1GCC(1) GNU GCC(1)
2
6 gcc - GNU project C and C++ compiler
7
9 gcc [-c⎪-S⎪-E] [-std=standard]
10 [-g] [-pg] [-Olevel]
11 [-Wwarn...] [-pedantic]
12 [-Idir...] [-Ldir...]
13 [-Dmacro[=defn]...] [-Umacro]
14 [-foption...] [-mmachine-option...]
15 [-o outfile] infile...
16 Only the most useful options are listed here; see below for the remain‐
18 der. g++ accepts mostly the same options as gcc.
19
21 When you invoke GCC, it normally does preprocessing, compilation,
22 assembly and linking. The "overall options" allow you to stop this
23 process at an intermediate stage. For example, the -c option says not
24 to run the linker. Then the output consists of object files output by
25 the assembler.
26 Other options are passed on to one stage of processing. Some options
28 control the preprocessor and others the compiler itself. Yet other
29 options control the assembler and linker; most of these are not docu‐
30 mented here, since you rarely need to use any of them.
31 Most of the command line options that you can use with GCC are useful
33 for C programs; when an option is only useful with another language
34 (usually C++), the explanation says so explicitly. If the description
35 for a particular option does not mention a source language, you can use
36 that option with all supported languages.
37 The gcc program accepts options and file names as operands. Many
39 options have multi-letter names; therefore multiple single-letter
40 options may not be grouped: -dr is very different from -d -r.
41 You can mix options and other arguments. For the most part, the order
43 you use doesn't matter. Order does matter when you use several options
44 of the same kind; for example, if you specify -L more than once, the
45 directories are searched in the order specified.
46 Many options have long names starting with -f or with -W---for example,
48 -fstrength-reduce, -Wformat and so on. Most of these have both posi‐
49 tive and negative forms; the negative form of -ffoo would be -fno-foo.
50 This manual documents only one of these two forms, whichever one is not
51 the default.
52
54 Option Summary
55 Here is a summary of all the options, grouped by type. Explanations
57 are in the following sections.
58 Overall Options
60 -c -S -E -o file -combine -pipe -pass-exit-codes -x language
61 -v -### --help --target-help --version
62 C Language Options
64 -ansi -std=standard -aux-info filename -fno-asm -fno-builtin
65 -fno-builtin-function -fhosted -ffreestanding -fms-extensions
66 -trigraphs -no-integrated-cpp -traditional -traditional-cpp
67 -fallow-single-precision -fcond-mismatch -fsigned-bitfields
68 -fsigned-char -funsigned-bitfields -funsigned-char
69 C++ Language Options
71 -fabi-version=n -fno-access-control -fcheck-new -fconserve-space
72 -ffriend-injection -fno-const-strings -fno-elide-constructors
73 -fno-enforce-eh-specs -ffor-scope -fno-for-scope -fno-gnu-key‐
74 words -fno-implicit-templates -fno-implicit-inline-templates
75 -fno-implement-inlines -fms-extensions -fno-nonansi-builtins
76 -fno-operator-names -fno-optional-diags -fpermissive -frepo
77 -fno-rtti -fstats -ftemplate-depth-n -fno-threadsafe-statics
78 -fuse-cxa-atexit -fno-weak -nostdinc++ -fno-default-inline
79 -fvisibility-inlines-hidden -Wabi -Wctor-dtor-privacy -Wnon-vir‐
80 tual-dtor -Wreorder -Weffc++ -Wno-deprecated -Wstrict-null-sen‐
81 tinel -Wno-non-template-friend -Wold-style-cast -Woverloaded-vir‐
82 tual -Wno-pmf-conversions -Wsign-promo
83 Objective-C and Objective-C++ Language Options
85 -fconstant-string-class=class-name -fgnu-runtime -fnext-runtime
86 -fno-nil-receivers -fobjc-call-cxx-cdtors -fobjc-direct-dispatch
87 -fobjc-exceptions -fobjc-gc -freplace-objc-classes -fzero-link
88 -gen-decls -Wassign-intercept -Wno-protocol -Wselector
89 -Wstrict-selector-match -Wundeclared-selector
90 Language Independent Options
92 -fmessage-length=n -fdiagnostics-show-location=[once⎪every-line]
93 -fdiagnostics-show-options
94 Warning Options
96 -fsyntax-only -pedantic -pedantic-errors -w -Wextra -Wall
97 -Waggregate-return -Wno-attributes -Wc++-compat -Wcast-align
98 -Wcast-qual -Wchar-subscripts -Wcomment -Wconversion -Wno-depre‐
99 cated-declarations -Wdisabled-optimization -Wno-div-by-zero
100 -Wno-endif-labels -Werror -Werror-implicit-function-declaration
101 -Wfatal-errors -Wfloat-equal -Wformat -Wformat=2 -Wno-for‐
102 mat-extra-args -Wformat-nonliteral -Wformat-security -Wformat-y2k
103 -Wimplicit -Wimplicit-function-declaration -Wimplicit-int -Wim‐
104 port -Wno-import -Winit-self -Winline -Wno-int-to-pointer-cast
105 -Wno-invalid-offsetof -Winvalid-pch -Wlarger-than-len -Wun‐
106 safe-loop-optimizations -Wlong-long -Wmain -Wmissing-braces
107 -Wmissing-field-initializers -Wmissing-format-attribute -Wmiss‐
108 ing-include-dirs -Wmissing-noreturn -Wno-multichar -Wnonnull
109 -Wpacked -Wpadded -Wparentheses -Wpointer-arith
110 -Wno-pointer-to-int-cast -Wredundant-decls -Wreturn-type -Wse‐
111 quence-point -Wshadow -Wsign-compare -Wstack-protector
112 -Wstrict-aliasing -Wstrict-aliasing=2 -Wswitch -Wswitch-default
113 -Wswitch-enum -Wsystem-headers -Wtrigraphs -Wundef -Wuninitial‐
114 ized -Wunknown-pragmas -Wno-pragmas -Wunreachable-code -Wunused
115 -Wunused-function -Wunused-label -Wunused-parameter
116 -Wunused-value -Wunused-variable -Wvariadic-macros
117 -Wvolatile-register-var -Wwrite-strings
118 C-only Warning Options
120 -Wbad-function-cast -Wmissing-declarations -Wmissing-prototypes
121 -Wnested-externs -Wold-style-definition -Wstrict-prototypes
122 -Wtraditional -Wdeclaration-after-statement -Wpointer-sign
123 Debugging Options
125 -dletters -dumpspecs -dumpmachine -dumpversion -fdump-unnumbered
126 -fdump-translation-unit[-n] -fdump-class-hierarchy[-n]
127 -fdump-ipa-all -fdump-ipa-cgraph -fdump-tree-all -fdump-tree-origi‐
128 nal[-n] -fdump-tree-optimized[-n] -fdump-tree-inlined[-n]
129 -fdump-tree-cfg -fdump-tree-vcg -fdump-tree-alias -fdump-tree-ch
130 -fdump-tree-ssa[-n] -fdump-tree-pre[-n] -fdump-tree-ccp[-n]
131 -fdump-tree-dce[-n] -fdump-tree-gimple[-raw] -fdump-tree-mud‐
132 flap[-n] -fdump-tree-dom[-n] -fdump-tree-dse[-n]
133 -fdump-tree-phiopt[-n] -fdump-tree-forwprop[-n] -fdump-tree-copyre‐
134 name[-n] -fdump-tree-nrv -fdump-tree-vect -fdump-tree-sink
135 -fdump-tree-sra[-n] -fdump-tree-salias -fdump-tree-fre[-n]
136 -fdump-tree-vrp[-n] -ftree-vectorizer-verbose=n
137 -fdump-tree-storeccp[-n] -feliminate-dwarf2-dups -felimi‐
138 nate-unused-debug-types -feliminate-unused-debug-symbols
139 -fmem-report -fprofile-arcs -frandom-seed=string -fsched-verbose=n
140 -ftest-coverage -ftime-report -fvar-tracking -g -glevel -gcoff
141 -gdwarf-2 -ggdb -gstabs -gstabs+ -gvms -gxcoff -gxcoff+ -p
142 -pg -print-file-name=library -print-libgcc-file-name
143 -print-multi-directory -print-multi-lib -print-prog-name=program
144 -print-search-dirs -Q -save-temps -time
145 Optimization Options
147 -falign-functions=n -falign-jumps=n -falign-labels=n
148 -falign-loops=n -fbounds-check -fmudflap -fmudflapth -fmudflapir
149 -fbranch-probabilities -fprofile-values -fvpt -fbranch-tar‐
150 get-load-optimize -fbranch-target-load-optimize2 -fbtr-bb-exclusive
151 -fcaller-saves -fcprop-registers -fcse-follow-jumps
152 -fcse-skip-blocks -fcx-limited-range -fdata-sections -fde‐
153 layed-branch -fdelete-null-pointer-checks -fearly-inlining -fex‐
154 pensive-optimizations -ffast-math -ffloat-store -fforce-addr
155 -ffunction-sections -fgcse -fgcse-lm -fgcse-sm -fgcse-las
156 -fgcse-after-reload -floop-optimize -fcrossjumping -fif-conversion
157 -fif-conversion2 -finline-functions -finline-functions-called-once
158 -finline-limit=n -fkeep-inline-functions -fkeep-static-consts
159 -fmerge-constants -fmerge-all-constants -fmodulo-sched
160 -fno-branch-count-reg -fno-default-inline -fno-defer-pop
161 -floop-optimize2 -fmove-loop-invariants -fno-function-cse
162 -fno-guess-branch-probability -fno-inline -fno-math-errno
163 -fno-peephole -fno-peephole2 -funsafe-math-optimizations -fun‐
164 safe-loop-optimizations -ffinite-math-only -fno-trapping-math
165 -fno-zero-initialized-in-bss -fomit-frame-pointer -foptimize-reg‐
166 ister-move -foptimize-sibling-calls -fprefetch-loop-arrays -fpro‐
167 file-generate -fprofile-use -fregmove -frename-registers -fre‐
168 order-blocks -freorder-blocks-and-partition -freorder-functions
169 -frerun-cse-after-loop -frerun-loop-opt -frounding-math -fsched‐
170 ule-insns -fschedule-insns2 -fno-sched-interblock -fno-sched-spec
171 -fsched-spec-load -fsched-spec-load-dangerous
172 -fsched-stalled-insns=n -fsched-stalled-insns-dep=n
173 -fsched2-use-superblocks -fsched2-use-traces -freschedule-mod‐
174 ulo-scheduled-loops -fsignaling-nans -fsingle-precision-constant
175 -fstack-protector -fstack-protector-all -fstrength-reduce
176 -fstrict-aliasing -ftracer -fthread-jumps -funroll-all-loops
177 -funroll-loops -fpeel-loops -fsplit-ivs-in-unroller
178 -funswitch-loops -fvariable-expansion-in-unroller -ftree-pre
179 -ftree-ccp -ftree-dce -ftree-loop-optimize -ftree-loop-linear
180 -ftree-loop-im -ftree-loop-ivcanon -fivopts -ftree-dominator-opts
181 -ftree-dse -ftree-copyrename -ftree-sink -ftree-ch -ftree-sra
182 -ftree-ter -ftree-lrs -ftree-fre -ftree-vectorize
183 -ftree-vect-loop-version -ftree-salias -fweb -ftree-copy-prop
184 -ftree-store-ccp -ftree-store-copy-prop -fwhole-program --param
185 name=value -O -O0 -O1 -O2 -O3 -Os
186 Preprocessor Options
188 -Aquestion=answer -A-question[=answer] -C -dD -dI -dM -dN
189 -Dmacro[=defn] -E -H -idirafter dir -include file -imacros file
190 -iprefix file -iwithprefix dir -iwithprefixbefore dir -isystem
191 dir -isysroot dir -M -MM -MF -MG -MP -MQ -MT -nostdinc -P
192 -fworking-directory -remap -trigraphs -undef -Umacro -Wp,option
193 -Xpreprocessor option
194 Assembler Option
196 -Wa,option -Xassembler option
197 Linker Options
199 object-file-name -llibrary -nostartfiles -nodefaultlibs -nost‐
200 dlib -pie -rdynamic -s -static -static-libgcc -shared
201 -shared-libgcc -symbolic -Wl,option -Xlinker option -u symbol
202 Directory Options
204 -Bprefix -Idir -iquotedir -Ldir -specs=file -I- --sysroot=dir
205 Target Options
207 -V version -b machine
208 Machine Dependent Options
210 ARC Options -EB -EL -mmangle-cpu -mcpu=cpu -mtext=text-section
211 -mdata=data-section -mrodata=readonly-data-section
212 ARM Options -mapcs-frame -mno-apcs-frame -mabi=name
214 -mapcs-stack-check -mno-apcs-stack-check -mapcs-float
215 -mno-apcs-float -mapcs-reentrant -mno-apcs-reentrant -msched-pro‐
216 log -mno-sched-prolog -mlittle-endian -mbig-endian -mwords-lit‐
217 tle-endian -mfloat-abi=name -msoft-float -mhard-float -mfpe
218 -mthumb-interwork -mno-thumb-interwork -mcpu=name -march=name
219 -mfpu=name -mstructure-size-boundary=n -mabort-on-noreturn
220 -mlong-calls -mno-long-calls -msingle-pic-base -mno-sin‐
221 gle-pic-base -mpic-register=reg -mnop-fun-dllimport -mcir‐
222 rus-fix-invalid-insns -mno-cirrus-fix-invalid-insns -mpoke-func‐
223 tion-name -mthumb -marm -mtpcs-frame -mtpcs-leaf-frame
224 -mcaller-super-interworking -mcallee-super-interworking -mtp=name
225 AVR Options -mmcu=mcu -msize -minit-stack=n -mno-interrupts
227 -mcall-prologues -mno-tablejump -mtiny-stack -mint8
228 Blackfin Options -momit-leaf-frame-pointer
230 -mno-omit-leaf-frame-pointer -mspecld-anomaly -mno-specld-anomaly
231 -mcsync-anomaly -mno-csync-anomaly -mlow-64k -mno-low64k
232 -mid-shared-library -mno-id-shared-library -mshared-library-id=n
233 -mlong-calls -mno-long-calls
234 CRIS Options -mcpu=cpu -march=cpu -mtune=cpu -mmax-stack-frame=n
236 -melinux-stacksize=n -metrax4 -metrax100 -mpdebug -mcc-init
237 -mno-side-effects -mstack-align -mdata-align -mconst-align
238 -m32-bit -m16-bit -m8-bit -mno-prologue-epilogue -mno-gotplt
239 -melf -maout -melinux -mlinux -sim -sim2 -mmul-bug-workaround
240 -mno-mul-bug-workaround
241 CRX Options -mmac -mpush-args
243 Darwin Options -all_load -allowable_client -arch
245 -arch_errors_fatal -arch_only -bind_at_load -bundle -bun‐
246 dle_loader -client_name -compatibility_version -current_version
247 -dead_strip -dependency-file -dylib_file -dylinker_install_name
248 -dynamic -dynamiclib -exported_symbols_list -filelist
249 -flat_namespace -force_cpusubtype_ALL -force_flat_namespace
250 -headerpad_max_install_names -image_base -init -install_name
251 -keep_private_externs -multi_module -multiply_defined -multi‐
252 ply_defined_unused -noall_load -no_dead_strip_inits_and_terms
253 -nofixprebinding -nomultidefs -noprebind -noseglinkedit
254 -pagezero_size -prebind -prebind_all_twolevel_modules -pri‐
255 vate_bundle -read_only_relocs -sectalign -sectobjectsymbols
256 -whyload -seg1addr -sectcreate -sectobjectsymbols -sectorder
257 -segaddr -segs_read_only_addr -segs_read_write_addr -seg_addr_table
258 -seg_addr_table_filename -seglinkedit -segprot
259 -segs_read_only_addr -segs_read_write_addr -single_module -static
260 -sub_library -sub_umbrella -twolevel_namespace -umbrella -unde‐
261 fined -unexported_symbols_list -weak_reference_mismatches -what‐
262 sloaded -F -gused -gfull -mmacosx-version-min=version
263 -mone-byte-bool
264 DEC Alpha Options -mno-fp-regs -msoft-float -malpha-as -mgas
266 -mieee -mieee-with-inexact -mieee-conformant -mfp-trap-mode=mode
267 -mfp-rounding-mode=mode -mtrap-precision=mode -mbuild-constants
268 -mcpu=cpu-type -mtune=cpu-type -mbwx -mmax -mfix -mcix
269 -mfloat-vax -mfloat-ieee -mexplicit-relocs -msmall-data
270 -mlarge-data -msmall-text -mlarge-text -mmemory-latency=time
271 DEC Alpha/VMS Options -mvms-return-codes
273 FRV Options -mgpr-32 -mgpr-64 -mfpr-32 -mfpr-64 -mhard-float
275 -msoft-float -malloc-cc -mfixed-cc -mdword -mno-dword -mdouble
276 -mno-double -mmedia -mno-media -mmuladd -mno-muladd -mfdpic
277 -minline-plt -mgprel-ro -multilib-library-pic -mlinked-fp
278 -mlong-calls -malign-labels -mlibrary-pic -macc-4 -macc-8 -mpack
279 -mno-pack -mno-eflags -mcond-move -mno-cond-move -moptimize-mem‐
280 bar -mno-optimize-membar -mscc -mno-scc -mcond-exec
281 -mno-cond-exec -mvliw-branch -mno-vliw-branch -mmulti-cond-exec
282 -mno-multi-cond-exec -mnested-cond-exec -mno-nested-cond-exec
283 -mtomcat-stats -mTLS -mtls -mcpu=cpu
284 H8/300 Options -mrelax -mh -ms -mn -mint32 -malign-300
286 HPPA Options -march=architecture-type -mbig-switch -mdis‐
288 able-fpregs -mdisable-indexing -mfast-indirect-calls -mgas
289 -mgnu-ld -mhp-ld -mfixed-range=register-range -mjump-in-delay
290 -mlinker-opt -mlong-calls -mlong-load-store -mno-big-switch
291 -mno-disable-fpregs -mno-disable-indexing -mno-fast-indirect-calls
292 -mno-gas -mno-jump-in-delay -mno-long-load-store -mno-porta‐
293 ble-runtime -mno-soft-float -mno-space-regs -msoft-float
294 -mpa-risc-1-0 -mpa-risc-1-1 -mpa-risc-2-0 -mportable-runtime
295 -mschedule=cpu-type -mspace-regs -msio -mwsio -munix=unix-std
296 -nolibdld -static -threads
297 i386 and x86-64 Options -mtune=cpu-type -march=cpu-type -mfp‐
299 math=unit -masm=dialect -mno-fancy-math-387 -mno-fp-ret-in-387
300 -msoft-float -msvr3-shlib -mno-wide-multiply -mrtd -malign-dou‐
301 ble -mpreferred-stack-boundary=num -mmmx -msse -msse2 -msse3
302 -m3dnow -mthreads -mno-align-stringops -minline-all-stringops
303 -mpush-args -maccumulate-outgoing-args -m128bit-long-double
304 -m96bit-long-double -mregparm=num -msseregparm
305 -momit-leaf-frame-pointer -mno-red-zone -mno-tls-direct-seg-refs
306 -mcmodel=code-model -m32 -m64 -mlarge-data-threshold=num
307 IA-64 Options -mbig-endian -mlittle-endian -mgnu-as -mgnu-ld
309 -mno-pic -mvolatile-asm-stop -mregister-names -mno-sdata -mcon‐
310 stant-gp -mauto-pic -minline-float-divide-min-latency -min‐
311 line-float-divide-max-throughput -minline-int-divide-min-latency
312 -minline-int-divide-max-throughput -minline-sqrt-min-latency -min‐
313 line-sqrt-max-throughput -mno-dwarf2-asm -mearly-stop-bits
314 -mfixed-range=register-range -mtls-size=tls-size -mtune=cpu-type
315 -mt -pthread -milp32 -mlp64
316 M32R/D Options -m32r2 -m32rx -m32r -mdebug -malign-loops
318 -mno-align-loops -missue-rate=number -mbranch-cost=number
319 -mmodel=code-size-model-type -msdata=sdata-type -mno-flush-func
320 -mflush-func=name -mno-flush-trap -mflush-trap=number -G num
321 M32C Options -mcpu=cpu -msim -memregs=number
323 M680x0 Options -m68000 -m68020 -m68020-40 -m68020-60 -m68030
325 -m68040 -m68060 -mcpu32 -m5200 -m68881 -mbitfield -mc68000
326 -mc68020 -mnobitfield -mrtd -mshort -msoft-float -mpcrel
327 -malign-int -mstrict-align -msep-data -mno-sep-data
328 -mshared-library-id=n -mid-shared-library -mno-id-shared-library
329 M68hc1x Options -m6811 -m6812 -m68hc11 -m68hc12 -m68hcs12
331 -mauto-incdec -minmax -mlong-calls -mshort
332 -msoft-reg-count=count
333 MCore Options -mhardlit -mno-hardlit -mdiv -mno-div -mre‐
335 lax-immediates -mno-relax-immediates -mwide-bitfields
336 -mno-wide-bitfields -m4byte-functions -mno-4byte-functions
337 -mcallgraph-data -mno-callgraph-data -mslow-bytes -mno-slow-bytes
338 -mno-lsim -mlittle-endian -mbig-endian -m210 -m340
339 -mstack-increment
340 MIPS Options -EL -EB -march=arch -mtune=arch -mips1 -mips2
342 -mips3 -mips4 -mips32 -mips32r2 -mips64 -mips16 -mno-mips16
343 -mabi=abi -mabicalls -mno-abicalls -mxgot -mno-xgot -mgp32
344 -mgp64 -mfp32 -mfp64 -mhard-float -msoft-float -msingle-float
345 -mdouble-float -mdsp -mpaired-single -mips3d -mlong64 -mlong32
346 -msym32 -mno-sym32 -Gnum -membedded-data -mno-embedded-data
347 -muninit-const-in-rodata -mno-uninit-const-in-rodata
348 -msplit-addresses -mno-split-addresses -mexplicit-relocs
349 -mno-explicit-relocs -mcheck-zero-division -mno-check-zero-divi‐
350 sion -mdivide-traps -mdivide-breaks -mmemcpy -mno-memcpy
351 -mlong-calls -mno-long-calls -mmad -mno-mad -mfused-madd
352 -mno-fused-madd -nocpp -mfix-r4000 -mno-fix-r4000 -mfix-r4400
353 -mno-fix-r4400 -mfix-vr4120 -mno-fix-vr4120 -mfix-vr4130
354 -mfix-sb1 -mno-fix-sb1 -mflush-func=func -mno-flush-func
355 -mbranch-likely -mno-branch-likely -mfp-exceptions -mno-fp-excep‐
356 tions -mvr4130-align -mno-vr4130-align
357 MMIX Options -mlibfuncs -mno-libfuncs -mepsilon -mno-epsilon
359 -mabi=gnu -mabi=mmixware -mzero-extend -mknuthdiv -mto‐
360 plevel-symbols -melf -mbranch-predict -mno-branch-predict
361 -mbase-addresses -mno-base-addresses -msingle-exit -mno-sin‐
362 gle-exit
363 MN10300 Options -mmult-bug -mno-mult-bug -mam33 -mno-am33
365 -mam33-2 -mno-am33-2 -mreturn-pointer-on-d0 -mno-crt0 -mrelax
366 MT Options -mno-crt0 -mbacc -msim -march=cpu-type
368 PDP-11 Options -mfpu -msoft-float -mac0 -mno-ac0 -m40 -m45
370 -m10 -mbcopy -mbcopy-builtin -mint32 -mno-int16 -mint16
371 -mno-int32 -mfloat32 -mno-float64 -mfloat64 -mno-float32 -mab‐
372 shi -mno-abshi -mbranch-expensive -mbranch-cheap -msplit
373 -mno-split -munix-asm -mdec-asm
374 PowerPC Options See RS/6000 and PowerPC Options.
376 RS/6000 and PowerPC Options -mcpu=cpu-type -mtune=cpu-type -mpower
378 -mno-power -mpower2 -mno-power2 -mpowerpc -mpowerpc64 -mno-pow‐
379 erpc -maltivec -mno-altivec -mpowerpc-gpopt -mno-powerpc-gpopt
380 -mpowerpc-gfxopt -mno-powerpc-gfxopt -mmfcrf -mno-mfcrf -mpopc‐
381 ntb -mno-popcntb -mfprnd -mno-fprnd -mnew-mnemonics
382 -mold-mnemonics -mfull-toc -mminimal-toc -mno-fp-in-toc
383 -mno-sum-in-toc -m64 -m32 -mxl-compat -mno-xl-compat -mpe
384 -malign-power -malign-natural -msoft-float -mhard-float -mmulti‐
385 ple -mno-multiple -mstring -mno-string -mupdate -mno-update
386 -mfused-madd -mno-fused-madd -mbit-align -mno-bit-align
387 -mstrict-align -mno-strict-align -mrelocatable -mno-relocatable
388 -mrelocatable-lib -mno-relocatable-lib -mtoc -mno-toc -mlittle
389 -mlittle-endian -mbig -mbig-endian -mdynamic-no-pic -maltivec
390 -mswdiv -mprioritize-restricted-insns=priority
391 -msched-costly-dep=dependence_type -minsert-sched-nops=scheme
392 -mcall-sysv -mcall-netbsd -maix-struct-return
393 -msvr4-struct-return -mabi=abi-type -msecure-plt -mbss-plt -misel
394 -mno-isel -misel=yes -misel=no -mspe -mno-spe -mspe=yes -mspe=no
395 -mvrsave -mno-vrsave -mfloat-gprs=yes -mfloat-gprs=no
396 -mfloat-gprs=single -mfloat-gprs=double -mprototype -mno-prototype
397 -msim -mmvme -mads -myellowknife -memb -msdata -msdata=opt
398 -mvxworks -mwindiss -G num -pthread
399 S/390 and zSeries Options -mtune=cpu-type -march=cpu-type
401 -mhard-float -msoft-float -mlong-double-64 -mlong-double-128
402 -mbackchain -mno-backchain -mpacked-stack -mno-packed-stack
403 -msmall-exec -mno-small-exec -mmvcle -mno-mvcle -m64 -m31 -mde‐
404 bug -mno-debug -mesa -mzarch -mtpf-trace -mno-tpf-trace
405 -mfused-madd -mno-fused-madd -mwarn-framesize -mwarn-dynamicstack
406 -mstack-size -mstack-guard
407 SH Options -m1 -m2 -m2e -m3 -m3e -m4-nofpu -m4-single-only
409 -m4-single -m4 -m4a-nofpu -m4a-single-only -m4a-single -m4a -m4al
410 -m5-64media -m5-64media-nofpu -m5-32media -m5-32media-nofpu
411 -m5-compact -m5-compact-nofpu -mb -ml -mdalign -mrelax
412 -mbigtable -mfmovd -mhitachi -mrenesas -mno-renesas -mnomacsave
413 -mieee -misize -mpadstruct -mspace -mprefergot -musermode
414 -multcost=number -mdiv=strategy -mdivsi3_libfunc=name -mad‐
415 just-unroll -mindexed-addressing -mgettrcost=number -mpt-fixed
416 -minvalid-symbols
417 SPARC Options -mcpu=cpu-type -mtune=cpu-type -mcmodel=code-model
419 -m32 -m64 -mapp-regs -mno-app-regs -mfaster-structs
420 -mno-faster-structs -mfpu -mno-fpu -mhard-float -msoft-float
421 -mhard-quad-float -msoft-quad-float -mimpure-text
422 -mno-impure-text -mlittle-endian -mstack-bias -mno-stack-bias
423 -munaligned-doubles -mno-unaligned-doubles -mv8plus -mno-v8plus
424 -mvis -mno-vis -threads -pthreads -pthread
425 System V Options -Qy -Qn -YP,paths -Ym,dir
427 TMS320C3x/C4x Options -mcpu=cpu -mbig -msmall -mregparm -mmem‐
429 parm -mfast-fix -mmpyi -mbk -mti -mdp-isr-reload -mrpts=count
430 -mrptb -mdb -mloop-unsigned -mparallel-insns -mparallel-mpy
431 -mpreserve-float
432 V850 Options -mlong-calls -mno-long-calls -mep -mno-ep -mpro‐
434 log-function -mno-prolog-function -mspace -mtda=n -msda=n
435 -mzda=n -mapp-regs -mno-app-regs -mdisable-callt -mno-dis‐
436 able-callt -mv850e1 -mv850e -mv850 -mbig-switch
437 VAX Options -mg -mgnu -munix
439 x86-64 Options See i386 and x86-64 Options.
441 Xstormy16 Options -msim
443 Xtensa Options -mconst16 -mno-const16 -mfused-madd -mno-fused-madd
445 -mtext-section-literals -mno-text-section-literals -mtarget-align
446 -mno-target-align -mlongcalls -mno-longcalls
447 zSeries Options See S/390 and zSeries Options.
449 Code Generation Options
451 -fcall-saved-reg -fcall-used-reg -ffixed-reg -fexceptions
452 -fnon-call-exceptions -funwind-tables -fasynchronous-unwind-tables
453 -finhibit-size-directive -finstrument-functions -fno-common
454 -fno-ident -fpcc-struct-return -fpic -fPIC -fpie -fPIE
455 -fno-jump-tables -freg-struct-return -fshared-data -fshort-enums
456 -fshort-double -fshort-wchar -fverbose-asm -fpack-struct[=n]
457 -fstack-check -fstack-limit-register=reg -fstack-limit-symbol=sym
458 -fargument-alias -fargument-noalias -fargument-noalias-global
459 -fleading-underscore -ftls-model=model -ftrapv -fwrapv
460 -fbounds-check -fvisibility
461 Options Controlling the Kind of Output
463 Compilation can involve up to four stages: preprocessing, compilation
465 proper, assembly and linking, always in that order. GCC is capable of
466 preprocessing and compiling several files either into several assembler
467 input files, or into one assembler input file; then each assembler
468 input file produces an object file, and linking combines all the object
469 files (those newly compiled, and those specified as input) into an exe‐
470 cutable file.
471 For any given input file, the file name suffix determines what kind of
473 compilation is done:
474 file.c
476 C source code which must be preprocessed.
477 file.i
479 C source code which should not be preprocessed.
480 file.ii
482 C++ source code which should not be preprocessed.
483 file.m
485 Objective-C source code. Note that you must link with the libobjc
486 library to make an Objective-C program work.
487 file.mi
489 Objective-C source code which should not be preprocessed.
490 file.mm
492 file.M
493 Objective-C++ source code. Note that you must link with the
494 libobjc library to make an Objective-C++ program work. Note that
495 .M refers to a literal capital M.
496 file.mii
498 Objective-C++ source code which should not be preprocessed.
499 file.h
501 C, C++, Objective-C or Objective-C++ header file to be turned into
502 a precompiled header.
503 file.cc
505 file.cp
506 file.cxx
507 file.cpp
508 file.CPP
509 file.c++
510 file.C
511 C++ source code which must be preprocessed. Note that in .cxx, the
512 last two letters must both be literally x. Likewise, .C refers to
513 a literal capital C.
514 file.mm
516 file.M
517 Objective-C++ source code which must be preprocessed.
518 file.mii
520 Objective-C++ source code which should not be preprocessed.
521 file.hh
523 file.H
524 C++ header file to be turned into a precompiled header.
525 file.f
527 file.for
528 file.FOR
529 Fixed form Fortran source code which should not be preprocessed.
530 file.F
532 file.fpp
533 file.FPP
534 Fixed form Fortran source code which must be preprocessed (with the
535 traditional preprocessor).
536 file.f90
538 file.f95
539 Free form Fortran source code which should not be preprocessed.
540 file.F90
542 file.F95
543 Free form Fortran source code which must be preprocessed (with the
544 traditional preprocessor).
545 file.ads
547 Ada source code file which contains a library unit declaration (a
548 declaration of a package, subprogram, or generic, or a generic
549 instantiation), or a library unit renaming declaration (a package,
550 generic, or subprogram renaming declaration). Such files are also
551 called specs.
552 file.adb
554 Ada source code file containing a library unit body (a subprogram
555 or package body). Such files are also called bodies.
556 file.s
558 Assembler code.
559 file.S
561 Assembler code which must be preprocessed.
562 other
564 An object file to be fed straight into linking. Any file name with
565 no recognized suffix is treated this way.
566 You can specify the input language explicitly with the -x option:
568 -x language
570 Specify explicitly the language for the following input files
571 (rather than letting the compiler choose a default based on the
572 file name suffix). This option applies to all following input
573 files until the next -x option. Possible values for language are:
574 c c-header c-cpp-output
576 c++ c++-header c++-cpp-output
577 objective-c objective-c-header objective-c-cpp-output
578 objective-c++ objective-c++-header objective-c++-cpp-output
579 assembler assembler-with-cpp
580 ada
581 f95 f95-cpp-input
582 java
583 treelang
584 -x none
586 Turn off any specification of a language, so that subsequent files
587 are handled according to their file name suffixes (as they are if
588 -x has not been used at all).
589 -pass-exit-codes
591 Normally the gcc program will exit with the code of 1 if any phase
592 of the compiler returns a non-success return code. If you specify
593 -pass-exit-codes, the gcc program will instead return with numeri‐
594 cally highest error produced by any phase that returned an error
595 indication.
596 If you only want some of the stages of compilation, you can use -x (or
598 filename suffixes) to tell gcc where to start, and one of the options
599 -c, -S, or -E to say where gcc is to stop. Note that some combinations
600 (for example, -x cpp-output -E) instruct gcc to do nothing at all.
601 -c Compile or assemble the source files, but do not link. The linking
603 stage simply is not done. The ultimate output is in the form of an
604 object file for each source file.
605 By default, the object file name for a source file is made by
607 replacing the suffix .c, .i, .s, etc., with .o.
608 Unrecognized input files, not requiring compilation or assembly,
610 are ignored.
611 -S Stop after the stage of compilation proper; do not assemble. The
613 output is in the form of an assembler code file for each non-assem‐
614 bler input file specified.
615 By default, the assembler file name for a source file is made by
617 replacing the suffix .c, .i, etc., with .s.
618 Input files that don't require compilation are ignored.
620 -E Stop after the preprocessing stage; do not run the compiler proper.
622 The output is in the form of preprocessed source code, which is
623 sent to the standard output.
624 Input files which don't require preprocessing are ignored.
626 -o file
628 Place output in file file. This applies regardless to whatever
629 sort of output is being produced, whether it be an executable file,
630 an object file, an assembler file or preprocessed C code.
631 If -o is not specified, the default is to put an executable file in
633 a.out, the object file for source.suffix in source.o, its assembler
634 file in source.s, a precompiled header file in source.suffix.gch,
635 and all preprocessed C source on standard output.
636 -v Print (on standard error output) the commands executed to run the
638 stages of compilation. Also print the version number of the com‐
639 piler driver program and of the preprocessor and the compiler
640 proper.
641 -###
643 Like -v except the commands are not executed and all command argu‐
644 ments are quoted. This is useful for shell scripts to capture the
645 driver-generated command lines.
646 -pipe
648 Use pipes rather than temporary files for communication between the
649 various stages of compilation. This fails to work on some systems
650 where the assembler is unable to read from a pipe; but the GNU
651 assembler has no trouble.
652 -combine
654 If you are compiling multiple source files, this option tells the
655 driver to pass all the source files to the compiler at once (for
656 those languages for which the compiler can handle this). This will
657 allow intermodule analysis (IMA) to be performed by the compiler.
658 Currently the only language for which this is supported is C. If
659 you pass source files for multiple languages to the driver, using
660 this option, the driver will invoke the compiler(s) that support
661 IMA once each, passing each compiler all the source files appropri‐
662 ate for it. For those languages that do not support IMA this
663 option will be ignored, and the compiler will be invoked once for
664 each source file in that language. If you use this option in con‐
665 junction with -save-temps, the compiler will generate multiple pre-
666 processed files (one for each source file), but only one (combined)
667 .o or .s file.
668 --help
670 Print (on the standard output) a description of the command line
671 options understood by gcc. If the -v option is also specified then
672 --help will also be passed on to the various processes invoked by
673 gcc, so that they can display the command line options they accept.
674 If the -Wextra option is also specified then command line options
675 which have no documentation associated with them will also be dis‐
676 played.
677 --target-help
679 Print (on the standard output) a description of target specific
680 command line options for each tool.
681 --version
683 Display the version number and copyrights of the invoked GCC.
684 Compiling C++ Programs
686 C++ source files conventionally use one of the suffixes .C, .cc, .cpp,
688 .CPP, .c++, .cp, or .cxx; C++ header files often use .hh or .H; and
689 preprocessed C++ files use the suffix .ii. GCC recognizes files with
690 these names and compiles them as C++ programs even if you call the com‐
691 piler the same way as for compiling C programs (usually with the name
692 gcc).
693 However, C++ programs often require class libraries as well as a com‐
695 piler that understands the C++ language---and under some circumstances,
696 you might want to compile programs or header files from standard input,
697 or otherwise without a suffix that flags them as C++ programs. You
698 might also like to precompile a C header file with a .h extension to be
699 used in C++ compilations. g++ is a program that calls GCC with the
700 default language set to C++, and automatically specifies linking
701 against the C++ library. On many systems, g++ is also installed with
702 the name c++.
703 When you compile C++ programs, you may specify many of the same com‐
705 mand-line options that you use for compiling programs in any language;
706 or command-line options meaningful for C and related languages; or
707 options that are meaningful only for C++ programs.
708 Options Controlling C Dialect
710 The following options control the dialect of C (or languages derived
712 from C, such as C++, Objective-C and Objective-C++) that the compiler
713 accepts:
714 -ansi
716 In C mode, support all ISO C90 programs. In C++ mode, remove GNU
717 extensions that conflict with ISO C++.
718 This turns off certain features of GCC that are incompatible with
720 ISO C90 (when compiling C code), or of standard C++ (when compiling
721 C++ code), such as the "asm" and "typeof" keywords, and predefined
722 macros such as "unix" and "vax" that identify the type of system
723 you are using. It also enables the undesirable and rarely used ISO
724 trigraph feature. For the C compiler, it disables recognition of
725 C++ style // comments as well as the "inline" keyword.
726 The alternate keywords "__asm__", "__extension__", "__inline__" and
728 "__typeof__" continue to work despite -ansi. You would not want to
729 use them in an ISO C program, of course, but it is useful to put
730 them in header files that might be included in compilations done
731 with -ansi. Alternate predefined macros such as "__unix__" and
732 "__vax__" are also available, with or without -ansi.
733 The -ansi option does not cause non-ISO programs to be rejected
735 gratuitously. For that, -pedantic is required in addition to
736 -ansi.
737 The macro "__STRICT_ANSI__" is predefined when the -ansi option is
739 used. Some header files may notice this macro and refrain from
740 declaring certain functions or defining certain macros that the ISO
741 standard doesn't call for; this is to avoid interfering with any
742 programs that might use these names for other things.
743 Functions which would normally be built in but do not have seman‐
745 tics defined by ISO C (such as "alloca" and "ffs") are not built-in
746 functions with -ansi is used.
747 -std=
749 Determine the language standard. This option is currently only
750 supported when compiling C or C++. A value for this option must be
751 provided; possible values are
752 c89
754 iso9899:1990
755 ISO C90 (same as -ansi).
756 iso9899:199409
758 ISO C90 as modified in amendment 1.
759 c99
761 c9x
762 iso9899:1999
763 iso9899:199x
764 ISO C99. Note that this standard is not yet fully supported;
765 see <http://gcc.gnu.org/gcc-4.1/c99status.html> for more infor‐
766 mation. The names c9x and iso9899:199x are deprecated.
767 gnu89
769 Default, ISO C90 plus GNU extensions (including some C99 fea‐
770 tures).
771 gnu99
773 gnu9x
774 ISO C99 plus GNU extensions. When ISO C99 is fully implemented
775 in GCC, this will become the default. The name gnu9x is depre‐
776 cated.
777 c++98
779 The 1998 ISO C++ standard plus amendments.
780 gnu++98
782 The same as -std=c++98 plus GNU extensions. This is the
783 default for C++ code.
784 Even when this option is not specified, you can still use some of
786 the features of newer standards in so far as they do not conflict
787 with previous C standards. For example, you may use "__restrict__"
788 even when -std=c99 is not specified.
789 The -std options specifying some version of ISO C have the same
791 effects as -ansi, except that features that were not in ISO C90 but
792 are in the specified version (for example, // comments and the
793 "inline" keyword in ISO C99) are not disabled.
794 -aux-info filename
796 Output to the given filename prototyped declarations for all func‐
797 tions declared and/or defined in a translation unit, including
798 those in header files. This option is silently ignored in any lan‐
799 guage other than C.
800 Besides declarations, the file indicates, in comments, the origin
802 of each declaration (source file and line), whether the declaration
803 was implicit, prototyped or unprototyped (I, N for new or O for
804 old, respectively, in the first character after the line number and
805 the colon), and whether it came from a declaration or a definition
806 (C or F, respectively, in the following character). In the case of
807 function definitions, a K&R-style list of arguments followed by
808 their declarations is also provided, inside comments, after the
809 declaration.
810 -fno-asm
812 Do not recognize "asm", "inline" or "typeof" as a keyword, so that
813 code can use these words as identifiers. You can use the keywords
814 "__asm__", "__inline__" and "__typeof__" instead. -ansi implies
815 -fno-asm.
816 In C++, this switch only affects the "typeof" keyword, since "asm"
818 and "inline" are standard keywords. You may want to use the
819 -fno-gnu-keywords flag instead, which has the same effect. In C99
820 mode (-std=c99 or -std=gnu99), this switch only affects the "asm"
821 and "typeof" keywords, since "inline" is a standard keyword in ISO
822 C99.
823 -fno-builtin
825 -fno-builtin-function
826 Don't recognize built-in functions that do not begin with
827 __builtin_ as prefix.
828 GCC normally generates special code to handle certain built-in
830 functions more efficiently; for instance, calls to "alloca" may
831 become single instructions that adjust the stack directly, and
832 calls to "memcpy" may become inline copy loops. The resulting code
833 is often both smaller and faster, but since the function calls no
834 longer appear as such, you cannot set a breakpoint on those calls,
835 nor can you change the behavior of the functions by linking with a
836 different library. In addition, when a function is recognized as a
837 built-in function, GCC may use information about that function to
838 warn about problems with calls to that function, or to generate
839 more efficient code, even if the resulting code still contains
840 calls to that function. For example, warnings are given with
841 -Wformat for bad calls to "printf", when "printf" is built in, and
842 "strlen" is known not to modify global memory.
843 With the -fno-builtin-function option only the built-in function
845 function is disabled. function must not begin with __builtin_. If
846 a function is named this is not built-in in this version of GCC,
847 this option is ignored. There is no corresponding -fbuiltin-func‐
848 tion option; if you wish to enable built-in functions selectively
849 when using -fno-builtin or -ffreestanding, you may define macros
850 such as:
851 #define abs(n) __builtin_abs ((n))
853 #define strcpy(d, s) __builtin_strcpy ((d), (s))
854 -fhosted
856 Assert that compilation takes place in a hosted environment. This
857 implies -fbuiltin. A hosted environment is one in which the entire
858 standard library is available, and in which "main" has a return
859 type of "int". Examples are nearly everything except a kernel.
860 This is equivalent to -fno-freestanding.
861 -ffreestanding
863 Assert that compilation takes place in a freestanding environment.
864 This implies -fno-builtin. A freestanding environment is one in
865 which the standard library may not exist, and program startup may
866 not necessarily be at "main". The most obvious example is an OS
867 kernel. This is equivalent to -fno-hosted.
868 -fms-extensions
870 Accept some non-standard constructs used in Microsoft header files.
871 Some cases of unnamed fields in structures and unions are only
873 accepted with this option.
874 -trigraphs
876 Support ISO C trigraphs. The -ansi option (and -std options for
877 strict ISO C conformance) implies -trigraphs.
878 -no-integrated-cpp
880 Performs a compilation in two passes: preprocessing and compiling.
881 This option allows a user supplied "cc1", "cc1plus", or "cc1obj"
882 via the -B option. The user supplied compilation step can then add
883 in an additional preprocessing step after normal preprocessing but
884 before compiling. The default is to use the integrated cpp (inter‐
885 nal cpp)
886 The semantics of this option will change if "cc1", "cc1plus", and
888 "cc1obj" are merged.
889 -traditional
891 -traditional-cpp
892 Formerly, these options caused GCC to attempt to emulate a pre-
893 standard C compiler. They are now only supported with the -E
894 switch. The preprocessor continues to support a pre-standard mode.
895 See the GNU CPP manual for details.
896 -fcond-mismatch
898 Allow conditional expressions with mismatched types in the second
899 and third arguments. The value of such an expression is void.
900 This option is not supported for C++.
901 -funsigned-char
903 Let the type "char" be unsigned, like "unsigned char".
904 Each kind of machine has a default for what "char" should be. It
906 is either like "unsigned char" by default or like "signed char" by
907 default.
908 Ideally, a portable program should always use "signed char" or
910 "unsigned char" when it depends on the signedness of an object.
911 But many programs have been written to use plain "char" and expect
912 it to be signed, or expect it to be unsigned, depending on the
913 machines they were written for. This option, and its inverse, let
914 you make such a program work with the opposite default.
915 The type "char" is always a distinct type from each of "signed
917 char" or "unsigned char", even though its behavior is always just
918 like one of those two.
919 -fsigned-char
921 Let the type "char" be signed, like "signed char".
922 Note that this is equivalent to -fno-unsigned-char, which is the
924 negative form of -funsigned-char. Likewise, the option
925 -fno-signed-char is equivalent to -funsigned-char.
926 -fsigned-bitfields
928 -funsigned-bitfields
929 -fno-signed-bitfields
930 -fno-unsigned-bitfields
931 These options control whether a bit-field is signed or unsigned,
932 when the declaration does not use either "signed" or "unsigned".
933 By default, such a bit-field is signed, because this is consistent:
934 the basic integer types such as "int" are signed types.
935 Options Controlling C++ Dialect
937 This section describes the command-line options that are only meaning‐
939 ful for C++ programs; but you can also use most of the GNU compiler
940 options regardless of what language your program is in. For example,
941 you might compile a file "firstClass.C" like this:
942 g++ -g -frepo -O -c firstClass.C
944 In this example, only -frepo is an option meant only for C++ programs;
946 you can use the other options with any language supported by GCC.
947 Here is a list of options that are only for compiling C++ programs:
949 -fabi-version=n
951 Use version n of the C++ ABI. Version 2 is the version of the C++
952 ABI that first appeared in G++ 3.4. Version 1 is the version of
953 the C++ ABI that first appeared in G++ 3.2. Version 0 will always
954 be the version that conforms most closely to the C++ ABI specifica‐
955 tion. Therefore, the ABI obtained using version 0 will change as
956 ABI bugs are fixed.
957 The default is version 2.
959 -fno-access-control
961 Turn off all access checking. This switch is mainly useful for
962 working around bugs in the access control code.
963 -fcheck-new
965 Check that the pointer returned by "operator new" is non-null
966 before attempting to modify the storage allocated. This check is
967 normally unnecessary because the C++ standard specifies that "oper‐
968 ator new" will only return 0 if it is declared throw(), in which
969 case the compiler will always check the return value even without
970 this option. In all other cases, when "operator new" has a non-
971 empty exception specification, memory exhaustion is signalled by
972 throwing "std::bad_alloc". See also new (nothrow).
973 -fconserve-space
975 Put uninitialized or runtime-initialized global variables into the
976 common segment, as C does. This saves space in the executable at
977 the cost of not diagnosing duplicate definitions. If you compile
978 with this flag and your program mysteriously crashes after "main()"
979 has completed, you may have an object that is being destroyed twice
980 because two definitions were merged.
981 This option is no longer useful on most targets, now that support
983 has been added for putting variables into BSS without making them
984 common.
985 -ffriend-injection
987 Inject friend functions into the enclosing namespace, so that they
988 are visible outside the scope of the class in which they are
989 declared. Friend functions were documented to work this way in the
990 old Annotated C++ Reference Manual, and versions of G++ before 4.1
991 always worked that way. However, in ISO C++ a friend function
992 which is not declared in an enclosing scope can only be found using
993 argument dependent lookup. This option causes friends to be
994 injected as they were in earlier releases.
995 This option is for compatibility, and may be removed in a future
997 release of G++.
998 -fno-const-strings
1000 Give string constants type "char *" instead of type "const char *".
1001 By default, G++ uses type "const char *" as required by the stan‐
1002 dard. Even if you use -fno-const-strings, you cannot actually mod‐
1003 ify the value of a string constant.
1004 This option might be removed in a future release of G++. For maxi‐
1006 mum portability, you should structure your code so that it works
1007 with string constants that have type "const char *".
1008 -fno-elide-constructors
1010 The C++ standard allows an implementation to omit creating a tempo‐
1011 rary which is only used to initialize another object of the same
1012 type. Specifying this option disables that optimization, and
1013 forces G++ to call the copy constructor in all cases.
1014 -fno-enforce-eh-specs
1016 Don't generate code to check for violation of exception specifica‐
1017 tions at runtime. This option violates the C++ standard, but may
1018 be useful for reducing code size in production builds, much like
1019 defining NDEBUG. This does not give user code permission to throw
1020 exceptions in violation of the exception specifications; the com‐
1021 piler will still optimize based on the specifications, so throwing
1022 an unexpected exception will result in undefined behavior.
1023 -ffor-scope
1025 -fno-for-scope
1026 If -ffor-scope is specified, the scope of variables declared in a
1027 for-init-statement is limited to the for loop itself, as specified
1028 by the C++ standard. If -fno-for-scope is specified, the scope of
1029 variables declared in a for-init-statement extends to the end of
1030 the enclosing scope, as was the case in old versions of G++, and
1031 other (traditional) implementations of C++.
1032 The default if neither flag is given to follow the standard, but to
1034 allow and give a warning for old-style code that would otherwise be
1035 invalid, or have different behavior.
1036 -fno-gnu-keywords
1038 Do not recognize "typeof" as a keyword, so that code can use this
1039 word as an identifier. You can use the keyword "__typeof__"
1040 instead. -ansi implies -fno-gnu-keywords.
1041 -fno-implicit-templates
1043 Never emit code for non-inline templates which are instantiated
1044 implicitly (i.e. by use); only emit code for explicit instantia‐
1045 tions.
1046 -fno-implicit-inline-templates
1048 Don't emit code for implicit instantiations of inline templates,
1049 either. The default is to handle inlines differently so that com‐
1050 piles with and without optimization will need the same set of
1051 explicit instantiations.
1052 -fno-implement-inlines
1054 To save space, do not emit out-of-line copies of inline functions
1055 controlled by #pragma implementation. This will cause linker
1056 errors if these functions are not inlined everywhere they are
1057 called.
1058 -fms-extensions
1060 Disable pedantic warnings about constructs used in MFC, such as
1061 implicit int and getting a pointer to member function via non-stan‐
1062 dard syntax.
1063 -fno-nonansi-builtins
1065 Disable built-in declarations of functions that are not mandated by
1066 ANSI/ISO C. These include "ffs", "alloca", "_exit", "index",
1067 "bzero", "conjf", and other related functions.
1068 -fno-operator-names
1070 Do not treat the operator name keywords "and", "bitand", "bitor",
1071 "compl", "not", "or" and "xor" as synonyms as keywords.
1072 -fno-optional-diags
1074 Disable diagnostics that the standard says a compiler does not need
1075 to issue. Currently, the only such diagnostic issued by G++ is the
1076 one for a name having multiple meanings within a class.
1077 -fpermissive
1079 Downgrade some diagnostics about nonconformant code from errors to
1080 warnings. Thus, using -fpermissive will allow some nonconforming
1081 code to compile.
1082 -frepo
1084 Enable automatic template instantiation at link time. This option
1085 also implies -fno-implicit-templates.
1086 -fno-rtti
1088 Disable generation of information about every class with virtual
1089 functions for use by the C++ runtime type identification features
1090 (dynamic_cast and typeid). If you don't use those parts of the
1091 language, you can save some space by using this flag. Note that
1092 exception handling uses the same information, but it will generate
1093 it as needed.
1094 -fstats
1096 Emit statistics about front-end processing at the end of the compi‐
1097 lation. This information is generally only useful to the G++
1098 development team.
1099 -ftemplate-depth-n
1101 Set the maximum instantiation depth for template classes to n. A
1102 limit on the template instantiation depth is needed to detect end‐
1103 less recursions during template class instantiation. ANSI/ISO C++
1104 conforming programs must not rely on a maximum depth greater than
1105 17.
1106 -fno-threadsafe-statics
1108 Do not emit the extra code to use the routines specified in the C++
1109 ABI for thread-safe initialization of local statics. You can use
1110 this option to reduce code size slightly in code that doesn't need
1111 to be thread-safe.
1112 -fuse-cxa-atexit
1114 Register destructors for objects with static storage duration with
1115 the "__cxa_atexit" function rather than the "atexit" function.
1116 This option is required for fully standards-compliant handling of
1117 static destructors, but will only work if your C library supports
1118 "__cxa_atexit".
1119 -fvisibility-inlines-hidden
1121 Causes all inlined methods to be marked with "__attribute__ ((visi‐
1122 bility ("hidden")))" so that they do not appear in the export table
1123 of a DSO and do not require a PLT indirection when used within the
1124 DSO. Enabling this option can have a dramatic effect on load and
1125 link times of a DSO as it massively reduces the size of the dynamic
1126 export table when the library makes heavy use of templates. While
1127 it can cause bloating through duplication of code within each DSO
1128 where it is used, often the wastage is less than the considerable
1129 space occupied by a long symbol name in the export table which is
1130 typical when using templates and namespaces. For even more sav‐
1131 ings, combine with the -fvisibility=hidden switch.
1132 -fno-weak
1134 Do not use weak symbol support, even if it is provided by the
1135 linker. By default, G++ will use weak symbols if they are avail‐
1136 able. This option exists only for testing, and should not be used
1137 by end-users; it will result in inferior code and has no benefits.
