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...] [-Wpedantic]
12 [-Idir...] [-Ldir...]
13 [-Dmacro[=defn]...] [-Umacro]
14 [-foption...] [-mmachine-option...]
15 [-o outfile] [@file] infile...
16 Only the most useful options are listed here; see below for the
18 remainder. 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 or more stages of processing. Some
28 options control the preprocessor and others the compiler itself. Yet
29 other options control the assembler and linker; most of these are not
30 documented 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 usual way to run GCC is to run the executable called gcc, or
39 machine-gcc when cross-compiling, or machine-gcc-version to run a
40 specific version of GCC. When you compile C++ programs, you should
41 invoke GCC as g++ instead.
42 The gcc program accepts options and file names as operands. Many
44 options have multi-letter names; therefore multiple single-letter
45 options may not be grouped: -dv is very different from -d -v.
46 You can mix options and other arguments. For the most part, the order
48 you use doesn't matter. Order does matter when you use several options
49 of the same kind; for example, if you specify -L more than once, the
50 directories are searched in the order specified. Also, the placement
51 of the -l option is significant.
52 Many options have long names starting with -f or with -W---for example,
54 -fmove-loop-invariants, -Wformat and so on. Most of these have both
55 positive and negative forms; the negative form of -ffoo is -fno-foo.
56 This manual documents only one of these two forms, whichever one is not
57 the default.
58
60 Option Summary
61 Here is a summary of all the options, grouped by type. Explanations
62 are in the following sections.
63 Overall Options
65 -c -S -E -o file -x language -v -### --help[=class[,...]]
66 --target-help --version -pass-exit-codes -pipe -specs=file
67 -wrapper @file -ffile-prefix-map=old=new -fplugin=file
68 -fplugin-arg-name=arg -fdump-ada-spec[-slim]
69 -fada-spec-parent=unit -fdump-go-spec=file
70 C Language Options
72 -ansi -std=standard -fgnu89-inline
73 -fpermitted-flt-eval-methods=standard -aux-info filename
74 -fallow-parameterless-variadic-functions -fno-asm -fno-builtin
75 -fno-builtin-function -fgimple -fhosted -ffreestanding -fopenacc
76 -fopenmp -fopenmp-simd -fms-extensions -fplan9-extensions
77 -fsso-struct=endianness -fallow-single-precision -fcond-mismatch
78 -flax-vector-conversions -fsigned-bitfields -fsigned-char
79 -funsigned-bitfields -funsigned-char
80 C++ Language Options
82 -fabi-version=n -fno-access-control -faligned-new=n
83 -fargs-in-order=n -fcheck-new -fconstexpr-depth=n
84 -fconstexpr-loop-limit=n -ffriend-injection -fno-elide-constructors
85 -fno-enforce-eh-specs -ffor-scope -fno-for-scope
86 -fno-gnu-keywords -fno-implicit-templates
87 -fno-implicit-inline-templates -fno-implement-inlines
88 -fms-extensions -fnew-inheriting-ctors -fnew-ttp-matching
89 -fno-nonansi-builtins -fnothrow-opt -fno-operator-names
90 -fno-optional-diags -fpermissive -fno-pretty-templates -frepo
91 -fno-rtti -fsized-deallocation -ftemplate-backtrace-limit=n
92 -ftemplate-depth=n -fno-threadsafe-statics -fuse-cxa-atexit
93 -fno-weak -nostdinc++ -fvisibility-inlines-hidden
94 -fvisibility-ms-compat -fext-numeric-literals -Wabi=n -Wabi-tag
95 -Wconversion-null -Wctor-dtor-privacy -Wdelete-non-virtual-dtor
96 -Wliteral-suffix -Wmultiple-inheritance -Wnamespaces -Wnarrowing
97 -Wnoexcept -Wnoexcept-type -Wclass-memaccess -Wnon-virtual-dtor
98 -Wreorder -Wregister -Weffc++ -Wstrict-null-sentinel -Wtemplates
99 -Wno-non-template-friend -Wold-style-cast -Woverloaded-virtual
100 -Wno-pmf-conversions -Wsign-promo -Wvirtual-inheritance
101 Objective-C and Objective-C++ Language Options
103 -fconstant-string-class=class-name -fgnu-runtime -fnext-runtime
104 -fno-nil-receivers -fobjc-abi-version=n -fobjc-call-cxx-cdtors
105 -fobjc-direct-dispatch -fobjc-exceptions -fobjc-gc -fobjc-nilcheck
106 -fobjc-std=objc1 -fno-local-ivars
107 -fivar-visibility=[public|protected|private|package]
108 -freplace-objc-classes -fzero-link -gen-decls -Wassign-intercept
109 -Wno-protocol -Wselector -Wstrict-selector-match
110 -Wundeclared-selector
111 Diagnostic Message Formatting Options
113 -fmessage-length=n -fdiagnostics-show-location=[once|every-line]
114 -fdiagnostics-color=[auto|never|always]
115 -fno-diagnostics-show-option -fno-diagnostics-show-caret
116 -fdiagnostics-parseable-fixits -fdiagnostics-generate-patch
117 -fdiagnostics-show-template-tree -fno-elide-type -fno-show-column
118 Warning Options
120 -fsyntax-only -fmax-errors=n -Wpedantic -pedantic-errors -w
121 -Wextra -Wall -Waddress -Waggregate-return -Waligned-new
122 -Walloc-zero -Walloc-size-larger-than=n -Walloca
123 -Walloca-larger-than=n -Wno-aggressive-loop-optimizations
124 -Warray-bounds -Warray-bounds=n -Wno-attributes -Wbool-compare
125 -Wbool-operation -Wno-builtin-declaration-mismatch
126 -Wno-builtin-macro-redefined -Wc90-c99-compat -Wc99-c11-compat
127 -Wc++-compat -Wc++11-compat -Wc++14-compat -Wcast-align
128 -Wcast-align=strict -Wcast-function-type -Wcast-qual
129 -Wchar-subscripts -Wchkp -Wcatch-value -Wcatch-value=n
130 -Wclobbered -Wcomment -Wconditionally-supported -Wconversion
131 -Wcoverage-mismatch -Wno-cpp -Wdangling-else -Wdate-time
132 -Wdelete-incomplete -Wno-deprecated -Wno-deprecated-declarations
133 -Wno-designated-init -Wdisabled-optimization
134 -Wno-discarded-qualifiers -Wno-discarded-array-qualifiers
135 -Wno-div-by-zero -Wdouble-promotion -Wduplicated-branches
136 -Wduplicated-cond -Wempty-body -Wenum-compare -Wno-endif-labels
137 -Wexpansion-to-defined -Werror -Werror=* -Wextra-semi
138 -Wfatal-errors -Wfloat-equal -Wformat -Wformat=2
139 -Wno-format-contains-nul -Wno-format-extra-args
140 -Wformat-nonliteral -Wformat-overflow=n -Wformat-security
141 -Wformat-signedness -Wformat-truncation=n -Wformat-y2k
142 -Wframe-address -Wframe-larger-than=len -Wno-free-nonheap-object
143 -Wjump-misses-init -Wif-not-aligned -Wignored-qualifiers
144 -Wignored-attributes -Wincompatible-pointer-types -Wimplicit
145 -Wimplicit-fallthrough -Wimplicit-fallthrough=n
146 -Wimplicit-function-declaration -Wimplicit-int -Winit-self
147 -Winline -Wno-int-conversion -Wint-in-bool-context
148 -Wno-int-to-pointer-cast -Winvalid-memory-model
149 -Wno-invalid-offsetof -Winvalid-pch -Wlarger-than=len -Wlogical-op
150 -Wlogical-not-parentheses -Wlong-long -Wmain
151 -Wmaybe-uninitialized -Wmemset-elt-size -Wmemset-transposed-args
152 -Wmisleading-indentation -Wmissing-attributes -Wmissing-braces
153 -Wmissing-field-initializers -Wmissing-include-dirs -Wno-multichar
154 -Wmultistatement-macros -Wnonnull -Wnonnull-compare
155 -Wnormalized=[none|id|nfc|nfkc] -Wnull-dereference -Wodr
156 -Wno-overflow -Wopenmp-simd -Woverride-init-side-effects
157 -Woverlength-strings -Wpacked -Wpacked-bitfield-compat
158 -Wpacked-not-aligned -Wpadded -Wparentheses
159 -Wno-pedantic-ms-format -Wplacement-new -Wplacement-new=n
160 -Wpointer-arith -Wpointer-compare -Wno-pointer-to-int-cast
161 -Wno-pragmas -Wredundant-decls -Wrestrict -Wno-return-local-addr
162 -Wreturn-type -Wsequence-point -Wshadow -Wno-shadow-ivar
163 -Wshadow=global, -Wshadow=local, -Wshadow=compatible-local
164 -Wshift-overflow -Wshift-overflow=n -Wshift-count-negative
165 -Wshift-count-overflow -Wshift-negative-value -Wsign-compare
166 -Wsign-conversion -Wfloat-conversion -Wno-scalar-storage-order
167 -Wsizeof-pointer-div -Wsizeof-pointer-memaccess
168 -Wsizeof-array-argument -Wstack-protector -Wstack-usage=len
169 -Wstrict-aliasing -Wstrict-aliasing=n -Wstrict-overflow
170 -Wstrict-overflow=n -Wstringop-overflow=n -Wstringop-truncation
171 -Wsuggest-attribute=[pure|const|noreturn|format|malloc]
172 -Wsuggest-final-types -Wsuggest-final-methods -Wsuggest-override
173 -Wmissing-format-attribute -Wsubobject-linkage -Wswitch
174 -Wswitch-bool -Wswitch-default -Wswitch-enum -Wswitch-unreachable
175 -Wsync-nand -Wsystem-headers -Wtautological-compare -Wtrampolines
176 -Wtrigraphs -Wtype-limits -Wundef -Wuninitialized
177 -Wunknown-pragmas -Wunsuffixed-float-constants -Wunused
178 -Wunused-function -Wunused-label -Wunused-local-typedefs
179 -Wunused-macros -Wunused-parameter -Wno-unused-result
180 -Wunused-value -Wunused-variable -Wunused-const-variable
181 -Wunused-const-variable=n -Wunused-but-set-parameter
182 -Wunused-but-set-variable -Wuseless-cast -Wvariadic-macros
183 -Wvector-operation-performance -Wvla -Wvla-larger-than=n
184 -Wvolatile-register-var -Wwrite-strings
185 -Wzero-as-null-pointer-constant -Whsa
186 C and Objective-C-only Warning Options
188 -Wbad-function-cast -Wmissing-declarations
189 -Wmissing-parameter-type -Wmissing-prototypes -Wnested-externs
190 -Wold-style-declaration -Wold-style-definition -Wstrict-prototypes
191 -Wtraditional -Wtraditional-conversion
192 -Wdeclaration-after-statement -Wpointer-sign
193 Debugging Options
195 -g -glevel -gdwarf -gdwarf-version -ggdb -grecord-gcc-switches
196 -gno-record-gcc-switches -gstabs -gstabs+ -gstrict-dwarf
197 -gno-strict-dwarf -gas-loc-support -gno-as-loc-support
198 -gas-locview-support -gno-as-locview-support -gcolumn-info
199 -gno-column-info -gstatement-frontiers -gno-statement-frontiers
200 -gvariable-location-views -gno-variable-location-views
201 -ginternal-reset-location-views -gno-internal-reset-location-views
202 -ginline-points -gno-inline-points -gvms -gxcoff -gxcoff+
203 -gz[=type] -fdebug-prefix-map=old=new -fdebug-types-section
204 -fno-eliminate-unused-debug-types -femit-struct-debug-baseonly
205 -femit-struct-debug-reduced -femit-struct-debug-detailed[=spec-
206 list] -feliminate-unused-debug-symbols -femit-class-debug-always
207 -fno-merge-debug-strings -fno-dwarf2-cfi-asm -fvar-tracking
208 -fvar-tracking-assignments
209 Optimization Options
211 -faggressive-loop-optimizations -falign-functions[=n]
212 -falign-jumps[=n] -falign-labels[=n] -falign-loops[=n]
213 -fassociative-math -fauto-profile -fauto-profile[=path]
214 -fauto-inc-dec -fbranch-probabilities
215 -fbranch-target-load-optimize -fbranch-target-load-optimize2
216 -fbtr-bb-exclusive -fcaller-saves -fcombine-stack-adjustments
217 -fconserve-stack -fcompare-elim -fcprop-registers -fcrossjumping
218 -fcse-follow-jumps -fcse-skip-blocks -fcx-fortran-rules
219 -fcx-limited-range -fdata-sections -fdce -fdelayed-branch
220 -fdelete-null-pointer-checks -fdevirtualize
221 -fdevirtualize-speculatively -fdevirtualize-at-ltrans -fdse
222 -fearly-inlining -fipa-sra -fexpensive-optimizations
223 -ffat-lto-objects -ffast-math -ffinite-math-only -ffloat-store
224 -fexcess-precision=style -fforward-propagate -ffp-contract=style
225 -ffunction-sections -fgcse -fgcse-after-reload -fgcse-las
226 -fgcse-lm -fgraphite-identity -fgcse-sm -fhoist-adjacent-loads
227 -fif-conversion -fif-conversion2 -findirect-inlining
228 -finline-functions -finline-functions-called-once
229 -finline-limit=n -finline-small-functions -fipa-cp -fipa-cp-clone
230 -fipa-bit-cp -fipa-vrp -fipa-pta -fipa-profile -fipa-pure-const
231 -fipa-reference -fipa-icf -fira-algorithm=algorithm
232 -fira-region=region -fira-hoist-pressure -fira-loop-pressure
233 -fno-ira-share-save-slots -fno-ira-share-spill-slots
234 -fisolate-erroneous-paths-dereference
235 -fisolate-erroneous-paths-attribute -fivopts
236 -fkeep-inline-functions -fkeep-static-functions
237 -fkeep-static-consts -flimit-function-alignment
238 -flive-range-shrinkage -floop-block -floop-interchange
239 -floop-strip-mine -floop-unroll-and-jam -floop-nest-optimize
240 -floop-parallelize-all -flra-remat -flto -flto-compression-level
241 -flto-partition=alg -fmerge-all-constants -fmerge-constants
242 -fmodulo-sched -fmodulo-sched-allow-regmoves
243 -fmove-loop-invariants -fno-branch-count-reg -fno-defer-pop
244 -fno-fp-int-builtin-inexact -fno-function-cse
245 -fno-guess-branch-probability -fno-inline -fno-math-errno
246 -fno-peephole -fno-peephole2 -fno-printf-return-value
247 -fno-sched-interblock -fno-sched-spec -fno-signed-zeros
248 -fno-toplevel-reorder -fno-trapping-math
249 -fno-zero-initialized-in-bss -fomit-frame-pointer
250 -foptimize-sibling-calls -fpartial-inlining -fpeel-loops
251 -fpredictive-commoning -fprefetch-loop-arrays -fprofile-correction
252 -fprofile-use -fprofile-use=path -fprofile-values
253 -fprofile-reorder-functions -freciprocal-math -free
254 -frename-registers -freorder-blocks
255 -freorder-blocks-algorithm=algorithm -freorder-blocks-and-partition
256 -freorder-functions -frerun-cse-after-loop
257 -freschedule-modulo-scheduled-loops -frounding-math
258 -fsched2-use-superblocks -fsched-pressure -fsched-spec-load
259 -fsched-spec-load-dangerous -fsched-stalled-insns-dep[=n]
260 -fsched-stalled-insns[=n] -fsched-group-heuristic
261 -fsched-critical-path-heuristic -fsched-spec-insn-heuristic
262 -fsched-rank-heuristic -fsched-last-insn-heuristic
263 -fsched-dep-count-heuristic -fschedule-fusion -fschedule-insns
264 -fschedule-insns2 -fsection-anchors -fselective-scheduling
265 -fselective-scheduling2 -fsel-sched-pipelining
266 -fsel-sched-pipelining-outer-loops -fsemantic-interposition
267 -fshrink-wrap -fshrink-wrap-separate -fsignaling-nans
268 -fsingle-precision-constant -fsplit-ivs-in-unroller -fsplit-loops
269 -fsplit-paths -fsplit-wide-types -fssa-backprop -fssa-phiopt
270 -fstdarg-opt -fstore-merging -fstrict-aliasing -fthread-jumps
271 -ftracer -ftree-bit-ccp -ftree-builtin-call-dce -ftree-ccp
272 -ftree-ch -ftree-coalesce-vars -ftree-copy-prop -ftree-dce
273 -ftree-dominator-opts -ftree-dse -ftree-forwprop -ftree-fre
274 -fcode-hoisting -ftree-loop-if-convert -ftree-loop-im
275 -ftree-phiprop -ftree-loop-distribution
276 -ftree-loop-distribute-patterns -ftree-loop-ivcanon
277 -ftree-loop-linear -ftree-loop-optimize -ftree-loop-vectorize
278 -ftree-parallelize-loops=n -ftree-pre -ftree-partial-pre
279 -ftree-pta -ftree-reassoc -ftree-sink -ftree-slsr -ftree-sra
280 -ftree-switch-conversion -ftree-tail-merge -ftree-ter
281 -ftree-vectorize -ftree-vrp -funconstrained-commons
282 -funit-at-a-time -funroll-all-loops -funroll-loops
283 -funsafe-math-optimizations -funswitch-loops -fipa-ra
284 -fvariable-expansion-in-unroller -fvect-cost-model -fvpt -fweb
285 -fwhole-program -fwpa -fuse-linker-plugin --param name=value -O
286 -O0 -O1 -O2 -O3 -Os -Ofast -Og
287 Program Instrumentation Options
289 -p -pg -fprofile-arcs --coverage -ftest-coverage
290 -fprofile-abs-path -fprofile-dir=path -fprofile-generate
291 -fprofile-generate=path -fsanitize=style -fsanitize-recover
292 -fsanitize-recover=style -fasan-shadow-offset=number
293 -fsanitize-sections=s1,s2,... -fsanitize-undefined-trap-on-error
294 -fbounds-check -fcheck-pointer-bounds -fchkp-check-incomplete-type
295 -fchkp-first-field-has-own-bounds -fchkp-narrow-bounds
296 -fchkp-narrow-to-innermost-array -fchkp-optimize
297 -fchkp-use-fast-string-functions -fchkp-use-nochk-string-functions
298 -fchkp-use-static-bounds -fchkp-use-static-const-bounds
299 -fchkp-treat-zero-dynamic-size-as-infinite -fchkp-check-read
300 -fchkp-check-read -fchkp-check-write -fchkp-store-bounds
301 -fchkp-instrument-calls -fchkp-instrument-marked-only
302 -fchkp-use-wrappers -fchkp-flexible-struct-trailing-arrays
303 -fcf-protection=[full|branch|return|none] -fstack-protector
304 -fstack-protector-all -fstack-protector-strong
305 -fstack-protector-explicit -fstack-check
306 -fstack-limit-register=reg -fstack-limit-symbol=sym
307 -fno-stack-limit -fsplit-stack -fvtable-verify=[std|preinit|none]
308 -fvtv-counts -fvtv-debug -finstrument-functions
309 -finstrument-functions-exclude-function-list=sym,sym,...
310 -finstrument-functions-exclude-file-list=file,file,...
311 Preprocessor Options
313 -Aquestion=answer -A-question[=answer] -C -CC -Dmacro[=defn] -dD
314 -dI -dM -dN -dU -fdebug-cpp -fdirectives-only
315 -fdollars-in-identifiers -fexec-charset=charset
316 -fextended-identifiers -finput-charset=charset
317 -fmacro-prefix-map=old=new -fno-canonical-system-headers
318 -fpch-deps -fpch-preprocess -fpreprocessed -ftabstop=width
319 -ftrack-macro-expansion -fwide-exec-charset=charset
320 -fworking-directory -H -imacros file -include file -M -MD -MF
321 -MG -MM -MMD -MP -MQ -MT -no-integrated-cpp -P -pthread
322 -remap -traditional -traditional-cpp -trigraphs -Umacro -undef
323 -Wp,option -Xpreprocessor option
324 Assembler Options
326 -Wa,option -Xassembler option
327 Linker Options
329 object-file-name -fuse-ld=linker -llibrary -nostartfiles
330 -nodefaultlibs -nostdlib -pie -pthread -rdynamic -s -static
331 -static-pie -static-libgcc -static-libstdc++ -static-libasan
332 -static-libtsan -static-liblsan -static-libubsan -static-libmpx
333 -static-libmpxwrappers -shared -shared-libgcc -symbolic -T script
334 -Wl,option -Xlinker option -u symbol -z keyword
335 Directory Options
337 -Bprefix -Idir -I- -idirafter dir -imacros file -imultilib dir
338 -iplugindir=dir -iprefix file -iquote dir -isysroot dir -isystem
339 dir -iwithprefix dir -iwithprefixbefore dir -Ldir
340 -no-canonical-prefixes --no-sysroot-suffix -nostdinc -nostdinc++
341 --sysroot=dir
342 Code Generation Options
344 -fcall-saved-reg -fcall-used-reg -ffixed-reg -fexceptions
345 -fnon-call-exceptions -fdelete-dead-exceptions -funwind-tables
346 -fasynchronous-unwind-tables -fno-gnu-unique
347 -finhibit-size-directive -fno-common -fno-ident
348 -fpcc-struct-return -fpic -fPIC -fpie -fPIE -fno-plt
349 -fno-jump-tables -frecord-gcc-switches -freg-struct-return
350 -fshort-enums -fshort-wchar -fverbose-asm -fpack-struct[=n]
351 -fleading-underscore -ftls-model=model -fstack-reuse=reuse_level
352 -ftrampolines -ftrapv -fwrapv
353 -fvisibility=[default|internal|hidden|protected]
354 -fstrict-volatile-bitfields -fsync-libcalls
355 Developer Options
357 -dletters -dumpspecs -dumpmachine -dumpversion -dumpfullversion
358 -fchecking -fchecking=n -fdbg-cnt-list -fdbg-cnt=counter-value-
359 list -fdisable-ipa-pass_name -fdisable-rtl-pass_name
360 -fdisable-rtl-pass-name=range-list -fdisable-tree-pass_name
361 -fdisable-tree-pass-name=range-list -fdump-noaddr
362 -fdump-unnumbered -fdump-unnumbered-links
363 -fdump-class-hierarchy[-n] -fdump-final-insns[=file] -fdump-ipa-all
364 -fdump-ipa-cgraph -fdump-ipa-inline -fdump-lang-all
365 -fdump-lang-switch -fdump-lang-switch-options
366 -fdump-lang-switch-options=filename -fdump-passes -fdump-rtl-pass
367 -fdump-rtl-pass=filename -fdump-statistics -fdump-tree-all
368 -fdump-tree-switch -fdump-tree-switch-options
369 -fdump-tree-switch-options=filename -fcompare-debug[=opts]
370 -fcompare-debug-second -fenable-kind-pass -fenable-kind-pass=range-
371 list -fira-verbose=n -flto-report -flto-report-wpa
372 -fmem-report-wpa -fmem-report -fpre-ipa-mem-report
373 -fpost-ipa-mem-report -fopt-info -fopt-info-options[=file]
374 -fprofile-report -frandom-seed=string -fsched-verbose=n
375 -fsel-sched-verbose -fsel-sched-dump-cfg
376 -fsel-sched-pipelining-verbose -fstats -fstack-usage
377 -ftime-report -ftime-report-details
378 -fvar-tracking-assignments-toggle -gtoggle
379 -print-file-name=library -print-libgcc-file-name
380 -print-multi-directory -print-multi-lib -print-multi-os-directory
381 -print-prog-name=program -print-search-dirs -Q -print-sysroot
382 -print-sysroot-headers-suffix -save-temps -save-temps=cwd
383 -save-temps=obj -time[=file]
384 Machine-Dependent Options
386 AArch64 Options -mabi=name -mbig-endian -mlittle-endian
387 -mgeneral-regs-only -mcmodel=tiny -mcmodel=small -mcmodel=large
388 -mstrict-align -momit-leaf-frame-pointer -mtls-dialect=desc
389 -mtls-dialect=traditional -mtls-size=size -mfix-cortex-a53-835769
390 -mfix-cortex-a53-843419 -mlow-precision-recip-sqrt
391 -mlow-precision-sqrt -mlow-precision-div
392 -mpc-relative-literal-loads -msign-return-address=scope -march=name
393 -mcpu=name -mtune=name -moverride=string -mverbose-cost-dump
394 Adapteva Epiphany Options -mhalf-reg-file -mprefer-short-insn-regs
396 -mbranch-cost=num -mcmove -mnops=num -msoft-cmpsf -msplit-lohi
397 -mpost-inc -mpost-modify -mstack-offset=num -mround-nearest
398 -mlong-calls -mshort-calls -msmall16 -mfp-mode=mode
399 -mvect-double -max-vect-align=num -msplit-vecmove-early
400 -m1reg-reg
401 ARC Options -mbarrel-shifter -mjli-always -mcpu=cpu -mA6 -mARC600
403 -mA7 -mARC700 -mdpfp -mdpfp-compact -mdpfp-fast -mno-dpfp-lrsr
404 -mea -mno-mpy -mmul32x16 -mmul64 -matomic -mnorm -mspfp
405 -mspfp-compact -mspfp-fast -msimd -msoft-float -mswap -mcrc
406 -mdsp-packa -mdvbf -mlock -mmac-d16 -mmac-24 -mrtsc -mswape
407 -mtelephony -mxy -misize -mannotate-align -marclinux
408 -marclinux_prof -mlong-calls -mmedium-calls -msdata
409 -mirq-ctrl-saved -mrgf-banked-regs -mlpc-width=width -G num
410 -mvolatile-cache -mtp-regno=regno -malign-call -mauto-modify-reg
411 -mbbit-peephole -mno-brcc -mcase-vector-pcrel -mcompact-casesi
412 -mno-cond-exec -mearly-cbranchsi -mexpand-adddi -mindexed-loads
413 -mlra -mlra-priority-none -mlra-priority-compact mlra-priority-
414 noncompact -mno-millicode -mmixed-code -mq-class -mRcq -mRcw
415 -msize-level=level -mtune=cpu -mmultcost=num
416 -munalign-prob-threshold=probability -mmpy-option=multo -mdiv-rem
417 -mcode-density -mll64 -mfpu=fpu -mrf16
418 ARM Options -mapcs-frame -mno-apcs-frame -mabi=name
420 -mapcs-stack-check -mno-apcs-stack-check -mapcs-reentrant
421 -mno-apcs-reentrant -msched-prolog -mno-sched-prolog
422 -mlittle-endian -mbig-endian -mbe8 -mbe32 -mfloat-abi=name
423 -mfp16-format=name -mthumb-interwork -mno-thumb-interwork
424 -mcpu=name -march=name -mfpu=name -mtune=name -mprint-tune-info
425 -mstructure-size-boundary=n -mabort-on-noreturn -mlong-calls
426 -mno-long-calls -msingle-pic-base -mno-single-pic-base
427 -mpic-register=reg -mnop-fun-dllimport -mpoke-function-name -mthumb
428 -marm -mflip-thumb -mtpcs-frame -mtpcs-leaf-frame
429 -mcaller-super-interworking -mcallee-super-interworking -mtp=name
430 -mtls-dialect=dialect -mword-relocations -mfix-cortex-m3-ldrd
431 -munaligned-access -mneon-for-64bits -mslow-flash-data
432 -masm-syntax-unified -mrestrict-it -mverbose-cost-dump -mpure-code
433 -mcmse
434 AVR Options -mmcu=mcu -mabsdata -maccumulate-args
436 -mbranch-cost=cost -mcall-prologues -mgas-isr-prologues -mint8
437 -mn_flash=size -mno-interrupts -mmain-is-OS_task -mrelax -mrmw
438 -mstrict-X -mtiny-stack -mfract-convert-truncate -mshort-calls
439 -nodevicelib -Waddr-space-convert -Wmisspelled-isr
440 Blackfin Options -mcpu=cpu[-sirevision] -msim
442 -momit-leaf-frame-pointer -mno-omit-leaf-frame-pointer
443 -mspecld-anomaly -mno-specld-anomaly -mcsync-anomaly
444 -mno-csync-anomaly -mlow-64k -mno-low64k -mstack-check-l1
445 -mid-shared-library -mno-id-shared-library -mshared-library-id=n
446 -mleaf-id-shared-library -mno-leaf-id-shared-library -msep-data
447 -mno-sep-data -mlong-calls -mno-long-calls -mfast-fp
448 -minline-plt -mmulticore -mcorea -mcoreb -msdram -micplb
449 C6X Options -mbig-endian -mlittle-endian -march=cpu -msim
451 -msdata=sdata-type
452 CRIS Options -mcpu=cpu -march=cpu -mtune=cpu -mmax-stack-frame=n
454 -melinux-stacksize=n -metrax4 -metrax100 -mpdebug -mcc-init
455 -mno-side-effects -mstack-align -mdata-align -mconst-align
456 -m32-bit -m16-bit -m8-bit -mno-prologue-epilogue -mno-gotplt
457 -melf -maout -melinux -mlinux -sim -sim2 -mmul-bug-workaround
458 -mno-mul-bug-workaround
459 CR16 Options -mmac -mcr16cplus -mcr16c -msim -mint32 -mbit-ops
461 -mdata-model=model
462 Darwin Options -all_load -allowable_client -arch
464 -arch_errors_fatal -arch_only -bind_at_load -bundle
465 -bundle_loader -client_name -compatibility_version
466 -current_version -dead_strip -dependency-file -dylib_file
467 -dylinker_install_name -dynamic -dynamiclib
468 -exported_symbols_list -filelist -flat_namespace
469 -force_cpusubtype_ALL -force_flat_namespace
470 -headerpad_max_install_names -iframework -image_base -init
471 -install_name -keep_private_externs -multi_module
472 -multiply_defined -multiply_defined_unused -noall_load
473 -no_dead_strip_inits_and_terms -nofixprebinding -nomultidefs
474 -noprebind -noseglinkedit -pagezero_size -prebind
475 -prebind_all_twolevel_modules -private_bundle -read_only_relocs
476 -sectalign -sectobjectsymbols -whyload -seg1addr -sectcreate
477 -sectobjectsymbols -sectorder -segaddr -segs_read_only_addr
478 -segs_read_write_addr -seg_addr_table -seg_addr_table_filename
479 -seglinkedit -segprot -segs_read_only_addr -segs_read_write_addr
480 -single_module -static -sub_library -sub_umbrella
481 -twolevel_namespace -umbrella -undefined -unexported_symbols_list
482 -weak_reference_mismatches -whatsloaded -F -gused -gfull
483 -mmacosx-version-min=version -mkernel -mone-byte-bool
484 DEC Alpha Options -mno-fp-regs -msoft-float -mieee
486 -mieee-with-inexact -mieee-conformant -mfp-trap-mode=mode
487 -mfp-rounding-mode=mode -mtrap-precision=mode -mbuild-constants
488 -mcpu=cpu-type -mtune=cpu-type -mbwx -mmax -mfix -mcix
489 -mfloat-vax -mfloat-ieee -mexplicit-relocs -msmall-data
490 -mlarge-data -msmall-text -mlarge-text -mmemory-latency=time
491 FR30 Options -msmall-model -mno-lsim
493 FT32 Options -msim -mlra -mnodiv -mft32b -mcompress -mnopm
495 FRV Options -mgpr-32 -mgpr-64 -mfpr-32 -mfpr-64 -mhard-float
497 -msoft-float -malloc-cc -mfixed-cc -mdword -mno-dword -mdouble
498 -mno-double -mmedia -mno-media -mmuladd -mno-muladd -mfdpic
499 -minline-plt -mgprel-ro -multilib-library-pic -mlinked-fp
500 -mlong-calls -malign-labels -mlibrary-pic -macc-4 -macc-8 -mpack
501 -mno-pack -mno-eflags -mcond-move -mno-cond-move
502 -moptimize-membar -mno-optimize-membar -mscc -mno-scc
503 -mcond-exec -mno-cond-exec -mvliw-branch -mno-vliw-branch
504 -mmulti-cond-exec -mno-multi-cond-exec -mnested-cond-exec
505 -mno-nested-cond-exec -mtomcat-stats -mTLS -mtls -mcpu=cpu
506 GNU/Linux Options -mglibc -muclibc -mmusl -mbionic -mandroid
508 -tno-android-cc -tno-android-ld
509 H8/300 Options -mrelax -mh -ms -mn -mexr -mno-exr -mint32
511 -malign-300
512 HPPA Options -march=architecture-type -mcaller-copies
514 -mdisable-fpregs -mdisable-indexing -mfast-indirect-calls -mgas
515 -mgnu-ld -mhp-ld -mfixed-range=register-range -mjump-in-delay
516 -mlinker-opt -mlong-calls -mlong-load-store -mno-disable-fpregs
517 -mno-disable-indexing -mno-fast-indirect-calls -mno-gas
518 -mno-jump-in-delay -mno-long-load-store -mno-portable-runtime
519 -mno-soft-float -mno-space-regs -msoft-float -mpa-risc-1-0
520 -mpa-risc-1-1 -mpa-risc-2-0 -mportable-runtime -mschedule=cpu-
521 type -mspace-regs -msio -mwsio -munix=unix-std -nolibdld
522 -static -threads
523 IA-64 Options -mbig-endian -mlittle-endian -mgnu-as -mgnu-ld
525 -mno-pic -mvolatile-asm-stop -mregister-names -msdata -mno-sdata
526 -mconstant-gp -mauto-pic -mfused-madd
527 -minline-float-divide-min-latency
528 -minline-float-divide-max-throughput -mno-inline-float-divide
529 -minline-int-divide-min-latency -minline-int-divide-max-throughput
530 -mno-inline-int-divide -minline-sqrt-min-latency
