18734 to a PPU variable from SPU code using the @code{__ea} named address space
18735 qualifier will not interfere with changes to other PPU variables residing
18736 in the same cache line from PPU code. If you do not use atomic updates,
18737 such interference may occur; however, writing back cache lines will be
18738 more efficient. The default behavior is to use atomic updates.
18741 @itemx -mdual-nops=@var{n}
18742 @opindex mdual-nops
18743 By default, GCC will insert nops to increase dual issue when it expects
18744 it to increase performance. @var{n} can be a value from 0 to 10. A
18745 smaller @var{n} will insert fewer nops. 10 is the default, 0 is the
18746 same as @option{-mno-dual-nops}. Disabled with @option{-Os}.
18748 @item -mhint-max-nops=@var{n}
18749 @opindex mhint-max-nops
18750 Maximum number of nops to insert for a branch hint. A branch hint must
18751 be at least 8 instructions away from the branch it is effecting. GCC
18752 will insert up to @var{n} nops to enforce this, otherwise it will not
18753 generate the branch hint.
18755 @item -mhint-max-distance=@var{n}
18756 @opindex mhint-max-distance
18757 The encoding of the branch hint instruction limits the hint to be within
18758 256 instructions of the branch it is effecting. By default, GCC makes
18759 sure it is within 125.
18762 @opindex msafe-hints
18763 Work around a hardware bug that causes the SPU to stall indefinitely.
18764 By default, GCC will insert the @code{hbrp} instruction to make sure
18765 this stall won't happen.
18769 @node System V Options
18770 @subsection Options for System V
18772 These additional options are available on System V Release 4 for
18773 compatibility with other compilers on those systems:
18778 Create a shared object.
18779 It is recommended that @option{-symbolic} or @option{-shared} be used instead.
18783 Identify the versions of each tool used by the compiler, in a
18784 @code{.ident} assembler directive in the output.
18788 Refrain from adding @code{.ident} directives to the output file (this is
18791 @item -YP,@var{dirs}
18793 Search the directories @var{dirs}, and no others, for libraries
18794 specified with @option{-l}.
18796 @item -Ym,@var{dir}
18798 Look in the directory @var{dir} to find the M4 preprocessor.
18799 The assembler uses this option.
18800 @c This is supposed to go with a -Yd for predefined M4 macro files, but
18801 @c the generic assembler that comes with Solaris takes just -Ym.
18804 @node TILE-Gx Options
18805 @subsection TILE-Gx Options
18806 @cindex TILE-Gx options
18808 These @samp{-m} options are supported on the TILE-Gx:
18811 @item -mcpu=@var{name}
18813 Selects the type of CPU to be targeted. Currently the only supported
18814 type is @samp{tilegx}.
18820 Generate code for a 32-bit or 64-bit environment. The 32-bit
18821 environment sets int, long, and pointer to 32 bits. The 64-bit
18822 environment sets int to 32 bits and long and pointer to 64 bits.
18825 @node TILEPro Options
18826 @subsection TILEPro Options
18827 @cindex TILEPro options
18829 These @samp{-m} options are supported on the TILEPro:
18832 @item -mcpu=@var{name}
18834 Selects the type of CPU to be targeted. Currently the only supported
18835 type is @samp{tilepro}.
18839 Generate code for a 32-bit environment, which sets int, long, and
18840 pointer to 32 bits. This is the only supported behavior so the flag
18841 is essentially ignored.
18845 @subsection V850 Options
18846 @cindex V850 Options
18848 These @samp{-m} options are defined for V850 implementations:
18852 @itemx -mno-long-calls
18853 @opindex mlong-calls
18854 @opindex mno-long-calls
18855 Treat all calls as being far away (near). If calls are assumed to be
18856 far away, the compiler will always load the functions address up into a
18857 register, and call indirect through the pointer.
18863 Do not optimize (do optimize) basic blocks that use the same index
18864 pointer 4 or more times to copy pointer into the @code{ep} register, and
18865 use the shorter @code{sld} and @code{sst} instructions. The @option{-mep}
18866 option is on by default if you optimize.
18868 @item -mno-prolog-function
18869 @itemx -mprolog-function
18870 @opindex mno-prolog-function
18871 @opindex mprolog-function
18872 Do not use (do use) external functions to save and restore registers
18873 at the prologue and epilogue of a function. The external functions
18874 are slower, but use less code space if more than one function saves
18875 the same number of registers. The @option{-mprolog-function} option
18876 is on by default if you optimize.
18880 Try to make the code as small as possible. At present, this just turns
18881 on the @option{-mep} and @option{-mprolog-function} options.
18883 @item -mtda=@var{n}
18885 Put static or global variables whose size is @var{n} bytes or less into
18886 the tiny data area that register @code{ep} points to. The tiny data
18887 area can hold up to 256 bytes in total (128 bytes for byte references).
18889 @item -msda=@var{n}
18891 Put static or global variables whose size is @var{n} bytes or less into
18892 the small data area that register @code{gp} points to. The small data
18893 area can hold up to 64 kilobytes.
18895 @item -mzda=@var{n}
18897 Put static or global variables whose size is @var{n} bytes or less into
18898 the first 32 kilobytes of memory.
18902 Specify that the target processor is the V850.
18905 @opindex mbig-switch
18906 Generate code suitable for big switch tables. Use this option only if
18907 the assembler/linker complain about out of range branches within a switch
18912 This option will cause r2 and r5 to be used in the code generated by
18913 the compiler. This setting is the default.
18915 @item -mno-app-regs
18916 @opindex mno-app-regs
18917 This option will cause r2 and r5 to be treated as fixed registers.
18921 Specify that the target processor is the V850E2V3. The preprocessor
18922 constants @samp{__v850e2v3__} will be defined if
18923 this option is used.
18927 Specify that the target processor is the V850E2. The preprocessor
18928 constants @samp{__v850e2__} will be defined if this option is used.
18932 Specify that the target processor is the V850E1. The preprocessor
18933 constants @samp{__v850e1__} and @samp{__v850e__} will be defined if
18934 this option is used.
18938 Specify that the target processor is the V850ES. This is an alias for
18939 the @option{-mv850e1} option.