1138 This option may be removed in a future release of G++.
1139 -nostdinc++
1141 Do not search for header files in the standard directories specific
1142 to C++, but do still search the other standard directories. (This
1143 option is used when building the C++ library.)
1144 In addition, these optimization, warning, and code generation options
1146 have meanings only for C++ programs:
1147 -fno-default-inline
1149 Do not assume inline for functions defined inside a class scope.
1150 Note that these functions will have linkage like inline func‐
1151 tions; they just won't be inlined by default.
1152 -Wabi (C++ only)
1154 Warn when G++ generates code that is probably not compatible with
1155 the vendor-neutral C++ ABI. Although an effort has been made to
1156 warn about all such cases, there are probably some cases that are
1157 not warned about, even though G++ is generating incompatible code.
1158 There may also be cases where warnings are emitted even though the
1159 code that is generated will be compatible.
1160 You should rewrite your code to avoid these warnings if you are
1162 concerned about the fact that code generated by G++ may not be
1163 binary compatible with code generated by other compilers.
1164 The known incompatibilities at this point include:
1166 * Incorrect handling of tail-padding for bit-fields. G++ may
1168 attempt to pack data into the same byte as a base class. For
1169 example:
1170 struct A { virtual void f(); int f1 : 1; };
1172 struct B : public A { int f2 : 1; };
1173 In this case, G++ will place "B::f2" into the same byte
1175 as"A::f1"; other compilers will not. You can avoid this prob‐
1176 lem by explicitly padding "A" so that its size is a multiple of
1177 the byte size on your platform; that will cause G++ and other
1178 compilers to layout "B" identically.
1179 * Incorrect handling of tail-padding for virtual bases. G++ does
1181 not use tail padding when laying out virtual bases. For exam‐
1182 ple:
1183 struct A { virtual void f(); char c1; };
1185 struct B { B(); char c2; };
1186 struct C : public A, public virtual B {};
1187 In this case, G++ will not place "B" into the tail-padding for
1189 "A"; other compilers will. You can avoid this problem by
1190 explicitly padding "A" so that its size is a multiple of its
1191 alignment (ignoring virtual base classes); that will cause G++
1192 and other compilers to layout "C" identically.
1193 * Incorrect handling of bit-fields with declared widths greater
1195 than that of their underlying types, when the bit-fields appear
1196 in a union. For example:
1197 union U { int i : 4096; };
1199 Assuming that an "int" does not have 4096 bits, G++ will make
1201 the union too small by the number of bits in an "int".
1202 * Empty classes can be placed at incorrect offsets. For example:
1204 struct A {};
1206 struct B {
1208 A a;
1209 virtual void f ();
1210 };
1211 struct C : public B, public A {};
1213 G++ will place the "A" base class of "C" at a nonzero offset;
1215 it should be placed at offset zero. G++ mistakenly believes
1216 that the "A" data member of "B" is already at offset zero.
1217 * Names of template functions whose types involve "typename" or
1219 template template parameters can be mangled incorrectly.
1220 template <typename Q>
1222 void f(typename Q::X) {}
1223 template <template <typename> class Q>
1225 void f(typename Q<int>::X) {}
1226 Instantiations of these templates may be mangled incorrectly.
1228 -Wctor-dtor-privacy (C++ only)
1230 Warn when a class seems unusable because all the constructors or
1231 destructors in that class are private, and it has neither friends
1232 nor public static member functions.
1233 -Wnon-virtual-dtor (C++ only)
1235 Warn when a class appears to be polymorphic, thereby requiring a
1236 virtual destructor, yet it declares a non-virtual one. This warn‐
1237 ing is enabled by -Wall.
1238 -Wreorder (C++ only)
1240 Warn when the order of member initializers given in the code does
1241 not match the order in which they must be executed. For instance:
1242 struct A {
1244 int i;
1245 int j;
1246 A(): j (0), i (1) { }
1247 };
1248 The compiler will rearrange the member initializers for i and j to
1250 match the declaration order of the members, emitting a warning to
1251 that effect. This warning is enabled by -Wall.
1252 The following -W... options are not affected by -Wall.
1254 -Weffc++ (C++ only)
1256 Warn about violations of the following style guidelines from Scott
1257 Meyers' Effective C++ book:
1258 * Item 11: Define a copy constructor and an assignment operator
1260 for classes with dynamically allocated memory.
1261 * Item 12: Prefer initialization to assignment in constructors.
1263 * Item 14: Make destructors virtual in base classes.
1265 * Item 15: Have "operator=" return a reference to *this.
1267 * Item 23: Don't try to return a reference when you must return
1269 an object.
1270 Also warn about violations of the following style guidelines from
1272 Scott Meyers' More Effective C++ book:
1273 * Item 6: Distinguish between prefix and postfix forms of incre‐
1275 ment and decrement operators.
1276 * Item 7: Never overload "&&", "⎪⎪", or ",".
1278 When selecting this option, be aware that the standard library
1280 headers do not obey all of these guidelines; use grep -v to filter
1281 out those warnings.
1282 -Wno-deprecated (C++ only)
1284 Do not warn about usage of deprecated features.
1285 -Wstrict-null-sentinel (C++ only)
1287 Warn also about the use of an uncasted "NULL" as sentinel. When
1288 compiling only with GCC this is a valid sentinel, as "NULL" is
1289 defined to "__null". Although it is a null pointer constant not a
1290 null pointer, it is guaranteed to of the same size as a pointer.
1291 But this use is not portable across different compilers.
1292 -Wno-non-template-friend (C++ only)
1294 Disable warnings when non-templatized friend functions are declared
1295 within a template. Since the advent of explicit template specifi‐
1296 cation support in G++, if the name of the friend is an unqualified-
1297 id (i.e., friend foo(int)), the C++ language specification demands
1298 that the friend declare or define an ordinary, nontemplate func‐
1299 tion. (Section 14.5.3). Before G++ implemented explicit specifi‐
1300 cation, unqualified-ids could be interpreted as a particular spe‐
1301 cialization of a templatized function. Because this non-conforming
1302 behavior is no longer the default behavior for G++, -Wnon-tem‐
1303 plate-friend allows the compiler to check existing code for poten‐
1304 tial trouble spots and is on by default. This new compiler behav‐
1305 ior can be turned off with -Wno-non-template-friend which keeps the
1306 conformant compiler code but disables the helpful warning.
1307 -Wold-style-cast (C++ only)
1309 Warn if an old-style (C-style) cast to a non-void type is used
1310 within a C++ program. The new-style casts (dynamic_cast,
1311 static_cast, reinterpret_cast, and const_cast) are less vulnerable
1312 to unintended effects and much easier to search for.
1313 -Woverloaded-virtual (C++ only)
1315 Warn when a function declaration hides virtual functions from a
1316 base class. For example, in:
1317 struct A {
1319 virtual void f();
1320 };
1321 struct B: public A {
1323 void f(int);
1324 };
1325 the "A" class version of "f" is hidden in "B", and code like:
1327 B* b;
1329 b->f();
1330 will fail to compile.
1332 -Wno-pmf-conversions (C++ only)
1334 Disable the diagnostic for converting a bound pointer to member
1335 function to a plain pointer.
1336 -Wsign-promo (C++ only)
1338 Warn when overload resolution chooses a promotion from unsigned or
1339 enumerated type to a signed type, over a conversion to an unsigned
1340 type of the same size. Previous versions of G++ would try to pre‐
1341 serve unsignedness, but the standard mandates the current behavior.
1342 struct A {
1344 operator int ();
1345 A& operator = (int);
1346 };
1347 main ()
1349 {
1350 A a,b;
1351 a = b;
1352 }
1353 In this example, G++ will synthesize a default A& operator = (const
1355 A&);, while cfront will use the user-defined operator =.
1356 Options Controlling Objective-C and Objective-C++ Dialects
1358 (NOTE: This manual does not describe the Objective-C and Objective-C++
1360 languages themselves. See
1361 This section describes the command-line options that are only meaning‐
1363 ful for Objective-C and Objective-C++ programs, but you can also use
1364 most of the language-independent GNU compiler options. For example,
1365 you might compile a file "some_class.m" like this:
1366 gcc -g -fgnu-runtime -O -c some_class.m
1368 In this example, -fgnu-runtime is an option meant only for Objective-C
1370 and Objective-C++ programs; you can use the other options with any lan‐
1371 guage supported by GCC.
1372 Note that since Objective-C is an extension of the C language, Objec‐
1374 tive-C compilations may also use options specific to the C front-end
1375 (e.g., -Wtraditional). Similarly, Objective-C++ compilations may use
1376 C++-specific options (e.g., -Wabi).
1377 Here is a list of options that are only for compiling Objective-C and
1379 Objective-C++ programs:
1380 -fconstant-string-class=class-name
1382 Use class-name as the name of the class to instantiate for each
1383 literal string specified with the syntax "@"..."". The default
1384 class name is "NXConstantString" if the GNU runtime is being used,
1385 and "NSConstantString" if the NeXT runtime is being used (see
1386 below). The -fconstant-cfstrings option, if also present, will
1387 override the -fconstant-string-class setting and cause "@"...""
1388 literals to be laid out as constant CoreFoundation strings.
1389 -fgnu-runtime
1391 Generate object code compatible with the standard GNU Objective-C
1392 runtime. This is the default for most types of systems.
1393 -fnext-runtime
1395 Generate output compatible with the NeXT runtime. This is the
1396 default for NeXT-based systems, including Darwin and Mac OS X. The
1397 macro "__NEXT_RUNTIME__" is predefined if (and only if) this option
1398 is used.
1399 -fno-nil-receivers
1401 Assume that all Objective-C message dispatches (e.g., "[receiver
1402 message:arg]") in this translation unit ensure that the receiver is
1403 not "nil". This allows for more efficient entry points in the run‐
1404 time to be used. Currently, this option is only available in con‐
1405 junction with the NeXT runtime on Mac OS X 10.3 and later.
1406 -fobjc-call-cxx-cdtors
1408 For each Objective-C class, check if any of its instance variables
1409 is a C++ object with a non-trivial default constructor. If so,
1410 synthesize a special "- (id) .cxx_construct" instance method that
1411 will run non-trivial default constructors on any such instance
1412 variables, in order, and then return "self". Similarly, check if
1413 any instance variable is a C++ object with a non-trivial destruc‐
1414 tor, and if so, synthesize a special "- (void) .cxx_destruct"
1415 method that will run all such default destructors, in reverse
1416 order.
1417 The "- (id) .cxx_construct" and/or "- (void) .cxx_destruct" methods
1419 thusly generated will only operate on instance variables declared
1420 in the current Objective-C class, and not those inherited from
1421 superclasses. It is the responsibility of the Objective-C runtime
1422 to invoke all such methods in an object's inheritance hierarchy.
1423 The "- (id) .cxx_construct" methods will be invoked by the runtime
1424 immediately after a new object instance is allocated; the "- (void)
1425 .cxx_destruct" methods will be invoked immediately before the run‐
1426 time deallocates an object instance.
1427 As of this writing, only the NeXT runtime on Mac OS X 10.4 and
1429 later has support for invoking the "- (id) .cxx_construct" and "-
1430 (void) .cxx_destruct" methods.
1431 -fobjc-direct-dispatch
1433 Allow fast jumps to the message dispatcher. On Darwin this is
1434 accomplished via the comm page.
1435 -fobjc-exceptions
1437 Enable syntactic support for structured exception handling in
1438 Objective-C, similar to what is offered by C++ and Java. Cur‐
1439 rently, this option is only available in conjunction with the NeXT
1440 runtime on Mac OS X 10.3 and later.
1441 @try {
1443 ...
1444 @throw expr;
1445 ...
1446 }
1447 @catch (AnObjCClass *exc) {
1448 ...
1449 @throw expr;
1450 ...
1451 @throw;
1452 ...
1453 }
1454 @catch (AnotherClass *exc) {
1455 ...
1456 }
1457 @catch (id allOthers) {
1458 ...
1459 }
1460 @finally {
1461 ...
1462 @throw expr;
1463 ...
1464 }
1465 The @throw statement may appear anywhere in an Objective-C or
1467 Objective-C++ program; when used inside of a @catch block, the
1468 @throw may appear without an argument (as shown above), in which
1469 case the object caught by the @catch will be rethrown.
1470 Note that only (pointers to) Objective-C objects may be thrown and
1472 caught using this scheme. When an object is thrown, it will be
1473 caught by the nearest @catch clause capable of handling objects of
1474 that type, analogously to how "catch" blocks work in C++ and Java.
1475 A "@catch(id ...)" clause (as shown above) may also be provided to
1476 catch any and all Objective-C exceptions not caught by previous
1477 @catch clauses (if any).
1478 The @finally clause, if present, will be executed upon exit from
1480 the immediately preceding "@try ... @catch" section. This will
1481 happen regardless of whether any exceptions are thrown, caught or
1482 rethrown inside the "@try ... @catch" section, analogously to the
1483 behavior of the "finally" clause in Java.
1484 There are several caveats to using the new exception mechanism:
1486 * Although currently designed to be binary compatible with
1488 "NS_HANDLER"-style idioms provided by the "NSException" class,
1489 the new exceptions can only be used on Mac OS X 10.3 (Panther)
1490 and later systems, due to additional functionality needed in
1491 the (NeXT) Objective-C runtime.
1492 * As mentioned above, the new exceptions do not support handling
1494 types other than Objective-C objects. Furthermore, when used
1495 from Objective-C++, the Objective-C exception model does not
1496 interoperate with C++ exceptions at this time. This means you
1497 cannot @throw an exception from Objective-C and "catch" it in
1498 C++, or vice versa (i.e., "throw ... @catch").
1499 The -fobjc-exceptions switch also enables the use of synchroniza‐
1501 tion blocks for thread-safe execution:
1502 @synchronized (ObjCClass *guard) {
1504 ...
1505 }
1506 Upon entering the @synchronized block, a thread of execution shall
1508 first check whether a lock has been placed on the corresponding
1509 "guard" object by another thread. If it has, the current thread
1510 shall wait until the other thread relinquishes its lock. Once
1511 "guard" becomes available, the current thread will place its own
1512 lock on it, execute the code contained in the @synchronized block,
1513 and finally relinquish the lock (thereby making "guard" available
1514 to other threads).
1515 Unlike Java, Objective-C does not allow for entire methods to be
1517 marked @synchronized. Note that throwing exceptions out of @syn‐
1518 chronized blocks is allowed, and will cause the guarding object to
1519 be unlocked properly.
1520 -fobjc-gc
1522 Enable garbage collection (GC) in Objective-C and Objective-C++
1523 programs.
1524 -freplace-objc-classes
1526 Emit a special marker instructing ld(1) not to statically link in
1527 the resulting object file, and allow dyld(1) to load it in at run
1528 time instead. This is used in conjunction with the Fix-and-Con‐
1529 tinue debugging mode, where the object file in question may be
1530 recompiled and dynamically reloaded in the course of program execu‐
1531 tion, without the need to restart the program itself. Currently,
1532 Fix-and-Continue functionality is only available in conjunction
1533 with the NeXT runtime on Mac OS X 10.3 and later.
1534 -fzero-link
1536 When compiling for the NeXT runtime, the compiler ordinarily
1537 replaces calls to "objc_getClass("...")" (when the name of the
1538 class is known at compile time) with static class references that
1539 get initialized at load time, which improves run-time performance.
1540 Specifying the -fzero-link flag suppresses this behavior and causes
1541 calls to "objc_getClass("...")" to be retained. This is useful in
1542 Zero-Link debugging mode, since it allows for individual class
1543 implementations to be modified during program execution.
1544 -gen-decls
1546 Dump interface declarations for all classes seen in the source file
1547 to a file named sourcename.decl.
1548 -Wassign-intercept
1550 Warn whenever an Objective-C assignment is being intercepted by the
1551 garbage collector.
1552 -Wno-protocol
1554 If a class is declared to implement a protocol, a warning is issued
1555 for every method in the protocol that is not implemented by the
1556 class. The default behavior is to issue a warning for every method
1557 not explicitly implemented in the class, even if a method implemen‐
1558 tation is inherited from the superclass. If you use the -Wno-pro‐
1559 tocol option, then methods inherited from the superclass are con‐
1560 sidered to be implemented, and no warning is issued for them.
1561 -Wselector
1563 Warn if multiple methods of different types for the same selector
1564 are found during compilation. The check is performed on the list
1565 of methods in the final stage of compilation. Additionally, a
1566 check is performed for each selector appearing in a "@selec‐
1567 tor(...)" expression, and a corresponding method for that selector
1568 has been found during compilation. Because these checks scan the
1569 method table only at the end of compilation, these warnings are not
1570 produced if the final stage of compilation is not reached, for
1571 example because an error is found during compilation, or because
1572 the -fsyntax-only option is being used.
1573 -Wstrict-selector-match
1575 Warn if multiple methods with differing argument and/or return
1576 types are found for a given selector when attempting to send a mes‐
1577 sage using this selector to a receiver of type "id" or "Class".
1578 When this flag is off (which is the default behavior), the compiler
1579 will omit such warnings if any differences found are confined to
1580 types which share the same size and alignment.
1581 -Wundeclared-selector
1583 Warn if a "@selector(...)" expression referring to an undeclared
1584 selector is found. A selector is considered undeclared if no
1585 method with that name has been declared before the "@selector(...)"
1586 expression, either explicitly in an @interface or @protocol decla‐
1587 ration, or implicitly in an @implementation section. This option
1588 always performs its checks as soon as a "@selector(...)" expression
1589 is found, while -Wselector only performs its checks in the final
1590 stage of compilation. This also enforces the coding style conven‐
1591 tion that methods and selectors must be declared before being used.
1592 -print-objc-runtime-info
1594 Generate C header describing the largest structure that is passed
1595 by value, if any.
1596 Options to Control Diagnostic Messages Formatting
1598 Traditionally, diagnostic messages have been formatted irrespective of
1600 the output device's aspect (e.g. its width, ...). The options
1601 described below can be used to control the diagnostic messages format‐
1602 ting algorithm, e.g. how many characters per line, how often source
1603 location information should be reported. Right now, only the C++ front
1604 end can honor these options. However it is expected, in the near
1605 future, that the remaining front ends would be able to digest them cor‐
1606 rectly.
1607 -fmessage-length=n
1609 Try to format error messages so that they fit on lines of about n
1610 characters. The default is 72 characters for g++ and 0 for the
1611 rest of the front ends supported by GCC. If n is zero, then no
1612 line-wrapping will be done; each error message will appear on a
1613 single line.
1614 -fdiagnostics-show-location=once
1616 Only meaningful in line-wrapping mode. Instructs the diagnostic
1617 messages reporter to emit once source location information; that
1618 is, in case the message is too long to fit on a single physical
1619 line and has to be wrapped, the source location won't be emitted
1620 (as prefix) again, over and over, in subsequent continuation lines.
1621 This is the default behavior.
1622 -fdiagnostics-show-location=every-line
1624 Only meaningful in line-wrapping mode. Instructs the diagnostic
1625 messages reporter to emit the same source location information (as
1626 prefix) for physical lines that result from the process of breaking
1627 a message which is too long to fit on a single line.
1628 -fdiagnostics-show-options
1630 This option instructs the diagnostic machinery to add text to each
1631 diagnostic emitted, which indicates which command line option
1632 directly controls that diagnostic, when such an option is known to
1633 the diagnostic machinery.
1634 Options to Request or Suppress Warnings
1636 Warnings are diagnostic messages that report constructions which are
1638 not inherently erroneous but which are risky or suggest there may have
1639 been an error.
1640 You can request many specific warnings with options beginning -W, for
1642 example -Wimplicit to request warnings on implicit declarations. Each
1643 of these specific warning options also has a negative form beginning
1644 -Wno- to turn off warnings; for example, -Wno-implicit. This manual
1645 lists only one of the two forms, whichever is not the default.
1646 The following options control the amount and kinds of warnings produced
1648 by GCC; for further, language-specific options also refer to C++
1649 Dialect Options and Objective-C and Objective-C++ Dialect Options.
1650 -fsyntax-only
1652 Check the code for syntax errors, but don't do anything beyond
1653 that.
1654 -pedantic
1656 Issue all the warnings demanded by strict ISO C and ISO C++; reject
1657 all programs that use forbidden extensions, and some other programs
1658 that do not follow ISO C and ISO C++. For ISO C, follows the ver‐
1659 sion of the ISO C standard specified by any -std option used.
1660 Valid ISO C and ISO C++ programs should compile properly with or
1662 without this option (though a rare few will require -ansi or a -std
1663 option specifying the required version of ISO C). However, without
1664 this option, certain GNU extensions and traditional C and C++ fea‐
1665 tures are supported as well. With this option, they are rejected.
1666 -pedantic does not cause warning messages for use of the alternate
1668 keywords whose names begin and end with __. Pedantic warnings are
1669 also disabled in the expression that follows "__extension__". How‐
1670 ever, only system header files should use these escape routes;
1671 application programs should avoid them.
1672 Some users try to use -pedantic to check programs for strict ISO C
1674 conformance. They soon find that it does not do quite what they
1675 want: it finds some non-ISO practices, but not all---only those for
1676 which ISO C requires a diagnostic, and some others for which diag‐
1677 nostics have been added.
1678 A feature to report any failure to conform to ISO C might be useful
1680 in some instances, but would require considerable additional work
1681 and would be quite different from -pedantic. We don't have plans
1682 to support such a feature in the near future.
1683 Where the standard specified with -std represents a GNU extended
1685 dialect of C, such as gnu89 or gnu99, there is a corresponding base
1686 standard, the version of ISO C on which the GNU extended dialect is
1687 based. Warnings from -pedantic are given where they are required
1688 by the base standard. (It would not make sense for such warnings
1689 to be given only for features not in the specified GNU C dialect,
1690 since by definition the GNU dialects of C include all features the
1691 compiler supports with the given option, and there would be nothing
1692 to warn about.)
1693 -pedantic-errors
1695 Like -pedantic, except that errors are produced rather than warn‐
1696 ings.
1697 -w Inhibit all warning messages.
1699 -Wno-import
1701 Inhibit warning messages about the use of #import.
1702 -Wchar-subscripts
1704 Warn if an array subscript has type "char". This is a common cause
1705 of error, as programmers often forget that this type is signed on
1706 some machines. This warning is enabled by -Wall.
1707 -Wcomment
1709 Warn whenever a comment-start sequence /* appears in a /* comment,
1710 or whenever a Backslash-Newline appears in a // comment. This
1711 warning is enabled by -Wall.
1712 -Wfatal-errors
1714 This option causes the compiler to abort compilation on the first
1715 error occurred rather than trying to keep going and printing fur‐
1716 ther error messages.
1717 -Wformat
1719 Check calls to "printf" and "scanf", etc., to make sure that the
1720 arguments supplied have types appropriate to the format string
1721 specified, and that the conversions specified in the format string
1722 make sense. This includes standard functions, and others specified
1723 by format attributes, in the "printf", "scanf", "strftime" and
1724 "strfmon" (an X/Open extension, not in the C standard) families (or
1725 other target-specific families). Which functions are checked with‐
1726 out format attributes having been specified depends on the standard
1727 version selected, and such checks of functions without the
1728 attribute specified are disabled by -ffreestanding or -fno-builtin.
1729 The formats are checked against the format features supported by
1731 GNU libc version 2.2. These include all ISO C90 and C99 features,
1732 as well as features from the Single Unix Specification and some BSD
1733 and GNU extensions. Other library implementations may not support
1734 all these features; GCC does not support warning about features
1735 that go beyond a particular library's limitations. However, if
1736 -pedantic is used with -Wformat, warnings will be given about for‐
1737 mat features not in the selected standard version (but not for
1738 "strfmon" formats, since those are not in any version of the C
1739 standard).
1740 Since -Wformat also checks for null format arguments for several
1742 functions, -Wformat also implies -Wnonnull.
1743 -Wformat is included in -Wall. For more control over some aspects
1745 of format checking, the options -Wformat-y2k, -Wno-for‐
1746 mat-extra-args, -Wno-format-zero-length, -Wformat-nonliteral,
1747 -Wformat-security, and -Wformat=2 are available, but are not
1748 included in -Wall.
1749 -Wformat-y2k
1751 If -Wformat is specified, also warn about "strftime" formats which
1752 may yield only a two-digit year.
1753 -Wno-format-extra-args
1755 If -Wformat is specified, do not warn about excess arguments to a
1756 "printf" or "scanf" format function. The C standard specifies that
1757 such arguments are ignored.
1758 Where the unused arguments lie between used arguments that are
1760 specified with $ operand number specifications, normally warnings
1761 are still given, since the implementation could not know what type
1762 to pass to "va_arg" to skip the unused arguments. However, in the
1763 case of "scanf" formats, this option will suppress the warning if
1764 the unused arguments are all pointers, since the Single Unix Speci‐
1765 fication says that such unused arguments are allowed.
1766 -Wno-format-zero-length
1768 If -Wformat is specified, do not warn about zero-length formats.
1769 The C standard specifies that zero-length formats are allowed.
1770 -Wformat-nonliteral
1772 If -Wformat is specified, also warn if the format string is not a
1773 string literal and so cannot be checked, unless the format function
1774 takes its format arguments as a "va_list".
1775 -Wformat-security
1777 If -Wformat is specified, also warn about uses of format functions
1778 that represent possible security problems. At present, this warns
1779 about calls to "printf" and "scanf" functions where the format
1780 string is not a string literal and there are no format arguments,
1781 as in "printf (foo);". This may be a security hole if the format
1782 string came from untrusted input and contains %n. (This is cur‐
1783 rently a subset of what -Wformat-nonliteral warns about, but in
1784 future warnings may be added to -Wformat-security that are not
1785 included in -Wformat-nonliteral.)
1786 -Wformat=2
1788 Enable -Wformat plus format checks not included in -Wformat. Cur‐
1789 rently equivalent to -Wformat -Wformat-nonliteral -Wformat-security
1790 -Wformat-y2k.
1791 -Wnonnull
1793 Warn about passing a null pointer for arguments marked as requiring
1794 a non-null value by the "nonnull" function attribute.
1795 -Wnonnull is included in -Wall and -Wformat. It can be disabled
1797 with the -Wno-nonnull option.
1798 -Winit-self (C, C++, Objective-C and Objective-C++ only)
1800 Warn about uninitialized variables which are initialized with them‐
1801 selves. Note this option can only be used with the -Wuninitialized
1802 option, which in turn only works with -O1 and above.
1803 For example, GCC will warn about "i" being uninitialized in the
1805 following snippet only when -Winit-self has been specified:
1806 int f()
1808 {
1809 int i = i;
1810 return i;
1811 }
1812 -Wimplicit-int
1814 Warn when a declaration does not specify a type. This warning is
1815 enabled by -Wall.
1816 -Wimplicit-function-declaration
1818 -Werror-implicit-function-declaration
1819 Give a warning (or error) whenever a function is used before being
1820 declared. The form -Wno-error-implicit-function-declaration is not
1821 supported. This warning is enabled by -Wall (as a warning, not an
1822 error).
1823 -Wimplicit
1825 Same as -Wimplicit-int and -Wimplicit-function-declaration. This
1826 warning is enabled by -Wall.
1827 -Wmain
1829 Warn if the type of main is suspicious. main should be a function
1830 with external linkage, returning int, taking either zero arguments,
1831 two, or three arguments of appropriate types. This warning is
1832 enabled by -Wall.
1833 -Wmissing-braces
1835 Warn if an aggregate or union initializer is not fully bracketed.
1836 In the following example, the initializer for a is not fully brack‐
1837 eted, but that for b is fully bracketed.
1838 int a[2][2] = { 0, 1, 2, 3 };
1840 int b[2][2] = { { 0, 1 }, { 2, 3 } };
1841 This warning is enabled by -Wall.
1843 -Wmissing-include-dirs (C, C++, Objective-C and Objective-C++ only)
1845 Warn if a user-supplied include directory does not exist.
1846 -Wparentheses
1848 Warn if parentheses are omitted in certain contexts, such as when
1849 there is an assignment in a context where a truth value is
1850 expected, or when operators are nested whose precedence people
1851 often get confused about. Only the warning for an assignment used
1852 as a truth value is supported when compiling C++; the other warn‐
1853 ings are only supported when compiling C.
1854 Also warn if a comparison like x<=y<=z appears; this is equivalent
1856 to (x<=y ? 1 : 0) <= z, which is a different interpretation from
1857 that of ordinary mathematical notation.
1858 Also warn about constructions where there may be confusion to which
1860 "if" statement an "else" branch belongs. Here is an example of
1861 such a case:
1862 {
1864 if (a)
1865 if (b)
1866 foo ();
1867 else
1868 bar ();
1869 }
1870 In C, every "else" branch belongs to the innermost possible "if"
1872 statement, which in this example is "if (b)". This is often not
1873 what the programmer expected, as illustrated in the above example
1874 by indentation the programmer chose. When there is the potential
1875 for this confusion, GCC will issue a warning when this flag is
1876 specified. To eliminate the warning, add explicit braces around
1877 the innermost "if" statement so there is no way the "else" could
1878 belong to the enclosing "if". The resulting code would look like
1879 this:
1880 {
1882 if (a)
1883 {
1884 if (b)
1885 foo ();
1886 else
1887 bar ();
1888 }
1889 }
1890 This warning is enabled by -Wall.
1892 -Wsequence-point
1894 Warn about code that may have undefined semantics because of viola‐
1895 tions of sequence point rules in the C standard.
1896 The C standard defines the order in which expressions in a C pro‐
1898 gram are evaluated in terms of sequence points, which represent a
1899 partial ordering between the execution of parts of the program:
1900 those executed before the sequence point, and those executed after
1901 it. These occur after the evaluation of a full expression (one
1902 which is not part of a larger expression), after the evaluation of
1903 the first operand of a "&&", "⎪⎪", "? :" or "," (comma) operator,
1904 before a function is called (but after the evaluation of its argu‐
1905 ments and the expression denoting the called function), and in cer‐
1906 tain other places. Other than as expressed by the sequence point
1907 rules, the order of evaluation of subexpressions of an expression
1908 is not specified. All these rules describe only a partial order
1909 rather than a total order, since, for example, if two functions are
1910 called within one expression with no sequence point between them,
1911 the order in which the functions are called is not specified. How‐
1912 ever, the standards committee have ruled that function calls do not
1913 overlap.
1914 It is not specified when between sequence points modifications to
1916 the values of objects take effect. Programs whose behavior depends
1917 on this have undefined behavior; the C standard specifies that
1918 "Between the previous and next sequence point an object shall have
1919 its stored value modified at most once by the evaluation of an
1920 expression. Furthermore, the prior value shall be read only to
1921 determine the value to be stored.". If a program breaks these
1922 rules, the results on any particular implementation are entirely
1923 unpredictable.
1924 Examples of code with undefined behavior are "a = a++;", "a[n] =
1926 b[n++]" and "a[i++] = i;". Some more complicated cases are not
1927 diagnosed by this option, and it may give an occasional false posi‐
1928 tive result, but in general it has been found fairly effective at
1929 detecting this sort of problem in programs.
1930 The present implementation of this option only works for C pro‐
1932 grams. A future implementation may also work for C++ programs.
1933 The C standard is worded confusingly, therefore there is some
1935 debate over the precise meaning of the sequence point rules in sub‐
1936 tle cases. Links to discussions of the problem, including proposed
1937 formal definitions, may be found on the GCC readings page, at
1938 <http://gcc.gnu.org/readings.html>.
1939 This warning is enabled by -Wall.
1941 -Wreturn-type
1943 Warn whenever a function is defined with a return-type that
1944 defaults to "int". Also warn about any "return" statement with no
1945 return-value in a function whose return-type is not "void".
1946 For C, also warn if the return type of a function has a type quali‐
1948 fier such as "const". Such a type qualifier has no effect, since
1949 the value returned by a function is not an lvalue. ISO C prohibits
1950 qualified "void" return types on function definitions, so such
1951 return types always receive a warning even without this option.
1952 For C++, a function without return type always produces a diagnos‐
1954 tic message, even when -Wno-return-type is specified. The only
1955 exceptions are main and functions defined in system headers.
1956 This warning is enabled by -Wall.
1958 -Wswitch
1960 Warn whenever a "switch" statement has an index of enumerated type
1961 and lacks a "case" for one or more of the named codes of that enu‐
1962 meration. (The presence of a "default" label prevents this warn‐
1963 ing.) "case" labels outside the enumeration range also provoke
1964 warnings when this option is used. This warning is enabled by
1965 -Wall.
1966 -Wswitch-default
1968 Warn whenever a "switch" statement does not have a "default" case.
1969 -Wswitch-enum
1971 Warn whenever a "switch" statement has an index of enumerated type
1972 and lacks a "case" for one or more of the named codes of that enu‐
1973 meration. "case" labels outside the enumeration range also provoke
1974 warnings when this option is used.
1975 -Wtrigraphs
1977 Warn if any trigraphs are encountered that might change the meaning
1978 of the program (trigraphs within comments are not warned about).
1979 This warning is enabled by -Wall.
1980 -Wunused-function
1982 Warn whenever a static function is declared but not defined or a
1983 non-inline static function is unused. This warning is enabled by
1984 -Wall.
1985 -Wunused-label
1987 Warn whenever a label is declared but not used. This warning is
1988 enabled by -Wall.
1989 To suppress this warning use the unused attribute.
1991 -Wunused-parameter
1993 Warn whenever a function parameter is unused aside from its decla‐
1994 ration.
1995 To suppress this warning use the unused attribute.
1997 -Wunused-variable
1999 Warn whenever a local variable or non-constant static variable is
2000 unused aside from its declaration. This warning is enabled by
2001 -Wall.
2002 To suppress this warning use the unused attribute.
2004 -Wunused-value
2006 Warn whenever a statement computes a result that is explicitly not
2007 used. This warning is enabled by -Wall.
2008 To suppress this warning cast the expression to void.
2010 -Wunused
2012 All the above -Wunused options combined.
2013 In order to get a warning about an unused function parameter, you
2015 must either specify -Wextra -Wunused (note that -Wall implies
2016 -Wunused), or separately specify -Wunused-parameter.
2017 -Wuninitialized
2019 Warn if an automatic variable is used without first being initial‐
2020 ized or if a variable may be clobbered by a "setjmp" call.
2021 These warnings are possible only in optimizing compilation, because
2023 they require data flow information that is computed only when opti‐
2024 mizing. If you don't specify -O, you simply won't get these warn‐
2025 ings.
2026 If you want to warn about code which uses the uninitialized value
2028 of the variable in its own initializer, use the -Winit-self option.
2029 These warnings occur for individual uninitialized or clobbered ele‐
2031 ments of structure, union or array variables as well as for vari‐
2032 ables which are uninitialized or clobbered as a whole. They do not
2033 occur for variables or elements declared "volatile". Because these
2034 warnings depend on optimization, the exact variables or elements
2035 for which there are warnings will depend on the precise optimiza‐
2036 tion options and version of GCC used.
2037 Note that there may be no warning about a variable that is used
2039 only to compute a value that itself is never used, because such
2040 computations may be deleted by data flow analysis before the warn‐
2041 ings are printed.
2042 These warnings are made optional because GCC is not smart enough to
2044 see all the reasons why the code might be correct despite appearing
2045 to have an error. Here is one example of how this can happen:
2046 {
2048 int x;
2049 switch (y)
2050 {
2051 case 1: x = 1;
2052 break;
2053 case 2: x = 4;
2054 break;
2055 case 3: x = 5;
2056 }
2057 foo (x);
2058 }
2059 If the value of "y" is always 1, 2 or 3, then "x" is always ini‐
2061 tialized, but GCC doesn't know this. Here is another common case:
2062 {
2064 int save_y;
2065 if (change_y) save_y = y, y = new_y;
2066 ...
2067 if (change_y) y = save_y;
2068 }
2069 This has no bug because "save_y" is used only if it is set.
2071 This option also warns when a non-volatile automatic variable might
2073 be changed by a call to "longjmp". These warnings as well are pos‐
2074 sible only in optimizing compilation.
2075 The compiler sees only the calls to "setjmp". It cannot know where
2077 "longjmp" will be called; in fact, a signal handler could call it
2078 at any point in the code. As a result, you may get a warning even
2079 when there is in fact no problem because "longjmp" cannot in fact
2080 be called at the place which would cause a problem.
2081 Some spurious warnings can be avoided if you declare all the func‐
2083 tions you use that never return as "noreturn".
2084 This warning is enabled by -Wall.
2086 -Wunknown-pragmas
2088 Warn when a #pragma directive is encountered which is not under‐
2089 stood by GCC. If this command line option is used, warnings will
2090 even be issued for unknown pragmas in system header files. This is
2091 not the case if the warnings were only enabled by the -Wall command
2092 line option.
2093 -Wno-pragmas
2095 Do not warn about misuses of pragmas, such as incorrect parameters,
2096 invalid syntax, or conflicts between pragmas. See also -Wun‐
2097 known-pragmas.
2098 -Wstrict-aliasing
2100 This option is only active when -fstrict-aliasing is active. It
2101 warns about code which might break the strict aliasing rules that
2102 the compiler is using for optimization. The warning does not catch
2103 all cases, but does attempt to catch the more common pitfalls. It
2104 is included in -Wall.
2105 -Wstrict-aliasing=2
2107 This option is only active when -fstrict-aliasing is active. It
2108 warns about code which might break the strict aliasing rules that
2109 the compiler is using for optimization. This warning catches more
2110 cases than -Wstrict-aliasing, but it will also give a warning for
2111 some ambiguous cases that are safe.
2112 -Wall
2114 All of the above -W options combined. This enables all the warn‐
2115 ings about constructions that some users consider questionable, and
2116 that are easy to avoid (or modify to prevent the warning), even in
2117 conjunction with macros. This also enables some language-specific
2118 warnings described in C++ Dialect Options and Objective-C and
2119 Objective-C++ Dialect Options.
2120 The following -W... options are not implied by -Wall. Some of them
2122 warn about constructions that users generally do not consider question‐
2123 able, but which occasionally you might wish to check for; others warn
2124 about constructions that are necessary or hard to avoid in some cases,
2125 and there is no simple way to modify the code to suppress the warning.
2126 -Wextra
2128 (This option used to be called -W. The older name is still sup‐
2129 ported, but the newer name is more descriptive.) Print extra warn‐
2130 ing messages for these events:
2131 * A function can return either with or without a value. (Falling
2133 off the end of the function body is considered returning with‐
2134 out a value.) For example, this function would evoke such a
2135 warning:
2136 foo (a)
2138 {
2139 if (a > 0)
2140 return a;
2141 }
2142 * An expression-statement or the left-hand side of a comma
2144 expression contains no side effects. To suppress the warning,
2145 cast the unused expression to void. For example, an expression
2146 such as x[i,j] will cause a warning, but x[(void)i,j] will not.
2147 * An unsigned value is compared against zero with < or >=.
2149 * Storage-class specifiers like "static" are not the first things
2151 in a declaration. According to the C Standard, this usage is
2152 obsolescent.
2153 * If -Wall or -Wunused is also specified, warn about unused argu‐
2155 ments.
2156 * A comparison between signed and unsigned values could produce
2158 an incorrect result when the signed value is converted to
2159 unsigned. (But don't warn if -Wno-sign-compare is also speci‐
2160 fied.)
2161 * An aggregate has an initializer which does not initialize all
2163 members. This warning can be independently controlled by
2164 -Wmissing-field-initializers.
2165 * A function parameter is declared without a type specifier in
2167 K&R-style functions:
2168 void foo(bar) { }
2170 * An empty body occurs in an if or else statement.
2172 * A pointer is compared against integer zero with <, <=, >, or
2174 >=.
2175 * A variable might be changed by longjmp or vfork.
2177 * Any of several floating-point events that often indicate
2179 errors, such as overflow, underflow, loss of precision, etc.
2180 *<(C++ only)>
2182 An enumerator and a non-enumerator both appear in a conditional
2183 expression.
2184 *<(C++ only)>
2186 A non-static reference or non-static const member appears in a
2187 class without constructors.
2188 *<(C++ only)>
2190 Ambiguous virtual bases.
2191 *<(C++ only)>
2193 Subscripting an array which has been declared register.
2194 *<(C++ only)>
2196 Taking the address of a variable which has been declared regis‐
2197 ter.
2198 *<(C++ only)>
2200 A base class is not initialized in a derived class' copy con‐
2201 structor.
2202 -Wno-div-by-zero
2204 Do not warn about compile-time integer division by zero. Floating
2205 point division by zero is not warned about, as it can be a legiti‐
2206 mate way of obtaining infinities and NaNs.
2207 -Wsystem-headers
2209 Print warning messages for constructs found in system header files.
2210 Warnings from system headers are normally suppressed, on the
2211 assumption that they usually do not indicate real problems and
2212 would only make the compiler output harder to read. Using this
2213 command line option tells GCC to emit warnings from system headers
2214 as if they occurred in user code. However, note that using -Wall
2215 in conjunction with this option will not warn about unknown pragmas
2216 in system headers---for that, -Wunknown-pragmas must also be used.
2217 -Wfloat-equal
2219 Warn if floating point values are used in equality comparisons.
2220 The idea behind this is that sometimes it is convenient (for the
2222 programmer) to consider floating-point values as approximations to
2223 infinitely precise real numbers. If you are doing this, then you
2224 need to compute (by analyzing the code, or in some other way) the
2225 maximum or likely maximum error that the computation introduces,
2226 and allow for it when performing comparisons (and when producing
2227 output, but that's a different problem). In particular, instead of
2228 testing for equality, you would check to see whether the two values
2229 have ranges that overlap; and this is done with the relational
2230 operators, so equality comparisons are probably mistaken.
2231 -Wtraditional (C only)
2233 Warn about certain constructs that behave differently in tradi‐
2234 tional and ISO C. Also warn about ISO C constructs that have no
2235 traditional C equivalent, and/or problematic constructs which
2236 should be avoided.
2237 * Macro parameters that appear within string literals in the
2239 macro body. In traditional C macro replacement takes place
2240 within string literals, but does not in ISO C.
2241 * In traditional C, some preprocessor directives did not exist.
2243 Traditional preprocessors would only consider a line to be a
2244 directive if the # appeared in column 1 on the line. Therefore
2245 -Wtraditional warns about directives that traditional C under‐
2246 stands but would ignore because the # does not appear as the
2247 first character on the line. It also suggests you hide direc‐
2248 tives like #pragma not understood by traditional C by indenting
2249 them. Some traditional implementations would not recognize
2250 #elif, so it suggests avoiding it altogether.
2251 * A function-like macro that appears without arguments.
2253 * The unary plus operator.
2255 * The U integer constant suffix, or the F or L floating point
2257 constant suffixes. (Traditional C does support the L suffix on
2258 integer constants.) Note, these suffixes appear in macros
2259 defined in the system headers of most modern systems, e.g. the
2260 _MIN/_MAX macros in "<limits.h>". Use of these macros in user
2261 code might normally lead to spurious warnings, however GCC's
2262 integrated preprocessor has enough context to avoid warning in
2263 these cases.
2264 * A function declared external in one block and then used after
2266 the end of the block.
2267 * A "switch" statement has an operand of type "long".
2269 * A non-"static" function declaration follows a "static" one.
2271 This construct is not accepted by some traditional C compilers.
2272 * The ISO type of an integer constant has a different width or
2274 signedness from its traditional type. This warning is only
2275 issued if the base of the constant is ten. I.e. hexadecimal or
2276 octal values, which typically represent bit patterns, are not
2277 warned about.
2278 * Usage of ISO string concatenation is detected.
2280 * Initialization of automatic aggregates.
2282 * Identifier conflicts with labels. Traditional C lacks a sepa‐
2284 rate namespace for labels.
2285 * Initialization of unions. If the initializer is zero, the
2287 warning is omitted. This is done under the assumption that the
2288 zero initializer in user code appears conditioned on e.g.
2289 "__STDC__" to avoid missing initializer warnings and relies on
2290 default initialization to zero in the traditional C case.
2291 * Conversions by prototypes between fixed/floating point values
2293 and vice versa. The absence of these prototypes when compiling
2294 with traditional C would cause serious problems. This is a
2295 subset of the possible conversion warnings, for the full set
2296 use -Wconversion.
2297 * Use of ISO C style function definitions. This warning inten‐
2299 tionally is not issued for prototype declarations or variadic
2300 functions because these ISO C features will appear in your code
2301 when using libiberty's traditional C compatibility macros,
2302 "PARAMS" and "VPARAMS". This warning is also bypassed for
2303 nested functions because that feature is already a GCC exten‐
2304 sion and thus not relevant to traditional C compatibility.
2305 -Wdeclaration-after-statement (C only)
2307 Warn when a declaration is found after a statement in a block.
2308 This construct, known from C++, was introduced with ISO C99 and is
2309 by default allowed in GCC. It is not supported by ISO C90 and was
2310 not supported by GCC versions before GCC 3.0.
2311 -Wundef
2313 Warn if an undefined identifier is evaluated in an #if directive.
2314 -Wno-endif-labels
2316 Do not warn whenever an #else or an #endif are followed by text.
2317 -Wshadow
2319 Warn whenever a local variable shadows another local variable,
2320 parameter or global variable or whenever a built-in function is
2321 shadowed.
2322 -Wlarger-than-len
2324 Warn whenever an object of larger than len bytes is defined.
2325 -Wunsafe-loop-optimizations
2327 Warn if the loop cannot be optimized because the compiler could not
2328 assume anything on the bounds of the loop indices. With -fun‐
2329 safe-loop-optimizations warn if the compiler made such assumptions.
2330 -Wpointer-arith
2332 Warn about anything that depends on the "size of" a function type
2333 or of "void". GNU C assigns these types a size of 1, for conve‐
2334 nience in calculations with "void *" pointers and pointers to func‐
2335 tions.
2336 -Wbad-function-cast (C only)
2338 Warn whenever a function call is cast to a non-matching type. For
2339 example, warn if "int malloc()" is cast to "anything *".
2340 -Wc++-compat
2342 Warn about ISO C constructs that are outside of the common subset
2343 of ISO C and ISO C++, e.g. request for implicit conversion from
2344 "void *" to a pointer to non-"void" type.
2345 -Wcast-qual
2347 Warn whenever a pointer is cast so as to remove a type qualifier
2348 from the target type. For example, warn if a "const char *" is
2349 cast to an ordinary "char *".
2350 -Wcast-align
2352 Warn whenever a pointer is cast such that the required alignment of
2353 the target is increased. For example, warn if a "char *" is cast
2354 to an "int *" on machines where integers can only be accessed at
2355 two- or four-byte boundaries.
2356 -Wwrite-strings
2358 When compiling C, give string constants the type "const
2359 char[length]" so that copying the address of one into a non-"const"
2360 "char *" pointer will get a warning; when compiling C++, warn about
2361 the deprecated conversion from string constants to "char *". These
2362 warnings will help you find at compile time code that can try to
2363 write into a string constant, but only if you have been very care‐
2364 ful about using "const" in declarations and prototypes. Otherwise,
2365 it will just be a nuisance; this is why we did not make -Wall
2366 request these warnings.
2367 -Wconversion
2369 Warn if a prototype causes a type conversion that is different from
2370 what would happen to the same argument in the absence of a proto‐
2371 type. This includes conversions of fixed point to floating and
2372 vice versa, and conversions changing the width or signedness of a
2373 fixed point argument except when the same as the default promotion.
2374 Also, warn if a negative integer constant expression is implicitly
2376 converted to an unsigned type. For example, warn about the assign‐
2377 ment "x = -1" if "x" is unsigned. But do not warn about explicit
2378 casts like "(unsigned) -1".
2379 -Wsign-compare
2381 Warn when a comparison between signed and unsigned values could
2382 produce an incorrect result when the signed value is converted to
2383 unsigned. This warning is also enabled by -Wextra; to get the
2384 other warnings of -Wextra without this warning, use -Wextra
2385 -Wno-sign-compare.
2386 -Waggregate-return
2388 Warn if any functions that return structures or unions are defined
2389 or called. (In languages where you can return an array, this also
2390 elicits a warning.)
2391 -Wno-attributes
2393 Do not warn if an unexpected "__attribute__" is used, such as
2394 unrecognized attributes, function attributes applied to variables,
2395 etc. This will not stop errors for incorrect use of supported
2396 attributes.
2397 -Wstrict-prototypes (C only)
2399 Warn if a function is declared or defined without specifying the
2400 argument types. (An old-style function definition is permitted
2401 without a warning if preceded by a declaration which specifies the
2402 argument types.)
2403 -Wold-style-definition (C only)
2405 Warn if an old-style function definition is used. A warning is
2406 given even if there is a previous prototype.
2407 -Wmissing-prototypes (C only)
2409 Warn if a global function is defined without a previous prototype
2410 declaration. This warning is issued even if the definition itself
2411 provides a prototype. The aim is to detect global functions that
2412 fail to be declared in header files.
2413 -Wmissing-declarations (C only)
2415 Warn if a global function is defined without a previous declara‐
2416 tion. Do so even if the definition itself provides a prototype.
2417 Use this option to detect global functions that are not declared in
2418 header files.
2419 -Wmissing-field-initializers
2421 Warn if a structure's initializer has some fields missing. For
2422 example, the following code would cause such a warning, because
2423 "x.h" is implicitly zero:
2424 struct s { int f, g, h; };
2426 struct s x = { 3, 4 };
2427 This option does not warn about designated initializers, so the
2429 following modification would not trigger a warning:
2430 struct s { int f, g, h; };
2432 struct s x = { .f = 3, .g = 4 };
2433 This warning is included in -Wextra. To get other -Wextra warnings
2435 without this one, use -Wextra -Wno-missing-field-initializers.
2436 -Wmissing-noreturn
2438 Warn about functions which might be candidates for attribute "nore‐
2439 turn". Note these are only possible candidates, not absolute ones.
2440 Care should be taken to manually verify functions actually do not
2441 ever return before adding the "noreturn" attribute, otherwise sub‐
2442 tle code generation bugs could be introduced. You will not get a
2443 warning for "main" in hosted C environments.
2444 -Wmissing-format-attribute
2446 Warn about function pointers which might be candidates for "format"
2447 attributes. Note these are only possible candidates, not absolute
2448 ones. GCC will guess that function pointers with "format"
2449 attributes that are used in assignment, initialization, parameter
2450 passing or return statements should have a corresponding "format"
2451 attribute in the resulting type. I.e. the left-hand side of the
2452 assignment or initialization, the type of the parameter variable,
2453 or the return type of the containing function respectively should
2454 also have a "format" attribute to avoid the warning.
2455 GCC will also warn about function definitions which might be candi‐
2457 dates for "format" attributes. Again, these are only possible can‐
2458 didates. GCC will guess that "format" attributes might be appro‐
2459 priate for any function that calls a function like "vprintf" or
2460 "vscanf", but this might not always be the case, and some functions
2461 for which "format" attributes are appropriate may not be detected.
2462 -Wno-multichar
2464 Do not warn if a multicharacter constant ('FOOF') is used. Usually
2465 they indicate a typo in the user's code, as they have implementa‐
2466 tion-defined values, and should not be used in portable code.
2467 -Wnormalized=<none⎪id⎪nfc⎪nfkc>
2469 In ISO C and ISO C++, two identifiers are different if they are
2470 different sequences of characters. However, sometimes when charac‐
2471 ters outside the basic ASCII character set are used, you can have
2472 two different character sequences that look the same. To avoid
2473 confusion, the ISO 10646 standard sets out some normalization rules
2474 which when applied ensure that two sequences that look the same are
2475 turned into the same sequence. GCC can warn you if you are using
2476 identifiers which have not been normalized; this option controls
2477 that warning.
2478 There are four levels of warning that GCC supports. The default is
2480 -Wnormalized=nfc, which warns about any identifier which is not in
2481 the ISO 10646 "C" normalized form, NFC. NFC is the recommended
2482 form for most uses.
2483 Unfortunately, there are some characters which ISO C and ISO C++
2485 allow in identifiers that when turned into NFC aren't allowable as
2486 identifiers. That is, there's no way to use these symbols in por‐
2487 table ISO C or C++ and have all your identifiers in NFC. -Wnormal‐
2488 ized=id suppresses the warning for these characters. It is hoped
2489 that future versions of the standards involved will correct this,
2490 which is why this option is not the default.
2491 You can switch the warning off for all characters by writing -Wnor‐
2493 malized=none. You would only want to do this if you were using
2494 some other normalization scheme (like "D"), because otherwise you
2495 can easily create bugs that are literally impossible to see.
2496 Some characters in ISO 10646 have distinct meanings but look iden‐
2498 tical in some fonts or display methodologies, especially once for‐
2499 matting has been applied. For instance "\u207F", "SUPERSCRIPT
2500 LATIN SMALL LETTER N", will display just like a regular "n" which
2501 has been placed in a superscript. ISO 10646 defines the NFKC nor‐
2502 malisation scheme to convert all these into a standard form as
2503 well, and GCC will warn if your code is not in NFKC if you use
2504 -Wnormalized=nfkc. This warning is comparable to warning about
2505 every identifier that contains the letter O because it might be
2506 confused with the digit 0, and so is not the default, but may be
2507 useful as a local coding convention if the programming environment
2508 is unable to be fixed to display these characters distinctly.
2509 -Wno-deprecated-declarations
2511 Do not warn about uses of functions, variables, and types marked as
2512 deprecated by using the "deprecated" attribute. (@pxref{Function
2513 Attributes}, @pxref{Variable Attributes}, @pxref{Type Attributes}.)
2514 -Wpacked
2516 Warn if a structure is given the packed attribute, but the packed
2517 attribute has no effect on the layout or size of the structure.
2518 Such structures may be mis-aligned for little benefit. For
2519 instance, in this code, the variable "f.x" in "struct bar" will be
2520 misaligned even though "struct bar" does not itself have the packed
2521 attribute:
2522 struct foo {
2524 int x;
2525 char a, b, c, d;
2526 } __attribute__((packed));
2527 struct bar {
2528 char z;
2529 struct foo f;
2530 };
2531 -Wpadded
2533 Warn if padding is included in a structure, either to align an ele‐
2534 ment of the structure or to align the whole structure. Sometimes
2535 when this happens it is possible to rearrange the fields of the
2536 structure to reduce the padding and so make the structure smaller.