531 -minline-sqrt-max-throughput -mno-inline-sqrt -mdwarf2-asm
532 -mearly-stop-bits -mfixed-range=register-range -mtls-size=tls-size
533 -mtune=cpu-type -milp32 -mlp64 -msched-br-data-spec
534 -msched-ar-data-spec -msched-control-spec -msched-br-in-data-spec
535 -msched-ar-in-data-spec -msched-in-control-spec -msched-spec-ldc
536 -msched-spec-control-ldc -msched-prefer-non-data-spec-insns
537 -msched-prefer-non-control-spec-insns
538 -msched-stop-bits-after-every-cycle
539 -msched-count-spec-in-critical-path
540 -msel-sched-dont-check-control-spec -msched-fp-mem-deps-zero-cost
541 -msched-max-memory-insns-hard-limit -msched-max-memory-insns=max-
542 insns
543 LM32 Options -mbarrel-shift-enabled -mdivide-enabled
545 -mmultiply-enabled -msign-extend-enabled -muser-enabled
546 M32R/D Options -m32r2 -m32rx -m32r -mdebug -malign-loops
548 -mno-align-loops -missue-rate=number -mbranch-cost=number
549 -mmodel=code-size-model-type -msdata=sdata-type -mno-flush-func
550 -mflush-func=name -mno-flush-trap -mflush-trap=number -G num
551 M32C Options -mcpu=cpu -msim -memregs=number
553 M680x0 Options -march=arch -mcpu=cpu -mtune=tune -m68000 -m68020
555 -m68020-40 -m68020-60 -m68030 -m68040 -m68060 -mcpu32 -m5200
556 -m5206e -m528x -m5307 -m5407 -mcfv4e -mbitfield -mno-bitfield
557 -mc68000 -mc68020 -mnobitfield -mrtd -mno-rtd -mdiv -mno-div
558 -mshort -mno-short -mhard-float -m68881 -msoft-float -mpcrel
559 -malign-int -mstrict-align -msep-data -mno-sep-data
560 -mshared-library-id=n -mid-shared-library -mno-id-shared-library
561 -mxgot -mno-xgot -mlong-jump-table-offsets
562 MCore Options -mhardlit -mno-hardlit -mdiv -mno-div
564 -mrelax-immediates -mno-relax-immediates -mwide-bitfields
565 -mno-wide-bitfields -m4byte-functions -mno-4byte-functions
566 -mcallgraph-data -mno-callgraph-data -mslow-bytes -mno-slow-bytes
567 -mno-lsim -mlittle-endian -mbig-endian -m210 -m340
568 -mstack-increment
569 MeP Options -mabsdiff -mall-opts -maverage -mbased=n -mbitops
571 -mc=n -mclip -mconfig=name -mcop -mcop32 -mcop64 -mivc2 -mdc
572 -mdiv -meb -mel -mio-volatile -ml -mleadz -mm -mminmax
573 -mmult -mno-opts -mrepeat -ms -msatur -msdram -msim
574 -msimnovec -mtf -mtiny=n
575 MicroBlaze Options -msoft-float -mhard-float -msmall-divides
577 -mcpu=cpu -mmemcpy -mxl-soft-mul -mxl-soft-div -mxl-barrel-shift
578 -mxl-pattern-compare -mxl-stack-check -mxl-gp-opt -mno-clearbss
579 -mxl-multiply-high -mxl-float-convert -mxl-float-sqrt
580 -mbig-endian -mlittle-endian -mxl-reorder -mxl-mode-app-model
581 MIPS Options -EL -EB -march=arch -mtune=arch -mips1 -mips2
583 -mips3 -mips4 -mips32 -mips32r2 -mips32r3 -mips32r5 -mips32r6
584 -mips64 -mips64r2 -mips64r3 -mips64r5 -mips64r6 -mips16
585 -mno-mips16 -mflip-mips16 -minterlink-compressed
586 -mno-interlink-compressed -minterlink-mips16 -mno-interlink-mips16
587 -mabi=abi -mabicalls -mno-abicalls -mshared -mno-shared -mplt
588 -mno-plt -mxgot -mno-xgot -mgp32 -mgp64 -mfp32 -mfpxx -mfp64
589 -mhard-float -msoft-float -mno-float -msingle-float
590 -mdouble-float -modd-spreg -mno-odd-spreg -mabs=mode
591 -mnan=encoding -mdsp -mno-dsp -mdspr2 -mno-dspr2 -mmcu
592 -mmno-mcu -meva -mno-eva -mvirt -mno-virt -mxpa -mno-xpa
593 -mmicromips -mno-micromips -mmsa -mno-msa -mfpu=fpu-type
594 -msmartmips -mno-smartmips -mpaired-single -mno-paired-single
595 -mdmx -mno-mdmx -mips3d -mno-mips3d -mmt -mno-mt -mllsc
596 -mno-llsc -mlong64 -mlong32 -msym32 -mno-sym32 -Gnum
597 -mlocal-sdata -mno-local-sdata -mextern-sdata -mno-extern-sdata
598 -mgpopt -mno-gopt -membedded-data -mno-embedded-data
599 -muninit-const-in-rodata -mno-uninit-const-in-rodata
600 -mcode-readable=setting -msplit-addresses -mno-split-addresses
601 -mexplicit-relocs -mno-explicit-relocs -mcheck-zero-division
602 -mno-check-zero-division -mdivide-traps -mdivide-breaks
603 -mload-store-pairs -mno-load-store-pairs -mmemcpy -mno-memcpy
604 -mlong-calls -mno-long-calls -mmad -mno-mad -mimadd -mno-imadd
605 -mfused-madd -mno-fused-madd -nocpp -mfix-24k -mno-fix-24k
606 -mfix-r4000 -mno-fix-r4000 -mfix-r4400 -mno-fix-r4400
607 -mfix-r10000 -mno-fix-r10000 -mfix-rm7000 -mno-fix-rm7000
608 -mfix-vr4120 -mno-fix-vr4120 -mfix-vr4130 -mno-fix-vr4130
609 -mfix-sb1 -mno-fix-sb1 -mflush-func=func -mno-flush-func
610 -mbranch-cost=num -mbranch-likely -mno-branch-likely
611 -mcompact-branches=policy -mfp-exceptions -mno-fp-exceptions
612 -mvr4130-align -mno-vr4130-align -msynci -mno-synci -mlxc1-sxc1
613 -mno-lxc1-sxc1 -mmadd4 -mno-madd4 -mrelax-pic-calls
614 -mno-relax-pic-calls -mmcount-ra-address -mframe-header-opt
615 -mno-frame-header-opt
616 MMIX Options -mlibfuncs -mno-libfuncs -mepsilon -mno-epsilon
618 -mabi=gnu -mabi=mmixware -mzero-extend -mknuthdiv
619 -mtoplevel-symbols -melf -mbranch-predict -mno-branch-predict
620 -mbase-addresses -mno-base-addresses -msingle-exit
621 -mno-single-exit
622 MN10300 Options -mmult-bug -mno-mult-bug -mno-am33 -mam33
624 -mam33-2 -mam34 -mtune=cpu-type -mreturn-pointer-on-d0 -mno-crt0
625 -mrelax -mliw -msetlb
626 Moxie Options -meb -mel -mmul.x -mno-crt0
628 MSP430 Options -msim -masm-hex -mmcu= -mcpu= -mlarge -msmall
630 -mrelax -mwarn-mcu -mcode-region= -mdata-region= -msilicon-errata=
631 -msilicon-errata-warn= -mhwmult= -minrt
632 NDS32 Options -mbig-endian -mlittle-endian -mreduced-regs
634 -mfull-regs -mcmov -mno-cmov -mext-perf -mno-ext-perf -mext-perf2
635 -mno-ext-perf2 -mext-string -mno-ext-string -mv3push -mno-v3push
636 -m16bit -mno-16bit -misr-vector-size=num -mcache-block-size=num
637 -march=arch -mcmodel=code-model -mctor-dtor -mrelax
638 Nios II Options -G num -mgpopt=option -mgpopt -mno-gpopt
640 -mgprel-sec=regexp -mr0rel-sec=regexp -mel -meb -mno-bypass-cache
641 -mbypass-cache -mno-cache-volatile -mcache-volatile
642 -mno-fast-sw-div -mfast-sw-div -mhw-mul -mno-hw-mul -mhw-mulx
643 -mno-hw-mulx -mno-hw-div -mhw-div -mcustom-insn=N
644 -mno-custom-insn -mcustom-fpu-cfg=name -mhal -msmallc
645 -msys-crt0=name -msys-lib=name -march=arch -mbmx -mno-bmx -mcdx
646 -mno-cdx
647 Nvidia PTX Options -m32 -m64 -mmainkernel -moptimize
649 PDP-11 Options -mfpu -msoft-float -mac0 -mno-ac0 -m40 -m45
651 -m10 -mbcopy -mbcopy-builtin -mint32 -mno-int16 -mint16
652 -mno-int32 -mfloat32 -mno-float64 -mfloat64 -mno-float32
653 -mabshi -mno-abshi -mbranch-expensive -mbranch-cheap -munix-asm
654 -mdec-asm
655 picoChip Options -mae=ae_type -mvliw-lookahead=N
657 -msymbol-as-address -mno-inefficient-warnings
658 PowerPC Options See RS/6000 and PowerPC Options.
660 PowerPC SPE Options -mcpu=cpu-type -mtune=cpu-type -mmfcrf
662 -mno-mfcrf -mpopcntb -mno-popcntb -mfull-toc -mminimal-toc
663 -mno-fp-in-toc -mno-sum-in-toc -m32 -mxl-compat -mno-xl-compat
664 -malign-power -malign-natural -msoft-float -mhard-float
665 -mmultiple -mno-multiple -msingle-float -mdouble-float -mupdate
666 -mno-update -mavoid-indexed-addresses -mno-avoid-indexed-addresses
667 -mstrict-align -mno-strict-align -mrelocatable -mno-relocatable
668 -mrelocatable-lib -mno-relocatable-lib -mtoc -mno-toc -mlittle
669 -mlittle-endian -mbig -mbig-endian -msingle-pic-base
670 -mprioritize-restricted-insns=priority
671 -msched-costly-dep=dependence_type -minsert-sched-nops=scheme
672 -mcall-sysv -mcall-netbsd -maix-struct-return
673 -msvr4-struct-return -mabi=abi-type -msecure-plt -mbss-plt
674 -mblock-move-inline-limit=num -misel -mno-isel -misel=yes
675 -misel=no -mspe -mno-spe -mspe=yes -mspe=no -mfloat-gprs=yes
676 -mfloat-gprs=no -mfloat-gprs=single -mfloat-gprs=double
677 -mprototype -mno-prototype -msim -mmvme -mads -myellowknife
678 -memb -msdata -msdata=opt -mvxworks -G num -mrecip -mrecip=opt
679 -mno-recip -mrecip-precision -mno-recip-precision
680 -mpointers-to-nested-functions -mno-pointers-to-nested-functions
681 -msave-toc-indirect -mno-save-toc-indirect -mcompat-align-parm
682 -mno-compat-align-parm -mfloat128 -mno-float128 -mgnu-attribute
683 -mno-gnu-attribute -mstack-protector-guard=guard
684 -mstack-protector-guard-reg=reg
685 -mstack-protector-guard-offset=offset
686 RISC-V Options -mbranch-cost=N-instruction -mplt -mno-plt
688 -mabi=ABI-string -mfdiv -mno-fdiv -mdiv -mno-div -march=ISA-
689 string -mtune=processor-string -mpreferred-stack-boundary=num
690 -msmall-data-limit=N-bytes -msave-restore -mno-save-restore
691 -mstrict-align -mno-strict-align -mcmodel=medlow -mcmodel=medany
692 -mexplicit-relocs -mno-explicit-relocs -mrelax -mno-relax
693 RL78 Options -msim -mmul=none -mmul=g13 -mmul=g14 -mallregs
695 -mcpu=g10 -mcpu=g13 -mcpu=g14 -mg10 -mg13 -mg14
696 -m64bit-doubles -m32bit-doubles -msave-mduc-in-interrupts
697 RS/6000 and PowerPC Options -mcpu=cpu-type -mtune=cpu-type
699 -mcmodel=code-model -mpowerpc64 -maltivec -mno-altivec
700 -mpowerpc-gpopt -mno-powerpc-gpopt -mpowerpc-gfxopt
701 -mno-powerpc-gfxopt -mmfcrf -mno-mfcrf -mpopcntb -mno-popcntb
702 -mpopcntd -mno-popcntd -mfprnd -mno-fprnd -mcmpb -mno-cmpb
703 -mmfpgpr -mno-mfpgpr -mhard-dfp -mno-hard-dfp -mfull-toc
704 -mminimal-toc -mno-fp-in-toc -mno-sum-in-toc -m64 -m32
705 -mxl-compat -mno-xl-compat -mpe -malign-power -malign-natural
706 -msoft-float -mhard-float -mmultiple -mno-multiple
707 -msingle-float -mdouble-float -msimple-fpu -mupdate -mno-update
708 -mavoid-indexed-addresses -mno-avoid-indexed-addresses
709 -mfused-madd -mno-fused-madd -mbit-align -mno-bit-align
710 -mstrict-align -mno-strict-align -mrelocatable -mno-relocatable
711 -mrelocatable-lib -mno-relocatable-lib -mtoc -mno-toc -mlittle
712 -mlittle-endian -mbig -mbig-endian -mdynamic-no-pic -maltivec
713 -mswdiv -msingle-pic-base -mprioritize-restricted-insns=priority
714 -msched-costly-dep=dependence_type -minsert-sched-nops=scheme
715 -mcall-aixdesc -mcall-eabi -mcall-freebsd -mcall-linux
716 -mcall-netbsd -mcall-openbsd -mcall-sysv -mcall-sysv-eabi
717 -mcall-sysv-noeabi -mtraceback=traceback_type -maix-struct-return
718 -msvr4-struct-return -mabi=abi-type -msecure-plt -mbss-plt
719 -mblock-move-inline-limit=num -mblock-compare-inline-limit=num
720 -mblock-compare-inline-loop-limit=num
721 -mstring-compare-inline-limit=num -misel -mno-isel -misel=yes
722 -misel=no -mpaired -mvrsave -mno-vrsave -mmulhw -mno-mulhw
723 -mdlmzb -mno-dlmzb -mprototype -mno-prototype -msim -mmvme
724 -mads -myellowknife -memb -msdata -msdata=opt
725 -mreadonly-in-sdata -mvxworks -G num -mrecip -mrecip=opt
726 -mno-recip -mrecip-precision -mno-recip-precision -mveclibabi=type
727 -mfriz -mno-friz -mpointers-to-nested-functions
728 -mno-pointers-to-nested-functions -msave-toc-indirect
729 -mno-save-toc-indirect -mpower8-fusion -mno-mpower8-fusion
730 -mpower8-vector -mno-power8-vector -mcrypto -mno-crypto -mhtm
731 -mno-htm -mquad-memory -mno-quad-memory -mquad-memory-atomic
732 -mno-quad-memory-atomic -mcompat-align-parm -mno-compat-align-parm
733 -mfloat128 -mno-float128 -mfloat128-hardware
734 -mno-float128-hardware -mgnu-attribute -mno-gnu-attribute
735 -mstack-protector-guard=guard -mstack-protector-guard-reg=reg
736 -mstack-protector-guard-offset=offset
737 RX Options -m64bit-doubles -m32bit-doubles -fpu -nofpu -mcpu=
739 -mbig-endian-data -mlittle-endian-data -msmall-data -msim
740 -mno-sim -mas100-syntax -mno-as100-syntax -mrelax
741 -mmax-constant-size= -mint-register= -mpid -mallow-string-insns
742 -mno-allow-string-insns -mjsr -mno-warn-multiple-fast-interrupts
743 -msave-acc-in-interrupts
744 S/390 and zSeries Options -mtune=cpu-type -march=cpu-type
746 -mhard-float -msoft-float -mhard-dfp -mno-hard-dfp
747 -mlong-double-64 -mlong-double-128 -mbackchain -mno-backchain
748 -mpacked-stack -mno-packed-stack -msmall-exec -mno-small-exec
749 -mmvcle -mno-mvcle -m64 -m31 -mdebug -mno-debug -mesa -mzarch
750 -mhtm -mvx -mzvector -mtpf-trace -mno-tpf-trace -mfused-madd
751 -mno-fused-madd -mwarn-framesize -mwarn-dynamicstack -mstack-size
752 -mstack-guard -mhotpatch=halfwords,halfwords
753 Score Options -meb -mel -mnhwloop -muls -mmac -mscore5 -mscore5u
755 -mscore7 -mscore7d
756 SH Options -m1 -m2 -m2e -m2a-nofpu -m2a-single-only -m2a-single
758 -m2a -m3 -m3e -m4-nofpu -m4-single-only -m4-single -m4
759 -m4a-nofpu -m4a-single-only -m4a-single -m4a -m4al -mb -ml
760 -mdalign -mrelax -mbigtable -mfmovd -mrenesas -mno-renesas
761 -mnomacsave -mieee -mno-ieee -mbitops -misize
762 -minline-ic_invalidate -mpadstruct -mprefergot -musermode
763 -multcost=number -mdiv=strategy -mdivsi3_libfunc=name
764 -mfixed-range=register-range -maccumulate-outgoing-args
765 -matomic-model=atomic-model -mbranch-cost=num -mzdcbranch
766 -mno-zdcbranch -mcbranch-force-delay-slot -mfused-madd
767 -mno-fused-madd -mfsca -mno-fsca -mfsrra -mno-fsrra
768 -mpretend-cmove -mtas
769 Solaris 2 Options -mclear-hwcap -mno-clear-hwcap -mimpure-text
771 -mno-impure-text -pthreads
772 SPARC Options -mcpu=cpu-type -mtune=cpu-type -mcmodel=code-model
774 -mmemory-model=mem-model -m32 -m64 -mapp-regs -mno-app-regs
775 -mfaster-structs -mno-faster-structs -mflat -mno-flat -mfpu
776 -mno-fpu -mhard-float -msoft-float -mhard-quad-float
777 -msoft-quad-float -mstack-bias -mno-stack-bias -mstd-struct-return
778 -mno-std-struct-return -munaligned-doubles -mno-unaligned-doubles
779 -muser-mode -mno-user-mode -mv8plus -mno-v8plus -mvis -mno-vis
780 -mvis2 -mno-vis2 -mvis3 -mno-vis3 -mvis4 -mno-vis4 -mvis4b
781 -mno-vis4b -mcbcond -mno-cbcond -mfmaf -mno-fmaf -mfsmuld
782 -mno-fsmuld -mpopc -mno-popc -msubxc -mno-subxc -mfix-at697f
783 -mfix-ut699 -mfix-ut700 -mfix-gr712rc -mlra -mno-lra
784 SPU Options -mwarn-reloc -merror-reloc -msafe-dma -munsafe-dma
786 -mbranch-hints -msmall-mem -mlarge-mem -mstdmain
787 -mfixed-range=register-range -mea32 -mea64
788 -maddress-space-conversion -mno-address-space-conversion
789 -mcache-size=cache-size -matomic-updates -mno-atomic-updates
790 System V Options -Qy -Qn -YP,paths -Ym,dir
792 TILE-Gx Options -mcpu=CPU -m32 -m64 -mbig-endian
794 -mlittle-endian -mcmodel=code-model
795 TILEPro Options -mcpu=cpu -m32
797 V850 Options -mlong-calls -mno-long-calls -mep -mno-ep
799 -mprolog-function -mno-prolog-function -mspace -mtda=n -msda=n
800 -mzda=n -mapp-regs -mno-app-regs -mdisable-callt
801 -mno-disable-callt -mv850e2v3 -mv850e2 -mv850e1 -mv850es -mv850e
802 -mv850 -mv850e3v5 -mloop -mrelax -mlong-jumps -msoft-float
803 -mhard-float -mgcc-abi -mrh850-abi -mbig-switch
804 VAX Options -mg -mgnu -munix
806 Visium Options -mdebug -msim -mfpu -mno-fpu -mhard-float
808 -msoft-float -mcpu=cpu-type -mtune=cpu-type -msv-mode
809 -muser-mode
810 VMS Options -mvms-return-codes -mdebug-main=prefix -mmalloc64
812 -mpointer-size=size
813 VxWorks Options -mrtp -non-static -Bstatic -Bdynamic -Xbind-lazy
815 -Xbind-now
816 x86 Options -mtune=cpu-type -march=cpu-type -mtune-ctrl=feature-
818 list -mdump-tune-features -mno-default -mfpmath=unit
819 -masm=dialect -mno-fancy-math-387 -mno-fp-ret-in-387 -m80387
820 -mhard-float -msoft-float -mno-wide-multiply -mrtd
821 -malign-double -mpreferred-stack-boundary=num
822 -mincoming-stack-boundary=num -mcld -mcx16 -msahf -mmovbe
823 -mcrc32 -mrecip -mrecip=opt -mvzeroupper -mprefer-avx128
824 -mprefer-vector-width=opt -mmmx -msse -msse2 -msse3 -mssse3
825 -msse4.1 -msse4.2 -msse4 -mavx -mavx2 -mavx512f -mavx512pf
826 -mavx512er -mavx512cd -mavx512vl -mavx512bw -mavx512dq
827 -mavx512ifma -mavx512vbmi -msha -maes -mpclmul -mfsgsbase
828 -mrdrnd -mf16c -mfma -mpconfig -mwbnoinvd -mprefetchwt1
829 -mclflushopt -mxsavec -mxsaves -msse4a -m3dnow -m3dnowa
830 -mpopcnt -mabm -mbmi -mtbm -mfma4 -mxop -mlzcnt -mbmi2
831 -mfxsr -mxsave -mxsaveopt -mrtm -mlwp -mmpx -mmwaitx -mclzero
832 -mpku -mthreads -mgfni -mvaes -mshstk -mforce-indirect-call
833 -mavx512vbmi2 -mvpclmulqdq -mavx512bitalg -mmovdiri -mmovdir64b
834 -mavx512vpopcntdq -mms-bitfields -mno-align-stringops
835 -minline-all-stringops -minline-stringops-dynamically
836 -mstringop-strategy=alg -mmemcpy-strategy=strategy
837 -mmemset-strategy=strategy -mpush-args -maccumulate-outgoing-args
838 -m128bit-long-double -m96bit-long-double -mlong-double-64
839 -mlong-double-80 -mlong-double-128 -mregparm=num -msseregparm
840 -mveclibabi=type -mvect8-ret-in-mem -mpc32 -mpc64 -mpc80
841 -mstackrealign -momit-leaf-frame-pointer -mno-red-zone
842 -mno-tls-direct-seg-refs -mcmodel=code-model -mabi=name
843 -maddress-mode=mode -m32 -m64 -mx32 -m16 -miamcu
844 -mlarge-data-threshold=num -msse2avx -mfentry -mrecord-mcount
845 -mnop-mcount -m8bit-idiv -mavx256-split-unaligned-load
846 -mavx256-split-unaligned-store -malign-data=type
847 -mstack-protector-guard=guard -mstack-protector-guard-reg=reg
848 -mstack-protector-guard-offset=offset
849 -mstack-protector-guard-symbol=symbol -mmitigate-rop
850 -mgeneral-regs-only -mcall-ms2sysv-xlogues -mindirect-branch=choice
851 -mfunction-return=choice -mindirect-branch-register
852 x86 Windows Options -mconsole -mcygwin -mno-cygwin -mdll
854 -mnop-fun-dllimport -mthread -municode -mwin32 -mwindows
855 -fno-set-stack-executable
856 Xstormy16 Options -msim
858 Xtensa Options -mconst16 -mno-const16 -mfused-madd
860 -mno-fused-madd -mforce-no-pic -mserialize-volatile
861 -mno-serialize-volatile -mtext-section-literals
862 -mno-text-section-literals -mauto-litpools -mno-auto-litpools
863 -mtarget-align -mno-target-align -mlongcalls -mno-longcalls
864 zSeries Options See S/390 and zSeries Options.
866 Options Controlling the Kind of Output
868 Compilation can involve up to four stages: preprocessing, compilation
869 proper, assembly and linking, always in that order. GCC is capable of
870 preprocessing and compiling several files either into several assembler
871 input files, or into one assembler input file; then each assembler
872 input file produces an object file, and linking combines all the object
873 files (those newly compiled, and those specified as input) into an
874 executable file.
875 For any given input file, the file name suffix determines what kind of
877 compilation is done:
878 file.c
880 C source code that must be preprocessed.
881 file.i
883 C source code that should not be preprocessed.
884 file.ii
886 C++ source code that should not be preprocessed.
887 file.m
889 Objective-C source code. Note that you must link with the libobjc
890 library to make an Objective-C program work.
891 file.mi
893 Objective-C source code that should not be preprocessed.
894 file.mm
896 file.M
897 Objective-C++ source code. Note that you must link with the
898 libobjc library to make an Objective-C++ program work. Note that
899 .M refers to a literal capital M.
900 file.mii
902 Objective-C++ source code that should not be preprocessed.
903 file.h
905 C, C++, Objective-C or Objective-C++ header file to be turned into
906 a precompiled header (default), or C, C++ header file to be turned
907 into an Ada spec (via the -fdump-ada-spec switch).
908 file.cc
910 file.cp
911 file.cxx
912 file.cpp
913 file.CPP
914 file.c++
915 file.C
916 C++ source code that must be preprocessed. Note that in .cxx, the
917 last two letters must both be literally x. Likewise, .C refers to
918 a literal capital C.
919 file.mm
921 file.M
922 Objective-C++ source code that must be preprocessed.
923 file.mii
925 Objective-C++ source code that should not be preprocessed.
926 file.hh
928 file.H
929 file.hp
930 file.hxx
931 file.hpp
932 file.HPP
933 file.h++
934 file.tcc
935 C++ header file to be turned into a precompiled header or Ada spec.
936 file.f
938 file.for
939 file.ftn
940 Fixed form Fortran source code that should not be preprocessed.
941 file.F
943 file.FOR
944 file.fpp
945 file.FPP
946 file.FTN
947 Fixed form Fortran source code that must be preprocessed (with the
948 traditional preprocessor).
949 file.f90
951 file.f95
952 file.f03
953 file.f08
954 Free form Fortran source code that should not be preprocessed.
955 file.F90
957 file.F95
958 file.F03
959 file.F08
960 Free form Fortran source code that must be preprocessed (with the
961 traditional preprocessor).
962 file.go
964 Go source code.
965 file.brig
967 BRIG files (binary representation of HSAIL).
968 file.ads
970 Ada source code file that contains a library unit declaration (a
971 declaration of a package, subprogram, or generic, or a generic
972 instantiation), or a library unit renaming declaration (a package,
973 generic, or subprogram renaming declaration). Such files are also
974 called specs.
975 file.adb
977 Ada source code file containing a library unit body (a subprogram
978 or package body). Such files are also called bodies.
979 file.s
981 Assembler code.
982 file.S
984 file.sx
985 Assembler code that must be preprocessed.
986 other
988 An object file to be fed straight into linking. Any file name with
989 no recognized suffix is treated this way.
990 You can specify the input language explicitly with the -x option:
992 -x language
994 Specify explicitly the language for the following input files
995 (rather than letting the compiler choose a default based on the
996 file name suffix). This option applies to all following input
997 files until the next -x option. Possible values for language are:
998 c c-header cpp-output
1000 c++ c++-header c++-cpp-output
1001 objective-c objective-c-header objective-c-cpp-output
1002 objective-c++ objective-c++-header objective-c++-cpp-output
1003 assembler assembler-with-cpp
1004 ada
1005 f77 f77-cpp-input f95 f95-cpp-input
1006 go
1007 brig
1008 -x none
1010 Turn off any specification of a language, so that subsequent files
1011 are handled according to their file name suffixes (as they are if
1012 -x has not been used at all).
1013 If you only want some of the stages of compilation, you can use -x (or
1015 filename suffixes) to tell gcc where to start, and one of the options
1016 -c, -S, or -E to say where gcc is to stop. Note that some combinations
1017 (for example, -x cpp-output -E) instruct gcc to do nothing at all.
1018 -c Compile or assemble the source files, but do not link. The linking
1020 stage simply is not done. The ultimate output is in the form of an
1021 object file for each source file.
1022 By default, the object file name for a source file is made by
1024 replacing the suffix .c, .i, .s, etc., with .o.
1025 Unrecognized input files, not requiring compilation or assembly,
1027 are ignored.
1028 -S Stop after the stage of compilation proper; do not assemble. The
1030 output is in the form of an assembler code file for each non-
1031 assembler input file specified.
1032 By default, the assembler file name for a source file is made by
1034 replacing the suffix .c, .i, etc., with .s.
1035 Input files that don't require compilation are ignored.
1037 -E Stop after the preprocessing stage; do not run the compiler proper.
1039 The output is in the form of preprocessed source code, which is
1040 sent to the standard output.
1041 Input files that don't require preprocessing are ignored.
1043 -o file
1045 Place output in file file. This applies to whatever sort of output
1046 is being produced, whether it be an executable file, an object
1047 file, an assembler file or preprocessed C code.
1048 If -o is not specified, the default is to put an executable file in
1050 a.out, the object file for source.suffix in source.o, its assembler
1051 file in source.s, a precompiled header file in source.suffix.gch,
1052 and all preprocessed C source on standard output.
1053 -v Print (on standard error output) the commands executed to run the
1055 stages of compilation. Also print the version number of the
1056 compiler driver program and of the preprocessor and the compiler
1057 proper.
1058 -###
1060 Like -v except the commands are not executed and arguments are
1061 quoted unless they contain only alphanumeric characters or "./-_".
1062 This is useful for shell scripts to capture the driver-generated
1063 command lines.
1064 --help
1066 Print (on the standard output) a description of the command-line
1067 options understood by gcc. If the -v option is also specified then
1068 --help is also passed on to the various processes invoked by gcc,
1069 so that they can display the command-line options they accept. If
1070 the -Wextra option has also been specified (prior to the --help
1071 option), then command-line options that have no documentation
1072 associated with them are also displayed.
1073 --target-help
1075 Print (on the standard output) a description of target-specific
1076 command-line options for each tool. For some targets extra target-
1077 specific information may also be printed.
1078 --help={class|[^]qualifier}[,...]
1080 Print (on the standard output) a description of the command-line
1081 options understood by the compiler that fit into all specified
1082 classes and qualifiers. These are the supported classes:
1083 optimizers
1085 Display all of the optimization options supported by the
1086 compiler.
1087 warnings
1089 Display all of the options controlling warning messages
1090 produced by the compiler.
1091 target
1093 Display target-specific options. Unlike the --target-help
1094 option however, target-specific options of the linker and
1095 assembler are not displayed. This is because those tools do
1096 not currently support the extended --help= syntax.
1097 params
1099 Display the values recognized by the --param option.
1100 language
1102 Display the options supported for language, where language is
1103 the name of one of the languages supported in this version of
1104 GCC.
1105 common
1107 Display the options that are common to all languages.
1108 These are the supported qualifiers:
1110 undocumented
1112 Display only those options that are undocumented.
1113 joined
1115 Display options taking an argument that appears after an equal
1116 sign in the same continuous piece of text, such as:
1117 --help=target.
1118 separate
1120 Display options taking an argument that appears as a separate
1121 word following the original option, such as: -o output-file.
1122 Thus for example to display all the undocumented target-specific
1124 switches supported by the compiler, use:
1125 --help=target,undocumented
1127 The sense of a qualifier can be inverted by prefixing it with the ^
1129 character, so for example to display all binary warning options
1130 (i.e., ones that are either on or off and that do not take an
1131 argument) that have a description, use:
1132 --help=warnings,^joined,^undocumented
1134 The argument to --help= should not consist solely of inverted
1136 qualifiers.
1137 Combining several classes is possible, although this usually
1139 restricts the output so much that there is nothing to display. One
1140 case where it does work, however, is when one of the classes is
1141 target. For example, to display all the target-specific
1142 optimization options, use:
1143 --help=target,optimizers
1145 The --help= option can be repeated on the command line. Each
1147 successive use displays its requested class of options, skipping
1148 those that have already been displayed.
1149 If the -Q option appears on the command line before the --help=
1151 option, then the descriptive text displayed by --help= is changed.