18943 Specify that the target processor is the V850E@. The preprocessor
18944 constant @samp{__v850e__} will be defined if this option is used.
18946 If neither @option{-mv850} nor @option{-mv850e} nor @option{-mv850e1}
18947 nor @option{-mv850e2} nor @option{-mv850e2v3}
18948 are defined then a default target processor will be chosen and the
18949 relevant @samp{__v850*__} preprocessor constant will be defined.
18951 The preprocessor constants @samp{__v850} and @samp{__v851__} are always
18952 defined, regardless of which processor variant is the target.
18954 @item -mdisable-callt
18955 @opindex mdisable-callt
18956 This option will suppress generation of the CALLT instruction for the
18957 v850e, v850e1, v850e2 and v850e2v3 flavors of the v850 architecture. The default is
18958 @option{-mno-disable-callt} which allows the CALLT instruction to be used.
18963 @subsection VAX Options
18964 @cindex VAX options
18966 These @samp{-m} options are defined for the VAX:
18971 Do not output certain jump instructions (@code{aobleq} and so on)
18972 that the Unix assembler for the VAX cannot handle across long
18977 Do output those jump instructions, on the assumption that you
18978 will assemble with the GNU assembler.
18982 Output code for G-format floating-point numbers instead of D-format.
18985 @node VxWorks Options
18986 @subsection VxWorks Options
18987 @cindex VxWorks Options
18989 The options in this section are defined for all VxWorks targets.
18990 Options specific to the target hardware are listed with the other
18991 options for that target.
18996 GCC can generate code for both VxWorks kernels and real time processes
18997 (RTPs). This option switches from the former to the latter. It also
18998 defines the preprocessor macro @code{__RTP__}.
19001 @opindex non-static
19002 Link an RTP executable against shared libraries rather than static
19003 libraries. The options @option{-static} and @option{-shared} can
19004 also be used for RTPs (@pxref{Link Options}); @option{-static}
19011 These options are passed down to the linker. They are defined for
19012 compatibility with Diab.
19015 @opindex Xbind-lazy
19016 Enable lazy binding of function calls. This option is equivalent to
19017 @option{-Wl,-z,now} and is defined for compatibility with Diab.
19021 Disable lazy binding of function calls. This option is the default and
19022 is defined for compatibility with Diab.
19025 @node x86-64 Options
19026 @subsection x86-64 Options
19027 @cindex x86-64 options
19029 These are listed under @xref{i386 and x86-64 Options}.
19031 @node Xstormy16 Options
19032 @subsection Xstormy16 Options
19033 @cindex Xstormy16 Options
19035 These options are defined for Xstormy16:
19040 Choose startup files and linker script suitable for the simulator.
19043 @node Xtensa Options
19044 @subsection Xtensa Options
19045 @cindex Xtensa Options
19047 These options are supported for Xtensa targets:
19051 @itemx -mno-const16
19053 @opindex mno-const16
19054 Enable or disable use of @code{CONST16} instructions for loading
19055 constant values. The @code{CONST16} instruction is currently not a
19056 standard option from Tensilica. When enabled, @code{CONST16}
19057 instructions are always used in place of the standard @code{L32R}
19058 instructions. The use of @code{CONST16} is enabled by default only if
19059 the @code{L32R} instruction is not available.
19062 @itemx -mno-fused-madd
19063 @opindex mfused-madd
19064 @opindex mno-fused-madd
19065 Enable or disable use of fused multiply/add and multiply/subtract
19066 instructions in the floating-point option. This has no effect if the
19067 floating-point option is not also enabled. Disabling fused multiply/add
19068 and multiply/subtract instructions forces the compiler to use separate
19069 instructions for the multiply and add/subtract operations. This may be
19070 desirable in some cases where strict IEEE 754-compliant results are
19071 required: the fused multiply add/subtract instructions do not round the
19072 intermediate result, thereby producing results with @emph{more} bits of
19073 precision than specified by the IEEE standard. Disabling fused multiply
19074 add/subtract instructions also ensures that the program output is not
19075 sensitive to the compiler's ability to combine multiply and add/subtract
19078 @item -mserialize-volatile
19079 @itemx -mno-serialize-volatile
19080 @opindex mserialize-volatile
19081 @opindex mno-serialize-volatile
19082 When this option is enabled, GCC inserts @code{MEMW} instructions before
19083 @code{volatile} memory references to guarantee sequential consistency.
19084 The default is @option{-mserialize-volatile}. Use
19085 @option{-mno-serialize-volatile} to omit the @code{MEMW} instructions.
19087 @item -mforce-no-pic
19088 @opindex mforce-no-pic
19089 For targets, like GNU/Linux, where all user-mode Xtensa code must be
19090 position-independent code (PIC), this option disables PIC for compiling
19093 @item -mtext-section-literals
19094 @itemx -mno-text-section-literals
19095 @opindex mtext-section-literals
19096 @opindex mno-text-section-literals
19097 Control the treatment of literal pools. The default is
19098 @option{-mno-text-section-literals}, which places literals in a separate
19099 section in the output file. This allows the literal pool to be placed
19100 in a data RAM/ROM, and it also allows the linker to combine literal
19101 pools from separate object files to remove redundant literals and
19102 improve code size. With @option{-mtext-section-literals}, the literals
19103 are interspersed in the text section in order to keep them as close as
19104 possible to their references. This may be necessary for large assembly
19107 @item -mtarget-align
19108 @itemx -mno-target-align
19109 @opindex mtarget-align
19110 @opindex mno-target-align
19111 When this option is enabled, GCC instructs the assembler to
19112 automatically align instructions to reduce branch penalties at the
19113 expense of some code density. The assembler attempts to widen density
19114 instructions to align branch targets and the instructions following call
19115 instructions. If there are not enough preceding safe density
19116 instructions to align a target, no widening will be performed. The
19117 default is @option{-mtarget-align}. These options do not affect the
19118 treatment of auto-aligned instructions like @code{LOOP}, which the
19119 assembler will always align, either by widening density instructions or
19120 by inserting no-op instructions.