2537 -Wredundant-decls
2539 Warn if anything is declared more than once in the same scope, even
2540 in cases where multiple declaration is valid and changes nothing.
2541 -Wnested-externs (C only)
2543 Warn if an "extern" declaration is encountered within a function.
2544 -Wunreachable-code
2546 Warn if the compiler detects that code will never be executed.
2547 This option is intended to warn when the compiler detects that at
2549 least a whole line of source code will never be executed, because
2550 some condition is never satisfied or because it is after a proce‐
2551 dure that never returns.
2552 It is possible for this option to produce a warning even though
2554 there are circumstances under which part of the affected line can
2555 be executed, so care should be taken when removing apparently-
2556 unreachable code.
2557 For instance, when a function is inlined, a warning may mean that
2559 the line is unreachable in only one inlined copy of the function.
2560 This option is not made part of -Wall because in a debugging ver‐
2562 sion of a program there is often substantial code which checks cor‐
2563 rect functioning of the program and is, hopefully, unreachable
2564 because the program does work. Another common use of unreachable
2565 code is to provide behavior which is selectable at compile-time.
2566 -Winline
2568 Warn if a function can not be inlined and it was declared as
2569 inline. Even with this option, the compiler will not warn about
2570 failures to inline functions declared in system headers.
2571 The compiler uses a variety of heuristics to determine whether or
2573 not to inline a function. For example, the compiler takes into
2574 account the size of the function being inlined and the amount of
2575 inlining that has already been done in the current function.
2576 Therefore, seemingly insignificant changes in the source program
2577 can cause the warnings produced by -Winline to appear or disappear.
2578 -Wno-invalid-offsetof (C++ only)
2580 Suppress warnings from applying the offsetof macro to a non-POD
2581 type. According to the 1998 ISO C++ standard, applying offsetof to
2582 a non-POD type is undefined. In existing C++ implementations, how‐
2583 ever, offsetof typically gives meaningful results even when applied
2584 to certain kinds of non-POD types. (Such as a simple struct that
2585 fails to be a POD type only by virtue of having a constructor.)
2586 This flag is for users who are aware that they are writing non‐
2587 portable code and who have deliberately chosen to ignore the warn‐
2588 ing about it.
2589 The restrictions on offsetof may be relaxed in a future version of
2591 the C++ standard.
2592 -Wno-int-to-pointer-cast (C only)
2594 Suppress warnings from casts to pointer type of an integer of a
2595 different size.
2596 -Wno-pointer-to-int-cast (C only)
2598 Suppress warnings from casts from a pointer to an integer type of a
2599 different size.
2600 -Winvalid-pch
2602 Warn if a precompiled header is found in the search path but can't
2603 be used.
2604 -Wlong-long
2606 Warn if long long type is used. This is default. To inhibit the
2607 warning messages, use -Wno-long-long. Flags -Wlong-long and
2608 -Wno-long-long are taken into account only when -pedantic flag is
2609 used.
2610 -Wvariadic-macros
2612 Warn if variadic macros are used in pedantic ISO C90 mode, or the
2613 GNU alternate syntax when in pedantic ISO C99 mode. This is
2614 default. To inhibit the warning messages, use -Wno-vari‐
2615 adic-macros.
2616 -Wvolatile-register-var
2618 Warn if a register variable is declared volatile. The volatile
2619 modifier does not inhibit all optimizations that may eliminate
2620 reads and/or writes to register variables.
2621 -Wdisabled-optimization
2623 Warn if a requested optimization pass is disabled. This warning
2624 does not generally indicate that there is anything wrong with your
2625 code; it merely indicates that GCC's optimizers were unable to han‐
2626 dle the code effectively. Often, the problem is that your code is
2627 too big or too complex; GCC will refuse to optimize programs when
2628 the optimization itself is likely to take inordinate amounts of
2629 time.
2630 -Wpointer-sign
2632 Warn for pointer argument passing or assignment with different
2633 signedness. This option is only supported for C and Objective-C.
2634 It is implied by -Wall and by -pedantic, which can be disabled with
2635 -Wno-pointer-sign.
2636 -Werror
2638 Make all warnings into errors.
2639 -Wstack-protector
2641 This option is only active when -fstack-protector is active. It
2642 warns about functions that will not be protected against stack
2643 smashing.
2644 Options for Debugging Your Program or GCC
2646 GCC has various special options that are used for debugging either your
2648 program or GCC:
2649 -g Produce debugging information in the operating system's native for‐
2651 mat (stabs, COFF, XCOFF, or DWARF 2). GDB can work with this
2652 debugging information.
2653 On most systems that use stabs format, -g enables use of extra
2655 debugging information that only GDB can use; this extra information
2656 makes debugging work better in GDB but will probably make other
2657 debuggers crash or refuse to read the program. If you want to con‐
2658 trol for certain whether to generate the extra information, use
2659 -gstabs+, -gstabs, -gxcoff+, -gxcoff, or -gvms (see below).
2660 GCC allows you to use -g with -O. The shortcuts taken by optimized
2662 code may occasionally produce surprising results: some variables
2663 you declared may not exist at all; flow of control may briefly move
2664 where you did not expect it; some statements may not be executed
2665 because they compute constant results or their values were already
2666 at hand; some statements may execute in different places because
2667 they were moved out of loops.
2668 Nevertheless it proves possible to debug optimized output. This
2670 makes it reasonable to use the optimizer for programs that might
2671 have bugs.
2672 The following options are useful when GCC is generated with the
2674 capability for more than one debugging format.
2675 -ggdb
2677 Produce debugging information for use by GDB. This means to use
2678 the most expressive format available (DWARF 2, stabs, or the native
2679 format if neither of those are supported), including GDB extensions
2680 if at all possible.
2681 -gstabs
2683 Produce debugging information in stabs format (if that is sup‐
2684 ported), without GDB extensions. This is the format used by DBX on
2685 most BSD systems. On MIPS, Alpha and System V Release 4 systems
2686 this option produces stabs debugging output which is not understood
2687 by DBX or SDB. On System V Release 4 systems this option requires
2688 the GNU assembler.
2689 -feliminate-unused-debug-symbols
2691 Produce debugging information in stabs format (if that is sup‐
2692 ported), for only symbols that are actually used.
2693 -gstabs+
2695 Produce debugging information in stabs format (if that is sup‐
2696 ported), using GNU extensions understood only by the GNU debugger
2697 (GDB). The use of these extensions is likely to make other debug‐
2698 gers crash or refuse to read the program.
2699 -gcoff
2701 Produce debugging information in COFF format (if that is sup‐
2702 ported). This is the format used by SDB on most System V systems
2703 prior to System V Release 4.
2704 -gxcoff
2706 Produce debugging information in XCOFF format (if that is sup‐
2707 ported). This is the format used by the DBX debugger on IBM
2708 RS/6000 systems.
2709 -gxcoff+
2711 Produce debugging information in XCOFF format (if that is sup‐
2712 ported), using GNU extensions understood only by the GNU debugger
2713 (GDB). The use of these extensions is likely to make other debug‐
2714 gers crash or refuse to read the program, and may cause assemblers
2715 other than the GNU assembler (GAS) to fail with an error.
2716 -gdwarf-2
2718 Produce debugging information in DWARF version 2 format (if that is
2719 supported). This is the format used by DBX on IRIX 6. With this
2720 option, GCC uses features of DWARF version 3 when they are useful;
2721 version 3 is upward compatible with version 2, but may still cause
2722 problems for older debuggers.
2723 -gvms
2725 Produce debugging information in VMS debug format (if that is sup‐
2726 ported). This is the format used by DEBUG on VMS systems.
2727 -glevel
2729 -ggdblevel
2730 -gstabslevel
2731 -gcofflevel
2732 -gxcofflevel
2733 -gvmslevel
2734 Request debugging information and also use level to specify how
2735 much information. The default level is 2.
2736 Level 1 produces minimal information, enough for making backtraces
2738 in parts of the program that you don't plan to debug. This
2739 includes descriptions of functions and external variables, but no
2740 information about local variables and no line numbers.
2741 Level 3 includes extra information, such as all the macro defini‐
2743 tions present in the program. Some debuggers support macro expan‐
2744 sion when you use -g3.
2745 -gdwarf-2 does not accept a concatenated debug level, because GCC
2747 used to support an option -gdwarf that meant to generate debug
2748 information in version 1 of the DWARF format (which is very differ‐
2749 ent from version 2), and it would have been too confusing. That
2750 debug format is long obsolete, but the option cannot be changed
2751 now. Instead use an additional -glevel option to change the debug
2752 level for DWARF2.
2753 -feliminate-dwarf2-dups
2755 Compress DWARF2 debugging information by eliminating duplicated
2756 information about each symbol. This option only makes sense when
2757 generating DWARF2 debugging information with -gdwarf-2.
2758 -p Generate extra code to write profile information suitable for the
2760 analysis program prof. You must use this option when compiling the
2761 source files you want data about, and you must also use it when
2762 linking.
2763 -pg Generate extra code to write profile information suitable for the
2765 analysis program gprof. You must use this option when compiling
2766 the source files you want data about, and you must also use it when
2767 linking.
2768 -Q Makes the compiler print out each function name as it is compiled,
2770 and print some statistics about each pass when it finishes.
2771 -ftime-report
2773 Makes the compiler print some statistics about the time consumed by
2774 each pass when it finishes.
2775 -fmem-report
2777 Makes the compiler print some statistics about permanent memory
2778 allocation when it finishes.
2779 -fprofile-arcs
2781 Add code so that program flow arcs are instrumented. During execu‐
2782 tion the program records how many times each branch and call is
2783 executed and how many times it is taken or returns. When the com‐
2784 piled program exits it saves this data to a file called aux‐
2785 name.gcda for each source file. The data may be used for profile-
2786 directed optimizations (-fbranch-probabilities), or for test cover‐
2787 age analysis (-ftest-coverage). Each object file's auxname is gen‐
2788 erated from the name of the output file, if explicitly specified
2789 and it is not the final executable, otherwise it is the basename of
2790 the source file. In both cases any suffix is removed (e.g.
2791 foo.gcda for input file dir/foo.c, or dir/foo.gcda for output file
2792 specified as -o dir/foo.o).
2793 --coverage
2795 This option is used to compile and link code instrumented for cov‐
2796 erage analysis. The option is a synonym for -fprofile-arcs
2797 -ftest-coverage (when compiling) and -lgcov (when linking). See
2798 the documentation for those options for more details.
2799 @bullet
2801 Compile the source files with -fprofile-arcs plus optimization
2802 and code generation options. For test coverage analysis, use
2803 the additional -ftest-coverage option. You do not need to pro‐
2804 file every source file in a program.
2805 @cvmmfu
2807 Link your object files with -lgcov or -fprofile-arcs (the lat‐
2808 ter implies the former).
2809 @dwnngv
2811 Run the program on a representative workload to generate the
2812 arc profile information. This may be repeated any number of
2813 times. You can run concurrent instances of your program, and
2814 provided that the file system supports locking, the data files
2815 will be correctly updated. Also "fork" calls are detected and
2816 correctly handled (double counting will not happen).
2817 @exoohw
2819 For profile-directed optimizations, compile the source files
2820 again with the same optimization and code generation options
2821 plus -fbranch-probabilities.
2822 @fyppix
2824 For test coverage analysis, use gcov to produce human readable
2825 information from the .gcno and .gcda files. Refer to the gcov
2826 documentation for further information.
2827 With -fprofile-arcs, for each function of your program GCC creates
2829 a program flow graph, then finds a spanning tree for the graph.
2830 Only arcs that are not on the spanning tree have to be instru‐
2831 mented: the compiler adds code to count the number of times that
2832 these arcs are executed. When an arc is the only exit or only
2833 entrance to a block, the instrumentation code can be added to the
2834 block; otherwise, a new basic block must be created to hold the
2835 instrumentation code.
2836 -ftest-coverage
2838 Produce a notes file that the gcov code-coverage utility can use to
2839 show program coverage. Each source file's note file is called aux‐
2840 name.gcno. Refer to the -fprofile-arcs option above for a descrip‐
2841 tion of auxname and instructions on how to generate test coverage
2842 data. Coverage data will match the source files more closely, if
2843 you do not optimize.
2844 -dletters
2846 -fdump-rtl-pass
2847 Says to make debugging dumps during compilation at times specified
2848 by letters. This is used for debugging the RTL-based passes of
2849 the compiler. The file names for most of the dumps are made by
2850 appending a pass number and a word to the dumpname. dumpname is
2851 generated from the name of the output file, if explicitly specified
2852 and it is not an executable, otherwise it is the basename of the
2853 source file.
2854 Most debug dumps can be enabled either passing a letter to the -d
2856 option, or with a long -fdump-rtl switch; here are the possible
2857 letters for use in letters and pass, and their meanings:
2858 -dA Annotate the assembler output with miscellaneous debugging
2860 information.
2861 -db
2863 -fdump-rtl-bp
2864 Dump after computing branch probabilities, to file.09.bp.
2865 -dB
2867 -fdump-rtl-bbro
2868 Dump after block reordering, to file.30.bbro.
2869 -dc
2871 -fdump-rtl-combine
2872 Dump after instruction combination, to the file file.17.com‐
2873 bine.
2874 -dC
2876 -fdump-rtl-ce1
2877 -fdump-rtl-ce2
2878 -dC and -fdump-rtl-ce1 enable dumping after the first if con‐
2879 version, to the file file.11.ce1. -dC and -fdump-rtl-ce2
2880 enable dumping after the second if conversion, to the file
2881 file.18.ce2.
2882 -dd
2884 -fdump-rtl-btl
2885 -fdump-rtl-dbr
2886 -dd and -fdump-rtl-btl enable dumping after branch target load
2887 optimization, to file.31.btl. -dd and -fdump-rtl-dbr enable
2888 dumping after delayed branch scheduling, to file.36.dbr.
2889 -dD Dump all macro definitions, at the end of preprocessing, in
2891 addition to normal output.
2892 -dE
2894 -fdump-rtl-ce3
2895 Dump after the third if conversion, to file.28.ce3.
2896 -df
2898 -fdump-rtl-cfg
2899 -fdump-rtl-life
2900 -df and -fdump-rtl-cfg enable dumping after control and data
2901 flow analysis, to file.08.cfg. -df and -fdump-rtl-cfg enable
2902 dumping dump after life analysis, to file.16.life.
2903 -dg
2905 -fdump-rtl-greg
2906 Dump after global register allocation, to file.23.greg.
2907 -dG
2909 -fdump-rtl-gcse
2910 -fdump-rtl-bypass
2911 -dG and -fdump-rtl-gcse enable dumping after GCSE, to
2912 file.05.gcse. -dG and -fdump-rtl-bypass enable dumping after
2913 jump bypassing and control flow optimizations, to
2914 file.07.bypass.
2915 -dh
2917 -fdump-rtl-eh
2918 Dump after finalization of EH handling code, to file.02.eh.
2919 -di
2921 -fdump-rtl-sibling
2922 Dump after sibling call optimizations, to file.01.sibling.
2923 -dj
2925 -fdump-rtl-jump
2926 Dump after the first jump optimization, to file.03.jump.
2927 -dk
2929 -fdump-rtl-stack
2930 Dump after conversion from registers to stack, to
2931 file.33.stack.
2932 -dl
2934 -fdump-rtl-lreg
2935 Dump after local register allocation, to file.22.lreg.
2936 -dL
2938 -fdump-rtl-loop
2939 -fdump-rtl-loop2
2940 -dL and -fdump-rtl-loop enable dumping after the first loop
2941 optimization pass, to file.06.loop. -dL and -fdump-rtl-loop2
2942 enable dumping after the second pass, to file.13.loop2.
2943 -dm
2945 -fdump-rtl-sms
2946 Dump after modulo scheduling, to file.20.sms.
2947 -dM
2949 -fdump-rtl-mach
2950 Dump after performing the machine dependent reorganization
2951 pass, to file.35.mach.
2952 -dn
2954 -fdump-rtl-rnreg
2955 Dump after register renumbering, to file.29.rnreg.
2956 -dN
2958 -fdump-rtl-regmove
2959 Dump after the register move pass, to file.19.regmove.
2960 -do
2962 -fdump-rtl-postreload
2963 Dump after post-reload optimizations, to file.24.postreload.
2964 -dr
2966 -fdump-rtl-expand
2967 Dump after RTL generation, to file.00.expand.
2968 -dR
2970 -fdump-rtl-sched2
2971 Dump after the second scheduling pass, to file.32.sched2.
2972 -ds
2974 -fdump-rtl-cse
2975 Dump after CSE (including the jump optimization that sometimes
2976 follows CSE), to file.04.cse.
2977 -dS
2979 -fdump-rtl-sched
2980 Dump after the first scheduling pass, to file.21.sched.
2981 -dt
2983 -fdump-rtl-cse2
2984 Dump after the second CSE pass (including the jump optimization
2985 that sometimes follows CSE), to file.15.cse2.
2986 -dT
2988 -fdump-rtl-tracer
2989 Dump after running tracer, to file.12.tracer.
2990 -dV
2992 -fdump-rtl-vpt
2993 -fdump-rtl-vartrack
2994 -dV and -fdump-rtl-vpt enable dumping after the value profile
2995 transformations, to file.10.vpt. -dV and -fdump-rtl-vartrack
2996 enable dumping after variable tracking, to file.34.vartrack.
2997 -dw
2999 -fdump-rtl-flow2
3000 Dump after the second flow pass, to file.26.flow2.
3001 -dz
3003 -fdump-rtl-peephole2
3004 Dump after the peephole pass, to file.27.peephole2.
3005 -dZ
3007 -fdump-rtl-web
3008 Dump after live range splitting, to file.14.web.
3009 -da
3011 -fdump-rtl-all
3012 Produce all the dumps listed above.
3013 -dH Produce a core dump whenever an error occurs.
3015 -dm Print statistics on memory usage, at the end of the run, to
3017 standard error.
3018 -dp Annotate the assembler output with a comment indicating which
3020 pattern and alternative was used. The length of each instruc‐
3021 tion is also printed.
3022 -dP Dump the RTL in the assembler output as a comment before each
3024 instruction. Also turns on -dp annotation.
3025 -dv For each of the other indicated dump files (either with -d or
3027 -fdump-rtl-pass), dump a representation of the control flow
3028 graph suitable for viewing with VCG to file.pass.vcg.
3029 -dx Just generate RTL for a function instead of compiling it. Usu‐
3031 ally used with r (-fdump-rtl-expand).
3032 -dy Dump debugging information during parsing, to standard error.
3034 -fdump-unnumbered
3036 When doing debugging dumps (see -d option above), suppress instruc‐
3037 tion numbers and line number note output. This makes it more fea‐
3038 sible to use diff on debugging dumps for compiler invocations with
3039 different options, in particular with and without -g.
3040 -fdump-translation-unit (C++ only)
3042 -fdump-translation-unit-options (C++ only)
3043 Dump a representation of the tree structure for the entire transla‐
3044 tion unit to a file. The file name is made by appending .tu to the
3045 source file name. If the -options form is used, options controls
3046 the details of the dump as described for the -fdump-tree options.
3047 -fdump-class-hierarchy (C++ only)
3049 -fdump-class-hierarchy-options (C++ only)
3050 Dump a representation of each class's hierarchy and virtual func‐
3051 tion table layout to a file. The file name is made by appending
3052 .class to the source file name. If the -options form is used,
3053 options controls the details of the dump as described for the
3054 -fdump-tree options.
3055 -fdump-ipa-switch
3057 Control the dumping at various stages of inter-procedural analysis
3058 language tree to a file. The file name is generated by appending a
3059 switch specific suffix to the source file name. The following
3060 dumps are possible:
3061 all Enables all inter-procedural analysis dumps; currently the only
3063 produced dump is the cgraph dump.
3064 cgraph
3066 Dumps information about call-graph optimization, unused func‐
3067 tion removal, and inlining decisions.
3068 -fdump-tree-switch
3070 -fdump-tree-switch-options
3071 Control the dumping at various stages of processing the intermedi‐
3072 ate language tree to a file. The file name is generated by append‐
3073 ing a switch specific suffix to the source file name. If the
3074 -options form is used, options is a list of - separated options
3075 that control the details of the dump. Not all options are applica‐
3076 ble to all dumps, those which are not meaningful will be ignored.
3077 The following options are available
3078 address
3080 Print the address of each node. Usually this is not meaningful
3081 as it changes according to the environment and source file.
3082 Its primary use is for tying up a dump file with a debug envi‐
3083 ronment.
3084 slim
3086 Inhibit dumping of members of a scope or body of a function
3087 merely because that scope has been reached. Only dump such
3088 items when they are directly reachable by some other path.
3089 When dumping pretty-printed trees, this option inhibits dumping
3090 the bodies of control structures.
3091 raw Print a raw representation of the tree. By default, trees are
3093 pretty-printed into a C-like representation.
3094 details
3096 Enable more detailed dumps (not honored by every dump option).
3097 stats
3099 Enable dumping various statistics about the pass (not honored
3100 by every dump option).
3101 blocks
3103 Enable showing basic block boundaries (disabled in raw dumps).
3104 vops
3106 Enable showing virtual operands for every statement.
3107 lineno
3109 Enable showing line numbers for statements.
3110 uid Enable showing the unique ID ("DECL_UID") for each variable.
3112 all Turn on all options, except raw, slim and lineno.
3114 The following tree dumps are possible:
3116 original
3118 Dump before any tree based optimization, to file.original.
3119 optimized
3121 Dump after all tree based optimization, to file.optimized.
3122 inlined
3124 Dump after function inlining, to file.inlined.
3125 gimple
3127 Dump each function before and after the gimplification pass to
3128 a file. The file name is made by appending .gimple to the
3129 source file name.
3130 cfg Dump the control flow graph of each function to a file. The
3132 file name is made by appending .cfg to the source file name.
3133 vcg Dump the control flow graph of each function to a file in VCG
3135 format. The file name is made by appending .vcg to the source
3136 file name. Note that if the file contains more than one func‐
3137 tion, the generated file cannot be used directly by VCG. You
3138 will need to cut and paste each function's graph into its own
3139 separate file first.
3140 ch Dump each function after copying loop headers. The file name
3142 is made by appending .ch to the source file name.
3143 ssa Dump SSA related information to a file. The file name is made
3145 by appending .ssa to the source file name.
3146 salias
3148 Dump structure aliasing variable information to a file. This
3149 file name is made by appending .salias to the source file name.
3150 alias
3152 Dump aliasing information for each function. The file name is
3153 made by appending .alias to the source file name.
3154 ccp Dump each function after CCP. The file name is made by append‐
3156 ing .ccp to the source file name.
3157 storeccp
3159 Dump each function after STORE-CCP. The file name is made by
3160 appending .storeccp to the source file name.
3161 pre Dump trees after partial redundancy elimination. The file name
3163 is made by appending .pre to the source file name.
3164 fre Dump trees after full redundancy elimination. The file name is
3166 made by appending .fre to the source file name.
3167 copyprop
3169 Dump trees after copy propagation. The file name is made by
3170 appending .copyprop to the source file name.
3171 store_copyprop
3173 Dump trees after store copy-propagation. The file name is made
3174 by appending .store_copyprop to the source file name.
3175 dce Dump each function after dead code elimination. The file name
3177 is made by appending .dce to the source file name.
3178 mudflap
3180 Dump each function after adding mudflap instrumentation. The
3181 file name is made by appending .mudflap to the source file
3182 name.
3183 sra Dump each function after performing scalar replacement of
3185 aggregates. The file name is made by appending .sra to the
3186 source file name.
3187 sink
3189 Dump each function after performing code sinking. The file
3190 name is made by appending .sink to the source file name.
3191 dom Dump each function after applying dominator tree optimizations.
3193 The file name is made by appending .dom to the source file
3194 name.
3195 dse Dump each function after applying dead store elimination. The
3197 file name is made by appending .dse to the source file name.
3198 phiopt
3200 Dump each function after optimizing PHI nodes into straightline
3201 code. The file name is made by appending .phiopt to the source
3202 file name.
3203 forwprop
3205 Dump each function after forward propagating single use vari‐
3206 ables. The file name is made by appending .forwprop to the
3207 source file name.
3208 copyrename
3210 Dump each function after applying the copy rename optimization.
3211 The file name is made by appending .copyrename to the source
3212 file name.
3213 nrv Dump each function after applying the named return value opti‐
3215 mization on generic trees. The file name is made by appending
3216 .nrv to the source file name.
3217 vect
3219 Dump each function after applying vectorization of loops. The
3220 file name is made by appending .vect to the source file name.
3221 vrp Dump each function after Value Range Propagation (VRP). The
3223 file name is made by appending .vrp to the source file name.
3224 all Enable all the available tree dumps with the flags provided in
3226 this option.
3227 -ftree-vectorizer-verbose=n
3229 This option controls the amount of debugging output the vectorizer
3230 prints. This information is written to standard error, unless
3231 -fdump-tree-all or -fdump-tree-vect is specified, in which case it
3232 is output to the usual dump listing file, .vect.
3233 -frandom-seed=string
3235 This option provides a seed that GCC uses when it would otherwise
3236 use random numbers. It is used to generate certain symbol names
3237 that have to be different in every compiled file. It is also used
3238 to place unique stamps in coverage data files and the object files
3239 that produce them. You can use the -frandom-seed option to produce
3240 reproducibly identical object files.
3241 The string should be different for every file you compile.
3243 -fsched-verbose=n
3245 On targets that use instruction scheduling, this option controls
3246 the amount of debugging output the scheduler prints. This informa‐
3247 tion is written to standard error, unless -dS or -dR is specified,
3248 in which case it is output to the usual dump listing file, .sched
3249 or .sched2 respectively. However for n greater than nine, the out‐
3250 put is always printed to standard error.
3251 For n greater than zero, -fsched-verbose outputs the same informa‐
3253 tion as -dRS. For n greater than one, it also output basic block
3254 probabilities, detailed ready list information and unit/insn info.
3255 For n greater than two, it includes RTL at abort point, control-
3256 flow and regions info. And for n over four, -fsched-verbose also
3257 includes dependence info.
3258 -save-temps
3260 Store the usual "temporary" intermediate files permanently; place
3261 them in the current directory and name them based on the source
3262 file. Thus, compiling foo.c with -c -save-temps would produce
3263 files foo.i and foo.s, as well as foo.o. This creates a prepro‐
3264 cessed foo.i output file even though the compiler now normally uses
3265 an integrated preprocessor.
3266 When used in combination with the -x command line option,
3268 -save-temps is sensible enough to avoid over writing an input
3269 source file with the same extension as an intermediate file. The
3270 corresponding intermediate file may be obtained by renaming the
3271 source file before using -save-temps.
3272 -time
3274 Report the CPU time taken by each subprocess in the compilation
3275 sequence. For C source files, this is the compiler proper and
3276 assembler (plus the linker if linking is done). The output looks
3277 like this:
3278 # cc1 0.12 0.01
3280 # as 0.00 0.01
3281 The first number on each line is the "user time", that is time
3283 spent executing the program itself. The second number is "system
3284 time", time spent executing operating system routines on behalf of
3285 the program. Both numbers are in seconds.
3286 -fvar-tracking
3288 Run variable tracking pass. It computes where variables are stored
3289 at each position in code. Better debugging information is then
3290 generated (if the debugging information format supports this infor‐
3291 mation).
3292 It is enabled by default when compiling with optimization (-Os, -O,
3294 -O2, ...), debugging information (-g) and the debug info format
3295 supports it.
3296 -print-file-name=library
3298 Print the full absolute name of the library file library that would
3299 be used when linking---and don't do anything else. With this
3300 option, GCC does not compile or link anything; it just prints the
3301 file name.
3302 -print-multi-directory
3304 Print the directory name corresponding to the multilib selected by
3305 any other switches present in the command line. This directory is
3306 supposed to exist in GCC_EXEC_PREFIX.
3307 -print-multi-lib
3309 Print the mapping from multilib directory names to compiler
3310 switches that enable them. The directory name is separated from
3311 the switches by ;, and each switch starts with an @} instead of the
3312 @samp{-, without spaces between multiple switches. This is sup‐
3313 posed to ease shell-processing.
3314 -print-prog-name=program
3316 Like -print-file-name, but searches for a program such as cpp.
3317 -print-libgcc-file-name
3319 Same as -print-file-name=libgcc.a.
3320 This is useful when you use -nostdlib or -nodefaultlibs but you do
3322 want to link with libgcc.a. You can do
3323 gcc -nostdlib <files>... `gcc -print-libgcc-file-name`
3325 -print-search-dirs
3327 Print the name of the configured installation directory and a list
3328 of program and library directories gcc will search---and don't do
3329 anything else.
3330 This is useful when gcc prints the error message installation prob‐
3332 lem, cannot exec cpp0: No such file or directory. To resolve this
3333 you either need to put cpp0 and the other compiler components where
3334 gcc expects to find them, or you can set the environment variable
3335 GCC_EXEC_PREFIX to the directory where you installed them. Don't
3336 forget the trailing /.
3337 -dumpmachine
3339 Print the compiler's target machine (for example,
3340 i686-pc-linux-gnu)---and don't do anything else.
3341 -dumpversion
3343 Print the compiler version (for example, 3.0)---and don't do any‐
3344 thing else.
3345 -dumpspecs
3347 Print the compiler's built-in specs---and don't do anything else.
3348 (This is used when GCC itself is being built.)
3349 -feliminate-unused-debug-types
3351 Normally, when producing DWARF2 output, GCC will emit debugging
3352 information for all types declared in a compilation unit, regard‐
3353 less of whether or not they are actually used in that compilation
3354 unit. Sometimes this is useful, such as if, in the debugger, you
3355 want to cast a value to a type that is not actually used in your
3356 program (but is declared). More often, however, this results in a
3357 significant amount of wasted space. With this option, GCC will
3358 avoid producing debug symbol output for types that are nowhere used
3359 in the source file being compiled.
3360 Options That Control Optimization
3362 These options control various sorts of optimizations.
3364 Without any optimization option, the compiler's goal is to reduce the
3366 cost of compilation and to make debugging produce the expected results.
3367 Statements are independent: if you stop the program with a breakpoint
3368 between statements, you can then assign a new value to any variable or
3369 change the program counter to any other statement in the function and
3370 get exactly the results you would expect from the source code.
3371 Turning on optimization flags makes the compiler attempt to improve the
3373 performance and/or code size at the expense of compilation time and
3374 possibly the ability to debug the program.
3375 The compiler performs optimization based on the knowledge it has of the
3377 program. Optimization levels -O and above, in particular, enable unit-
3378 at-a-time mode, which allows the compiler to consider information
3379 gained from later functions in the file when compiling a function.
3380 Compiling multiple files at once to a single output file in unit-at-a-
3381 time mode allows the compiler to use information gained from all of the
3382 files when compiling each of them.
3383 Not all optimizations are controlled directly by a flag. Only opti‐
3385 mizations that have a flag are listed.
3386 -O
3388 -O1 Optimize. Optimizing compilation takes somewhat more time, and a
3389 lot more memory for a large function.
3390 With -O, the compiler tries to reduce code size and execution time,
3392 without performing any optimizations that take a great deal of com‐
3393 pilation time.
3394 -O turns on the following optimization flags: -fdefer-pop -fde‐
3396 layed-branch -fguess-branch-probability -fcprop-registers
3397 -floop-optimize -fif-conversion -fif-conversion2 -ftree-ccp
3398 -ftree-dce -ftree-dominator-opts -ftree-dse -ftree-ter -ftree-lrs
3399 -ftree-sra -ftree-copyrename -ftree-fre -ftree-ch -funit-at-a-time
3400 -fmerge-constants
3401 -O also turns on -fomit-frame-pointer on machines where doing so
3403 does not interfere with debugging.
3404 -O doesn't turn on -ftree-sra for the Ada compiler. This option
3406 must be explicitly specified on the command line to be enabled for
3407 the Ada compiler.
3408 -O2 Optimize even more. GCC performs nearly all supported optimiza‐
3410 tions that do not involve a space-speed tradeoff. The compiler
3411 does not perform loop unrolling or function inlining when you spec‐
3412 ify -O2. As compared to -O, this option increases both compilation
3413 time and the performance of the generated code.
3414 -O2 turns on all optimization flags specified by -O. It also turns
3416 on the following optimization flags: -fthread-jumps -fcrossjumping
3417 -foptimize-sibling-calls -fcse-follow-jumps -fcse-skip-blocks
3418 -fgcse -fgcse-lm -fexpensive-optimizations -fstrength-reduce -fre‐
3419 run-cse-after-loop -frerun-loop-opt -fcaller-saves -fpeephole2
3420 -fschedule-insns -fschedule-insns2 -fsched-interblock
3421 -fsched-spec -fregmove -fstrict-aliasing
3422 -fdelete-null-pointer-checks -freorder-blocks -freorder-functions
3423 -falign-functions -falign-jumps -falign-loops -falign-labels
3424 -ftree-vrp -ftree-pre
3425 Please note the warning under -fgcse about invoking -O2 on programs
3427 that use computed gotos.
3428 -O3 Optimize yet more. -O3 turns on all optimizations specified by -O2
3430 and also turns on the -finline-functions, -funswitch-loops and
3431 -fgcse-after-reload options.
3432 -O0 Do not optimize. This is the default.
3434 -Os Optimize for size. -Os enables all -O2 optimizations that do not
3436 typically increase code size. It also performs further optimiza‐
3437 tions designed to reduce code size.
3438 -Os disables the following optimization flags: -falign-functions
3440 -falign-jumps -falign-loops -falign-labels -freorder-blocks
3441 -freorder-blocks-and-partition -fprefetch-loop-arrays
3442 -ftree-vect-loop-version
3443 If you use multiple -O options, with or without level numbers, the
3445 last such option is the one that is effective.
3446 Options of the form -fflag specify machine-independent flags. Most
3448 flags have both positive and negative forms; the negative form of -ffoo
3449 would be -fno-foo. In the table below, only one of the forms is
3450 listed---the one you typically will use. You can figure out the other
3451 form by either removing no- or adding it.
3452 The following options control specific optimizations. They are either
3454 activated by -O options or are related to ones that are. You can use
3455 the following flags in the rare cases when "fine-tuning" of optimiza‐
3456 tions to be performed is desired.
3457 -fno-default-inline
3459 Do not make member functions inline by default merely because they
3460 are defined inside the class scope (C++ only). Otherwise, when you
3461 specify -O, member functions defined inside class scope are com‐
3462 piled inline by default; i.e., you don't need to add inline in
3463 front of the member function name.
3464 -fno-defer-pop
3466 Always pop the arguments to each function call as soon as that
3467 function returns. For machines which must pop arguments after a
3468 function call, the compiler normally lets arguments accumulate on
3469 the stack for several function calls and pops them all at once.
3470 Disabled at levels -O, -O2, -O3, -Os.
3472 -fforce-mem
3474 Force memory operands to be copied into registers before doing
3475 arithmetic on them. This produces better code by making all memory
3476 references potential common subexpressions. When they are not com‐
3477 mon subexpressions, instruction combination should eliminate the
3478 separate register-load. This option is now a nop and will be
3479 removed in 4.2.
3480 -fforce-addr
3482 Force memory address constants to be copied into registers before
3483 doing arithmetic on them.
3484 -fomit-frame-pointer
3486 Don't keep the frame pointer in a register for functions that don't
3487 need one. This avoids the instructions to save, set up and restore
3488 frame pointers; it also makes an extra register available in many
3489 functions. It also makes debugging impossible on some machines.
3490 On some machines, such as the VAX, this flag has no effect, because
3492 the standard calling sequence automatically handles the frame
3493 pointer and nothing is saved by pretending it doesn't exist. The
3494 machine-description macro "FRAME_POINTER_REQUIRED" controls whether
3495 a target machine supports this flag.
3496 Enabled at levels -O, -O2, -O3, -Os.
3498 -foptimize-sibling-calls
3500 Optimize sibling and tail recursive calls.
3501 Enabled at levels -O2, -O3, -Os.
3503 -fno-inline
3505 Don't pay attention to the "inline" keyword. Normally this option
3506 is used to keep the compiler from expanding any functions inline.
3507 Note that if you are not optimizing, no functions can be expanded
3508 inline.
3509 -finline-functions
3511 Integrate all simple functions into their callers. The compiler
3512 heuristically decides which functions are simple enough to be worth
3513 integrating in this way.
3514 If all calls to a given function are integrated, and the function
3516 is declared "static", then the function is normally not output as
3517 assembler code in its own right.
3518 Enabled at level -O3.
3520 -finline-functions-called-once
3522 Consider all "static" functions called once for inlining into their
3523 caller even if they are not marked "inline". If a call to a given
3524 function is integrated, then the function is not output as assem‐
3525 bler code in its own right.
3526 Enabled if -funit-at-a-time is enabled.
3528 -fearly-inlining
3530 Inline functions marked by "always_inline" and functions whose body
3531 seems smaller than the function call overhead early before doing
3532 -fprofile-generate instrumentation and real inlining pass. Doing
3533 so makes profiling significantly cheaper and usually inlining
3534 faster on programs having large chains of nested wrapper functions.
3535 Enabled by default.
3537 -finline-limit=n
3539 By default, GCC limits the size of functions that can be inlined.
3540 This flag allows the control of this limit for functions that are
3541 explicitly marked as inline (i.e., marked with the inline keyword
3542 or defined within the class definition in c++). n is the size of
3543 functions that can be inlined in number of pseudo instructions (not
3544 counting parameter handling). The default value of n is 600.
3545 Increasing this value can result in more inlined code at the cost
3546 of compilation time and memory consumption. Decreasing usually
3547 makes the compilation faster and less code will be inlined (which
3548 presumably means slower programs). This option is particularly
3549 useful for programs that use inlining heavily such as those based
3550 on recursive templates with C++.
3551 Inlining is actually controlled by a number of parameters, which
3553 may be specified individually by using --param name=value. The
3554 -finline-limit=n option sets some of these parameters as follows:
3555 max-inline-insns-single
3557 is set to I<n>/2.
3558 max-inline-insns-auto
3560 is set to I<n>/2.
3561 min-inline-insns
3563 is set to 130 or I<n>/4, whichever is smaller.
3564 max-inline-insns-rtl
3566 is set to I<n>.
3567 See below for a documentation of the individual parameters control‐
3569 ling inlining.
3570 Note: pseudo instruction represents, in this particular context, an
3572 abstract measurement of function's size. In no way does it repre‐
3573 sent a count of assembly instructions and as such its exact meaning
3574 might change from one release to an another.
3575 -fkeep-inline-functions
3577 In C, emit "static" functions that are declared "inline" into the
3578 object file, even if the function has been inlined into all of its
3579 callers. This switch does not affect functions using the "extern
3580 inline" extension in GNU C. In C++, emit any and all inline func‐
3581 tions into the object file.
3582 -fkeep-static-consts
3584 Emit variables declared "static const" when optimization isn't
3585 turned on, even if the variables aren't referenced.
3586 GCC enables this option by default. If you want to force the com‐
3588 piler to check if the variable was referenced, regardless of
3589 whether or not optimization is turned on, use the
3590 -fno-keep-static-consts option.
3591 -fmerge-constants
3593 Attempt to merge identical constants (string constants and floating
3594 point constants) across compilation units.
3595 This option is the default for optimized compilation if the assem‐
3597 bler and linker support it. Use -fno-merge-constants to inhibit
3598 this behavior.
3599 Enabled at levels -O, -O2, -O3, -Os.
3601 -fmerge-all-constants
3603 Attempt to merge identical constants and identical variables.
3604 This option implies -fmerge-constants. In addition to -fmerge-con‐
3606 stants this considers e.g. even constant initialized arrays or ini‐
3607 tialized constant variables with integral or floating point types.
3608 Languages like C or C++ require each non-automatic variable to have
3609 distinct location, so using this option will result in non-conform‐
3610 ing behavior.
3611 -fmodulo-sched
3613 Perform swing modulo scheduling immediately before the first sched‐
3614 uling pass. This pass looks at innermost loops and reorders their
3615 instructions by overlapping different iterations.
3616 -fno-branch-count-reg
3618 Do not use "decrement and branch" instructions on a count register,
3619 but instead generate a sequence of instructions that decrement a
3620 register, compare it against zero, then branch based upon the
3621 result. This option is only meaningful on architectures that sup‐
3622 port such instructions, which include x86, PowerPC, IA-64 and
3623 S/390.
3624 The default is -fbranch-count-reg, enabled when -fstrength-reduce
3626 is enabled.
3627 -fno-function-cse
3629 Do not put function addresses in registers; make each instruction
3630 that calls a constant function contain the function's address
3631 explicitly.
3632 This option results in less efficient code, but some strange hacks
3634 that alter the assembler output may be confused by the optimiza‐
3635 tions performed when this option is not used.
3636 The default is -ffunction-cse
3638 -fno-zero-initialized-in-bss
3640 If the target supports a BSS section, GCC by default puts variables
3641 that are initialized to zero into BSS. This can save space in the
3642 resulting code.
3643 This option turns off this behavior because some programs explic‐
3645 itly rely on variables going to the data section. E.g., so that
3646 the resulting executable can find the beginning of that section
3647 and/or make assumptions based on that.
3648 The default is -fzero-initialized-in-bss.
3650 -fbounds-check
3652 For front-ends that support it, generate additional code to check
3653 that indices used to access arrays are within the declared range.
3654 This is currently only supported by the Java and Fortran
3655 front-ends, where this option defaults to true and false respec‐
3656 tively.
3657 -fmudflap -fmudflapth -fmudflapir
3659 For front-ends that support it (C and C++), instrument all risky
3660 pointer/array dereferencing operations, some standard library
3661 string/heap functions, and some other associated constructs with
3662 range/validity tests. Modules so instrumented should be immune to
3663 buffer overflows, invalid heap use, and some other classes of C/C++
3664 programming errors. The instrumentation relies on a separate run‐
3665 time library (libmudflap), which will be linked into a program if
3666 -fmudflap is given at link time. Run-time behavior of the instru‐
3667 mented program is controlled by the MUDFLAP_OPTIONS environment
3668 variable. See "env MUDFLAP_OPTIONS=-help a.out" for its options.
3669 Use -fmudflapth instead of -fmudflap to compile and to link if your
3671 program is multi-threaded. Use -fmudflapir, in addition to -fmud‐
3672 flap or -fmudflapth, if instrumentation should ignore pointer
3673 reads. This produces less instrumentation (and therefore faster
3674 execution) and still provides some protection against outright mem‐
3675 ory corrupting writes, but allows erroneously read data to propa‐
3676 gate within a program.
3677 -fstrength-reduce
3679 Perform the optimizations of loop strength reduction and elimina‐
3680 tion of iteration variables.
3681 Enabled at levels -O2, -O3, -Os.
3683 -fthread-jumps
3685 Perform optimizations where we check to see if a jump branches to a
3686 location where another comparison subsumed by the first is found.
3687 If so, the first branch is redirected to either the destination of
3688 the second branch or a point immediately following it, depending on
3689 whether the condition is known to be true or false.
3690 Enabled at levels -O2, -O3, -Os.
3692 -fcse-follow-jumps
3694 In common subexpression elimination, scan through jump instructions
3695 when the target of the jump is not reached by any other path. For
3696 example, when CSE encounters an "if" statement with an "else"
3697 clause, CSE will follow the jump when the condition tested is
3698 false.
3699 Enabled at levels -O2, -O3, -Os.
3701 -fcse-skip-blocks
3703 This is similar to -fcse-follow-jumps, but causes CSE to follow
3704 jumps which conditionally skip over blocks. When CSE encounters a
3705 simple "if" statement with no else clause, -fcse-skip-blocks causes
3706 CSE to follow the jump around the body of the "if".
3707 Enabled at levels -O2, -O3, -Os.
3709 -frerun-cse-after-loop
3711 Re-run common subexpression elimination after loop optimizations
3712 has been performed.
3713 Enabled at levels -O2, -O3, -Os.
3715 -frerun-loop-opt
3717 Run the loop optimizer twice.
3718 Enabled at levels -O2, -O3, -Os.
3720 -fgcse
3722 Perform a global common subexpression elimination pass. This pass
3723 also performs global constant and copy propagation.
3724 Note: When compiling a program using computed gotos, a GCC exten‐
3726 sion, you may get better runtime performance if you disable the
3727 global common subexpression elimination pass by adding -fno-gcse to
3728 the command line.
3729 Enabled at levels -O2, -O3, -Os.
3731 -fgcse-lm
3733 When -fgcse-lm is enabled, global common subexpression elimination
3734 will attempt to move loads which are only killed by stores into
3735 themselves. This allows a loop containing a load/store sequence to
3736 be changed to a load outside the loop, and a copy/store within the
3737 loop.
3738 Enabled by default when gcse is enabled.
3740 -fgcse-sm
3742 When -fgcse-sm is enabled, a store motion pass is run after global
3743 common subexpression elimination. This pass will attempt to move
3744 stores out of loops. When used in conjunction with -fgcse-lm,
3745 loops containing a load/store sequence can be changed to a load
3746 before the loop and a store after the loop.
3747 Not enabled at any optimization level.
3749 -fgcse-las
3751 When -fgcse-las is enabled, the global common subexpression elimi‐
3752 nation pass eliminates redundant loads that come after stores to
3753 the same memory location (both partial and full redundancies).
3754 Not enabled at any optimization level.
3756 -fgcse-after-reload
3758 When -fgcse-after-reload is enabled, a redundant load elimination
3759 pass is performed after reload. The purpose of this pass is to
3760 cleanup redundant spilling.
3761 -floop-optimize
3763 Perform loop optimizations: move constant expressions out of loops,
3764 simplify exit test conditions and optionally do strength-reduction
3765 as well.
3766 Enabled at levels -O, -O2, -O3, -Os.
3768 -floop-optimize2
3770 Perform loop optimizations using the new loop optimizer. The opti‐
3771 mizations (loop unrolling, peeling and unswitching, loop invariant
3772 motion) are enabled by separate flags.
3773 -funsafe-loop-optimizations
3775 If given, the loop optimizer will assume that loop indices do not
3776 overflow, and that the loops with nontrivial exit condition are not
3777 infinite. This enables a wider range of loop optimizations even if
3778 the loop optimizer itself cannot prove that these assumptions are
3779 valid. Using -Wunsafe-loop-optimizations, the compiler will warn
3780 you if it finds this kind of loop.
3781 -fcrossjumping
3783 Perform cross-jumping transformation. This transformation unifies
3784 equivalent code and save code size. The resulting code may or may
3785 not perform better than without cross-jumping.
3786 Enabled at levels -O2, -O3, -Os.
3788 -fif-conversion
3790 Attempt to transform conditional jumps into branch-less equiva‐
3791 lents. This include use of conditional moves, min, max, set flags
3792 and abs instructions, and some tricks doable by standard arith‐
3793 metics. The use of conditional execution on chips where it is
3794 available is controlled by "if-conversion2".
3795 Enabled at levels -O, -O2, -O3, -Os.
3797 -fif-conversion2
3799 Use conditional execution (where available) to transform condi‐
3800 tional jumps into branch-less equivalents.
3801 Enabled at levels -O, -O2, -O3, -Os.
3803 -fdelete-null-pointer-checks
3805 Use global dataflow analysis to identify and eliminate useless
3806 checks for null pointers. The compiler assumes that dereferencing
3807 a null pointer would have halted the program. If a pointer is
3808 checked after it has already been dereferenced, it cannot be null.
3809 In some environments, this assumption is not true, and programs can
3811 safely dereference null pointers. Use
3812 -fno-delete-null-pointer-checks to disable this optimization for
3813 programs which depend on that behavior.
3814 Enabled at levels -O2, -O3, -Os.
3816 -fexpensive-optimizations
3818 Perform a number of minor optimizations that are relatively expen‐
3819 sive.
3820 Enabled at levels -O2, -O3, -Os.
3822 -foptimize-register-move
3824 -fregmove
3825 Attempt to reassign register numbers in move instructions and as
3826 operands of other simple instructions in order to maximize the
3827 amount of register tying. This is especially helpful on machines
3828 with two-operand instructions.
3829 Note -fregmove and -foptimize-register-move are the same optimiza‐
3831 tion.
3832 Enabled at levels -O2, -O3, -Os.
3834 -fdelayed-branch
3836 If supported for the target machine, attempt to reorder instruc‐
3837 tions to exploit instruction slots available after delayed branch
3838 instructions.
3839 Enabled at levels -O, -O2, -O3, -Os.
3841 -fschedule-insns
3843 If supported for the target machine, attempt to reorder instruc‐
3844 tions to eliminate execution stalls due to required data being
3845 unavailable. This helps machines that have slow floating point or
3846 memory load instructions by allowing other instructions to be
3847 issued until the result of the load or floating point instruction
3848 is required.
3849 Enabled at levels -O2, -O3, -Os.
3851 -fschedule-insns2
3853 Similar to -fschedule-insns, but requests an additional pass of
3854 instruction scheduling after register allocation has been done.
3855 This is especially useful on machines with a relatively small num‐
3856 ber of registers and where memory load instructions take more than
3857 one cycle.
3858 Enabled at levels -O2, -O3, -Os.
3860 -fno-sched-interblock
3862 Don't schedule instructions across basic blocks. This is normally
3863 enabled by default when scheduling before register allocation, i.e.
3864 with -fschedule-insns or at -O2 or higher.
3865 -fno-sched-spec
3867 Don't allow speculative motion of non-load instructions. This is
3868 normally enabled by default when scheduling before register alloca‐
3869 tion, i.e. with -fschedule-insns or at -O2 or higher.
3870 -fsched-spec-load
3872 Allow speculative motion of some load instructions. This only
3873 makes sense when scheduling before register allocation, i.e. with
3874 -fschedule-insns or at -O2 or higher.
3875 -fsched-spec-load-dangerous
3877 Allow speculative motion of more load instructions. This only
3878 makes sense when scheduling before register allocation, i.e. with
3879 -fschedule-insns or at -O2 or higher.
3880 -fsched-stalled-insns=n
3882 Define how many insns (if any) can be moved prematurely from the
3883 queue of stalled insns into the ready list, during the second
3884 scheduling pass.
3885 -fsched-stalled-insns-dep=n
3887 Define how many insn groups (cycles) will be examined for a depen‐
3888 dency on a stalled insn that is candidate for premature removal
3889 from the queue of stalled insns. Has an effect only during the
3890 second scheduling pass, and only if -fsched-stalled-insns is used
3891 and its value is not zero.
3892 -fsched2-use-superblocks
3894 When scheduling after register allocation, do use superblock sched‐
3895 uling algorithm. Superblock scheduling allows motion across basic
3896 block boundaries resulting on faster schedules. This option is
3897 experimental, as not all machine descriptions used by GCC model the
3898 CPU closely enough to avoid unreliable results from the algorithm.
3899 This only makes sense when scheduling after register allocation,
3901 i.e. with -fschedule-insns2 or at -O2 or higher.
3902 -fsched2-use-traces
3904 Use -fsched2-use-superblocks algorithm when scheduling after regis‐
3905 ter allocation and additionally perform code duplication in order
3906 to increase the size of superblocks using tracer pass. See
3907 -ftracer for details on trace formation.
3908 This mode should produce faster but significantly longer programs.
3910 Also without -fbranch-probabilities the traces constructed may not
3911 match the reality and hurt the performance. This only makes sense
3912 when scheduling after register allocation, i.e. with -fsched‐
3913 ule-insns2 or at -O2 or higher.
3914 -freschedule-modulo-scheduled-loops
3916 The modulo scheduling comes before the traditional scheduling, if a
3917 loop was modulo scheduled we may want to prevent the later schedul‐
3918 ing passes from changing its schedule, we use this option to con‐
3919 trol that.
3920 -fcaller-saves
3922 Enable values to be allocated in registers that will be clobbered
3923 by function calls, by emitting extra instructions to save and
3924 restore the registers around such calls. Such allocation is done
3925 only when it seems to result in better code than would otherwise be
3926 produced.
3927 This option is always enabled by default on certain machines, usu‐
3929 ally those which have no call-preserved registers to use instead.
3930 Enabled at levels -O2, -O3, -Os.
3932 -ftree-pre
3934 Perform Partial Redundancy Elimination (PRE) on trees. This flag
3935 is enabled by default at -O2 and -O3.
3936 -ftree-fre
3938 Perform Full Redundancy Elimination (FRE) on trees. The difference
3939 between FRE and PRE is that FRE only considers expressions that are
3940 computed on all paths leading to the redundant computation. This
3941 analysis faster than PRE, though it exposes fewer redundancies.
3942 This flag is enabled by default at -O and higher.
3943 -ftree-copy-prop
3945 Perform copy propagation on trees. This pass eliminates unneces‐
3946 sary copy operations. This flag is enabled by default at -O and
3947 higher.
3948 -ftree-store-copy-prop
3950 Perform copy propagation of memory loads and stores. This pass
3951 eliminates unnecessary copy operations in memory references (struc‐
3952 tures, global variables, arrays, etc). This flag is enabled by
3953 default at -O2 and higher.
3954 -ftree-salias
3956 Perform structural alias analysis on trees. This flag is enabled
3957 by default at -O and higher.
3958 -ftree-sink
3960 Perform forward store motion on trees. This flag is enabled by
3961 default at -O and higher.
3962 -ftree-ccp
3964 Perform sparse conditional constant propagation (CCP) on trees.
3965 This pass only operates on local scalar variables and is enabled by
3966 default at -O and higher.
3967 -ftree-store-ccp
3969 Perform sparse conditional constant propagation (CCP) on trees.
3970 This pass operates on both local scalar variables and memory stores
3971 and loads (global variables, structures, arrays, etc). This flag
3972 is enabled by default at -O2 and higher.
3973 -ftree-dce
3975 Perform dead code elimination (DCE) on trees. This flag is enabled
3976 by default at -O and higher.
3977 -ftree-dominator-opts
3979 Perform a variety of simple scalar cleanups (constant/copy propaga‐
3980 tion, redundancy elimination, range propagation and expression sim‐
3981 plification) based on a dominator tree traversal. This also per‐
3982 forms jump threading (to reduce jumps to jumps). This flag is
3983 enabled by default at -O and higher.