1152 Instead of describing the displayed options, an indication is given
1153 as to whether the option is enabled, disabled or set to a specific
1154 value (assuming that the compiler knows this at the point where the
1155 --help= option is used).
1156 Here is a truncated example from the ARM port of gcc:
1158 % gcc -Q -mabi=2 --help=target -c
1160 The following options are target specific:
1161 -mabi= 2
1162 -mabort-on-noreturn [disabled]
1163 -mapcs [disabled]
1164 The output is sensitive to the effects of previous command-line
1166 options, so for example it is possible to find out which
1167 optimizations are enabled at -O2 by using:
1168 -Q -O2 --help=optimizers
1170 Alternatively you can discover which binary optimizations are
1172 enabled by -O3 by using:
1173 gcc -c -Q -O3 --help=optimizers > /tmp/O3-opts
1175 gcc -c -Q -O2 --help=optimizers > /tmp/O2-opts
1176 diff /tmp/O2-opts /tmp/O3-opts | grep enabled
1177 --version
1179 Display the version number and copyrights of the invoked GCC.
1180 -pass-exit-codes
1182 Normally the gcc program exits with the code of 1 if any phase of
1183 the compiler returns a non-success return code. If you specify
1184 -pass-exit-codes, the gcc program instead returns with the
1185 numerically highest error produced by any phase returning an error
1186 indication. The C, C++, and Fortran front ends return 4 if an
1187 internal compiler error is encountered.
1188 -pipe
1190 Use pipes rather than temporary files for communication between the
1191 various stages of compilation. This fails to work on some systems
1192 where the assembler is unable to read from a pipe; but the GNU
1193 assembler has no trouble.
1194 -specs=file
1196 Process file after the compiler reads in the standard specs file,
1197 in order to override the defaults which the gcc driver program uses
1198 when determining what switches to pass to cc1, cc1plus, as, ld,
1199 etc. More than one -specs=file can be specified on the command
1200 line, and they are processed in order, from left to right.
1201 -wrapper
1203 Invoke all subcommands under a wrapper program. The name of the
1204 wrapper program and its parameters are passed as a comma separated
1205 list.
1206 gcc -c t.c -wrapper gdb,--args
1208 This invokes all subprograms of gcc under gdb --args, thus the
1210 invocation of cc1 is gdb --args cc1 ....
1211 -ffile-prefix-map=old=new
1213 When compiling files residing in directory old, record any
1214 references to them in the result of the compilation as if the files
1215 resided in directory new instead. Specifying this option is
1216 equivalent to specifying all the individual -f*-prefix-map options.
1217 This can be used to make reproducible builds that are location
1218 independent. See also -fmacro-prefix-map and -fdebug-prefix-map.
1219 -fplugin=name.so
1221 Load the plugin code in file name.so, assumed to be a shared object
1222 to be dlopen'd by the compiler. The base name of the shared object
1223 file is used to identify the plugin for the purposes of argument
1224 parsing (See -fplugin-arg-name-key=value below). Each plugin
1225 should define the callback functions specified in the Plugins API.
1226 -fplugin-arg-name-key=value
1228 Define an argument called key with a value of value for the plugin
1229 called name.
1230 -fdump-ada-spec[-slim]
1232 For C and C++ source and include files, generate corresponding Ada
1233 specs.
1234 -fada-spec-parent=unit
1236 In conjunction with -fdump-ada-spec[-slim] above, generate Ada
1237 specs as child units of parent unit.
1238 -fdump-go-spec=file
1240 For input files in any language, generate corresponding Go
1241 declarations in file. This generates Go "const", "type", "var",
1242 and "func" declarations which may be a useful way to start writing
1243 a Go interface to code written in some other language.
1244 @file
1246 Read command-line options from file. The options read are inserted
1247 in place of the original @file option. If file does not exist, or
1248 cannot be read, then the option will be treated literally, and not
1249 removed.
1250 Options in file are separated by whitespace. A whitespace
1252 character may be included in an option by surrounding the entire
1253 option in either single or double quotes. Any character (including
1254 a backslash) may be included by prefixing the character to be
1255 included with a backslash. The file may itself contain additional
1256 @file options; any such options will be processed recursively.
1257 Compiling C++ Programs
1259 C++ source files conventionally use one of the suffixes .C, .cc, .cpp,
1260 .CPP, .c++, .cp, or .cxx; C++ header files often use .hh, .hpp, .H, or
1261 (for shared template code) .tcc; and preprocessed C++ files use the
1262 suffix .ii. GCC recognizes files with these names and compiles them as
1263 C++ programs even if you call the compiler the same way as for
1264 compiling C programs (usually with the name gcc).
1265 However, the use of gcc does not add the C++ library. g++ is a program
1267 that calls GCC and automatically specifies linking against the C++
1268 library. It treats .c, .h and .i files as C++ source files instead of
1269 C source files unless -x is used. This program is also useful when
1270 precompiling a C header file with a .h extension for use in C++
1271 compilations. On many systems, g++ is also installed with the name
1272 c++.
1273 When you compile C++ programs, you may specify many of the same
1275 command-line options that you use for compiling programs in any
1276 language; or command-line options meaningful for C and related
1277 languages; or options that are meaningful only for C++ programs.
1278 Options Controlling C Dialect
1280 The following options control the dialect of C (or languages derived
1281 from C, such as C++, Objective-C and Objective-C++) that the compiler
1282 accepts:
1283 -ansi
1285 In C mode, this is equivalent to -std=c90. In C++ mode, it is
1286 equivalent to -std=c++98.
1287 This turns off certain features of GCC that are incompatible with
1289 ISO C90 (when compiling C code), or of standard C++ (when compiling
1290 C++ code), such as the "asm" and "typeof" keywords, and predefined
1291 macros such as "unix" and "vax" that identify the type of system
1292 you are using. It also enables the undesirable and rarely used ISO
1293 trigraph feature. For the C compiler, it disables recognition of
1294 C++ style // comments as well as the "inline" keyword.
1295 The alternate keywords "__asm__", "__extension__", "__inline__" and
1297 "__typeof__" continue to work despite -ansi. You would not want to
1298 use them in an ISO C program, of course, but it is useful to put
1299 them in header files that might be included in compilations done
1300 with -ansi. Alternate predefined macros such as "__unix__" and
1301 "__vax__" are also available, with or without -ansi.
1302 The -ansi option does not cause non-ISO programs to be rejected
1304 gratuitously. For that, -Wpedantic is required in addition to
1305 -ansi.
1306 The macro "__STRICT_ANSI__" is predefined when the -ansi option is
1308 used. Some header files may notice this macro and refrain from
1309 declaring certain functions or defining certain macros that the ISO
1310 standard doesn't call for; this is to avoid interfering with any
1311 programs that might use these names for other things.
1312 Functions that are normally built in but do not have semantics
1314 defined by ISO C (such as "alloca" and "ffs") are not built-in
1315 functions when -ansi is used.
1316 -std=
1318 Determine the language standard. This option is currently only
1319 supported when compiling C or C++.
1320 The compiler can accept several base standards, such as c90 or
1322 c++98, and GNU dialects of those standards, such as gnu90 or
1323 gnu++98. When a base standard is specified, the compiler accepts
1324 all programs following that standard plus those using GNU
1325 extensions that do not contradict it. For example, -std=c90 turns
1326 off certain features of GCC that are incompatible with ISO C90,
1327 such as the "asm" and "typeof" keywords, but not other GNU
1328 extensions that do not have a meaning in ISO C90, such as omitting
1329 the middle term of a "?:" expression. On the other hand, when a GNU
1330 dialect of a standard is specified, all features supported by the
1331 compiler are enabled, even when those features change the meaning
1332 of the base standard. As a result, some strict-conforming programs
1333 may be rejected. The particular standard is used by -Wpedantic to
1334 identify which features are GNU extensions given that version of
1335 the standard. For example -std=gnu90 -Wpedantic warns about C++
1336 style // comments, while -std=gnu99 -Wpedantic does not.
1337 A value for this option must be provided; possible values are
1339 c90
1341 c89
1342 iso9899:1990
1343 Support all ISO C90 programs (certain GNU extensions that
1344 conflict with ISO C90 are disabled). Same as -ansi for C code.
1345 iso9899:199409
1347 ISO C90 as modified in amendment 1.
1348 c99
1350 c9x
1351 iso9899:1999
1352 iso9899:199x
1353 ISO C99. This standard is substantially completely supported,
1354 modulo bugs and floating-point issues (mainly but not entirely
1355 relating to optional C99 features from Annexes F and G). See
1356 <http://gcc.gnu.org/c99status.html> for more information. The
1357 names c9x and iso9899:199x are deprecated.
1358 c11
1360 c1x
1361 iso9899:2011
1362 ISO C11, the 2011 revision of the ISO C standard. This
1363 standard is substantially completely supported, modulo bugs,
1364 floating-point issues (mainly but not entirely relating to
1365 optional C11 features from Annexes F and G) and the optional
1366 Annexes K (Bounds-checking interfaces) and L (Analyzability).
1367 The name c1x is deprecated.
1368 c17
1370 c18
1371 iso9899:2017
1372 iso9899:2018
1373 ISO C17, the 2017 revision of the ISO C standard (expected to
1374 be published in 2018). This standard is same as C11 except for
1375 corrections of defects (all of which are also applied with
1376 -std=c11) and a new value of "__STDC_VERSION__", and so is
1377 supported to the same extent as C11.
1378 gnu90
1380 gnu89
1381 GNU dialect of ISO C90 (including some C99 features).
1382 gnu99
1384 gnu9x
1385 GNU dialect of ISO C99. The name gnu9x is deprecated.
1386 gnu11
1388 gnu1x
1389 GNU dialect of ISO C11. The name gnu1x is deprecated.
1390 gnu17
1392 gnu18
1393 GNU dialect of ISO C17. This is the default for C code.
1394 c++98
1396 c++03
1397 The 1998 ISO C++ standard plus the 2003 technical corrigendum
1398 and some additional defect reports. Same as -ansi for C++ code.
1399 gnu++98
1401 gnu++03
1402 GNU dialect of -std=c++98.
1403 c++11
1405 c++0x
1406 The 2011 ISO C++ standard plus amendments. The name c++0x is
1407 deprecated.
1408 gnu++11
1410 gnu++0x
1411 GNU dialect of -std=c++11. The name gnu++0x is deprecated.
1412 c++14
1414 c++1y
1415 The 2014 ISO C++ standard plus amendments. The name c++1y is
1416 deprecated.
1417 gnu++14
1419 gnu++1y
1420 GNU dialect of -std=c++14. This is the default for C++ code.
1421 The name gnu++1y is deprecated.
1422 c++17
1424 c++1z
1425 The 2017 ISO C++ standard plus amendments. The name c++1z is
1426 deprecated.
1427 gnu++17
1429 gnu++1z
1430 GNU dialect of -std=c++17. The name gnu++1z is deprecated.
1431 c++2a
1433 The next revision of the ISO C++ standard, tentatively planned
1434 for 2020. Support is highly experimental, and will almost
1435 certainly change in incompatible ways in future releases.
1436 gnu++2a
1438 GNU dialect of -std=c++2a. Support is highly experimental, and
1439 will almost certainly change in incompatible ways in future
1440 releases.
1441 -fgnu89-inline
1443 The option -fgnu89-inline tells GCC to use the traditional GNU
1444 semantics for "inline" functions when in C99 mode.
1445 Using this option is roughly equivalent to adding the "gnu_inline"
1447 function attribute to all inline functions.
1448 The option -fno-gnu89-inline explicitly tells GCC to use the C99
1450 semantics for "inline" when in C99 or gnu99 mode (i.e., it
1451 specifies the default behavior). This option is not supported in
1452 -std=c90 or -std=gnu90 mode.
1453 The preprocessor macros "__GNUC_GNU_INLINE__" and
1455 "__GNUC_STDC_INLINE__" may be used to check which semantics are in
1456 effect for "inline" functions.
1457 -fpermitted-flt-eval-methods=style
1459 ISO/IEC TS 18661-3 defines new permissible values for
1460 "FLT_EVAL_METHOD" that indicate that operations and constants with
1461 a semantic type that is an interchange or extended format should be
1462 evaluated to the precision and range of that type. These new
1463 values are a superset of those permitted under C99/C11, which does
1464 not specify the meaning of other positive values of
1465 "FLT_EVAL_METHOD". As such, code conforming to C11 may not have
1466 been written expecting the possibility of the new values.
1467 -fpermitted-flt-eval-methods specifies whether the compiler should
1469 allow only the values of "FLT_EVAL_METHOD" specified in C99/C11, or
1470 the extended set of values specified in ISO/IEC TS 18661-3.
1471 style is either "c11" or "ts-18661-3" as appropriate.
1473 The default when in a standards compliant mode (-std=c11 or
1475 similar) is -fpermitted-flt-eval-methods=c11. The default when in
1476 a GNU dialect (-std=gnu11 or similar) is
1477 -fpermitted-flt-eval-methods=ts-18661-3.
1478 -aux-info filename
1480 Output to the given filename prototyped declarations for all
1481 functions declared and/or defined in a translation unit, including
1482 those in header files. This option is silently ignored in any
1483 language other than C.
1484 Besides declarations, the file indicates, in comments, the origin
1486 of each declaration (source file and line), whether the declaration
1487 was implicit, prototyped or unprototyped (I, N for new or O for
1488 old, respectively, in the first character after the line number and
1489 the colon), and whether it came from a declaration or a definition
1490 (C or F, respectively, in the following character). In the case of
1491 function definitions, a K&R-style list of arguments followed by
1492 their declarations is also provided, inside comments, after the
1493 declaration.
1494 -fallow-parameterless-variadic-functions
1496 Accept variadic functions without named parameters.
1497 Although it is possible to define such a function, this is not very
1499 useful as it is not possible to read the arguments. This is only
1500 supported for C as this construct is allowed by C++.
1501 -fno-asm
1503 Do not recognize "asm", "inline" or "typeof" as a keyword, so that
1504 code can use these words as identifiers. You can use the keywords
1505 "__asm__", "__inline__" and "__typeof__" instead. -ansi implies
1506 -fno-asm.
1507 In C++, this switch only affects the "typeof" keyword, since "asm"
1509 and "inline" are standard keywords. You may want to use the
1510 -fno-gnu-keywords flag instead, which has the same effect. In C99
1511 mode (-std=c99 or -std=gnu99), this switch only affects the "asm"
1512 and "typeof" keywords, since "inline" is a standard keyword in ISO
1513 C99.
1514 -fno-builtin
1516 -fno-builtin-function
1517 Don't recognize built-in functions that do not begin with
1518 __builtin_ as prefix.
1519 GCC normally generates special code to handle certain built-in
1521 functions more efficiently; for instance, calls to "alloca" may
1522 become single instructions which adjust the stack directly, and
1523 calls to "memcpy" may become inline copy loops. The resulting code
1524 is often both smaller and faster, but since the function calls no
1525 longer appear as such, you cannot set a breakpoint on those calls,
1526 nor can you change the behavior of the functions by linking with a
1527 different library. In addition, when a function is recognized as a
1528 built-in function, GCC may use information about that function to
1529 warn about problems with calls to that function, or to generate
1530 more efficient code, even if the resulting code still contains
1531 calls to that function. For example, warnings are given with
1532 -Wformat for bad calls to "printf" when "printf" is built in and
1533 "strlen" is known not to modify global memory.
1534 With the -fno-builtin-function option only the built-in function
1536 function is disabled. function must not begin with __builtin_. If
1537 a function is named that is not built-in in this version of GCC,
1538 this option is ignored. There is no corresponding
1539 -fbuiltin-function option; if you wish to enable built-in functions
1540 selectively when using -fno-builtin or -ffreestanding, you may
1541 define macros such as:
1542 #define abs(n) __builtin_abs ((n))
1544 #define strcpy(d, s) __builtin_strcpy ((d), (s))
1545 -fgimple
1547 Enable parsing of function definitions marked with "__GIMPLE".
1548 This is an experimental feature that allows unit testing of GIMPLE
1549 passes.
1550 -fhosted
1552 Assert that compilation targets a hosted environment. This implies
1553 -fbuiltin. A hosted environment is one in which the entire
1554 standard library is available, and in which "main" has a return
1555 type of "int". Examples are nearly everything except a kernel.
1556 This is equivalent to -fno-freestanding.
1557 -ffreestanding
1559 Assert that compilation targets a freestanding environment. This
1560 implies -fno-builtin. A freestanding environment is one in which
1561 the standard library may not exist, and program startup may not
1562 necessarily be at "main". The most obvious example is an OS
1563 kernel. This is equivalent to -fno-hosted.
1564 -fopenacc
1566 Enable handling of OpenACC directives "#pragma acc" in C/C++ and
1567 "!$acc" in Fortran. When -fopenacc is specified, the compiler
1568 generates accelerated code according to the OpenACC Application
1569 Programming Interface v2.0 <https://www.openacc.org>. This option
1570 implies -pthread, and thus is only supported on targets that have
1571 support for -pthread.
1572 -fopenacc-dim=geom
1574 Specify default compute dimensions for parallel offload regions
1575 that do not explicitly specify. The geom value is a triple of
1576 ':'-separated sizes, in order 'gang', 'worker' and, 'vector'. A
1577 size can be omitted, to use a target-specific default value.
1578 -fopenmp
1580 Enable handling of OpenMP directives "#pragma omp" in C/C++ and
1581 "!$omp" in Fortran. When -fopenmp is specified, the compiler
1582 generates parallel code according to the OpenMP Application Program
1583 Interface v4.5 <http://www.openmp.org/>. This option implies
1584 -pthread, and thus is only supported on targets that have support
1585 for -pthread. -fopenmp implies -fopenmp-simd.
1586 -fopenmp-simd
1588 Enable handling of OpenMP's SIMD directives with "#pragma omp" in
1589 C/C++ and "!$omp" in Fortran. Other OpenMP directives are ignored.
1590 -fgnu-tm
1592 When the option -fgnu-tm is specified, the compiler generates code
1593 for the Linux variant of Intel's current Transactional Memory ABI
1594 specification document (Revision 1.1, May 6 2009). This is an
1595 experimental feature whose interface may change in future versions
1596 of GCC, as the official specification changes. Please note that
1597 not all architectures are supported for this feature.
1598 For more information on GCC's support for transactional memory,
1600 Note that the transactional memory feature is not supported with
1602 non-call exceptions (-fnon-call-exceptions).
1603 -fms-extensions
1605 Accept some non-standard constructs used in Microsoft header files.
1606 In C++ code, this allows member names in structures to be similar
1608 to previous types declarations.
1609 typedef int UOW;
1611 struct ABC {
1612 UOW UOW;
1613 };
1614 Some cases of unnamed fields in structures and unions are only
1616 accepted with this option.
1617 Note that this option is off for all targets but x86 targets using
1619 ms-abi.
1620 -fplan9-extensions
1622 Accept some non-standard constructs used in Plan 9 code.
1623 This enables -fms-extensions, permits passing pointers to
1625 structures with anonymous fields to functions that expect pointers
1626 to elements of the type of the field, and permits referring to
1627 anonymous fields declared using a typedef. This is only
1628 supported for C, not C++.
1629 -fcond-mismatch
1631 Allow conditional expressions with mismatched types in the second
1632 and third arguments. The value of such an expression is void.
1633 This option is not supported for C++.
1634 -flax-vector-conversions
1636 Allow implicit conversions between vectors with differing numbers
1637 of elements and/or incompatible element types. This option should
1638 not be used for new code.
1639 -funsigned-char
1641 Let the type "char" be unsigned, like "unsigned char".
1642 Each kind of machine has a default for what "char" should be. It
1644 is either like "unsigned char" by default or like "signed char" by
1645 default.
1646 Ideally, a portable program should always use "signed char" or
1648 "unsigned char" when it depends on the signedness of an object.
1649 But many programs have been written to use plain "char" and expect
1650 it to be signed, or expect it to be unsigned, depending on the
1651 machines they were written for. This option, and its inverse, let
1652 you make such a program work with the opposite default.
1653 The type "char" is always a distinct type from each of "signed
1655 char" or "unsigned char", even though its behavior is always just
1656 like one of those two.
1657 -fsigned-char
1659 Let the type "char" be signed, like "signed char".
1660 Note that this is equivalent to -fno-unsigned-char, which is the
1662 negative form of -funsigned-char. Likewise, the option
1663 -fno-signed-char is equivalent to -funsigned-char.
1664 -fsigned-bitfields
1666 -funsigned-bitfields
1667 -fno-signed-bitfields
1668 -fno-unsigned-bitfields
1669 These options control whether a bit-field is signed or unsigned,
1670 when the declaration does not use either "signed" or "unsigned".
1671 By default, such a bit-field is signed, because this is consistent:
1672 the basic integer types such as "int" are signed types.
1673 -fsso-struct=endianness
1675 Set the default scalar storage order of structures and unions to
1676 the specified endianness. The accepted values are big-endian,
1677 little-endian and native for the native endianness of the target
1678 (the default). This option is not supported for C++.
1679 Warning: the -fsso-struct switch causes GCC to generate code that
1681 is not binary compatible with code generated without it if the
1682 specified endianness is not the native endianness of the target.
1683 Options Controlling C++ Dialect
1685 This section describes the command-line options that are only
1686 meaningful for C++ programs. You can also use most of the GNU compiler
1687 options regardless of what language your program is in. For example,
1688 you might compile a file firstClass.C like this:
1689 g++ -g -fstrict-enums -O -c firstClass.C
1691 In this example, only -fstrict-enums is an option meant only for C++
1693 programs; you can use the other options with any language supported by
1694 GCC.
1695 Some options for compiling C programs, such as -std, are also relevant
1697 for C++ programs.
1698 Here is a list of options that are only for compiling C++ programs:
1700 -fabi-version=n
1702 Use version n of the C++ ABI. The default is version 0.
1703 Version 0 refers to the version conforming most closely to the C++
1705 ABI specification. Therefore, the ABI obtained using version 0
1706 will change in different versions of G++ as ABI bugs are fixed.
1707 Version 1 is the version of the C++ ABI that first appeared in G++
1709 3.2.
1710 Version 2 is the version of the C++ ABI that first appeared in G++
1712 3.4, and was the default through G++ 4.9.
1713 Version 3 corrects an error in mangling a constant address as a
1715 template argument.
1716 Version 4, which first appeared in G++ 4.5, implements a standard
1718 mangling for vector types.
1719 Version 5, which first appeared in G++ 4.6, corrects the mangling
1721 of attribute const/volatile on function pointer types, decltype of
1722 a plain decl, and use of a function parameter in the declaration of
1723 another parameter.
1724 Version 6, which first appeared in G++ 4.7, corrects the promotion
1726 behavior of C++11 scoped enums and the mangling of template
1727 argument packs, const/static_cast, prefix ++ and --, and a class
1728 scope function used as a template argument.
1729 Version 7, which first appeared in G++ 4.8, that treats nullptr_t
1731 as a builtin type and corrects the mangling of lambdas in default
1732 argument scope.
1733 Version 8, which first appeared in G++ 4.9, corrects the
1735 substitution behavior of function types with function-cv-
1736 qualifiers.
1737 Version 9, which first appeared in G++ 5.2, corrects the alignment
1739 of "nullptr_t".
1740 Version 10, which first appeared in G++ 6.1, adds mangling of
1742 attributes that affect type identity, such as ia32 calling
1743 convention attributes (e.g. stdcall).
1744 Version 11, which first appeared in G++ 7, corrects the mangling of
1746 sizeof... expressions and operator names. For multiple entities
1747 with the same name within a function, that are declared in
1748 different scopes, the mangling now changes starting with the
1749 twelfth occurrence. It also implies -fnew-inheriting-ctors.
1750 Version 12, which first appeared in G++ 8, corrects the calling
1752 conventions for empty classes on the x86_64 target and for classes
1753 with only deleted copy/move constructors. It accidentally changes
1754 the calling convention for classes with a deleted copy constructor
1755 and a trivial move constructor.
1756 Version 13, which first appeared in G++ 8.2, fixes the accidental
1758 change in version 12.
1759 See also -Wabi.
1761 -fabi-compat-version=n
1763 On targets that support strong aliases, G++ works around mangling
1764 changes by creating an alias with the correct mangled name when
1765 defining a symbol with an incorrect mangled name. This switch
1766 specifies which ABI version to use for the alias.
1767 With -fabi-version=0 (the default), this defaults to 11 (GCC 7
1769 compatibility). If another ABI version is explicitly selected,
1770 this defaults to 0. For compatibility with GCC versions 3.2
1771 through 4.9, use -fabi-compat-version=2.
1772 If this option is not provided but -Wabi=n is, that version is used
1774 for compatibility aliases. If this option is provided along with
1775 -Wabi (without the version), the version from this option is used
1776 for the warning.
1777 -fno-access-control
1779 Turn off all access checking. This switch is mainly useful for
1780 working around bugs in the access control code.
1781 -faligned-new
1783 Enable support for C++17 "new" of types that require more alignment
1784 than "void* ::operator new(std::size_t)" provides. A numeric
1785 argument such as "-faligned-new=32" can be used to specify how much
1786 alignment (in bytes) is provided by that function, but few users
1787 will need to override the default of "alignof(std::max_align_t)".
1788 This flag is enabled by default for -std=c++17.
1790 -fcheck-new
1792 Check that the pointer returned by "operator new" is non-null
1793 before attempting to modify the storage allocated. This check is
1794 normally unnecessary because the C++ standard specifies that
1795 "operator new" only returns 0 if it is declared "throw()", in which
1796 case the compiler always checks the return value even without this
1797 option. In all other cases, when "operator new" has a non-empty
1798 exception specification, memory exhaustion is signalled by throwing
1799 "std::bad_alloc". See also new (nothrow).
1800 -fconcepts
1802 Enable support for the C++ Extensions for Concepts Technical
1803 Specification, ISO 19217 (2015), which allows code like
1804 template <class T> concept bool Addable = requires (T t) { t + t; };
1806 template <Addable T> T add (T a, T b) { return a + b; }
1807 -fconstexpr-depth=n
1809 Set the maximum nested evaluation depth for C++11 constexpr
1810 functions to n. A limit is needed to detect endless recursion
1811 during constant expression evaluation. The minimum specified by
1812 the standard is 512.
1813 -fconstexpr-loop-limit=n
1815 Set the maximum number of iterations for a loop in C++14 constexpr
1816 functions to n. A limit is needed to detect infinite loops during
1817 constant expression evaluation. The default is 262144 (1<<18).
1818 -fdeduce-init-list
1820 Enable deduction of a template type parameter as
1821 "std::initializer_list" from a brace-enclosed initializer list,
1822 i.e.
1823 template <class T> auto forward(T t) -> decltype (realfn (t))
1825 {
1826 return realfn (t);
1827 }
1828 void f()
1830 {
1831 forward({1,2}); // call forward<std::initializer_list<int>>
1832 }
1833 This deduction was implemented as a possible extension to the
1835 originally proposed semantics for the C++11 standard, but was not
1836 part of the final standard, so it is disabled by default. This
1837 option is deprecated, and may be removed in a future version of
1838 G++.
1839 -ffriend-injection
1841 Inject friend functions into the enclosing namespace, so that they
1842 are visible outside the scope of the class in which they are
1843 declared. Friend functions were documented to work this way in the
1844 old Annotated C++ Reference Manual. However, in ISO C++ a friend
1845 function that is not declared in an enclosing scope can only be
1846 found using argument dependent lookup. GCC defaults to the
1847 standard behavior.
1848 This option is deprecated and will be removed.
1850 -fno-elide-constructors
1852 The C++ standard allows an implementation to omit creating a
1853 temporary that is only used to initialize another object of the
1854 same type. Specifying this option disables that optimization, and
1855 forces G++ to call the copy constructor in all cases. This option
1856 also causes G++ to call trivial member functions which otherwise
1857 would be expanded inline.
1858 In C++17, the compiler is required to omit these temporaries, but
1860 this option still affects trivial member functions.
1861 -fno-enforce-eh-specs
1863 Don't generate code to check for violation of exception
1864 specifications at run time. This option violates the C++ standard,
1865 but may be useful for reducing code size in production builds, much
1866 like defining "NDEBUG". This does not give user code permission to
1867 throw exceptions in violation of the exception specifications; the
1868 compiler still optimizes based on the specifications, so throwing
1869 an unexpected exception results in undefined behavior at run time.
1870 -fextern-tls-init
1872 -fno-extern-tls-init
1873 The C++11 and OpenMP standards allow "thread_local" and
1874 "threadprivate" variables to have dynamic (runtime) initialization.
1875 To support this, any use of such a variable goes through a wrapper
1876 function that performs any necessary initialization. When the use
1877 and definition of the variable are in the same translation unit,
1878 this overhead can be optimized away, but when the use is in a
1879 different translation unit there is significant overhead even if
1880 the variable doesn't actually need dynamic initialization. If the
1881 programmer can be sure that no use of the variable in a non-
1882 defining TU needs to trigger dynamic initialization (either because
1883 the variable is statically initialized, or a use of the variable in
1884 the defining TU will be executed before any uses in another TU),
1885 they can avoid this overhead with the -fno-extern-tls-init option.
1886 On targets that support symbol aliases, the default is
1888 -fextern-tls-init. On targets that do not support symbol aliases,
1889 the default is -fno-extern-tls-init.
1890 -ffor-scope
1892 -fno-for-scope
1893 If -ffor-scope is specified, the scope of variables declared in a
1894 for-init-statement is limited to the "for" loop itself, as
1895 specified by the C++ standard. If -fno-for-scope is specified, the
1896 scope of variables declared in a for-init-statement extends to the
1897 end of the enclosing scope, as was the case in old versions of G++,
1898 and other (traditional) implementations of C++.
1899 This option is deprecated and the associated non-standard
1901 functionality will be removed.
1902 -fno-gnu-keywords
1904 Do not recognize "typeof" as a keyword, so that code can use this
1905 word as an identifier. You can use the keyword "__typeof__"
1906 instead. This option is implied by the strict ISO C++ dialects:
1907 -ansi, -std=c++98, -std=c++11, etc.
1908 -fno-implicit-templates
1910 Never emit code for non-inline templates that are instantiated
1911 implicitly (i.e. by use); only emit code for explicit
1912 instantiations.
1913 -fno-implicit-inline-templates
1915 Don't emit code for implicit instantiations of inline templates,
1916 either. The default is to handle inlines differently so that
1917 compiles with and without optimization need the same set of
1918 explicit instantiations.
1919 -fno-implement-inlines
1921 To save space, do not emit out-of-line copies of inline functions
1922 controlled by "#pragma implementation". This causes linker errors
1923 if these functions are not inlined everywhere they are called.
1924 -fms-extensions
1926 Disable Wpedantic warnings about constructs used in MFC, such as
1927 implicit int and getting a pointer to member function via non-
1928 standard syntax.
1929 -fnew-inheriting-ctors
1931 Enable the P0136 adjustment to the semantics of C++11 constructor
1932 inheritance. This is part of C++17 but also considered to be a
1933 Defect Report against C++11 and C++14. This flag is enabled by
1934 default unless -fabi-version=10 or lower is specified.
1935 -fnew-ttp-matching
1937 Enable the P0522 resolution to Core issue 150, template template
1938 parameters and default arguments: this allows a template with
1939 default template arguments as an argument for a template template
1940 parameter with fewer template parameters. This flag is enabled by
1941 default for -std=c++17.
1942 -fno-nonansi-builtins
1944 Disable built-in declarations of functions that are not mandated by
1945 ANSI/ISO C. These include "ffs", "alloca", "_exit", "index",
1946 "bzero", "conjf", and other related functions.
1947 -fnothrow-opt
1949 Treat a "throw()" exception specification as if it were a
1950 "noexcept" specification to reduce or eliminate the text size
1951 overhead relative to a function with no exception specification.
1952 If the function has local variables of types with non-trivial
1953 destructors, the exception specification actually makes the
1954 function smaller because the EH cleanups for those variables can be
1955 optimized away. The semantic effect is that an exception thrown
1956 out of a function with such an exception specification results in a
1957 call to "terminate" rather than "unexpected".
1958 -fno-operator-names
1960 Do not treat the operator name keywords "and", "bitand", "bitor",
1961 "compl", "not", "or" and "xor" as synonyms as keywords.
1962 -fno-optional-diags
1964 Disable diagnostics that the standard says a compiler does not need
1965 to issue. Currently, the only such diagnostic issued by G++ is the
1966 one for a name having multiple meanings within a class.
1967 -fpermissive
1969 Downgrade some diagnostics about nonconformant code from errors to
1970 warnings. Thus, using -fpermissive allows some nonconforming code
1971 to compile.
1972 -fno-pretty-templates
1974 When an error message refers to a specialization of a function
1975 template, the compiler normally prints the signature of the
1976 template followed by the template arguments and any typedefs or
1977 typenames in the signature (e.g. "void f(T) [with T = int]" rather
1978 than "void f(int)") so that it's clear which template is involved.