19123 @itemx -mno-longcalls
19124 @opindex mlongcalls
19125 @opindex mno-longcalls
19126 When this option is enabled, GCC instructs the assembler to translate
19127 direct calls to indirect calls unless it can determine that the target
19128 of a direct call is in the range allowed by the call instruction. This
19129 translation typically occurs for calls to functions in other source
19130 files. Specifically, the assembler translates a direct @code{CALL}
19131 instruction into an @code{L32R} followed by a @code{CALLX} instruction.
19132 The default is @option{-mno-longcalls}. This option should be used in
19133 programs where the call target can potentially be out of range. This
19134 option is implemented in the assembler, not the compiler, so the
19135 assembly code generated by GCC will still show direct call
19136 instructions---look at the disassembled object code to see the actual
19137 instructions. Note that the assembler will use an indirect call for
19138 every cross-file call, not just those that really will be out of range.
19141 @node zSeries Options
19142 @subsection zSeries Options
19143 @cindex zSeries options
19145 These are listed under @xref{S/390 and zSeries Options}.
19147 @node Code Gen Options
19148 @section Options for Code Generation Conventions
19149 @cindex code generation conventions
19150 @cindex options, code generation
19151 @cindex run-time options
19153 These machine-independent options control the interface conventions
19154 used in code generation.
19156 Most of them have both positive and negative forms; the negative form
19157 of @option{-ffoo} would be @option{-fno-foo}. In the table below, only
19158 one of the forms is listed---the one that is not the default. You
19159 can figure out the other form by either removing @samp{no-} or adding
19163 @item -fbounds-check
19164 @opindex fbounds-check
19165 For front ends that support it, generate additional code to check that
19166 indices used to access arrays are within the declared range. This is
19167 currently only supported by the Java and Fortran front ends, where
19168 this option defaults to true and false respectively.
19172 This option generates traps for signed overflow on addition, subtraction,
19173 multiplication operations.
19177 This option instructs the compiler to assume that signed arithmetic
19178 overflow of addition, subtraction and multiplication wraps around
19179 using twos-complement representation. This flag enables some optimizations
19180 and disables others. This option is enabled by default for the Java
19181 front end, as required by the Java language specification.
19184 @opindex fexceptions
19185 Enable exception handling. Generates extra code needed to propagate
19186 exceptions. For some targets, this implies GCC will generate frame
19187 unwind information for all functions, which can produce significant data
19188 size overhead, although it does not affect execution. If you do not
19189 specify this option, GCC will enable it by default for languages like
19190 C++ that normally require exception handling, and disable it for
19191 languages like C that do not normally require it. However, you may need
19192 to enable this option when compiling C code that needs to interoperate
19193 properly with exception handlers written in C++. You may also wish to
19194 disable this option if you are compiling older C++ programs that don't
19195 use exception handling.
19197 @item -fnon-call-exceptions
19198 @opindex fnon-call-exceptions
19199 Generate code that allows trapping instructions to throw exceptions.
19200 Note that this requires platform-specific runtime support that does
19201 not exist everywhere. Moreover, it only allows @emph{trapping}
19202 instructions to throw exceptions, i.e.@: memory references or floating-point
19203 instructions. It does not allow exceptions to be thrown from
19204 arbitrary signal handlers such as @code{SIGALRM}.
19206 @item -funwind-tables
19207 @opindex funwind-tables
19208 Similar to @option{-fexceptions}, except that it will just generate any needed
19209 static data, but will not affect the generated code in any other way.
19210 You will normally not enable this option; instead, a language processor
19211 that needs this handling would enable it on your behalf.
19213 @item -fasynchronous-unwind-tables
19214 @opindex fasynchronous-unwind-tables
19215 Generate unwind table in dwarf2 format, if supported by target machine. The
19216 table is exact at each instruction boundary, so it can be used for stack
19217 unwinding from asynchronous events (such as debugger or garbage collector).
19219 @item -fpcc-struct-return
19220 @opindex fpcc-struct-return
19221 Return ``short'' @code{struct} and @code{union} values in memory like
19222 longer ones, rather than in registers. This convention is less
19223 efficient, but it has the advantage of allowing intercallability between
19224 GCC-compiled files and files compiled with other compilers, particularly
19225 the Portable C Compiler (pcc).
19227 The precise convention for returning structures in memory depends
19228 on the target configuration macros.
19230 Short structures and unions are those whose size and alignment match
19231 that of some integer type.
19233 @strong{Warning:} code compiled with the @option{-fpcc-struct-return}
19234 switch is not binary compatible with code compiled with the
19235 @option{-freg-struct-return} switch.
19236 Use it to conform to a non-default application binary interface.
19238 @item -freg-struct-return
19239 @opindex freg-struct-return
19240 Return @code{struct} and @code{union} values in registers when possible.
19241 This is more efficient for small structures than
19242 @option{-fpcc-struct-return}.
19244 If you specify neither @option{-fpcc-struct-return} nor
19245 @option{-freg-struct-return}, GCC defaults to whichever convention is
19246 standard for the target. If there is no standard convention, GCC
19247 defaults to @option{-fpcc-struct-return}, except on targets where GCC is
19248 the principal compiler. In those cases, we can choose the standard, and
19249 we chose the more efficient register return alternative.
19251 @strong{Warning:} code compiled with the @option{-freg-struct-return}
19252 switch is not binary compatible with code compiled with the
19253 @option{-fpcc-struct-return} switch.
19254 Use it to conform to a non-default application binary interface.
19256 @item -fshort-enums
19257 @opindex fshort-enums
19258 Allocate to an @code{enum} type only as many bytes as it needs for the
19259 declared range of possible values. Specifically, the @code{enum} type
19260 will be equivalent to the smallest integer type that has enough room.
19262 @strong{Warning:} the @option{-fshort-enums} switch causes GCC to generate
19263 code that is not binary compatible with code generated without that switch.
19264 Use it to conform to a non-default application binary interface.
19266 @item -fshort-double
19267 @opindex fshort-double
19268 Use the same size for @code{double} as for @code{float}.
19270 @strong{Warning:} the @option{-fshort-double} switch causes GCC to generate
19271 code that is not binary compatible with code generated without that switch.