3984 -ftree-ch
3986 Perform loop header copying on trees. This is beneficial since it
3987 increases effectiveness of code motion optimizations. It also
3988 saves one jump. This flag is enabled by default at -O and higher.
3989 It is not enabled for -Os, since it usually increases code size.
3990 -ftree-loop-optimize
3992 Perform loop optimizations on trees. This flag is enabled by
3993 default at -O and higher.
3994 -ftree-loop-linear
3996 Perform linear loop transformations on tree. This flag can improve
3997 cache performance and allow further loop optimizations to take
3998 place.
3999 -ftree-loop-im
4001 Perform loop invariant motion on trees. This pass moves only
4002 invariants that would be hard to handle at RTL level (function
4003 calls, operations that expand to nontrivial sequences of insns).
4004 With -funswitch-loops it also moves operands of conditions that are
4005 invariant out of the loop, so that we can use just trivial invari‐
4006 antness analysis in loop unswitching. The pass also includes store
4007 motion.
4008 -ftree-loop-ivcanon
4010 Create a canonical counter for number of iterations in the loop for
4011 that determining number of iterations requires complicated analy‐
4012 sis. Later optimizations then may determine the number easily.
4013 Useful especially in connection with unrolling.
4014 -fivopts
4016 Perform induction variable optimizations (strength reduction,
4017 induction variable merging and induction variable elimination) on
4018 trees.
4019 -ftree-sra
4021 Perform scalar replacement of aggregates. This pass replaces
4022 structure references with scalars to prevent committing structures
4023 to memory too early. This flag is enabled by default at -O and
4024 higher.
4025 -ftree-copyrename
4027 Perform copy renaming on trees. This pass attempts to rename com‐
4028 piler temporaries to other variables at copy locations, usually
4029 resulting in variable names which more closely resemble the origi‐
4030 nal variables. This flag is enabled by default at -O and higher.
4031 -ftree-ter
4033 Perform temporary expression replacement during the SSA->normal
4034 phase. Single use/single def temporaries are replaced at their use
4035 location with their defining expression. This results in non-GIM‐
4036 PLE code, but gives the expanders much more complex trees to work
4037 on resulting in better RTL generation. This is enabled by default
4038 at -O and higher.
4039 -ftree-lrs
4041 Perform live range splitting during the SSA->normal phase. Dis‐
4042 tinct live ranges of a variable are split into unique variables,
4043 allowing for better optimization later. This is enabled by default
4044 at -O and higher.
4045 -ftree-vectorize
4047 Perform loop vectorization on trees.
4048 -ftree-vect-loop-version
4050 Perform loop versioning when doing loop vectorization on trees.
4051 When a loop appears to be vectorizable except that data alignment
4052 or data dependence cannot be determined at compile time then vec‐
4053 torized and non-vectorized versions of the loop are generated along
4054 with runtime checks for alignment or dependence to control which
4055 version is executed. This option is enabled by default except at
4056 level -Os where it is disabled.
4057 -ftree-vrp
4059 Perform Value Range Propagation on trees. This is similar to the
4060 constant propagation pass, but instead of values, ranges of values
4061 are propagated. This allows the optimizers to remove unnecessary
4062 range checks like array bound checks and null pointer checks. This
4063 is enabled by default at -O2 and higher. Null pointer check elimi‐
4064 nation is only done if -fdelete-null-pointer-checks is enabled.
4065 -ftracer
4067 Perform tail duplication to enlarge superblock size. This trans‐
4068 formation simplifies the control flow of the function allowing
4069 other optimizations to do better job.
4070 -funroll-loops
4072 Unroll loops whose number of iterations can be determined at com‐
4073 pile time or upon entry to the loop. -funroll-loops implies both
4074 -fstrength-reduce and -frerun-cse-after-loop. This option makes
4075 code larger, and may or may not make it run faster.
4076 -funroll-all-loops
4078 Unroll all loops, even if their number of iterations is uncertain
4079 when the loop is entered. This usually makes programs run more
4080 slowly. -funroll-all-loops implies the same options as -fun‐
4081 roll-loops,
4082 -fsplit-ivs-in-unroller
4084 Enables expressing of values of induction variables in later itera‐
4085 tions of the unrolled loop using the value in the first iteration.
4086 This breaks long dependency chains, thus improving efficiency of
4087 the scheduling passes.
4088 Combination of -fweb and CSE is often sufficient to obtain the same
4090 effect. However in cases the loop body is more complicated than a
4091 single basic block, this is not reliable. It also does not work at
4092 all on some of the architectures due to restrictions in the CSE
4093 pass.
4094 This optimization is enabled by default.
4096 -fvariable-expansion-in-unroller
4098 With this option, the compiler will create multiple copies of some
4099 local variables when unrolling a loop which can result in superior
4100 code.
4101 -fprefetch-loop-arrays
4103 If supported by the target machine, generate instructions to
4104 prefetch memory to improve the performance of loops that access
4105 large arrays.
4106 These options may generate better or worse code; results are highly
4108 dependent on the structure of loops within the source code.
4109 -fno-peephole
4111 -fno-peephole2
4112 Disable any machine-specific peephole optimizations. The differ‐
4113 ence between -fno-peephole and -fno-peephole2 is in how they are
4114 implemented in the compiler; some targets use one, some use the
4115 other, a few use both.
4116 -fpeephole is enabled by default. -fpeephole2 enabled at levels
4118 -O2, -O3, -Os.
4119 -fno-guess-branch-probability
4121 Do not guess branch probabilities using heuristics.
4122 GCC will use heuristics to guess branch probabilities if they are
4124 not provided by profiling feedback (-fprofile-arcs). These heuris‐
4125 tics are based on the control flow graph. If some branch probabil‐
4126 ities are specified by __builtin_expect, then the heuristics will
4127 be used to guess branch probabilities for the rest of the control
4128 flow graph, taking the __builtin_expect info into account. The
4129 interactions between the heuristics and __builtin_expect can be
4130 complex, and in some cases, it may be useful to disable the heuris‐
4131 tics so that the effects of __builtin_expect are easier to under‐
4132 stand.
4133 The default is -fguess-branch-probability at levels -O, -O2, -O3,
4135 -Os.
4136 -freorder-blocks
4138 Reorder basic blocks in the compiled function in order to reduce
4139 number of taken branches and improve code locality.
4140 Enabled at levels -O2, -O3.
4142 -freorder-blocks-and-partition
4144 In addition to reordering basic blocks in the compiled function, in
4145 order to reduce number of taken branches, partitions hot and cold
4146 basic blocks into separate sections of the assembly and .o files,
4147 to improve paging and cache locality performance.
4148 This optimization is automatically turned off in the presence of
4150 exception handling, for linkonce sections, for functions with a
4151 user-defined section attribute and on any architecture that does
4152 not support named sections.
4153 -freorder-functions
4155 Reorder functions in the object file in order to improve code
4156 locality. This is implemented by using special subsections
4157 ".text.hot" for most frequently executed functions and
4158 ".text.unlikely" for unlikely executed functions. Reordering is
4159 done by the linker so object file format must support named sec‐
4160 tions and linker must place them in a reasonable way.
4161 Also profile feedback must be available in to make this option
4163 effective. See -fprofile-arcs for details.
4164 Enabled at levels -O2, -O3, -Os.
4166 -fstrict-aliasing
4168 Allows the compiler to assume the strictest aliasing rules applica‐
4169 ble to the language being compiled. For C (and C++), this acti‐
4170 vates optimizations based on the type of expressions. In particu‐
4171 lar, an object of one type is assumed never to reside at the same
4172 address as an object of a different type, unless the types are
4173 almost the same. For example, an "unsigned int" can alias an
4174 "int", but not a "void*" or a "double". A character type may alias
4175 any other type.
4176 Pay special attention to code like this:
4178 union a_union {
4180 int i;
4181 double d;
4182 };
4183 int f() {
4185 a_union t;
4186 t.d = 3.0;
4187 return t.i;
4188 }
4189 The practice of reading from a different union member than the one
4191 most recently written to (called "type-punning") is common. Even
4192 with -fstrict-aliasing, type-punning is allowed, provided the mem‐
4193 ory is accessed through the union type. So, the code above will
4194 work as expected. However, this code might not:
4195 int f() {
4197 a_union t;
4198 int* ip;
4199 t.d = 3.0;
4200 ip = &t.i;
4201 return *ip;
4202 }
4203 Every language that wishes to perform language-specific alias anal‐
4205 ysis should define a function that computes, given an "tree" node,
4206 an alias set for the node. Nodes in different alias sets are not
4207 allowed to alias. For an example, see the C front-end function
4208 "c_get_alias_set".
4209 Enabled at levels -O2, -O3, -Os.
4211 -falign-functions
4213 -falign-functions=n
4214 Align the start of functions to the next power-of-two greater than
4215 n, skipping up to n bytes. For instance, -falign-functions=32
4216 aligns functions to the next 32-byte boundary, but -falign-func‐
4217 tions=24 would align to the next 32-byte boundary only if this can
4218 be done by skipping 23 bytes or less.
4219 -fno-align-functions and -falign-functions=1 are equivalent and
4221 mean that functions will not be aligned.
4222 Some assemblers only support this flag when n is a power of two; in
4224 that case, it is rounded up.
4225 If n is not specified or is zero, use a machine-dependent default.
4227 Enabled at levels -O2, -O3.
4229 -falign-labels
4231 -falign-labels=n
4232 Align all branch targets to a power-of-two boundary, skipping up to
4233 n bytes like -falign-functions. This option can easily make code
4234 slower, because it must insert dummy operations for when the branch
4235 target is reached in the usual flow of the code.
4236 -fno-align-labels and -falign-labels=1 are equivalent and mean that
4238 labels will not be aligned.
4239 If -falign-loops or -falign-jumps are applicable and are greater
4241 than this value, then their values are used instead.
4242 If n is not specified or is zero, use a machine-dependent default
4244 which is very likely to be 1, meaning no alignment.
4245 Enabled at levels -O2, -O3.
4247 -falign-loops
4249 -falign-loops=n
4250 Align loops to a power-of-two boundary, skipping up to n bytes like
4251 -falign-functions. The hope is that the loop will be executed many
4252 times, which will make up for any execution of the dummy opera‐
4253 tions.
4254 -fno-align-loops and -falign-loops=1 are equivalent and mean that
4256 loops will not be aligned.
4257 If n is not specified or is zero, use a machine-dependent default.
4259 Enabled at levels -O2, -O3.
4261 -falign-jumps
4263 -falign-jumps=n
4264 Align branch targets to a power-of-two boundary, for branch targets
4265 where the targets can only be reached by jumping, skipping up to n
4266 bytes like -falign-functions. In this case, no dummy operations
4267 need be executed.
4268 -fno-align-jumps and -falign-jumps=1 are equivalent and mean that
4270 loops will not be aligned.
4271 If n is not specified or is zero, use a machine-dependent default.
4273 Enabled at levels -O2, -O3.
4275 -funit-at-a-time
4277 Parse the whole compilation unit before starting to produce code.
4278 This allows some extra optimizations to take place but consumes
4279 more memory (in general). There are some compatibility issues with
4280 unit-at-at-time mode:
4281 * enabling unit-at-a-time mode may change the order in which
4283 functions, variables, and top-level "asm" statements are emit‐
4284 ted, and will likely break code relying on some particular
4285 ordering. The majority of such top-level "asm" statements,
4286 though, can be replaced by "section" attributes.
4287 * unit-at-a-time mode removes unreferenced static variables and
4289 functions. This may result in undefined references when an
4290 "asm" statement refers directly to variables or functions that
4291 are otherwise unused. In that case either the variable/func‐
4292 tion shall be listed as an operand of the "asm" statement oper‐
4293 and or, in the case of top-level "asm" statements the attribute
4294 "used" shall be used on the declaration.
4295 * Static functions now can use non-standard passing conventions
4297 that may break "asm" statements calling functions directly.
4298 Again, attribute "used" will prevent this behavior.
4299 As a temporary workaround, -fno-unit-at-a-time can be used, but
4301 this scheme may not be supported by future releases of GCC.
4302 Enabled at levels -O, -O2, -O3, -Os.
4304 -fweb
4306 Constructs webs as commonly used for register allocation purposes
4307 and assign each web individual pseudo register. This allows the
4308 register allocation pass to operate on pseudos directly, but also
4309 strengthens several other optimization passes, such as CSE, loop
4310 optimizer and trivial dead code remover. It can, however, make
4311 debugging impossible, since variables will no longer stay in a
4312 "home register".
4313 Enabled by default with -funroll-loops.
4315 -fwhole-program
4317 Assume that the current compilation unit represents whole program
4318 being compiled. All public functions and variables with the excep‐
4319 tion of "main" and those merged by attribute "externally_visible"
4320 become static functions and in a affect gets more aggressively
4321 optimized by interprocedural optimizers. While this option is
4322 equivalent to proper use of "static" keyword for programs consist‐
4323 ing of single file, in combination with option --combine this flag
4324 can be used to compile most of smaller scale C programs since the
4325 functions and variables become local for the whole combined compi‐
4326 lation unit, not for the single source file itself.
4327 -fno-cprop-registers
4329 After register allocation and post-register allocation instruction
4330 splitting, we perform a copy-propagation pass to try to reduce
4331 scheduling dependencies and occasionally eliminate the copy.
4332 Disabled at levels -O, -O2, -O3, -Os.
4334 -fprofile-generate
4336 Enable options usually used for instrumenting application to pro‐
4337 duce profile useful for later recompilation with profile feedback
4338 based optimization. You must use -fprofile-generate both when com‐
4339 piling and when linking your program.
4340 The following options are enabled: "-fprofile-arcs", "-fpro‐
4342 file-values", "-fvpt".
4343 -fprofile-use
4345 Enable profile feedback directed optimizations, and optimizations
4346 generally profitable only with profile feedback available.
4347 The following options are enabled: "-fbranch-probabilities",
4349 "-fvpt", "-funroll-loops", "-fpeel-loops", "-ftracer",
4350 "-fno-loop-optimize".
4351 The following options control compiler behavior regarding floating
4353 point arithmetic. These options trade off between speed and correct‐
4354 ness. All must be specifically enabled.
4355 -ffloat-store
4357 Do not store floating point variables in registers, and inhibit
4358 other options that might change whether a floating point value is
4359 taken from a register or memory.
4360 This option prevents undesirable excess precision on machines such
4362 as the 68000 where the floating registers (of the 68881) keep more
4363 precision than a "double" is supposed to have. Similarly for the
4364 x86 architecture. For most programs, the excess precision does
4365 only good, but a few programs rely on the precise definition of
4366 IEEE floating point. Use -ffloat-store for such programs, after
4367 modifying them to store all pertinent intermediate computations
4368 into variables.
4369 -ffast-math
4371 Sets -fno-math-errno, -funsafe-math-optimizations, -fno-trap‐
4372 ping-math, -ffinite-math-only, -fno-rounding-math, -fno-signal‐
4373 ing-nans and fcx-limited-range.
4374 This option causes the preprocessor macro "__FAST_MATH__" to be
4376 defined.
4377 This option should never be turned on by any -O option since it can
4379 result in incorrect output for programs which depend on an exact
4380 implementation of IEEE or ISO rules/specifications for math func‐
4381 tions.
4382 -fno-math-errno
4384 Do not set ERRNO after calling math functions that are executed
4385 with a single instruction, e.g., sqrt. A program that relies on
4386 IEEE exceptions for math error handling may want to use this flag
4387 for speed while maintaining IEEE arithmetic compatibility.
4388 This option should never be turned on by any -O option since it can
4390 result in incorrect output for programs which depend on an exact
4391 implementation of IEEE or ISO rules/specifications for math func‐
4392 tions.
4393 The default is -fmath-errno.
4395 On Darwin systems, the math library never sets "errno". There is
4397 therefore no reason for the compiler to consider the possibility
4398 that it might, and -fno-math-errno is the default.
4399 -funsafe-math-optimizations
4401 Allow optimizations for floating-point arithmetic that (a) assume
4402 that arguments and results are valid and (b) may violate IEEE or
4403 ANSI standards. When used at link-time, it may include libraries
4404 or startup files that change the default FPU control word or other
4405 similar optimizations.
4406 This option should never be turned on by any -O option since it can
4408 result in incorrect output for programs which depend on an exact
4409 implementation of IEEE or ISO rules/specifications for math func‐
4410 tions.
4411 The default is -fno-unsafe-math-optimizations.
4413 -ffinite-math-only
4415 Allow optimizations for floating-point arithmetic that assume that
4416 arguments and results are not NaNs or +-Infs.
4417 This option should never be turned on by any -O option since it can
4419 result in incorrect output for programs which depend on an exact
4420 implementation of IEEE or ISO rules/specifications.
4421 The default is -fno-finite-math-only.
4423 -fno-trapping-math
4425 Compile code assuming that floating-point operations cannot gener‐
4426 ate user-visible traps. These traps include division by zero,
4427 overflow, underflow, inexact result and invalid operation. This
4428 option implies -fno-signaling-nans. Setting this option may allow
4429 faster code if one relies on "non-stop" IEEE arithmetic, for exam‐
4430 ple.
4431 This option should never be turned on by any -O option since it can
4433 result in incorrect output for programs which depend on an exact
4434 implementation of IEEE or ISO rules/specifications for math func‐
4435 tions.
4436 The default is -ftrapping-math.
4438 -frounding-math
4440 Disable transformations and optimizations that assume default
4441 floating point rounding behavior. This is round-to-zero for all
4442 floating point to integer conversions, and round-to-nearest for all
4443 other arithmetic truncations. This option should be specified for
4444 programs that change the FP rounding mode dynamically, or that may
4445 be executed with a non-default rounding mode. This option disables
4446 constant folding of floating point expressions at compile-time
4447 (which may be affected by rounding mode) and arithmetic transforma‐
4448 tions that are unsafe in the presence of sign-dependent rounding
4449 modes.
4450 The default is -fno-rounding-math.
4452 This option is experimental and does not currently guarantee to
4454 disable all GCC optimizations that are affected by rounding mode.
4455 Future versions of GCC may provide finer control of this setting
4456 using C99's "FENV_ACCESS" pragma. This command line option will be
4457 used to specify the default state for "FENV_ACCESS".
4458 -fsignaling-nans
4460 Compile code assuming that IEEE signaling NaNs may generate user-
4461 visible traps during floating-point operations. Setting this
4462 option disables optimizations that may change the number of excep‐
4463 tions visible with signaling NaNs. This option implies -ftrap‐
4464 ping-math.
4465 This option causes the preprocessor macro "__SUPPORT_SNAN__" to be
4467 defined.
4468 The default is -fno-signaling-nans.
4470 This option is experimental and does not currently guarantee to
4472 disable all GCC optimizations that affect signaling NaN behavior.
4473 -fsingle-precision-constant
4475 Treat floating point constant as single precision constant instead
4476 of implicitly converting it to double precision constant.
4477 -fcx-limited-range
4479 -fno-cx-limited-range
4480 When enabled, this option states that a range reduction step is not
4481 needed when performing complex division. The default is
4482 -fno-cx-limited-range, but is enabled by -ffast-math.
4483 This option controls the default setting of the ISO C99 "CX_LIM‐
4485 ITED_RANGE" pragma. Nevertheless, the option applies to all lan‐
4486 guages.
4487 The following options control optimizations that may improve perfor‐
4489 mance, but are not enabled by any -O options. This section includes
4490 experimental options that may produce broken code.
4491 -fbranch-probabilities
4493 After running a program compiled with -fprofile-arcs, you can com‐
4494 pile it a second time using -fbranch-probabilities, to improve
4495 optimizations based on the number of times each branch was taken.
4496 When the program compiled with -fprofile-arcs exits it saves arc
4497 execution counts to a file called sourcename.gcda for each source
4498 file The information in this data file is very dependent on the
4499 structure of the generated code, so you must use the same source
4500 code and the same optimization options for both compilations.
4501 With -fbranch-probabilities, GCC puts a REG_BR_PROB note on each
4503 JUMP_INSN and CALL_INSN. These can be used to improve optimiza‐
4504 tion. Currently, they are only used in one place: in reorg.c,
4505 instead of guessing which path a branch is mostly to take, the
4506 REG_BR_PROB values are used to exactly determine which path is
4507 taken more often.
4508 -fprofile-values
4510 If combined with -fprofile-arcs, it adds code so that some data
4511 about values of expressions in the program is gathered.
4512 With -fbranch-probabilities, it reads back the data gathered from
4514 profiling values of expressions and adds REG_VALUE_PROFILE notes to
4515 instructions for their later usage in optimizations.
4516 Enabled with -fprofile-generate and -fprofile-use.
4518 -fvpt
4520 If combined with -fprofile-arcs, it instructs the compiler to add a
4521 code to gather information about values of expressions.
4522 With -fbranch-probabilities, it reads back the data gathered and
4524 actually performs the optimizations based on them. Currently the
4525 optimizations include specialization of division operation using
4526 the knowledge about the value of the denominator.
4527 -frename-registers
4529 Attempt to avoid false dependencies in scheduled code by making use
4530 of registers left over after register allocation. This optimiza‐
4531 tion will most benefit processors with lots of registers. Depend‐
4532 ing on the debug information format adopted by the target, however,
4533 it can make debugging impossible, since variables will no longer
4534 stay in a "home register".
4535 Enabled by default with -funroll-loops.
4537 -ftracer
4539 Perform tail duplication to enlarge superblock size. This trans‐
4540 formation simplifies the control flow of the function allowing
4541 other optimizations to do better job.
4542 Enabled with -fprofile-use.
4544 -funroll-loops
4546 Unroll loops whose number of iterations can be determined at com‐
4547 pile time or upon entry to the loop. -funroll-loops implies -fre‐
4548 run-cse-after-loop, -fweb and -frename-registers. It also turns on
4549 complete loop peeling (i.e. complete removal of loops with small
4550 constant number of iterations). This option makes code larger, and
4551 may or may not make it run faster.
4552 Enabled with -fprofile-use.
4554 -funroll-all-loops
4556 Unroll all loops, even if their number of iterations is uncertain
4557 when the loop is entered. This usually makes programs run more
4558 slowly. -funroll-all-loops implies the same options as -fun‐
4559 roll-loops.
4560 -fpeel-loops
4562 Peels the loops for that there is enough information that they do
4563 not roll much (from profile feedback). It also turns on complete
4564 loop peeling (i.e. complete removal of loops with small constant
4565 number of iterations).
4566 Enabled with -fprofile-use.
4568 -fmove-loop-invariants
4570 Enables the loop invariant motion pass in the new loop optimizer.
4571 Enabled at level -O1
4572 -funswitch-loops
4574 Move branches with loop invariant conditions out of the loop, with
4575 duplicates of the loop on both branches (modified according to
4576 result of the condition).
4577 -fprefetch-loop-arrays
4579 If supported by the target machine, generate instructions to
4580 prefetch memory to improve the performance of loops that access
4581 large arrays.
4582 Disabled at level -Os.
4584 -ffunction-sections
4586 -fdata-sections
4587 Place each function or data item into its own section in the output
4588 file if the target supports arbitrary sections. The name of the
4589 function or the name of the data item determines the section's name
4590 in the output file.
4591 Use these options on systems where the linker can perform optimiza‐
4593 tions to improve locality of reference in the instruction space.
4594 Most systems using the ELF object format and SPARC processors run‐
4595 ning Solaris 2 have linkers with such optimizations. AIX may have
4596 these optimizations in the future.
4597 Only use these options when there are significant benefits from
4599 doing so. When you specify these options, the assembler and linker
4600 will create larger object and executable files and will also be
4601 slower. You will not be able to use "gprof" on all systems if you
4602 specify this option and you may have problems with debugging if you
4603 specify both this option and -g.
4604 -fbranch-target-load-optimize
4606 Perform branch target register load optimization before prologue /
4607 epilogue threading. The use of target registers can typically be
4608 exposed only during reload, thus hoisting loads out of loops and
4609 doing inter-block scheduling needs a separate optimization pass.
4610 -fbranch-target-load-optimize2
4612 Perform branch target register load optimization after prologue /
4613 epilogue threading.
4614 -fbtr-bb-exclusive
4616 When performing branch target register load optimization, don't re‐
4617 use branch target registers in within any basic block.
4618 -fstack-protector
4620 Emit extra code to check for buffer overflows, such as stack smash‐
4621 ing attacks. This is done by adding a guard variable to functions
4622 with vulnerable objects. This includes functions that call alloca,
4623 and functions with buffers larger than 8 bytes. The guards are
4624 initialized when a function is entered and then checked when the
4625 function exits. If a guard check fails, an error message is
4626 printed and the program exits.
4627 -fstack-protector-all
4629 Like -fstack-protector except that all functions are protected.
4630 --param name=value
4632 In some places, GCC uses various constants to control the amount of
4633 optimization that is done. For example, GCC will not inline func‐
4634 tions that contain more that a certain number of instructions. You
4635 can control some of these constants on the command-line using the
4636 --param option.
4637 The names of specific parameters, and the meaning of the values,
4639 are tied to the internals of the compiler, and are subject to
4640 change without notice in future releases.
4641 In each case, the value is an integer. The allowable choices for
4643 name are given in the following table:
4644 salias-max-implicit-fields
4646 The maximum number of fields in a variable without direct
4647 structure accesses for which structure aliasing will consider
4648 trying to track each field. The default is 5
4649 sra-max-structure-size
4651 The maximum structure size, in bytes, at which the scalar
4652 replacement of aggregates (SRA) optimization will perform block
4653 copies. The default value, 0, implies that GCC will select the
4654 most appropriate size itself.
4655 sra-field-structure-ratio
4657 The threshold ratio (as a percentage) between instantiated
4658 fields and the complete structure size. We say that if the
4659 ratio of the number of bytes in instantiated fields to the num‐
4660 ber of bytes in the complete structure exceeds this parameter,
4661 then block copies are not used. The default is 75.
4662 max-crossjump-edges
4664 The maximum number of incoming edges to consider for crossjump‐
4665 ing. The algorithm used by -fcrossjumping is O(N^2) in the
4666 number of edges incoming to each block. Increasing values mean
4667 more aggressive optimization, making the compile time increase
4668 with probably small improvement in executable size.
4669 min-crossjump-insns
4671 The minimum number of instructions which must be matched at the
4672 end of two blocks before crossjumping will be performed on
4673 them. This value is ignored in the case where all instructions
4674 in the block being crossjumped from are matched. The default
4675 value is 5.
4676 max-grow-copy-bb-insns
4678 The maximum code size expansion factor when copying basic
4679 blocks instead of jumping. The expansion is relative to a jump
4680 instruction. The default value is 8.
4681 max-goto-duplication-insns
4683 The maximum number of instructions to duplicate to a block that
4684 jumps to a computed goto. To avoid O(N^2) behavior in a number
4685 of passes, GCC factors computed gotos early in the compilation
4686 process, and unfactors them as late as possible. Only computed
4687 jumps at the end of a basic blocks with no more than max-goto-
4688 duplication-insns are unfactored. The default value is 8.
4689 max-delay-slot-insn-search
4691 The maximum number of instructions to consider when looking for
4692 an instruction to fill a delay slot. If more than this arbi‐
4693 trary number of instructions is searched, the time savings from
4694 filling the delay slot will be minimal so stop searching.
4695 Increasing values mean more aggressive optimization, making the
4696 compile time increase with probably small improvement in exe‐
4697 cutable run time.
4698 max-delay-slot-live-search
4700 When trying to fill delay slots, the maximum number of instruc‐
4701 tions to consider when searching for a block with valid live
4702 register information. Increasing this arbitrarily chosen value
4703 means more aggressive optimization, increasing the compile
4704 time. This parameter should be removed when the delay slot
4705 code is rewritten to maintain the control-flow graph.
4706 max-gcse-memory
4708 The approximate maximum amount of memory that will be allocated
4709 in order to perform the global common subexpression elimination
4710 optimization. If more memory than specified is required, the
4711 optimization will not be done.
4712 max-gcse-passes
4714 The maximum number of passes of GCSE to run. The default is 1.
4715 max-pending-list-length
4717 The maximum number of pending dependencies scheduling will
4718 allow before flushing the current state and starting over.
4719 Large functions with few branches or calls can create exces‐
4720 sively large lists which needlessly consume memory and
4721 resources.
4722 max-inline-insns-single
4724 Several parameters control the tree inliner used in gcc. This
4725 number sets the maximum number of instructions (counted in
4726 GCC's internal representation) in a single function that the
4727 tree inliner will consider for inlining. This only affects
4728 functions declared inline and methods implemented in a class
4729 declaration (C++). The default value is 450.
4730 max-inline-insns-auto
4732 When you use -finline-functions (included in -O3), a lot of
4733 functions that would otherwise not be considered for inlining
4734 by the compiler will be investigated. To those functions, a
4735 different (more restrictive) limit compared to functions
4736 declared inline can be applied. The default value is 90.
4737 large-function-insns
4739 The limit specifying really large functions. For functions
4740 larger than this limit after inlining inlining is constrained
4741 by --param large-function-growth. This parameter is useful
4742 primarily to avoid extreme compilation time caused by non-lin‐
4743 ear algorithms used by the backend. This parameter is ignored
4744 when -funit-at-a-time is not used. The default value is 2700.
4745 large-function-growth
4747 Specifies maximal growth of large function caused by inlining
4748 in percents. This parameter is ignored when -funit-at-a-time
4749 is not used. The default value is 100 which limits large func‐
4750 tion growth to 2.0 times the original size.
4751 large-unit-insns
4753 The limit specifying large translation unit. Growth caused by
4754 inlining of units larger than this limit is limited by --param
4755 inline-unit-growth. For small units this might be too tight
4756 (consider unit consisting of function A that is inline and B
4757 that just calls A three time. If B is small relative to A, the
4758 growth of unit is 300\% and yet such inlining is very sane.
4759 For very large units consisting of small inlininable functions
4760 however the overall unit growth limit is needed to avoid expo‐
4761 nential explosion of code size. Thus for smaller units, the
4762 size is increased to --param large-unit-insns before aplying
4763 --param inline-unit-growth. The default is 10000
4764 inline-unit-growth
4766 Specifies maximal overall growth of the compilation unit caused
4767 by inlining. This parameter is ignored when -funit-at-a-time
4768 is not used. The default value is 50 which limits unit growth
4769 to 1.5 times the original size.
4770 max-inline-insns-recursive
4772 max-inline-insns-recursive-auto
4773 Specifies maximum number of instructions out-of-line copy of
4774 self recursive inline function can grow into by performing
4775 recursive inlining.
4776 For functions declared inline --param max-inline-insns-recur‐
4778 sive is taken into account. For function not declared inline,
4779 recursive inlining happens only when -finline-functions
4780 (included in -O3) is enabled and --param max-inline-insns-
4781 recursive-auto is used. The default value is 450.
4782 max-inline-recursive-depth
4784 max-inline-recursive-depth-auto
4785 Specifies maximum recursion depth used by the recursive inlin‐
4786 ing.
4787 For functions declared inline --param max-inline-recursive-
4789 depth is taken into account. For function not declared inline,
4790 recursive inlining happens only when -finline-functions
4791 (included in -O3) is enabled and --param max-inline-recursive-
4792 depth-auto is used. The default value is 450.
4793 min-inline-recursive-probability
4795 Recursive inlining is profitable only for function having deep
4796 recursion in average and can hurt for function having little
4797 recursion depth by increasing the prologue size or complexity
4798 of function body to other optimizers.
4799 When profile feedback is available (see -fprofile-generate) the
4801 actual recursion depth can be guessed from probability that
4802 function will recurse via given call expression. This parame‐
4803 ter limits inlining only to call expression whose probability
4804 exceeds given threshold (in percents). The default value is
4805 10.
4806 inline-call-cost
4808 Specify cost of call instruction relative to simple arithmetics
4809 operations (having cost of 1). Increasing this cost disquali‐
4810 fies inlining of non-leaf functions and at the same time
4811 increases size of leaf function that is believed to reduce
4812 function size by being inlined. In effect it increases amount
4813 of inlining for code having large abstraction penalty (many
4814 functions that just pass the arguments to other functions) and
4815 decrease inlining for code with low abstraction penalty. The
4816 default value is 16.
4817 max-unrolled-insns
4819 The maximum number of instructions that a loop should have if
4820 that loop is unrolled, and if the loop is unrolled, it deter‐
4821 mines how many times the loop code is unrolled.
4822 max-average-unrolled-insns
4824 The maximum number of instructions biased by probabilities of
4825 their execution that a loop should have if that loop is
4826 unrolled, and if the loop is unrolled, it determines how many
4827 times the loop code is unrolled.
4828 max-unroll-times
4830 The maximum number of unrollings of a single loop.
4831 max-peeled-insns
4833 The maximum number of instructions that a loop should have if
4834 that loop is peeled, and if the loop is peeled, it determines
4835 how many times the loop code is peeled.
4836 max-peel-times
4838 The maximum number of peelings of a single loop.
4839 max-completely-peeled-insns
4841 The maximum number of insns of a completely peeled loop.
4842 max-completely-peel-times
4844 The maximum number of iterations of a loop to be suitable for
4845 complete peeling.
4846 max-unswitch-insns
4848 The maximum number of insns of an unswitched loop.
4849 max-unswitch-level
4851 The maximum number of branches unswitched in a single loop.
4852 lim-expensive
4854 The minimum cost of an expensive expression in the loop invari‐
4855 ant motion.
4856 iv-consider-all-candidates-bound
4858 Bound on number of candidates for induction variables below
4859 that all candidates are considered for each use in induction
4860 variable optimizations. Only the most relevant candidates are
4861 considered if there are more candidates, to avoid quadratic
4862 time complexity.
4863 iv-max-considered-uses
4865 The induction variable optimizations give up on loops that con‐
4866 tain more induction variable uses.
4867 iv-always-prune-cand-set-bound
4869 If number of candidates in the set is smaller than this value,
4870 we always try to remove unnecessary ivs from the set during its
4871 optimization when a new iv is added to the set.
4872 scev-max-expr-size
4874 Bound on size of expressions used in the scalar evolutions ana‐
4875 lyzer. Large expressions slow the analyzer.
4876 vect-max-version-checks
4878 The maximum number of runtime checks that can be performed when
4879 doing loop versioning in the vectorizer. See option ftree-
4880 vect-loop-version for more information.
4881 max-iterations-to-track
4883 The maximum number of iterations of a loop the brute force
4884 algorithm for analysis of # of iterations of the loop tries to
4885 evaluate.
4886 hot-bb-count-fraction
4888 Select fraction of the maximal count of repetitions of basic
4889 block in program given basic block needs to have to be consid‐
4890 ered hot.
4891 hot-bb-frequency-fraction
4893 Select fraction of the maximal frequency of executions of basic
4894 block in function given basic block needs to have to be consid‐
4895 ered hot
4896 max-predicted-iterations
4898 The maximum number of loop iterations we predict statically.
4899 This is useful in cases where function contain single loop with
4900 known bound and other loop with unknown. We predict the known
4901 number of iterations correctly, while the unknown number of
4902 iterations average to roughly 10. This means that the loop
4903 without bounds would appear artificially cold relative to the
4904 other one.
4905 tracer-dynamic-coverage
4907 tracer-dynamic-coverage-feedback
4908 This value is used to limit superblock formation once the given
4909 percentage of executed instructions is covered. This limits
4910 unnecessary code size expansion.
4911 The tracer-dynamic-coverage-feedback is used only when profile
4913 feedback is available. The real profiles (as opposed to stati‐
4914 cally estimated ones) are much less balanced allowing the
4915 threshold to be larger value.
4916 tracer-max-code-growth
4918 Stop tail duplication once code growth has reached given per‐
4919 centage. This is rather hokey argument, as most of the dupli‐
4920 cates will be eliminated later in cross jumping, so it may be
4921 set to much higher values than is the desired code growth.
4922 tracer-min-branch-ratio
4924 Stop reverse growth when the reverse probability of best edge
4925 is less than this threshold (in percent).
4926 tracer-min-branch-ratio
4928 tracer-min-branch-ratio-feedback
4929 Stop forward growth if the best edge do have probability lower
4930 than this threshold.
4931 Similarly to tracer-dynamic-coverage two values are present,
4933 one for compilation for profile feedback and one for compila‐
4934 tion without. The value for compilation with profile feedback
4935 needs to be more conservative (higher) in order to make tracer
4936 effective.
4937 max-cse-path-length
4939 Maximum number of basic blocks on path that cse considers. The
4940 default is 10.
4941 max-cse-insns
4943 The maximum instructions CSE process before flushing. The
4944 default is 1000.
4945 global-var-threshold
4947 Counts the number of function calls (n) and the number of call-
4948 clobbered variables (v). If nxv is larger than this limit, a
4949 single artificial variable will be created to represent all the
4950 call-clobbered variables at function call sites. This artifi‐
4951 cial variable will then be made to alias every call-clobbered
4952 variable. (done as "int * size_t" on the host machine; beware
4953 overflow).
4954 max-aliased-vops
4956 Maximum number of virtual operands allowed to represent aliases
4957 before triggering the alias grouping heuristic. Alias grouping
4958 reduces compile times and memory consumption needed for alias‐
4959 ing at the expense of precision loss in alias information.
4960 ggc-min-expand
4962 GCC uses a garbage collector to manage its own memory alloca‐
4963 tion. This parameter specifies the minimum percentage by which
4964 the garbage collector's heap should be allowed to expand
4965 between collections. Tuning this may improve compilation
4966 speed; it has no effect on code generation.
4967 The default is 30% + 70% * (RAM/1GB) with an upper bound of
4969 100% when RAM >= 1GB. If "getrlimit" is available, the notion
4970 of "RAM" is the smallest of actual RAM and "RLIMIT_DATA" or
4971 "RLIMIT_AS". If GCC is not able to calculate RAM on a particu‐
4972 lar platform, the lower bound of 30% is used. Setting this
4973 parameter and ggc-min-heapsize to zero causes a full collection
4974 to occur at every opportunity. This is extremely slow, but can
4975 be useful for debugging.
4976 ggc-min-heapsize
4978 Minimum size of the garbage collector's heap before it begins
4979 bothering to collect garbage. The first collection occurs
4980 after the heap expands by ggc-min-expand% beyond ggc-min-heap‐
4981 size. Again, tuning this may improve compilation speed, and
4982 has no effect on code generation.
4983 The default is the smaller of RAM/8, RLIMIT_RSS, or a limit
4985 which tries to ensure that RLIMIT_DATA or RLIMIT_AS are not
4986 exceeded, but with a lower bound of 4096 (four megabytes) and
4987 an upper bound of 131072 (128 megabytes). If GCC is not able
4988 to calculate RAM on a particular platform, the lower bound is
4989 used. Setting this parameter very large effectively disables
4990 garbage collection. Setting this parameter and ggc-min-expand
4991 to zero causes a full collection to occur at every opportunity.
4992 max-reload-search-insns
4994 The maximum number of instruction reload should look backward
4995 for equivalent register. Increasing values mean more aggres‐
4996 sive optimization, making the compile time increase with proba‐
4997 bly slightly better performance. The default value is 100.
4998 max-cselib-memory-location
5000 The maximum number of memory locations cselib should take into
5001 account. Increasing values mean more aggressive optimization,
5002 making the compile time increase with probably slightly better
5003 performance. The default value is 500.
5004 max-flow-memory-location
5006 Similar as max-cselib-memory-location but for dataflow live‐
5007 ness. The default value is 100.
5008 reorder-blocks-duplicate
5010 reorder-blocks-duplicate-feedback
5011 Used by basic block reordering pass to decide whether to use
5012 unconditional branch or duplicate the code on its destination.
5013 Code is duplicated when its estimated size is smaller than this
5014 value multiplied by the estimated size of unconditional jump in
5015 the hot spots of the program.
5016 The reorder-block-duplicate-feedback is used only when profile
5018 feedback is available and may be set to higher values than
5019 reorder-block-duplicate since information about the hot spots
5020 is more accurate.
5021 max-sched-region-blocks
5023 The maximum number of blocks in a region to be considered for
5024 interblock scheduling. The default value is 10.
5025 max-sched-region-insns
5027 The maximum number of insns in a region to be considered for
5028 interblock scheduling. The default value is 100.
5029 min-sched-prob
5031 The minimum probability of reaching a source block for
5032 interblock speculative scheduling. The default value is 40.
5033 max-last-value-rtl
5035 The maximum size measured as number of RTLs that can be
5036 recorded in an expression in combiner for a pseudo register as
5037 last known value of that register. The default is 10000.
5038 integer-share-limit
5040 Small integer constants can use a shared data structure, reduc‐
5041 ing the compiler's memory usage and increasing its speed. This
5042 sets the maximum value of a shared integer constant's. The
5043 default value is 256.
5044 min-virtual-mappings
5046 Specifies the minimum number of virtual mappings in the incre‐
5047 mental SSA updater that should be registered to trigger the
5048 virtual mappings heuristic defined by virtual-mappings-ratio.
5049 The default value is 100.
5050 virtual-mappings-ratio
5052 If the number of virtual mappings is virtual-mappings-ratio
5053 bigger than the number of virtual symbols to be updated, then
5054 the incremental SSA updater switches to a full update for those
5055 symbols. The default ratio is 3.
5056 ssp-buffer-size
5058 The minimum size of buffers (i.e. arrays) that will receive
5059 stack smashing protection when -fstack-protection is used.
5060 max-jump-thread-duplication-stmts
5062 Maximum number of statements allowed in a block that needs to
5063 be duplicated when threading jumps.
5064 max-fields-for-field-sensitive
5066 Maximum number of fields in a structure we will treat in a
5067 field sensitive manner during pointer analysis.
5068 Options Controlling the Preprocessor
5070 These options control the C preprocessor, which is run on each C source
5072 file before actual compilation.
5073 If you use the -E option, nothing is done except preprocessing. Some
5075 of these options make sense only together with -E because they cause
5076 the preprocessor output to be unsuitable for actual compilation.
5077 You can use -Wp,option to bypass the compiler driver and pass
5079 option directly through to the preprocessor. If option contains
5080 commas, it is split into multiple options at the commas. However,
5081 many options are modified, translated or interpreted by the com‐
5082 piler driver before being passed to the preprocessor, and -Wp
5083 forcibly bypasses this phase. The preprocessor's direct interface
5084 is undocumented and subject to change, so whenever possible you
5085 should avoid using -Wp and let the driver handle the options
5086 instead.
5087 -Xpreprocessor option
5089 Pass option as an option to the preprocessor. You can use this to
5090 supply system-specific preprocessor options which GCC does not know
5091 how to recognize.
5092 If you want to pass an option that takes an argument, you must use
5094 -Xpreprocessor twice, once for the option and once for the argu‐
5095 ment.
5096 -D name
5098 Predefine name as a macro, with definition 1.
5099 -D name=definition
5101 The contents of definition are tokenized and processed as if they
5102 appeared during translation phase three in a #define directive. In
5103 particular, the definition will be truncated by embedded newline
5104 characters.
5105 If you are invoking the preprocessor from a shell or shell-like
5107 program you may need to use the shell's quoting syntax to protect
5108 characters such as spaces that have a meaning in the shell syntax.
5109 If you wish to define a function-like macro on the command line,
5111 write its argument list with surrounding parentheses before the
5112 equals sign (if any). Parentheses are meaningful to most shells,
5113 so you will need to quote the option. With sh and csh,
5114 -D'name(args...)=definition' works.
5115 -D and -U options are processed in the order they are given on the
5117 command line. All -imacros file and -include file options are pro‐
5118 cessed after all -D and -U options.
5119 -U name
5121 Cancel any previous definition of name, either built in or provided
5122 with a -D option.
5123 -undef
5125 Do not predefine any system-specific or GCC-specific macros. The
5126 standard predefined macros remain defined.
5127 -I dir
5129 Add the directory dir to the list of directories to be searched for
5130 header files. Directories named by -I are searched before the
5131 standard system include directories. If the directory dir is a
5132 standard system include directory, the option is ignored to ensure
5133 that the default search order for system directories and the spe‐
5134 cial treatment of system headers are not defeated .
5135 -o file
5137 Write output to file. This is the same as specifying file as the
5138 second non-option argument to cpp. gcc has a different interpreta‐
5139 tion of a second non-option argument, so you must use -o to specify
5140 the output file.
5141 -Wall
5143 Turns on all optional warnings which are desirable for normal code.
5144 At present this is -Wcomment, -Wtrigraphs, -Wmultichar and a warn‐
5145 ing about integer promotion causing a change of sign in "#if"
5146 expressions. Note that many of the preprocessor's warnings are on
5147 by default and have no options to control them.
5148 -Wcomment
5150 -Wcomments
5151 Warn whenever a comment-start sequence /* appears in a /* comment,
5152 or whenever a backslash-newline appears in a // comment. (Both
5153 forms have the same effect.)
5154 -Wtrigraphs
5156 @anchor{Wtrigraphs} Most trigraphs in comments cannot affect the
5157 meaning of the program. However, a trigraph that would form an
5158 escaped newline (??/ at the end of a line) can, by changing where
5159 the comment begins or ends. Therefore, only trigraphs that would
5160 form escaped newlines produce warnings inside a comment.
5161 This option is implied by -Wall. If -Wall is not given, this
5163 option is still enabled unless trigraphs are enabled. To get tri‐
5164 graph conversion without warnings, but get the other -Wall warn‐
5165 ings, use -trigraphs -Wall -Wno-trigraphs.
5166 -Wtraditional
5168 Warn about certain constructs that behave differently in tradi‐
5169 tional and ISO C. Also warn about ISO C constructs that have no
5170 traditional C equivalent, and problematic constructs which should
5171 be avoided.
5172 -Wimport
5174 Warn the first time #import is used.
5175 -Wundef
5177 Warn whenever an identifier which is not a macro is encountered in
5178 an #if directive, outside of defined. Such identifiers are
5179 replaced with zero.
5180 -Wunused-macros
5182 Warn about macros defined in the main file that are unused. A
5183 macro is used if it is expanded or tested for existence at least
5184 once. The preprocessor will also warn if the macro has not been
5185 used at the time it is redefined or undefined.
5186 Built-in macros, macros defined on the command line, and macros
5188 defined in include files are not warned about.
5189 Note: If a macro is actually used, but only used in skipped condi‐
5191 tional blocks, then CPP will report it as unused. To avoid the
5192 warning in such a case, you might improve the scope of the macro's
5193 definition by, for example, moving it into the first skipped block.
5194 Alternatively, you could provide a dummy use with something like:
5195 #if defined the_macro_causing_the_warning
5197 #endif
5198 -Wendif-labels
5200 Warn whenever an #else or an #endif are followed by text. This
5201 usually happens in code of the form
5202 #if FOO
5204 ...
5205 #else FOO
5206 ...
5207 #endif FOO
5208 The second and third "FOO" should be in comments, but often are not
5210 in older programs. This warning is on by default.
5211 -Werror
5213 Make all warnings into hard errors. Source code which triggers
5214 warnings will be rejected.
5215 -Wsystem-headers
5217 Issue warnings for code in system headers. These are normally
5218 unhelpful in finding bugs in your own code, therefore suppressed.
5219 If you are responsible for the system library, you may want to see
5220 them.
5221 -w Suppress all warnings, including those which GNU CPP issues by
5223 default.
5224 -pedantic
5226 Issue all the mandatory diagnostics listed in the C standard. Some
5227 of them are left out by default, since they trigger frequently on
5228 harmless code.
5229 -pedantic-errors
5231 Issue all the mandatory diagnostics, and make all mandatory diag‐
5232 nostics into errors. This includes mandatory diagnostics that GCC
5233 issues without -pedantic but treats as warnings.
5234 -M Instead of outputting the result of preprocessing, output a rule
5236 suitable for make describing the dependencies of the main source
5237 file. The preprocessor outputs one make rule containing the object
5238 file name for that source file, a colon, and the names of all the
5239 included files, including those coming from -include or -imacros
5240 command line options.
5241 Unless specified explicitly (with -MT or -MQ), the object file name
5243 consists of the basename of the source file with any suffix
5244 replaced with object file suffix. If there are many included files
5245 then the rule is split into several lines using \-newline. The
5246 rule has no commands.
5247 This option does not suppress the preprocessor's debug output, such
5249 as -dM. To avoid mixing such debug output with the dependency
5250 rules you should explicitly specify the dependency output file with
5251 -MF, or use an environment variable like DEPENDENCIES_OUTPUT.
5252 Debug output will still be sent to the regular output stream as
5253 normal.
5254 Passing -M to the driver implies -E, and suppresses warnings with
5256 an implicit -w.
5257 -MM Like -M but do not mention header files that are found in system
5259 header directories, nor header files that are included, directly or
5260 indirectly, from such a header.
5261 This implies that the choice of angle brackets or double quotes in
5263 an #include directive does not in itself determine whether that
5264 header will appear in -MM dependency output. This is a slight
5265 change in semantics from GCC versions 3.0 and earlier.
5266 @anchor{dashMF}
5268 -MF file
5270 When used with -M or -MM, specifies a file to write the dependen‐
5271 cies to. If no -MF switch is given the preprocessor sends the
5272 rules to the same place it would have sent preprocessed output.
5273 When used with the driver options -MD or -MMD, -MF overrides the
5275 default dependency output file.
5276 -MG In conjunction with an option such as -M requesting dependency gen‐
5278 eration, -MG assumes missing header files are generated files and
5279 adds them to the dependency list without raising an error. The
5280 dependency filename is taken directly from the "#include" directive
5281 without prepending any path. -MG also suppresses preprocessed out‐
5282 put, as a missing header file renders this useless.
5283 This feature is used in automatic updating of makefiles.
5285 -MP This option instructs CPP to add a phony target for each dependency
5287 other than the main file, causing each to depend on nothing. These
5288 dummy rules work around errors make gives if you remove header
5289 files without updating the Makefile to match.
5290 This is typical output:
5292 test.o: test.c test.h
5294 test.h:
5296 -MT target
5298 Change the target of the rule emitted by dependency generation. By
5299 default CPP takes the name of the main input file, including any
5300 path, deletes any file suffix such as .c, and appends the plat‐
5301 form's usual object suffix. The result is the target.
5302 An -MT option will set the target to be exactly the string you
5304 specify. If you want multiple targets, you can specify them as a
5305 single argument to -MT, or use multiple -MT options.
5306 For example, -MT '$(objpfx)foo.o' might give
5308 $(objpfx)foo.o: foo.c
5310 -MQ target
5312 Same as -MT, but it quotes any characters which are special to
5313 Make. -MQ '$(objpfx)foo.o' gives
5314 $$(objpfx)foo.o: foo.c
5316 The default target is automatically quoted, as if it were given
5318 with -MQ.
5319 -MD -MD is equivalent to -M -MF file, except that -E is not implied.
5321 The driver determines file based on whether an -o option is given.
5322 If it is, the driver uses its argument but with a suffix of .d,
5323 otherwise it take the basename of the input file and applies a .d
5324 suffix.
5325 If -MD is used in conjunction with -E, any -o switch is understood
5327 to specify the dependency output file (but @pxref{dashMF,,-MF}),
5328 but if used without -E, each -o is understood to specify a target
5329 object file.
5330 Since -E is not implied, -MD can be used to generate a dependency
5332 output file as a side-effect of the compilation process.
5333 -MMD
5335 Like -MD except mention only user header files, not system header
5336 files.
5337 -fpch-deps
5339 When using precompiled headers, this flag will cause the depen‐
5340 dency-output flags to also list the files from the precompiled
5341 header's dependencies. If not specified only the precompiled
5342 header would be listed and not the files that were used to create
5343 it because those files are not consulted when a precompiled header
5344 is used.
5345 -fpch-preprocess
5347 This option allows use of a precompiled header together with -E.
5348 It inserts a special "#pragma", "#pragma GCC pch_preprocess "<file‐
5349 name>"" in the output to mark the place where the precompiled
5350 header was found, and its filename. When -fpreprocessed is in use,
5351 GCC recognizes this "#pragma" and loads the PCH.
5352 This option is off by default, because the resulting preprocessed
5354 output is only really suitable as input to GCC. It is switched on
5355 by -save-temps.
5356 You should not write this "#pragma" in your own code, but it is
5358 safe to edit the filename if the PCH file is available in a differ‐
5359 ent location. The filename may be absolute or it may be relative
5360 to GCC's current directory.
5361 -x c
5363 -x c++
5364 -x objective-c
5365 -x assembler-with-cpp
5366 Specify the source language: C, C++, Objective-C, or assembly.
5367 This has nothing to do with standards conformance or extensions; it
5368 merely selects which base syntax to expect. If you give none of
5369 these options, cpp will deduce the language from the extension of
5370 the source file: .c, .cc, .m, or .S. Some other common extensions
5371 for C++ and assembly are also recognized. If cpp does not recog‐
5372 nize the extension, it will treat the file as C; this is the most
5373 generic mode.
5374 Note: Previous versions of cpp accepted a -lang option which
5376 selected both the language and the standards conformance level.
5377 This option has been removed, because it conflicts with the -l
5378 option.
5379 -std=standard
5381 -ansi
5382 Specify the standard to which the code should conform. Currently
5383 CPP knows about C and C++ standards; others may be added in the
5384 future.
5385 standard may be one of:
5387 "iso9899:1990"
5389 "c89"
5390 The ISO C standard from 1990. c89 is the customary shorthand
5391 for this version of the standard.
5392 The -ansi option is equivalent to -std=c89.
5394 "iso9899:199409"
5396 The 1990 C standard, as amended in 1994.
5397 "iso9899:1999"
5399 "c99"
5400 "iso9899:199x"
5401 "c9x"
5402 The revised ISO C standard, published in December 1999. Before
5403 publication, this was known as C9X.
5404 "gnu89"
5406 The 1990 C standard plus GNU extensions. This is the default.
5407 "gnu99"
5409 "gnu9x"
5410 The 1999 C standard plus GNU extensions.
5411 "c++98"
5413 The 1998 ISO C++ standard plus amendments.
5414 "gnu++98"
5416 The same as -std=c++98 plus GNU extensions. This is the
5417 default for C++ code.