1979 When an error message refers to a specialization of a class
1980 template, the compiler omits any template arguments that match the
1981 default template arguments for that template. If either of these
1982 behaviors make it harder to understand the error message rather
1983 than easier, you can use -fno-pretty-templates to disable them.
1984 -frepo
1986 Enable automatic template instantiation at link time. This option
1987 also implies -fno-implicit-templates.
1988 -fno-rtti
1990 Disable generation of information about every class with virtual
1991 functions for use by the C++ run-time type identification features
1992 ("dynamic_cast" and "typeid"). If you don't use those parts of the
1993 language, you can save some space by using this flag. Note that
1994 exception handling uses the same information, but G++ generates it
1995 as needed. The "dynamic_cast" operator can still be used for casts
1996 that do not require run-time type information, i.e. casts to "void
1997 *" or to unambiguous base classes.
1998 -fsized-deallocation
2000 Enable the built-in global declarations
2001 void operator delete (void *, std::size_t) noexcept;
2003 void operator delete[] (void *, std::size_t) noexcept;
2004 as introduced in C++14. This is useful for user-defined
2006 replacement deallocation functions that, for example, use the size
2007 of the object to make deallocation faster. Enabled by default
2008 under -std=c++14 and above. The flag -Wsized-deallocation warns
2009 about places that might want to add a definition.
2010 -fstrict-enums
2012 Allow the compiler to optimize using the assumption that a value of
2013 enumerated type can only be one of the values of the enumeration
2014 (as defined in the C++ standard; basically, a value that can be
2015 represented in the minimum number of bits needed to represent all
2016 the enumerators). This assumption may not be valid if the program
2017 uses a cast to convert an arbitrary integer value to the enumerated
2018 type.
2019 -fstrong-eval-order
2021 Evaluate member access, array subscripting, and shift expressions
2022 in left-to-right order, and evaluate assignment in right-to-left
2023 order, as adopted for C++17. Enabled by default with -std=c++17.
2024 -fstrong-eval-order=some enables just the ordering of member access
2025 and shift expressions, and is the default without -std=c++17.
2026 -ftemplate-backtrace-limit=n
2028 Set the maximum number of template instantiation notes for a single
2029 warning or error to n. The default value is 10.
2030 -ftemplate-depth=n
2032 Set the maximum instantiation depth for template classes to n. A
2033 limit on the template instantiation depth is needed to detect
2034 endless recursions during template class instantiation. ANSI/ISO
2035 C++ conforming programs must not rely on a maximum depth greater
2036 than 17 (changed to 1024 in C++11). The default value is 900, as
2037 the compiler can run out of stack space before hitting 1024 in some
2038 situations.
2039 -fno-threadsafe-statics
2041 Do not emit the extra code to use the routines specified in the C++
2042 ABI for thread-safe initialization of local statics. You can use
2043 this option to reduce code size slightly in code that doesn't need
2044 to be thread-safe.
2045 -fuse-cxa-atexit
2047 Register destructors for objects with static storage duration with
2048 the "__cxa_atexit" function rather than the "atexit" function.
2049 This option is required for fully standards-compliant handling of
2050 static destructors, but only works if your C library supports
2051 "__cxa_atexit".
2052 -fno-use-cxa-get-exception-ptr
2054 Don't use the "__cxa_get_exception_ptr" runtime routine. This
2055 causes "std::uncaught_exception" to be incorrect, but is necessary
2056 if the runtime routine is not available.
2057 -fvisibility-inlines-hidden
2059 This switch declares that the user does not attempt to compare
2060 pointers to inline functions or methods where the addresses of the
2061 two functions are taken in different shared objects.
2062 The effect of this is that GCC may, effectively, mark inline
2064 methods with "__attribute__ ((visibility ("hidden")))" so that they
2065 do not appear in the export table of a DSO and do not require a PLT
2066 indirection when used within the DSO. Enabling this option can
2067 have a dramatic effect on load and link times of a DSO as it
2068 massively reduces the size of the dynamic export table when the
2069 library makes heavy use of templates.
2070 The behavior of this switch is not quite the same as marking the
2072 methods as hidden directly, because it does not affect static
2073 variables local to the function or cause the compiler to deduce
2074 that the function is defined in only one shared object.
2075 You may mark a method as having a visibility explicitly to negate
2077 the effect of the switch for that method. For example, if you do
2078 want to compare pointers to a particular inline method, you might
2079 mark it as having default visibility. Marking the enclosing class
2080 with explicit visibility has no effect.
2081 Explicitly instantiated inline methods are unaffected by this
2083 option as their linkage might otherwise cross a shared library
2084 boundary.
2085 -fvisibility-ms-compat
2087 This flag attempts to use visibility settings to make GCC's C++
2088 linkage model compatible with that of Microsoft Visual Studio.
2089 The flag makes these changes to GCC's linkage model:
2091 1. It sets the default visibility to "hidden", like
2093 -fvisibility=hidden.
2094 2. Types, but not their members, are not hidden by default.
2096 3. The One Definition Rule is relaxed for types without explicit
2098 visibility specifications that are defined in more than one
2099 shared object: those declarations are permitted if they are
2100 permitted when this option is not used.
2101 In new code it is better to use -fvisibility=hidden and export
2103 those classes that are intended to be externally visible.
2104 Unfortunately it is possible for code to rely, perhaps
2105 accidentally, on the Visual Studio behavior.
2106 Among the consequences of these changes are that static data
2108 members of the same type with the same name but defined in
2109 different shared objects are different, so changing one does not
2110 change the other; and that pointers to function members defined in
2111 different shared objects may not compare equal. When this flag is
2112 given, it is a violation of the ODR to define types with the same
2113 name differently.
2114 -fno-weak
2116 Do not use weak symbol support, even if it is provided by the
2117 linker. By default, G++ uses weak symbols if they are available.
2118 This option exists only for testing, and should not be used by end-
2119 users; it results in inferior code and has no benefits. This
2120 option may be removed in a future release of G++.
2121 -nostdinc++
2123 Do not search for header files in the standard directories specific
2124 to C++, but do still search the other standard directories. (This
2125 option is used when building the C++ library.)
2126 In addition, these optimization, warning, and code generation options
2128 have meanings only for C++ programs:
2129 -Wabi (C, Objective-C, C++ and Objective-C++ only)
2131 Warn when G++ it generates code that is probably not compatible
2132 with the vendor-neutral C++ ABI. Since G++ now defaults to
2133 updating the ABI with each major release, normally -Wabi will warn
2134 only if there is a check added later in a release series for an ABI
2135 issue discovered since the initial release. -Wabi will warn about
2136 more things if an older ABI version is selected (with
2137 -fabi-version=n).
2138 -Wabi can also be used with an explicit version number to warn
2140 about compatibility with a particular -fabi-version level, e.g.
2141 -Wabi=2 to warn about changes relative to -fabi-version=2.
2142 If an explicit version number is provided and -fabi-compat-version
2144 is not specified, the version number from this option is used for
2145 compatibility aliases. If no explicit version number is provided
2146 with this option, but -fabi-compat-version is specified, that
2147 version number is used for ABI warnings.
2148 Although an effort has been made to warn about all such cases,
2150 there are probably some cases that are not warned about, even
2151 though G++ is generating incompatible code. There may also be
2152 cases where warnings are emitted even though the code that is
2153 generated is compatible.
2154 You should rewrite your code to avoid these warnings if you are
2156 concerned about the fact that code generated by G++ may not be
2157 binary compatible with code generated by other compilers.
2158 Known incompatibilities in -fabi-version=2 (which was the default
2160 from GCC 3.4 to 4.9) include:
2161 * A template with a non-type template parameter of reference type
2163 was mangled incorrectly:
2164 extern int N;
2166 template <int &> struct S {};
2167 void n (S<N>) {2}
2168 This was fixed in -fabi-version=3.
2170 * SIMD vector types declared using "__attribute ((vector_size))"
2172 were mangled in a non-standard way that does not allow for
2173 overloading of functions taking vectors of different sizes.
2174 The mangling was changed in -fabi-version=4.
2176 * "__attribute ((const))" and "noreturn" were mangled as type
2178 qualifiers, and "decltype" of a plain declaration was folded
2179 away.
2180 These mangling issues were fixed in -fabi-version=5.
2182 * Scoped enumerators passed as arguments to a variadic function
2184 are promoted like unscoped enumerators, causing "va_arg" to
2185 complain. On most targets this does not actually affect the
2186 parameter passing ABI, as there is no way to pass an argument
2187 smaller than "int".
2188 Also, the ABI changed the mangling of template argument packs,
2190 "const_cast", "static_cast", prefix increment/decrement, and a
2191 class scope function used as a template argument.
2192 These issues were corrected in -fabi-version=6.
2194 * Lambdas in default argument scope were mangled incorrectly, and
2196 the ABI changed the mangling of "nullptr_t".
2197 These issues were corrected in -fabi-version=7.
2199 * When mangling a function type with function-cv-qualifiers, the
2201 un-qualified function type was incorrectly treated as a
2202 substitution candidate.
2203 This was fixed in -fabi-version=8, the default for GCC 5.1.
2205 * "decltype(nullptr)" incorrectly had an alignment of 1, leading
2207 to unaligned accesses. Note that this did not affect the ABI
2208 of a function with a "nullptr_t" parameter, as parameters have
2209 a minimum alignment.
2210 This was fixed in -fabi-version=9, the default for GCC 5.2.
2212 * Target-specific attributes that affect the identity of a type,
2214 such as ia32 calling conventions on a function type (stdcall,
2215 regparm, etc.), did not affect the mangled name, leading to
2216 name collisions when function pointers were used as template
2217 arguments.
2218 This was fixed in -fabi-version=10, the default for GCC 6.1.
2220 It also warns about psABI-related changes. The known psABI changes
2222 at this point include:
2223 * For SysV/x86-64, unions with "long double" members are passed
2225 in memory as specified in psABI. For example:
2226 union U {
2228 long double ld;
2229 int i;
2230 };
2231 "union U" is always passed in memory.
2233 -Wabi-tag (C++ and Objective-C++ only)
2235 Warn when a type with an ABI tag is used in a context that does not
2236 have that ABI tag. See C++ Attributes for more information about
2237 ABI tags.
2238 -Wctor-dtor-privacy (C++ and Objective-C++ only)
2240 Warn when a class seems unusable because all the constructors or
2241 destructors in that class are private, and it has neither friends
2242 nor public static member functions. Also warn if there are no non-
2243 private methods, and there's at least one private member function
2244 that isn't a constructor or destructor.
2245 -Wdelete-non-virtual-dtor (C++ and Objective-C++ only)
2247 Warn when "delete" is used to destroy an instance of a class that
2248 has virtual functions and non-virtual destructor. It is unsafe to
2249 delete an instance of a derived class through a pointer to a base
2250 class if the base class does not have a virtual destructor. This
2251 warning is enabled by -Wall.
2252 -Wliteral-suffix (C++ and Objective-C++ only)
2254 Warn when a string or character literal is followed by a ud-suffix
2255 which does not begin with an underscore. As a conforming
2256 extension, GCC treats such suffixes as separate preprocessing
2257 tokens in order to maintain backwards compatibility with code that
2258 uses formatting macros from "<inttypes.h>". For example:
2259 #define __STDC_FORMAT_MACROS
2261 #include <inttypes.h>
2262 #include <stdio.h>
2263 int main() {
2265 int64_t i64 = 123;
2266 printf("My int64: %" PRId64"\n", i64);
2267 }
2268 In this case, "PRId64" is treated as a separate preprocessing
2270 token.
2271 Additionally, warn when a user-defined literal operator is declared
2273 with a literal suffix identifier that doesn't begin with an
2274 underscore. Literal suffix identifiers that don't begin with an
2275 underscore are reserved for future standardization.
2276 This warning is enabled by default.
2278 -Wlto-type-mismatch
2280 During the link-time optimization warn about type mismatches in
2281 global declarations from different compilation units. Requires
2282 -flto to be enabled. Enabled by default.
2283 -Wno-narrowing (C++ and Objective-C++ only)
2285 For C++11 and later standards, narrowing conversions are diagnosed
2286 by default, as required by the standard. A narrowing conversion
2287 from a constant produces an error, and a narrowing conversion from
2288 a non-constant produces a warning, but -Wno-narrowing suppresses
2289 the diagnostic. Note that this does not affect the meaning of
2290 well-formed code; narrowing conversions are still considered ill-
2291 formed in SFINAE contexts.
2292 With -Wnarrowing in C++98, warn when a narrowing conversion
2294 prohibited by C++11 occurs within { }, e.g.
2295 int i = { 2.2 }; // error: narrowing from double to int
2297 This flag is included in -Wall and -Wc++11-compat.
2299 -Wnoexcept (C++ and Objective-C++ only)
2301 Warn when a noexcept-expression evaluates to false because of a
2302 call to a function that does not have a non-throwing exception
2303 specification (i.e. "throw()" or "noexcept") but is known by the
2304 compiler to never throw an exception.
2305 -Wnoexcept-type (C++ and Objective-C++ only)
2307 Warn if the C++17 feature making "noexcept" part of a function type
2308 changes the mangled name of a symbol relative to C++14. Enabled by
2309 -Wabi and -Wc++17-compat.
2310 As an example:
2312 template <class T> void f(T t) { t(); };
2314 void g() noexcept;
2315 void h() { f(g); }
2316 In C++14, "f" calls "f<void(*)()>", but in C++17 it calls
2318 "f<void(*)()noexcept>".
2319 -Wclass-memaccess (C++ and Objective-C++ only)
2321 Warn when the destination of a call to a raw memory function such
2322 as "memset" or "memcpy" is an object of class type, and when
2323 writing into such an object might bypass the class non-trivial or
2324 deleted constructor or copy assignment, violate const-correctness
2325 or encapsulation, or corrupt virtual table pointers. Modifying the
2326 representation of such objects may violate invariants maintained by
2327 member functions of the class. For example, the call to "memset"
2328 below is undefined because it modifies a non-trivial class object
2329 and is, therefore, diagnosed. The safe way to either initialize or
2330 clear the storage of objects of such types is by using the
2331 appropriate constructor or assignment operator, if one is
2332 available.
2333 std::string str = "abc";
2335 memset (&str, 0, sizeof str);
2336 The -Wclass-memaccess option is enabled by -Wall. Explicitly
2338 casting the pointer to the class object to "void *" or to a type
2339 that can be safely accessed by the raw memory function suppresses
2340 the warning.
2341 -Wnon-virtual-dtor (C++ and Objective-C++ only)
2343 Warn when a class has virtual functions and an accessible non-
2344 virtual destructor itself or in an accessible polymorphic base
2345 class, in which case it is possible but unsafe to delete an
2346 instance of a derived class through a pointer to the class itself
2347 or base class. This warning is automatically enabled if -Weffc++
2348 is specified.
2349 -Wregister (C++ and Objective-C++ only)
2351 Warn on uses of the "register" storage class specifier, except when
2352 it is part of the GNU Explicit Register Variables extension. The
2353 use of the "register" keyword as storage class specifier has been
2354 deprecated in C++11 and removed in C++17. Enabled by default with
2355 -std=c++17.
2356 -Wreorder (C++ and Objective-C++ only)
2358 Warn when the order of member initializers given in the code does
2359 not match the order in which they must be executed. For instance:
2360 struct A {
2362 int i;
2363 int j;
2364 A(): j (0), i (1) { }
2365 };
2366 The compiler rearranges the member initializers for "i" and "j" to
2368 match the declaration order of the members, emitting a warning to
2369 that effect. This warning is enabled by -Wall.
2370 -fext-numeric-literals (C++ and Objective-C++ only)
2372 Accept imaginary, fixed-point, or machine-defined literal number
2373 suffixes as GNU extensions. When this option is turned off these
2374 suffixes are treated as C++11 user-defined literal numeric
2375 suffixes. This is on by default for all pre-C++11 dialects and all
2376 GNU dialects: -std=c++98, -std=gnu++98, -std=gnu++11, -std=gnu++14.
2377 This option is off by default for ISO C++11 onwards (-std=c++11,
2378 ...).
2379 The following -W... options are not affected by -Wall.
2381 -Weffc++ (C++ and Objective-C++ only)
2383 Warn about violations of the following style guidelines from Scott
2384 Meyers' Effective C++ series of books:
2385 * Define a copy constructor and an assignment operator for
2387 classes with dynamically-allocated memory.
2388 * Prefer initialization to assignment in constructors.
2390 * Have "operator=" return a reference to *this.
2392 * Don't try to return a reference when you must return an object.
2394 * Distinguish between prefix and postfix forms of increment and
2396 decrement operators.
2397 * Never overload "&&", "||", or ",".
2399 This option also enables -Wnon-virtual-dtor, which is also one of
2401 the effective C++ recommendations. However, the check is extended
2402 to warn about the lack of virtual destructor in accessible non-
2403 polymorphic bases classes too.
2404 When selecting this option, be aware that the standard library
2406 headers do not obey all of these guidelines; use grep -v to filter
2407 out those warnings.
2408 -Wstrict-null-sentinel (C++ and Objective-C++ only)
2410 Warn about the use of an uncasted "NULL" as sentinel. When
2411 compiling only with GCC this is a valid sentinel, as "NULL" is
2412 defined to "__null". Although it is a null pointer constant rather
2413 than a null pointer, it is guaranteed to be of the same size as a
2414 pointer. But this use is not portable across different compilers.
2415 -Wno-non-template-friend (C++ and Objective-C++ only)
2417 Disable warnings when non-template friend functions are declared
2418 within a template. In very old versions of GCC that predate
2419 implementation of the ISO standard, declarations such as friend int
2420 foo(int), where the name of the friend is an unqualified-id, could
2421 be interpreted as a particular specialization of a template
2422 function; the warning exists to diagnose compatibility problems,
2423 and is enabled by default.
2424 -Wold-style-cast (C++ and Objective-C++ only)
2426 Warn if an old-style (C-style) cast to a non-void type is used
2427 within a C++ program. The new-style casts ("dynamic_cast",
2428 "static_cast", "reinterpret_cast", and "const_cast") are less
2429 vulnerable to unintended effects and much easier to search for.
2430 -Woverloaded-virtual (C++ and Objective-C++ only)
2432 Warn when a function declaration hides virtual functions from a
2433 base class. For example, in:
2434 struct A {
2436 virtual void f();
2437 };
2438 struct B: public A {
2440 void f(int);
2441 };
2442 the "A" class version of "f" is hidden in "B", and code like:
2444 B* b;
2446 b->f();
2447 fails to compile.
2449 -Wno-pmf-conversions (C++ and Objective-C++ only)
2451 Disable the diagnostic for converting a bound pointer to member
2452 function to a plain pointer.
2453 -Wsign-promo (C++ and Objective-C++ only)
2455 Warn when overload resolution chooses a promotion from unsigned or
2456 enumerated type to a signed type, over a conversion to an unsigned
2457 type of the same size. Previous versions of G++ tried to preserve
2458 unsignedness, but the standard mandates the current behavior.
2459 -Wtemplates (C++ and Objective-C++ only)
2461 Warn when a primary template declaration is encountered. Some
2462 coding rules disallow templates, and this may be used to enforce
2463 that rule. The warning is inactive inside a system header file,
2464 such as the STL, so one can still use the STL. One may also
2465 instantiate or specialize templates.
2466 -Wmultiple-inheritance (C++ and Objective-C++ only)
2468 Warn when a class is defined with multiple direct base classes.
2469 Some coding rules disallow multiple inheritance, and this may be
2470 used to enforce that rule. The warning is inactive inside a system
2471 header file, such as the STL, so one can still use the STL. One
2472 may also define classes that indirectly use multiple inheritance.
2473 -Wvirtual-inheritance
2475 Warn when a class is defined with a virtual direct base class.
2476 Some coding rules disallow multiple inheritance, and this may be
2477 used to enforce that rule. The warning is inactive inside a system
2478 header file, such as the STL, so one can still use the STL. One
2479 may also define classes that indirectly use virtual inheritance.
2480 -Wnamespaces
2482 Warn when a namespace definition is opened. Some coding rules
2483 disallow namespaces, and this may be used to enforce that rule.
2484 The warning is inactive inside a system header file, such as the
2485 STL, so one can still use the STL. One may also use using
2486 directives and qualified names.
2487 -Wno-terminate (C++ and Objective-C++ only)
2489 Disable the warning about a throw-expression that will immediately
2490 result in a call to "terminate".
2491 Options Controlling Objective-C and Objective-C++ Dialects
2493 (NOTE: This manual does not describe the Objective-C and Objective-C++
2494 languages themselves.
2495 This section describes the command-line options that are only
2497 meaningful for Objective-C and Objective-C++ programs. You can also
2498 use most of the language-independent GNU compiler options. For
2499 example, you might compile a file some_class.m like this:
2500 gcc -g -fgnu-runtime -O -c some_class.m
2502 In this example, -fgnu-runtime is an option meant only for Objective-C
2504 and Objective-C++ programs; you can use the other options with any
2505 language supported by GCC.
2506 Note that since Objective-C is an extension of the C language,
2508 Objective-C compilations may also use options specific to the C front-
2509 end (e.g., -Wtraditional). Similarly, Objective-C++ compilations may
2510 use C++-specific options (e.g., -Wabi).
2511 Here is a list of options that are only for compiling Objective-C and
2513 Objective-C++ programs:
2514 -fconstant-string-class=class-name
2516 Use class-name as the name of the class to instantiate for each
2517 literal string specified with the syntax "@"..."". The default
2518 class name is "NXConstantString" if the GNU runtime is being used,
2519 and "NSConstantString" if the NeXT runtime is being used (see
2520 below). The -fconstant-cfstrings option, if also present,
2521 overrides the -fconstant-string-class setting and cause "@"...""
2522 literals to be laid out as constant CoreFoundation strings.
2523 -fgnu-runtime
2525 Generate object code compatible with the standard GNU Objective-C
2526 runtime. This is the default for most types of systems.
2527 -fnext-runtime
2529 Generate output compatible with the NeXT runtime. This is the
2530 default for NeXT-based systems, including Darwin and Mac OS X. The
2531 macro "__NEXT_RUNTIME__" is predefined if (and only if) this option
2532 is used.
2533 -fno-nil-receivers
2535 Assume that all Objective-C message dispatches ("[receiver
2536 message:arg]") in this translation unit ensure that the receiver is
2537 not "nil". This allows for more efficient entry points in the
2538 runtime to be used. This option is only available in conjunction
2539 with the NeXT runtime and ABI version 0 or 1.
2540 -fobjc-abi-version=n
2542 Use version n of the Objective-C ABI for the selected runtime.
2543 This option is currently supported only for the NeXT runtime. In
2544 that case, Version 0 is the traditional (32-bit) ABI without
2545 support for properties and other Objective-C 2.0 additions.
2546 Version 1 is the traditional (32-bit) ABI with support for
2547 properties and other Objective-C 2.0 additions. Version 2 is the
2548 modern (64-bit) ABI. If nothing is specified, the default is
2549 Version 0 on 32-bit target machines, and Version 2 on 64-bit target
2550 machines.
2551 -fobjc-call-cxx-cdtors
2553 For each Objective-C class, check if any of its instance variables
2554 is a C++ object with a non-trivial default constructor. If so,
2555 synthesize a special "- (id) .cxx_construct" instance method which
2556 runs non-trivial default constructors on any such instance
2557 variables, in order, and then return "self". Similarly, check if
2558 any instance variable is a C++ object with a non-trivial
2559 destructor, and if so, synthesize a special "- (void)
2560 .cxx_destruct" method which runs all such default destructors, in
2561 reverse order.
2562 The "- (id) .cxx_construct" and "- (void) .cxx_destruct" methods
2564 thusly generated only operate on instance variables declared in the
2565 current Objective-C class, and not those inherited from
2566 superclasses. It is the responsibility of the Objective-C runtime
2567 to invoke all such methods in an object's inheritance hierarchy.
2568 The "- (id) .cxx_construct" methods are invoked by the runtime
2569 immediately after a new object instance is allocated; the "- (void)
2570 .cxx_destruct" methods are invoked immediately before the runtime
2571 deallocates an object instance.
2572 As of this writing, only the NeXT runtime on Mac OS X 10.4 and
2574 later has support for invoking the "- (id) .cxx_construct" and "-
2575 (void) .cxx_destruct" methods.
2576 -fobjc-direct-dispatch
2578 Allow fast jumps to the message dispatcher. On Darwin this is
2579 accomplished via the comm page.
2580 -fobjc-exceptions
2582 Enable syntactic support for structured exception handling in
2583 Objective-C, similar to what is offered by C++. This option is
2584 required to use the Objective-C keywords @try, @throw, @catch,
2585 @finally and @synchronized. This option is available with both the
2586 GNU runtime and the NeXT runtime (but not available in conjunction
2587 with the NeXT runtime on Mac OS X 10.2 and earlier).
2588 -fobjc-gc
2590 Enable garbage collection (GC) in Objective-C and Objective-C++
2591 programs. This option is only available with the NeXT runtime; the
2592 GNU runtime has a different garbage collection implementation that
2593 does not require special compiler flags.
2594 -fobjc-nilcheck
2596 For the NeXT runtime with version 2 of the ABI, check for a nil
2597 receiver in method invocations before doing the actual method call.
2598 This is the default and can be disabled using -fno-objc-nilcheck.
2599 Class methods and super calls are never checked for nil in this way
2600 no matter what this flag is set to. Currently this flag does
2601 nothing when the GNU runtime, or an older version of the NeXT
2602 runtime ABI, is used.
2603 -fobjc-std=objc1
2605 Conform to the language syntax of Objective-C 1.0, the language
2606 recognized by GCC 4.0. This only affects the Objective-C additions
2607 to the C/C++ language; it does not affect conformance to C/C++
2608 standards, which is controlled by the separate C/C++ dialect option
2609 flags. When this option is used with the Objective-C or
2610 Objective-C++ compiler, any Objective-C syntax that is not
2611 recognized by GCC 4.0 is rejected. This is useful if you need to
2612 make sure that your Objective-C code can be compiled with older
2613 versions of GCC.
2614 -freplace-objc-classes
2616 Emit a special marker instructing ld(1) not to statically link in
2617 the resulting object file, and allow dyld(1) to load it in at run
2618 time instead. This is used in conjunction with the Fix-and-
2619 Continue debugging mode, where the object file in question may be
2620 recompiled and dynamically reloaded in the course of program
2621 execution, without the need to restart the program itself.
2622 Currently, Fix-and-Continue functionality is only available in
2623 conjunction with the NeXT runtime on Mac OS X 10.3 and later.
2624 -fzero-link
2626 When compiling for the NeXT runtime, the compiler ordinarily
2627 replaces calls to "objc_getClass("...")" (when the name of the
2628 class is known at compile time) with static class references that
2629 get initialized at load time, which improves run-time performance.
2630 Specifying the -fzero-link flag suppresses this behavior and causes
2631 calls to "objc_getClass("...")" to be retained. This is useful in
2632 Zero-Link debugging mode, since it allows for individual class
2633 implementations to be modified during program execution. The GNU
2634 runtime currently always retains calls to "objc_get_class("...")"
2635 regardless of command-line options.
2636 -fno-local-ivars
2638 By default instance variables in Objective-C can be accessed as if
2639 they were local variables from within the methods of the class
2640 they're declared in. This can lead to shadowing between instance
2641 variables and other variables declared either locally inside a
2642 class method or globally with the same name. Specifying the
2643 -fno-local-ivars flag disables this behavior thus avoiding variable
2644 shadowing issues.
2645 -fivar-visibility=[public|protected|private|package]
2647 Set the default instance variable visibility to the specified
2648 option so that instance variables declared outside the scope of any
2649 access modifier directives default to the specified visibility.
2650 -gen-decls
2652 Dump interface declarations for all classes seen in the source file
2653 to a file named sourcename.decl.
2654 -Wassign-intercept (Objective-C and Objective-C++ only)
2656 Warn whenever an Objective-C assignment is being intercepted by the
2657 garbage collector.
2658 -Wno-protocol (Objective-C and Objective-C++ only)
2660 If a class is declared to implement a protocol, a warning is issued
2661 for every method in the protocol that is not implemented by the
2662 class. The default behavior is to issue a warning for every method
2663 not explicitly implemented in the class, even if a method
2664 implementation is inherited from the superclass. If you use the
2665 -Wno-protocol option, then methods inherited from the superclass
2666 are considered to be implemented, and no warning is issued for
2667 them.
2668 -Wselector (Objective-C and Objective-C++ only)
2670 Warn if multiple methods of different types for the same selector
2671 are found during compilation. The check is performed on the list
2672 of methods in the final stage of compilation. Additionally, a
2673 check is performed for each selector appearing in a
2674 "@selector(...)" expression, and a corresponding method for that
2675 selector has been found during compilation. Because these checks
2676 scan the method table only at the end of compilation, these
2677 warnings are not produced if the final stage of compilation is not
2678 reached, for example because an error is found during compilation,
2679 or because the -fsyntax-only option is being used.
2680 -Wstrict-selector-match (Objective-C and Objective-C++ only)
2682 Warn if multiple methods with differing argument and/or return
2683 types are found for a given selector when attempting to send a
2684 message using this selector to a receiver of type "id" or "Class".
2685 When this flag is off (which is the default behavior), the compiler
2686 omits such warnings if any differences found are confined to types
2687 that share the same size and alignment.
2688 -Wundeclared-selector (Objective-C and Objective-C++ only)
2690 Warn if a "@selector(...)" expression referring to an undeclared
2691 selector is found. A selector is considered undeclared if no
2692 method with that name has been declared before the "@selector(...)"
2693 expression, either explicitly in an @interface or @protocol
2694 declaration, or implicitly in an @implementation section. This
2695 option always performs its checks as soon as a "@selector(...)"
2696 expression is found, while -Wselector only performs its checks in
2697 the final stage of compilation. This also enforces the coding
2698 style convention that methods and selectors must be declared before
2699 being used.
2700 -print-objc-runtime-info
2702 Generate C header describing the largest structure that is passed
2703 by value, if any.
2704 Options to Control Diagnostic Messages Formatting
2706 Traditionally, diagnostic messages have been formatted irrespective of
2707 the output device's aspect (e.g. its width, ...). You can use the
2708 options described below to control the formatting algorithm for
2709 diagnostic messages, e.g. how many characters per line, how often
2710 source location information should be reported. Note that some
2711 language front ends may not honor these options.
2712 -fmessage-length=n
2714 Try to format error messages so that they fit on lines of about n
2715 characters. If n is zero, then no line-wrapping is done; each
2716 error message appears on a single line. This is the default for
2717 all front ends.
2718 -fdiagnostics-show-location=once
2720 Only meaningful in line-wrapping mode. Instructs the diagnostic
2721 messages reporter to emit source location information once; that
2722 is, in case the message is too long to fit on a single physical
2723 line and has to be wrapped, the source location won't be emitted
2724 (as prefix) again, over and over, in subsequent continuation lines.
2725 This is the default behavior.
2726 -fdiagnostics-show-location=every-line
2728 Only meaningful in line-wrapping mode. Instructs the diagnostic
2729 messages reporter to emit the same source location information (as
2730 prefix) for physical lines that result from the process of breaking
2731 a message which is too long to fit on a single line.
2732 -fdiagnostics-color[=WHEN]
2734 -fno-diagnostics-color
2735 Use color in diagnostics. WHEN is never, always, or auto. The
2736 default depends on how the compiler has been configured, it can be
2737 any of the above WHEN options or also never if GCC_COLORS
2738 environment variable isn't present in the environment, and auto
2739 otherwise. auto means to use color only when the standard error is
2740 a terminal. The forms -fdiagnostics-color and
2741 -fno-diagnostics-color are aliases for -fdiagnostics-color=always
2742 and -fdiagnostics-color=never, respectively.
2743 The colors are defined by the environment variable GCC_COLORS. Its
2745 value is a colon-separated list of capabilities and Select Graphic
2746 Rendition (SGR) substrings. SGR commands are interpreted by the
2747 terminal or terminal emulator. (See the section in the
2748 documentation of your text terminal for permitted values and their
2749 meanings as character attributes.) These substring values are
2750 integers in decimal representation and can be concatenated with
2751 semicolons. Common values to concatenate include 1 for bold, 4 for
2752 underline, 5 for blink, 7 for inverse, 39 for default foreground
2753 color, 30 to 37 for foreground colors, 90 to 97 for 16-color mode
2754 foreground colors, 38;5;0 to 38;5;255 for 88-color and 256-color
2755 modes foreground colors, 49 for default background color, 40 to 47
2756 for background colors, 100 to 107 for 16-color mode background
2757 colors, and 48;5;0 to 48;5;255 for 88-color and 256-color modes
2758 background colors.
2759 The default GCC_COLORS is
2761 error=01;31:warning=01;35:note=01;36:range1=32:range2=34:locus=01:\
2763 quote=01:fixit-insert=32:fixit-delete=31:\
2764 diff-filename=01:diff-hunk=32:diff-delete=31:diff-insert=32:\
2765 type-diff=01;32
2766 where 01;31 is bold red, 01;35 is bold magenta, 01;36 is bold cyan,
2768 32 is green, 34 is blue, 01 is bold, and 31 is red. Setting
2769 GCC_COLORS to the empty string disables colors. Supported
2770 capabilities are as follows.