19272 Use it to conform to a non-default application binary interface.
19274 @item -fshort-wchar
19275 @opindex fshort-wchar
19276 Override the underlying type for @samp{wchar_t} to be @samp{short
19277 unsigned int} instead of the default for the target. This option is
19278 useful for building programs to run under WINE@.
19280 @strong{Warning:} the @option{-fshort-wchar} switch causes GCC to generate
19281 code that is not binary compatible with code generated without that switch.
19282 Use it to conform to a non-default application binary interface.
19285 @opindex fno-common
19286 In C code, controls the placement of uninitialized global variables.
19287 Unix C compilers have traditionally permitted multiple definitions of
19288 such variables in different compilation units by placing the variables
19290 This is the behavior specified by @option{-fcommon}, and is the default
19291 for GCC on most targets.
19292 On the other hand, this behavior is not required by ISO C, and on some
19293 targets may carry a speed or code size penalty on variable references.
19294 The @option{-fno-common} option specifies that the compiler should place
19295 uninitialized global variables in the data section of the object file,
19296 rather than generating them as common blocks.
19297 This has the effect that if the same variable is declared
19298 (without @code{extern}) in two different compilations,
19299 you will get a multiple-definition error when you link them.
19300 In this case, you must compile with @option{-fcommon} instead.
19301 Compiling with @option{-fno-common} is useful on targets for which
19302 it provides better performance, or if you wish to verify that the
19303 program will work on other systems that always treat uninitialized
19304 variable declarations this way.
19308 Ignore the @samp{#ident} directive.
19310 @item -finhibit-size-directive
19311 @opindex finhibit-size-directive
19312 Don't output a @code{.size} assembler directive, or anything else that
19313 would cause trouble if the function is split in the middle, and the
19314 two halves are placed at locations far apart in memory. This option is
19315 used when compiling @file{crtstuff.c}; you should not need to use it
19318 @item -fverbose-asm
19319 @opindex fverbose-asm
19320 Put extra commentary information in the generated assembly code to
19321 make it more readable. This option is generally only of use to those
19322 who actually need to read the generated assembly code (perhaps while
19323 debugging the compiler itself).
19325 @option{-fno-verbose-asm}, the default, causes the
19326 extra information to be omitted and is useful when comparing two assembler
19329 @item -frecord-gcc-switches
19330 @opindex frecord-gcc-switches
19331 This switch causes the command line that was used to invoke the
19332 compiler to be recorded into the object file that is being created.
19333 This switch is only implemented on some targets and the exact format
19334 of the recording is target and binary file format dependent, but it
19335 usually takes the form of a section containing ASCII text. This
19336 switch is related to the @option{-fverbose-asm} switch, but that
19337 switch only records information in the assembler output file as
19338 comments, so it never reaches the object file.
19339 See also @option{-grecord-gcc-switches} for another
19340 way of storing compiler options into the object file.
19344 @cindex global offset table
19346 Generate position-independent code (PIC) suitable for use in a shared
19347 library, if supported for the target machine. Such code accesses all
19348 constant addresses through a global offset table (GOT)@. The dynamic
19349 loader resolves the GOT entries when the program starts (the dynamic
19350 loader is not part of GCC; it is part of the operating system). If
19351 the GOT size for the linked executable exceeds a machine-specific
19352 maximum size, you get an error message from the linker indicating that
19353 @option{-fpic} does not work; in that case, recompile with @option{-fPIC}
19354 instead. (These maximums are 8k on the SPARC and 32k
19355 on the m68k and RS/6000. The 386 has no such limit.)
19357 Position-independent code requires special support, and therefore works
19358 only on certain machines. For the 386, GCC supports PIC for System V
19359 but not for the Sun 386i. Code generated for the IBM RS/6000 is always
19360 position-independent.
19362 When this flag is set, the macros @code{__pic__} and @code{__PIC__}
19367 If supported for the target machine, emit position-independent code,
19368 suitable for dynamic linking and avoiding any limit on the size of the
19369 global offset table. This option makes a difference on the m68k,
19370 PowerPC and SPARC@.
19372 Position-independent code requires special support, and therefore works
19373 only on certain machines.
19375 When this flag is set, the macros @code{__pic__} and @code{__PIC__}
19382 These options are similar to @option{-fpic} and @option{-fPIC}, but
19383 generated position independent code can be only linked into executables.
19384 Usually these options are used when @option{-pie} GCC option will be
19385 used during linking.
19387 @option{-fpie} and @option{-fPIE} both define the macros
19388 @code{__pie__} and @code{__PIE__}. The macros have the value 1
19389 for @option{-fpie} and 2 for @option{-fPIE}.
19391 @item -fno-jump-tables
19392 @opindex fno-jump-tables
19393 Do not use jump tables for switch statements even where it would be
19394 more efficient than other code generation strategies. This option is
19395 of use in conjunction with @option{-fpic} or @option{-fPIC} for
19396 building code that forms part of a dynamic linker and cannot
19397 reference the address of a jump table. On some targets, jump tables
19398 do not require a GOT and this option is not needed.
19400 @item -ffixed-@var{reg}
19402 Treat the register named @var{reg} as a fixed register; generated code
19403 should never refer to it (except perhaps as a stack pointer, frame
19404 pointer or in some other fixed role).
19406 @var{reg} must be the name of a register. The register names accepted
19407 are machine-specific and are defined in the @code{REGISTER_NAMES}
19408 macro in the machine description macro file.
19410 This flag does not have a negative form, because it specifies a
19413 @item -fcall-used-@var{reg}
19414 @opindex fcall-used
19415 Treat the register named @var{reg} as an allocable register that is
19416 clobbered by function calls. It may be allocated for temporaries or
19417 variables that do not live across a call. Functions compiled this way
19418 will not save and restore the register @var{reg}.
19420 It is an error to used this flag with the frame pointer or stack pointer.
19421 Use of this flag for other registers that have fixed pervasive roles in
19422 the machine's execution model will produce disastrous results.