5418 -I- Split the include path. Any directories specified with -I options
5420 before -I- are searched only for headers requested with
5421 "#include "file""; they are not searched for "#include <file>". If
5422 additional directories are specified with -I options after the -I-,
5423 those directories are searched for all #include directives.
5424 In addition, -I- inhibits the use of the directory of the current
5426 file directory as the first search directory for "#include "file"".
5427 This option has been deprecated.
5428 -nostdinc
5430 Do not search the standard system directories for header files.
5431 Only the directories you have specified with -I options (and the
5432 directory of the current file, if appropriate) are searched.
5433 -nostdinc++
5435 Do not search for header files in the C++-specific standard direc‐
5436 tories, but do still search the other standard directories. (This
5437 option is used when building the C++ library.)
5438 -include file
5440 Process file as if "#include "file"" appeared as the first line of
5441 the primary source file. However, the first directory searched for
5442 file is the preprocessor's working directory instead of the direc‐
5443 tory containing the main source file. If not found there, it is
5444 searched for in the remainder of the "#include "..."" search chain
5445 as normal.
5446 If multiple -include options are given, the files are included in
5448 the order they appear on the command line.
5449 -imacros file
5451 Exactly like -include, except that any output produced by scanning
5452 file is thrown away. Macros it defines remain defined. This
5453 allows you to acquire all the macros from a header without also
5454 processing its declarations.
5455 All files specified by -imacros are processed before all files
5457 specified by -include.
5458 -idirafter dir
5460 Search dir for header files, but do it after all directories speci‐
5461 fied with -I and the standard system directories have been
5462 exhausted. dir is treated as a system include directory.
5463 -iprefix prefix
5465 Specify prefix as the prefix for subsequent -iwithprefix options.
5466 If the prefix represents a directory, you should include the final
5467 /.
5468 -iwithprefix dir
5470 -iwithprefixbefore dir
5471 Append dir to the prefix specified previously with -iprefix, and
5472 add the resulting directory to the include search path. -iwithpre‐
5473 fixbefore puts it in the same place -I would; -iwithprefix puts it
5474 where -idirafter would.
5475 -isysroot dir
5477 This option is like the --sysroot option, but applies only to
5478 header files. See the --sysroot option for more information.
5479 -isystem dir
5481 Search dir for header files, after all directories specified by -I
5482 but before the standard system directories. Mark it as a system
5483 directory, so that it gets the same special treatment as is applied
5484 to the standard system directories.
5485 -iquote dir
5487 Search dir only for header files requested with "#include "file"";
5488 they are not searched for "#include <file>", before all directories
5489 specified by -I and before the standard system directories.
5490 -fdollars-in-identifiers
5492 @anchor{fdollars-in-identifiers} Accept $ in identifiers.
5493 -fextended-identifiers
5495 Accept universal character names in identifiers. This option is
5496 experimental; in a future version of GCC, it will be enabled by
5497 default for C99 and C++.
5498 -fpreprocessed
5500 Indicate to the preprocessor that the input file has already been
5501 preprocessed. This suppresses things like macro expansion, tri‐
5502 graph conversion, escaped newline splicing, and processing of most
5503 directives. The preprocessor still recognizes and removes com‐
5504 ments, so that you can pass a file preprocessed with -C to the com‐
5505 piler without problems. In this mode the integrated preprocessor
5506 is little more than a tokenizer for the front ends.
5507 -fpreprocessed is implicit if the input file has one of the exten‐
5509 sions .i, .ii or .mi. These are the extensions that GCC uses for
5510 preprocessed files created by -save-temps.
5511 -ftabstop=width
5513 Set the distance between tab stops. This helps the preprocessor
5514 report correct column numbers in warnings or errors, even if tabs
5515 appear on the line. If the value is less than 1 or greater than
5516 100, the option is ignored. The default is 8.
5517 -fexec-charset=charset
5519 Set the execution character set, used for string and character con‐
5520 stants. The default is UTF-8. charset can be any encoding sup‐
5521 ported by the system's "iconv" library routine.
5522 -fwide-exec-charset=charset
5524 Set the wide execution character set, used for wide string and
5525 character constants. The default is UTF-32 or UTF-16, whichever
5526 corresponds to the width of "wchar_t". As with -fexec-charset,
5527 charset can be any encoding supported by the system's "iconv"
5528 library routine; however, you will have problems with encodings
5529 that do not fit exactly in "wchar_t".
5530 -finput-charset=charset
5532 Set the input character set, used for translation from the charac‐
5533 ter set of the input file to the source character set used by GCC.
5534 If the locale does not specify, or GCC cannot get this information
5535 from the locale, the default is UTF-8. This can be overridden by
5536 either the locale or this command line option. Currently the com‐
5537 mand line option takes precedence if there's a conflict. charset
5538 can be any encoding supported by the system's "iconv" library rou‐
5539 tine.
5540 -fworking-directory
5542 Enable generation of linemarkers in the preprocessor output that
5543 will let the compiler know the current working directory at the
5544 time of preprocessing. When this option is enabled, the preproces‐
5545 sor will emit, after the initial linemarker, a second linemarker
5546 with the current working directory followed by two slashes. GCC
5547 will use this directory, when it's present in the preprocessed
5548 input, as the directory emitted as the current working directory in
5549 some debugging information formats. This option is implicitly
5550 enabled if debugging information is enabled, but this can be inhib‐
5551 ited with the negated form -fno-working-directory. If the -P flag
5552 is present in the command line, this option has no effect, since no
5553 "#line" directives are emitted whatsoever.
5554 -fno-show-column
5556 Do not print column numbers in diagnostics. This may be necessary
5557 if diagnostics are being scanned by a program that does not under‐
5558 stand the column numbers, such as dejagnu.
5559 -A predicate=answer
5561 Make an assertion with the predicate predicate and answer answer.
5562 This form is preferred to the older form -A predicate(answer),
5563 which is still supported, because it does not use shell special
5564 characters.
5565 -A -predicate=answer
5567 Cancel an assertion with the predicate predicate and answer answer.
5568 -dCHARS
5570 CHARS is a sequence of one or more of the following characters, and
5571 must not be preceded by a space. Other characters are interpreted
5572 by the compiler proper, or reserved for future versions of GCC, and
5573 so are silently ignored. If you specify characters whose behavior
5574 conflicts, the result is undefined.
5575 M Instead of the normal output, generate a list of #define direc‐
5577 tives for all the macros defined during the execution of the
5578 preprocessor, including predefined macros. This gives you a
5579 way of finding out what is predefined in your version of the
5580 preprocessor. Assuming you have no file foo.h, the command
5581 touch foo.h; cpp -dM foo.h
5583 will show all the predefined macros.
5585 D Like M except in two respects: it does not include the prede‐
5587 fined macros, and it outputs both the #define directives and
5588 the result of preprocessing. Both kinds of output go to the
5589 standard output file.
5590 N Like D, but emit only the macro names, not their expansions.
5592 I Output #include directives in addition to the result of prepro‐
5594 cessing.
5595 -P Inhibit generation of linemarkers in the output from the preproces‐
5597 sor. This might be useful when running the preprocessor on some‐
5598 thing that is not C code, and will be sent to a program which might
5599 be confused by the linemarkers.
5600 -C Do not discard comments. All comments are passed through to the
5602 output file, except for comments in processed directives, which are
5603 deleted along with the directive.
5604 You should be prepared for side effects when using -C; it causes
5606 the preprocessor to treat comments as tokens in their own right.
5607 For example, comments appearing at the start of what would be a
5608 directive line have the effect of turning that line into an ordi‐
5609 nary source line, since the first token on the line is no longer a
5610 #.
5611 -CC Do not discard comments, including during macro expansion. This is
5613 like -C, except that comments contained within macros are also
5614 passed through to the output file where the macro is expanded.
5615 In addition to the side-effects of the -C option, the -CC option
5617 causes all C++-style comments inside a macro to be converted to
5618 C-style comments. This is to prevent later use of that macro from
5619 inadvertently commenting out the remainder of the source line.
5620 The -CC option is generally used to support lint comments.
5622 -traditional-cpp
5624 Try to imitate the behavior of old-fashioned C preprocessors, as
5625 opposed to ISO C preprocessors.
5626 -trigraphs
5628 Process trigraph sequences. These are three-character sequences,
5629 all starting with ??, that are defined by ISO C to stand for single
5630 characters. For example, ??/ stands for \, so '??/n' is a charac‐
5631 ter constant for a newline. By default, GCC ignores trigraphs, but
5632 in standard-conforming modes it converts them. See the -std and
5633 -ansi options.
5634 The nine trigraphs and their replacements are
5636 Trigraph: ??( ??) ??< ??> ??= ??/ ??' ??! ??-
5638 Replacement: [ ] { } # \ ^ ⎪ ~
5639 -remap
5641 Enable special code to work around file systems which only permit
5642 very short file names, such as MS-DOS.
5643 --help
5645 --target-help
5646 Print text describing all the command line options instead of pre‐
5647 processing anything.
5648 -v Verbose mode. Print out GNU CPP's version number at the beginning
5650 of execution, and report the final form of the include path.
5651 -H Print the name of each header file used, in addition to other nor‐
5653 mal activities. Each name is indented to show how deep in the
5654 #include stack it is. Precompiled header files are also printed,
5655 even if they are found to be invalid; an invalid precompiled header
5656 file is printed with ...x and a valid one with ...! .
5657 -version
5659 --version
5660 Print out GNU CPP's version number. With one dash, proceed to pre‐
5661 process as normal. With two dashes, exit immediately.
5662 Passing Options to the Assembler
5664 You can pass options to the assembler.
5666 -Wa,option
5668 Pass option as an option to the assembler. If option contains com‐
5669 mas, it is split into multiple options at the commas.
5670 -Xassembler option
5672 Pass option as an option to the assembler. You can use this to
5673 supply system-specific assembler options which GCC does not know
5674 how to recognize.
5675 If you want to pass an option that takes an argument, you must use
5677 -Xassembler twice, once for the option and once for the argument.
5678 Options for Linking
5680 These options come into play when the compiler links object files into
5682 an executable output file. They are meaningless if the compiler is not
5683 doing a link step.
5684 object-file-name
5686 A file name that does not end in a special recognized suffix is
5687 considered to name an object file or library. (Object files are
5688 distinguished from libraries by the linker according to the file
5689 contents.) If linking is done, these object files are used as
5690 input to the linker.
5691 -c
5693 -S
5694 -E If any of these options is used, then the linker is not run, and
5695 object file names should not be used as arguments.
5696 -llibrary
5698 -l library
5699 Search the library named library when linking. (The second alter‐
5700 native with the library as a separate argument is only for POSIX
5701 compliance and is not recommended.)
5702 It makes a difference where in the command you write this option;
5704 the linker searches and processes libraries and object files in the
5705 order they are specified. Thus, foo.o -lz bar.o searches library z
5706 after file foo.o but before bar.o. If bar.o refers to functions in
5707 z, those functions may not be loaded.
5708 The linker searches a standard list of directories for the library,
5710 which is actually a file named liblibrary.a. The linker then uses
5711 this file as if it had been specified precisely by name.
5712 The directories searched include several standard system directo‐
5714 ries plus any that you specify with -L.
5715 Normally the files found this way are library files---archive files
5717 whose members are object files. The linker handles an archive file
5718 by scanning through it for members which define symbols that have
5719 so far been referenced but not defined. But if the file that is
5720 found is an ordinary object file, it is linked in the usual fash‐
5721 ion. The only difference between using an -l option and specifying
5722 a file name is that -l surrounds library with lib and .a and
5723 searches several directories.
5724 -lobjc
5726 You need this special case of the -l option in order to link an
5727 Objective-C or Objective-C++ program.
5728 -nostartfiles
5730 Do not use the standard system startup files when linking. The
5731 standard system libraries are used normally, unless -nostdlib or
5732 -nodefaultlibs is used.
5733 -nodefaultlibs
5735 Do not use the standard system libraries when linking. Only the
5736 libraries you specify will be passed to the linker. The standard
5737 startup files are used normally, unless -nostartfiles is used. The
5738 compiler may generate calls to "memcmp", "memset", "memcpy" and
5739 "memmove". These entries are usually resolved by entries in libc.
5740 These entry points should be supplied through some other mechanism
5741 when this option is specified.
5742 -nostdlib
5744 Do not use the standard system startup files or libraries when
5745 linking. No startup files and only the libraries you specify will
5746 be passed to the linker. The compiler may generate calls to "mem‐
5747 cmp", "memset", "memcpy" and "memmove". These entries are usually
5748 resolved by entries in libc. These entry points should be supplied
5749 through some other mechanism when this option is specified.
5750 One of the standard libraries bypassed by -nostdlib and -nodefault‐
5752 libs is libgcc.a, a library of internal subroutines that GCC uses
5753 to overcome shortcomings of particular machines, or special needs
5754 for some languages.
5755 In most cases, you need libgcc.a even when you want to avoid other
5757 standard libraries. In other words, when you specify -nostdlib or
5758 -nodefaultlibs you should usually specify -lgcc as well. This
5759 ensures that you have no unresolved references to internal GCC
5760 library subroutines. (For example, __main, used to ensure C++ con‐
5761 structors will be called.)
5762 -pie
5764 Produce a position independent executable on targets which support
5765 it. For predictable results, you must also specify the same set of
5766 options that were used to generate code (-fpie, -fPIE, or model
5767 suboptions) when you specify this option.
5768 -rdynamic
5770 Pass the flag -export-dynamic to the ELF linker, on targets that
5771 support it. This instructs the linker to add all symbols, not only
5772 used ones, to the dynamic symbol table. This option is needed for
5773 some uses of "dlopen" or to allow obtaining backtraces from within
5774 a program.
5775 -s Remove all symbol table and relocation information from the exe‐
5777 cutable.
5778 -static
5780 On systems that support dynamic linking, this prevents linking with
5781 the shared libraries. On other systems, this option has no effect.
5782 -shared
5784 Produce a shared object which can then be linked with other objects
5785 to form an executable. Not all systems support this option. For
5786 predictable results, you must also specify the same set of options
5787 that were used to generate code (-fpic, -fPIC, or model suboptions)
5788 when you specify this option.[1]
5789 -shared-libgcc
5791 -static-libgcc
5792 On systems that provide libgcc as a shared library, these options
5793 force the use of either the shared or static version respectively.
5794 If no shared version of libgcc was built when the compiler was con‐
5795 figured, these options have no effect.
5796 There are several situations in which an application should use the
5798 shared libgcc instead of the static version. The most common of
5799 these is when the application wishes to throw and catch exceptions
5800 across different shared libraries. In that case, each of the
5801 libraries as well as the application itself should use the shared
5802 libgcc.
5803 Therefore, the G++ and GCJ drivers automatically add -shared-libgcc
5805 whenever you build a shared library or a main executable, because
5806 C++ and Java programs typically use exceptions, so this is the
5807 right thing to do.
5808 If, instead, you use the GCC driver to create shared libraries, you
5810 may find that they will not always be linked with the shared
5811 libgcc. If GCC finds, at its configuration time, that you have a
5812 non-GNU linker or a GNU linker that does not support option
5813 --eh-frame-hdr, it will link the shared version of libgcc into
5814 shared libraries by default. Otherwise, it will take advantage of
5815 the linker and optimize away the linking with the shared version of
5816 libgcc, linking with the static version of libgcc by default. This
5817 allows exceptions to propagate through such shared libraries, with‐
5818 out incurring relocation costs at library load time.
5819 However, if a library or main executable is supposed to throw or
5821 catch exceptions, you must link it using the G++ or GCJ driver, as
5822 appropriate for the languages used in the program, or using the
5823 option -shared-libgcc, such that it is linked with the shared
5824 libgcc.
5825 -symbolic
5827 Bind references to global symbols when building a shared object.
5828 Warn about any unresolved references (unless overridden by the link
5829 editor option -Xlinker -z -Xlinker defs). Only a few systems sup‐
5830 port this option.
5831 -Xlinker option
5833 Pass option as an option to the linker. You can use this to supply
5834 system-specific linker options which GCC does not know how to rec‐
5835 ognize.
5836 If you want to pass an option that takes an argument, you must use
5838 -Xlinker twice, once for the option and once for the argument. For
5839 example, to pass -assert definitions, you must write -Xlinker
5840 -assert -Xlinker definitions. It does not work to write -Xlinker
5841 "-assert definitions", because this passes the entire string as a
5842 single argument, which is not what the linker expects.
5843 -Wl,option
5845 Pass option as an option to the linker. If option contains commas,
5846 it is split into multiple options at the commas.
5847 -u symbol
5849 Pretend the symbol symbol is undefined, to force linking of library
5850 modules to define it. You can use -u multiple times with different
5851 symbols to force loading of additional library modules.
5852 Options for Directory Search
5854 These options specify directories to search for header files, for
5856 libraries and for parts of the compiler:
5857 -Idir
5859 Add the directory dir to the head of the list of directories to be
5860 searched for header files. This can be used to override a system
5861 header file, substituting your own version, since these directories
5862 are searched before the system header file directories. However,
5863 you should not use this option to add directories that contain ven‐
5864 dor-supplied system header files (use -isystem for that). If you
5865 use more than one -I option, the directories are scanned in left-
5866 to-right order; the standard system directories come after.
5867 If a standard system include directory, or a directory specified
5869 with -isystem, is also specified with -I, the -I option will be
5870 ignored. The directory will still be searched but as a system
5871 directory at its normal position in the system include chain. This
5872 is to ensure that GCC's procedure to fix buggy system headers and
5873 the ordering for the include_next directive are not inadvertently
5874 changed. If you really need to change the search order for system
5875 directories, use the -nostdinc and/or -isystem options.
5876 -iquotedir
5878 Add the directory dir to the head of the list of directories to be
5879 searched for header files only for the case of #include "file";
5880 they are not searched for #include <file>, otherwise just like -I.
5881 -Ldir
5883 Add directory dir to the list of directories to be searched for -l.
5884 -Bprefix
5886 This option specifies where to find the executables, libraries,
5887 include files, and data files of the compiler itself.
5888 The compiler driver program runs one or more of the subprograms
5890 cpp, cc1, as and ld. It tries prefix as a prefix for each program
5891 it tries to run, both with and without machine/version/.
5892 For each subprogram to be run, the compiler driver first tries the
5894 -B prefix, if any. If that name is not found, or if -B was not
5895 specified, the driver tries two standard prefixes, which are
5896 /usr/lib/gcc/ and /usr/local/lib/gcc/. If neither of those results
5897 in a file name that is found, the unmodified program name is
5898 searched for using the directories specified in your PATH environ‐
5899 ment variable.
5900 The compiler will check to see if the path provided by the -B
5902 refers to a directory, and if necessary it will add a directory
5903 separator character at the end of the path.
5904 -B prefixes that effectively specify directory names also apply to
5906 libraries in the linker, because the compiler translates these
5907 options into -L options for the linker. They also apply to
5908 includes files in the preprocessor, because the compiler translates
5909 these options into -isystem options for the preprocessor. In this
5910 case, the compiler appends include to the prefix.
5911 The run-time support file libgcc.a can also be searched for using
5913 the -B prefix, if needed. If it is not found there, the two stan‐
5914 dard prefixes above are tried, and that is all. The file is left
5915 out of the link if it is not found by those means.
5916 Another way to specify a prefix much like the -B prefix is to use
5918 the environment variable GCC_EXEC_PREFIX.
5919 As a special kludge, if the path provided by -B is [dir/]stageN/,
5921 where N is a number in the range 0 to 9, then it will be replaced
5922 by [dir/]include. This is to help with boot-strapping the com‐
5923 piler.
5924 -specs=file
5926 Process file after the compiler reads in the standard specs file,
5927 in order to override the defaults that the gcc driver program uses
5928 when determining what switches to pass to cc1, cc1plus, as, ld,
5929 etc. More than one -specs=file can be specified on the command
5930 line, and they are processed in order, from left to right.
5931 --sysroot=dir
5933 Use dir as the logical root directory for headers and libraries.
5934 For example, if the compiler would normally search for headers in
5935 /usr/include and libraries in /usr/lib, it will instead search
5936 dir/usr/include and dir/usr/lib.
5937 If you use both this option and the -isysroot option, then the
5939 --sysroot option will apply to libraries, but the -isysroot option
5940 will apply to header files.
5941 The GNU linker (beginning with version 2.16) has the necessary sup‐
5943 port for this option. If your linker does not support this option,
5944 the header file aspect of --sysroot will still work, but the
5945 library aspect will not.
5946 -I- This option has been deprecated. Please use -iquote instead for -I
5948 directories before the -I- and remove the -I-. Any directories you
5949 specify with -I options before the -I- option are searched only for
5950 the case of #include "file"; they are not searched for #include
5951 <file>.
5952 If additional directories are specified with -I options after the
5954 -I-, these directories are searched for all #include directives.
5955 (Ordinarily all -I directories are used this way.)
5956 In addition, the -I- option inhibits the use of the current direc‐
5958 tory (where the current input file came from) as the first search
5959 directory for #include "file". There is no way to override this
5960 effect of -I-. With -I. you can specify searching the directory
5961 which was current when the compiler was invoked. That is not
5962 exactly the same as what the preprocessor does by default, but it
5963 is often satisfactory.
5964 -I- does not inhibit the use of the standard system directories for
5966 header files. Thus, -I- and -nostdinc are independent.
5967 Specifying Target Machine and Compiler Version
5969 The usual way to run GCC is to run the executable called gcc, or
5971 <machine>-gcc when cross-compiling, or <machine>-gcc-<version> to run a
5972 version other than the one that was installed last. Sometimes this is
5973 inconvenient, so GCC provides options that will switch to another
5974 cross-compiler or version.
5975 -b machine
5977 The argument machine specifies the target machine for compilation.
5978 The value to use for machine is the same as was specified as the
5980 machine type when configuring GCC as a cross-compiler. For exam‐
5981 ple, if a cross-compiler was configured with configure arm-elf,
5982 meaning to compile for an arm processor with elf binaries, then you
5983 would specify -b arm-elf to run that cross compiler. Because there
5984 are other options beginning with -b, the configuration must contain
5985 a hyphen.
5986 -V version
5988 The argument version specifies which version of GCC to run. This
5989 is useful when multiple versions are installed. For example, ver‐
5990 sion might be 4.0, meaning to run GCC version 4.0.
5991 The -V and -b options work by running the <machine>-gcc-<version> exe‐
5993 cutable, so there's no real reason to use them if you can just run that
5994 directly.
5995 Hardware Models and Configurations
5997 Earlier we discussed the standard option -b which chooses among differ‐
5999 ent installed compilers for completely different target machines, such
6000 as VAX vs. 68000 vs. 80386.
6001 In addition, each of these target machine types can have its own spe‐
6003 cial options, starting with -m, to choose among various hardware models
6004 or configurations---for example, 68010 vs 68020, floating coprocessor
6005 or none. A single installed version of the compiler can compile for
6006 any model or configuration, according to the options specified.
6007 Some configurations of the compiler also support additional special
6009 options, usually for compatibility with other compilers on the same
6010 platform.
6011 ARC Options
6013 These options are defined for ARC implementations:
6015 -EL Compile code for little endian mode. This is the default.
6017 -EB Compile code for big endian mode.
6019 -mmangle-cpu
6021 Prepend the name of the cpu to all public symbol names. In multi‐
6022 ple-processor systems, there are many ARC variants with different
6023 instruction and register set characteristics. This flag prevents
6024 code compiled for one cpu to be linked with code compiled for
6025 another. No facility exists for handling variants that are "almost
6026 identical". This is an all or nothing option.
6027 -mcpu=cpu
6029 Compile code for ARC variant cpu. Which variants are supported
6030 depend on the configuration. All variants support -mcpu=base, this
6031 is the default.
6032 -mtext=text-section
6034 -mdata=data-section
6035 -mrodata=readonly-data-section
6036 Put functions, data, and readonly data in text-section, data-sec‐
6037 tion, and readonly-data-section respectively by default. This can
6038 be overridden with the "section" attribute.
6039 ARM Options
6041 These -m options are defined for Advanced RISC Machines (ARM) architec‐
6043 tures:
6044 -mabi=name
6046 Generate code for the specified ABI. Permissible values are: apcs-
6047 gnu, atpcs, aapcs, aapcs-linux and iwmmxt.
6048 -mapcs-frame
6050 Generate a stack frame that is compliant with the ARM Procedure
6051 Call Standard for all functions, even if this is not strictly nec‐
6052 essary for correct execution of the code. Specifying
6053 -fomit-frame-pointer with this option will cause the stack frames
6054 not to be generated for leaf functions. The default is
6055 -mno-apcs-frame.
6056 -mapcs
6058 This is a synonym for -mapcs-frame.
6059 -mthumb-interwork
6061 Generate code which supports calling between the ARM and Thumb
6062 instruction sets. Without this option the two instruction sets
6063 cannot be reliably used inside one program. The default is
6064 -mno-thumb-interwork, since slightly larger code is generated when
6065 -mthumb-interwork is specified.
6066 -mno-sched-prolog
6068 Prevent the reordering of instructions in the function prolog, or
6069 the merging of those instruction with the instructions in the func‐
6070 tion's body. This means that all functions will start with a rec‐
6071 ognizable set of instructions (or in fact one of a choice from a
6072 small set of different function prologues), and this information
6073 can be used to locate the start if functions inside an executable
6074 piece of code. The default is -msched-prolog.
6075 -mhard-float
6077 Generate output containing floating point instructions. This is
6078 the default.
6079 -msoft-float
6081 Generate output containing library calls for floating point. Warn‐
6082 ing: the requisite libraries are not available for all ARM targets.
6083 Normally the facilities of the machine's usual C compiler are used,
6084 but this cannot be done directly in cross-compilation. You must
6085 make your own arrangements to provide suitable library functions
6086 for cross-compilation.
6087 -msoft-float changes the calling convention in the output file;
6089 therefore, it is only useful if you compile all of a program with
6090 this option. In particular, you need to compile libgcc.a, the
6091 library that comes with GCC, with -msoft-float in order for this to
6092 work.
6093 -mfloat-abi=name
6095 Specifies which ABI to use for floating point values. Permissible
6096 values are: soft, softfp and hard.
6097 soft and hard are equivalent to -msoft-float and -mhard-float
6099 respectively. softfp allows the generation of floating point
6100 instructions, but still uses the soft-float calling conventions.
6101 -mlittle-endian
6103 Generate code for a processor running in little-endian mode. This
6104 is the default for all standard configurations.
6105 -mbig-endian
6107 Generate code for a processor running in big-endian mode; the
6108 default is to compile code for a little-endian processor.
6109 -mwords-little-endian
6111 This option only applies when generating code for big-endian pro‐
6112 cessors. Generate code for a little-endian word order but a big-
6113 endian byte order. That is, a byte order of the form 32107654.
6114 Note: this option should only be used if you require compatibility
6115 with code for big-endian ARM processors generated by versions of
6116 the compiler prior to 2.8.
6117 -mcpu=name
6119 This specifies the name of the target ARM processor. GCC uses this
6120 name to determine what kind of instructions it can emit when gener‐
6121 ating assembly code. Permissible names are: arm2, arm250, arm3,
6122 arm6, arm60, arm600, arm610, arm620, arm7, arm7m, arm7d, arm7dm,
6123 arm7di, arm7dmi, arm70, arm700, arm700i, arm710, arm710c, arm7100,
6124 arm7500, arm7500fe, arm7tdmi, arm7tdmi-s, arm8, strongarm, stron‐
6125 garm110, strongarm1100, arm8, arm810, arm9, arm9e, arm920, arm920t,
6126 arm922t, arm946e-s, arm966e-s, arm968e-s, arm926ej-s, arm940t,
6127 arm9tdmi, arm10tdmi, arm1020t, arm1026ej-s, arm10e, arm1020e,
6128 arm1022e, arm1136j-s, arm1136jf-s, mpcore, mpcorenovfp,
6129 arm1176jz-s, arm1176jzf-s, xscale, iwmmxt, ep9312.
6130 -mtune=name
6132 This option is very similar to the -mcpu= option, except that
6133 instead of specifying the actual target processor type, and hence
6134 restricting which instructions can be used, it specifies that GCC
6135 should tune the performance of the code as if the target were of
6136 the type specified in this option, but still choosing the instruc‐
6137 tions that it will generate based on the cpu specified by a -mcpu=
6138 option. For some ARM implementations better performance can be
6139 obtained by using this option.
6140 -march=name
6142 This specifies the name of the target ARM architecture. GCC uses
6143 this name to determine what kind of instructions it can emit when
6144 generating assembly code. This option can be used in conjunction
6145 with or instead of the -mcpu= option. Permissible names are:
6146 armv2, armv2a, armv3, armv3m, armv4, armv4t, armv5, armv5t,
6147 armv5te, armv6, armv6j, iwmmxt, ep9312.
6148 -mfpu=name
6150 -mfpe=number
6151 -mfp=number
6152 This specifies what floating point hardware (or hardware emulation)
6153 is available on the target. Permissible names are: fpa, fpe2,
6154 fpe3, maverick, vfp. -mfp and -mfpe are synonyms for -mfpu=fpenum‐
6155 ber, for compatibility with older versions of GCC.
6156 If -msoft-float is specified this specifies the format of floating
6158 point values.
6159 -mstructure-size-boundary=n
6161 The size of all structures and unions will be rounded up to a mul‐
6162 tiple of the number of bits set by this option. Permissible values
6163 are 8, 32 and 64. The default value varies for different
6164 toolchains. For the COFF targeted toolchain the default value is
6165 8. A value of 64 is only allowed if the underlying ABI supports
6166 it.
6167 Specifying the larger number can produce faster, more efficient
6169 code, but can also increase the size of the program. Different
6170 values are potentially incompatible. Code compiled with one value
6171 cannot necessarily expect to work with code or libraries compiled
6172 with another value, if they exchange information using structures
6173 or unions.
6174 -mabort-on-noreturn
6176 Generate a call to the function "abort" at the end of a "noreturn"
6177 function. It will be executed if the function tries to return.
6178 -mlong-calls
6180 -mno-long-calls
6181 Tells the compiler to perform function calls by first loading the
6182 address of the function into a register and then performing a sub‐
6183 routine call on this register. This switch is needed if the target
6184 function will lie outside of the 64 megabyte addressing range of
6185 the offset based version of subroutine call instruction.
6186 Even if this switch is enabled, not all function calls will be
6188 turned into long calls. The heuristic is that static functions,
6189 functions which have the short-call attribute, functions that are
6190 inside the scope of a #pragma no_long_calls directive and functions
6191 whose definitions have already been compiled within the current
6192 compilation unit, will not be turned into long calls. The excep‐
6193 tion to this rule is that weak function definitions, functions with
6194 the long-call attribute or the section attribute, and functions
6195 that are within the scope of a #pragma long_calls directive, will
6196 always be turned into long calls.
6197 This feature is not enabled by default. Specifying -mno-long-calls
6199 will restore the default behavior, as will placing the function
6200 calls within the scope of a #pragma long_calls_off directive. Note
6201 these switches have no effect on how the compiler generates code to
6202 handle function calls via function pointers.
6203 -mnop-fun-dllimport
6205 Disable support for the "dllimport" attribute.
6206 -msingle-pic-base
6208 Treat the register used for PIC addressing as read-only, rather
6209 than loading it in the prologue for each function. The run-time
6210 system is responsible for initializing this register with an appro‐
6211 priate value before execution begins.
6212 -mpic-register=reg
6214 Specify the register to be used for PIC addressing. The default is
6215 R10 unless stack-checking is enabled, when R9 is used.
6216 -mcirrus-fix-invalid-insns
6218 Insert NOPs into the instruction stream to in order to work around
6219 problems with invalid Maverick instruction combinations. This
6220 option is only valid if the -mcpu=ep9312 option has been used to
6221 enable generation of instructions for the Cirrus Maverick floating
6222 point co-processor. This option is not enabled by default, since
6223 the problem is only present in older Maverick implementations. The
6224 default can be re-enabled by use of the -mno-cir‐
6225 rus-fix-invalid-insns switch.
6226 -mpoke-function-name
6228 Write the name of each function into the text section, directly
6229 preceding the function prologue. The generated code is similar to
6230 this:
6231 t0
6233 .ascii "arm_poke_function_name", 0
6234 .align
6235 t1
6236 .word 0xff000000 + (t1 - t0)
6237 arm_poke_function_name
6238 mov ip, sp
6239 stmfd sp!, {fp, ip, lr, pc}
6240 sub fp, ip, #4
6241 When performing a stack backtrace, code can inspect the value of
6243 "pc" stored at "fp + 0". If the trace function then looks at loca‐
6244 tion "pc - 12" and the top 8 bits are set, then we know that there
6245 is a function name embedded immediately preceding this location and
6246 has length "((pc[-3]) & 0xff000000)".
6247 -mthumb
6249 Generate code for the 16-bit Thumb instruction set. The default is
6250 to use the 32-bit ARM instruction set.
6251 -mtpcs-frame
6253 Generate a stack frame that is compliant with the Thumb Procedure
6254 Call Standard for all non-leaf functions. (A leaf function is one
6255 that does not call any other functions.) The default is
6256 -mno-tpcs-frame.
6257 -mtpcs-leaf-frame
6259 Generate a stack frame that is compliant with the Thumb Procedure
6260 Call Standard for all leaf functions. (A leaf function is one that
6261 does not call any other functions.) The default is
6262 -mno-apcs-leaf-frame.
6263 -mcallee-super-interworking
6265 Gives all externally visible functions in the file being compiled
6266 an ARM instruction set header which switches to Thumb mode before
6267 executing the rest of the function. This allows these functions to
6268 be called from non-interworking code.
6269 -mcaller-super-interworking
6271 Allows calls via function pointers (including virtual functions) to
6272 execute correctly regardless of whether the target code has been
6273 compiled for interworking or not. There is a small overhead in the
6274 cost of executing a function pointer if this option is enabled.
6275 -mtp=name
6277 Specify the access model for the thread local storage pointer. The
6278 valid models are soft, which generates calls to "__aeabi_read_tp",
6279 cp15, which fetches the thread pointer from "cp15" directly (sup‐
6280 ported in the arm6k architecture), and auto, which uses the best
6281 available method for the selected processor. The default setting
6282 is auto.
6283 AVR Options
6285 These options are defined for AVR implementations:
6287 -mmcu=mcu
6289 Specify ATMEL AVR instruction set or MCU type.
6290 Instruction set avr1 is for the minimal AVR core, not supported by
6292 the C compiler, only for assembler programs (MCU types: at90s1200,
6293 attiny10, attiny11, attiny12, attiny15, attiny28).
6294 Instruction set avr2 (default) is for the classic AVR core with up
6296 to 8K program memory space (MCU types: at90s2313, at90s2323,
6297 attiny22, at90s2333, at90s2343, at90s4414, at90s4433, at90s4434,
6298 at90s8515, at90c8534, at90s8535).
6299 Instruction set avr3 is for the classic AVR core with up to 128K
6301 program memory space (MCU types: atmega103, atmega603, at43usb320,
6302 at76c711).
6303 Instruction set avr4 is for the enhanced AVR core with up to 8K
6305 program memory space (MCU types: atmega8, atmega83, atmega85).
6306 Instruction set avr5 is for the enhanced AVR core with up to 128K
6308 program memory space (MCU types: atmega16, atmega161, atmega163,
6309 atmega32, atmega323, atmega64, atmega128, at43usb355, at94k).
6310 -msize
6312 Output instruction sizes to the asm file.
6313 -minit-stack=N
6315 Specify the initial stack address, which may be a symbol or numeric
6316 value, __stack is the default.
6317 -mno-interrupts
6319 Generated code is not compatible with hardware interrupts. Code
6320 size will be smaller.
6321 -mcall-prologues
6323 Functions prologues/epilogues expanded as call to appropriate sub‐
6324 routines. Code size will be smaller.
6325 -mno-tablejump
6327 Do not generate tablejump insns which sometimes increase code size.
6328 -mtiny-stack
6330 Change only the low 8 bits of the stack pointer.
6331 -mint8
6333 Assume int to be 8 bit integer. This affects the sizes of all
6334 types: A char will be 1 byte, an int will be 1 byte, an long will
6335 be 2 bytes and long long will be 4 bytes. Please note that this
6336 option does not comply to the C standards, but it will provide you
6337 with smaller code size.
6338 Blackfin Options
6340 -momit-leaf-frame-pointer
6342 Don't keep the frame pointer in a register for leaf functions.
6343 This avoids the instructions to save, set up and restore frame
6344 pointers and makes an extra register available in leaf functions.
6345 The option -fomit-frame-pointer removes the frame pointer for all
6346 functions which might make debugging harder.
6347 -mspecld-anomaly
6349 When enabled, the compiler will ensure that the generated code does
6350 not contain speculative loads after jump instructions. This option
6351 is enabled by default.
6352 -mno-specld-anomaly
6354 Don't generate extra code to prevent speculative loads from occur‐
6355 ring.
6356 -mcsync-anomaly
6358 When enabled, the compiler will ensure that the generated code does
6359 not contain CSYNC or SSYNC instructions too soon after conditional
6360 branches. This option is enabled by default.
6361 -mno-csync-anomaly
6363 Don't generate extra code to prevent CSYNC or SSYNC instructions
6364 from occurring too soon after a conditional branch.
6365 -mlow-64k
6367 When enabled, the compiler is free to take advantage of the knowl‐
6368 edge that the entire program fits into the low 64k of memory.
6369 -mno-low-64k
6371 Assume that the program is arbitrarily large. This is the default.
6372 -mid-shared-library
6374 Generate code that supports shared libraries via the library ID
6375 method. This allows for execute in place and shared libraries in
6376 an environment without virtual memory management. This option
6377 implies -fPIC.
6378 -mno-id-shared-library
6380 Generate code that doesn't assume ID based shared libraries are
6381 being used. This is the default.
6382 -mshared-library-id=n
6384 Specified the identification number of the ID based shared library
6385 being compiled. Specifying a value of 0 will generate more compact
6386 code, specifying other values will force the allocation of that
6387 number to the current library but is no more space or time effi‐
6388 cient than omitting this option.
6389 -mlong-calls
6391 -mno-long-calls
6392 Tells the compiler to perform function calls by first loading the
6393 address of the function into a register and then performing a sub‐
6394 routine call on this register. This switch is needed if the target
6395 function will lie outside of the 24 bit addressing range of the
6396 offset based version of subroutine call instruction.
6397 This feature is not enabled by default. Specifying -mno-long-calls
6399 will restore the default behavior. Note these switches have no
6400 effect on how the compiler generates code to handle function calls
6401 via function pointers.
6402 CRIS Options
6404 These options are defined specifically for the CRIS ports.
6406 -march=architecture-type
6408 -mcpu=architecture-type
6409 Generate code for the specified architecture. The choices for
6410 architecture-type are v3, v8 and v10 for respectively ETRAX 4,
6411 ETRAX 100, and ETRAX 100 LX. Default is v0 except for
6412 cris-axis-linux-gnu, where the default is v10.
6413 -mtune=architecture-type
6415 Tune to architecture-type everything applicable about the generated
6416 code, except for the ABI and the set of available instructions.
6417 The choices for architecture-type are the same as for -march=archi‐
6418 tecture-type.
6419 -mmax-stack-frame=n
6421 Warn when the stack frame of a function exceeds n bytes.
6422 -melinux-stacksize=n
6424 Only available with the cris-axis-aout target. Arranges for indi‐
6425 cations in the program to the kernel loader that the stack of the
6426 program should be set to n bytes.
6427 -metrax4
6429 -metrax100
6430 The options -metrax4 and -metrax100 are synonyms for -march=v3 and
6431 -march=v8 respectively.
6432 -mmul-bug-workaround
6434 -mno-mul-bug-workaround
6435 Work around a bug in the "muls" and "mulu" instructions for CPU
6436 models where it applies. This option is active by default.
6437 -mpdebug
6439 Enable CRIS-specific verbose debug-related information in the
6440 assembly code. This option also has the effect to turn off the
6441 #NO_APP formatted-code indicator to the assembler at the beginning
6442 of the assembly file.
6443 -mcc-init
6445 Do not use condition-code results from previous instruction; always
6446 emit compare and test instructions before use of condition codes.
6447 -mno-side-effects
6449 Do not emit instructions with side-effects in addressing modes
6450 other than post-increment.
6451 -mstack-align
6453 -mno-stack-align
6454 -mdata-align
6455 -mno-data-align
6456 -mconst-align
6457 -mno-const-align
6458 These options (no-options) arranges (eliminate arrangements) for
6459 the stack-frame, individual data and constants to be aligned for
6460 the maximum single data access size for the chosen CPU model. The
6461 default is to arrange for 32-bit alignment. ABI details such as
6462 structure layout are not affected by these options.
6463 -m32-bit
6465 -m16-bit
6466 -m8-bit
6467 Similar to the stack- data- and const-align options above, these
6468 options arrange for stack-frame, writable data and constants to all
6469 be 32-bit, 16-bit or 8-bit aligned. The default is 32-bit align‐
6470 ment.
6471 -mno-prologue-epilogue
6473 -mprologue-epilogue
6474 With -mno-prologue-epilogue, the normal function prologue and epi‐
6475 logue that sets up the stack-frame are omitted and no return
6476 instructions or return sequences are generated in the code. Use
6477 this option only together with visual inspection of the compiled
6478 code: no warnings or errors are generated when call-saved registers
6479 must be saved, or storage for local variable needs to be allocated.
6480 -mno-gotplt
6482 -mgotplt
6483 With -fpic and -fPIC, don't generate (do generate) instruction
6484 sequences that load addresses for functions from the PLT part of
6485 the GOT rather than (traditional on other architectures) calls to
6486 the PLT. The default is -mgotplt.
6487 -maout
6489 Legacy no-op option only recognized with the cris-axis-aout target.
6490 -melf
6492 Legacy no-op option only recognized with the cris-axis-elf and
6493 cris-axis-linux-gnu targets.
6494 -melinux
6496 Only recognized with the cris-axis-aout target, where it selects a
6497 GNU/linux-like multilib, include files and instruction set for
6498 -march=v8.
6499 -mlinux
6501 Legacy no-op option only recognized with the cris-axis-linux-gnu
6502 target.
6503 -sim
6505 This option, recognized for the cris-axis-aout and cris-axis-elf
6506 arranges to link with input-output functions from a simulator
6507 library. Code, initialized data and zero-initialized data are
6508 allocated consecutively.
6509 -sim2
6511 Like -sim, but pass linker options to locate initialized data at
6512 0x40000000 and zero-initialized data at 0x80000000.
6513 CRX Options
6515 These options are defined specifically for the CRX ports.
6517 -mmac
6519 Enable the use of multiply-accumulate instructions. Disabled by
6520 default.
6521 -mpush-args
6523 Push instructions will be used to pass outgoing arguments when
6524 functions are called. Enabled by default.
6525 Darwin Options
6527 These options are defined for all architectures running the Darwin
6529 operating system.
6530 FSF GCC on Darwin does not create "fat" object files; it will create an
6532 object file for the single architecture that it was built to target.
6533 Apple's GCC on Darwin does create "fat" files if multiple -arch options
6534 are used; it does so by running the compiler or linker multiple times
6535 and joining the results together with lipo.
6536 The subtype of the file created (like ppc7400 or ppc970 or i686) is
6538 determined by the flags that specify the ISA that GCC is targetting,
6539 like -mcpu or -march. The -force_cpusubtype_ALL option can be used to
6540 override this.
6541 The Darwin tools vary in their behavior when presented with an ISA mis‐
6543 match. The assembler, as, will only permit instructions to be used
6544 that are valid for the subtype of the file it is generating, so you
6545 cannot put 64-bit instructions in an ppc750 object file. The linker
6546 for shared libraries, /usr/bin/libtool, will fail and print an error if
6547 asked to create a shared library with a less restrictive subtype than
6548 its input files (for instance, trying to put a ppc970 object file in a
6549 ppc7400 library). The linker for executables, ld, will quietly give
6550 the executable the most restrictive subtype of any of its input files.
6551 -Fdir
6553 Add the framework directory dir to the head of the list of directo‐
6554 ries to be searched for header files. These directories are inter‐
6555 leaved with those specified by -I options and are scanned in a
6556 left-to-right order.
6557 A framework directory is a directory with frameworks in it. A
6559 framework is a directory with a "Headers" and/or "PrivateHeaders"
6560 directory contained directly in it that ends in ".framework". The
6561 name of a framework is the name of this directory excluding the
6562 ".framework". Headers associated with the framework are found in
6563 one of those two directories, with "Headers" being searched first.
6564 A subframework is a framework directory that is in a framework's
6565 "Frameworks" directory. Includes of subframework headers can only
6566 appear in a header of a framework that contains the subframework,
6567 or in a sibling subframework header. Two subframeworks are sib‐
6568 lings if they occur in the same framework. A subframework should
6569 not have the same name as a framework, a warning will be issued if
6570 this is violated. Currently a subframework cannot have subframe‐
6571 works, in the future, the mechanism may be extended to support
6572 this. The standard frameworks can be found in "/Sys‐
6573 tem/Library/Frameworks" and "/Library/Frameworks". An example
6574 include looks like "#include <Framework/header.h>", where Framework
6575 denotes the name of the framework and header.h is found in the
6576 "PrivateHeaders" or "Headers" directory.
6577 -gused
6579 Emit debugging information for symbols that are used. For STABS
6580 debugging format, this enables -feliminate-unused-debug-symbols.
6581 This is by default ON.
6582 -gfull
6584 Emit debugging information for all symbols and types.
6585 -mmacosx-version-min=version
6587 The earliest version of MacOS X that this executable will run on is
6588 version. Typical values of version include 10.1, 10.2, and 10.3.9.
6589 The default for this option is to make choices that seem to be most
6591 useful.
6592 -mone-byte-bool
6594 Override the defaults for bool so that sizeof(bool)==1. By default
6595 sizeof(bool) is 4 when compiling for Darwin/PowerPC and 1 when com‐
6596 piling for Darwin/x86, so this option has no effect on x86.
6597 Warning: The -mone-byte-bool switch causes GCC to generate code
6599 that is not binary compatible with code generated without that
6600 switch. Using this switch may require recompiling all other mod‐
6601 ules in a program, including system libraries. Use this switch to
6602 conform to a non-default data model.
6603 -mfix-and-continue
6605 -ffix-and-continue
6606 -findirect-data
6607 Generate code suitable for fast turn around development. Needed to
6608 enable gdb to dynamically load ".o" files into already running pro‐
6609 grams. -findirect-data and -ffix-and-continue are provided for
6610 backwards compatibility.
6611 -all_load
6613 Loads all members of static archive libraries. See man ld(1) for
6614 more information.
6615 -arch_errors_fatal
6617 Cause the errors having to do with files that have the wrong archi‐
6618 tecture to be fatal.
6619 -bind_at_load
6621 Causes the output file to be marked such that the dynamic linker
6622 will bind all undefined references when the file is loaded or
6623 launched.
6624 -bundle
6626 Produce a Mach-o bundle format file. See man ld(1) for more infor‐
6627 mation.
6628 -bundle_loader executable
6630 This option specifies the executable that will be loading the build
6631 output file being linked. See man ld(1) for more information.
6632 -dynamiclib
6634 When passed this option, GCC will produce a dynamic library instead
6635 of an executable when linking, using the Darwin libtool command.
6636 -force_cpusubtype_ALL
6638 This causes GCC's output file to have the ALL subtype, instead of
6639 one controlled by the -mcpu or -march option.
6640 -allowable_client client_name
6642 -client_name
6643 -compatibility_version
6644 -current_version
6645 -dead_strip
6646 -dependency-file
6647 -dylib_file
6648 -dylinker_install_name
6649 -dynamic
6650 -exported_symbols_list
6651 -filelist
6652 -flat_namespace
6653 -force_flat_namespace
6654 -headerpad_max_install_names
6655 -image_base
6656 -init
6657 -install_name
6658 -keep_private_externs
6659 -multi_module
6660 -multiply_defined
6661 -multiply_defined_unused
6662 -noall_load
6663 -no_dead_strip_inits_and_terms
6664 -nofixprebinding
6665 -nomultidefs
6666 -noprebind
6667 -noseglinkedit
6668 -pagezero_size
6669 -prebind
6670 -prebind_all_twolevel_modules
6671 -private_bundle
6672 -read_only_relocs
6673 -sectalign
6674 -sectobjectsymbols
6675 -whyload
6676 -seg1addr
6677 -sectcreate
6678 -sectobjectsymbols
6679 -sectorder
6680 -segaddr
6681 -segs_read_only_addr
6682 -segs_read_write_addr
6683 -seg_addr_table
6684 -seg_addr_table_filename
6685 -seglinkedit
6686 -segprot
6687 -segs_read_only_addr
6688 -segs_read_write_addr
6689 -single_module
6690 -static
6691 -sub_library
6692 -sub_umbrella
6693 -twolevel_namespace
6694 -umbrella
6695 -undefined
6696 -unexported_symbols_list
6697 -weak_reference_mismatches
6698 -whatsloaded
6699 These options are passed to the Darwin linker. The Darwin linker
6700 man page describes them in detail.
6701 DEC Alpha Options
6703 These -m options are defined for the DEC Alpha implementations:
6705 -mno-soft-float
6707 -msoft-float
6708 Use (do not use) the hardware floating-point instructions for
6709 floating-point operations. When -msoft-float is specified, func‐
6710 tions in libgcc.a will be used to perform floating-point opera‐
6711 tions. Unless they are replaced by routines that emulate the
6712 floating-point operations, or compiled in such a way as to call
6713 such emulations routines, these routines will issue floating-point
6714 operations. If you are compiling for an Alpha without floating-
6715 point operations, you must ensure that the library is built so as
6716 not to call them.
6717 Note that Alpha implementations without floating-point operations
6719 are required to have floating-point registers.
6720 -mfp-reg
6722 -mno-fp-regs
6723 Generate code that uses (does not use) the floating-point register
6724 set. -mno-fp-regs implies -msoft-float. If the floating-point
6725 register set is not used, floating point operands are passed in
6726 integer registers as if they were integers and floating-point
6727 results are passed in $0 instead of $f0. This is a non-standard
6728 calling sequence, so any function with a floating-point argument or
6729 return value called by code compiled with -mno-fp-regs must also be
6730 compiled with that option.
6731 A typical use of this option is building a kernel that does not
6733 use, and hence need not save and restore, any floating-point regis‐
6734 ters.
6735 -mieee
6737 The Alpha architecture implements floating-point hardware optimized
6738 for maximum performance. It is mostly compliant with the IEEE
6739 floating point standard. However, for full compliance, software
6740 assistance is required. This option generates code fully IEEE com‐
6741 pliant code except that the inexact-flag is not maintained (see
6742 below). If this option is turned on, the preprocessor macro
6743 "_IEEE_FP" is defined during compilation. The resulting code is
6744 less efficient but is able to correctly support denormalized num‐
6745 bers and exceptional IEEE values such as not-a-number and
6746 plus/minus infinity. Other Alpha compilers call this option
6747 -ieee_with_no_inexact.
6748 -mieee-with-inexact
6750 This is like -mieee except the generated code also maintains the
6751 IEEE inexact-flag. Turning on this option causes the generated
6752 code to implement fully-compliant IEEE math. In addition to
6753 "_IEEE_FP", "_IEEE_FP_EXACT" is defined as a preprocessor macro.
6754 On some Alpha implementations the resulting code may execute sig‐
6755 nificantly slower than the code generated by default. Since there
6756 is very little code that depends on the inexact-flag, you should
6757 normally not specify this option. Other Alpha compilers call this
6758 option -ieee_with_inexact.
6759 -mfp-trap-mode=trap-mode
6761 This option controls what floating-point related traps are enabled.
6762 Other Alpha compilers call this option -fptm trap-mode. The trap
6763 mode can be set to one of four values:
6764 n This is the default (normal) setting. The only traps that are
6766 enabled are the ones that cannot be disabled in software (e.g.,
6767 division by zero trap).
6768 u In addition to the traps enabled by n, underflow traps are
6770 enabled as well.
6771 su Like su, but the instructions are marked to be safe for soft‐
6773 ware completion (see Alpha architecture manual for details).
6774 sui Like su, but inexact traps are enabled as well.
6776 -mfp-rounding-mode=rounding-mode
6778 Selects the IEEE rounding mode. Other Alpha compilers call this
6779 option -fprm rounding-mode. The rounding-mode can be one of:
6780 n Normal IEEE rounding mode. Floating point numbers are rounded
6782 towards the nearest machine number or towards the even machine
6783 number in case of a tie.
6784 m Round towards minus infinity.
6786 c Chopped rounding mode. Floating point numbers are rounded
6788 towards zero.
6789 d Dynamic rounding mode. A field in the floating point control
6791 register (fpcr, see Alpha architecture reference manual) con‐
6792 trols the rounding mode in effect. The C library initializes
6793 this register for rounding towards plus infinity. Thus, unless
6794 your program modifies the fpcr, d corresponds to round towards
6795 plus infinity.
6796 -mtrap-precision=trap-precision
6798 In the Alpha architecture, floating point traps are imprecise.
6799 This means without software assistance it is impossible to recover
6800 from a floating trap and program execution normally needs to be
6801 terminated. GCC can generate code that can assist operating system
6802 trap handlers in determining the exact location that caused a
6803 floating point trap. Depending on the requirements of an applica‐
6804 tion, different levels of precisions can be selected:
6805 p Program precision. This option is the default and means a trap
6807 handler can only identify which program caused a floating point
6808 exception.
6809 f Function precision. The trap handler can determine the func‐
6811 tion that caused a floating point exception.
6812 i Instruction precision. The trap handler can determine the
6814 exact instruction that caused a floating point exception.
6815 Other Alpha compilers provide the equivalent options called
6817 -scope_safe and -resumption_safe.
6818 -mieee-conformant
6820 This option marks the generated code as IEEE conformant. You must
6821 not use this option unless you also specify -mtrap-precision=i and
6822 either -mfp-trap-mode=su or -mfp-trap-mode=sui. Its only effect is
6823 to emit the line .eflag 48 in the function prologue of the gener‐
6824 ated assembly file. Under DEC Unix, this has the effect that IEEE-
6825 conformant math library routines will be linked in.