2771 "error="
2773 SGR substring for error: markers.
2774 "warning="
2776 SGR substring for warning: markers.
2777 "note="
2779 SGR substring for note: markers.
2780 "range1="
2782 SGR substring for first additional range.
2783 "range2="
2785 SGR substring for second additional range.
2786 "locus="
2788 SGR substring for location information, file:line or
2789 file:line:column etc.
2790 "quote="
2792 SGR substring for information printed within quotes.
2793 "fixit-insert="
2795 SGR substring for fix-it hints suggesting text to be inserted
2796 or replaced.
2797 "fixit-delete="
2799 SGR substring for fix-it hints suggesting text to be deleted.
2800 "diff-filename="
2802 SGR substring for filename headers within generated patches.
2803 "diff-hunk="
2805 SGR substring for the starts of hunks within generated patches.
2806 "diff-delete="
2808 SGR substring for deleted lines within generated patches.
2809 "diff-insert="
2811 SGR substring for inserted lines within generated patches.
2812 "type-diff="
2814 SGR substring for highlighting mismatching types within
2815 template arguments in the C++ frontend.
2816 -fno-diagnostics-show-option
2818 By default, each diagnostic emitted includes text indicating the
2819 command-line option that directly controls the diagnostic (if such
2820 an option is known to the diagnostic machinery). Specifying the
2821 -fno-diagnostics-show-option flag suppresses that behavior.
2822 -fno-diagnostics-show-caret
2824 By default, each diagnostic emitted includes the original source
2825 line and a caret ^ indicating the column. This option suppresses
2826 this information. The source line is truncated to n characters, if
2827 the -fmessage-length=n option is given. When the output is done to
2828 the terminal, the width is limited to the width given by the
2829 COLUMNS environment variable or, if not set, to the terminal width.
2830 -fdiagnostics-parseable-fixits
2832 Emit fix-it hints in a machine-parseable format, suitable for
2833 consumption by IDEs. For each fix-it, a line will be printed after
2834 the relevant diagnostic, starting with the string "fix-it:". For
2835 example:
2836 fix-it:"test.c":{45:3-45:21}:"gtk_widget_show_all"
2838 The location is expressed as a half-open range, expressed as a
2840 count of bytes, starting at byte 1 for the initial column. In the
2841 above example, bytes 3 through 20 of line 45 of "test.c" are to be
2842 replaced with the given string:
2843 00000000011111111112222222222
2845 12345678901234567890123456789
2846 gtk_widget_showall (dlg);
2847 ^^^^^^^^^^^^^^^^^^
2848 gtk_widget_show_all
2849 The filename and replacement string escape backslash as "\\", tab
2851 as "\t", newline as "\n", double quotes as "\"", non-printable
2852 characters as octal (e.g. vertical tab as "\013").
2853 An empty replacement string indicates that the given range is to be
2855 removed. An empty range (e.g. "45:3-45:3") indicates that the
2856 string is to be inserted at the given position.
2857 -fdiagnostics-generate-patch
2859 Print fix-it hints to stderr in unified diff format, after any
2860 diagnostics are printed. For example:
2861 --- test.c
2863 +++ test.c
2864 @ -42,5 +42,5 @
2865 void show_cb(GtkDialog *dlg)
2867 {
2868 - gtk_widget_showall(dlg);
2869 + gtk_widget_show_all(dlg);
2870 }
2871 The diff may or may not be colorized, following the same rules as
2873 for diagnostics (see -fdiagnostics-color).
2874 -fdiagnostics-show-template-tree
2876 In the C++ frontend, when printing diagnostics showing mismatching
2877 template types, such as:
2878 could not convert 'std::map<int, std::vector<double> >()'
2880 from 'map<[...],vector<double>>' to 'map<[...],vector<float>>
2881 the -fdiagnostics-show-template-tree flag enables printing a tree-
2883 like structure showing the common and differing parts of the types,
2884 such as:
2885 map<
2887 [...],
2888 vector<
2889 [double != float]>>
2890 The parts that differ are highlighted with color ("double" and
2892 "float" in this case).
2893 -fno-elide-type
2895 By default when the C++ frontend prints diagnostics showing
2896 mismatching template types, common parts of the types are printed
2897 as "[...]" to simplify the error message. For example:
2898 could not convert 'std::map<int, std::vector<double> >()'
2900 from 'map<[...],vector<double>>' to 'map<[...],vector<float>>
2901 Specifying the -fno-elide-type flag suppresses that behavior. This
2903 flag also affects the output of the
2904 -fdiagnostics-show-template-tree flag.
2905 -fno-show-column
2907 Do not print column numbers in diagnostics. This may be necessary
2908 if diagnostics are being scanned by a program that does not
2909 understand the column numbers, such as dejagnu.
2910 Options to Request or Suppress Warnings
2912 Warnings are diagnostic messages that report constructions that are not
2913 inherently erroneous but that are risky or suggest there may have been
2914 an error.
2915 The following language-independent options do not enable specific
2917 warnings but control the kinds of diagnostics produced by GCC.
2918 -fsyntax-only
2920 Check the code for syntax errors, but don't do anything beyond
2921 that.
2922 -fmax-errors=n
2924 Limits the maximum number of error messages to n, at which point
2925 GCC bails out rather than attempting to continue processing the
2926 source code. If n is 0 (the default), there is no limit on the
2927 number of error messages produced. If -Wfatal-errors is also
2928 specified, then -Wfatal-errors takes precedence over this option.
2929 -w Inhibit all warning messages.
2931 -Werror
2933 Make all warnings into errors.
2934 -Werror=
2936 Make the specified warning into an error. The specifier for a
2937 warning is appended; for example -Werror=switch turns the warnings
2938 controlled by -Wswitch into errors. This switch takes a negative
2939 form, to be used to negate -Werror for specific warnings; for
2940 example -Wno-error=switch makes -Wswitch warnings not be errors,
2941 even when -Werror is in effect.
2942 The warning message for each controllable warning includes the
2944 option that controls the warning. That option can then be used
2945 with -Werror= and -Wno-error= as described above. (Printing of the
2946 option in the warning message can be disabled using the
2947 -fno-diagnostics-show-option flag.)
2948 Note that specifying -Werror=foo automatically implies -Wfoo.
2950 However, -Wno-error=foo does not imply anything.
2951 -Wfatal-errors
2953 This option causes the compiler to abort compilation on the first
2954 error occurred rather than trying to keep going and printing
2955 further error messages.
2956 You can request many specific warnings with options beginning with -W,
2958 for example -Wimplicit to request warnings on implicit declarations.
2959 Each of these specific warning options also has a negative form
2960 beginning -Wno- to turn off warnings; for example, -Wno-implicit. This
2961 manual lists only one of the two forms, whichever is not the default.
2962 For further language-specific options also refer to C++ Dialect Options
2963 and Objective-C and Objective-C++ Dialect Options.
2964 Some options, such as -Wall and -Wextra, turn on other options, such as
2966 -Wunused, which may turn on further options, such as -Wunused-value.
2967 The combined effect of positive and negative forms is that more
2968 specific options have priority over less specific ones, independently
2969 of their position in the command-line. For options of the same
2970 specificity, the last one takes effect. Options enabled or disabled via
2971 pragmas take effect as if they appeared at the end of the command-line.
2972 When an unrecognized warning option is requested (e.g.,
2974 -Wunknown-warning), GCC emits a diagnostic stating that the option is
2975 not recognized. However, if the -Wno- form is used, the behavior is
2976 slightly different: no diagnostic is produced for -Wno-unknown-warning
2977 unless other diagnostics are being produced. This allows the use of
2978 new -Wno- options with old compilers, but if something goes wrong, the
2979 compiler warns that an unrecognized option is present.
2980 -Wpedantic
2982 -pedantic
2983 Issue all the warnings demanded by strict ISO C and ISO C++; reject
2984 all programs that use forbidden extensions, and some other programs
2985 that do not follow ISO C and ISO C++. For ISO C, follows the
2986 version of the ISO C standard specified by any -std option used.
2987 Valid ISO C and ISO C++ programs should compile properly with or
2989 without this option (though a rare few require -ansi or a -std
2990 option specifying the required version of ISO C). However, without
2991 this option, certain GNU extensions and traditional C and C++
2992 features are supported as well. With this option, they are
2993 rejected.
2994 -Wpedantic does not cause warning messages for use of the alternate
2996 keywords whose names begin and end with __. Pedantic warnings are
2997 also disabled in the expression that follows "__extension__".
2998 However, only system header files should use these escape routes;
2999 application programs should avoid them.
3000 Some users try to use -Wpedantic to check programs for strict ISO C
3002 conformance. They soon find that it does not do quite what they
3003 want: it finds some non-ISO practices, but not all---only those for
3004 which ISO C requires a diagnostic, and some others for which
3005 diagnostics have been added.
3006 A feature to report any failure to conform to ISO C might be useful
3008 in some instances, but would require considerable additional work
3009 and would be quite different from -Wpedantic. We don't have plans
3010 to support such a feature in the near future.
3011 Where the standard specified with -std represents a GNU extended
3013 dialect of C, such as gnu90 or gnu99, there is a corresponding base
3014 standard, the version of ISO C on which the GNU extended dialect is
3015 based. Warnings from -Wpedantic are given where they are required
3016 by the base standard. (It does not make sense for such warnings to
3017 be given only for features not in the specified GNU C dialect,
3018 since by definition the GNU dialects of C include all features the
3019 compiler supports with the given option, and there would be nothing
3020 to warn about.)
3021 -pedantic-errors
3023 Give an error whenever the base standard (see -Wpedantic) requires
3024 a diagnostic, in some cases where there is undefined behavior at
3025 compile-time and in some other cases that do not prevent
3026 compilation of programs that are valid according to the standard.
3027 This is not equivalent to -Werror=pedantic, since there are errors
3028 enabled by this option and not enabled by the latter and vice
3029 versa.
3030 -Wall
3032 This enables all the warnings about constructions that some users
3033 consider questionable, and that are easy to avoid (or modify to
3034 prevent the warning), even in conjunction with macros. This also
3035 enables some language-specific warnings described in C++ Dialect
3036 Options and Objective-C and Objective-C++ Dialect Options.
3037 -Wall turns on the following warning flags:
3039 -Waddress -Warray-bounds=1 (only with -O2) -Wbool-compare
3041 -Wbool-operation -Wc++11-compat -Wc++14-compat -Wcatch-value (C++
3042 and Objective-C++ only) -Wchar-subscripts -Wcomment
3043 -Wduplicate-decl-specifier (C and Objective-C only) -Wenum-compare
3044 (in C/ObjC; this is on by default in C++) -Wformat
3045 -Wint-in-bool-context -Wimplicit (C and Objective-C only)
3046 -Wimplicit-int (C and Objective-C only)
3047 -Wimplicit-function-declaration (C and Objective-C only)
3048 -Winit-self (only for C++) -Wlogical-not-parentheses -Wmain (only
3049 for C/ObjC and unless -ffreestanding) -Wmaybe-uninitialized
3050 -Wmemset-elt-size -Wmemset-transposed-args -Wmisleading-indentation
3051 (only for C/C++) -Wmissing-attributes -Wmissing-braces (only for
3052 C/ObjC) -Wmultistatement-macros -Wnarrowing (only for C++)
3053 -Wnonnull -Wnonnull-compare -Wopenmp-simd -Wparentheses
3054 -Wpointer-sign -Wreorder -Wrestrict -Wreturn-type -Wsequence-point
3055 -Wsign-compare (only in C++) -Wsizeof-pointer-div
3056 -Wsizeof-pointer-memaccess -Wstrict-aliasing -Wstrict-overflow=1
3057 -Wstringop-truncation -Wswitch -Wtautological-compare -Wtrigraphs
3058 -Wuninitialized -Wunknown-pragmas -Wunused-function -Wunused-label
3059 -Wunused-value -Wunused-variable -Wvolatile-register-var
3060 Note that some warning flags are not implied by -Wall. Some of
3062 them warn about constructions that users generally do not consider
3063 questionable, but which occasionally you might wish to check for;
3064 others warn about constructions that are necessary or hard to avoid
3065 in some cases, and there is no simple way to modify the code to
3066 suppress the warning. Some of them are enabled by -Wextra but many
3067 of them must be enabled individually.
3068 -Wextra
3070 This enables some extra warning flags that are not enabled by
3071 -Wall. (This option used to be called -W. The older name is still
3072 supported, but the newer name is more descriptive.)
3073 -Wclobbered -Wcast-function-type -Wempty-body -Wignored-qualifiers
3075 -Wimplicit-fallthrough=3 -Wmissing-field-initializers
3076 -Wmissing-parameter-type (C only) -Wold-style-declaration (C only)
3077 -Woverride-init -Wsign-compare (C only) -Wtype-limits
3078 -Wuninitialized -Wshift-negative-value (in C++03 and in C99 and
3079 newer) -Wunused-parameter (only with -Wunused or -Wall)
3080 -Wunused-but-set-parameter (only with -Wunused or -Wall)
3081 The option -Wextra also prints warning messages for the following
3083 cases:
3084 * A pointer is compared against integer zero with "<", "<=", ">",
3086 or ">=".
3087 * (C++ only) An enumerator and a non-enumerator both appear in a
3089 conditional expression.
3090 * (C++ only) Ambiguous virtual bases.
3092 * (C++ only) Subscripting an array that has been declared
3094 "register".
3095 * (C++ only) Taking the address of a variable that has been
3097 declared "register".
3098 * (C++ only) A base class is not initialized in the copy
3100 constructor of a derived class.
3101 -Wchar-subscripts
3103 Warn if an array subscript has type "char". This is a common cause
3104 of error, as programmers often forget that this type is signed on
3105 some machines. This warning is enabled by -Wall.
3106 -Wchkp
3108 Warn about an invalid memory access that is found by Pointer Bounds
3109 Checker (-fcheck-pointer-bounds).
3110 -Wno-coverage-mismatch
3112 Warn if feedback profiles do not match when using the -fprofile-use
3113 option. If a source file is changed between compiling with
3114 -fprofile-gen and with -fprofile-use, the files with the profile
3115 feedback can fail to match the source file and GCC cannot use the
3116 profile feedback information. By default, this warning is enabled
3117 and is treated as an error. -Wno-coverage-mismatch can be used to
3118 disable the warning or -Wno-error=coverage-mismatch can be used to
3119 disable the error. Disabling the error for this warning can result
3120 in poorly optimized code and is useful only in the case of very
3121 minor changes such as bug fixes to an existing code-base.
3122 Completely disabling the warning is not recommended.
3123 -Wno-cpp
3125 (C, Objective-C, C++, Objective-C++ and Fortran only)
3126 Suppress warning messages emitted by "#warning" directives.
3128 -Wdouble-promotion (C, C++, Objective-C and Objective-C++ only)
3130 Give a warning when a value of type "float" is implicitly promoted
3131 to "double". CPUs with a 32-bit "single-precision" floating-point
3132 unit implement "float" in hardware, but emulate "double" in
3133 software. On such a machine, doing computations using "double"
3134 values is much more expensive because of the overhead required for
3135 software emulation.
3136 It is easy to accidentally do computations with "double" because
3138 floating-point literals are implicitly of type "double". For
3139 example, in:
3140 float area(float radius)
3142 {
3143 return 3.14159 * radius * radius;
3144 }
3145 the compiler performs the entire computation with "double" because
3147 the floating-point literal is a "double".
3148 -Wduplicate-decl-specifier (C and Objective-C only)
3150 Warn if a declaration has duplicate "const", "volatile", "restrict"
3151 or "_Atomic" specifier. This warning is enabled by -Wall.
3152 -Wformat
3154 -Wformat=n
3155 Check calls to "printf" and "scanf", etc., to make sure that the
3156 arguments supplied have types appropriate to the format string
3157 specified, and that the conversions specified in the format string
3158 make sense. This includes standard functions, and others specified
3159 by format attributes, in the "printf", "scanf", "strftime" and
3160 "strfmon" (an X/Open extension, not in the C standard) families (or
3161 other target-specific families). Which functions are checked
3162 without format attributes having been specified depends on the
3163 standard version selected, and such checks of functions without the
3164 attribute specified are disabled by -ffreestanding or -fno-builtin.
3165 The formats are checked against the format features supported by
3167 GNU libc version 2.2. These include all ISO C90 and C99 features,
3168 as well as features from the Single Unix Specification and some BSD
3169 and GNU extensions. Other library implementations may not support
3170 all these features; GCC does not support warning about features
3171 that go beyond a particular library's limitations. However, if
3172 -Wpedantic is used with -Wformat, warnings are given about format
3173 features not in the selected standard version (but not for
3174 "strfmon" formats, since those are not in any version of the C
3175 standard).
3176 -Wformat=1
3178 -Wformat
3179 Option -Wformat is equivalent to -Wformat=1, and -Wno-format is
3180 equivalent to -Wformat=0. Since -Wformat also checks for null
3181 format arguments for several functions, -Wformat also implies
3182 -Wnonnull. Some aspects of this level of format checking can
3183 be disabled by the options: -Wno-format-contains-nul,
3184 -Wno-format-extra-args, and -Wno-format-zero-length. -Wformat
3185 is enabled by -Wall.
3186 -Wno-format-contains-nul
3188 If -Wformat is specified, do not warn about format strings that
3189 contain NUL bytes.
3190 -Wno-format-extra-args
3192 If -Wformat is specified, do not warn about excess arguments to
3193 a "printf" or "scanf" format function. The C standard
3194 specifies that such arguments are ignored.
3195 Where the unused arguments lie between used arguments that are
3197 specified with $ operand number specifications, normally
3198 warnings are still given, since the implementation could not
3199 know what type to pass to "va_arg" to skip the unused
3200 arguments. However, in the case of "scanf" formats, this
3201 option suppresses the warning if the unused arguments are all
3202 pointers, since the Single Unix Specification says that such
3203 unused arguments are allowed.
3204 -Wformat-overflow
3206 -Wformat-overflow=level
3207 Warn about calls to formatted input/output functions such as
3208 "sprintf" and "vsprintf" that might overflow the destination
3209 buffer. When the exact number of bytes written by a format
3210 directive cannot be determined at compile-time it is estimated
3211 based on heuristics that depend on the level argument and on
3212 optimization. While enabling optimization will in most cases
3213 improve the accuracy of the warning, it may also result in
3214 false positives.
3215 -Wformat-overflow
3217 -Wformat-overflow=1
3218 Level 1 of -Wformat-overflow enabled by -Wformat employs a
3219 conservative approach that warns only about calls that most
3220 likely overflow the buffer. At this level, numeric
3221 arguments to format directives with unknown values are
3222 assumed to have the value of one, and strings of unknown
3223 length to be empty. Numeric arguments that are known to be
3224 bounded to a subrange of their type, or string arguments
3225 whose output is bounded either by their directive's
3226 precision or by a finite set of string literals, are
3227 assumed to take on the value within the range that results
3228 in the most bytes on output. For example, the call to
3229 "sprintf" below is diagnosed because even with both a and b
3230 equal to zero, the terminating NUL character ('\0')
3231 appended by the function to the destination buffer will be
3232 written past its end. Increasing the size of the buffer by
3233 a single byte is sufficient to avoid the warning, though it
3234 may not be sufficient to avoid the overflow.
3235 void f (int a, int b)
3237 {
3238 char buf [13];
3239 sprintf (buf, "a = %i, b = %i\n", a, b);
3240 }
3241 -Wformat-overflow=2
3243 Level 2 warns also about calls that might overflow the
3244 destination buffer given an argument of sufficient length
3245 or magnitude. At level 2, unknown numeric arguments are
3246 assumed to have the minimum representable value for signed
3247 types with a precision greater than 1, and the maximum
3248 representable value otherwise. Unknown string arguments
3249 whose length cannot be assumed to be bounded either by the
3250 directive's precision, or by a finite set of string
3251 literals they may evaluate to, or the character array they
3252 may point to, are assumed to be 1 character long.
3253 At level 2, the call in the example above is again
3255 diagnosed, but this time because with a equal to a 32-bit
3256 "INT_MIN" the first %i directive will write some of its
3257 digits beyond the end of the destination buffer. To make
3258 the call safe regardless of the values of the two
3259 variables, the size of the destination buffer must be
3260 increased to at least 34 bytes. GCC includes the minimum
3261 size of the buffer in an informational note following the
3262 warning.
3263 An alternative to increasing the size of the destination
3265 buffer is to constrain the range of formatted values. The
3266 maximum length of string arguments can be bounded by
3267 specifying the precision in the format directive. When
3268 numeric arguments of format directives can be assumed to be
3269 bounded by less than the precision of their type, choosing
3270 an appropriate length modifier to the format specifier will
3271 reduce the required buffer size. For example, if a and b
3272 in the example above can be assumed to be within the
3273 precision of the "short int" type then using either the %hi
3274 format directive or casting the argument to "short" reduces
3275 the maximum required size of the buffer to 24 bytes.
3276 void f (int a, int b)
3278 {
3279 char buf [23];
3280 sprintf (buf, "a = %hi, b = %i\n", a, (short)b);
3281 }
3282 -Wno-format-zero-length
3284 If -Wformat is specified, do not warn about zero-length
3285 formats. The C standard specifies that zero-length formats are
3286 allowed.
3287 -Wformat=2
3289 Enable -Wformat plus additional format checks. Currently
3290 equivalent to -Wformat -Wformat-nonliteral -Wformat-security
3291 -Wformat-y2k.
3292 -Wformat-nonliteral
3294 If -Wformat is specified, also warn if the format string is not
3295 a string literal and so cannot be checked, unless the format
3296 function takes its format arguments as a "va_list".
3297 -Wformat-security
3299 If -Wformat is specified, also warn about uses of format
3300 functions that represent possible security problems. At
3301 present, this warns about calls to "printf" and "scanf"
3302 functions where the format string is not a string literal and
3303 there are no format arguments, as in "printf (foo);". This may
3304 be a security hole if the format string came from untrusted
3305 input and contains %n. (This is currently a subset of what
3306 -Wformat-nonliteral warns about, but in future warnings may be
3307 added to -Wformat-security that are not included in
3308 -Wformat-nonliteral.)
3309 -Wformat-signedness
3311 If -Wformat is specified, also warn if the format string
3312 requires an unsigned argument and the argument is signed and
3313 vice versa.
3314 -Wformat-truncation
3316 -Wformat-truncation=level
3317 Warn about calls to formatted input/output functions such as
3318 "snprintf" and "vsnprintf" that might result in output
3319 truncation. When the exact number of bytes written by a format
3320 directive cannot be determined at compile-time it is estimated
3321 based on heuristics that depend on the level argument and on
3322 optimization. While enabling optimization will in most cases
3323 improve the accuracy of the warning, it may also result in
3324 false positives. Except as noted otherwise, the option uses
3325 the same logic -Wformat-overflow.
3326 -Wformat-truncation
3328 -Wformat-truncation=1
3329 Level 1 of -Wformat-truncation enabled by -Wformat employs
3330 a conservative approach that warns only about calls to
3331 bounded functions whose return value is unused and that
3332 will most likely result in output truncation.
3333 -Wformat-truncation=2
3335 Level 2 warns also about calls to bounded functions whose
3336 return value is used and that might result in truncation
3337 given an argument of sufficient length or magnitude.
3338 -Wformat-y2k
3340 If -Wformat is specified, also warn about "strftime" formats
3341 that may yield only a two-digit year.
3342 -Wnonnull
3344 Warn about passing a null pointer for arguments marked as requiring
3345 a non-null value by the "nonnull" function attribute.
3346 -Wnonnull is included in -Wall and -Wformat. It can be disabled
3348 with the -Wno-nonnull option.
3349 -Wnonnull-compare
3351 Warn when comparing an argument marked with the "nonnull" function
3352 attribute against null inside the function.
3353 -Wnonnull-compare is included in -Wall. It can be disabled with
3355 the -Wno-nonnull-compare option.
3356 -Wnull-dereference
3358 Warn if the compiler detects paths that trigger erroneous or
3359 undefined behavior due to dereferencing a null pointer. This
3360 option is only active when -fdelete-null-pointer-checks is active,
3361 which is enabled by optimizations in most targets. The precision
3362 of the warnings depends on the optimization options used.
3363 -Winit-self (C, C++, Objective-C and Objective-C++ only)
3365 Warn about uninitialized variables that are initialized with
3366 themselves. Note this option can only be used with the
3367 -Wuninitialized option.
3368 For example, GCC warns about "i" being uninitialized in the
3370 following snippet only when -Winit-self has been specified:
3371 int f()
3373 {
3374 int i = i;
3375 return i;
3376 }
3377 This warning is enabled by -Wall in C++.
3379 -Wimplicit-int (C and Objective-C only)
3381 Warn when a declaration does not specify a type. This warning is
3382 enabled by -Wall.
3383 -Wimplicit-function-declaration (C and Objective-C only)
3385 Give a warning whenever a function is used before being declared.
3386 In C99 mode (-std=c99 or -std=gnu99), this warning is enabled by
3387 default and it is made into an error by -pedantic-errors. This
3388 warning is also enabled by -Wall.
3389 -Wimplicit (C and Objective-C only)
3391 Same as -Wimplicit-int and -Wimplicit-function-declaration. This
3392 warning is enabled by -Wall.
3393 -Wimplicit-fallthrough
3395 -Wimplicit-fallthrough is the same as -Wimplicit-fallthrough=3 and
3396 -Wno-implicit-fallthrough is the same as -Wimplicit-fallthrough=0.
3397 -Wimplicit-fallthrough=n
3399 Warn when a switch case falls through. For example:
3400 switch (cond)
3402 {
3403 case 1:
3404 a = 1;
3405 break;
3406 case 2:
3407 a = 2;
3408 case 3:
3409 a = 3;
3410 break;
3411 }
3412 This warning does not warn when the last statement of a case cannot
3414 fall through, e.g. when there is a return statement or a call to
3415 function declared with the noreturn attribute.
3416 -Wimplicit-fallthrough= also takes into account control flow
3417 statements, such as ifs, and only warns when appropriate. E.g.
3418 switch (cond)
3420 {
3421 case 1:
3422 if (i > 3) {
3423 bar (5);
3424 break;
3425 } else if (i < 1) {
3426 bar (0);
3427 } else
3428 return;
3429 default:
3430 ...
3431 }
3432 Since there are occasions where a switch case fall through is
3434 desirable, GCC provides an attribute, "__attribute__
3435 ((fallthrough))", that is to be used along with a null statement to
3436 suppress this warning that would normally occur:
3437 switch (cond)
3439 {
3440 case 1:
3441 bar (0);
3442 __attribute__ ((fallthrough));
3443 default:
3444 ...
3445 }
3446 C++17 provides a standard way to suppress the
3448 -Wimplicit-fallthrough warning using "[[fallthrough]];" instead of
3449 the GNU attribute. In C++11 or C++14 users can use
3450 "[[gnu::fallthrough]];", which is a GNU extension. Instead of
3451 these attributes, it is also possible to add a fallthrough comment
3452 to silence the warning. The whole body of the C or C++ style
3453 comment should match the given regular expressions listed below.
3454 The option argument n specifies what kind of comments are accepted:
3455 *<-Wimplicit-fallthrough=0 disables the warning altogether.>
3457 *<-Wimplicit-fallthrough=1 matches ".*" regular>
3458 expression, any comment is used as fallthrough comment.
3459 *<-Wimplicit-fallthrough=2 case insensitively matches>
3461 ".*falls?[ \t-]*thr(ough|u).*" regular expression.
3462 *<-Wimplicit-fallthrough=3 case sensitively matches one of the>
3464 following regular expressions:
3465 *<"-fallthrough">
3467 *<"@fallthrough@">
3468 *<"lint -fallthrough[ \t]*">
3469 *<"[ \t.!]*(ELSE,? |INTENTIONAL(LY)? )?FALL(S |
3470 |-)?THR(OUGH|U)[ \t.!]*(-[^\n\r]*)?">
3471 *<"[ \t.!]*(Else,? |Intentional(ly)? )?Fall((s |
3472 |-)[Tt]|t)hr(ough|u)[ \t.!]*(-[^\n\r]*)?">
3473 *<"[ \t.!]*([Ee]lse,? |[Ii]ntentional(ly)? )?fall(s |
3474 |-)?thr(ough|u)[ \t.!]*(-[^\n\r]*)?">
3475 *<-Wimplicit-fallthrough=4 case sensitively matches one of the>
3476 following regular expressions:
3477 *<"-fallthrough">
3479 *<"@fallthrough@">
3480 *<"lint -fallthrough[ \t]*">
3481 *<"[ \t]*FALLTHR(OUGH|U)[ \t]*">
3482 *<-Wimplicit-fallthrough=5 doesn't recognize any comments as>
3483 fallthrough comments, only attributes disable the warning.
3484 The comment needs to be followed after optional whitespace and
3486 other comments by "case" or "default" keywords or by a user label
3487 that precedes some "case" or "default" label.
3488 switch (cond)
3490 {
3491 case 1:
3492 bar (0);
3493 /* FALLTHRU */
3494 default:
3495 ...
3496 }
3497 The -Wimplicit-fallthrough=3 warning is enabled by -Wextra.
3499 -Wif-not-aligned (C, C++, Objective-C and Objective-C++ only)
3501 Control if warning triggered by the "warn_if_not_aligned" attribute
3502 should be issued. This is enabled by default. Use
3503 -Wno-if-not-aligned to disable it.
3504 -Wignored-qualifiers (C and C++ only)
3506 Warn if the return type of a function has a type qualifier such as
3507 "const". For ISO C such a type qualifier has no effect, since the
3508 value returned by a function is not an lvalue. For C++, the
3509 warning is only emitted for scalar types or "void". ISO C
3510 prohibits qualified "void" return types on function definitions, so
3511 such return types always receive a warning even without this
3512 option.
3513 This warning is also enabled by -Wextra.
3515 -Wignored-attributes (C and C++ only)
3517 Warn when an attribute is ignored. This is different from the
3518 -Wattributes option in that it warns whenever the compiler decides
3519 to drop an attribute, not that the attribute is either unknown,
3520 used in a wrong place, etc. This warning is enabled by default.
3521 -Wmain
3523 Warn if the type of "main" is suspicious. "main" should be a
3524 function with external linkage, returning int, taking either zero
3525 arguments, two, or three arguments of appropriate types. This
3526 warning is enabled by default in C++ and is enabled by either -Wall
3527 or -Wpedantic.
3528 -Wmisleading-indentation (C and C++ only)
3530 Warn when the indentation of the code does not reflect the block
3531 structure. Specifically, a warning is issued for "if", "else",
3532 "while", and "for" clauses with a guarded statement that does not
3533 use braces, followed by an unguarded statement with the same
3534 indentation.
3535 In the following example, the call to "bar" is misleadingly
3537 indented as if it were guarded by the "if" conditional.
3538 if (some_condition ())
3540 foo ();
3541 bar (); /* Gotcha: this is not guarded by the "if". */
3542 In the case of mixed tabs and spaces, the warning uses the
3544 -ftabstop= option to determine if the statements line up
3545 (defaulting to 8).
3546 The warning is not issued for code involving multiline preprocessor
3548 logic such as the following example.
3549 if (flagA)
3551 foo (0);
3552 #if SOME_CONDITION_THAT_DOES_NOT_HOLD
3553 if (flagB)
3554 #endif
3555 foo (1);
3556 The warning is not issued after a "#line" directive, since this
3558 typically indicates autogenerated code, and no assumptions can be
3559 made about the layout of the file that the directive references.
3560 This warning is enabled by -Wall in C and C++.
3562 -Wmissing-attributes
3564 Warn when a declaration of a function is missing one or more
3565 attributes that a related function is declared with and whose
3566 absence may adversely affect the correctness or efficiency of
3567 generated code. For example, in C++, the warning is issued when an
3568 explicit specialization of a primary template declared with
3569 attribute "alloc_align", "alloc_size", "assume_aligned", "format",
3570 "format_arg", "malloc", or "nonnull" is declared without it.
3571 Attributes "deprecated", "error", and "warning" suppress the
3572 warning..
3573 -Wmissing-attributes is enabled by -Wall.
3575 For example, since the declaration of the primary function template
3577 below makes use of both attribute "malloc" and "alloc_size" the
3578 declaration of the explicit specialization of the template is
3579 diagnosed because it is missing one of the attributes.
3580 template <class T>
3582 T* __attribute__ ((malloc, alloc_size (1)))
3583 allocate (size_t);
3584 template <>
3586 void* __attribute__ ((malloc)) // missing alloc_size
3587 allocate<void> (size_t);
3588 -Wmissing-braces
3590 Warn if an aggregate or union initializer is not fully bracketed.
3591 In the following example, the initializer for "a" is not fully
3592 bracketed, but that for "b" is fully bracketed. This warning is
3593 enabled by -Wall in C.