19424 This flag does not have a negative form, because it specifies a
19427 @item -fcall-saved-@var{reg}
19428 @opindex fcall-saved
19429 Treat the register named @var{reg} as an allocable register saved by
19430 functions. It may be allocated even for temporaries or variables that
19431 live across a call. Functions compiled this way will save and restore
19432 the register @var{reg} if they use it.
19434 It is an error to used this flag with the frame pointer or stack pointer.
19435 Use of this flag for other registers that have fixed pervasive roles in
19436 the machine's execution model will produce disastrous results.
19438 A different sort of disaster will result from the use of this flag for
19439 a register in which function values may be returned.
19441 This flag does not have a negative form, because it specifies a
19444 @item -fpack-struct[=@var{n}]
19445 @opindex fpack-struct
19446 Without a value specified, pack all structure members together without
19447 holes. When a value is specified (which must be a small power of two), pack
19448 structure members according to this value, representing the maximum
19449 alignment (that is, objects with default alignment requirements larger than
19450 this will be output potentially unaligned at the next fitting location.
19452 @strong{Warning:} the @option{-fpack-struct} switch causes GCC to generate
19453 code that is not binary compatible with code generated without that switch.
19454 Additionally, it makes the code suboptimal.
19455 Use it to conform to a non-default application binary interface.
19457 @item -finstrument-functions
19458 @opindex finstrument-functions
19459 Generate instrumentation calls for entry and exit to functions. Just
19460 after function entry and just before function exit, the following
19461 profiling functions will be called with the address of the current
19462 function and its call site. (On some platforms,
19463 @code{__builtin_return_address} does not work beyond the current
19464 function, so the call site information may not be available to the
19465 profiling functions otherwise.)
19468 void __cyg_profile_func_enter (void *this_fn,
19470 void __cyg_profile_func_exit (void *this_fn,
19474 The first argument is the address of the start of the current function,
19475 which may be looked up exactly in the symbol table.
19477 This instrumentation is also done for functions expanded inline in other
19478 functions. The profiling calls will indicate where, conceptually, the
19479 inline function is entered and exited. This means that addressable
19480 versions of such functions must be available. If all your uses of a
19481 function are expanded inline, this may mean an additional expansion of
19482 code size. If you use @samp{extern inline} in your C code, an
19483 addressable version of such functions must be provided. (This is
19484 normally the case anyways, but if you get lucky and the optimizer always
19485 expands the functions inline, you might have gotten away without
19486 providing static copies.)
19488 A function may be given the attribute @code{no_instrument_function}, in
19489 which case this instrumentation will not be done. This can be used, for
19490 example, for the profiling functions listed above, high-priority
19491 interrupt routines, and any functions from which the profiling functions
19492 cannot safely be called (perhaps signal handlers, if the profiling
19493 routines generate output or allocate memory).
19495 @item -finstrument-functions-exclude-file-list=@var{file},@var{file},@dots{}
19496 @opindex finstrument-functions-exclude-file-list
19498 Set the list of functions that are excluded from instrumentation (see
19499 the description of @code{-finstrument-functions}). If the file that
19500 contains a function definition matches with one of @var{file}, then
19501 that function is not instrumented. The match is done on substrings:
19502 if the @var{file} parameter is a substring of the file name, it is
19503 considered to be a match.
19508 -finstrument-functions-exclude-file-list=/bits/stl,include/sys
19512 will exclude any inline function defined in files whose pathnames
19513 contain @code{/bits/stl} or @code{include/sys}.
19515 If, for some reason, you want to include letter @code{','} in one of
19516 @var{sym}, write @code{'\,'}. For example,
19517 @code{-finstrument-functions-exclude-file-list='\,\,tmp'}
19518 (note the single quote surrounding the option).
19520 @item -finstrument-functions-exclude-function-list=@var{sym},@var{sym},@dots{}
19521 @opindex finstrument-functions-exclude-function-list
19523 This is similar to @code{-finstrument-functions-exclude-file-list},
19524 but this option sets the list of function names to be excluded from
19525 instrumentation. The function name to be matched is its user-visible
19526 name, such as @code{vector<int> blah(const vector<int> &)}, not the
19527 internal mangled name (e.g., @code{_Z4blahRSt6vectorIiSaIiEE}). The
19528 match is done on substrings: if the @var{sym} parameter is a substring
19529 of the function name, it is considered to be a match. For C99 and C++
19530 extended identifiers, the function name must be given in UTF-8, not
19531 using universal character names.
19533 @item -fstack-check
19534 @opindex fstack-check
19535 Generate code to verify that you do not go beyond the boundary of the
19536 stack. You should specify this flag if you are running in an
19537 environment with multiple threads, but only rarely need to specify it in
19538 a single-threaded environment since stack overflow is automatically
19539 detected on nearly all systems if there is only one stack.
19541 Note that this switch does not actually cause checking to be done; the
19542 operating system or the language runtime must do that. The switch causes
19543 generation of code to ensure that they see the stack being extended.
19545 You can additionally specify a string parameter: @code{no} means no
19546 checking, @code{generic} means force the use of old-style checking,
19547 @code{specific} means use the best checking method and is equivalent
19548 to bare @option{-fstack-check}.
19550 Old-style checking is a generic mechanism that requires no specific
19551 target support in the compiler but comes with the following drawbacks:
19555 Modified allocation strategy for large objects: they will always be
19556 allocated dynamically if their size exceeds a fixed threshold.
19559 Fixed limit on the size of the static frame of functions: when it is
19560 topped by a particular function, stack checking is not reliable and
19561 a warning is issued by the compiler.
19564 Inefficiency: because of both the modified allocation strategy and the
19565 generic implementation, the performances of the code are hampered.
19568 Note that old-style stack checking is also the fallback method for
19569 @code{specific} if no target support has been added in the compiler.
19571 @item -fstack-limit-register=@var{reg}
19572 @itemx -fstack-limit-symbol=@var{sym}
19573 @itemx -fno-stack-limit
19574 @opindex fstack-limit-register
19575 @opindex fstack-limit-symbol
19576 @opindex fno-stack-limit
19577 Generate code to ensure that the stack does not grow beyond a certain value,
19578 either the value of a register or the address of a symbol. If the stack
19579 would grow beyond the value, a signal is raised. For most targets,
19580 the signal is raised before the stack overruns the boundary, so
19581 it is possible to catch the signal without taking special precautions.