6826 -mbuild-constants
6828 Normally GCC examines a 32- or 64-bit integer constant to see if it
6829 can construct it from smaller constants in two or three instruc‐
6830 tions. If it cannot, it will output the constant as a literal and
6831 generate code to load it from the data segment at runtime.
6832 Use this option to require GCC to construct all integer constants
6834 using code, even if it takes more instructions (the maximum is
6835 six).
6836 You would typically use this option to build a shared library
6838 dynamic loader. Itself a shared library, it must relocate itself
6839 in memory before it can find the variables and constants in its own
6840 data segment.
6841 -malpha-as
6843 -mgas
6844 Select whether to generate code to be assembled by the vendor-sup‐
6845 plied assembler (-malpha-as) or by the GNU assembler -mgas.
6846 -mbwx
6848 -mno-bwx
6849 -mcix
6850 -mno-cix
6851 -mfix
6852 -mno-fix
6853 -mmax
6854 -mno-max
6855 Indicate whether GCC should generate code to use the optional BWX,
6856 CIX, FIX and MAX instruction sets. The default is to use the
6857 instruction sets supported by the CPU type specified via -mcpu=
6858 option or that of the CPU on which GCC was built if none was speci‐
6859 fied.
6860 -mfloat-vax
6862 -mfloat-ieee
6863 Generate code that uses (does not use) VAX F and G floating point
6864 arithmetic instead of IEEE single and double precision.
6865 -mexplicit-relocs
6867 -mno-explicit-relocs
6868 Older Alpha assemblers provided no way to generate symbol reloca‐
6869 tions except via assembler macros. Use of these macros does not
6870 allow optimal instruction scheduling. GNU binutils as of version
6871 2.12 supports a new syntax that allows the compiler to explicitly
6872 mark which relocations should apply to which instructions. This
6873 option is mostly useful for debugging, as GCC detects the capabili‐
6874 ties of the assembler when it is built and sets the default accord‐
6875 ingly.
6876 -msmall-data
6878 -mlarge-data
6879 When -mexplicit-relocs is in effect, static data is accessed via
6880 gp-relative relocations. When -msmall-data is used, objects 8
6881 bytes long or smaller are placed in a small data area (the ".sdata"
6882 and ".sbss" sections) and are accessed via 16-bit relocations off
6883 of the $gp register. This limits the size of the small data area
6884 to 64KB, but allows the variables to be directly accessed via a
6885 single instruction.
6886 The default is -mlarge-data. With this option the data area is
6888 limited to just below 2GB. Programs that require more than 2GB of
6889 data must use "malloc" or "mmap" to allocate the data in the heap
6890 instead of in the program's data segment.
6891 When generating code for shared libraries, -fpic implies
6893 -msmall-data and -fPIC implies -mlarge-data.
6894 -msmall-text
6896 -mlarge-text
6897 When -msmall-text is used, the compiler assumes that the code of
6898 the entire program (or shared library) fits in 4MB, and is thus
6899 reachable with a branch instruction. When -msmall-data is used,
6900 the compiler can assume that all local symbols share the same $gp
6901 value, and thus reduce the number of instructions required for a
6902 function call from 4 to 1.
6903 The default is -mlarge-text.
6905 -mcpu=cpu_type
6907 Set the instruction set and instruction scheduling parameters for
6908 machine type cpu_type. You can specify either the EV style name or
6909 the corresponding chip number. GCC supports scheduling parameters
6910 for the EV4, EV5 and EV6 family of processors and will choose the
6911 default values for the instruction set from the processor you spec‐
6912 ify. If you do not specify a processor type, GCC will default to
6913 the processor on which the compiler was built.
6914 Supported values for cpu_type are
6916 ev4
6918 ev45
6919 21064
6920 Schedules as an EV4 and has no instruction set extensions.
6921 ev5
6923 21164
6924 Schedules as an EV5 and has no instruction set extensions.
6925 ev56
6927 21164a
6928 Schedules as an EV5 and supports the BWX extension.
6929 pca56
6931 21164pc
6932 21164PC
6933 Schedules as an EV5 and supports the BWX and MAX extensions.
6934 ev6
6936 21264
6937 Schedules as an EV6 and supports the BWX, FIX, and MAX exten‐
6938 sions.
6939 ev67
6941 21264a
6942 Schedules as an EV6 and supports the BWX, CIX, FIX, and MAX
6943 extensions.
6944 -mtune=cpu_type
6946 Set only the instruction scheduling parameters for machine type
6947 cpu_type. The instruction set is not changed.
6948 -mmemory-latency=time
6950 Sets the latency the scheduler should assume for typical memory
6951 references as seen by the application. This number is highly
6952 dependent on the memory access patterns used by the application and
6953 the size of the external cache on the machine.
6954 Valid options for time are
6956 number
6958 A decimal number representing clock cycles.
6959 L1
6961 L2
6962 L3
6963 main
6964 The compiler contains estimates of the number of clock cycles
6965 for "typical" EV4 & EV5 hardware for the Level 1, 2 & 3 caches
6966 (also called Dcache, Scache, and Bcache), as well as to main
6967 memory. Note that L3 is only valid for EV5.
6968 DEC Alpha/VMS Options
6970 These -m options are defined for the DEC Alpha/VMS implementations:
6972 -mvms-return-codes
6974 Return VMS condition codes from main. The default is to return
6975 POSIX style condition (e.g. error) codes.
6976 FRV Options
6978 -mgpr-32
6980 Only use the first 32 general purpose registers.
6981 -mgpr-64
6983 Use all 64 general purpose registers.
6984 -mfpr-32
6986 Use only the first 32 floating point registers.
6987 -mfpr-64
6989 Use all 64 floating point registers
6990 -mhard-float
6992 Use hardware instructions for floating point operations.
6993 -msoft-float
6995 Use library routines for floating point operations.
6996 -malloc-cc
6998 Dynamically allocate condition code registers.
6999 -mfixed-cc
7001 Do not try to dynamically allocate condition code registers, only
7002 use "icc0" and "fcc0".
7003 -mdword
7005 Change ABI to use double word insns.
7006 -mno-dword
7008 Do not use double word instructions.
7009 -mdouble
7011 Use floating point double instructions.
7012 -mno-double
7014 Do not use floating point double instructions.
7015 -mmedia
7017 Use media instructions.
7018 -mno-media
7020 Do not use media instructions.
7021 -mmuladd
7023 Use multiply and add/subtract instructions.
7024 -mno-muladd
7026 Do not use multiply and add/subtract instructions.
7027 -mfdpic
7029 Select the FDPIC ABI, that uses function descriptors to represent
7030 pointers to functions. Without any PIC/PIE-related options, it
7031 implies -fPIE. With -fpic or -fpie, it assumes GOT entries and
7032 small data are within a 12-bit range from the GOT base address;
7033 with -fPIC or -fPIE, GOT offsets are computed with 32 bits.
7034 -minline-plt
7036 Enable inlining of PLT entries in function calls to functions that
7037 are not known to bind locally. It has no effect without -mfdpic.
7038 It's enabled by default if optimizing for speed and compiling for
7039 shared libraries (i.e., -fPIC or -fpic), or when an optimization
7040 option such as -O3 or above is present in the command line.
7041 -mTLS
7043 Assume a large TLS segment when generating thread-local code.
7044 -mtls
7046 Do not assume a large TLS segment when generating thread-local
7047 code.
7048 -mgprel-ro
7050 Enable the use of "GPREL" relocations in the FDPIC ABI for data
7051 that is known to be in read-only sections. It's enabled by
7052 default, except for -fpic or -fpie: even though it may help make
7053 the global offset table smaller, it trades 1 instruction for 4.
7054 With -fPIC or -fPIE, it trades 3 instructions for 4, one of which
7055 may be shared by multiple symbols, and it avoids the need for a GOT
7056 entry for the referenced symbol, so it's more likely to be a win.
7057 If it is not, -mno-gprel-ro can be used to disable it.
7058 -multilib-library-pic
7060 Link with the (library, not FD) pic libraries. It's implied by
7061 -mlibrary-pic, as well as by -fPIC and -fpic without -mfdpic. You
7062 should never have to use it explicitly.
7063 -mlinked-fp
7065 Follow the EABI requirement of always creating a frame pointer
7066 whenever a stack frame is allocated. This option is enabled by
7067 default and can be disabled with -mno-linked-fp.
7068 -mlong-calls
7070 Use indirect addressing to call functions outside the current com‐
7071 pilation unit. This allows the functions to be placed anywhere
7072 within the 32-bit address space.
7073 -malign-labels
7075 Try to align labels to an 8-byte boundary by inserting nops into
7076 the previous packet. This option only has an effect when VLIW
7077 packing is enabled. It doesn't create new packets; it merely adds
7078 nops to existing ones.
7079 -mlibrary-pic
7081 Generate position-independent EABI code.
7082 -macc-4
7084 Use only the first four media accumulator registers.
7085 -macc-8
7087 Use all eight media accumulator registers.
7088 -mpack
7090 Pack VLIW instructions.
7091 -mno-pack
7093 Do not pack VLIW instructions.
7094 -mno-eflags
7096 Do not mark ABI switches in e_flags.
7097 -mcond-move
7099 Enable the use of conditional-move instructions (default).
7100 This switch is mainly for debugging the compiler and will likely be
7102 removed in a future version.
7103 -mno-cond-move
7105 Disable the use of conditional-move instructions.
7106 This switch is mainly for debugging the compiler and will likely be
7108 removed in a future version.
7109 -mscc
7111 Enable the use of conditional set instructions (default).
7112 This switch is mainly for debugging the compiler and will likely be
7114 removed in a future version.
7115 -mno-scc
7117 Disable the use of conditional set instructions.
7118 This switch is mainly for debugging the compiler and will likely be
7120 removed in a future version.
7121 -mcond-exec
7123 Enable the use of conditional execution (default).
7124 This switch is mainly for debugging the compiler and will likely be
7126 removed in a future version.
7127 -mno-cond-exec
7129 Disable the use of conditional execution.
7130 This switch is mainly for debugging the compiler and will likely be
7132 removed in a future version.
7133 -mvliw-branch
7135 Run a pass to pack branches into VLIW instructions (default).
7136 This switch is mainly for debugging the compiler and will likely be
7138 removed in a future version.
7139 -mno-vliw-branch
7141 Do not run a pass to pack branches into VLIW instructions.
7142 This switch is mainly for debugging the compiler and will likely be
7144 removed in a future version.
7145 -mmulti-cond-exec
7147 Enable optimization of "&&" and "⎪⎪" in conditional execution
7148 (default).
7149 This switch is mainly for debugging the compiler and will likely be
7151 removed in a future version.
7152 -mno-multi-cond-exec
7154 Disable optimization of "&&" and "⎪⎪" in conditional execution.
7155 This switch is mainly for debugging the compiler and will likely be
7157 removed in a future version.
7158 -mnested-cond-exec
7160 Enable nested conditional execution optimizations (default).
7161 This switch is mainly for debugging the compiler and will likely be
7163 removed in a future version.
7164 -mno-nested-cond-exec
7166 Disable nested conditional execution optimizations.
7167 This switch is mainly for debugging the compiler and will likely be
7169 removed in a future version.
7170 -moptimize-membar
7172 This switch removes redundant "membar" instructions from the com‐
7173 piler generated code. It is enabled by default.
7174 -mno-optimize-membar
7176 This switch disables the automatic removal of redundant "membar"
7177 instructions from the generated code.
7178 -mtomcat-stats
7180 Cause gas to print out tomcat statistics.
7181 -mcpu=cpu
7183 Select the processor type for which to generate code. Possible
7184 values are frv, fr550, tomcat, fr500, fr450, fr405, fr400, fr300
7185 and simple.
7186 H8/300 Options
7188 These -m options are defined for the H8/300 implementations:
7190 -mrelax
7192 Shorten some address references at link time, when possible; uses
7193 the linker option -relax.
7194 -mh Generate code for the H8/300H.
7196 -ms Generate code for the H8S.
7198 -mn Generate code for the H8S and H8/300H in the normal mode. This
7200 switch must be used either with -mh or -ms.
7201 -ms2600
7203 Generate code for the H8S/2600. This switch must be used with -ms.
7204 -mint32
7206 Make "int" data 32 bits by default.
7207 -malign-300
7209 On the H8/300H and H8S, use the same alignment rules as for the
7210 H8/300. The default for the H8/300H and H8S is to align longs and
7211 floats on 4 byte boundaries. -malign-300 causes them to be aligned
7212 on 2 byte boundaries. This option has no effect on the H8/300.
7213 HPPA Options
7215 These -m options are defined for the HPPA family of computers:
7217 -march=architecture-type
7219 Generate code for the specified architecture. The choices for
7220 architecture-type are 1.0 for PA 1.0, 1.1 for PA 1.1, and 2.0 for
7221 PA 2.0 processors. Refer to /usr/lib/sched.models on an HP-UX sys‐
7222 tem to determine the proper architecture option for your machine.
7223 Code compiled for lower numbered architectures will run on higher
7224 numbered architectures, but not the other way around.
7225 -mpa-risc-1-0
7227 -mpa-risc-1-1
7228 -mpa-risc-2-0
7229 Synonyms for -march=1.0, -march=1.1, and -march=2.0 respectively.
7230 -mbig-switch
7232 Generate code suitable for big switch tables. Use this option only
7233 if the assembler/linker complain about out of range branches within
7234 a switch table.
7235 -mjump-in-delay
7237 Fill delay slots of function calls with unconditional jump instruc‐
7238 tions by modifying the return pointer for the function call to be
7239 the target of the conditional jump.
7240 -mdisable-fpregs
7242 Prevent floating point registers from being used in any manner.
7243 This is necessary for compiling kernels which perform lazy context
7244 switching of floating point registers. If you use this option and
7245 attempt to perform floating point operations, the compiler will
7246 abort.
7247 -mdisable-indexing
7249 Prevent the compiler from using indexing address modes. This
7250 avoids some rather obscure problems when compiling MIG generated
7251 code under MACH.
7252 -mno-space-regs
7254 Generate code that assumes the target has no space registers. This
7255 allows GCC to generate faster indirect calls and use unscaled index
7256 address modes.
7257 Such code is suitable for level 0 PA systems and kernels.
7259 -mfast-indirect-calls
7261 Generate code that assumes calls never cross space boundaries.
7262 This allows GCC to emit code which performs faster indirect calls.
7263 This option will not work in the presence of shared libraries or
7265 nested functions.
7266 -mfixed-range=register-range
7268 Generate code treating the given register range as fixed registers.
7269 A fixed register is one that the register allocator can not use.
7270 This is useful when compiling kernel code. A register range is
7271 specified as two registers separated by a dash. Multiple register
7272 ranges can be specified separated by a comma.
7273 -mlong-load-store
7275 Generate 3-instruction load and store sequences as sometimes
7276 required by the HP-UX 10 linker. This is equivalent to the +k
7277 option to the HP compilers.
7278 -mportable-runtime
7280 Use the portable calling conventions proposed by HP for ELF sys‐
7281 tems.
7282 -mgas
7284 Enable the use of assembler directives only GAS understands.
7285 -mschedule=cpu-type
7287 Schedule code according to the constraints for the machine type
7288 cpu-type. The choices for cpu-type are 700 7100, 7100LC, 7200,
7289 7300 and 8000. Refer to /usr/lib/sched.models on an HP-UX system
7290 to determine the proper scheduling option for your machine. The
7291 default scheduling is 8000.
7292 -mlinker-opt
7294 Enable the optimization pass in the HP-UX linker. Note this makes
7295 symbolic debugging impossible. It also triggers a bug in the HP-UX
7296 8 and HP-UX 9 linkers in which they give bogus error messages when
7297 linking some programs.
7298 -msoft-float
7300 Generate output containing library calls for floating point. Warn‐
7301 ing: the requisite libraries are not available for all HPPA tar‐
7302 gets. Normally the facilities of the machine's usual C compiler
7303 are used, but this cannot be done directly in cross-compilation.
7304 You must make your own arrangements to provide suitable library
7305 functions for cross-compilation. The embedded target hppa1.1-*-pro
7306 does provide software floating point support.
7307 -msoft-float changes the calling convention in the output file;
7309 therefore, it is only useful if you compile all of a program with
7310 this option. In particular, you need to compile libgcc.a, the
7311 library that comes with GCC, with -msoft-float in order for this to
7312 work.
7313 -msio
7315 Generate the predefine, "_SIO", for server IO. The default is
7316 -mwsio. This generates the predefines, "__hp9000s700",
7317 "__hp9000s700__" and "_WSIO", for workstation IO. These options
7318 are available under HP-UX and HI-UX.
7319 -mgnu-ld
7321 Use GNU ld specific options. This passes -shared to ld when build‐
7322 ing a shared library. It is the default when GCC is configured,
7323 explicitly or implicitly, with the GNU linker. This option does
7324 not have any affect on which ld is called, it only changes what
7325 parameters are passed to that ld. The ld that is called is deter‐
7326 mined by the --with-ld configure option, GCC's program search path,
7327 and finally by the user's PATH. The linker used by GCC can be
7328 printed using which `gcc -print-prog-name=ld`. This option is only
7329 available on the 64 bit HP-UX GCC, i.e. configured with
7330 hppa*64*-*-hpux*.
7331 -mhp-ld
7333 Use HP ld specific options. This passes -b to ld when building a
7334 shared library and passes +Accept TypeMismatch to ld on all links.
7335 It is the default when GCC is configured, explicitly or implicitly,
7336 with the HP linker. This option does not have any affect on which
7337 ld is called, it only changes what parameters are passed to that
7338 ld. The ld that is called is determined by the --with-ld configure
7339 option, GCC's program search path, and finally by the user's PATH.
7340 The linker used by GCC can be printed using which `gcc
7341 -print-prog-name=ld`. This option is only available on the 64 bit
7342 HP-UX GCC, i.e. configured with hppa*64*-*-hpux*.
7343 -mlong-calls
7345 Generate code that uses long call sequences. This ensures that a
7346 call is always able to reach linker generated stubs. The default
7347 is to generate long calls only when the distance from the call site
7348 to the beginning of the function or translation unit, as the case
7349 may be, exceeds a predefined limit set by the branch type being
7350 used. The limits for normal calls are 7,600,000 and 240,000 bytes,
7351 respectively for the PA 2.0 and PA 1.X architectures. Sibcalls are
7352 always limited at 240,000 bytes.
7353 Distances are measured from the beginning of functions when using
7355 the -ffunction-sections option, or when using the -mgas and
7356 -mno-portable-runtime options together under HP-UX with the SOM
7357 linker.
7358 It is normally not desirable to use this option as it will degrade
7360 performance. However, it may be useful in large applications, par‐
7361 ticularly when partial linking is used to build the application.
7362 The types of long calls used depends on the capabilities of the
7364 assembler and linker, and the type of code being generated. The
7365 impact on systems that support long absolute calls, and long pic
7366 symbol-difference or pc-relative calls should be relatively small.
7367 However, an indirect call is used on 32-bit ELF systems in pic code
7368 and it is quite long.
7369 -munix=unix-std
7371 Generate compiler predefines and select a startfile for the speci‐
7372 fied UNIX standard. The choices for unix-std are 93, 95 and 98.
7373 93 is supported on all HP-UX versions. 95 is available on HP-UX
7374 10.10 and later. 98 is available on HP-UX 11.11 and later. The
7375 default values are 93 for HP-UX 10.00, 95 for HP-UX 10.10 though to
7376 11.00, and 98 for HP-UX 11.11 and later.
7377 -munix=93 provides the same predefines as GCC 3.3 and 3.4.
7379 -munix=95 provides additional predefines for "XOPEN_UNIX" and
7380 "_XOPEN_SOURCE_EXTENDED", and the startfile unix95.o. -munix=98
7381 provides additional predefines for "_XOPEN_UNIX",
7382 "_XOPEN_SOURCE_EXTENDED", "_INCLUDE__STDC_A1_SOURCE" and
7383 "_INCLUDE_XOPEN_SOURCE_500", and the startfile unix98.o.
7384 It is important to note that this option changes the interfaces for
7386 various library routines. It also affects the operational behavior
7387 of the C library. Thus, extreme care is needed in using this
7388 option.
7389 Library code that is intended to operate with more than one UNIX
7391 standard must test, set and restore the variable
7392 __xpg4_extended_mask as appropriate. Most GNU software doesn't
7393 provide this capability.
7394 -nolibdld
7396 Suppress the generation of link options to search libdld.sl when
7397 the -static option is specified on HP-UX 10 and later.
7398 -static
7400 The HP-UX implementation of setlocale in libc has a dependency on
7401 libdld.sl. There isn't an archive version of libdld.sl. Thus,
7402 when the -static option is specified, special link options are
7403 needed to resolve this dependency.
7404 On HP-UX 10 and later, the GCC driver adds the necessary options to
7406 link with libdld.sl when the -static option is specified. This
7407 causes the resulting binary to be dynamic. On the 64-bit port, the
7408 linkers generate dynamic binaries by default in any case. The
7409 -nolibdld option can be used to prevent the GCC driver from adding
7410 these link options.
7411 -threads
7413 Add support for multithreading with the dce thread library under
7414 HP-UX. This option sets flags for both the preprocessor and
7415 linker.
7416 Intel 386 and AMD x86-64 Options
7418 These -m options are defined for the i386 and x86-64 family of comput‐
7420 ers:
7421 -mtune=cpu-type
7423 Tune to cpu-type everything applicable about the generated code,
7424 except for the ABI and the set of available instructions. The
7425 choices for cpu-type are:
7426 i386
7428 Original Intel's i386 CPU.
7429 i486
7431 Intel's i486 CPU. (No scheduling is implemented for this
7432 chip.)
7433 i586, pentium
7435 Intel Pentium CPU with no MMX support.
7436 pentium-mmx
7438 Intel PentiumMMX CPU based on Pentium core with MMX instruction
7439 set support.
7440 i686, pentiumpro
7442 Intel PentiumPro CPU.
7443 pentium2
7445 Intel Pentium2 CPU based on PentiumPro core with MMX instruc‐
7446 tion set support.
7447 pentium3, pentium3m
7449 Intel Pentium3 CPU based on PentiumPro core with MMX and SSE
7450 instruction set support.
7451 pentium-m
7453 Low power version of Intel Pentium3 CPU with MMX, SSE and SSE2
7454 instruction set support. Used by Centrino notebooks.
7455 pentium4, pentium4m
7457 Intel Pentium4 CPU with MMX, SSE and SSE2 instruction set sup‐
7458 port.
7459 prescott
7461 Improved version of Intel Pentium4 CPU with MMX, SSE, SSE2 and
7462 SSE3 instruction set support.
7463 nocona
7465 Improved version of Intel Pentium4 CPU with 64-bit extensions,
7466 MMX, SSE, SSE2 and SSE3 instruction set support.
7467 k6 AMD K6 CPU with MMX instruction set support.
7469 k6-2, k6-3
7471 Improved versions of AMD K6 CPU with MMX and 3dNOW! instruction
7472 set support.
7473 athlon, athlon-tbird
7475 AMD Athlon CPU with MMX, 3dNOW!, enhanced 3dNOW! and SSE
7476 prefetch instructions support.
7477 athlon-4, athlon-xp, athlon-mp
7479 Improved AMD Athlon CPU with MMX, 3dNOW!, enhanced 3dNOW! and
7480 full SSE instruction set support.
7481 k8, opteron, athlon64, athlon-fx
7483 AMD K8 core based CPUs with x86-64 instruction set support.
7484 (This supersets MMX, SSE, SSE2, 3dNOW!, enhanced 3dNOW! and
7485 64-bit instruction set extensions.)
7486 winchip-c6
7488 IDT Winchip C6 CPU, dealt in same way as i486 with additional
7489 MMX instruction set support.
7490 winchip2
7492 IDT Winchip2 CPU, dealt in same way as i486 with additional MMX
7493 and 3dNOW! instruction set support.
7494 c3 Via C3 CPU with MMX and 3dNOW! instruction set support. (No
7496 scheduling is implemented for this chip.)
7497 c3-2
7499 Via C3-2 CPU with MMX and SSE instruction set support. (No
7500 scheduling is implemented for this chip.)
7501 While picking a specific cpu-type will schedule things appropri‐
7503 ately for that particular chip, the compiler will not generate any
7504 code that does not run on the i386 without the -march=cpu-type
7505 option being used.
7506 -march=cpu-type
7508 Generate instructions for the machine type cpu-type. The choices
7509 for cpu-type are the same as for -mtune. Moreover, specifying
7510 -march=cpu-type implies -mtune=cpu-type.
7511 -mcpu=cpu-type
7513 A deprecated synonym for -mtune.
7514 -m386
7516 -m486
7517 -mpentium
7518 -mpentiumpro
7519 These options are synonyms for -mtune=i386, -mtune=i486,
7520 -mtune=pentium, and -mtune=pentiumpro respectively. These synonyms
7521 are deprecated.
7522 -mfpmath=unit
7524 Generate floating point arithmetics for selected unit unit. The
7525 choices for unit are:
7526 387 Use the standard 387 floating point coprocessor present major‐
7528 ity of chips and emulated otherwise. Code compiled with this
7529 option will run almost everywhere. The temporary results are
7530 computed in 80bit precision instead of precision specified by
7531 the type resulting in slightly different results compared to
7532 most of other chips. See -ffloat-store for more detailed
7533 description.
7534 This is the default choice for i386 compiler.
7536 sse Use scalar floating point instructions present in the SSE
7538 instruction set. This instruction set is supported by Pentium3
7539 and newer chips, in the AMD line by Athlon-4, Athlon-xp and
7540 Athlon-mp chips. The earlier version of SSE instruction set
7541 supports only single precision arithmetics, thus the double and
7542 extended precision arithmetics is still done using 387. Later
7543 version, present only in Pentium4 and the future AMD x86-64
7544 chips supports double precision arithmetics too.
7545 For the i386 compiler, you need to use -march=cpu-type, -msse
7547 or -msse2 switches to enable SSE extensions and make this
7548 option effective. For the x86-64 compiler, these extensions
7549 are enabled by default.
7550 The resulting code should be considerably faster in the major‐
7552 ity of cases and avoid the numerical instability problems of
7553 387 code, but may break some existing code that expects tempo‐
7554 raries to be 80bit.
7555 This is the default choice for the x86-64 compiler.
7557 sse,387
7559 Attempt to utilize both instruction sets at once. This effec‐
7560 tively double the amount of available registers and on chips
7561 with separate execution units for 387 and SSE the execution
7562 resources too. Use this option with care, as it is still
7563 experimental, because the GCC register allocator does not model
7564 separate functional units well resulting in instable perfor‐
7565 mance.
7566 -masm=dialect
7568 Output asm instructions using selected dialect. Supported choices
7569 are intel or att (the default one). Darwin does not support intel.
7570 -mieee-fp
7572 -mno-ieee-fp
7573 Control whether or not the compiler uses IEEE floating point com‐
7574 parisons. These handle correctly the case where the result of a
7575 comparison is unordered.
7576 -msoft-float
7578 Generate output containing library calls for floating point. Warn‐
7579 ing: the requisite libraries are not part of GCC. Normally the
7580 facilities of the machine's usual C compiler are used, but this
7581 can't be done directly in cross-compilation. You must make your
7582 own arrangements to provide suitable library functions for
7583 cross-compilation.
7584 On machines where a function returns floating point results in the
7586 80387 register stack, some floating point opcodes may be emitted
7587 even if -msoft-float is used.
7588 -mno-fp-ret-in-387
7590 Do not use the FPU registers for return values of functions.
7591 The usual calling convention has functions return values of types
7593 "float" and "double" in an FPU register, even if there is no FPU.
7594 The idea is that the operating system should emulate an FPU.
7595 The option -mno-fp-ret-in-387 causes such values to be returned in
7597 ordinary CPU registers instead.
7598 -mno-fancy-math-387
7600 Some 387 emulators do not support the "sin", "cos" and "sqrt"
7601 instructions for the 387. Specify this option to avoid generating
7602 those instructions. This option is the default on FreeBSD, OpenBSD
7603 and NetBSD. This option is overridden when -march indicates that
7604 the target cpu will always have an FPU and so the instruction will
7605 not need emulation. As of revision 2.6.1, these instructions are
7606 not generated unless you also use the -funsafe-math-optimizations
7607 switch.
7608 -malign-double
7610 -mno-align-double
7611 Control whether GCC aligns "double", "long double", and "long long"
7612 variables on a two word boundary or a one word boundary. Aligning
7613 "double" variables on a two word boundary will produce code that
7614 runs somewhat faster on a Pentium at the expense of more memory.
7615 On x86-64, -malign-double is enabled by default.
7617 Warning: if you use the -malign-double switch, structures contain‐
7619 ing the above types will be aligned differently than the published
7620 application binary interface specifications for the 386 and will
7621 not be binary compatible with structures in code compiled without
7622 that switch.
7623 -m96bit-long-double
7625 -m128bit-long-double
7626 These switches control the size of "long double" type. The i386
7627 application binary interface specifies the size to be 96 bits, so
7628 -m96bit-long-double is the default in 32 bit mode.
7629 Modern architectures (Pentium and newer) would prefer "long double"
7631 to be aligned to an 8 or 16 byte boundary. In arrays or structures
7632 conforming to the ABI, this would not be possible. So specifying a
7633 -m128bit-long-double will align "long double" to a 16 byte boundary
7634 by padding the "long double" with an additional 32 bit zero.
7635 In the x86-64 compiler, -m128bit-long-double is the default choice
7637 as its ABI specifies that "long double" is to be aligned on 16 byte
7638 boundary.
7639 Notice that neither of these options enable any extra precision
7641 over the x87 standard of 80 bits for a "long double".
7642 Warning: if you override the default value for your target ABI, the
7644 structures and arrays containing "long double" variables will
7645 change their size as well as function calling convention for func‐
7646 tion taking "long double" will be modified. Hence they will not be
7647 binary compatible with arrays or structures in code compiled with‐
7648 out that switch.
7649 -mmlarge-data-threshold=number
7651 When -mcmodel=medium is specified, the data greater than threshold
7652 are placed in large data section. This value must be the same
7653 across all object linked into the binary and defaults to 65535.
7654 -msvr3-shlib
7656 -mno-svr3-shlib
7657 Control whether GCC places uninitialized local variables into the
7658 "bss" or "data" segments. -msvr3-shlib places them into "bss".
7659 These options are meaningful only on System V Release 3.
7660 -mrtd
7662 Use a different function-calling convention, in which functions
7663 that take a fixed number of arguments return with the "ret" num
7664 instruction, which pops their arguments while returning. This
7665 saves one instruction in the caller since there is no need to pop
7666 the arguments there.
7667 You can specify that an individual function is called with this
7669 calling sequence with the function attribute stdcall. You can also
7670 override the -mrtd option by using the function attribute cdecl.
7671 Warning: this calling convention is incompatible with the one nor‐
7673 mally used on Unix, so you cannot use it if you need to call
7674 libraries compiled with the Unix compiler.
7675 Also, you must provide function prototypes for all functions that
7677 take variable numbers of arguments (including "printf"); otherwise
7678 incorrect code will be generated for calls to those functions.
7679 In addition, seriously incorrect code will result if you call a
7681 function with too many arguments. (Normally, extra arguments are
7682 harmlessly ignored.)
7683 -mregparm=num
7685 Control how many registers are used to pass integer arguments. By
7686 default, no registers are used to pass arguments, and at most 3
7687 registers can be used. You can control this behavior for a spe‐
7688 cific function by using the function attribute regparm.
7689 Warning: if you use this switch, and num is nonzero, then you must
7691 build all modules with the same value, including any libraries.
7692 This includes the system libraries and startup modules.
7693 -msseregparm
7695 Use SSE register passing conventions for float and double arguments
7696 and return values. You can control this behavior for a specific
7697 function by using the function attribute sseregparm.
7698 Warning: if you use this switch then you must build all modules
7700 with the same value, including any libraries. This includes the
7701 system libraries and startup modules.
7702 -mpreferred-stack-boundary=num
7704 Attempt to keep the stack boundary aligned to a 2 raised to num
7705 byte boundary. If -mpreferred-stack-boundary is not specified, the
7706 default is 4 (16 bytes or 128 bits).
7707 On Pentium and PentiumPro, "double" and "long double" values should
7709 be aligned to an 8 byte boundary (see -malign-double) or suffer
7710 significant run time performance penalties. On Pentium III, the
7711 Streaming SIMD Extension (SSE) data type "__m128" may not work
7712 properly if it is not 16 byte aligned.
7713 To ensure proper alignment of this values on the stack, the stack
7715 boundary must be as aligned as that required by any value stored on
7716 the stack. Further, every function must be generated such that it
7717 keeps the stack aligned. Thus calling a function compiled with a
7718 higher preferred stack boundary from a function compiled with a
7719 lower preferred stack boundary will most likely misalign the stack.
7720 It is recommended that libraries that use callbacks always use the
7721 default setting.
7722 This extra alignment does consume extra stack space, and generally
7724 increases code size. Code that is sensitive to stack space usage,
7725 such as embedded systems and operating system kernels, may want to
7726 reduce the preferred alignment to -mpreferred-stack-boundary=2.
7727 -mmmx
7729 -mno-mmx
7730 -msse
7731 -mno-sse
7732 -msse2
7733 -mno-sse2
7734 -msse3
7735 -mno-sse3
7736 -m3dnow
7737 -mno-3dnow
7738 These switches enable or disable the use of instructions in the
7739 MMX, SSE, SSE2 or 3DNow! extended instruction sets. These exten‐
7740 sions are also available as built-in functions: see X86 Built-in
7741 Functions, for details of the functions enabled and disabled by
7742 these switches.
7743 To have SSE/SSE2 instructions generated automatically from float‐
7745 ing-point code (as opposed to 387 instructions), see -mfpmath=sse.
7746 These options will enable GCC to use these extended instructions in
7748 generated code, even without -mfpmath=sse. Applications which per‐
7749 form runtime CPU detection must compile separate files for each
7750 supported architecture, using the appropriate flags. In particu‐
7751 lar, the file containing the CPU detection code should be compiled
7752 without these options.
7753 -mpush-args
7755 -mno-push-args
7756 Use PUSH operations to store outgoing parameters. This method is
7757 shorter and usually equally fast as method using SUB/MOV operations
7758 and is enabled by default. In some cases disabling it may improve
7759 performance because of improved scheduling and reduced dependen‐
7760 cies.
7761 -maccumulate-outgoing-args
7763 If enabled, the maximum amount of space required for outgoing argu‐
7764 ments will be computed in the function prologue. This is faster on
7765 most modern CPUs because of reduced dependencies, improved schedul‐
7766 ing and reduced stack usage when preferred stack boundary is not
7767 equal to 2. The drawback is a notable increase in code size. This
7768 switch implies -mno-push-args.
7769 -mthreads
7771 Support thread-safe exception handling on Mingw32. Code that
7772 relies on thread-safe exception handling must compile and link all
7773 code with the -mthreads option. When compiling, -mthreads defines
7774 -D_MT; when linking, it links in a special thread helper library
7775 -lmingwthrd which cleans up per thread exception handling data.
7776 -mno-align-stringops
7778 Do not align destination of inlined string operations. This switch
7779 reduces code size and improves performance in case the destination
7780 is already aligned, but GCC doesn't know about it.
7781 -minline-all-stringops
7783 By default GCC inlines string operations only when destination is
7784 known to be aligned at least to 4 byte boundary. This enables more
7785 inlining, increase code size, but may improve performance of code
7786 that depends on fast memcpy, strlen and memset for short lengths.
7787 -momit-leaf-frame-pointer
7789 Don't keep the frame pointer in a register for leaf functions.
7790 This avoids the instructions to save, set up and restore frame
7791 pointers and makes an extra register available in leaf functions.
7792 The option -fomit-frame-pointer removes the frame pointer for all
7793 functions which might make debugging harder.
7794 -mtls-direct-seg-refs
7796 -mno-tls-direct-seg-refs
7797 Controls whether TLS variables may be accessed with offsets from
7798 the TLS segment register (%gs for 32-bit, %fs for 64-bit), or
7799 whether the thread base pointer must be added. Whether or not this
7800 is legal depends on the operating system, and whether it maps the
7801 segment to cover the entire TLS area.
7802 For systems that use GNU libc, the default is on.
7804 These -m switches are supported in addition to the above on AMD x86-64
7806 processors in 64-bit environments.
7807 -m32
7809 -m64
7810 Generate code for a 32-bit or 64-bit environment. The 32-bit envi‐
7811 ronment sets int, long and pointer to 32 bits and generates code
7812 that runs on any i386 system. The 64-bit environment sets int to
7813 32 bits and long and pointer to 64 bits and generates code for
7814 AMD's x86-64 architecture.
7815 -mno-red-zone
7817 Do not use a so called red zone for x86-64 code. The red zone is
7818 mandated by the x86-64 ABI, it is a 128-byte area beyond the loca‐
7819 tion of the stack pointer that will not be modified by signal or
7820 interrupt handlers and therefore can be used for temporary data
7821 without adjusting the stack pointer. The flag -mno-red-zone dis‐
7822 ables this red zone.
7823 -mcmodel=small
7825 Generate code for the small code model: the program and its symbols
7826 must be linked in the lower 2 GB of the address space. Pointers
7827 are 64 bits. Programs can be statically or dynamically linked.
7828 This is the default code model.
7829 -mcmodel=kernel
7831 Generate code for the kernel code model. The kernel runs in the
7832 negative 2 GB of the address space. This model has to be used for
7833 Linux kernel code.
7834 -mcmodel=medium
7836 Generate code for the medium model: The program is linked in the
7837 lower 2 GB of the address space but symbols can be located anywhere
7838 in the address space. Programs can be statically or dynamically
7839 linked, but building of shared libraries are not supported with the
7840 medium model.
7841 -mcmodel=large
7843 Generate code for the large model: This model makes no assumptions
7844 about addresses and sizes of sections. Currently GCC does not
7845 implement this model.
7846 IA-64 Options
7848 These are the -m options defined for the Intel IA-64 architecture.
7850 -mbig-endian
7852 Generate code for a big endian target. This is the default for
7853 HP-UX.
7854 -mlittle-endian
7856 Generate code for a little endian target. This is the default for
7857 AIX5 and GNU/Linux.
7858 -mgnu-as
7860 -mno-gnu-as
7861 Generate (or don't) code for the GNU assembler. This is the
7862 default.
7863 -mgnu-ld
7865 -mno-gnu-ld
7866 Generate (or don't) code for the GNU linker. This is the default.
7867 -mno-pic
7869 Generate code that does not use a global pointer register. The
7870 result is not position independent code, and violates the IA-64
7871 ABI.
7872 -mvolatile-asm-stop
7874 -mno-volatile-asm-stop
7875 Generate (or don't) a stop bit immediately before and after
7876 volatile asm statements.
7877 -mregister-names
7879 -mno-register-names
7880 Generate (or don't) in, loc, and out register names for the stacked
7881 registers. This may make assembler output more readable.
7882 -mno-sdata
7884 -msdata
7885 Disable (or enable) optimizations that use the small data section.
7886 This may be useful for working around optimizer bugs.
7887 -mconstant-gp
7889 Generate code that uses a single constant global pointer value.
7890 This is useful when compiling kernel code.
7891 -mauto-pic
7893 Generate code that is self-relocatable. This implies -mcon‐
7894 stant-gp. This is useful when compiling firmware code.
7895 -minline-float-divide-min-latency
7897 Generate code for inline divides of floating point values using the
7898 minimum latency algorithm.
7899 -minline-float-divide-max-throughput
7901 Generate code for inline divides of floating point values using the
7902 maximum throughput algorithm.
7903 -minline-int-divide-min-latency
7905 Generate code for inline divides of integer values using the mini‐
7906 mum latency algorithm.
7907 -minline-int-divide-max-throughput
7909 Generate code for inline divides of integer values using the maxi‐
7910 mum throughput algorithm.
7911 -minline-sqrt-min-latency
7913 Generate code for inline square roots using the minimum latency
7914 algorithm.
7915 -minline-sqrt-max-throughput
7917 Generate code for inline square roots using the maximum throughput
7918 algorithm.
7919 -mno-dwarf2-asm
7921 -mdwarf2-asm
7922 Don't (or do) generate assembler code for the DWARF2 line number
7923 debugging info. This may be useful when not using the GNU assem‐
7924 bler.
7925 -mearly-stop-bits
7927 -mno-early-stop-bits
7928 Allow stop bits to be placed earlier than immediately preceding the
7929 instruction that triggered the stop bit. This can improve instruc‐
7930 tion scheduling, but does not always do so.
7931 -mfixed-range=register-range
7933 Generate code treating the given register range as fixed registers.
7934 A fixed register is one that the register allocator can not use.
7935 This is useful when compiling kernel code. A register range is
7936 specified as two registers separated by a dash. Multiple register
7937 ranges can be specified separated by a comma.
7938 -mtls-size=tls-size
7940 Specify bit size of immediate TLS offsets. Valid values are 14,
7941 22, and 64.
7942 -mtune=cpu-type
7944 Tune the instruction scheduling for a particular CPU, Valid values
7945 are itanium, itanium1, merced, itanium2, and mckinley.
7946 -mt
7948 -pthread
7949 Add support for multithreading using the POSIX threads library.
7950 This option sets flags for both the preprocessor and linker. It
7951 does not affect the thread safety of object code produced by the
7952 compiler or that of libraries supplied with it. These are HP-UX
7953 specific flags.
7954 -milp32
7956 -mlp64
7957 Generate code for a 32-bit or 64-bit environment. The 32-bit envi‐
7958 ronment sets int, long and pointer to 32 bits. The 64-bit environ‐
7959 ment sets int to 32 bits and long and pointer to 64 bits. These
7960 are HP-UX specific flags.
7961 M32C Options
7963 -mcpu=name
7965 Select the CPU for which code is generated. name may be one of r8c
7966 for the R8C/Tiny series, m16c for the M16C (up to /60) series,
7967 m32cm for the M16C/80 series, or m32c for the M32C/80 series.
7968 -msim
7970 Specifies that the program will be run on the simulator. This
7971 causes an alternate runtime library to be linked in which supports,
7972 for example, file I/O. You must not use this option when generat‐
7973 ing programs that will run on real hardware; you must provide your
7974 own runtime library for whatever I/O functions are needed.
7975 -memregs=number
7977 Specifies the number of memory-based pseudo-registers GCC will use
7978 during code generation. These pseudo-registers will be used like
7979 real registers, so there is a tradeoff between GCC's ability to fit
7980 the code into available registers, and the performance penalty of
7981 using memory instead of registers. Note that all modules in a pro‐
7982 gram must be compiled with the same value for this option. Because
7983 of that, you must not use this option with the default runtime
7984 libraries gcc builds.
7985 M32R/D Options
7987 These -m options are defined for Renesas M32R/D architectures:
7989 -m32r2
7991 Generate code for the M32R/2.
7992 -m32rx
7994 Generate code for the M32R/X.
7995 -m32r
7997 Generate code for the M32R. This is the default.
7998 -mmodel=small
8000 Assume all objects live in the lower 16MB of memory (so that their
8001 addresses can be loaded with the "ld24" instruction), and assume
8002 all subroutines are reachable with the "bl" instruction. This is
8003 the default.
8004 The addressability of a particular object can be set with the
8006 "model" attribute.
8007 -mmodel=medium
8009 Assume objects may be anywhere in the 32-bit address space (the
8010 compiler will generate "seth/add3" instructions to load their
8011 addresses), and assume all subroutines are reachable with the "bl"
8012 instruction.
8013 -mmodel=large
8015 Assume objects may be anywhere in the 32-bit address space (the
8016 compiler will generate "seth/add3" instructions to load their
8017 addresses), and assume subroutines may not be reachable with the
8018 "bl" instruction (the compiler will generate the much slower
8019 "seth/add3/jl" instruction sequence).
8020 -msdata=none
8022 Disable use of the small data area. Variables will be put into one
8023 of .data, bss, or .rodata (unless the "section" attribute has been
8024 specified). This is the default.
8025 The small data area consists of sections .sdata and .sbss. Objects
8027 may be explicitly put in the small data area with the "section"
8028 attribute using one of these sections.
8029 -msdata=sdata
8031 Put small global and static data in the small data area, but do not
8032 generate special code to reference them.
8033 -msdata=use
8035 Put small global and static data in the small data area, and gener‐
8036 ate special instructions to reference them.
8037 -G num
8039 Put global and static objects less than or equal to num bytes into
8040 the small data or bss sections instead of the normal data or bss
8041 sections. The default value of num is 8. The -msdata option must
8042 be set to one of sdata or use for this option to have any effect.
8043 All modules should be compiled with the same -G num value. Compil‐
8045 ing with different values of num may or may not work; if it doesn't
8046 the linker will give an error message---incorrect code will not be
8047 generated.
8048 -mdebug
8050 Makes the M32R specific code in the compiler display some statis‐
8051 tics that might help in debugging programs.
8052 -malign-loops
8054 Align all loops to a 32-byte boundary.
8055 -mno-align-loops
8057 Do not enforce a 32-byte alignment for loops. This is the default.
8058 -missue-rate=number
8060 Issue number instructions per cycle. number can only be 1 or 2.
8061 -mbranch-cost=number
8063 number can only be 1 or 2. If it is 1 then branches will be pre‐
8064 ferred over conditional code, if it is 2, then the opposite will
8065 apply.
8066 -mflush-trap=number
8068 Specifies the trap number to use to flush the cache. The default
8069 is 12. Valid numbers are between 0 and 15 inclusive.
8070 -mno-flush-trap
8072 Specifies that the cache cannot be flushed by using a trap.
8073 -mflush-func=name
8075 Specifies the name of the operating system function to call to
8076 flush the cache. The default is _flush_cache, but a function call
8077 will only be used if a trap is not available.
8078 -mno-flush-func
8080 Indicates that there is no OS function for flushing the cache.
8081 M680x0 Options
8083 These are the -m options defined for the 68000 series. The default
8085 values for these options depends on which style of 68000 was selected
8086 when the compiler was configured; the defaults for the most common
8087 choices are given below.
8088 -m68000
8090 -mc68000
8091 Generate output for a 68000. This is the default when the compiler
8092 is configured for 68000-based systems.
8093 Use this option for microcontrollers with a 68000 or EC000 core,
8095 including the 68008, 68302, 68306, 68307, 68322, 68328 and 68356.
8096 -m68020
8098 -mc68020
8099 Generate output for a 68020. This is the default when the compiler
8100 is configured for 68020-based systems.
8101 -m68881
8103 Generate output containing 68881 instructions for floating point.
8104 This is the default for most 68020 systems unless --nfp was speci‐
8105 fied when the compiler was configured.
8106 -m68030
8108 Generate output for a 68030. This is the default when the compiler
8109 is configured for 68030-based systems.
8110 -m68040
8112 Generate output for a 68040. This is the default when the compiler
8113 is configured for 68040-based systems.
8114 This option inhibits the use of 68881/68882 instructions that have
8116 to be emulated by software on the 68040. Use this option if your
8117 68040 does not have code to emulate those instructions.
8118 -m68060
8120 Generate output for a 68060. This is the default when the compiler
8121 is configured for 68060-based systems.
8122 This option inhibits the use of 68020 and 68881/68882 instructions
8124 that have to be emulated by software on the 68060. Use this option
8125 if your 68060 does not have code to emulate those instructions.
8126 -mcpu32
8128 Generate output for a CPU32. This is the default when the compiler
8129 is configured for CPU32-based systems.
8130 Use this option for microcontrollers with a CPU32 or CPU32+ core,
8132 including the 68330, 68331, 68332, 68333, 68334, 68336, 68340,
8133 68341, 68349 and 68360.
8134 -m5200
8136 Generate output for a 520X "coldfire" family cpu. This is the
8137 default when the compiler is configured for 520X-based systems.
8138 Use this option for microcontroller with a 5200 core, including the
8140 MCF5202, MCF5203, MCF5204 and MCF5202.
8141 -m68020-40
8143 Generate output for a 68040, without using any of the new instruc‐
8144 tions. This results in code which can run relatively efficiently
8145 on either a 68020/68881 or a 68030 or a 68040. The generated code
8146 does use the 68881 instructions that are emulated on the 68040.
8147 -m68020-60
8149 Generate output for a 68060, without using any of the new instruc‐
8150 tions. This results in code which can run relatively efficiently
8151 on either a 68020/68881 or a 68030 or a 68040. The generated code
8152 does use the 68881 instructions that are emulated on the 68060.
8153 -msoft-float
8155 Generate output containing library calls for floating point. Warn‐
8156 ing: the requisite libraries are not available for all m68k tar‐
8157 gets. Normally the facilities of the machine's usual C compiler
8158 are used, but this can't be done directly in cross-compilation.
8159 You must make your own arrangements to provide suitable library
8160 functions for cross-compilation. The embedded targets m68k-*-aout
8161 and m68k-*-coff do provide software floating point support.
8162 -mshort
8164 Consider type "int" to be 16 bits wide, like "short int". Addi‐
8165 tionally, parameters passed on the stack are also aligned to a
8166 16-bit boundary even on targets whose API mandates promotion to
8167 32-bit.
8168 -mnobitfield
8170 Do not use the bit-field instructions. The -m68000, -mcpu32 and
8171 -m5200 options imply -mnobitfield.
8172 -mbitfield
8174 Do use the bit-field instructions. The -m68020 option implies
8175 -mbitfield. This is the default if you use a configuration
8176 designed for a 68020.
8177 -mrtd
8179 Use a different function-calling convention, in which functions
8180 that take a fixed number of arguments return with the "rtd"
8181 instruction, which pops their arguments while returning. This
8182 saves one instruction in the caller since there is no need to pop
8183 the arguments there.
8184 This calling convention is incompatible with the one normally used
8186 on Unix, so you cannot use it if you need to call libraries com‐
8187 piled with the Unix compiler.
8188 Also, you must provide function prototypes for all functions that
8190 take variable numbers of arguments (including "printf"); otherwise
8191 incorrect code will be generated for calls to those functions.
8192 In addition, seriously incorrect code will result if you call a
8194 function with too many arguments. (Normally, extra arguments are
8195 harmlessly ignored.)
8196 The "rtd" instruction is supported by the 68010, 68020, 68030,
8198 68040, 68060 and CPU32 processors, but not by the 68000 or 5200.
8199 -malign-int
8201 -mno-align-int
8202 Control whether GCC aligns "int", "long", "long long", "float",
8203 "double", and "long double" variables on a 32-bit boundary
8204 (-malign-int) or a 16-bit boundary (-mno-align-int). Aligning
8205 variables on 32-bit boundaries produces code that runs somewhat
8206 faster on processors with 32-bit busses at the expense of more mem‐
8207 ory.
8208 Warning: if you use the -malign-int switch, GCC will align struc‐
8210 tures containing the above types differently than most published
8211 application binary interface specifications for the m68k.
8212 -mpcrel
8214 Use the pc-relative addressing mode of the 68000 directly, instead
8215 of using a global offset table. At present, this option implies
8216 -fpic, allowing at most a 16-bit offset for pc-relative addressing.
8217 -fPIC is not presently supported with -mpcrel, though this could be
8218 supported for 68020 and higher processors.
8219 -mno-strict-align
8221 -mstrict-align
8222 Do not (do) assume that unaligned memory references will be handled
8223 by the system.
8224 -msep-data
8226 Generate code that allows the data segment to be located in a dif‐
8227 ferent area of memory from the text segment. This allows for exe‐
8228 cute in place in an environment without virtual memory management.
8229 This option implies -fPIC.
8230 -mno-sep-data
8232 Generate code that assumes that the data segment follows the text
8233 segment. This is the default.
8234 -mid-shared-library
8236 Generate code that supports shared libraries via the library ID
8237 method. This allows for execute in place and shared libraries in
8238 an environment without virtual memory management. This option
8239 implies -fPIC.
8240 -mno-id-shared-library
8242 Generate code that doesn't assume ID based shared libraries are
8243 being used. This is the default.
8244 -mshared-library-id=n
8246 Specified the identification number of the ID based shared library
8247 being compiled. Specifying a value of 0 will generate more compact
8248 code, specifying other values will force the allocation of that
8249 number to the current library but is no more space or time effi‐
8250 cient than omitting this option.
8251 M68hc1x Options
8253 These are the -m options defined for the 68hc11 and 68hc12 microcon‐
8255 trollers. The default values for these options depends on which style
8256 of microcontroller was selected when the compiler was configured; the
8257 defaults for the most common choices are given below.
8258 -m6811
8260 -m68hc11
8261 Generate output for a 68HC11. This is the default when the com‐
8262 piler is configured for 68HC11-based systems.
8263 -m6812
8265 -m68hc12
8266 Generate output for a 68HC12. This is the default when the com‐
8267 piler is configured for 68HC12-based systems.
8268 -m68S12
8270 -m68hcs12
8271 Generate output for a 68HCS12.
8272 -mauto-incdec
8274 Enable the use of 68HC12 pre and post auto-increment and auto-
8275 decrement addressing modes.
8276 -minmax
8278 -nominmax
8279 Enable the use of 68HC12 min and max instructions.
8280 -mlong-calls
8282 -mno-long-calls
8283 Treat all calls as being far away (near). If calls are assumed to
8284 be far away, the compiler will use the "call" instruction to call a
8285 function and the "rtc" instruction for returning.
8286 -mshort
8288 Consider type "int" to be 16 bits wide, like "short int".
8289 -msoft-reg-count=count
8291 Specify the number of pseudo-soft registers which are used for the
8292 code generation. The maximum number is 32. Using more pseudo-soft
8293 register may or may not result in better code depending on the pro‐
8294 gram. The default is 4 for 68HC11 and 2 for 68HC12.
8295 MCore Options
8297 These are the -m options defined for the Motorola M*Core processors.
8299 -mhardlit
8301 -mno-hardlit
8302 Inline constants into the code stream if it can be done in two
8303 instructions or less.