3594 int a[2][2] = { 0, 1, 2, 3 };
3596 int b[2][2] = { { 0, 1 }, { 2, 3 } };
3597 This warning is enabled by -Wall.
3599 -Wmissing-include-dirs (C, C++, Objective-C and Objective-C++ only)
3601 Warn if a user-supplied include directory does not exist.
3602 -Wmultistatement-macros
3604 Warn about unsafe multiple statement macros that appear to be
3605 guarded by a clause such as "if", "else", "for", "switch", or
3606 "while", in which only the first statement is actually guarded
3607 after the macro is expanded.
3608 For example:
3610 #define DOIT x++; y++
3612 if (c)
3613 DOIT;
3614 will increment "y" unconditionally, not just when "c" holds. The
3616 can usually be fixed by wrapping the macro in a do-while loop:
3617 #define DOIT do { x++; y++; } while (0)
3619 if (c)
3620 DOIT;
3621 This warning is enabled by -Wall in C and C++.
3623 -Wparentheses
3625 Warn if parentheses are omitted in certain contexts, such as when
3626 there is an assignment in a context where a truth value is
3627 expected, or when operators are nested whose precedence people
3628 often get confused about.
3629 Also warn if a comparison like "x<=y<=z" appears; this is
3631 equivalent to "(x<=y ? 1 : 0) <= z", which is a different
3632 interpretation from that of ordinary mathematical notation.
3633 Also warn for dangerous uses of the GNU extension to "?:" with
3635 omitted middle operand. When the condition in the "?": operator is
3636 a boolean expression, the omitted value is always 1. Often
3637 programmers expect it to be a value computed inside the conditional
3638 expression instead.
3639 For C++ this also warns for some cases of unnecessary parentheses
3641 in declarations, which can indicate an attempt at a function call
3642 instead of a declaration:
3643 {
3645 // Declares a local variable called mymutex.
3646 std::unique_lock<std::mutex> (mymutex);
3647 // User meant std::unique_lock<std::mutex> lock (mymutex);
3648 }
3649 This warning is enabled by -Wall.
3651 -Wsequence-point
3653 Warn about code that may have undefined semantics because of
3654 violations of sequence point rules in the C and C++ standards.
3655 The C and C++ standards define the order in which expressions in a
3657 C/C++ program are evaluated in terms of sequence points, which
3658 represent a partial ordering between the execution of parts of the
3659 program: those executed before the sequence point, and those
3660 executed after it. These occur after the evaluation of a full
3661 expression (one which is not part of a larger expression), after
3662 the evaluation of the first operand of a "&&", "||", "? :" or ","
3663 (comma) operator, before a function is called (but after the
3664 evaluation of its arguments and the expression denoting the called
3665 function), and in certain other places. Other than as expressed by
3666 the sequence point rules, the order of evaluation of subexpressions
3667 of an expression is not specified. All these rules describe only a
3668 partial order rather than a total order, since, for example, if two
3669 functions are called within one expression with no sequence point
3670 between them, the order in which the functions are called is not
3671 specified. However, the standards committee have ruled that
3672 function calls do not overlap.
3673 It is not specified when between sequence points modifications to
3675 the values of objects take effect. Programs whose behavior depends
3676 on this have undefined behavior; the C and C++ standards specify
3677 that "Between the previous and next sequence point an object shall
3678 have its stored value modified at most once by the evaluation of an
3679 expression. Furthermore, the prior value shall be read only to
3680 determine the value to be stored.". If a program breaks these
3681 rules, the results on any particular implementation are entirely
3682 unpredictable.
3683 Examples of code with undefined behavior are "a = a++;", "a[n] =
3685 b[n++]" and "a[i++] = i;". Some more complicated cases are not
3686 diagnosed by this option, and it may give an occasional false
3687 positive result, but in general it has been found fairly effective
3688 at detecting this sort of problem in programs.
3689 The C++17 standard will define the order of evaluation of operands
3691 in more cases: in particular it requires that the right-hand side
3692 of an assignment be evaluated before the left-hand side, so the
3693 above examples are no longer undefined. But this warning will
3694 still warn about them, to help people avoid writing code that is
3695 undefined in C and earlier revisions of C++.
3696 The standard is worded confusingly, therefore there is some debate
3698 over the precise meaning of the sequence point rules in subtle
3699 cases. Links to discussions of the problem, including proposed
3700 formal definitions, may be found on the GCC readings page, at
3701 <http://gcc.gnu.org/readings.html>.
3702 This warning is enabled by -Wall for C and C++.
3704 -Wno-return-local-addr
3706 Do not warn about returning a pointer (or in C++, a reference) to a
3707 variable that goes out of scope after the function returns.
3708 -Wreturn-type
3710 Warn whenever a function is defined with a return type that
3711 defaults to "int". Also warn about any "return" statement with no
3712 return value in a function whose return type is not "void" (falling
3713 off the end of the function body is considered returning without a
3714 value).
3715 For C only, warn about a "return" statement with an expression in a
3717 function whose return type is "void", unless the expression type is
3718 also "void". As a GNU extension, the latter case is accepted
3719 without a warning unless -Wpedantic is used.
3720 For C++, a function without return type always produces a
3722 diagnostic message, even when -Wno-return-type is specified. The
3723 only exceptions are "main" and functions defined in system headers.
3724 This warning is enabled by default for C++ and is enabled by -Wall.
3726 -Wshift-count-negative
3728 Warn if shift count is negative. This warning is enabled by
3729 default.
3730 -Wshift-count-overflow
3732 Warn if shift count >= width of type. This warning is enabled by
3733 default.
3734 -Wshift-negative-value
3736 Warn if left shifting a negative value. This warning is enabled by
3737 -Wextra in C99 and C++11 modes (and newer).
3738 -Wshift-overflow
3740 -Wshift-overflow=n
3741 Warn about left shift overflows. This warning is enabled by
3742 default in C99 and C++11 modes (and newer).
3743 -Wshift-overflow=1
3745 This is the warning level of -Wshift-overflow and is enabled by
3746 default in C99 and C++11 modes (and newer). This warning level
3747 does not warn about left-shifting 1 into the sign bit.
3748 (However, in C, such an overflow is still rejected in contexts
3749 where an integer constant expression is required.)
3750 -Wshift-overflow=2
3752 This warning level also warns about left-shifting 1 into the
3753 sign bit, unless C++14 mode is active.
3754 -Wswitch
3756 Warn whenever a "switch" statement has an index of enumerated type
3757 and lacks a "case" for one or more of the named codes of that
3758 enumeration. (The presence of a "default" label prevents this
3759 warning.) "case" labels outside the enumeration range also provoke
3760 warnings when this option is used (even if there is a "default"
3761 label). This warning is enabled by -Wall.
3762 -Wswitch-default
3764 Warn whenever a "switch" statement does not have a "default" case.
3765 -Wswitch-enum
3767 Warn whenever a "switch" statement has an index of enumerated type
3768 and lacks a "case" for one or more of the named codes of that
3769 enumeration. "case" labels outside the enumeration range also
3770 provoke warnings when this option is used. The only difference
3771 between -Wswitch and this option is that this option gives a
3772 warning about an omitted enumeration code even if there is a
3773 "default" label.
3774 -Wswitch-bool
3776 Warn whenever a "switch" statement has an index of boolean type and
3777 the case values are outside the range of a boolean type. It is
3778 possible to suppress this warning by casting the controlling
3779 expression to a type other than "bool". For example:
3780 switch ((int) (a == 4))
3782 {
3783 ...
3784 }
3785 This warning is enabled by default for C and C++ programs.
3787 -Wswitch-unreachable
3789 Warn whenever a "switch" statement contains statements between the
3790 controlling expression and the first case label, which will never
3791 be executed. For example:
3792 switch (cond)
3794 {
3795 i = 15;
3796 ...
3797 case 5:
3798 ...
3799 }
3800 -Wswitch-unreachable does not warn if the statement between the
3802 controlling expression and the first case label is just a
3803 declaration:
3804 switch (cond)
3806 {
3807 int i;
3808 ...
3809 case 5:
3810 i = 5;
3811 ...
3812 }
3813 This warning is enabled by default for C and C++ programs.
3815 -Wsync-nand (C and C++ only)
3817 Warn when "__sync_fetch_and_nand" and "__sync_nand_and_fetch"
3818 built-in functions are used. These functions changed semantics in
3819 GCC 4.4.
3820 -Wunused-but-set-parameter
3822 Warn whenever a function parameter is assigned to, but otherwise
3823 unused (aside from its declaration).
3824 To suppress this warning use the "unused" attribute.
3826 This warning is also enabled by -Wunused together with -Wextra.
3828 -Wunused-but-set-variable
3830 Warn whenever a local variable is assigned to, but otherwise unused
3831 (aside from its declaration). This warning is enabled by -Wall.
3832 To suppress this warning use the "unused" attribute.
3834 This warning is also enabled by -Wunused, which is enabled by
3836 -Wall.
3837 -Wunused-function
3839 Warn whenever a static function is declared but not defined or a
3840 non-inline static function is unused. This warning is enabled by
3841 -Wall.
3842 -Wunused-label
3844 Warn whenever a label is declared but not used. This warning is
3845 enabled by -Wall.
3846 To suppress this warning use the "unused" attribute.
3848 -Wunused-local-typedefs (C, Objective-C, C++ and Objective-C++ only)
3850 Warn when a typedef locally defined in a function is not used.
3851 This warning is enabled by -Wall.
3852 -Wunused-parameter
3854 Warn whenever a function parameter is unused aside from its
3855 declaration.
3856 To suppress this warning use the "unused" attribute.
3858 -Wno-unused-result
3860 Do not warn if a caller of a function marked with attribute
3861 "warn_unused_result" does not use its return value. The default is
3862 -Wunused-result.
3863 -Wunused-variable
3865 Warn whenever a local or static variable is unused aside from its
3866 declaration. This option implies -Wunused-const-variable=1 for C,
3867 but not for C++. This warning is enabled by -Wall.
3868 To suppress this warning use the "unused" attribute.
3870 -Wunused-const-variable
3872 -Wunused-const-variable=n
3873 Warn whenever a constant static variable is unused aside from its
3874 declaration. -Wunused-const-variable=1 is enabled by
3875 -Wunused-variable for C, but not for C++. In C this declares
3876 variable storage, but in C++ this is not an error since const
3877 variables take the place of "#define"s.
3878 To suppress this warning use the "unused" attribute.
3880 -Wunused-const-variable=1
3882 This is the warning level that is enabled by -Wunused-variable
3883 for C. It warns only about unused static const variables
3884 defined in the main compilation unit, but not about static
3885 const variables declared in any header included.
3886 -Wunused-const-variable=2
3888 This warning level also warns for unused constant static
3889 variables in headers (excluding system headers). This is the
3890 warning level of -Wunused-const-variable and must be explicitly
3891 requested since in C++ this isn't an error and in C it might be
3892 harder to clean up all headers included.
3893 -Wunused-value
3895 Warn whenever a statement computes a result that is explicitly not
3896 used. To suppress this warning cast the unused expression to
3897 "void". This includes an expression-statement or the left-hand side
3898 of a comma expression that contains no side effects. For example,
3899 an expression such as "x[i,j]" causes a warning, while
3900 "x[(void)i,j]" does not.
3901 This warning is enabled by -Wall.
3903 -Wunused
3905 All the above -Wunused options combined.
3906 In order to get a warning about an unused function parameter, you
3908 must either specify -Wextra -Wunused (note that -Wall implies
3909 -Wunused), or separately specify -Wunused-parameter.
3910 -Wuninitialized
3912 Warn if an automatic variable is used without first being
3913 initialized or if a variable may be clobbered by a "setjmp" call.
3914 In C++, warn if a non-static reference or non-static "const" member
3915 appears in a class without constructors.
3916 If you want to warn about code that uses the uninitialized value of
3918 the variable in its own initializer, use the -Winit-self option.
3919 These warnings occur for individual uninitialized or clobbered
3921 elements of structure, union or array variables as well as for
3922 variables that are uninitialized or clobbered as a whole. They do
3923 not occur for variables or elements declared "volatile". Because
3924 these warnings depend on optimization, the exact variables or
3925 elements for which there are warnings depends on the precise
3926 optimization options and version of GCC used.
3927 Note that there may be no warning about a variable that is used
3929 only to compute a value that itself is never used, because such
3930 computations may be deleted by data flow analysis before the
3931 warnings are printed.
3932 -Winvalid-memory-model
3934 Warn for invocations of __atomic Builtins, __sync Builtins, and the
3935 C11 atomic generic functions with a memory consistency argument
3936 that is either invalid for the operation or outside the range of
3937 values of the "memory_order" enumeration. For example, since the
3938 "__atomic_store" and "__atomic_store_n" built-ins are only defined
3939 for the relaxed, release, and sequentially consistent memory orders
3940 the following code is diagnosed:
3941 void store (int *i)
3943 {
3944 __atomic_store_n (i, 0, memory_order_consume);
3945 }
3946 -Winvalid-memory-model is enabled by default.
3948 -Wmaybe-uninitialized
3950 For an automatic (i.e. local) variable, if there exists a path from
3951 the function entry to a use of the variable that is initialized,
3952 but there exist some other paths for which the variable is not
3953 initialized, the compiler emits a warning if it cannot prove the
3954 uninitialized paths are not executed at run time.
3955 These warnings are only possible in optimizing compilation, because
3957 otherwise GCC does not keep track of the state of variables.
3958 These warnings are made optional because GCC may not be able to
3960 determine when the code is correct in spite of appearing to have an
3961 error. Here is one example of how this can happen:
3962 {
3964 int x;
3965 switch (y)
3966 {
3967 case 1: x = 1;
3968 break;
3969 case 2: x = 4;
3970 break;
3971 case 3: x = 5;
3972 }
3973 foo (x);
3974 }
3975 If the value of "y" is always 1, 2 or 3, then "x" is always
3977 initialized, but GCC doesn't know this. To suppress the warning,
3978 you need to provide a default case with assert(0) or similar code.
3979 This option also warns when a non-volatile automatic variable might
3981 be changed by a call to "longjmp". The compiler sees only the
3982 calls to "setjmp". It cannot know where "longjmp" will be called;
3983 in fact, a signal handler could call it at any point in the code.
3984 As a result, you may get a warning even when there is in fact no
3985 problem because "longjmp" cannot in fact be called at the place
3986 that would cause a problem.
3987 Some spurious warnings can be avoided if you declare all the
3989 functions you use that never return as "noreturn".
3990 This warning is enabled by -Wall or -Wextra.
3992 -Wunknown-pragmas
3994 Warn when a "#pragma" directive is encountered that is not
3995 understood by GCC. If this command-line option is used, warnings
3996 are even issued for unknown pragmas in system header files. This
3997 is not the case if the warnings are only enabled by the -Wall
3998 command-line option.
3999 -Wno-pragmas
4001 Do not warn about misuses of pragmas, such as incorrect parameters,
4002 invalid syntax, or conflicts between pragmas. See also
4003 -Wunknown-pragmas.
4004 -Wstrict-aliasing
4006 This option is only active when -fstrict-aliasing is active. It
4007 warns about code that might break the strict aliasing rules that
4008 the compiler is using for optimization. The warning does not catch
4009 all cases, but does attempt to catch the more common pitfalls. It
4010 is included in -Wall. It is equivalent to -Wstrict-aliasing=3
4011 -Wstrict-aliasing=n
4013 This option is only active when -fstrict-aliasing is active. It
4014 warns about code that might break the strict aliasing rules that
4015 the compiler is using for optimization. Higher levels correspond
4016 to higher accuracy (fewer false positives). Higher levels also
4017 correspond to more effort, similar to the way -O works.
4018 -Wstrict-aliasing is equivalent to -Wstrict-aliasing=3.
4019 Level 1: Most aggressive, quick, least accurate. Possibly useful
4021 when higher levels do not warn but -fstrict-aliasing still breaks
4022 the code, as it has very few false negatives. However, it has many
4023 false positives. Warns for all pointer conversions between
4024 possibly incompatible types, even if never dereferenced. Runs in
4025 the front end only.
4026 Level 2: Aggressive, quick, not too precise. May still have many
4028 false positives (not as many as level 1 though), and few false
4029 negatives (but possibly more than level 1). Unlike level 1, it
4030 only warns when an address is taken. Warns about incomplete types.
4031 Runs in the front end only.
4032 Level 3 (default for -Wstrict-aliasing): Should have very few false
4034 positives and few false negatives. Slightly slower than levels 1
4035 or 2 when optimization is enabled. Takes care of the common
4036 pun+dereference pattern in the front end: "*(int*)&some_float". If
4037 optimization is enabled, it also runs in the back end, where it
4038 deals with multiple statement cases using flow-sensitive points-to
4039 information. Only warns when the converted pointer is
4040 dereferenced. Does not warn about incomplete types.
4041 -Wstrict-overflow
4043 -Wstrict-overflow=n
4044 This option is only active when signed overflow is undefined. It
4045 warns about cases where the compiler optimizes based on the
4046 assumption that signed overflow does not occur. Note that it does
4047 not warn about all cases where the code might overflow: it only
4048 warns about cases where the compiler implements some optimization.
4049 Thus this warning depends on the optimization level.
4050 An optimization that assumes that signed overflow does not occur is
4052 perfectly safe if the values of the variables involved are such
4053 that overflow never does, in fact, occur. Therefore this warning
4054 can easily give a false positive: a warning about code that is not
4055 actually a problem. To help focus on important issues, several
4056 warning levels are defined. No warnings are issued for the use of
4057 undefined signed overflow when estimating how many iterations a
4058 loop requires, in particular when determining whether a loop will
4059 be executed at all.
4060 -Wstrict-overflow=1
4062 Warn about cases that are both questionable and easy to avoid.
4063 For example the compiler simplifies "x + 1 > x" to 1. This
4064 level of -Wstrict-overflow is enabled by -Wall; higher levels
4065 are not, and must be explicitly requested.
4066 -Wstrict-overflow=2
4068 Also warn about other cases where a comparison is simplified to
4069 a constant. For example: "abs (x) >= 0". This can only be
4070 simplified when signed integer overflow is undefined, because
4071 "abs (INT_MIN)" overflows to "INT_MIN", which is less than
4072 zero. -Wstrict-overflow (with no level) is the same as
4073 -Wstrict-overflow=2.
4074 -Wstrict-overflow=3
4076 Also warn about other cases where a comparison is simplified.
4077 For example: "x + 1 > 1" is simplified to "x > 0".
4078 -Wstrict-overflow=4
4080 Also warn about other simplifications not covered by the above
4081 cases. For example: "(x * 10) / 5" is simplified to "x * 2".
4082 -Wstrict-overflow=5
4084 Also warn about cases where the compiler reduces the magnitude
4085 of a constant involved in a comparison. For example: "x + 2 >
4086 y" is simplified to "x + 1 >= y". This is reported only at the
4087 highest warning level because this simplification applies to
4088 many comparisons, so this warning level gives a very large
4089 number of false positives.
4090 -Wstringop-overflow
4092 -Wstringop-overflow=type
4093 Warn for calls to string manipulation functions such as "memcpy"
4094 and "strcpy" that are determined to overflow the destination
4095 buffer. The optional argument is one greater than the type of
4096 Object Size Checking to perform to determine the size of the
4097 destination. The argument is meaningful only for functions that
4098 operate on character arrays but not for raw memory functions like
4099 "memcpy" which always make use of Object Size type-0. The option
4100 also warns for calls that specify a size in excess of the largest
4101 possible object or at most "SIZE_MAX / 2" bytes. The option
4102 produces the best results with optimization enabled but can detect
4103 a small subset of simple buffer overflows even without optimization
4104 in calls to the GCC built-in functions like "__builtin_memcpy" that
4105 correspond to the standard functions. In any case, the option
4106 warns about just a subset of buffer overflows detected by the
4107 corresponding overflow checking built-ins. For example, the option
4108 will issue a warning for the "strcpy" call below because it copies
4109 at least 5 characters (the string "blue" including the terminating
4110 NUL) into the buffer of size 4.
4111 enum Color { blue, purple, yellow };
4113 const char* f (enum Color clr)
4114 {
4115 static char buf [4];
4116 const char *str;
4117 switch (clr)
4118 {
4119 case blue: str = "blue"; break;
4120 case purple: str = "purple"; break;
4121 case yellow: str = "yellow"; break;
4122 }
4123 return strcpy (buf, str); // warning here
4125 }
4126 Option -Wstringop-overflow=2 is enabled by default.
4128 -Wstringop-overflow
4130 -Wstringop-overflow=1
4131 The -Wstringop-overflow=1 option uses type-zero Object Size
4132 Checking to determine the sizes of destination objects. This
4133 is the default setting of the option. At this setting the
4134 option will not warn for writes past the end of subobjects of
4135 larger objects accessed by pointers unless the size of the
4136 largest surrounding object is known. When the destination may
4137 be one of several objects it is assumed to be the largest one
4138 of them. On Linux systems, when optimization is enabled at
4139 this setting the option warns for the same code as when the
4140 "_FORTIFY_SOURCE" macro is defined to a non-zero value.
4141 -Wstringop-overflow=2
4143 The -Wstringop-overflow=2 option uses type-one Object Size
4144 Checking to determine the sizes of destination objects. At
4145 this setting the option will warn about overflows when writing
4146 to members of the largest complete objects whose exact size is
4147 known. It will, however, not warn for excessive writes to the
4148 same members of unknown objects referenced by pointers since
4149 they may point to arrays containing unknown numbers of
4150 elements.
4151 -Wstringop-overflow=3
4153 The -Wstringop-overflow=3 option uses type-two Object Size
4154 Checking to determine the sizes of destination objects. At
4155 this setting the option warns about overflowing the smallest
4156 object or data member. This is the most restrictive setting of
4157 the option that may result in warnings for safe code.
4158 -Wstringop-overflow=4
4160 The -Wstringop-overflow=4 option uses type-three Object Size
4161 Checking to determine the sizes of destination objects. At
4162 this setting the option will warn about overflowing any data
4163 members, and when the destination is one of several objects it
4164 uses the size of the largest of them to decide whether to issue
4165 a warning. Similarly to -Wstringop-overflow=3 this setting of
4166 the option may result in warnings for benign code.
4167 -Wstringop-truncation
4169 Warn for calls to bounded string manipulation functions such as
4170 "strncat", "strncpy", and "stpncpy" that may either truncate the
4171 copied string or leave the destination unchanged.
4172 In the following example, the call to "strncat" specifies a bound
4174 that is less than the length of the source string. As a result,
4175 the copy of the source will be truncated and so the call is
4176 diagnosed. To avoid the warning use "bufsize - strlen (buf) - 1)"
4177 as the bound.
4178 void append (char *buf, size_t bufsize)
4180 {
4181 strncat (buf, ".txt", 3);
4182 }
4183 As another example, the following call to "strncpy" results in
4185 copying to "d" just the characters preceding the terminating NUL,
4186 without appending the NUL to the end. Assuming the result of
4187 "strncpy" is necessarily a NUL-terminated string is a common
4188 mistake, and so the call is diagnosed. To avoid the warning when
4189 the result is not expected to be NUL-terminated, call "memcpy"
4190 instead.
4191 void copy (char *d, const char *s)
4193 {
4194 strncpy (d, s, strlen (s));
4195 }
4196 In the following example, the call to "strncpy" specifies the size
4198 of the destination buffer as the bound. If the length of the
4199 source string is equal to or greater than this size the result of
4200 the copy will not be NUL-terminated. Therefore, the call is also
4201 diagnosed. To avoid the warning, specify "sizeof buf - 1" as the
4202 bound and set the last element of the buffer to "NUL".
4203 void copy (const char *s)
4205 {
4206 char buf[80];
4207 strncpy (buf, s, sizeof buf);
4208 ...
4209 }
4210 In situations where a character array is intended to store a
4212 sequence of bytes with no terminating "NUL" such an array may be
4213 annotated with attribute "nonstring" to avoid this warning. Such
4214 arrays, however, are not suitable arguments to functions that
4215 expect "NUL"-terminated strings. To help detect accidental misuses
4216 of such arrays GCC issues warnings unless it can prove that the use
4217 is safe.
4218 Option -Wstringop-truncation is enabled by -Wall.
4220 -Wsuggest-attribute=[pure|const|noreturn|format|cold|malloc]
4222 Warn for cases where adding an attribute may be beneficial. The
4223 attributes currently supported are listed below.
4224 -Wsuggest-attribute=pure
4226 -Wsuggest-attribute=const
4227 -Wsuggest-attribute=noreturn
4228 -Wsuggest-attribute=malloc
4229 Warn about functions that might be candidates for attributes
4230 "pure", "const" or "noreturn" or "malloc". The compiler only
4231 warns for functions visible in other compilation units or (in
4232 the case of "pure" and "const") if it cannot prove that the
4233 function returns normally. A function returns normally if it
4234 doesn't contain an infinite loop or return abnormally by
4235 throwing, calling "abort" or trapping. This analysis requires
4236 option -fipa-pure-const, which is enabled by default at -O and
4237 higher. Higher optimization levels improve the accuracy of the
4238 analysis.
4239 -Wsuggest-attribute=format
4241 -Wmissing-format-attribute
4242 Warn about function pointers that might be candidates for
4243 "format" attributes. Note these are only possible candidates,
4244 not absolute ones. GCC guesses that function pointers with
4245 "format" attributes that are used in assignment,
4246 initialization, parameter passing or return statements should
4247 have a corresponding "format" attribute in the resulting type.
4248 I.e. the left-hand side of the assignment or initialization,
4249 the type of the parameter variable, or the return type of the
4250 containing function respectively should also have a "format"
4251 attribute to avoid the warning.
4252 GCC also warns about function definitions that might be
4254 candidates for "format" attributes. Again, these are only
4255 possible candidates. GCC guesses that "format" attributes
4256 might be appropriate for any function that calls a function
4257 like "vprintf" or "vscanf", but this might not always be the
4258 case, and some functions for which "format" attributes are
4259 appropriate may not be detected.
4260 -Wsuggest-attribute=cold
4262 Warn about functions that might be candidates for "cold"
4263 attribute. This is based on static detection and generally
4264 will only warn about functions which always leads to a call to
4265 another "cold" function such as wrappers of C++ "throw" or
4266 fatal error reporting functions leading to "abort".
4267 -Wsuggest-final-types
4269 Warn about types with virtual methods where code quality would be
4270 improved if the type were declared with the C++11 "final"
4271 specifier, or, if possible, declared in an anonymous namespace.
4272 This allows GCC to more aggressively devirtualize the polymorphic
4273 calls. This warning is more effective with link time optimization,
4274 where the information about the class hierarchy graph is more
4275 complete.
4276 -Wsuggest-final-methods
4278 Warn about virtual methods where code quality would be improved if
4279 the method were declared with the C++11 "final" specifier, or, if
4280 possible, its type were declared in an anonymous namespace or with
4281 the "final" specifier. This warning is more effective with link-
4282 time optimization, where the information about the class hierarchy
4283 graph is more complete. It is recommended to first consider
4284 suggestions of -Wsuggest-final-types and then rebuild with new
4285 annotations.
4286 -Wsuggest-override
4288 Warn about overriding virtual functions that are not marked with
4289 the override keyword.
4290 -Walloc-zero
4292 Warn about calls to allocation functions decorated with attribute
4293 "alloc_size" that specify zero bytes, including those to the built-
4294 in forms of the functions "aligned_alloc", "alloca", "calloc",
4295 "malloc", and "realloc". Because the behavior of these functions
4296 when called with a zero size differs among implementations (and in
4297 the case of "realloc" has been deprecated) relying on it may result
4298 in subtle portability bugs and should be avoided.
4299 -Walloc-size-larger-than=n
4301 Warn about calls to functions decorated with attribute "alloc_size"
4302 that attempt to allocate objects larger than the specified number
4303 of bytes, or where the result of the size computation in an integer
4304 type with infinite precision would exceed "SIZE_MAX / 2". The
4305 option argument n may end in one of the standard suffixes
4306 designating a multiple of bytes such as "kB" and "KiB" for kilobyte
4307 and kibibyte, respectively, "MB" and "MiB" for megabyte and
4308 mebibyte, and so on. -Walloc-size-larger-than=PTRDIFF_MAX is
4309 enabled by default. Warnings controlled by the option can be
4310 disabled by specifying n of SIZE_MAX or more.
4311 -Walloca
4313 This option warns on all uses of "alloca" in the source.
4314 -Walloca-larger-than=n
4316 This option warns on calls to "alloca" that are not bounded by a
4317 controlling predicate limiting its argument of integer type to at
4318 most n bytes, or calls to "alloca" where the bound is unknown.
4319 Arguments of non-integer types are considered unbounded even if
4320 they appear to be constrained to the expected range.
4321 For example, a bounded case of "alloca" could be:
4323 void func (size_t n)
4325 {
4326 void *p;
4327 if (n <= 1000)
4328 p = alloca (n);
4329 else
4330 p = malloc (n);
4331 f (p);
4332 }
4333 In the above example, passing "-Walloca-larger-than=1000" would not
4335 issue a warning because the call to "alloca" is known to be at most
4336 1000 bytes. However, if "-Walloca-larger-than=500" were passed,
4337 the compiler would emit a warning.
4338 Unbounded uses, on the other hand, are uses of "alloca" with no
4340 controlling predicate constraining its integer argument. For
4341 example:
4342 void func ()
4344 {
4345 void *p = alloca (n);
4346 f (p);
4347 }
4348 If "-Walloca-larger-than=500" were passed, the above would trigger
4350 a warning, but this time because of the lack of bounds checking.
4351 Note, that even seemingly correct code involving signed integers
4353 could cause a warning:
4354 void func (signed int n)
4356 {
4357 if (n < 500)
4358 {
4359 p = alloca (n);
4360 f (p);
4361 }
4362 }
4363 In the above example, n could be negative, causing a larger than
4365 expected argument to be implicitly cast into the "alloca" call.
4366 This option also warns when "alloca" is used in a loop.
4368 This warning is not enabled by -Wall, and is only active when
4370 -ftree-vrp is active (default for -O2 and above).
4371 See also -Wvla-larger-than=n.
4373 -Warray-bounds
4375 -Warray-bounds=n
4376 This option is only active when -ftree-vrp is active (default for
4377 -O2 and above). It warns about subscripts to arrays that are always
4378 out of bounds. This warning is enabled by -Wall.
4379 -Warray-bounds=1
4381 This is the warning level of -Warray-bounds and is enabled by
4382 -Wall; higher levels are not, and must be explicitly requested.
4383 -Warray-bounds=2
4385 This warning level also warns about out of bounds access for
4386 arrays at the end of a struct and for arrays accessed through
4387 pointers. This warning level may give a larger number of false
4388 positives and is deactivated by default.
4389 -Wattribute-alias
4391 Warn about declarations using the "alias" and similar attributes
4392 whose target is incompatible with the type of the alias.
4393 -Wbool-compare
4395 Warn about boolean expression compared with an integer value
4396 different from "true"/"false". For instance, the following
4397 comparison is always false:
4398 int n = 5;
4400 ...
4401 if ((n > 1) == 2) { ... }
4402 This warning is enabled by -Wall.
4404 -Wbool-operation
4406 Warn about suspicious operations on expressions of a boolean type.
4407 For instance, bitwise negation of a boolean is very likely a bug in
4408 the program. For C, this warning also warns about incrementing or
4409 decrementing a boolean, which rarely makes sense. (In C++,
4410 decrementing a boolean is always invalid. Incrementing a boolean
4411 is invalid in C++17, and deprecated otherwise.)
4412 This warning is enabled by -Wall.
4414 -Wduplicated-branches
4416 Warn when an if-else has identical branches. This warning detects
4417 cases like
4418 if (p != NULL)
4420 return 0;
4421 else
4422 return 0;
4423 It doesn't warn when both branches contain just a null statement.
4425 This warning also warn for conditional operators:
4426 int i = x ? *p : *p;
4428 -Wduplicated-cond
4430 Warn about duplicated conditions in an if-else-if chain. For
4431 instance, warn for the following code:
4432 if (p->q != NULL) { ... }
4434 else if (p->q != NULL) { ... }
4435 -Wframe-address
4437 Warn when the __builtin_frame_address or __builtin_return_address
4438 is called with an argument greater than 0. Such calls may return
4439 indeterminate values or crash the program. The warning is included
4440 in -Wall.
4441 -Wno-discarded-qualifiers (C and Objective-C only)
4443 Do not warn if type qualifiers on pointers are being discarded.
4444 Typically, the compiler warns if a "const char *" variable is
4445 passed to a function that takes a "char *" parameter. This option
4446 can be used to suppress such a warning.
4447 -Wno-discarded-array-qualifiers (C and Objective-C only)
4449 Do not warn if type qualifiers on arrays which are pointer targets
4450 are being discarded. Typically, the compiler warns if a "const int
4451 (*)[]" variable is passed to a function that takes a "int (*)[]"
4452 parameter. This option can be used to suppress such a warning.