19583 For instance, if the stack starts at absolute address @samp{0x80000000}
19584 and grows downwards, you can use the flags
19585 @option{-fstack-limit-symbol=__stack_limit} and
19586 @option{-Wl,--defsym,__stack_limit=0x7ffe0000} to enforce a stack limit
19587 of 128KB@. Note that this may only work with the GNU linker.
19589 @item -fsplit-stack
19590 @opindex fsplit-stack
19591 Generate code to automatically split the stack before it overflows.
19592 The resulting program has a discontiguous stack which can only
19593 overflow if the program is unable to allocate any more memory. This
19594 is most useful when running threaded programs, as it is no longer
19595 necessary to calculate a good stack size to use for each thread. This
19596 is currently only implemented for the i386 and x86_64 back ends running
19599 When code compiled with @option{-fsplit-stack} calls code compiled
19600 without @option{-fsplit-stack}, there may not be much stack space
19601 available for the latter code to run. If compiling all code,
19602 including library code, with @option{-fsplit-stack} is not an option,
19603 then the linker can fix up these calls so that the code compiled
19604 without @option{-fsplit-stack} always has a large stack. Support for
19605 this is implemented in the gold linker in GNU binutils release 2.21
19608 @item -fleading-underscore
19609 @opindex fleading-underscore
19610 This option and its counterpart, @option{-fno-leading-underscore}, forcibly
19611 change the way C symbols are represented in the object file. One use
19612 is to help link with legacy assembly code.
19614 @strong{Warning:} the @option{-fleading-underscore} switch causes GCC to
19615 generate code that is not binary compatible with code generated without that
19616 switch. Use it to conform to a non-default application binary interface.
19617 Not all targets provide complete support for this switch.
19619 @item -ftls-model=@var{model}
19620 @opindex ftls-model
19621 Alter the thread-local storage model to be used (@pxref{Thread-Local}).
19622 The @var{model} argument should be one of @code{global-dynamic},
19623 @code{local-dynamic}, @code{initial-exec} or @code{local-exec}.
19625 The default without @option{-fpic} is @code{initial-exec}; with
19626 @option{-fpic} the default is @code{global-dynamic}.
19628 @item -fvisibility=@var{default|internal|hidden|protected}
19629 @opindex fvisibility
19630 Set the default ELF image symbol visibility to the specified option---all
19631 symbols will be marked with this unless overridden within the code.
19632 Using this feature can very substantially improve linking and
19633 load times of shared object libraries, produce more optimized
19634 code, provide near-perfect API export and prevent symbol clashes.
19635 It is @strong{strongly} recommended that you use this in any shared objects
19638 Despite the nomenclature, @code{default} always means public; i.e.,
19639 available to be linked against from outside the shared object.
19640 @code{protected} and @code{internal} are pretty useless in real-world
19641 usage so the only other commonly used option will be @code{hidden}.
19642 The default if @option{-fvisibility} isn't specified is
19643 @code{default}, i.e., make every
19644 symbol public---this causes the same behavior as previous versions of
19647 A good explanation of the benefits offered by ensuring ELF
19648 symbols have the correct visibility is given by ``How To Write
19649 Shared Libraries'' by Ulrich Drepper (which can be found at
19650 @w{@uref{http://people.redhat.com/~drepper/}})---however a superior
19651 solution made possible by this option to marking things hidden when
19652 the default is public is to make the default hidden and mark things
19653 public. This is the norm with DLL's on Windows and with @option{-fvisibility=hidden}
19654 and @code{__attribute__ ((visibility("default")))} instead of
19655 @code{__declspec(dllexport)} you get almost identical semantics with
19656 identical syntax. This is a great boon to those working with
19657 cross-platform projects.
19659 For those adding visibility support to existing code, you may find
19660 @samp{#pragma GCC visibility} of use. This works by you enclosing
19661 the declarations you wish to set visibility for with (for example)
19662 @samp{#pragma GCC visibility push(hidden)} and
19663 @samp{#pragma GCC visibility pop}.
19664 Bear in mind that symbol visibility should be viewed @strong{as
19665 part of the API interface contract} and thus all new code should
19666 always specify visibility when it is not the default; i.e., declarations
19667 only for use within the local DSO should @strong{always} be marked explicitly
19668 as hidden as so to avoid PLT indirection overheads---making this
19669 abundantly clear also aids readability and self-documentation of the code.
19670 Note that due to ISO C++ specification requirements, operator new and
19671 operator delete must always be of default visibility.
19673 Be aware that headers from outside your project, in particular system
19674 headers and headers from any other library you use, may not be
19675 expecting to be compiled with visibility other than the default. You
19676 may need to explicitly say @samp{#pragma GCC visibility push(default)}
19677 before including any such headers.
19679 @samp{extern} declarations are not affected by @samp{-fvisibility}, so
19680 a lot of code can be recompiled with @samp{-fvisibility=hidden} with
19681 no modifications. However, this means that calls to @samp{extern}
19682 functions with no explicit visibility will use the PLT, so it is more
19683 effective to use @samp{__attribute ((visibility))} and/or
19684 @samp{#pragma GCC visibility} to tell the compiler which @samp{extern}
19685 declarations should be treated as hidden.
19687 Note that @samp{-fvisibility} does affect C++ vague linkage
19688 entities. This means that, for instance, an exception class that will
19689 be thrown between DSOs must be explicitly marked with default
19690 visibility so that the @samp{type_info} nodes will be unified between
19693 An overview of these techniques, their benefits and how to use them
19694 is at @uref{http://gcc.gnu.org/@/wiki/@/Visibility}.
19696 @item -fstrict-volatile-bitfields
19697 @opindex fstrict-volatile-bitfields
19698 This option should be used if accesses to volatile bit-fields (or other
19699 structure fields, although the compiler usually honors those types
19700 anyway) should use a single access of the width of the
19701 field's type, aligned to a natural alignment if possible. For
19702 example, targets with memory-mapped peripheral registers might require
19703 all such accesses to be 16 bits wide; with this flag the user could
19704 declare all peripheral bit-fields as ``unsigned short'' (assuming short
19705 is 16 bits on these targets) to force GCC to use 16-bit accesses
19706 instead of, perhaps, a more efficient 32-bit access.