8304 -mdiv
8306 -mno-div
8307 Use the divide instruction. (Enabled by default).
8308 -mrelax-immediate
8310 -mno-relax-immediate
8311 Allow arbitrary sized immediates in bit operations.
8312 -mwide-bitfields
8314 -mno-wide-bitfields
8315 Always treat bit-fields as int-sized.
8316 -m4byte-functions
8318 -mno-4byte-functions
8319 Force all functions to be aligned to a four byte boundary.
8320 -mcallgraph-data
8322 -mno-callgraph-data
8323 Emit callgraph information.
8324 -mslow-bytes
8326 -mno-slow-bytes
8327 Prefer word access when reading byte quantities.
8328 -mlittle-endian
8330 -mbig-endian
8331 Generate code for a little endian target.
8332 -m210
8334 -m340
8335 Generate code for the 210 processor.
8336 MIPS Options
8338 -EB Generate big-endian code.
8340 -EL Generate little-endian code. This is the default for mips*el-*-*
8342 configurations.
8343 -march=arch
8345 Generate code that will run on arch, which can be the name of a
8346 generic MIPS ISA, or the name of a particular processor. The ISA
8347 names are: mips1, mips2, mips3, mips4, mips32, mips32r2, and
8348 mips64. The processor names are: 4kc, 4km, 4kp, 5kc, 5kf, 20kc,
8349 24k, 24kc, 24kf, 24kx, m4k, orion, r2000, r3000, r3900, r4000,
8350 r4400, r4600, r4650, r6000, r8000, rm7000, rm9000, sb1, sr71000,
8351 vr4100, vr4111, vr4120, vr4130, vr4300, vr5000, vr5400 and vr5500.
8352 The special value from-abi selects the most compatible architecture
8353 for the selected ABI (that is, mips1 for 32-bit ABIs and mips3 for
8354 64-bit ABIs).
8355 In processor names, a final 000 can be abbreviated as k (for exam‐
8357 ple, -march=r2k). Prefixes are optional, and vr may be written r.
8358 GCC defines two macros based on the value of this option. The
8360 first is _MIPS_ARCH, which gives the name of target architecture,
8361 as a string. The second has the form _MIPS_ARCH_foo, where foo is
8362 the capitalized value of _MIPS_ARCH. For example, -march=r2000
8363 will set _MIPS_ARCH to "r2000" and define the macro
8364 _MIPS_ARCH_R2000.
8365 Note that the _MIPS_ARCH macro uses the processor names given
8367 above. In other words, it will have the full prefix and will not
8368 abbreviate 000 as k. In the case of from-abi, the macro names the
8369 resolved architecture (either "mips1" or "mips3"). It names the
8370 default architecture when no -march option is given.
8371 -mtune=arch
8373 Optimize for arch. Among other things, this option controls the
8374 way instructions are scheduled, and the perceived cost of arith‐
8375 metic operations. The list of arch values is the same as for
8376 -march.
8377 When this option is not used, GCC will optimize for the processor
8379 specified by -march. By using -march and -mtune together, it is
8380 possible to generate code that will run on a family of processors,
8381 but optimize the code for one particular member of that family.
8382 -mtune defines the macros _MIPS_TUNE and _MIPS_TUNE_foo, which work
8384 in the same way as the -march ones described above.
8385 -mips1
8387 Equivalent to -march=mips1.
8388 -mips2
8390 Equivalent to -march=mips2.
8391 -mips3
8393 Equivalent to -march=mips3.
8394 -mips4
8396 Equivalent to -march=mips4.
8397 -mips32
8399 Equivalent to -march=mips32.
8400 -mips32r2
8402 Equivalent to -march=mips32r2.
8403 -mips64
8405 Equivalent to -march=mips64.
8406 -mips16
8408 -mno-mips16
8409 Generate (do not generate) MIPS16 code. If GCC is targetting a
8410 MIPS32 or MIPS64 architecture, it will make use of the MIPS16e ASE.
8411 -mabi=32
8413 -mabi=o64
8414 -mabi=n32
8415 -mabi=64
8416 -mabi=eabi
8417 Generate code for the given ABI.
8418 Note that the EABI has a 32-bit and a 64-bit variant. GCC normally
8420 generates 64-bit code when you select a 64-bit architecture, but
8421 you can use -mgp32 to get 32-bit code instead.
8422 For information about the O64 ABI, see
8424 <http://gcc.gnu.org/projects/mipso64-abi.html>.
8425 -mabicalls
8427 -mno-abicalls
8428 Generate (do not generate) SVR4-style position-independent code.
8429 -mabicalls is the default for SVR4-based systems.
8430 -mxgot
8432 -mno-xgot
8433 Lift (do not lift) the usual restrictions on the size of the global
8434 offset table.
8435 GCC normally uses a single instruction to load values from the GOT.
8437 While this is relatively efficient, it will only work if the GOT is
8438 smaller than about 64k. Anything larger will cause the linker to
8439 report an error such as:
8440 relocation truncated to fit: R_MIPS_GOT16 foobar
8442 If this happens, you should recompile your code with -mxgot. It
8444 should then work with very large GOTs, although it will also be
8445 less efficient, since it will take three instructions to fetch the
8446 value of a global symbol.
8447 Note that some linkers can create multiple GOTs. If you have such
8449 a linker, you should only need to use -mxgot when a single object
8450 file accesses more than 64k's worth of GOT entries. Very few do.
8451 These options have no effect unless GCC is generating position
8453 independent code.
8454 -mgp32
8456 Assume that general-purpose registers are 32 bits wide.
8457 -mgp64
8459 Assume that general-purpose registers are 64 bits wide.
8460 -mfp32
8462 Assume that floating-point registers are 32 bits wide.
8463 -mfp64
8465 Assume that floating-point registers are 64 bits wide.
8466 -mhard-float
8468 Use floating-point coprocessor instructions.
8469 -msoft-float
8471 Do not use floating-point coprocessor instructions. Implement
8472 floating-point calculations using library calls instead.
8473 -msingle-float
8475 Assume that the floating-point coprocessor only supports single-
8476 precision operations.
8477 -mdouble-float
8479 Assume that the floating-point coprocessor supports double-preci‐
8480 sion operations. This is the default.
8481 -mdsp
8483 -mno-dsp
8484 Use (do not use) the MIPS DSP ASE.
8485 -mpaired-single
8487 -mno-paired-single
8488 Use (do not use) paired-single floating-point instructions.
8489 This option can only be used when generating 64-bit code and
8490 requires hardware floating-point support to be enabled.
8491 -mips3d
8493 -mno-mips3d
8494 Use (do not use) the MIPS-3D ASE. The option -mips3d implies
8495 -mpaired-single.
8496 -mlong64
8498 Force "long" types to be 64 bits wide. See -mlong32 for an expla‐
8499 nation of the default and the way that the pointer size is deter‐
8500 mined.
8501 -mlong32
8503 Force "long", "int", and pointer types to be 32 bits wide.
8504 The default size of "int"s, "long"s and pointers depends on the
8506 ABI. All the supported ABIs use 32-bit "int"s. The n64 ABI uses
8507 64-bit "long"s, as does the 64-bit EABI; the others use 32-bit
8508 "long"s. Pointers are the same size as "long"s, or the same size
8509 as integer registers, whichever is smaller.
8510 -msym32
8512 -mno-sym32
8513 Assume (do not assume) that all symbols have 32-bit values, regard‐
8514 less of the selected ABI. This option is useful in combination
8515 with -mabi=64 and -mno-abicalls because it allows GCC to generate
8516 shorter and faster references to symbolic addresses.
8517 -G num
8519 Put global and static items less than or equal to num bytes into
8520 the small data or bss section instead of the normal data or bss
8521 section. This allows the data to be accessed using a single
8522 instruction.
8523 All modules should be compiled with the same -G num value.
8525 -membedded-data
8527 -mno-embedded-data
8528 Allocate variables to the read-only data section first if possible,
8529 then next in the small data section if possible, otherwise in data.
8530 This gives slightly slower code than the default, but reduces the
8531 amount of RAM required when executing, and thus may be preferred
8532 for some embedded systems.
8533 -muninit-const-in-rodata
8535 -mno-uninit-const-in-rodata
8536 Put uninitialized "const" variables in the read-only data section.
8537 This option is only meaningful in conjunction with -membedded-data.
8538 -msplit-addresses
8540 -mno-split-addresses
8541 Enable (disable) use of the "%hi()" and "%lo()" assembler reloca‐
8542 tion operators. This option has been superseded by -mex‐
8543 plicit-relocs but is retained for backwards compatibility.
8544 -mexplicit-relocs
8546 -mno-explicit-relocs
8547 Use (do not use) assembler relocation operators when dealing with
8548 symbolic addresses. The alternative, selected by
8549 -mno-explicit-relocs, is to use assembler macros instead.
8550 -mexplicit-relocs is the default if GCC was configured to use an
8552 assembler that supports relocation operators.
8553 -mcheck-zero-division
8555 -mno-check-zero-division
8556 Trap (do not trap) on integer division by zero. The default is
8557 -mcheck-zero-division.
8558 -mdivide-traps
8560 -mdivide-breaks
8561 MIPS systems check for division by zero by generating either a con‐
8562 ditional trap or a break instruction. Using traps results in
8563 smaller code, but is only supported on MIPS II and later. Also,
8564 some versions of the Linux kernel have a bug that prevents trap
8565 from generating the proper signal ("SIGFPE"). Use -mdivide-traps
8566 to allow conditional traps on architectures that support them and
8567 -mdivide-breaks to force the use of breaks.
8568 The default is usually -mdivide-traps, but this can be overridden
8570 at configure time using --with-divide=breaks. Divide-by-zero
8571 checks can be completely disabled using -mno-check-zero-division.
8572 -mmemcpy
8574 -mno-memcpy
8575 Force (do not force) the use of "memcpy()" for non-trivial block
8576 moves. The default is -mno-memcpy, which allows GCC to inline most
8577 constant-sized copies.
8578 -mlong-calls
8580 -mno-long-calls
8581 Disable (do not disable) use of the "jal" instruction. Calling
8582 functions using "jal" is more efficient but requires the caller and
8583 callee to be in the same 256 megabyte segment.
8584 This option has no effect on abicalls code. The default is
8586 -mno-long-calls.
8587 -mmad
8589 -mno-mad
8590 Enable (disable) use of the "mad", "madu" and "mul" instructions,
8591 as provided by the R4650 ISA.
8592 -mfused-madd
8594 -mno-fused-madd
8595 Enable (disable) use of the floating point multiply-accumulate
8596 instructions, when they are available. The default is
8597 -mfused-madd.
8598 When multiply-accumulate instructions are used, the intermediate
8600 product is calculated to infinite precision and is not subject to
8601 the FCSR Flush to Zero bit. This may be undesirable in some cir‐
8602 cumstances.
8603 -nocpp
8605 Tell the MIPS assembler to not run its preprocessor over user
8606 assembler files (with a .s suffix) when assembling them.
8607 -mfix-r4000
8609 -mno-fix-r4000
8610 Work around certain R4000 CPU errata:
8611 - A double-word or a variable shift may give an incorrect result
8613 if executed immediately after starting an integer division.
8614 - A double-word or a variable shift may give an incorrect result
8616 if executed while an integer multiplication is in progress.
8617 - An integer division may give an incorrect result if started in
8619 a delay slot of a taken branch or a jump.
8620 -mfix-r4400
8622 -mno-fix-r4400
8623 Work around certain R4400 CPU errata:
8624 - A double-word or a variable shift may give an incorrect result
8626 if executed immediately after starting an integer division.
8627 -mfix-vr4120
8629 -mno-fix-vr4120
8630 Work around certain VR4120 errata:
8631 - "dmultu" does not always produce the correct result.
8633 - "div" and "ddiv" do not always produce the correct result if
8635 one of the operands is negative.
8636 The workarounds for the division errata rely on special functions
8638 in libgcc.a. At present, these functions are only provided by the
8639 "mips64vr*-elf" configurations.
8640 Other VR4120 errata require a nop to be inserted between certain
8642 pairs of instructions. These errata are handled by the assembler,
8643 not by GCC itself.
8644 -mfix-vr4130
8646 Work around the VR4130 "mflo"/"mfhi" errata. The workarounds are
8647 implemented by the assembler rather than by GCC, although GCC will
8648 avoid using "mflo" and "mfhi" if the VR4130 "macc", "macchi",
8649 "dmacc" and "dmacchi" instructions are available instead.
8650 -mfix-sb1
8652 -mno-fix-sb1
8653 Work around certain SB-1 CPU core errata. (This flag currently
8654 works around the SB-1 revision 2 "F1" and "F2" floating point
8655 errata.)
8656 -mflush-func=func
8658 -mno-flush-func
8659 Specifies the function to call to flush the I and D caches, or to
8660 not call any such function. If called, the function must take the
8661 same arguments as the common "_flush_func()", that is, the address
8662 of the memory range for which the cache is being flushed, the size
8663 of the memory range, and the number 3 (to flush both caches). The
8664 default depends on the target GCC was configured for, but commonly
8665 is either _flush_func or __cpu_flush.
8666 -mbranch-likely
8668 -mno-branch-likely
8669 Enable or disable use of Branch Likely instructions, regardless of
8670 the default for the selected architecture. By default, Branch
8671 Likely instructions may be generated if they are supported by the
8672 selected architecture. An exception is for the MIPS32 and MIPS64
8673 architectures and processors which implement those architectures;
8674 for those, Branch Likely instructions will not be generated by
8675 default because the MIPS32 and MIPS64 architectures specifically
8676 deprecate their use.
8677 -mfp-exceptions
8679 -mno-fp-exceptions
8680 Specifies whether FP exceptions are enabled. This affects how we
8681 schedule FP instructions for some processors. The default is that
8682 FP exceptions are enabled.
8683 For instance, on the SB-1, if FP exceptions are disabled, and we
8685 are emitting 64-bit code, then we can use both FP pipes. Other‐
8686 wise, we can only use one FP pipe.
8687 -mvr4130-align
8689 -mno-vr4130-align
8690 The VR4130 pipeline is two-way superscalar, but can only issue two
8691 instructions together if the first one is 8-byte aligned. When
8692 this option is enabled, GCC will align pairs of instructions that
8693 it thinks should execute in parallel.
8694 This option only has an effect when optimizing for the VR4130. It
8696 normally makes code faster, but at the expense of making it bigger.
8697 It is enabled by default at optimization level -O3.
8698 MMIX Options
8700 These options are defined for the MMIX:
8702 -mlibfuncs
8704 -mno-libfuncs
8705 Specify that intrinsic library functions are being compiled, pass‐
8706 ing all values in registers, no matter the size.
8707 -mepsilon
8709 -mno-epsilon
8710 Generate floating-point comparison instructions that compare with
8711 respect to the "rE" epsilon register.
8712 -mabi=mmixware
8714 -mabi=gnu
8715 Generate code that passes function parameters and return values
8716 that (in the called function) are seen as registers $0 and up, as
8717 opposed to the GNU ABI which uses global registers $231 and up.
8718 -mzero-extend
8720 -mno-zero-extend
8721 When reading data from memory in sizes shorter than 64 bits, use
8722 (do not use) zero-extending load instructions by default, rather
8723 than sign-extending ones.
8724 -mknuthdiv
8726 -mno-knuthdiv
8727 Make the result of a division yielding a remainder have the same
8728 sign as the divisor. With the default, -mno-knuthdiv, the sign of
8729 the remainder follows the sign of the dividend. Both methods are
8730 arithmetically valid, the latter being almost exclusively used.
8731 -mtoplevel-symbols
8733 -mno-toplevel-symbols
8734 Prepend (do not prepend) a : to all global symbols, so the assembly
8735 code can be used with the "PREFIX" assembly directive.
8736 -melf
8738 Generate an executable in the ELF format, rather than the default
8739 mmo format used by the mmix simulator.
8740 -mbranch-predict
8742 -mno-branch-predict
8743 Use (do not use) the probable-branch instructions, when static
8744 branch prediction indicates a probable branch.
8745 -mbase-addresses
8747 -mno-base-addresses
8748 Generate (do not generate) code that uses base addresses. Using a
8749 base address automatically generates a request (handled by the
8750 assembler and the linker) for a constant to be set up in a global
8751 register. The register is used for one or more base address
8752 requests within the range 0 to 255 from the value held in the reg‐
8753 ister. The generally leads to short and fast code, but the number
8754 of different data items that can be addressed is limited. This
8755 means that a program that uses lots of static data may require
8756 -mno-base-addresses.
8757 -msingle-exit
8759 -mno-single-exit
8760 Force (do not force) generated code to have a single exit point in
8761 each function.
8762 MN10300 Options
8764 These -m options are defined for Matsushita MN10300 architectures:
8766 -mmult-bug
8768 Generate code to avoid bugs in the multiply instructions for the
8769 MN10300 processors. This is the default.
8770 -mno-mult-bug
8772 Do not generate code to avoid bugs in the multiply instructions for
8773 the MN10300 processors.
8774 -mam33
8776 Generate code which uses features specific to the AM33 processor.
8777 -mno-am33
8779 Do not generate code which uses features specific to the AM33 pro‐
8780 cessor. This is the default.
8781 -mreturn-pointer-on-d0
8783 When generating a function which returns a pointer, return the
8784 pointer in both "a0" and "d0". Otherwise, the pointer is returned
8785 only in a0, and attempts to call such functions without a prototype
8786 would result in errors. Note that this option is on by default;
8787 use -mno-return-pointer-on-d0 to disable it.
8788 -mno-crt0
8790 Do not link in the C run-time initialization object file.
8791 -mrelax
8793 Indicate to the linker that it should perform a relaxation opti‐
8794 mization pass to shorten branches, calls and absolute memory
8795 addresses. This option only has an effect when used on the command
8796 line for the final link step.
8797 This option makes symbolic debugging impossible.
8799 MT Options
8801 These -m options are defined for Morpho MT architectures:
8803 -march=cpu-type
8805 Generate code that will run on cpu-type, which is the name of a
8806 system representing a certain processor type. Possible values for
8807 cpu-type are ms1-64-001, ms1-16-002, ms1-16-003 and ms2.
8808 When this option is not used, the default is -march=ms1-16-002.
8810 -mbacc
8812 Use byte loads and stores when generating code.
8813 -mno-bacc
8815 Do not use byte loads and stores when generating code.
8816 -msim
8818 Use simulator runtime
8819 -mno-crt0
8821 Do not link in the C run-time initialization object file crti.o.
8822 Other run-time initialization and termination files such as
8823 startup.o and exit.o are still included on the linker command line.
8824 PDP-11 Options
8826 These options are defined for the PDP-11:
8828 -mfpu
8830 Use hardware FPP floating point. This is the default. (FIS float‐
8831 ing point on the PDP-11/40 is not supported.)
8832 -msoft-float
8834 Do not use hardware floating point.
8835 -mac0
8837 Return floating-point results in ac0 (fr0 in Unix assembler syn‐
8838 tax).
8839 -mno-ac0
8841 Return floating-point results in memory. This is the default.
8842 -m40
8844 Generate code for a PDP-11/40.
8845 -m45
8847 Generate code for a PDP-11/45. This is the default.
8848 -m10
8850 Generate code for a PDP-11/10.
8851 -mbcopy-builtin
8853 Use inline "movmemhi" patterns for copying memory. This is the
8854 default.
8855 -mbcopy
8857 Do not use inline "movmemhi" patterns for copying memory.
8858 -mint16
8860 -mno-int32
8861 Use 16-bit "int". This is the default.
8862 -mint32
8864 -mno-int16
8865 Use 32-bit "int".
8866 -mfloat64
8868 -mno-float32
8869 Use 64-bit "float". This is the default.
8870 -mfloat32
8872 -mno-float64
8873 Use 32-bit "float".
8874 -mabshi
8876 Use "abshi2" pattern. This is the default.
8877 -mno-abshi
8879 Do not use "abshi2" pattern.
8880 -mbranch-expensive
8882 Pretend that branches are expensive. This is for experimenting
8883 with code generation only.
8884 -mbranch-cheap
8886 Do not pretend that branches are expensive. This is the default.
8887 -msplit
8889 Generate code for a system with split I&D.
8890 -mno-split
8892 Generate code for a system without split I&D. This is the default.
8893 -munix-asm
8895 Use Unix assembler syntax. This is the default when configured for
8896 pdp11-*-bsd.
8897 -mdec-asm
8899 Use DEC assembler syntax. This is the default when configured for
8900 any PDP-11 target other than pdp11-*-bsd.
8901 PowerPC Options
8903 These are listed under
8905 IBM RS/6000 and PowerPC Options
8907 These -m options are defined for the IBM RS/6000 and PowerPC:
8909 -mpower
8911 -mno-power
8912 -mpower2
8913 -mno-power2
8914 -mpowerpc
8915 -mno-powerpc
8916 -mpowerpc-gpopt
8917 -mno-powerpc-gpopt
8918 -mpowerpc-gfxopt
8919 -mno-powerpc-gfxopt
8920 -mpowerpc64
8921 -mno-powerpc64
8922 -mmfcrf
8923 -mno-mfcrf
8924 -mpopcntb
8925 -mno-popcntb
8926 -mfprnd
8927 -mno-fprnd
8928 GCC supports two related instruction set architectures for the
8929 RS/6000 and PowerPC. The POWER instruction set are those instruc‐
8930 tions supported by the rios chip set used in the original RS/6000
8931 systems and the PowerPC instruction set is the architecture of the
8932 Freescale MPC5xx, MPC6xx, MPC8xx microprocessors, and the IBM 4xx,
8933 6xx, and follow-on microprocessors.
8934 Neither architecture is a subset of the other. However there is a
8936 large common subset of instructions supported by both. An MQ reg‐
8937 ister is included in processors supporting the POWER architecture.
8938 You use these options to specify which instructions are available
8940 on the processor you are using. The default value of these options
8941 is determined when configuring GCC. Specifying the -mcpu=cpu_type
8942 overrides the specification of these options. We recommend you use
8943 the -mcpu=cpu_type option rather than the options listed above.
8944 The -mpower option allows GCC to generate instructions that are
8946 found only in the POWER architecture and to use the MQ register.
8947 Specifying -mpower2 implies -power and also allows GCC to generate
8948 instructions that are present in the POWER2 architecture but not
8949 the original POWER architecture.
8950 The -mpowerpc option allows GCC to generate instructions that are
8952 found only in the 32-bit subset of the PowerPC architecture. Spec‐
8953 ifying -mpowerpc-gpopt implies -mpowerpc and also allows GCC to use
8954 the optional PowerPC architecture instructions in the General Pur‐
8955 pose group, including floating-point square root. Specifying
8956 -mpowerpc-gfxopt implies -mpowerpc and also allows GCC to use the
8957 optional PowerPC architecture instructions in the Graphics group,
8958 including floating-point select.
8959 The -mmfcrf option allows GCC to generate the move from condition
8961 register field instruction implemented on the POWER4 processor and
8962 other processors that support the PowerPC V2.01 architecture. The
8963 -mpopcntb option allows GCC to generate the popcount and double
8964 precision FP reciprocal estimate instruction implemented on the
8965 POWER5 processor and other processors that support the PowerPC
8966 V2.02 architecture. The -mfprnd option allows GCC to generate the
8967 FP round to integer instructions implemented on the POWER5+ proces‐
8968 sor and other processors that support the PowerPC V2.03 architec‐
8969 ture.
8970 The -mpowerpc64 option allows GCC to generate the additional 64-bit
8972 instructions that are found in the full PowerPC64 architecture and
8973 to treat GPRs as 64-bit, doubleword quantities. GCC defaults to
8974 -mno-powerpc64.
8975 If you specify both -mno-power and -mno-powerpc, GCC will use only
8977 the instructions in the common subset of both architectures plus
8978 some special AIX common-mode calls, and will not use the MQ regis‐
8979 ter. Specifying both -mpower and -mpowerpc permits GCC to use any
8980 instruction from either architecture and to allow use of the MQ
8981 register; specify this for the Motorola MPC601.
8982 -mnew-mnemonics
8984 -mold-mnemonics
8985 Select which mnemonics to use in the generated assembler code.
8986 With -mnew-mnemonics, GCC uses the assembler mnemonics defined for
8987 the PowerPC architecture. With -mold-mnemonics it uses the assem‐
8988 bler mnemonics defined for the POWER architecture. Instructions
8989 defined in only one architecture have only one mnemonic; GCC uses
8990 that mnemonic irrespective of which of these options is specified.
8991 GCC defaults to the mnemonics appropriate for the architecture in
8993 use. Specifying -mcpu=cpu_type sometimes overrides the value of
8994 these option. Unless you are building a cross-compiler, you should
8995 normally not specify either -mnew-mnemonics or -mold-mnemonics, but
8996 should instead accept the default.
8997 -mcpu=cpu_type
8999 Set architecture type, register usage, choice of mnemonics, and
9000 instruction scheduling parameters for machine type cpu_type. Sup‐
9001 ported values for cpu_type are 401, 403, 405, 405fp, 440, 440fp,
9002 505, 601, 602, 603, 603e, 604, 604e, 620, 630, 740, 7400, 7450,
9003 750, 801, 821, 823, 860, 970, 8540, ec603e, G3, G4, G5, power,
9004 power2, power3, power4, power5, power5+, power6, common, powerpc,
9005 powerpc64, rios, rios1, rios2, rsc, and rs64.
9006 -mcpu=common selects a completely generic processor. Code gener‐
9008 ated under this option will run on any POWER or PowerPC processor.
9009 GCC will use only the instructions in the common subset of both
9010 architectures, and will not use the MQ register. GCC assumes a
9011 generic processor model for scheduling purposes.
9012 -mcpu=power, -mcpu=power2, -mcpu=powerpc, and -mcpu=powerpc64 spec‐
9014 ify generic POWER, POWER2, pure 32-bit PowerPC (i.e., not MPC601),
9015 and 64-bit PowerPC architecture machine types, with an appropriate,
9016 generic processor model assumed for scheduling purposes.
9017 The other options specify a specific processor. Code generated
9019 under those options will run best on that processor, and may not
9020 run at all on others.
9021 The -mcpu options automatically enable or disable the following
9023 options: -maltivec, -mfprnd, -mhard-float, -mmfcrf, -mmultiple,
9024 -mnew-mnemonics, -mpopcntb, -mpower, -mpower2, -mpowerpc64, -mpow‐
9025 erpc-gpopt, -mpowerpc-gfxopt, -mstring. The particular options set
9026 for any particular CPU will vary between compiler versions, depend‐
9027 ing on what setting seems to produce optimal code for that CPU; it
9028 doesn't necessarily reflect the actual hardware's capabilities. If
9029 you wish to set an individual option to a particular value, you may
9030 specify it after the -mcpu option, like -mcpu=970 -mno-altivec.
9031 On AIX, the -maltivec and -mpowerpc64 options are not enabled or
9033 disabled by the -mcpu option at present because AIX does not have
9034 full support for these options. You may still enable or disable
9035 them individually if you're sure it'll work in your environment.
9036 -mtune=cpu_type
9038 Set the instruction scheduling parameters for machine type
9039 cpu_type, but do not set the architecture type, register usage, or
9040 choice of mnemonics, as -mcpu=cpu_type would. The same values for
9041 cpu_type are used for -mtune as for -mcpu. If both are specified,
9042 the code generated will use the architecture, registers, and
9043 mnemonics set by -mcpu, but the scheduling parameters set by
9044 -mtune.
9045 -mswdiv
9047 -mno-swdiv
9048 Generate code to compute division as reciprocal estimate and itera‐
9049 tive refinement, creating opportunities for increased throughput.
9050 This feature requires: optional PowerPC Graphics instruction set
9051 for single precision and FRE instruction for double precision,
9052 assuming divides cannot generate user-visible traps, and the domain
9053 values not include Infinities, denormals or zero denominator.
9054 -maltivec
9056 -mno-altivec
9057 Generate code that uses (does not use) AltiVec instructions, and
9058 also enable the use of built-in functions that allow more direct
9059 access to the AltiVec instruction set. You may also need to set
9060 -mabi=altivec to adjust the current ABI with AltiVec ABI enhance‐
9061 ments.
9062 -mvrsave
9064 -mno-vrsave
9065 Generate VRSAVE instructions when generating AltiVec code.
9066 -msecure-plt
9068 Generate code that allows ld and ld.so to build executables and
9069 shared libraries with non-exec .plt and .got sections. This is a
9070 PowerPC 32-bit SYSV ABI option.
9071 -mbss-plt
9073 Generate code that uses a BSS .plt section that ld.so fills in, and
9074 requires .plt and .got sections that are both writable and exe‐
9075 cutable. This is a PowerPC 32-bit SYSV ABI option.
9076 -misel
9078 -mno-isel
9079 This switch enables or disables the generation of ISEL instruc‐
9080 tions.
9081 -misel=yes/no
9083 This switch has been deprecated. Use -misel and -mno-isel instead.
9084 -mspe
9086 -mno-isel
9087 This switch enables or disables the generation of SPE simd instruc‐
9088 tions.
9089 -mspe=yes/no
9091 This option has been deprecated. Use -mspe and -mno-spe instead.
9092 -mfloat-gprs=yes/single/double/no
9094 -mfloat-gprs
9095 This switch enables or disables the generation of floating point
9096 operations on the general purpose registers for architectures that
9097 support it.
9098 The argument yes or single enables the use of single-precision
9100 floating point operations.
9101 The argument double enables the use of single and double-precision
9103 floating point operations.
9104 The argument no disables floating point operations on the general
9106 purpose registers.
9107 This option is currently only available on the MPC854x.
9109 -m32
9111 -m64
9112 Generate code for 32-bit or 64-bit environments of Darwin and SVR4
9113 targets (including GNU/Linux). The 32-bit environment sets int,
9114 long and pointer to 32 bits and generates code that runs on any
9115 PowerPC variant. The 64-bit environment sets int to 32 bits and
9116 long and pointer to 64 bits, and generates code for PowerPC64, as
9117 for -mpowerpc64.
9118 -mfull-toc
9120 -mno-fp-in-toc
9121 -mno-sum-in-toc
9122 -mminimal-toc
9123 Modify generation of the TOC (Table Of Contents), which is created
9124 for every executable file. The -mfull-toc option is selected by
9125 default. In that case, GCC will allocate at least one TOC entry
9126 for each unique non-automatic variable reference in your program.
9127 GCC will also place floating-point constants in the TOC. However,
9128 only 16,384 entries are available in the TOC.
9129 If you receive a linker error message that saying you have over‐
9131 flowed the available TOC space, you can reduce the amount of TOC
9132 space used with the -mno-fp-in-toc and -mno-sum-in-toc options.
9133 -mno-fp-in-toc prevents GCC from putting floating-point constants
9134 in the TOC and -mno-sum-in-toc forces GCC to generate code to cal‐
9135 culate the sum of an address and a constant at run-time instead of
9136 putting that sum into the TOC. You may specify one or both of
9137 these options. Each causes GCC to produce very slightly slower and
9138 larger code at the expense of conserving TOC space.
9139 If you still run out of space in the TOC even when you specify both
9141 of these options, specify -mminimal-toc instead. This option
9142 causes GCC to make only one TOC entry for every file. When you
9143 specify this option, GCC will produce code that is slower and
9144 larger but which uses extremely little TOC space. You may wish to
9145 use this option only on files that contain less frequently executed
9146 code.
9147 -maix64
9149 -maix32
9150 Enable 64-bit AIX ABI and calling convention: 64-bit pointers,
9151 64-bit "long" type, and the infrastructure needed to support them.
9152 Specifying -maix64 implies -mpowerpc64 and -mpowerpc, while -maix32
9153 disables the 64-bit ABI and implies -mno-powerpc64. GCC defaults
9154 to -maix32.
9155 -mxl-compat
9157 -mno-xl-compat
9158 Produce code that conforms more closely to IBM XL compiler seman‐
9159 tics when using AIX-compatible ABI. Pass floating-point arguments
9160 to prototyped functions beyond the register save area (RSA) on the
9161 stack in addition to argument FPRs. Do not assume that most sig‐
9162 nificant double in 128-bit long double value is properly rounded
9163 when comparing values and converting to double. Use XL symbol
9164 names for long double support routines.
9165 The AIX calling convention was extended but not initially docu‐
9167 mented to handle an obscure K&R C case of calling a function that
9168 takes the address of its arguments with fewer arguments than
9169 declared. IBM XL compilers access floating point arguments which
9170 do not fit in the RSA from the stack when a subroutine is compiled
9171 without optimization. Because always storing floating-point argu‐
9172 ments on the stack is inefficient and rarely needed, this option is
9173 not enabled by default and only is necessary when calling subrou‐
9174 tines compiled by IBM XL compilers without optimization.
9175 -mpe
9177 Support IBM RS/6000 SP Parallel Environment (PE). Link an applica‐
9178 tion written to use message passing with special startup code to
9179 enable the application to run. The system must have PE installed
9180 in the standard location (/usr/lpp/ppe.poe/), or the specs file
9181 must be overridden with the -specs= option to specify the appropri‐
9182 ate directory location. The Parallel Environment does not support
9183 threads, so the -mpe option and the -pthread option are incompati‐
9184 ble.
9185 -malign-natural
9187 -malign-power
9188 On AIX, 32-bit Darwin, and 64-bit PowerPC GNU/Linux, the option
9189 -malign-natural overrides the ABI-defined alignment of larger
9190 types, such as floating-point doubles, on their natural size-based
9191 boundary. The option -malign-power instructs GCC to follow the
9192 ABI-specified alignment rules. GCC defaults to the standard align‐
9193 ment defined in the ABI.
9194 On 64-bit Darwin, natural alignment is the default, and
9196 -malign-power is not supported.
9197 -msoft-float
9199 -mhard-float
9200 Generate code that does not use (uses) the floating-point register
9201 set. Software floating point emulation is provided if you use the
9202 -msoft-float option, and pass the option to GCC when linking.
9203 -mmultiple
9205 -mno-multiple
9206 Generate code that uses (does not use) the load multiple word
9207 instructions and the store multiple word instructions. These
9208 instructions are generated by default on POWER systems, and not
9209 generated on PowerPC systems. Do not use -mmultiple on little
9210 endian PowerPC systems, since those instructions do not work when
9211 the processor is in little endian mode. The exceptions are PPC740
9212 and PPC750 which permit the instructions usage in little endian
9213 mode.
9214 -mstring
9216 -mno-string
9217 Generate code that uses (does not use) the load string instructions
9218 and the store string word instructions to save multiple registers
9219 and do small block moves. These instructions are generated by
9220 default on POWER systems, and not generated on PowerPC systems. Do
9221 not use -mstring on little endian PowerPC systems, since those
9222 instructions do not work when the processor is in little endian
9223 mode. The exceptions are PPC740 and PPC750 which permit the
9224 instructions usage in little endian mode.
9225 -mupdate
9227 -mno-update
9228 Generate code that uses (does not use) the load or store instruc‐
9229 tions that update the base register to the address of the calcu‐
9230 lated memory location. These instructions are generated by
9231 default. If you use -mno-update, there is a small window between
9232 the time that the stack pointer is updated and the address of the
9233 previous frame is stored, which means code that walks the stack
9234 frame across interrupts or signals may get corrupted data.
9235 -mfused-madd
9237 -mno-fused-madd
9238 Generate code that uses (does not use) the floating point multiply
9239 and accumulate instructions. These instructions are generated by
9240 default if hardware floating is used.
9241 -mno-bit-align
9243 -mbit-align
9244 On System V.4 and embedded PowerPC systems do not (do) force struc‐
9245 tures and unions that contain bit-fields to be aligned to the base
9246 type of the bit-field.
9247 For example, by default a structure containing nothing but 8
9249 "unsigned" bit-fields of length 1 would be aligned to a 4 byte
9250 boundary and have a size of 4 bytes. By using -mno-bit-align, the
9251 structure would be aligned to a 1 byte boundary and be one byte in
9252 size.
9253 -mno-strict-align
9255 -mstrict-align
9256 On System V.4 and embedded PowerPC systems do not (do) assume that
9257 unaligned memory references will be handled by the system.
9258 -mrelocatable
9260 -mno-relocatable
9261 On embedded PowerPC systems generate code that allows (does not
9262 allow) the program to be relocated to a different address at run‐
9263 time. If you use -mrelocatable on any module, all objects linked
9264 together must be compiled with -mrelocatable or -mrelocatable-lib.
9265 -mrelocatable-lib
9267 -mno-relocatable-lib
9268 On embedded PowerPC systems generate code that allows (does not
9269 allow) the program to be relocated to a different address at run‐
9270 time. Modules compiled with -mrelocatable-lib can be linked with
9271 either modules compiled without -mrelocatable and -mrelocatable-lib
9272 or with modules compiled with the -mrelocatable options.
9273 -mno-toc
9275 -mtoc
9276 On System V.4 and embedded PowerPC systems do not (do) assume that
9277 register 2 contains a pointer to a global area pointing to the
9278 addresses used in the program.
9279 -mlittle
9281 -mlittle-endian
9282 On System V.4 and embedded PowerPC systems compile code for the
9283 processor in little endian mode. The -mlittle-endian option is the
9284 same as -mlittle.
9285 -mbig
9287 -mbig-endian
9288 On System V.4 and embedded PowerPC systems compile code for the
9289 processor in big endian mode. The -mbig-endian option is the same
9290 as -mbig.
9291 -mdynamic-no-pic
9293 On Darwin and Mac OS X systems, compile code so that it is not
9294 relocatable, but that its external references are relocatable. The
9295 resulting code is suitable for applications, but not shared
9296 libraries.
9297 -mprioritize-restricted-insns=priority
9299 This option controls the priority that is assigned to dispatch-slot
9300 restricted instructions during the second scheduling pass. The
9301 argument priority takes the value 0/1/2 to assign no/highest/sec‐
9302 ond-highest priority to dispatch slot restricted instructions.
9303 -msched-costly-dep=dependence_type
9305 This option controls which dependences are considered costly by the
9306 target during instruction scheduling. The argument dependence_type
9307 takes one of the following values: no: no dependence is costly,
9308 all: all dependences are costly, true_store_to_load: a true depen‐
9309 dence from store to load is costly, store_to_load: any dependence
9310 from store to load is costly, number: any dependence which latency
9311 >= number is costly.
9312 -minsert-sched-nops=scheme
9314 This option controls which nop insertion scheme will be used during
9315 the second scheduling pass. The argument scheme takes one of the
9316 following values: no: Don't insert nops. pad: Pad with nops any
9317 dispatch group which has vacant issue slots, according to the
9318 scheduler's grouping. regroup_exact: Insert nops to force costly
9319 dependent insns into separate groups. Insert exactly as many nops
9320 as needed to force an insn to a new group, according to the esti‐
9321 mated processor grouping. number: Insert nops to force costly
9322 dependent insns into separate groups. Insert number nops to force
9323 an insn to a new group.
9324 -mcall-sysv
9326 On System V.4 and embedded PowerPC systems compile code using call‐
9327 ing conventions that adheres to the March 1995 draft of the System
9328 V Application Binary Interface, PowerPC processor supplement. This
9329 is the default unless you configured GCC using powerpc-*-eabiaix.
9330 -mcall-sysv-eabi
9332 Specify both -mcall-sysv and -meabi options.
9333 -mcall-sysv-noeabi
9335 Specify both -mcall-sysv and -mno-eabi options.
9336 -mcall-solaris
9338 On System V.4 and embedded PowerPC systems compile code for the
9339 Solaris operating system.
9340 -mcall-linux
9342 On System V.4 and embedded PowerPC systems compile code for the
9343 Linux-based GNU system.
9344 -mcall-gnu
9346 On System V.4 and embedded PowerPC systems compile code for the
9347 Hurd-based GNU system.
9348 -mcall-netbsd
9350 On System V.4 and embedded PowerPC systems compile code for the
9351 NetBSD operating system.
9352 -maix-struct-return
9354 Return all structures in memory (as specified by the AIX ABI).
9355 -msvr4-struct-return
9357 Return structures smaller than 8 bytes in registers (as specified
9358 by the SVR4 ABI).
9359 -mabi=abi-type
9361 Extend the current ABI with a particular extension, or remove such
9362 extension. Valid values are altivec, no-altivec, spe, no-spe, ibm‐
9363 longdouble, ieeelongdouble.
9364 -mabi=spe
9366 Extend the current ABI with SPE ABI extensions. This does not
9367 change the default ABI, instead it adds the SPE ABI extensions to
9368 the current ABI.
9369 -mabi=no-spe
9371 Disable Booke SPE ABI extensions for the current ABI.
9372 -mabi=ibmlongdouble
9374 Change the current ABI to use IBM extended precision long double.
9375 This is a PowerPC 32-bit SYSV ABI option.
9376 -mabi=ieeelongdouble
9378 Change the current ABI to use IEEE extended precision long double.
9379 This is a PowerPC 32-bit Linux ABI option.
9380 -mprototype
9382 -mno-prototype
9383 On System V.4 and embedded PowerPC systems assume that all calls to
9384 variable argument functions are properly prototyped. Otherwise,
9385 the compiler must insert an instruction before every non prototyped
9386 call to set or clear bit 6 of the condition code register (CR) to
9387 indicate whether floating point values were passed in the floating
9388 point registers in case the function takes a variable arguments.
9389 With -mprototype, only calls to prototyped variable argument func‐
9390 tions will set or clear the bit.
9391 -msim
9393 On embedded PowerPC systems, assume that the startup module is
9394 called sim-crt0.o and that the standard C libraries are libsim.a
9395 and libc.a. This is the default for powerpc-*-eabisim. configura‐
9396 tions.
9397 -mmvme
9399 On embedded PowerPC systems, assume that the startup module is
9400 called crt0.o and the standard C libraries are libmvme.a and
9401 libc.a.
9402 -mads
9404 On embedded PowerPC systems, assume that the startup module is
9405 called crt0.o and the standard C libraries are libads.a and libc.a.
9406 -myellowknife
9408 On embedded PowerPC systems, assume that the startup module is
9409 called crt0.o and the standard C libraries are libyk.a and libc.a.
9410 -mvxworks
9412 On System V.4 and embedded PowerPC systems, specify that you are
9413 compiling for a VxWorks system.
9414 -mwindiss
9416 Specify that you are compiling for the WindISS simulation environ‐
9417 ment.
9418 -memb
9420 On embedded PowerPC systems, set the PPC_EMB bit in the ELF flags
9421 header to indicate that eabi extended relocations are used.
9422 -meabi
9424 -mno-eabi
9425 On System V.4 and embedded PowerPC systems do (do not) adhere to
9426 the Embedded Applications Binary Interface (eabi) which is a set of
9427 modifications to the System V.4 specifications. Selecting -meabi
9428 means that the stack is aligned to an 8 byte boundary, a function
9429 "__eabi" is called to from "main" to set up the eabi environment,
9430 and the -msdata option can use both "r2" and "r13" to point to two
9431 separate small data areas. Selecting -mno-eabi means that the
9432 stack is aligned to a 16 byte boundary, do not call an initializa‐
9433 tion function from "main", and the -msdata option will only use
9434 "r13" to point to a single small data area. The -meabi option is
9435 on by default if you configured GCC using one of the pow‐
9436 erpc*-*-eabi* options.
9437 -msdata=eabi
9439 On System V.4 and embedded PowerPC systems, put small initialized
9440 "const" global and static data in the .sdata2 section, which is
9441 pointed to by register "r2". Put small initialized non-"const"
9442 global and static data in the .sdata section, which is pointed to
9443 by register "r13". Put small uninitialized global and static data
9444 in the .sbss section, which is adjacent to the .sdata section. The
9445 -msdata=eabi option is incompatible with the -mrelocatable option.
9446 The -msdata=eabi option also sets the -memb option.
9447 -msdata=sysv
9449 On System V.4 and embedded PowerPC systems, put small global and
9450 static data in the .sdata section, which is pointed to by register
9451 "r13". Put small uninitialized global and static data in the .sbss
9452 section, which is adjacent to the .sdata section. The -msdata=sysv
9453 option is incompatible with the -mrelocatable option.
9454 -msdata=default
9456 -msdata
9457 On System V.4 and embedded PowerPC systems, if -meabi is used, com‐
9458 pile code the same as -msdata=eabi, otherwise compile code the same
9459 as -msdata=sysv.
9460 -msdata-data
9462 On System V.4 and embedded PowerPC systems, put small global and
9463 static data in the .sdata section. Put small uninitialized global
9464 and static data in the .sbss section. Do not use register "r13" to
9465 address small data however. This is the default behavior unless
9466 other -msdata options are used.
9467 -msdata=none
9469 -mno-sdata
9470 On embedded PowerPC systems, put all initialized global and static
9471 data in the .data section, and all uninitialized data in the .bss
9472 section.
9473 -G num
9475 On embedded PowerPC systems, put global and static items less than
9476 or equal to num bytes into the small data or bss sections instead
9477 of the normal data or bss section. By default, num is 8. The -G
9478 num switch is also passed to the linker. All modules should be
9479 compiled with the same -G num value.
9480 -mregnames
9482 -mno-regnames
9483 On System V.4 and embedded PowerPC systems do (do not) emit regis‐
9484 ter names in the assembly language output using symbolic forms.
9485 -mlongcall
9487 -mno-longcall
9488 Default to making all function calls indirectly, using a register,
9489 so that functions which reside further than 32 megabytes
9490 (33,554,432 bytes) from the current location can be called. This
9491 setting can be overridden by the "shortcall" function attribute, or
9492 by "#pragma longcall(0)".
9493 Some linkers are capable of detecting out-of-range calls and gener‐
9495 ating glue code on the fly. On these systems, long calls are
9496 unnecessary and generate slower code. As of this writing, the AIX
9497 linker can do this, as can the GNU linker for PowerPC/64. It is
9498 planned to add this feature to the GNU linker for 32-bit PowerPC
9499 systems as well.
9500 On Darwin/PPC systems, "#pragma longcall" will generate "jbsr
9502 callee, L42", plus a "branch island" (glue code). The two target
9503 addresses represent the callee and the "branch island". The Dar‐
9504 win/PPC linker will prefer the first address and generate a "bl
9505 callee" if the PPC "bl" instruction will reach the callee directly;
9506 otherwise, the linker will generate "bl L42" to call the "branch
9507 island". The "branch island" is appended to the body of the call‐
9508 ing function; it computes the full 32-bit address of the callee and
9509 jumps to it.
9510 On Mach-O (Darwin) systems, this option directs the compiler emit
9512 to the glue for every direct call, and the Darwin linker decides
9513 whether to use or discard it.
9514 In the future, we may cause GCC to ignore all longcall specifica‐
9516 tions when the linker is known to generate glue.
9517 -pthread
9519 Adds support for multithreading with the pthreads library. This
9520 option sets flags for both the preprocessor and linker.
9521 S/390 and zSeries Options
9523 These are the -m options defined for the S/390 and zSeries architec‐
9525 ture.
9526 -mhard-float
9528 -msoft-float
9529 Use (do not use) the hardware floating-point instructions and reg‐
9530 isters for floating-point operations. When -msoft-float is speci‐
9531 fied, functions in libgcc.a will be used to perform floating-point
9532 operations. When -mhard-float is specified, the compiler generates
9533 IEEE floating-point instructions. This is the default.
9534 -mlong-double-64
9536 -mlong-double-128
9537 These switches control the size of "long double" type. A size of
9538 64bit makes the "long double" type equivalent to the "double" type.
9539 This is the default.
9540 -mbackchain
9542 -mno-backchain
9543 Store (do not store) the address of the caller's frame as backchain
9544 pointer into the callee's stack frame. A backchain may be needed
9545 to allow debugging using tools that do not understand DWARF-2 call
9546 frame information. When -mno-packed-stack is in effect, the
9547 backchain pointer is stored at the bottom of the stack frame; when
9548 -mpacked-stack is in effect, the backchain is placed into the top‐
9549 most word of the 96/160 byte register save area.
9550 In general, code compiled with -mbackchain is call-compatible with
9552 code compiled with -mmo-backchain; however, use of the backchain
9553 for debugging purposes usually requires that the whole binary is
9554 built with -mbackchain. Note that the combination of -mbackchain,
9555 -mpacked-stack and -mhard-float is not supported. In order to
9556 build a linux kernel use -msoft-float.
9557 The default is to not maintain the backchain.
9559 -mpacked-stack
9561 -mno-packed-stack
9562 Use (do not use) the packed stack layout. When -mno-packed-stack
9563 is specified, the compiler uses the all fields of the 96/160 byte
9564 register save area only for their default purpose; unused fields
9565 still take up stack space. When -mpacked-stack is specified, reg‐
9566 ister save slots are densely packed at the top of the register save
9567 area; unused space is reused for other purposes, allowing for more
9568 efficient use of the available stack space. However, when
9569 -mbackchain is also in effect, the topmost word of the save area is
9570 always used to store the backchain, and the return address register
9571 is always saved two words below the backchain.
9572 As long as the stack frame backchain is not used, code generated
9574 with -mpacked-stack is call-compatible with code generated with
9575 -mno-packed-stack. Note that some non-FSF releases of GCC 2.95 for
9576 S/390 or zSeries generated code that uses the stack frame backchain
9577 at run time, not just for debugging purposes. Such code is not
9578 call-compatible with code compiled with -mpacked-stack. Also, note
9579 that the combination of -mbackchain, -mpacked-stack and
9580 -mhard-float is not supported. In order to build a linux kernel
9581 use -msoft-float.
9582 The default is to not use the packed stack layout.
9584 -msmall-exec
9586 -mno-small-exec
9587 Generate (or do not generate) code using the "bras" instruction to
9588 do subroutine calls. This only works reliably if the total exe‐
9589 cutable size does not exceed 64k. The default is to use the "basr"
9590 instruction instead, which does not have this limitation.
9591 -m64
9593 -m31
9594 When -m31 is specified, generate code compliant to the GNU/Linux
9595 for S/390 ABI. When -m64 is specified, generate code compliant to
9596 the GNU/Linux for zSeries ABI. This allows GCC in particular to
9597 generate 64-bit instructions. For the s390 targets, the default is
9598 -m31, while the s390x targets default to -m64.
9599 -mzarch
9601 -mesa
9602 When -mzarch is specified, generate code using the instructions
9603 available on z/Architecture. When -mesa is specified, generate
9604 code using the instructions available on ESA/390. Note that -mesa
9605 is not possible with -m64. When generating code compliant to the
9606 GNU/Linux for S/390 ABI, the default is -mesa. When generating
9607 code compliant to the GNU/Linux for zSeries ABI, the default is
9608 -mzarch.
9609 -mmvcle
9611 -mno-mvcle
9612 Generate (or do not generate) code using the "mvcle" instruction to
9613 perform block moves. When -mno-mvcle is specified, use a "mvc"
9614 loop instead. This is the default unless optimizing for size.
9615 -mdebug
9617 -mno-debug
9618 Print (or do not print) additional debug information when compil‐
9619 ing. The default is to not print debug information.
9620 -march=cpu-type
9622 Generate code that will run on cpu-type, which is the name of a
9623 system representing a certain processor type. Possible values for
9624 cpu-type are g5, g6, z900, and z990. When generating code using
9625 the instructions available on z/Architecture, the default is
9626 -march=z900. Otherwise, the default is -march=g5.
9627 -mtune=cpu-type
9629 Tune to cpu-type everything applicable about the generated code,
9630 except for the ABI and the set of available instructions. The list
9631 of cpu-type values is the same as for -march. The default is the
9632 value used for -march.
9633 -mtpf-trace
9635 -mno-tpf-trace
9636 Generate code that adds (does not add) in TPF OS specific branches
9637 to trace routines in the operating system. This option is off by
9638 default, even when compiling for the TPF OS.
9639 -mfused-madd
9641 -mno-fused-madd
9642 Generate code that uses (does not use) the floating point multiply
9643 and accumulate instructions. These instructions are generated by
9644 default if hardware floating point is used.
9645 -mwarn-framesize=framesize
9647 Emit a warning if the current function exceeds the given frame
9648 size. Because this is a compile time check it doesn't need to be a
9649 real problem when the program runs. It is intended to identify
9650 functions which most probably cause a stack overflow. It is useful
9651 to be used in an environment with limited stack size e.g. the linux
9652 kernel.
9653 -mwarn-dynamicstack
9655 Emit a warning if the function calls alloca or uses dynamically
9656 sized arrays. This is generally a bad idea with a limited stack
9657 size.
9658 -mstack-guard=stack-guard
9660 -mstack-size=stack-size
9661 These arguments always have to be used in conjunction. If they are
9662 present the s390 back end emits additional instructions in the
9663 function prologue which trigger a trap if the stack size is stack-
9664 guard bytes above the stack-size (remember that the stack on s390
9665 grows downward). These options are intended to be used to help
9666 debugging stack overflow problems. The additionally emitted code
9667 causes only little overhead and hence can also be used in produc‐
9668 tion like systems without greater performance degradation. The
9669 given values have to be exact powers of 2 and stack-size has to be
9670 greater than stack-guard without exceeding 64k. In order to be
9671 efficient the extra code makes the assumption that the stack starts
9672 at an address aligned to the value given by stack-size.
9673 SH Options
9675 These -m options are defined for the SH implementations:
9677 -m1 Generate code for the SH1.
9679 -m2 Generate code for the SH2.
9681 -m2e
9683 Generate code for the SH2e.
9684 -m3 Generate code for the SH3.
9686 -m3e
9688 Generate code for the SH3e.
9689 -m4-nofpu
9691 Generate code for the SH4 without a floating-point unit.
9692 -m4-single-only
9694 Generate code for the SH4 with a floating-point unit that only sup‐
9695 ports single-precision arithmetic.
9696 -m4-single
9698 Generate code for the SH4 assuming the floating-point unit is in
9699 single-precision mode by default.
9700 -m4 Generate code for the SH4.
9702 -m4a-nofpu
9704 Generate code for the SH4al-dsp, or for a SH4a in such a way that
9705 the floating-point unit is not used.
9706 -m4a-single-only
9708 Generate code for the SH4a, in such a way that no double-precision
9709 floating point operations are used.
9710 -m4a-single
9712 Generate code for the SH4a assuming the floating-point unit is in
9713 single-precision mode by default.