4453 -Wno-incompatible-pointer-types (C and Objective-C only)
4455 Do not warn when there is a conversion between pointers that have
4456 incompatible types. This warning is for cases not covered by
4457 -Wno-pointer-sign, which warns for pointer argument passing or
4458 assignment with different signedness.
4459 -Wno-int-conversion (C and Objective-C only)
4461 Do not warn about incompatible integer to pointer and pointer to
4462 integer conversions. This warning is about implicit conversions;
4463 for explicit conversions the warnings -Wno-int-to-pointer-cast and
4464 -Wno-pointer-to-int-cast may be used.
4465 -Wno-div-by-zero
4467 Do not warn about compile-time integer division by zero. Floating-
4468 point division by zero is not warned about, as it can be a
4469 legitimate way of obtaining infinities and NaNs.
4470 -Wsystem-headers
4472 Print warning messages for constructs found in system header files.
4473 Warnings from system headers are normally suppressed, on the
4474 assumption that they usually do not indicate real problems and
4475 would only make the compiler output harder to read. Using this
4476 command-line option tells GCC to emit warnings from system headers
4477 as if they occurred in user code. However, note that using -Wall
4478 in conjunction with this option does not warn about unknown pragmas
4479 in system headers---for that, -Wunknown-pragmas must also be used.
4480 -Wtautological-compare
4482 Warn if a self-comparison always evaluates to true or false. This
4483 warning detects various mistakes such as:
4484 int i = 1;
4486 ...
4487 if (i > i) { ... }
4488 This warning also warns about bitwise comparisons that always
4490 evaluate to true or false, for instance:
4491 if ((a & 16) == 10) { ... }
4493 will always be false.
4495 This warning is enabled by -Wall.
4497 -Wtrampolines
4499 Warn about trampolines generated for pointers to nested functions.
4500 A trampoline is a small piece of data or code that is created at
4501 run time on the stack when the address of a nested function is
4502 taken, and is used to call the nested function indirectly. For
4503 some targets, it is made up of data only and thus requires no
4504 special treatment. But, for most targets, it is made up of code
4505 and thus requires the stack to be made executable in order for the
4506 program to work properly.
4507 -Wfloat-equal
4509 Warn if floating-point values are used in equality comparisons.
4510 The idea behind this is that sometimes it is convenient (for the
4512 programmer) to consider floating-point values as approximations to
4513 infinitely precise real numbers. If you are doing this, then you
4514 need to compute (by analyzing the code, or in some other way) the
4515 maximum or likely maximum error that the computation introduces,
4516 and allow for it when performing comparisons (and when producing
4517 output, but that's a different problem). In particular, instead of
4518 testing for equality, you should check to see whether the two
4519 values have ranges that overlap; and this is done with the
4520 relational operators, so equality comparisons are probably
4521 mistaken.
4522 -Wtraditional (C and Objective-C only)
4524 Warn about certain constructs that behave differently in
4525 traditional and ISO C. Also warn about ISO C constructs that have
4526 no traditional C equivalent, and/or problematic constructs that
4527 should be avoided.
4528 * Macro parameters that appear within string literals in the
4530 macro body. In traditional C macro replacement takes place
4531 within string literals, but in ISO C it does not.
4532 * In traditional C, some preprocessor directives did not exist.
4534 Traditional preprocessors only considered a line to be a
4535 directive if the # appeared in column 1 on the line. Therefore
4536 -Wtraditional warns about directives that traditional C
4537 understands but ignores because the # does not appear as the
4538 first character on the line. It also suggests you hide
4539 directives like "#pragma" not understood by traditional C by
4540 indenting them. Some traditional implementations do not
4541 recognize "#elif", so this option suggests avoiding it
4542 altogether.
4543 * A function-like macro that appears without arguments.
4545 * The unary plus operator.
4547 * The U integer constant suffix, or the F or L floating-point
4549 constant suffixes. (Traditional C does support the L suffix on
4550 integer constants.) Note, these suffixes appear in macros
4551 defined in the system headers of most modern systems, e.g. the
4552 _MIN/_MAX macros in "<limits.h>". Use of these macros in user
4553 code might normally lead to spurious warnings, however GCC's
4554 integrated preprocessor has enough context to avoid warning in
4555 these cases.
4556 * A function declared external in one block and then used after
4558 the end of the block.
4559 * A "switch" statement has an operand of type "long".
4561 * A non-"static" function declaration follows a "static" one.
4563 This construct is not accepted by some traditional C compilers.
4564 * The ISO type of an integer constant has a different width or
4566 signedness from its traditional type. This warning is only
4567 issued if the base of the constant is ten. I.e. hexadecimal or
4568 octal values, which typically represent bit patterns, are not
4569 warned about.
4570 * Usage of ISO string concatenation is detected.
4572 * Initialization of automatic aggregates.
4574 * Identifier conflicts with labels. Traditional C lacks a
4576 separate namespace for labels.
4577 * Initialization of unions. If the initializer is zero, the
4579 warning is omitted. This is done under the assumption that the
4580 zero initializer in user code appears conditioned on e.g.
4581 "__STDC__" to avoid missing initializer warnings and relies on
4582 default initialization to zero in the traditional C case.
4583 * Conversions by prototypes between fixed/floating-point values
4585 and vice versa. The absence of these prototypes when compiling
4586 with traditional C causes serious problems. This is a subset
4587 of the possible conversion warnings; for the full set use
4588 -Wtraditional-conversion.
4589 * Use of ISO C style function definitions. This warning
4591 intentionally is not issued for prototype declarations or
4592 variadic functions because these ISO C features appear in your
4593 code when using libiberty's traditional C compatibility macros,
4594 "PARAMS" and "VPARAMS". This warning is also bypassed for
4595 nested functions because that feature is already a GCC
4596 extension and thus not relevant to traditional C compatibility.
4597 -Wtraditional-conversion (C and Objective-C only)
4599 Warn if a prototype causes a type conversion that is different from
4600 what would happen to the same argument in the absence of a
4601 prototype. This includes conversions of fixed point to floating
4602 and vice versa, and conversions changing the width or signedness of
4603 a fixed-point argument except when the same as the default
4604 promotion.
4605 -Wdeclaration-after-statement (C and Objective-C only)
4607 Warn when a declaration is found after a statement in a block.
4608 This construct, known from C++, was introduced with ISO C99 and is
4609 by default allowed in GCC. It is not supported by ISO C90.
4610 -Wshadow
4612 Warn whenever a local variable or type declaration shadows another
4613 variable, parameter, type, class member (in C++), or instance
4614 variable (in Objective-C) or whenever a built-in function is
4615 shadowed. Note that in C++, the compiler warns if a local variable
4616 shadows an explicit typedef, but not if it shadows a
4617 struct/class/enum. Same as -Wshadow=global.
4618 -Wno-shadow-ivar (Objective-C only)
4620 Do not warn whenever a local variable shadows an instance variable
4621 in an Objective-C method.
4622 -Wshadow=global
4624 The default for -Wshadow. Warns for any (global) shadowing.
4625 -Wshadow=local
4627 Warn when a local variable shadows another local variable or
4628 parameter. This warning is enabled by -Wshadow=global.
4629 -Wshadow=compatible-local
4631 Warn when a local variable shadows another local variable or
4632 parameter whose type is compatible with that of the shadowing
4633 variable. In C++, type compatibility here means the type of the
4634 shadowing variable can be converted to that of the shadowed
4635 variable. The creation of this flag (in addition to -Wshadow=local)
4636 is based on the idea that when a local variable shadows another one
4637 of incompatible type, it is most likely intentional, not a bug or
4638 typo, as shown in the following example:
4639 for (SomeIterator i = SomeObj.begin(); i != SomeObj.end(); ++i)
4641 {
4642 for (int i = 0; i < N; ++i)
4643 {
4644 ...
4645 }
4646 ...
4647 }
4648 Since the two variable "i" in the example above have incompatible
4650 types, enabling only -Wshadow=compatible-local will not emit a
4651 warning. Because their types are incompatible, if a programmer
4652 accidentally uses one in place of the other, type checking will
4653 catch that and emit an error or warning. So not warning (about
4654 shadowing) in this case will not lead to undetected bugs. Use of
4655 this flag instead of -Wshadow=local can possibly reduce the number
4656 of warnings triggered by intentional shadowing.
4657 This warning is enabled by -Wshadow=local.
4659 -Wlarger-than=len
4661 Warn whenever an object of larger than len bytes is defined.
4662 -Wframe-larger-than=len
4664 Warn if the size of a function frame is larger than len bytes. The
4665 computation done to determine the stack frame size is approximate
4666 and not conservative. The actual requirements may be somewhat
4667 greater than len even if you do not get a warning. In addition,
4668 any space allocated via "alloca", variable-length arrays, or
4669 related constructs is not included by the compiler when determining
4670 whether or not to issue a warning.
4671 -Wno-free-nonheap-object
4673 Do not warn when attempting to free an object that was not
4674 allocated on the heap.
4675 -Wstack-usage=len
4677 Warn if the stack usage of a function might be larger than len
4678 bytes. The computation done to determine the stack usage is
4679 conservative. Any space allocated via "alloca", variable-length
4680 arrays, or related constructs is included by the compiler when
4681 determining whether or not to issue a warning.
4682 The message is in keeping with the output of -fstack-usage.
4684 * If the stack usage is fully static but exceeds the specified
4686 amount, it's:
4687 warning: stack usage is 1120 bytes
4689 * If the stack usage is (partly) dynamic but bounded, it's:
4691 warning: stack usage might be 1648 bytes
4693 * If the stack usage is (partly) dynamic and not bounded, it's:
4695 warning: stack usage might be unbounded
4697 -Wno-pedantic-ms-format (MinGW targets only)
4699 When used in combination with -Wformat and -pedantic without GNU
4700 extensions, this option disables the warnings about non-ISO
4701 "printf" / "scanf" format width specifiers "I32", "I64", and "I"
4702 used on Windows targets, which depend on the MS runtime.
4703 -Waligned-new
4705 Warn about a new-expression of a type that requires greater
4706 alignment than the "alignof(std::max_align_t)" but uses an
4707 allocation function without an explicit alignment parameter. This
4708 option is enabled by -Wall.
4709 Normally this only warns about global allocation functions, but
4711 -Waligned-new=all also warns about class member allocation
4712 functions.
4713 -Wplacement-new
4715 -Wplacement-new=n
4716 Warn about placement new expressions with undefined behavior, such
4717 as constructing an object in a buffer that is smaller than the type
4718 of the object. For example, the placement new expression below is
4719 diagnosed because it attempts to construct an array of 64 integers
4720 in a buffer only 64 bytes large.
4721 char buf [64];
4723 new (buf) int[64];
4724 This warning is enabled by default.
4726 -Wplacement-new=1
4728 This is the default warning level of -Wplacement-new. At this
4729 level the warning is not issued for some strictly undefined
4730 constructs that GCC allows as extensions for compatibility with
4731 legacy code. For example, the following "new" expression is
4732 not diagnosed at this level even though it has undefined
4733 behavior according to the C++ standard because it writes past
4734 the end of the one-element array.
4735 struct S { int n, a[1]; };
4737 S *s = (S *)malloc (sizeof *s + 31 * sizeof s->a[0]);
4738 new (s->a)int [32]();
4739 -Wplacement-new=2
4741 At this level, in addition to diagnosing all the same
4742 constructs as at level 1, a diagnostic is also issued for
4743 placement new expressions that construct an object in the last
4744 member of structure whose type is an array of a single element
4745 and whose size is less than the size of the object being
4746 constructed. While the previous example would be diagnosed,
4747 the following construct makes use of the flexible member array
4748 extension to avoid the warning at level 2.
4749 struct S { int n, a[]; };
4751 S *s = (S *)malloc (sizeof *s + 32 * sizeof s->a[0]);
4752 new (s->a)int [32]();
4753 -Wpointer-arith
4755 Warn about anything that depends on the "size of" a function type
4756 or of "void". GNU C assigns these types a size of 1, for
4757 convenience in calculations with "void *" pointers and pointers to
4758 functions. In C++, warn also when an arithmetic operation involves
4759 "NULL". This warning is also enabled by -Wpedantic.
4760 -Wpointer-compare
4762 Warn if a pointer is compared with a zero character constant. This
4763 usually means that the pointer was meant to be dereferenced. For
4764 example:
4765 const char *p = foo ();
4767 if (p == '\0')
4768 return 42;
4769 Note that the code above is invalid in C++11.
4771 This warning is enabled by default.
4773 -Wtype-limits
4775 Warn if a comparison is always true or always false due to the
4776 limited range of the data type, but do not warn for constant
4777 expressions. For example, warn if an unsigned variable is compared
4778 against zero with "<" or ">=". This warning is also enabled by
4779 -Wextra.
4780 -Wcomment
4782 -Wcomments
4783 Warn whenever a comment-start sequence /* appears in a /* comment,
4784 or whenever a backslash-newline appears in a // comment. This
4785 warning is enabled by -Wall.
4786 -Wtrigraphs
4788 Warn if any trigraphs are encountered that might change the meaning
4789 of the program. Trigraphs within comments are not warned about,
4790 except those that would form escaped newlines.
4791 This option is implied by -Wall. If -Wall is not given, this
4793 option is still enabled unless trigraphs are enabled. To get
4794 trigraph conversion without warnings, but get the other -Wall
4795 warnings, use -trigraphs -Wall -Wno-trigraphs.
4796 -Wundef
4798 Warn if an undefined identifier is evaluated in an "#if" directive.
4799 Such identifiers are replaced with zero.
4800 -Wexpansion-to-defined
4802 Warn whenever defined is encountered in the expansion of a macro
4803 (including the case where the macro is expanded by an #if
4804 directive). Such usage is not portable. This warning is also
4805 enabled by -Wpedantic and -Wextra.
4806 -Wunused-macros
4808 Warn about macros defined in the main file that are unused. A
4809 macro is used if it is expanded or tested for existence at least
4810 once. The preprocessor also warns if the macro has not been used
4811 at the time it is redefined or undefined.
4812 Built-in macros, macros defined on the command line, and macros
4814 defined in include files are not warned about.
4815 Note: If a macro is actually used, but only used in skipped
4817 conditional blocks, then the preprocessor reports it as unused. To
4818 avoid the warning in such a case, you might improve the scope of
4819 the macro's definition by, for example, moving it into the first
4820 skipped block. Alternatively, you could provide a dummy use with
4821 something like:
4822 #if defined the_macro_causing_the_warning
4824 #endif
4825 -Wno-endif-labels
4827 Do not warn whenever an "#else" or an "#endif" are followed by
4828 text. This sometimes happens in older programs with code of the
4829 form
4830 #if FOO
4832 ...
4833 #else FOO
4834 ...
4835 #endif FOO
4836 The second and third "FOO" should be in comments. This warning is
4838 on by default.
4839 -Wbad-function-cast (C and Objective-C only)
4841 Warn when a function call is cast to a non-matching type. For
4842 example, warn if a call to a function returning an integer type is
4843 cast to a pointer type.
4844 -Wc90-c99-compat (C and Objective-C only)
4846 Warn about features not present in ISO C90, but present in ISO C99.
4847 For instance, warn about use of variable length arrays, "long long"
4848 type, "bool" type, compound literals, designated initializers, and
4849 so on. This option is independent of the standards mode. Warnings
4850 are disabled in the expression that follows "__extension__".
4851 -Wc99-c11-compat (C and Objective-C only)
4853 Warn about features not present in ISO C99, but present in ISO C11.
4854 For instance, warn about use of anonymous structures and unions,
4855 "_Atomic" type qualifier, "_Thread_local" storage-class specifier,
4856 "_Alignas" specifier, "Alignof" operator, "_Generic" keyword, and
4857 so on. This option is independent of the standards mode. Warnings
4858 are disabled in the expression that follows "__extension__".
4859 -Wc++-compat (C and Objective-C only)
4861 Warn about ISO C constructs that are outside of the common subset
4862 of ISO C and ISO C++, e.g. request for implicit conversion from
4863 "void *" to a pointer to non-"void" type.
4864 -Wc++11-compat (C++ and Objective-C++ only)
4866 Warn about C++ constructs whose meaning differs between ISO C++
4867 1998 and ISO C++ 2011, e.g., identifiers in ISO C++ 1998 that are
4868 keywords in ISO C++ 2011. This warning turns on -Wnarrowing and is
4869 enabled by -Wall.
4870 -Wc++14-compat (C++ and Objective-C++ only)
4872 Warn about C++ constructs whose meaning differs between ISO C++
4873 2011 and ISO C++ 2014. This warning is enabled by -Wall.
4874 -Wc++17-compat (C++ and Objective-C++ only)
4876 Warn about C++ constructs whose meaning differs between ISO C++
4877 2014 and ISO C++ 2017. This warning is enabled by -Wall.
4878 -Wcast-qual
4880 Warn whenever a pointer is cast so as to remove a type qualifier
4881 from the target type. For example, warn if a "const char *" is
4882 cast to an ordinary "char *".
4883 Also warn when making a cast that introduces a type qualifier in an
4885 unsafe way. For example, casting "char **" to "const char **" is
4886 unsafe, as in this example:
4887 /* p is char ** value. */
4889 const char **q = (const char **) p;
4890 /* Assignment of readonly string to const char * is OK. */
4891 *q = "string";
4892 /* Now char** pointer points to read-only memory. */
4893 **p = 'b';
4894 -Wcast-align
4896 Warn whenever a pointer is cast such that the required alignment of
4897 the target is increased. For example, warn if a "char *" is cast
4898 to an "int *" on machines where integers can only be accessed at
4899 two- or four-byte boundaries.
4900 -Wcast-align=strict
4902 Warn whenever a pointer is cast such that the required alignment of
4903 the target is increased. For example, warn if a "char *" is cast
4904 to an "int *" regardless of the target machine.
4905 -Wcast-function-type
4907 Warn when a function pointer is cast to an incompatible function
4908 pointer. In a cast involving function types with a variable
4909 argument list only the types of initial arguments that are provided
4910 are considered. Any parameter of pointer-type matches any other
4911 pointer-type. Any benign differences in integral types are
4912 ignored, like "int" vs. "long" on ILP32 targets. Likewise type
4913 qualifiers are ignored. The function type "void (*) (void)" is
4914 special and matches everything, which can be used to suppress this
4915 warning. In a cast involving pointer to member types this warning
4916 warns whenever the type cast is changing the pointer to member
4917 type. This warning is enabled by -Wextra.
4918 -Wwrite-strings
4920 When compiling C, give string constants the type "const
4921 char[length]" so that copying the address of one into a non-"const"
4922 "char *" pointer produces a warning. These warnings help you find
4923 at compile time code that can try to write into a string constant,
4924 but only if you have been very careful about using "const" in
4925 declarations and prototypes. Otherwise, it is just a nuisance.
4926 This is why we did not make -Wall request these warnings.
4927 When compiling C++, warn about the deprecated conversion from
4929 string literals to "char *". This warning is enabled by default
4930 for C++ programs.
4931 -Wcatch-value
4933 -Wcatch-value=n (C++ and Objective-C++ only)
4934 Warn about catch handlers that do not catch via reference. With
4935 -Wcatch-value=1 (or -Wcatch-value for short) warn about polymorphic
4936 class types that are caught by value. With -Wcatch-value=2 warn
4937 about all class types that are caught by value. With
4938 -Wcatch-value=3 warn about all types that are not caught by
4939 reference. -Wcatch-value is enabled by -Wall.
4940 -Wclobbered
4942 Warn for variables that might be changed by "longjmp" or "vfork".
4943 This warning is also enabled by -Wextra.
4944 -Wconditionally-supported (C++ and Objective-C++ only)
4946 Warn for conditionally-supported (C++11 [intro.defs]) constructs.
4947 -Wconversion
4949 Warn for implicit conversions that may alter a value. This includes
4950 conversions between real and integer, like "abs (x)" when "x" is
4951 "double"; conversions between signed and unsigned, like "unsigned
4952 ui = -1"; and conversions to smaller types, like "sqrtf (M_PI)". Do
4953 not warn for explicit casts like "abs ((int) x)" and "ui =
4954 (unsigned) -1", or if the value is not changed by the conversion
4955 like in "abs (2.0)". Warnings about conversions between signed and
4956 unsigned integers can be disabled by using -Wno-sign-conversion.
4957 For C++, also warn for confusing overload resolution for user-
4959 defined conversions; and conversions that never use a type
4960 conversion operator: conversions to "void", the same type, a base
4961 class or a reference to them. Warnings about conversions between
4962 signed and unsigned integers are disabled by default in C++ unless
4963 -Wsign-conversion is explicitly enabled.
4964 -Wno-conversion-null (C++ and Objective-C++ only)
4966 Do not warn for conversions between "NULL" and non-pointer types.
4967 -Wconversion-null is enabled by default.
4968 -Wzero-as-null-pointer-constant (C++ and Objective-C++ only)
4970 Warn when a literal 0 is used as null pointer constant. This can
4971 be useful to facilitate the conversion to "nullptr" in C++11.
4972 -Wsubobject-linkage (C++ and Objective-C++ only)
4974 Warn if a class type has a base or a field whose type uses the
4975 anonymous namespace or depends on a type with no linkage. If a
4976 type A depends on a type B with no or internal linkage, defining it
4977 in multiple translation units would be an ODR violation because the
4978 meaning of B is different in each translation unit. If A only
4979 appears in a single translation unit, the best way to silence the
4980 warning is to give it internal linkage by putting it in an
4981 anonymous namespace as well. The compiler doesn't give this
4982 warning for types defined in the main .C file, as those are
4983 unlikely to have multiple definitions. -Wsubobject-linkage is
4984 enabled by default.
4985 -Wdangling-else
4987 Warn about constructions where there may be confusion to which "if"
4988 statement an "else" branch belongs. Here is an example of such a
4989 case:
4990 {
4992 if (a)
4993 if (b)
4994 foo ();
4995 else
4996 bar ();
4997 }
4998 In C/C++, every "else" branch belongs to the innermost possible
5000 "if" statement, which in this example is "if (b)". This is often
5001 not what the programmer expected, as illustrated in the above
5002 example by indentation the programmer chose. When there is the
5003 potential for this confusion, GCC issues a warning when this flag
5004 is specified. To eliminate the warning, add explicit braces around
5005 the innermost "if" statement so there is no way the "else" can
5006 belong to the enclosing "if". The resulting code looks like this:
5007 {
5009 if (a)
5010 {
5011 if (b)
5012 foo ();
5013 else
5014 bar ();
5015 }
5016 }
5017 This warning is enabled by -Wparentheses.
5019 -Wdate-time
5021 Warn when macros "__TIME__", "__DATE__" or "__TIMESTAMP__" are
5022 encountered as they might prevent bit-wise-identical reproducible
5023 compilations.
5024 -Wdelete-incomplete (C++ and Objective-C++ only)
5026 Warn when deleting a pointer to incomplete type, which may cause
5027 undefined behavior at runtime. This warning is enabled by default.
5028 -Wuseless-cast (C++ and Objective-C++ only)
5030 Warn when an expression is casted to its own type.
5031 -Wempty-body
5033 Warn if an empty body occurs in an "if", "else" or "do while"
5034 statement. This warning is also enabled by -Wextra.
5035 -Wenum-compare
5037 Warn about a comparison between values of different enumerated
5038 types. In C++ enumerated type mismatches in conditional
5039 expressions are also diagnosed and the warning is enabled by
5040 default. In C this warning is enabled by -Wall.
5041 -Wextra-semi (C++, Objective-C++ only)
5043 Warn about redundant semicolon after in-class function definition.
5044 -Wjump-misses-init (C, Objective-C only)
5046 Warn if a "goto" statement or a "switch" statement jumps forward
5047 across the initialization of a variable, or jumps backward to a
5048 label after the variable has been initialized. This only warns
5049 about variables that are initialized when they are declared. This
5050 warning is only supported for C and Objective-C; in C++ this sort
5051 of branch is an error in any case.
5052 -Wjump-misses-init is included in -Wc++-compat. It can be disabled
5054 with the -Wno-jump-misses-init option.
5055 -Wsign-compare
5057 Warn when a comparison between signed and unsigned values could
5058 produce an incorrect result when the signed value is converted to
5059 unsigned. In C++, this warning is also enabled by -Wall. In C, it
5060 is also enabled by -Wextra.
5061 -Wsign-conversion
5063 Warn for implicit conversions that may change the sign of an
5064 integer value, like assigning a signed integer expression to an
5065 unsigned integer variable. An explicit cast silences the warning.
5066 In C, this option is enabled also by -Wconversion.
5067 -Wfloat-conversion
5069 Warn for implicit conversions that reduce the precision of a real
5070 value. This includes conversions from real to integer, and from
5071 higher precision real to lower precision real values. This option
5072 is also enabled by -Wconversion.
5073 -Wno-scalar-storage-order
5075 Do not warn on suspicious constructs involving reverse scalar
5076 storage order.
5077 -Wsized-deallocation (C++ and Objective-C++ only)
5079 Warn about a definition of an unsized deallocation function
5080 void operator delete (void *) noexcept;
5082 void operator delete[] (void *) noexcept;
5083 without a definition of the corresponding sized deallocation
5085 function
5086 void operator delete (void *, std::size_t) noexcept;
5088 void operator delete[] (void *, std::size_t) noexcept;
5089 or vice versa. Enabled by -Wextra along with -fsized-deallocation.
5091 -Wsizeof-pointer-div
5093 Warn for suspicious divisions of two sizeof expressions that divide
5094 the pointer size by the element size, which is the usual way to
5095 compute the array size but won't work out correctly with pointers.
5096 This warning warns e.g. about "sizeof (ptr) / sizeof (ptr[0])" if
5097 "ptr" is not an array, but a pointer. This warning is enabled by
5098 -Wall.
5099 -Wsizeof-pointer-memaccess
5101 Warn for suspicious length parameters to certain string and memory
5102 built-in functions if the argument uses "sizeof". This warning
5103 triggers for example for "memset (ptr, 0, sizeof (ptr));" if "ptr"
5104 is not an array, but a pointer, and suggests a possible fix, or
5105 about "memcpy (&foo, ptr, sizeof (&foo));".
5106 -Wsizeof-pointer-memaccess also warns about calls to bounded string
5107 copy functions like "strncat" or "strncpy" that specify as the
5108 bound a "sizeof" expression of the source array. For example, in
5109 the following function the call to "strncat" specifies the size of
5110 the source string as the bound. That is almost certainly a mistake
5111 and so the call is diagnosed.
5112 void make_file (const char *name)
5114 {
5115 char path[PATH_MAX];
5116 strncpy (path, name, sizeof path - 1);
5117 strncat (path, ".text", sizeof ".text");
5118 ...
5119 }
5120 The -Wsizeof-pointer-memaccess option is enabled by -Wall.
5122 -Wsizeof-array-argument
5124 Warn when the "sizeof" operator is applied to a parameter that is
5125 declared as an array in a function definition. This warning is
5126 enabled by default for C and C++ programs.
5127 -Wmemset-elt-size
5129 Warn for suspicious calls to the "memset" built-in function, if the
5130 first argument references an array, and the third argument is a
5131 number equal to the number of elements, but not equal to the size
5132 of the array in memory. This indicates that the user has omitted a
5133 multiplication by the element size. This warning is enabled by
5134 -Wall.
5135 -Wmemset-transposed-args
5137 Warn for suspicious calls to the "memset" built-in function, if the
5138 second argument is not zero and the third argument is zero. This
5139 warns e.g.@ about "memset (buf, sizeof buf, 0)" where most probably
5140 "memset (buf, 0, sizeof buf)" was meant instead. The diagnostics
5141 is only emitted if the third argument is literal zero. If it is
5142 some expression that is folded to zero, a cast of zero to some
5143 type, etc., it is far less likely that the user has mistakenly
5144 exchanged the arguments and no warning is emitted. This warning is
5145 enabled by -Wall.
5146 -Waddress
5148 Warn about suspicious uses of memory addresses. These include using
5149 the address of a function in a conditional expression, such as
5150 "void func(void); if (func)", and comparisons against the memory
5151 address of a string literal, such as "if (x == "abc")". Such uses
5152 typically indicate a programmer error: the address of a function
5153 always evaluates to true, so their use in a conditional usually
5154 indicate that the programmer forgot the parentheses in a function
5155 call; and comparisons against string literals result in unspecified
5156 behavior and are not portable in C, so they usually indicate that
5157 the programmer intended to use "strcmp". This warning is enabled
5158 by -Wall.
5159 -Wlogical-op
5161 Warn about suspicious uses of logical operators in expressions.
5162 This includes using logical operators in contexts where a bit-wise
5163 operator is likely to be expected. Also warns when the operands of
5164 a logical operator are the same:
5165 extern int a;
5167 if (a < 0 && a < 0) { ... }
5168 -Wlogical-not-parentheses
5170 Warn about logical not used on the left hand side operand of a
5171 comparison. This option does not warn if the right operand is
5172 considered to be a boolean expression. Its purpose is to detect
5173 suspicious code like the following:
5174 int a;
5176 ...
5177 if (!a > 1) { ... }
5178 It is possible to suppress the warning by wrapping the LHS into
5180 parentheses:
5181 if ((!a) > 1) { ... }
5183 This warning is enabled by -Wall.
5185 -Waggregate-return
5187 Warn if any functions that return structures or unions are defined
5188 or called. (In languages where you can return an array, this also
5189 elicits a warning.)
5190 -Wno-aggressive-loop-optimizations
5192 Warn if in a loop with constant number of iterations the compiler
5193 detects undefined behavior in some statement during one or more of
5194 the iterations.
5195 -Wno-attributes
5197 Do not warn if an unexpected "__attribute__" is used, such as
5198 unrecognized attributes, function attributes applied to variables,
5199 etc. This does not stop errors for incorrect use of supported
5200 attributes.
5201 -Wno-builtin-declaration-mismatch
5203 Warn if a built-in function is declared with the wrong signature or
5204 as non-function. This warning is enabled by default.
5205 -Wno-builtin-macro-redefined
5207 Do not warn if certain built-in macros are redefined. This
5208 suppresses warnings for redefinition of "__TIMESTAMP__",
5209 "__TIME__", "__DATE__", "__FILE__", and "__BASE_FILE__".
5210 -Wstrict-prototypes (C and Objective-C only)
5212 Warn if a function is declared or defined without specifying the
5213 argument types. (An old-style function definition is permitted
5214 without a warning if preceded by a declaration that specifies the
5215 argument types.)
5216 -Wold-style-declaration (C and Objective-C only)
5218 Warn for obsolescent usages, according to the C Standard, in a
5219 declaration. For example, warn if storage-class specifiers like
5220 "static" are not the first things in a declaration. This warning
5221 is also enabled by -Wextra.
5222 -Wold-style-definition (C and Objective-C only)
5224 Warn if an old-style function definition is used. A warning is
5225 given even if there is a previous prototype.
5226 -Wmissing-parameter-type (C and Objective-C only)
5228 A function parameter is declared without a type specifier in
5229 K&R-style functions:
5230 void foo(bar) { }
5232 This warning is also enabled by -Wextra.
5234 -Wmissing-prototypes (C and Objective-C only)
5236 Warn if a global function is defined without a previous prototype
5237 declaration. This warning is issued even if the definition itself
5238 provides a prototype. Use this option to detect global functions
5239 that do not have a matching prototype declaration in a header file.
5240 This option is not valid for C++ because all function declarations
5241 provide prototypes and a non-matching declaration declares an
5242 overload rather than conflict with an earlier declaration. Use
5243 -Wmissing-declarations to detect missing declarations in C++.
5244 -Wmissing-declarations
5246 Warn if a global function is defined without a previous
5247 declaration. Do so even if the definition itself provides a
5248 prototype. Use this option to detect global functions that are not
5249 declared in header files. In C, no warnings are issued for
5250 functions with previous non-prototype declarations; use
5251 -Wmissing-prototypes to detect missing prototypes. In C++, no
5252 warnings are issued for function templates, or for inline
5253 functions, or for functions in anonymous namespaces.
5254 -Wmissing-field-initializers
5256 Warn if a structure's initializer has some fields missing. For
5257 example, the following code causes such a warning, because "x.h" is
5258 implicitly zero:
5259 struct s { int f, g, h; };
5261 struct s x = { 3, 4 };
5262 This option does not warn about designated initializers, so the
5264 following modification does not trigger a warning:
5265 struct s { int f, g, h; };
5267 struct s x = { .f = 3, .g = 4 };
5268 In C this option does not warn about the universal zero initializer
5270 { 0 }:
5271 struct s { int f, g, h; };
5273 struct s x = { 0 };
5274 Likewise, in C++ this option does not warn about the empty { }
5276 initializer, for example:
5277 struct s { int f, g, h; };
5279 s x = { };
5280 This warning is included in -Wextra. To get other -Wextra warnings
5282 without this one, use -Wextra -Wno-missing-field-initializers.