19708 If this option is disabled, the compiler will use the most efficient
19709 instruction. In the previous example, that might be a 32-bit load
19710 instruction, even though that will access bytes that do not contain
19711 any portion of the bit-field, or memory-mapped registers unrelated to
19712 the one being updated.
19714 If the target requires strict alignment, and honoring the field
19715 type would require violating this alignment, a warning is issued.
19716 If the field has @code{packed} attribute, the access is done without
19717 honoring the field type. If the field doesn't have @code{packed}
19718 attribute, the access is done honoring the field type. In both cases,
19719 GCC assumes that the user knows something about the target hardware
19720 that it is unaware of.
19722 The default value of this option is determined by the application binary
19723 interface for the target processor.
19729 @node Environment Variables
19730 @section Environment Variables Affecting GCC
19731 @cindex environment variables
19733 @c man begin ENVIRONMENT
19734 This section describes several environment variables that affect how GCC
19735 operates. Some of them work by specifying directories or prefixes to use
19736 when searching for various kinds of files. Some are used to specify other
19737 aspects of the compilation environment.
19739 Note that you can also specify places to search using options such as
19740 @option{-B}, @option{-I} and @option{-L} (@pxref{Directory Options}). These
19741 take precedence over places specified using environment variables, which
19742 in turn take precedence over those specified by the configuration of GCC@.
19743 @xref{Driver,, Controlling the Compilation Driver @file{gcc}, gccint,
19744 GNU Compiler Collection (GCC) Internals}.
19749 @c @itemx LC_COLLATE
19751 @c @itemx LC_MONETARY
19752 @c @itemx LC_NUMERIC
19757 @c @findex LC_COLLATE
19758 @findex LC_MESSAGES
19759 @c @findex LC_MONETARY
19760 @c @findex LC_NUMERIC
19764 These environment variables control the way that GCC uses
19765 localization information which allows GCC to work with different
19766 national conventions. GCC inspects the locale categories
19767 @env{LC_CTYPE} and @env{LC_MESSAGES} if it has been configured to do
19768 so. These locale categories can be set to any value supported by your
19769 installation. A typical value is @samp{en_GB.UTF-8} for English in the United
19770 Kingdom encoded in UTF-8.
19772 The @env{LC_CTYPE} environment variable specifies character
19773 classification. GCC uses it to determine the character boundaries in
19774 a string; this is needed for some multibyte encodings that contain quote
19775 and escape characters that would otherwise be interpreted as a string
19778 The @env{LC_MESSAGES} environment variable specifies the language to
19779 use in diagnostic messages.
19781 If the @env{LC_ALL} environment variable is set, it overrides the value
19782 of @env{LC_CTYPE} and @env{LC_MESSAGES}; otherwise, @env{LC_CTYPE}
19783 and @env{LC_MESSAGES} default to the value of the @env{LANG}
19784 environment variable. If none of these variables are set, GCC
19785 defaults to traditional C English behavior.
19789 If @env{TMPDIR} is set, it specifies the directory to use for temporary
19790 files. GCC uses temporary files to hold the output of one stage of
19791 compilation which is to be used as input to the next stage: for example,
19792 the output of the preprocessor, which is the input to the compiler
19795 @item GCC_COMPARE_DEBUG
19796 @findex GCC_COMPARE_DEBUG
19797 Setting @env{GCC_COMPARE_DEBUG} is nearly equivalent to passing
19798 @option{-fcompare-debug} to the compiler driver. See the documentation
19799 of this option for more details.
19801 @item GCC_EXEC_PREFIX
19802 @findex GCC_EXEC_PREFIX
19803 If @env{GCC_EXEC_PREFIX} is set, it specifies a prefix to use in the
19804 names of the subprograms executed by the compiler. No slash is added
19805 when this prefix is combined with the name of a subprogram, but you can
19806 specify a prefix that ends with a slash if you wish.
19808 If @env{GCC_EXEC_PREFIX} is not set, GCC will attempt to figure out
19809 an appropriate prefix to use based on the pathname it was invoked with.
19811 If GCC cannot find the subprogram using the specified prefix, it
19812 tries looking in the usual places for the subprogram.
19814 The default value of @env{GCC_EXEC_PREFIX} is
19815 @file{@var{prefix}/lib/gcc/} where @var{prefix} is the prefix to
19816 the installed compiler. In many cases @var{prefix} is the value
19817 of @code{prefix} when you ran the @file{configure} script.
19819 Other prefixes specified with @option{-B} take precedence over this prefix.
19821 This prefix is also used for finding files such as @file{crt0.o} that are
19824 In addition, the prefix is used in an unusual way in finding the
19825 directories to search for header files. For each of the standard
19826 directories whose name normally begins with @samp{/usr/local/lib/gcc}
19827 (more precisely, with the value of @env{GCC_INCLUDE_DIR}), GCC tries
19828 replacing that beginning with the specified prefix to produce an
19829 alternate directory name. Thus, with @option{-Bfoo/}, GCC will search
19830 @file{foo/bar} where it would normally search @file{/usr/local/lib/bar}.
19831 These alternate directories are searched first; the standard directories
19832 come next. If a standard directory begins with the configured
19833 @var{prefix} then the value of @var{prefix} is replaced by
19834 @env{GCC_EXEC_PREFIX} when looking for header files.
19836 @item COMPILER_PATH
19837 @findex COMPILER_PATH
19838 The value of @env{COMPILER_PATH} is a colon-separated list of
19839 directories, much like @env{PATH}. GCC tries the directories thus
19840 specified when searching for subprograms, if it can't find the
19841 subprograms using @env{GCC_EXEC_PREFIX}.
19844 @findex LIBRARY_PATH
19845 The value of @env{LIBRARY_PATH} is a colon-separated list of
19846 directories, much like @env{PATH}. When configured as a native compiler,
19847 GCC tries the directories thus specified when searching for special
19848 linker files, if it can't find them using @env{GCC_EXEC_PREFIX}. Linking
19849 using GCC also uses these directories when searching for ordinary
19850 libraries for the @option{-l} option (but directories specified with
19851 @option{-L} come first).