9714 -m4a
9716 Generate code for the SH4a.
9717 -m4al
9719 Same as -m4a-nofpu, except that it implicitly passes -dsp to the
9720 assembler. GCC doesn't generate any DSP instructions at the
9721 moment.
9722 -mb Compile code for the processor in big endian mode.
9724 -ml Compile code for the processor in little endian mode.
9726 -mdalign
9728 Align doubles at 64-bit boundaries. Note that this changes the
9729 calling conventions, and thus some functions from the standard C
9730 library will not work unless you recompile it first with -mdalign.
9731 -mrelax
9733 Shorten some address references at link time, when possible; uses
9734 the linker option -relax.
9735 -mbigtable
9737 Use 32-bit offsets in "switch" tables. The default is to use
9738 16-bit offsets.
9739 -mfmovd
9741 Enable the use of the instruction "fmovd".
9742 -mhitachi
9744 Comply with the calling conventions defined by Renesas.
9745 -mrenesas
9747 Comply with the calling conventions defined by Renesas.
9748 -mno-renesas
9750 Comply with the calling conventions defined for GCC before the
9751 Renesas conventions were available. This option is the default for
9752 all targets of the SH toolchain except for sh-symbianelf.
9753 -mnomacsave
9755 Mark the "MAC" register as call-clobbered, even if -mhitachi is
9756 given.
9757 -mieee
9759 Increase IEEE-compliance of floating-point code. At the moment,
9760 this is equivalent to -fno-finite-math-only. When generating 16
9761 bit SH opcodes, getting IEEE-conforming results for comparisons of
9762 NANs / infinities incurs extra overhead in every floating point
9763 comparison, therefore the default is set to -ffinite-math-only.
9764 -misize
9766 Dump instruction size and location in the assembly code.
9767 -mpadstruct
9769 This option is deprecated. It pads structures to multiple of 4
9770 bytes, which is incompatible with the SH ABI.
9771 -mspace
9773 Optimize for space instead of speed. Implied by -Os.
9774 -mprefergot
9776 When generating position-independent code, emit function calls
9777 using the Global Offset Table instead of the Procedure Linkage Ta‐
9778 ble.
9779 -musermode
9781 Generate a library function call to invalidate instruction cache
9782 entries, after fixing up a trampoline. This library function call
9783 doesn't assume it can write to the whole memory address space.
9784 This is the default when the target is "sh-*-linux*".
9785 -multcost=number
9787 Set the cost to assume for a multiply insn.
9788 -mdiv=strategy
9790 Set the division strategy to use for SHmedia code. strategy must
9791 be one of: call, call2, fp, inv, inv:minlat, inv20u, inv20l,
9792 inv:call, inv:call2, inv:fp . "fp" performs the operation in
9793 floating point. This has a very high latency, but needs only a few
9794 instructions, so it might be a good choice if your code has enough
9795 easily exploitable ILP to allow the compiler to schedule the float‐
9796 ing point instructions together with other instructions. Division
9797 by zero causes a floating point exception. "inv" uses integer
9798 operations to calculate the inverse of the divisor, and then multi‐
9799 plies the dividend with the inverse. This strategy allows cse and
9800 hoisting of the inverse calculation. Division by zero calculates
9801 an unspecified result, but does not trap. "inv:minlat" is a vari‐
9802 ant of "inv" where if no cse / hoisting opportunities have been
9803 found, or if the entire operation has been hoisted to the same
9804 place, the last stages of the inverse calculation are intertwined
9805 with the final multiply to reduce the overall latency, at the
9806 expense of using a few more instructions, and thus offering fewer
9807 scheduling opportunities with other code. "call" calls a library
9808 function that usually implements the inv:minlat strategy. This
9809 gives high code density for m5-*media-nofpu compilations. "call2"
9810 uses a different entry point of the same library function, where it
9811 assumes that a pointer to a lookup table has already been set up,
9812 which exposes the pointer load to cse / code hoisting optimiza‐
9813 tions. "inv:call", "inv:call2" and "inv:fp" all use the "inv"
9814 algorithm for initial code generation, but if the code stays unop‐
9815 timized, revert to the "call", "call2", or "fp" strategies, respec‐
9816 tively. Note that the potentially-trapping side effect of division
9817 by zero is carried by a separate instruction, so it is possible
9818 that all the integer instructions are hoisted out, but the marker
9819 for the side effect stays where it is. A recombination to fp oper‐
9820 ations or a call is not possible in that case. "inv20u" and
9821 "inv20l" are variants of the "inv:minlat" strategy. In the case
9822 that the inverse calculation was nor separated from the multiply,
9823 they speed up division where the dividend fits into 20 bits (plus
9824 sign where applicable), by inserting a test to skip a number of
9825 operations in this case; this test slows down the case of larger
9826 dividends. inv20u assumes the case of a such a small dividend to
9827 be unlikely, and inv20l assumes it to be likely.
9828 -mdivsi3_libfunc=name
9830 Set the name of the library function used for 32 bit signed divi‐
9831 sion to name. This only affect the name used in the call and
9832 inv:call division strategies, and the compiler will still expect
9833 the same sets of input/output/clobbered registers as if this option
9834 was not present.
9835 -madjust-unroll
9837 Throttle unrolling to avoid thrashing target registers. This
9838 option only has an effect if the gcc code base supports the TAR‐
9839 GET_ADJUST_UNROLL_MAX target hook.
9840 -mindexed-addressing
9842 Enable the use of the indexed addressing mode for SHmedia32/SHcom‐
9843 pact. This is only safe if the hardware and/or OS implement 32 bit
9844 wrap-around semantics for the indexed addressing mode. The archi‐
9845 tecture allows the implementation of processors with 64 bit MMU,
9846 which the OS could use to get 32 bit addressing, but since no cur‐
9847 rent hardware implementation supports this or any other way to make
9848 the indexed addressing mode safe to use in the 32 bit ABI, the
9849 default is -mno-indexed-addressing.
9850 -mgettrcost=number
9852 Set the cost assumed for the gettr instruction to number. The
9853 default is 2 if -mpt-fixed is in effect, 100 otherwise.
9854 -mpt-fixed
9856 Assume pt* instructions won't trap. This will generally generate
9857 better scheduled code, but is unsafe on current hardware. The cur‐
9858 rent architecture definition says that ptabs and ptrel trap when
9859 the target anded with 3 is 3. This has the unintentional effect of
9860 making it unsafe to schedule ptabs / ptrel before a branch, or
9861 hoist it out of a loop. For example, __do_global_ctors, a part of
9862 libgcc that runs constructors at program startup, calls functions
9863 in a list which is delimited by -1. With the -mpt-fixed option,
9864 the ptabs will be done before testing against -1. That means that
9865 all the constructors will be run a bit quicker, but when the loop
9866 comes to the end of the list, the program crashes because ptabs
9867 loads -1 into a target register. Since this option is unsafe for
9868 any hardware implementing the current architecture specification,
9869 the default is -mno-pt-fixed. Unless the user specifies a specific
9870 cost with -mgettrcost, -mno-pt-fixed also implies -mgettrcost=100;
9871 this deters register allocation using target registers for storing
9872 ordinary integers.
9873 -minvalid-symbols
9875 Assume symbols might be invalid. Ordinary function symbols gener‐
9876 ated by the compiler will always be valid to load with
9877 movi/shori/ptabs or movi/shori/ptrel, but with assembler and/or
9878 linker tricks it is possible to generate symbols that will cause
9879 ptabs / ptrel to trap. This option is only meaningful when
9880 -mno-pt-fixed is in effect. It will then prevent cross-basic-block
9881 cse, hoisting and most scheduling of symbol loads. The default is
9882 -mno-invalid-symbols.
9883 SPARC Options
9885 These -m options are supported on the SPARC:
9887 -mno-app-regs
9889 -mapp-regs
9890 Specify -mapp-regs to generate output using the global registers 2
9891 through 4, which the SPARC SVR4 ABI reserves for applications.
9892 This is the default.
9893 To be fully SVR4 ABI compliant at the cost of some performance
9895 loss, specify -mno-app-regs. You should compile libraries and sys‐
9896 tem software with this option.
9897 -mfpu
9899 -mhard-float
9900 Generate output containing floating point instructions. This is
9901 the default.
9902 -mno-fpu
9904 -msoft-float
9905 Generate output containing library calls for floating point. Warn‐
9906 ing: the requisite libraries are not available for all SPARC tar‐
9907 gets. Normally the facilities of the machine's usual C compiler
9908 are used, but this cannot be done directly in cross-compilation.
9909 You must make your own arrangements to provide suitable library
9910 functions for cross-compilation. The embedded targets sparc-*-aout
9911 and sparclite-*-* do provide software floating point support.
9912 -msoft-float changes the calling convention in the output file;
9914 therefore, it is only useful if you compile all of a program with
9915 this option. In particular, you need to compile libgcc.a, the
9916 library that comes with GCC, with -msoft-float in order for this to
9917 work.
9918 -mhard-quad-float
9920 Generate output containing quad-word (long double) floating point
9921 instructions.
9922 -msoft-quad-float
9924 Generate output containing library calls for quad-word (long dou‐
9925 ble) floating point instructions. The functions called are those
9926 specified in the SPARC ABI. This is the default.
9927 As of this writing, there are no SPARC implementations that have
9929 hardware support for the quad-word floating point instructions.
9930 They all invoke a trap handler for one of these instructions, and
9931 then the trap handler emulates the effect of the instruction.
9932 Because of the trap handler overhead, this is much slower than
9933 calling the ABI library routines. Thus the -msoft-quad-float
9934 option is the default.
9935 -mno-unaligned-doubles
9937 -munaligned-doubles
9938 Assume that doubles have 8 byte alignment. This is the default.
9939 With -munaligned-doubles, GCC assumes that doubles have 8 byte
9941 alignment only if they are contained in another type, or if they
9942 have an absolute address. Otherwise, it assumes they have 4 byte
9943 alignment. Specifying this option avoids some rare compatibility
9944 problems with code generated by other compilers. It is not the
9945 default because it results in a performance loss, especially for
9946 floating point code.
9947 -mno-faster-structs
9949 -mfaster-structs
9950 With -mfaster-structs, the compiler assumes that structures should
9951 have 8 byte alignment. This enables the use of pairs of "ldd" and
9952 "std" instructions for copies in structure assignment, in place of
9953 twice as many "ld" and "st" pairs. However, the use of this
9954 changed alignment directly violates the SPARC ABI. Thus, it's
9955 intended only for use on targets where the developer acknowledges
9956 that their resulting code will not be directly in line with the
9957 rules of the ABI.
9958 -mimpure-text
9960 -mimpure-text, used in addition to -shared, tells the compiler to
9961 not pass -z text to the linker when linking a shared object. Using
9962 this option, you can link position-dependent code into a shared
9963 object.
9964 -mimpure-text suppresses the "relocations remain against allocat‐
9966 able but non-writable sections" linker error message. However, the
9967 necessary relocations will trigger copy-on-write, and the shared
9968 object is not actually shared across processes. Instead of using
9969 -mimpure-text, you should compile all source code with -fpic or
9970 -fPIC.
9971 This option is only available on SunOS and Solaris.
9973 -mcpu=cpu_type
9975 Set the instruction set, register set, and instruction scheduling
9976 parameters for machine type cpu_type. Supported values for
9977 cpu_type are v7, cypress, v8, supersparc, sparclite, f930, f934,
9978 hypersparc, sparclite86x, sparclet, tsc701, v9, ultrasparc, and
9979 ultrasparc3.
9980 Default instruction scheduling parameters are used for values that
9982 select an architecture and not an implementation. These are v7,
9983 v8, sparclite, sparclet, v9.
9984 Here is a list of each supported architecture and their supported
9986 implementations.
9987 v7: cypress
9989 v8: supersparc, hypersparc
9990 sparclite: f930, f934, sparclite86x
9991 sparclet: tsc701
9992 v9: ultrasparc, ultrasparc3
9993 By default (unless configured otherwise), GCC generates code for
9995 the V7 variant of the SPARC architecture. With -mcpu=cypress, the
9996 compiler additionally optimizes it for the Cypress CY7C602 chip, as
9997 used in the SPARCStation/SPARCServer 3xx series. This is also
9998 appropriate for the older SPARCStation 1, 2, IPX etc.
9999 With -mcpu=v8, GCC generates code for the V8 variant of the SPARC
10001 architecture. The only difference from V7 code is that the com‐
10002 piler emits the integer multiply and integer divide instructions
10003 which exist in SPARC-V8 but not in SPARC-V7. With -mcpu=super‐
10004 sparc, the compiler additionally optimizes it for the SuperSPARC
10005 chip, as used in the SPARCStation 10, 1000 and 2000 series.
10006 With -mcpu=sparclite, GCC generates code for the SPARClite variant
10008 of the SPARC architecture. This adds the integer multiply, integer
10009 divide step and scan ("ffs") instructions which exist in SPARClite
10010 but not in SPARC-V7. With -mcpu=f930, the compiler additionally
10011 optimizes it for the Fujitsu MB86930 chip, which is the original
10012 SPARClite, with no FPU. With -mcpu=f934, the compiler additionally
10013 optimizes it for the Fujitsu MB86934 chip, which is the more recent
10014 SPARClite with FPU.
10015 With -mcpu=sparclet, GCC generates code for the SPARClet variant of
10017 the SPARC architecture. This adds the integer multiply, multi‐
10018 ply/accumulate, integer divide step and scan ("ffs") instructions
10019 which exist in SPARClet but not in SPARC-V7. With -mcpu=tsc701,
10020 the compiler additionally optimizes it for the TEMIC SPARClet chip.
10021 With -mcpu=v9, GCC generates code for the V9 variant of the SPARC
10023 architecture. This adds 64-bit integer and floating-point move
10024 instructions, 3 additional floating-point condition code registers
10025 and conditional move instructions. With -mcpu=ultrasparc, the com‐
10026 piler additionally optimizes it for the Sun UltraSPARC I/II chips.
10027 With -mcpu=ultrasparc3, the compiler additionally optimizes it for
10028 the Sun UltraSPARC III chip.
10029 -mtune=cpu_type
10031 Set the instruction scheduling parameters for machine type
10032 cpu_type, but do not set the instruction set or register set that
10033 the option -mcpu=cpu_type would.
10034 The same values for -mcpu=cpu_type can be used for -mtune=cpu_type,
10036 but the only useful values are those that select a particular cpu
10037 implementation. Those are cypress, supersparc, hypersparc, f930,
10038 f934, sparclite86x, tsc701, ultrasparc, and ultrasparc3.
10039 -mv8plus
10041 -mno-v8plus
10042 With -mv8plus, GCC generates code for the SPARC-V8+ ABI. The dif‐
10043 ference from the V8 ABI is that the global and out registers are
10044 considered 64-bit wide. This is enabled by default on Solaris in
10045 32-bit mode for all SPARC-V9 processors.
10046 -mvis
10048 -mno-vis
10049 With -mvis, GCC generates code that takes advantage of the Ultra‐
10050 SPARC Visual Instruction Set extensions. The default is -mno-vis.
10051 These -m options are supported in addition to the above on SPARC-V9
10053 processors in 64-bit environments:
10054 -mlittle-endian
10056 Generate code for a processor running in little-endian mode. It is
10057 only available for a few configurations and most notably not on
10058 Solaris and Linux.
10059 -m32
10061 -m64
10062 Generate code for a 32-bit or 64-bit environment. The 32-bit envi‐
10063 ronment sets int, long and pointer to 32 bits. The 64-bit environ‐
10064 ment sets int to 32 bits and long and pointer to 64 bits.
10065 -mcmodel=medlow
10067 Generate code for the Medium/Low code model: 64-bit addresses, pro‐
10068 grams must be linked in the low 32 bits of memory. Programs can be
10069 statically or dynamically linked.
10070 -mcmodel=medmid
10072 Generate code for the Medium/Middle code model: 64-bit addresses,
10073 programs must be linked in the low 44 bits of memory, the text and
10074 data segments must be less than 2GB in size and the data segment
10075 must be located within 2GB of the text segment.
10076 -mcmodel=medany
10078 Generate code for the Medium/Anywhere code model: 64-bit addresses,
10079 programs may be linked anywhere in memory, the text and data seg‐
10080 ments must be less than 2GB in size and the data segment must be
10081 located within 2GB of the text segment.
10082 -mcmodel=embmedany
10084 Generate code for the Medium/Anywhere code model for embedded sys‐
10085 tems: 64-bit addresses, the text and data segments must be less
10086 than 2GB in size, both starting anywhere in memory (determined at
10087 link time). The global register %g4 points to the base of the data
10088 segment. Programs are statically linked and PIC is not supported.
10089 -mstack-bias
10091 -mno-stack-bias
10092 With -mstack-bias, GCC assumes that the stack pointer, and frame
10093 pointer if present, are offset by -2047 which must be added back
10094 when making stack frame references. This is the default in 64-bit
10095 mode. Otherwise, assume no such offset is present.
10096 These switches are supported in addition to the above on Solaris:
10098 -threads
10100 Add support for multithreading using the Solaris threads library.
10101 This option sets flags for both the preprocessor and linker. This
10102 option does not affect the thread safety of object code produced by
10103 the compiler or that of libraries supplied with it.
10104 -pthreads
10106 Add support for multithreading using the POSIX threads library.
10107 This option sets flags for both the preprocessor and linker. This
10108 option does not affect the thread safety of object code produced
10109 by the compiler or that of libraries supplied with it.
10110 -pthread
10112 This is a synonym for -pthreads.
10113 Options for System V
10115 These additional options are available on System V Release 4 for com‐
10117 patibility with other compilers on those systems:
10118 -G Create a shared object. It is recommended that -symbolic or
10120 -shared be used instead.
10121 -Qy Identify the versions of each tool used by the compiler, in a
10123 ".ident" assembler directive in the output.
10124 -Qn Refrain from adding ".ident" directives to the output file (this is
10126 the default).
10127 -YP,dirs
10129 Search the directories dirs, and no others, for libraries specified
10130 with -l.
10131 -Ym,dir
10133 Look in the directory dir to find the M4 preprocessor. The assem‐
10134 bler uses this option.
10135 TMS320C3x/C4x Options
10137 These -m options are defined for TMS320C3x/C4x implementations:
10139 -mcpu=cpu_type
10141 Set the instruction set, register set, and instruction scheduling
10142 parameters for machine type cpu_type. Supported values for
10143 cpu_type are c30, c31, c32, c40, and c44. The default is c40 to
10144 generate code for the TMS320C40.
10145 -mbig-memory
10147 -mbig
10148 -msmall-memory
10149 -msmall
10150 Generates code for the big or small memory model. The small memory
10151 model assumed that all data fits into one 64K word page. At run-
10152 time the data page (DP) register must be set to point to the 64K
10153 page containing the .bss and .data program sections. The big mem‐
10154 ory model is the default and requires reloading of the DP register
10155 for every direct memory access.
10156 -mbk
10158 -mno-bk
10159 Allow (disallow) allocation of general integer operands into the
10160 block count register BK.
10161 -mdb
10163 -mno-db
10164 Enable (disable) generation of code using decrement and branch,
10165 DBcond(D), instructions. This is enabled by default for the C4x.
10166 To be on the safe side, this is disabled for the C3x, since the
10167 maximum iteration count on the C3x is 2^{23 + 1} (but who iterates
10168 loops more than 2^{23} times on the C3x?). Note that GCC will try
10169 to reverse a loop so that it can utilize the decrement and branch
10170 instruction, but will give up if there is more than one memory ref‐
10171 erence in the loop. Thus a loop where the loop counter is decre‐
10172 mented can generate slightly more efficient code, in cases where
10173 the RPTB instruction cannot be utilized.
10174 -mdp-isr-reload
10176 -mparanoid
10177 Force the DP register to be saved on entry to an interrupt service
10178 routine (ISR), reloaded to point to the data section, and restored
10179 on exit from the ISR. This should not be required unless someone
10180 has violated the small memory model by modifying the DP register,
10181 say within an object library.
10182 -mmpyi
10184 -mno-mpyi
10185 For the C3x use the 24-bit MPYI instruction for integer multiplies
10186 instead of a library call to guarantee 32-bit results. Note that
10187 if one of the operands is a constant, then the multiplication will
10188 be performed using shifts and adds. If the -mmpyi option is not
10189 specified for the C3x, then squaring operations are performed
10190 inline instead of a library call.
10191 -mfast-fix
10193 -mno-fast-fix
10194 The C3x/C4x FIX instruction to convert a floating point value to an
10195 integer value chooses the nearest integer less than or equal to the
10196 floating point value rather than to the nearest integer. Thus if
10197 the floating point number is negative, the result will be incor‐
10198 rectly truncated an additional code is necessary to detect and cor‐
10199 rect this case. This option can be used to disable generation of
10200 the additional code required to correct the result.
10201 -mrptb
10203 -mno-rptb
10204 Enable (disable) generation of repeat block sequences using the
10205 RPTB instruction for zero overhead looping. The RPTB construct is
10206 only used for innermost loops that do not call functions or jump
10207 across the loop boundaries. There is no advantage having nested
10208 RPTB loops due to the overhead required to save and restore the RC,
10209 RS, and RE registers. This is enabled by default with -O2.
10210 -mrpts=count
10212 -mno-rpts
10213 Enable (disable) the use of the single instruction repeat instruc‐
10214 tion RPTS. If a repeat block contains a single instruction, and
10215 the loop count can be guaranteed to be less than the value count,
10216 GCC will emit a RPTS instruction instead of a RPTB. If no value is
10217 specified, then a RPTS will be emitted even if the loop count can‐
10218 not be determined at compile time. Note that the repeated instruc‐
10219 tion following RPTS does not have to be reloaded from memory each
10220 iteration, thus freeing up the CPU buses for operands. However,
10221 since interrupts are blocked by this instruction, it is disabled by
10222 default.
10223 -mloop-unsigned
10225 -mno-loop-unsigned
10226 The maximum iteration count when using RPTS and RPTB (and DB on the
10227 C40) is 2^{31 + 1} since these instructions test if the iteration
10228 count is negative to terminate the loop. If the iteration count is
10229 unsigned there is a possibility than the 2^{31 + 1} maximum itera‐
10230 tion count may be exceeded. This switch allows an unsigned itera‐
10231 tion count.
10232 -mti
10234 Try to emit an assembler syntax that the TI assembler (asm30) is
10235 happy with. This also enforces compatibility with the API employed
10236 by the TI C3x C compiler. For example, long doubles are passed as
10237 structures rather than in floating point registers.
10238 -mregparm
10240 -mmemparm
10241 Generate code that uses registers (stack) for passing arguments to
10242 functions. By default, arguments are passed in registers where
10243 possible rather than by pushing arguments on to the stack.
10244 -mparallel-insns
10246 -mno-parallel-insns
10247 Allow the generation of parallel instructions. This is enabled by
10248 default with -O2.
10249 -mparallel-mpy
10251 -mno-parallel-mpy
10252 Allow the generation of MPY⎪⎪ADD and MPY⎪⎪SUB parallel instruc‐
10253 tions, provided -mparallel-insns is also specified. These instruc‐
10254 tions have tight register constraints which can pessimize the code
10255 generation of large functions.
10256 V850 Options
10258 These -m options are defined for V850 implementations:
10260 -mlong-calls
10262 -mno-long-calls
10263 Treat all calls as being far away (near). If calls are assumed to
10264 be far away, the compiler will always load the functions address up
10265 into a register, and call indirect through the pointer.
10266 -mno-ep
10268 -mep
10269 Do not optimize (do optimize) basic blocks that use the same index
10270 pointer 4 or more times to copy pointer into the "ep" register, and
10271 use the shorter "sld" and "sst" instructions. The -mep option is
10272 on by default if you optimize.
10273 -mno-prolog-function
10275 -mprolog-function
10276 Do not use (do use) external functions to save and restore regis‐
10277 ters at the prologue and epilogue of a function. The external
10278 functions are slower, but use less code space if more than one
10279 function saves the same number of registers. The -mprolog-function
10280 option is on by default if you optimize.
10281 -mspace
10283 Try to make the code as small as possible. At present, this just
10284 turns on the -mep and -mprolog-function options.
10285 -mtda=n
10287 Put static or global variables whose size is n bytes or less into
10288 the tiny data area that register "ep" points to. The tiny data
10289 area can hold up to 256 bytes in total (128 bytes for byte refer‐
10290 ences).
10291 -msda=n
10293 Put static or global variables whose size is n bytes or less into
10294 the small data area that register "gp" points to. The small data
10295 area can hold up to 64 kilobytes.
10296 -mzda=n
10298 Put static or global variables whose size is n bytes or less into
10299 the first 32 kilobytes of memory.
10300 -mv850
10302 Specify that the target processor is the V850.
10303 -mbig-switch
10305 Generate code suitable for big switch tables. Use this option only
10306 if the assembler/linker complain about out of range branches within
10307 a switch table.
10308 -mapp-regs
10310 This option will cause r2 and r5 to be used in the code generated
10311 by the compiler. This setting is the default.
10312 -mno-app-regs
10314 This option will cause r2 and r5 to be treated as fixed registers.
10315 -mv850e1
10317 Specify that the target processor is the V850E1. The preprocessor
10318 constants __v850e1__ and __v850e__ will be defined if this option
10319 is used.
10320 -mv850e
10322 Specify that the target processor is the V850E. The preprocessor
10323 constant __v850e__ will be defined if this option is used.
10324 If neither -mv850 nor -mv850e nor -mv850e1 are defined then a
10326 default target processor will be chosen and the relevant __v850*__
10327 preprocessor constant will be defined.
10328 The preprocessor constants __v850 and __v851__ are always defined,
10330 regardless of which processor variant is the target.
10331 -mdisable-callt
10333 This option will suppress generation of the CALLT instruction for
10334 the v850e and v850e1 flavors of the v850 architecture. The default
10335 is -mno-disable-callt which allows the CALLT instruction to be
10336 used.
10337 VAX Options
10339 These -m options are defined for the VAX:
10341 -munix
10343 Do not output certain jump instructions ("aobleq" and so on) that
10344 the Unix assembler for the VAX cannot handle across long ranges.
10345 -mgnu
10347 Do output those jump instructions, on the assumption that you will
10348 assemble with the GNU assembler.
10349 -mg Output code for g-format floating point numbers instead of d-for‐
10351 mat.
10352 x86-64 Options
10354 These are listed under
10356 Xstormy16 Options
10358 These options are defined for Xstormy16:
10360 -msim
10362 Choose startup files and linker script suitable for the simulator.
10363 Xtensa Options
10365 These options are supported for Xtensa targets:
10367 -mconst16
10369 -mno-const16
10370 Enable or disable use of "CONST16" instructions for loading con‐
10371 stant values. The "CONST16" instruction is currently not a stan‐
10372 dard option from Tensilica. When enabled, "CONST16" instructions
10373 are always used in place of the standard "L32R" instructions. The
10374 use of "CONST16" is enabled by default only if the "L32R" instruc‐
10375 tion is not available.
10376 -mfused-madd
10378 -mno-fused-madd
10379 Enable or disable use of fused multiply/add and multiply/subtract
10380 instructions in the floating-point option. This has no effect if
10381 the floating-point option is not also enabled. Disabling fused
10382 multiply/add and multiply/subtract instructions forces the compiler
10383 to use separate instructions for the multiply and add/subtract
10384 operations. This may be desirable in some cases where strict IEEE
10385 754-compliant results are required: the fused multiply add/subtract
10386 instructions do not round the intermediate result, thereby produc‐
10387 ing results with more bits of precision than specified by the IEEE
10388 standard. Disabling fused multiply add/subtract instructions also
10389 ensures that the program output is not sensitive to the compiler's
10390 ability to combine multiply and add/subtract operations.
10391 -mtext-section-literals
10393 -mno-text-section-literals
10394 Control the treatment of literal pools. The default is
10395 -mno-text-section-literals, which places literals in a separate
10396 section in the output file. This allows the literal pool to be
10397 placed in a data RAM/ROM, and it also allows the linker to combine
10398 literal pools from separate object files to remove redundant liter‐
10399 als and improve code size. With -mtext-section-literals, the lit‐
10400 erals are interspersed in the text section in order to keep them as
10401 close as possible to their references. This may be necessary for
10402 large assembly files.
10403 -mtarget-align
10405 -mno-target-align
10406 When this option is enabled, GCC instructs the assembler to auto‐
10407 matically align instructions to reduce branch penalties at the
10408 expense of some code density. The assembler attempts to widen den‐
10409 sity instructions to align branch targets and the instructions fol‐
10410 lowing call instructions. If there are not enough preceding safe
10411 density instructions to align a target, no widening will be per‐
10412 formed. The default is -mtarget-align. These options do not
10413 affect the treatment of auto-aligned instructions like "LOOP",
10414 which the assembler will always align, either by widening density
10415 instructions or by inserting no-op instructions.
10416 -mlongcalls
10418 -mno-longcalls
10419 When this option is enabled, GCC instructs the assembler to trans‐
10420 late direct calls to indirect calls unless it can determine that
10421 the target of a direct call is in the range allowed by the call
10422 instruction. This translation typically occurs for calls to func‐
10423 tions in other source files. Specifically, the assembler trans‐
10424 lates a direct "CALL" instruction into an "L32R" followed by a
10425 "CALLX" instruction. The default is -mno-longcalls. This option
10426 should be used in programs where the call target can potentially be
10427 out of range. This option is implemented in the assembler, not the
10428 compiler, so the assembly code generated by GCC will still show
10429 direct call instructions---look at the disassembled object code to
10430 see the actual instructions. Note that the assembler will use an
10431 indirect call for every cross-file call, not just those that really
10432 will be out of range.
10433 zSeries Options
10435 These are listed under
10437 Options for Code Generation Conventions
10439 These machine-independent options control the interface conventions
10441 used in code generation.
10442 Most of them have both positive and negative forms; the negative form
10444 of -ffoo would be -fno-foo. In the table below, only one of the forms
10445 is listed---the one which is not the default. You can figure out the
10446 other form by either removing no- or adding it.
10447 -fbounds-check
10449 For front-ends that support it, generate additional code to check
10450 that indices used to access arrays are within the declared range.
10451 This is currently only supported by the Java and Fortran
10452 front-ends, where this option defaults to true and false respec‐
10453 tively.
10454 -ftrapv
10456 This option generates traps for signed overflow on addition, sub‐
10457 traction, multiplication operations.
10458 -fwrapv
10460 This option instructs the compiler to assume that signed arithmetic
10461 overflow of addition, subtraction and multiplication wraps around
10462 using twos-complement representation. This flag enables some opti‐
10463 mizations and disables others. This option is enabled by default
10464 for the Java front-end, as required by the Java language specifica‐
10465 tion.
10466 -fexceptions
10468 Enable exception handling. Generates extra code needed to propa‐
10469 gate exceptions. For some targets, this implies GCC will generate
10470 frame unwind information for all functions, which can produce sig‐
10471 nificant data size overhead, although it does not affect execution.
10472 If you do not specify this option, GCC will enable it by default
10473 for languages like C++ which normally require exception handling,
10474 and disable it for languages like C that do not normally require
10475 it. However, you may need to enable this option when compiling C
10476 code that needs to interoperate properly with exception handlers
10477 written in C++. You may also wish to disable this option if you
10478 are compiling older C++ programs that don't use exception handling.
10479 -fnon-call-exceptions
10481 Generate code that allows trapping instructions to throw excep‐
10482 tions. Note that this requires platform-specific runtime support
10483 that does not exist everywhere. Moreover, it only allows trapping
10484 instructions to throw exceptions, i.e. memory references or float‐
10485 ing point instructions. It does not allow exceptions to be thrown
10486 from arbitrary signal handlers such as "SIGALRM".
10487 -funwind-tables
10489 Similar to -fexceptions, except that it will just generate any
10490 needed static data, but will not affect the generated code in any
10491 other way. You will normally not enable this option; instead, a
10492 language processor that needs this handling would enable it on your
10493 behalf.
10494 -fasynchronous-unwind-tables
10496 Generate unwind table in dwarf2 format, if supported by target
10497 machine. The table is exact at each instruction boundary, so it
10498 can be used for stack unwinding from asynchronous events (such as
10499 debugger or garbage collector).
10500 -fpcc-struct-return
10502 Return "short" "struct" and "union" values in memory like longer
10503 ones, rather than in registers. This convention is less efficient,
10504 but it has the advantage of allowing intercallability between GCC-
10505 compiled files and files compiled with other compilers, particu‐
10506 larly the Portable C Compiler (pcc).
10507 The precise convention for returning structures in memory depends
10509 on the target configuration macros.
10510 Short structures and unions are those whose size and alignment
10512 match that of some integer type.
10513 Warning: code compiled with the -fpcc-struct-return switch is not
10515 binary compatible with code compiled with the -freg-struct-return
10516 switch. Use it to conform to a non-default application binary
10517 interface.
10518 -freg-struct-return
10520 Return "struct" and "union" values in registers when possible.
10521 This is more efficient for small structures than
10522 -fpcc-struct-return.
10523 If you specify neither -fpcc-struct-return nor -freg-struct-return,
10525 GCC defaults to whichever convention is standard for the target.
10526 If there is no standard convention, GCC defaults to
10527 -fpcc-struct-return, except on targets where GCC is the principal
10528 compiler. In those cases, we can choose the standard, and we chose
10529 the more efficient register return alternative.
10530 Warning: code compiled with the -freg-struct-return switch is not
10532 binary compatible with code compiled with the -fpcc-struct-return
10533 switch. Use it to conform to a non-default application binary
10534 interface.
10535 -fshort-enums
10537 Allocate to an "enum" type only as many bytes as it needs for the
10538 declared range of possible values. Specifically, the "enum" type
10539 will be equivalent to the smallest integer type which has enough
10540 room.
10541 Warning: the -fshort-enums switch causes GCC to generate code that
10543 is not binary compatible with code generated without that switch.
10544 Use it to conform to a non-default application binary interface.
10545 -fshort-double
10547 Use the same size for "double" as for "float".
10548 Warning: the -fshort-double switch causes GCC to generate code that
10550 is not binary compatible with code generated without that switch.
10551 Use it to conform to a non-default application binary interface.
10552 -fshort-wchar
10554 Override the underlying type for wchar_t to be short unsigned int
10555 instead of the default for the target. This option is useful for
10556 building programs to run under WINE.
10557 Warning: the -fshort-wchar switch causes GCC to generate code that
10559 is not binary compatible with code generated without that switch.
10560 Use it to conform to a non-default application binary interface.
10561 -fshared-data
10563 Requests that the data and non-"const" variables of this compila‐
10564 tion be shared data rather than private data. The distinction
10565 makes sense only on certain operating systems, where shared data is
10566 shared between processes running the same program, while private
10567 data exists in one copy per process.
10568 -fno-common
10570 In C, allocate even uninitialized global variables in the data sec‐
10571 tion of the object file, rather than generating them as common
10572 blocks. This has the effect that if the same variable is declared
10573 (without "extern") in two different compilations, you will get an
10574 error when you link them. The only reason this might be useful is
10575 if you wish to verify that the program will work on other systems
10576 which always work this way.
10577 -fno-ident
10579 Ignore the #ident directive.
10580 -finhibit-size-directive
10582 Don't output a ".size" assembler directive, or anything else that
10583 would cause trouble if the function is split in the middle, and the
10584 two halves are placed at locations far apart in memory. This
10585 option is used when compiling crtstuff.c; you should not need to
10586 use it for anything else.
10587 -fverbose-asm
10589 Put extra commentary information in the generated assembly code to
10590 make it more readable. This option is generally only of use to
10591 those who actually need to read the generated assembly code (per‐
10592 haps while debugging the compiler itself).
10593 -fno-verbose-asm, the default, causes the extra information to be
10595 omitted and is useful when comparing two assembler files.
10596 -fpic
10598 Generate position-independent code (PIC) suitable for use in a
10599 shared library, if supported for the target machine. Such code
10600 accesses all constant addresses through a global offset table
10601 (GOT). The dynamic loader resolves the GOT entries when the pro‐
10602 gram starts (the dynamic loader is not part of GCC; it is part of
10603 the operating system). If the GOT size for the linked executable
10604 exceeds a machine-specific maximum size, you get an error message
10605 from the linker indicating that -fpic does not work; in that case,
10606 recompile with -fPIC instead. (These maximums are 8k on the SPARC
10607 and 32k on the m68k and RS/6000. The 386 has no such limit.)
10608 Position-independent code requires special support, and therefore
10610 works only on certain machines. For the 386, GCC supports PIC for
10611 System V but not for the Sun 386i. Code generated for the IBM
10612 RS/6000 is always position-independent.
10613 -fPIC
10615 If supported for the target machine, emit position-independent
10616 code, suitable for dynamic linking and avoiding any limit on the
10617 size of the global offset table. This option makes a difference on
10618 the m68k, PowerPC and SPARC.
10619 Position-independent code requires special support, and therefore
10621 works only on certain machines.
10622 -fpie
10624 -fPIE
10625 These options are similar to -fpic and -fPIC, but generated posi‐
10626 tion independent code can be only linked into executables. Usually
10627 these options are used when -pie GCC option will be used during
10628 linking.
10629 -fno-jump-tables
10631 Do not use jump tables for switch statements even where it would be
10632 more efficient than other code generation strategies. This option
10633 is of use in conjunction with -fpic or -fPIC for building code
10634 which forms part of a dynamic linker and cannot reference the
10635 address of a jump table. On some targets, jump tables do not
10636 require a GOT and this option is not needed.
10637 -ffixed-reg
10639 Treat the register named reg as a fixed register; generated code
10640 should never refer to it (except perhaps as a stack pointer, frame
10641 pointer or in some other fixed role).
10642 reg must be the name of a register. The register names accepted
10644 are machine-specific and are defined in the "REGISTER_NAMES" macro
10645 in the machine description macro file.
10646 This flag does not have a negative form, because it specifies a
10648 three-way choice.
10649 -fcall-used-reg
10651 Treat the register named reg as an allocable register that is clob‐
10652 bered by function calls. It may be allocated for temporaries or
10653 variables that do not live across a call. Functions compiled this
10654 way will not save and restore the register reg.
10655 It is an error to used this flag with the frame pointer or stack
10657 pointer. Use of this flag for other registers that have fixed per‐
10658 vasive roles in the machine's execution model will produce disas‐
10659 trous results.
10660 This flag does not have a negative form, because it specifies a
10662 three-way choice.
10663 -fcall-saved-reg
10665 Treat the register named reg as an allocable register saved by
10666 functions. It may be allocated even for temporaries or variables
10667 that live across a call. Functions compiled this way will save and
10668 restore the register reg if they use it.
10669 It is an error to used this flag with the frame pointer or stack
10671 pointer. Use of this flag for other registers that have fixed per‐
10672 vasive roles in the machine's execution model will produce disas‐
10673 trous results.
10674 A different sort of disaster will result from the use of this flag
10676 for a register in which function values may be returned.
10677 This flag does not have a negative form, because it specifies a
10679 three-way choice.
10680 -fpack-struct[=n]
10682 Without a value specified, pack all structure members together
10683 without holes. When a value is specified (which must be a small
10684 power of two), pack structure members according to this value, rep‐
10685 resenting the maximum alignment (that is, objects with default
10686 alignment requirements larger than this will be output potentially
10687 unaligned at the next fitting location.
10688 Warning: the -fpack-struct switch causes GCC to generate code that
10690 is not binary compatible with code generated without that switch.
10691 Additionally, it makes the code suboptimal. Use it to conform to a
10692 non-default application binary interface.
10693 -finstrument-functions
10695 Generate instrumentation calls for entry and exit to functions.
10696 Just after function entry and just before function exit, the fol‐
10697 lowing profiling functions will be called with the address of the
10698 current function and its call site. (On some platforms,
10699 "__builtin_return_address" does not work beyond the current func‐
10700 tion, so the call site information may not be available to the pro‐
10701 filing functions otherwise.)
10702 void __cyg_profile_func_enter (void *this_fn,
10704 void *call_site);
10705 void __cyg_profile_func_exit (void *this_fn,
10706 void *call_site);
10707 The first argument is the address of the start of the current func‐
10709 tion, which may be looked up exactly in the symbol table.
10710 This instrumentation is also done for functions expanded inline in
10712 other functions. The profiling calls will indicate where, concep‐
10713 tually, the inline function is entered and exited. This means that
10714 addressable versions of such functions must be available. If all
10715 your uses of a function are expanded inline, this may mean an addi‐
10716 tional expansion of code size. If you use extern inline in your C
10717 code, an addressable version of such functions must be provided.
10718 (This is normally the case anyways, but if you get lucky and the
10719 optimizer always expands the functions inline, you might have got‐
10720 ten away without providing static copies.)
10721 A function may be given the attribute "no_instrument_function", in
10723 which case this instrumentation will not be done. This can be
10724 used, for example, for the profiling functions listed above, high-
10725 priority interrupt routines, and any functions from which the pro‐
10726 filing functions cannot safely be called (perhaps signal handlers,
10727 if the profiling routines generate output or allocate memory).
10728 -fstack-check
10730 Generate code to verify that you do not go beyond the boundary of
10731 the stack. You should specify this flag if you are running in an
10732 environment with multiple threads, but only rarely need to specify
10733 it in a single-threaded environment since stack overflow is auto‐
10734 matically detected on nearly all systems if there is only one
10735 stack.
10736 Note that this switch does not actually cause checking to be done;
10738 the operating system must do that. The switch causes generation of
10739 code to ensure that the operating system sees the stack being
10740 extended.
10741 -fstack-limit-register=reg
10743 -fstack-limit-symbol=sym
10744 -fno-stack-limit
10745 Generate code to ensure that the stack does not grow beyond a cer‐
10746 tain value, either the value of a register or the address of a sym‐
10747 bol. If the stack would grow beyond the value, a signal is raised.
10748 For most targets, the signal is raised before the stack overruns
10749 the boundary, so it is possible to catch the signal without taking
10750 special precautions.
10751 For instance, if the stack starts at absolute address 0x80000000
10753 and grows downwards, you can use the flags -fstack-limit-sym‐
10754 bol=__stack_limit and -Wl,--defsym,__stack_limit=0x7ffe0000 to
10755 enforce a stack limit of 128KB. Note that this may only work with
10756 the GNU linker.
10757 -fargument-alias
10759 -fargument-noalias
10760 -fargument-noalias-global
10761 Specify the possible relationships among parameters and between
10762 parameters and global data.
10763 -fargument-alias specifies that arguments (parameters) may alias
10765 each other and may alias global storage.-fargument-noalias speci‐
10766 fies that arguments do not alias each other, but may alias global
10767 storage.-fargument-noalias-global specifies that arguments do not
10768 alias each other and do not alias global storage.
10769 Each language will automatically use whatever option is required by
10771 the language standard. You should not need to use these options
10772 yourself.
10773 -fleading-underscore
10775 This option and its counterpart, -fno-leading-underscore, forcibly
10776 change the way C symbols are represented in the object file. One
10777 use is to help link with legacy assembly code.
10778 Warning: the -fleading-underscore switch causes GCC to generate
10780 code that is not binary compatible with code generated without that
10781 switch. Use it to conform to a non-default application binary
10782 interface. Not all targets provide complete support for this
10783 switch.
10784 -ftls-model=model
10786 Alter the thread-local storage model to be used. The model argu‐
10787 ment should be one of "global-dynamic", "local-dynamic", "ini‐
10788 tial-exec" or "local-exec".
10789 The default without -fpic is "initial-exec"; with -fpic the default
10791 is "global-dynamic".
10792 -fvisibility=default⎪internal⎪hidden⎪protected
10794 Set the default ELF image symbol visibility to the specified
10795 option---all symbols will be marked with this unless overridden
10796 within the code. Using this feature can very substantially improve
10797 linking and load times of shared object libraries, produce more
10798 optimized code, provide near-perfect API export and prevent symbol
10799 clashes. It is strongly recommended that you use this in any
10800 shared objects you distribute.
10801 Despite the nomenclature, "default" always means public ie; avail‐
10803 able to be linked against from outside the shared object. "pro‐
10804 tected" and "internal" are pretty useless in real-world usage so
10805 the only other commonly used option will be "hidden". The default
10806 if -fvisibility isn't specified is "default", i.e., make every sym‐
10807 bol public---this causes the same behavior as previous versions of
10808 GCC.
10809 A good explanation of the benefits offered by ensuring ELF symbols
10811 have the correct visibility is given by "How To Write Shared
10812 Libraries" by Ulrich Drepper (which can be found at <http://peo‐
10813 ple.redhat.com/~drepper/>)---however a superior solution made pos‐
10814 sible by this option to marking things hidden when the default is
10815 public is to make the default hidden and mark things public. This
10816 is the norm with DLL's on Windows and with -fvisibility=hidden and
10817 "__attribute__ ((visibility("default")))" instead of
10818 "__declspec(dllexport)" you get almost identical semantics with
10819 identical syntax. This is a great boon to those working with
10820 cross-platform projects.
10821 For those adding visibility support to existing code, you may find
10823 #pragma GCC visibility of use. This works by you enclosing the
10824 declarations you wish to set visibility for with (for example)
10825 #pragma GCC visibility push(hidden) and #pragma GCC visibility pop.
10826 Bear in mind that symbol visibility should be viewed as part of the
10827 API interface contract and thus all new code should always specify
10828 visibility when it is not the default ie; declarations only for use
10829 within the local DSO should always be marked explicitly as hidden
10830 as so to avoid PLT indirection overheads---making this abundantly
10831 clear also aids readability and self-documentation of the code.
10832 Note that due to ISO C++ specification requirements, operator new
10833 and operator delete must always be of default visibility.
10834 An overview of these techniques, their benefits and how to use them
10836 is at <http://gcc.gnu.org/wiki/Visibility>.
10837
10839 This section describes several environment variables that affect how
10840 GCC operates. Some of them work by specifying directories or prefixes
10841 to use when searching for various kinds of files. Some are used to
10842 specify other aspects of the compilation environment.
10843 Note that you can also specify places to search using options such as
10845 -B, -I and -L. These take precedence over places specified using envi‐
10846 ronment variables, which in turn take precedence over those specified
10847 by the configuration of GCC.
10848 LANG
10850 LC_CTYPE
10851 LC_MESSAGES
10852 LC_ALL
10853 These environment variables control the way that GCC uses localiza‐
10854 tion information that allow GCC to work with different national
10855 conventions. GCC inspects the locale categories LC_CTYPE and
10856 LC_MESSAGES if it has been configured to do so. These locale cate‐
10857 gories can be set to any value supported by your installation. A
10858 typical value is en_GB.UTF-8 for English in the United Kingdom
10859 encoded in UTF-8.
10860 The LC_CTYPE environment variable specifies character classifica‐
10862 tion. GCC uses it to determine the character boundaries in a
10863 string; this is needed for some multibyte encodings that contain
10864 quote and escape characters that would otherwise be interpreted as
10865 a string end or escape.
10866 The LC_MESSAGES environment variable specifies the language to use
10868 in diagnostic messages.
10869 If the LC_ALL environment variable is set, it overrides the value
10871 of LC_CTYPE and LC_MESSAGES; otherwise, LC_CTYPE and LC_MESSAGES
10872 default to the value of the LANG environment variable. If none of
10873 these variables are set, GCC defaults to traditional C English
10874 behavior.
10875 TMPDIR
10877 If TMPDIR is set, it specifies the directory to use for temporary
10878 files. GCC uses temporary files to hold the output of one stage of
10879 compilation which is to be used as input to the next stage: for
10880 example, the output of the preprocessor, which is the input to the
10881 compiler proper.
10882 GCC_EXEC_PREFIX
10884 If GCC_EXEC_PREFIX is set, it specifies a prefix to use in the
10885 names of the subprograms executed by the compiler. No slash is
10886 added when this prefix is combined with the name of a subprogram,
10887 but you can specify a prefix that ends with a slash if you wish.
10888 If GCC_EXEC_PREFIX is not set, GCC will attempt to figure out an
10890 appropriate prefix to use based on the pathname it was invoked
10891 with.
10892 If GCC cannot find the subprogram using the specified prefix, it
10894 tries looking in the usual places for the subprogram.
10895 The default value of GCC_EXEC_PREFIX is prefix/lib/gcc/ where pre‐
10897 fix is the value of "prefix" when you ran the configure script.
10898 Other prefixes specified with -B take precedence over this prefix.
10900 This prefix is also used for finding files such as crt0.o that are
10902 used for linking.
10903 In addition, the prefix is used in an unusual way in finding the
10905 directories to search for header files. For each of the standard
10906 directories whose name normally begins with /usr/local/lib/gcc
10907 (more precisely, with the value of GCC_INCLUDE_DIR), GCC tries
10908 replacing that beginning with the specified prefix to produce an
10909 alternate directory name. Thus, with -Bfoo/, GCC will search
10910 foo/bar where it would normally search /usr/local/lib/bar. These
10911 alternate directories are searched first; the standard directories
10912 come next.
10913 COMPILER_PATH
10915 The value of COMPILER_PATH is a colon-separated list of directo‐
10916 ries, much like PATH. GCC tries the directories thus specified
10917 when searching for subprograms, if it can't find the subprograms
10918 using GCC_EXEC_PREFIX.
10919 LIBRARY_PATH
10921 The value of LIBRARY_PATH is a colon-separated list of directories,
10922 much like PATH. When configured as a native compiler, GCC tries
10923 the directories thus specified when searching for special linker
10924 files, if it can't find them using GCC_EXEC_PREFIX. Linking using
10925 GCC also uses these directories when searching for ordinary
10926 libraries for the -l option (but directories specified with -L come
10927 first).
10928 LANG
10930 This variable is used to pass locale information to the compiler.
10931 One way in which this information is used is to determine the char‐
10932 acter set to be used when character literals, string literals and
10933 comments are parsed in C and C++. When the compiler is configured
10934 to allow multibyte characters, the following values for LANG are
10935 recognized:
10936 C-JIS
10938 Recognize JIS characters.
10939 C-SJIS
10941 Recognize SJIS characters.
10942 C-EUCJP
10944 Recognize EUCJP characters.
10945 If LANG is not defined, or if it has some other value, then the
10947 compiler will use mblen and mbtowc as defined by the default locale
10948 to recognize and translate multibyte characters.
10949 Some additional environments variables affect the behavior of the pre‐
10951 processor.
10952 CPATH
10954 C_INCLUDE_PATH
10955 CPLUS_INCLUDE_PATH
10956 OBJC_INCLUDE_PATH
10957 Each variable's value is a list of directories separated by a spe‐
10958 cial character, much like PATH, in which to look for header files.
10959 The special character, "PATH_SEPARATOR", is target-dependent and
10960 determined at GCC build time. For Microsoft Windows-based targets
10961 it is a semicolon, and for almost all other targets it is a colon.
10962 CPATH specifies a list of directories to be searched as if speci‐
10964 fied with -I, but after any paths given with -I options on the com‐
10965 mand line. This environment variable is used regardless of which
10966 language is being preprocessed.
10967 The remaining environment variables apply only when preprocessing
10969 the particular language indicated. Each specifies a list of direc‐
10970 tories to be searched as if specified with -isystem, but after any
10971 paths given with -isystem options on the command line.
10972 In all these variables, an empty element instructs the compiler to
10974 search its current working directory. Empty elements can appear at
10975 the beginning or end of a path. For instance, if the value of
10976 CPATH is ":/special/include", that has the same effect as
10977 -I. -I/special/include.
10978 DEPENDENCIES_OUTPUT
10980 If this variable is set, its value specifies how to output depen‐
10981 dencies for Make based on the non-system header files processed by
10982 the compiler. System header files are ignored in the dependency
10983 output.
10984 The value of DEPENDENCIES_OUTPUT can be just a file name, in which
10986 case the Make rules are written to that file, guessing the target
10987 name from the source file name. Or the value can have the form
10988 file target, in which case the rules are written to file file using
10989 target as the target name.
10990 In other words, this environment variable is equivalent to combin‐
10992 ing the options -MM and -MF, with an optional -MT switch too.
10993 SUNPRO_DEPENDENCIES
10995 This variable is the same as DEPENDENCIES_OUTPUT (see above),
10996 except that system header files are not ignored, so it implies -M
10997 rather than -MM. However, the dependence on the main input file is
10998 omitted.
10999
11001 For instructions on reporting bugs, see <http://gcc.gnu.org/bugs.html>.
11002
11004 1. On some systems, gcc -shared needs to build supplementary stub code
11005 for constructors to work. On multi-libbed systems, gcc -shared
11006 must select the correct support libraries to link against. Failing
11007 to supply the correct flags may lead to subtle defects. Supplying
11008 them in cases where they are not necessary is innocuous.
11009
11011 gpl(7), gfdl(7), fsf-funding(7), cpp(1), gcov(1), as(1), ld(1), gdb(1),
11012 adb(1), dbx(1), sdb(1) and the Info entries for gcc, cpp, as, ld, binu‐
11013 tils and gdb.
11014
11016 See the Info entry for gcc, or <http://gcc.gnu.org/onlinedocs/gcc/Con‐
11017 tributors.html>, for contributors to GCC.
11018
11020 Copyright (c) 1988, 1989, 1992, 1993, 1994, 1995, 1996, 1997, 1998,
11021 1999, 2000, 2001, 2002, 2003, 2004, 2005 Free Software Foundation, Inc.
11022 Permission is granted to copy, distribute and/or modify this document
11024 under the terms of the GNU Free Documentation License, Version 1.2 or
11025 any later version published by the Free Software Foundation; with the
11026 Invariant Sections being "GNU General Public License" and "Funding Free
11027 Software", the Front-Cover texts being (a) (see below), and with the
11028 Back-Cover Texts being (b) (see below). A copy of the license is
11029 included in the gfdl(7) man page.
11030 (a) The FSF's Front-Cover Text is:
11032 A GNU Manual
11034 (b) The FSF's Back-Cover Text is:
11036 You have freedom to copy and modify this GNU Manual, like GNU
11038 software. Copies published by the Free Software Foundation raise
11039 funds for GNU development.
11040gcc-4.1.2 2007-02-14 GCC(1)