19855 @cindex locale definition
19856 This variable is used to pass locale information to the compiler. One way in
19857 which this information is used is to determine the character set to be used
19858 when character literals, string literals and comments are parsed in C and C++.
19859 When the compiler is configured to allow multibyte characters,
19860 the following values for @env{LANG} are recognized:
19864 Recognize JIS characters.
19866 Recognize SJIS characters.
19868 Recognize EUCJP characters.
19871 If @env{LANG} is not defined, or if it has some other value, then the
19872 compiler will use mblen and mbtowc as defined by the default locale to
19873 recognize and translate multibyte characters.
19877 Some additional environments variables affect the behavior of the
19880 @include cppenv.texi
19884 @node Precompiled Headers
19885 @section Using Precompiled Headers
19886 @cindex precompiled headers
19887 @cindex speed of compilation
19889 Often large projects have many header files that are included in every
19890 source file. The time the compiler takes to process these header files
19891 over and over again can account for nearly all of the time required to
19892 build the project. To make builds faster, GCC allows users to
19893 `precompile' a header file; then, if builds can use the precompiled
19894 header file they will be much faster.
19896 To create a precompiled header file, simply compile it as you would any
19897 other file, if necessary using the @option{-x} option to make the driver
19898 treat it as a C or C++ header file. You will probably want to use a
19899 tool like @command{make} to keep the precompiled header up-to-date when
19900 the headers it contains change.
19902 A precompiled header file will be searched for when @code{#include} is
19903 seen in the compilation. As it searches for the included file
19904 (@pxref{Search Path,,Search Path,cpp,The C Preprocessor}) the
19905 compiler looks for a precompiled header in each directory just before it
19906 looks for the include file in that directory. The name searched for is
19907 the name specified in the @code{#include} with @samp{.gch} appended. If
19908 the precompiled header file can't be used, it is ignored.
19910 For instance, if you have @code{#include "all.h"}, and you have
19911 @file{all.h.gch} in the same directory as @file{all.h}, then the
19912 precompiled header file will be used if possible, and the original
19913 header will be used otherwise.
19915 Alternatively, you might decide to put the precompiled header file in a
19916 directory and use @option{-I} to ensure that directory is searched
19917 before (or instead of) the directory containing the original header.
19918 Then, if you want to check that the precompiled header file is always
19919 used, you can put a file of the same name as the original header in this
19920 directory containing an @code{#error} command.
19922 This also works with @option{-include}. So yet another way to use
19923 precompiled headers, good for projects not designed with precompiled
19924 header files in mind, is to simply take most of the header files used by
19925 a project, include them from another header file, precompile that header
19926 file, and @option{-include} the precompiled header. If the header files
19927 have guards against multiple inclusion, they will be skipped because
19928 they've already been included (in the precompiled header).
19930 If you need to precompile the same header file for different
19931 languages, targets, or compiler options, you can instead make a
19932 @emph{directory} named like @file{all.h.gch}, and put each precompiled
19933 header in the directory, perhaps using @option{-o}. It doesn't matter
19934 what you call the files in the directory, every precompiled header in
19935 the directory will be considered. The first precompiled header
19936 encountered in the directory that is valid for this compilation will
19937 be used; they're searched in no particular order.
19939 There are many other possibilities, limited only by your imagination,
19940 good sense, and the constraints of your build system.
19942 A precompiled header file can be used only when these conditions apply:
19946 Only one precompiled header can be used in a particular compilation.
19949 A precompiled header can't be used once the first C token is seen. You
19950 can have preprocessor directives before a precompiled header; you can
19951 even include a precompiled header from inside another header, so long as
19952 there are no C tokens before the @code{#include}.
19955 The precompiled header file must be produced for the same language as
19956 the current compilation. You can't use a C precompiled header for a C++
19960 The precompiled header file must have been produced by the same compiler
19961 binary as the current compilation is using.
19964 Any macros defined before the precompiled header is included must
19965 either be defined in the same way as when the precompiled header was
19966 generated, or must not affect the precompiled header, which usually
19967 means that they don't appear in the precompiled header at all.
19969 The @option{-D} option is one way to define a macro before a
19970 precompiled header is included; using a @code{#define} can also do it.
19971 There are also some options that define macros implicitly, like
19972 @option{-O} and @option{-Wdeprecated}; the same rule applies to macros
19975 @item If debugging information is output when using the precompiled
19976 header, using @option{-g} or similar, the same kind of debugging information
19977 must have been output when building the precompiled header. However,
19978 a precompiled header built using @option{-g} can be used in a compilation
19979 when no debugging information is being output.
19981 @item The same @option{-m} options must generally be used when building
19982 and using the precompiled header. @xref{Submodel Options},
19983 for any cases where this rule is relaxed.
19985 @item Each of the following options must be the same when building and using
19986 the precompiled header:
19988 @gccoptlist{-fexceptions}
19991 Some other command-line options starting with @option{-f},
19992 @option{-p}, or @option{-O} must be defined in the same way as when
19993 the precompiled header was generated. At present, it's not clear
19994 which options are safe to change and which are not; the safest choice
19995 is to use exactly the same options when generating and using the
19996 precompiled header. The following are known to be safe:
19998 @gccoptlist{-fmessage-length= -fpreprocessed -fsched-interblock @gol
19999 -fsched-spec -fsched-spec-load -fsched-spec-load-dangerous @gol
20000 -fsched-verbose=@var{number} -fschedule-insns -fvisibility= @gol
20005 For all of these except the last, the compiler will automatically
20006 ignore the precompiled header if the conditions aren't met. If you
20007 find an option combination that doesn't work and doesn't cause the
20008 precompiled header to be ignored, please consider filing a bug report,
20011 If you do use differing options when generating and using the
20012 precompiled header, the actual behavior will be a mixture of the
20013 behavior for the options. For instance, if you use @option{-g} to
20014 generate the precompiled header but not when using it, you may or may
20015 not get debugging information for routines in the precompiled header.