5283 -Wno-multichar
5285 Do not warn if a multicharacter constant ('FOOF') is used. Usually
5286 they indicate a typo in the user's code, as they have
5287 implementation-defined values, and should not be used in portable
5288 code.
5289 -Wnormalized=[none|id|nfc|nfkc]
5291 In ISO C and ISO C++, two identifiers are different if they are
5292 different sequences of characters. However, sometimes when
5293 characters outside the basic ASCII character set are used, you can
5294 have two different character sequences that look the same. To
5295 avoid confusion, the ISO 10646 standard sets out some normalization
5296 rules which when applied ensure that two sequences that look the
5297 same are turned into the same sequence. GCC can warn you if you
5298 are using identifiers that have not been normalized; this option
5299 controls that warning.
5300 There are four levels of warning supported by GCC. The default is
5302 -Wnormalized=nfc, which warns about any identifier that is not in
5303 the ISO 10646 "C" normalized form, NFC. NFC is the recommended
5304 form for most uses. It is equivalent to -Wnormalized.
5305 Unfortunately, there are some characters allowed in identifiers by
5307 ISO C and ISO C++ that, when turned into NFC, are not allowed in
5308 identifiers. That is, there's no way to use these symbols in
5309 portable ISO C or C++ and have all your identifiers in NFC.
5310 -Wnormalized=id suppresses the warning for these characters. It is
5311 hoped that future versions of the standards involved will correct
5312 this, which is why this option is not the default.
5313 You can switch the warning off for all characters by writing
5315 -Wnormalized=none or -Wno-normalized. You should only do this if
5316 you are using some other normalization scheme (like "D"), because
5317 otherwise you can easily create bugs that are literally impossible
5318 to see.
5319 Some characters in ISO 10646 have distinct meanings but look
5321 identical in some fonts or display methodologies, especially once
5322 formatting has been applied. For instance "\u207F", "SUPERSCRIPT
5323 LATIN SMALL LETTER N", displays just like a regular "n" that has
5324 been placed in a superscript. ISO 10646 defines the NFKC
5325 normalization scheme to convert all these into a standard form as
5326 well, and GCC warns if your code is not in NFKC if you use
5327 -Wnormalized=nfkc. This warning is comparable to warning about
5328 every identifier that contains the letter O because it might be
5329 confused with the digit 0, and so is not the default, but may be
5330 useful as a local coding convention if the programming environment
5331 cannot be fixed to display these characters distinctly.
5332 -Wno-deprecated
5334 Do not warn about usage of deprecated features.
5335 -Wno-deprecated-declarations
5337 Do not warn about uses of functions, variables, and types marked as
5338 deprecated by using the "deprecated" attribute.
5339 -Wno-overflow
5341 Do not warn about compile-time overflow in constant expressions.
5342 -Wno-odr
5344 Warn about One Definition Rule violations during link-time
5345 optimization. Requires -flto-odr-type-merging to be enabled.
5346 Enabled by default.
5347 -Wopenmp-simd
5349 Warn if the vectorizer cost model overrides the OpenMP simd
5350 directive set by user. The -fsimd-cost-model=unlimited option can
5351 be used to relax the cost model.
5352 -Woverride-init (C and Objective-C only)
5354 Warn if an initialized field without side effects is overridden
5355 when using designated initializers.
5356 This warning is included in -Wextra. To get other -Wextra warnings
5358 without this one, use -Wextra -Wno-override-init.
5359 -Woverride-init-side-effects (C and Objective-C only)
5361 Warn if an initialized field with side effects is overridden when
5362 using designated initializers. This warning is enabled by default.
5363 -Wpacked
5365 Warn if a structure is given the packed attribute, but the packed
5366 attribute has no effect on the layout or size of the structure.
5367 Such structures may be mis-aligned for little benefit. For
5368 instance, in this code, the variable "f.x" in "struct bar" is
5369 misaligned even though "struct bar" does not itself have the packed
5370 attribute:
5371 struct foo {
5373 int x;
5374 char a, b, c, d;
5375 } __attribute__((packed));
5376 struct bar {
5377 char z;
5378 struct foo f;
5379 };
5380 -Wpacked-bitfield-compat
5382 The 4.1, 4.2 and 4.3 series of GCC ignore the "packed" attribute on
5383 bit-fields of type "char". This has been fixed in GCC 4.4 but the
5384 change can lead to differences in the structure layout. GCC
5385 informs you when the offset of such a field has changed in GCC 4.4.
5386 For example there is no longer a 4-bit padding between field "a"
5387 and "b" in this structure:
5388 struct foo
5390 {
5391 char a:4;
5392 char b:8;
5393 } __attribute__ ((packed));
5394 This warning is enabled by default. Use
5396 -Wno-packed-bitfield-compat to disable this warning.
5397 -Wpacked-not-aligned (C, C++, Objective-C and Objective-C++ only)
5399 Warn if a structure field with explicitly specified alignment in a
5400 packed struct or union is misaligned. For example, a warning will
5401 be issued on "struct S", like, "warning: alignment 1 of 'struct S'
5402 is less than 8", in this code:
5403 struct __attribute__ ((aligned (8))) S8 { char a[8]; };
5405 struct __attribute__ ((packed)) S {
5406 struct S8 s8;
5407 };
5408 This warning is enabled by -Wall.
5410 -Wpadded
5412 Warn if padding is included in a structure, either to align an
5413 element of the structure or to align the whole structure.
5414 Sometimes when this happens it is possible to rearrange the fields
5415 of the structure to reduce the padding and so make the structure
5416 smaller.
5417 -Wredundant-decls
5419 Warn if anything is declared more than once in the same scope, even
5420 in cases where multiple declaration is valid and changes nothing.
5421 -Wno-restrict
5423 Warn when an object referenced by a "restrict"-qualified parameter
5424 (or, in C++, a "__restrict"-qualified parameter) is aliased by
5425 another argument, or when copies between such objects overlap. For
5426 example, the call to the "strcpy" function below attempts to
5427 truncate the string by replacing its initial characters with the
5428 last four. However, because the call writes the terminating NUL
5429 into "a[4]", the copies overlap and the call is diagnosed.
5430 void foo (void)
5432 {
5433 char a[] = "abcd1234";
5434 strcpy (a, a + 4);
5435 ...
5436 }
5437 The -Wrestrict option detects some instances of simple overlap even
5439 without optimization but works best at -O2 and above. It is
5440 included in -Wall.
5441 -Wnested-externs (C and Objective-C only)
5443 Warn if an "extern" declaration is encountered within a function.
5444 -Wno-inherited-variadic-ctor
5446 Suppress warnings about use of C++11 inheriting constructors when
5447 the base class inherited from has a C variadic constructor; the
5448 warning is on by default because the ellipsis is not inherited.
5449 -Winline
5451 Warn if a function that is declared as inline cannot be inlined.
5452 Even with this option, the compiler does not warn about failures to
5453 inline functions declared in system headers.
5454 The compiler uses a variety of heuristics to determine whether or
5456 not to inline a function. For example, the compiler takes into
5457 account the size of the function being inlined and the amount of
5458 inlining that has already been done in the current function.
5459 Therefore, seemingly insignificant changes in the source program
5460 can cause the warnings produced by -Winline to appear or disappear.
5461 -Wno-invalid-offsetof (C++ and Objective-C++ only)
5463 Suppress warnings from applying the "offsetof" macro to a non-POD
5464 type. According to the 2014 ISO C++ standard, applying "offsetof"
5465 to a non-standard-layout type is undefined. In existing C++
5466 implementations, however, "offsetof" typically gives meaningful
5467 results. This flag is for users who are aware that they are
5468 writing nonportable code and who have deliberately chosen to ignore
5469 the warning about it.
5470 The restrictions on "offsetof" may be relaxed in a future version
5472 of the C++ standard.
5473 -Wint-in-bool-context
5475 Warn for suspicious use of integer values where boolean values are
5476 expected, such as conditional expressions (?:) using non-boolean
5477 integer constants in boolean context, like "if (a <= b ? 2 : 3)".
5478 Or left shifting of signed integers in boolean context, like "for
5479 (a = 0; 1 << a; a++);". Likewise for all kinds of multiplications
5480 regardless of the data type. This warning is enabled by -Wall.
5481 -Wno-int-to-pointer-cast
5483 Suppress warnings from casts to pointer type of an integer of a
5484 different size. In C++, casting to a pointer type of smaller size
5485 is an error. Wint-to-pointer-cast is enabled by default.
5486 -Wno-pointer-to-int-cast (C and Objective-C only)
5488 Suppress warnings from casts from a pointer to an integer type of a
5489 different size.
5490 -Winvalid-pch
5492 Warn if a precompiled header is found in the search path but cannot
5493 be used.
5494 -Wlong-long
5496 Warn if "long long" type is used. This is enabled by either
5497 -Wpedantic or -Wtraditional in ISO C90 and C++98 modes. To inhibit
5498 the warning messages, use -Wno-long-long.
5499 -Wvariadic-macros
5501 Warn if variadic macros are used in ISO C90 mode, or if the GNU
5502 alternate syntax is used in ISO C99 mode. This is enabled by
5503 either -Wpedantic or -Wtraditional. To inhibit the warning
5504 messages, use -Wno-variadic-macros.
5505 -Wvarargs
5507 Warn upon questionable usage of the macros used to handle variable
5508 arguments like "va_start". This is default. To inhibit the
5509 warning messages, use -Wno-varargs.
5510 -Wvector-operation-performance
5512 Warn if vector operation is not implemented via SIMD capabilities
5513 of the architecture. Mainly useful for the performance tuning.
5514 Vector operation can be implemented "piecewise", which means that
5515 the scalar operation is performed on every vector element; "in
5516 parallel", which means that the vector operation is implemented
5517 using scalars of wider type, which normally is more performance
5518 efficient; and "as a single scalar", which means that vector fits
5519 into a scalar type.
5520 -Wno-virtual-move-assign
5522 Suppress warnings about inheriting from a virtual base with a non-
5523 trivial C++11 move assignment operator. This is dangerous because
5524 if the virtual base is reachable along more than one path, it is
5525 moved multiple times, which can mean both objects end up in the
5526 moved-from state. If the move assignment operator is written to
5527 avoid moving from a moved-from object, this warning can be
5528 disabled.
5529 -Wvla
5531 Warn if a variable-length array is used in the code. -Wno-vla
5532 prevents the -Wpedantic warning of the variable-length array.
5533 -Wvla-larger-than=n
5535 If this option is used, the compiler will warn on uses of variable-
5536 length arrays where the size is either unbounded, or bounded by an
5537 argument that can be larger than n bytes. This is similar to how
5538 -Walloca-larger-than=n works, but with variable-length arrays.
5539 Note that GCC may optimize small variable-length arrays of a known
5541 value into plain arrays, so this warning may not get triggered for
5542 such arrays.
5543 This warning is not enabled by -Wall, and is only active when
5545 -ftree-vrp is active (default for -O2 and above).
5546 See also -Walloca-larger-than=n.
5548 -Wvolatile-register-var
5550 Warn if a register variable is declared volatile. The volatile
5551 modifier does not inhibit all optimizations that may eliminate
5552 reads and/or writes to register variables. This warning is enabled
5553 by -Wall.
5554 -Wdisabled-optimization
5556 Warn if a requested optimization pass is disabled. This warning
5557 does not generally indicate that there is anything wrong with your
5558 code; it merely indicates that GCC's optimizers are unable to
5559 handle the code effectively. Often, the problem is that your code
5560 is too big or too complex; GCC refuses to optimize programs when
5561 the optimization itself is likely to take inordinate amounts of
5562 time.
5563 -Wpointer-sign (C and Objective-C only)
5565 Warn for pointer argument passing or assignment with different
5566 signedness. This option is only supported for C and Objective-C.
5567 It is implied by -Wall and by -Wpedantic, which can be disabled
5568 with -Wno-pointer-sign.
5569 -Wstack-protector
5571 This option is only active when -fstack-protector is active. It
5572 warns about functions that are not protected against stack
5573 smashing.
5574 -Woverlength-strings
5576 Warn about string constants that are longer than the "minimum
5577 maximum" length specified in the C standard. Modern compilers
5578 generally allow string constants that are much longer than the
5579 standard's minimum limit, but very portable programs should avoid
5580 using longer strings.
5581 The limit applies after string constant concatenation, and does not
5583 count the trailing NUL. In C90, the limit was 509 characters; in
5584 C99, it was raised to 4095. C++98 does not specify a normative
5585 minimum maximum, so we do not diagnose overlength strings in C++.
5586 This option is implied by -Wpedantic, and can be disabled with
5588 -Wno-overlength-strings.
5589 -Wunsuffixed-float-constants (C and Objective-C only)
5591 Issue a warning for any floating constant that does not have a
5592 suffix. When used together with -Wsystem-headers it warns about
5593 such constants in system header files. This can be useful when
5594 preparing code to use with the "FLOAT_CONST_DECIMAL64" pragma from
5595 the decimal floating-point extension to C99.
5596 -Wno-designated-init (C and Objective-C only)
5598 Suppress warnings when a positional initializer is used to
5599 initialize a structure that has been marked with the
5600 "designated_init" attribute.
5601 -Whsa
5603 Issue a warning when HSAIL cannot be emitted for the compiled
5604 function or OpenMP construct.
5605 Options for Debugging Your Program
5607 To tell GCC to emit extra information for use by a debugger, in almost
5608 all cases you need only to add -g to your other options.
5609 GCC allows you to use -g with -O. The shortcuts taken by optimized
5611 code may occasionally be surprising: some variables you declared may
5612 not exist at all; flow of control may briefly move where you did not
5613 expect it; some statements may not be executed because they compute
5614 constant results or their values are already at hand; some statements
5615 may execute in different places because they have been moved out of
5616 loops. Nevertheless it is possible to debug optimized output. This
5617 makes it reasonable to use the optimizer for programs that might have
5618 bugs.
5619 If you are not using some other optimization option, consider using -Og
5621 with -g. With no -O option at all, some compiler passes that collect
5622 information useful for debugging do not run at all, so that -Og may
5623 result in a better debugging experience.
5624 -g Produce debugging information in the operating system's native
5626 format (stabs, COFF, XCOFF, or DWARF). GDB can work with this
5627 debugging information.
5628 On most systems that use stabs format, -g enables use of extra
5630 debugging information that only GDB can use; this extra information
5631 makes debugging work better in GDB but probably makes other
5632 debuggers crash or refuse to read the program. If you want to
5633 control for certain whether to generate the extra information, use
5634 -gstabs+, -gstabs, -gxcoff+, -gxcoff, or -gvms (see below).
5635 -ggdb
5637 Produce debugging information for use by GDB. This means to use
5638 the most expressive format available (DWARF, stabs, or the native
5639 format if neither of those are supported), including GDB extensions
5640 if at all possible.
5641 -gdwarf
5643 -gdwarf-version
5644 Produce debugging information in DWARF format (if that is
5645 supported). The value of version may be either 2, 3, 4 or 5; the
5646 default version for most targets is 4. DWARF Version 5 is only
5647 experimental.
5648 Note that with DWARF Version 2, some ports require and always use
5650 some non-conflicting DWARF 3 extensions in the unwind tables.
5651 Version 4 may require GDB 7.0 and -fvar-tracking-assignments for
5653 maximum benefit.
5654 GCC no longer supports DWARF Version 1, which is substantially
5656 different than Version 2 and later. For historical reasons, some
5657 other DWARF-related options such as -fno-dwarf2-cfi-asm) retain a
5658 reference to DWARF Version 2 in their names, but apply to all
5659 currently-supported versions of DWARF.
5660 -gstabs
5662 Produce debugging information in stabs format (if that is
5663 supported), without GDB extensions. This is the format used by DBX
5664 on most BSD systems. On MIPS, Alpha and System V Release 4 systems
5665 this option produces stabs debugging output that is not understood
5666 by DBX. On System V Release 4 systems this option requires the GNU
5667 assembler.
5668 -gstabs+
5670 Produce debugging information in stabs format (if that is
5671 supported), using GNU extensions understood only by the GNU
5672 debugger (GDB). The use of these extensions is likely to make
5673 other debuggers crash or refuse to read the program.
5674 -gxcoff
5676 Produce debugging information in XCOFF format (if that is
5677 supported). This is the format used by the DBX debugger on IBM
5678 RS/6000 systems.
5679 -gxcoff+
5681 Produce debugging information in XCOFF format (if that is
5682 supported), using GNU extensions understood only by the GNU
5683 debugger (GDB). The use of these extensions is likely to make
5684 other debuggers crash or refuse to read the program, and may cause
5685 assemblers other than the GNU assembler (GAS) to fail with an
5686 error.
5687 -gvms
5689 Produce debugging information in Alpha/VMS debug format (if that is
5690 supported). This is the format used by DEBUG on Alpha/VMS systems.
5691 -glevel
5693 -ggdblevel
5694 -gstabslevel
5695 -gxcofflevel
5696 -gvmslevel
5697 Request debugging information and also use level to specify how
5698 much information. The default level is 2.
5699 Level 0 produces no debug information at all. Thus, -g0 negates
5701 -g.
5702 Level 1 produces minimal information, enough for making backtraces
5704 in parts of the program that you don't plan to debug. This
5705 includes descriptions of functions and external variables, and line
5706 number tables, but no information about local variables.
5707 Level 3 includes extra information, such as all the macro
5709 definitions present in the program. Some debuggers support macro
5710 expansion when you use -g3.
5711 -gdwarf does not accept a concatenated debug level, to avoid
5713 confusion with -gdwarf-level. Instead use an additional -glevel
5714 option to change the debug level for DWARF.
5715 -feliminate-unused-debug-symbols
5717 Produce debugging information in stabs format (if that is
5718 supported), for only symbols that are actually used.
5719 -femit-class-debug-always
5721 Instead of emitting debugging information for a C++ class in only
5722 one object file, emit it in all object files using the class. This
5723 option should be used only with debuggers that are unable to handle
5724 the way GCC normally emits debugging information for classes
5725 because using this option increases the size of debugging
5726 information by as much as a factor of two.
5727 -fno-merge-debug-strings
5729 Direct the linker to not merge together strings in the debugging
5730 information that are identical in different object files. Merging
5731 is not supported by all assemblers or linkers. Merging decreases
5732 the size of the debug information in the output file at the cost of
5733 increasing link processing time. Merging is enabled by default.
5734 -fdebug-prefix-map=old=new
5736 When compiling files residing in directory old, record debugging
5737 information describing them as if the files resided in directory
5738 new instead. This can be used to replace a build-time path with an
5739 install-time path in the debug info. It can also be used to change
5740 an absolute path to a relative path by using . for new. This can
5741 give more reproducible builds, which are location independent, but
5742 may require an extra command to tell GDB where to find the source
5743 files. See also -ffile-prefix-map.
5744 -fvar-tracking
5746 Run variable tracking pass. It computes where variables are stored
5747 at each position in code. Better debugging information is then
5748 generated (if the debugging information format supports this
5749 information).
5750 It is enabled by default when compiling with optimization (-Os, -O,
5752 -O2, ...), debugging information (-g) and the debug info format
5753 supports it.
5754 -fvar-tracking-assignments
5756 Annotate assignments to user variables early in the compilation and
5757 attempt to carry the annotations over throughout the compilation
5758 all the way to the end, in an attempt to improve debug information
5759 while optimizing. Use of -gdwarf-4 is recommended along with it.
5760 It can be enabled even if var-tracking is disabled, in which case
5762 annotations are created and maintained, but discarded at the end.
5763 By default, this flag is enabled together with -fvar-tracking,
5764 except when selective scheduling is enabled.
5765 -gsplit-dwarf
5767 Separate as much DWARF debugging information as possible into a
5768 separate output file with the extension .dwo. This option allows
5769 the build system to avoid linking files with debug information. To
5770 be useful, this option requires a debugger capable of reading .dwo
5771 files.
5772 -gpubnames
5774 Generate DWARF ".debug_pubnames" and ".debug_pubtypes" sections.
5775 -ggnu-pubnames
5777 Generate ".debug_pubnames" and ".debug_pubtypes" sections in a
5778 format suitable for conversion into a GDB index. This option is
5779 only useful with a linker that can produce GDB index version 7.
5780 -fdebug-types-section
5782 When using DWARF Version 4 or higher, type DIEs can be put into
5783 their own ".debug_types" section instead of making them part of the
5784 ".debug_info" section. It is more efficient to put them in a
5785 separate comdat sections since the linker can then remove
5786 duplicates. But not all DWARF consumers support ".debug_types"
5787 sections yet and on some objects ".debug_types" produces larger
5788 instead of smaller debugging information.
5789 -grecord-gcc-switches
5791 -gno-record-gcc-switches
5792 This switch causes the command-line options used to invoke the
5793 compiler that may affect code generation to be appended to the
5794 DW_AT_producer attribute in DWARF debugging information. The
5795 options are concatenated with spaces separating them from each
5796 other and from the compiler version. It is enabled by default.
5797 See also -frecord-gcc-switches for another way of storing compiler
5798 options into the object file.
5799 -gstrict-dwarf
5801 Disallow using extensions of later DWARF standard version than
5802 selected with -gdwarf-version. On most targets using non-
5803 conflicting DWARF extensions from later standard versions is
5804 allowed.
5805 -gno-strict-dwarf
5807 Allow using extensions of later DWARF standard version than
5808 selected with -gdwarf-version.
5809 -gas-loc-support
5811 Inform the compiler that the assembler supports ".loc" directives.
5812 It may then use them for the assembler to generate DWARF2+ line
5813 number tables.
5814 This is generally desirable, because assembler-generated line-
5816 number tables are a lot more compact than those the compiler can
5817 generate itself.
5818 This option will be enabled by default if, at GCC configure time,
5820 the assembler was found to support such directives.
5821 -gno-as-loc-support
5823 Force GCC to generate DWARF2+ line number tables internally, if
5824 DWARF2+ line number tables are to be generated.
5825 gas-locview-support
5827 Inform the compiler that the assembler supports "view" assignment
5828 and reset assertion checking in ".loc" directives.
5829 This option will be enabled by default if, at GCC configure time,
5831 the assembler was found to support them.
5832 gno-as-locview-support
5834 Force GCC to assign view numbers internally, if
5835 -gvariable-location-views are explicitly requested.
5836 -gcolumn-info
5838 -gno-column-info
5839 Emit location column information into DWARF debugging information,
5840 rather than just file and line. This option is enabled by default.
5841 -gstatement-frontiers
5843 -gno-statement-frontiers
5844 This option causes GCC to create markers in the internal
5845 representation at the beginning of statements, and to keep them
5846 roughly in place throughout compilation, using them to guide the
5847 output of "is_stmt" markers in the line number table. This is
5848 enabled by default when compiling with optimization (-Os, -O, -O2,
5849 ...), and outputting DWARF 2 debug information at the normal level.
5850 -gvariable-location-views
5852 -gvariable-location-views=incompat5
5853 -gno-variable-location-views
5854 Augment variable location lists with progressive view numbers
5855 implied from the line number table. This enables debug information
5856 consumers to inspect state at certain points of the program, even
5857 if no instructions associated with the corresponding source
5858 locations are present at that point. If the assembler lacks
5859 support for view numbers in line number tables, this will cause the
5860 compiler to emit the line number table, which generally makes them
5861 somewhat less compact. The augmented line number tables and
5862 location lists are fully backward-compatible, so they can be
5863 consumed by debug information consumers that are not aware of these
5864 augmentations, but they won't derive any benefit from them either.
5865 This is enabled by default when outputting DWARF 2 debug
5867 information at the normal level, as long as there is assembler
5868 support, -fvar-tracking-assignments is enabled and -gstrict-dwarf
5869 is not. When assembler support is not available, this may still be
5870 enabled, but it will force GCC to output internal line number
5871 tables, and if -ginternal-reset-location-views is not enabled, that
5872 will most certainly lead to silently mismatching location views.
5873 There is a proposed representation for view numbers that is not
5875 backward compatible with the location list format introduced in
5876 DWARF 5, that can be enabled with
5877 -gvariable-location-views=incompat5. This option may be removed in
5878 the future, is only provided as a reference implementation of the
5879 proposed representation. Debug information consumers are not
5880 expected to support this extended format, and they would be
5881 rendered unable to decode location lists using it.
5882 -ginternal-reset-location-views
5884 -gnointernal-reset-location-views
5885 Attempt to determine location views that can be omitted from
5886 location view lists. This requires the compiler to have very
5887 accurate insn length estimates, which isn't always the case, and it
5888 may cause incorrect view lists to be generated silently when using
5889 an assembler that does not support location view lists. The GNU
5890 assembler will flag any such error as a "view number mismatch".
5891 This is only enabled on ports that define a reliable estimation
5892 function.
5893 -ginline-points
5895 -gno-inline-points
5896 Generate extended debug information for inlined functions.
5897 Location view tracking markers are inserted at inlined entry
5898 points, so that address and view numbers can be computed and output
5899 in debug information. This can be enabled independently of
5900 location views, in which case the view numbers won't be output, but
5901 it can only be enabled along with statement frontiers, and it is
5902 only enabled by default if location views are enabled.
5903 -gz[=type]
5905 Produce compressed debug sections in DWARF format, if that is
5906 supported. If type is not given, the default type depends on the
5907 capabilities of the assembler and linker used. type may be one of
5908 none (don't compress debug sections), zlib (use zlib compression in
5909 ELF gABI format), or zlib-gnu (use zlib compression in traditional
5910 GNU format). If the linker doesn't support writing compressed
5911 debug sections, the option is rejected. Otherwise, if the
5912 assembler does not support them, -gz is silently ignored when
5913 producing object files.
5914 -femit-struct-debug-baseonly
5916 Emit debug information for struct-like types only when the base
5917 name of the compilation source file matches the base name of file
5918 in which the struct is defined.
5919 This option substantially reduces the size of debugging
5921 information, but at significant potential loss in type information
5922 to the debugger. See -femit-struct-debug-reduced for a less
5923 aggressive option. See -femit-struct-debug-detailed for more
5924 detailed control.
5925 This option works only with DWARF debug output.
5927 -femit-struct-debug-reduced
5929 Emit debug information for struct-like types only when the base
5930 name of the compilation source file matches the base name of file
5931 in which the type is defined, unless the struct is a template or
5932 defined in a system header.
5933 This option significantly reduces the size of debugging
5935 information, with some potential loss in type information to the
5936 debugger. See -femit-struct-debug-baseonly for a more aggressive
5937 option. See -femit-struct-debug-detailed for more detailed
5938 control.
5939 This option works only with DWARF debug output.
5941 -femit-struct-debug-detailed[=spec-list]
5943 Specify the struct-like types for which the compiler generates
5944 debug information. The intent is to reduce duplicate struct debug
5945 information between different object files within the same program.
5946 This option is a detailed version of -femit-struct-debug-reduced
5948 and -femit-struct-debug-baseonly, which serves for most needs.
5949 A specification has the
5951 syntax[dir:|ind:][ord:|gen:](any|sys|base|none)
5952 The optional first word limits the specification to structs that
5954 are used directly (dir:) or used indirectly (ind:). A struct type
5955 is used directly when it is the type of a variable, member.
5956 Indirect uses arise through pointers to structs. That is, when use
5957 of an incomplete struct is valid, the use is indirect. An example
5958 is struct one direct; struct two * indirect;.
5959 The optional second word limits the specification to ordinary
5961 structs (ord:) or generic structs (gen:). Generic structs are a
5962 bit complicated to explain. For C++, these are non-explicit
5963 specializations of template classes, or non-template classes within
5964 the above. Other programming languages have generics, but
5965 -femit-struct-debug-detailed does not yet implement them.
5966 The third word specifies the source files for those structs for
5968 which the compiler should emit debug information. The values none
5969 and any have the normal meaning. The value base means that the
5970 base of name of the file in which the type declaration appears must
5971 match the base of the name of the main compilation file. In
5972 practice, this means that when compiling foo.c, debug information
5973 is generated for types declared in that file and foo.h, but not
5974 other header files. The value sys means those types satisfying
5975 base or declared in system or compiler headers.
5976 You may need to experiment to determine the best settings for your
5978 application.
5979 The default is -femit-struct-debug-detailed=all.
5981 This option works only with DWARF debug output.
5983 -fno-dwarf2-cfi-asm
5985 Emit DWARF unwind info as compiler generated ".eh_frame" section
5986 instead of using GAS ".cfi_*" directives.
5987 -fno-eliminate-unused-debug-types
5989 Normally, when producing DWARF output, GCC avoids producing debug
5990 symbol output for types that are nowhere used in the source file
5991 being compiled. Sometimes it is useful to have GCC emit debugging
5992 information for all types declared in a compilation unit,
5993 regardless of whether or not they are actually used in that
5994 compilation unit, for example if, in the debugger, you want to cast
5995 a value to a type that is not actually used in your program (but is
5996 declared). More often, however, this results in a significant
5997 amount of wasted space.
5998 Options That Control Optimization
6000 These options control various sorts of optimizations.
6001 Without any optimization option, the compiler's goal is to reduce the
6003 cost of compilation and to make debugging produce the expected results.
6004 Statements are independent: if you stop the program with a breakpoint
6005 between statements, you can then assign a new value to any variable or
6006 change the program counter to any other statement in the function and
6007 get exactly the results you expect from the source code.
6008 Turning on optimization flags makes the compiler attempt to improve the
6010 performance and/or code size at the expense of compilation time and
6011 possibly the ability to debug the program.
6012 The compiler performs optimization based on the knowledge it has of the
6014 program. Compiling multiple files at once to a single output file mode
6015 allows the compiler to use information gained from all of the files
6016 when compiling each of them.
6017 Not all optimizations are controlled directly by a flag. Only
6019 optimizations that have a flag are listed in this section.
6020 Most optimizations are only enabled if an -O level is set on the
6022 command line. Otherwise they are disabled, even if individual
6023 optimization flags are specified.
6024 Depending on the target and how GCC was configured, a slightly
6026 different set of optimizations may be enabled at each -O level than
6027 those listed here. You can invoke GCC with -Q --help=optimizers to
6028 find out the exact set of optimizations that are enabled at each level.
6029 -O
6031 -O1 Optimize. Optimizing compilation takes somewhat more time, and a
6032 lot more memory for a large function.
6033 With -O, the compiler tries to reduce code size and execution time,
6035 without performing any optimizations that take a great deal of
6036 compilation time.
6037 -O turns on the following optimization flags:
6039 -fauto-inc-dec -fbranch-count-reg -fcombine-stack-adjustments
6041 -fcompare-elim -fcprop-registers -fdce -fdefer-pop -fdelayed-branch
6042 -fdse -fforward-propagate -fguess-branch-probability
6043 -fif-conversion2 -fif-conversion -finline-functions-called-once
6044 -fipa-pure-const -fipa-profile -fipa-reference -fmerge-constants
6045 -fmove-loop-invariants -fomit-frame-pointer -freorder-blocks
6046 -fshrink-wrap -fshrink-wrap-separate -fsplit-wide-types
6047 -fssa-backprop -fssa-phiopt -ftree-bit-ccp -ftree-ccp -ftree-ch
6048 -ftree-coalesce-vars -ftree-copy-prop -ftree-dce
6049 -ftree-dominator-opts -ftree-dse -ftree-forwprop -ftree-fre
6050 -ftree-phiprop -ftree-sink -ftree-slsr -ftree-sra -ftree-pta
6051 -ftree-ter -funit-at-a-time
6052 -O2 Optimize even more. GCC performs nearly all supported
6054 optimizations that do not involve a space-speed tradeoff. As
6055 compared to -O, this option increases both compilation time and the
6056 performance of the generated code.
6057 -O2 turns on all optimization flags specified by -O. It also turns
6059 on the following optimization flags: -fthread-jumps
6060 -falign-functions -falign-jumps -falign-loops -falign-labels
6061 -fcaller-saves -fcrossjumping -fcse-follow-jumps -fcse-skip-blocks
6062 -fdelete-null-pointer-checks -fdevirtualize
6063 -fdevirtualize-speculatively -fexpensive-optimizations -fgcse
6064 -fgcse-lm -fhoist-adjacent-loads -finline-small-functions
6065 -findirect-inlining -fipa-cp -fipa-bit-cp -fipa-vrp -fipa-sra
6066 -fipa-icf -fisolate-erroneous-paths-dereference -flra-remat
6067 -foptimize-sibling-calls -foptimize-strlen -fpartial-inlining
6068 -fpeephole2 -freorder-blocks-algorithm=stc
6069 -freorder-blocks-and-partition -freorder-functions
6070 -frerun-cse-after-loop -fsched-interblock -fsched-spec
6071 -fschedule-insns -fschedule-insns2 -fstore-merging
6072 -fstrict-aliasing -ftree-builtin-call-dce -ftree-switch-conversion
6073 -ftree-tail-merge -fcode-hoisting -ftree-pre -ftree-vrp -fipa-ra
6074 Please note the warning under -fgcse about invoking -O2 on programs
6076 that use computed gotos.
6077 -O3 Optimize yet more. -O3 turns on all optimizations specified by -O2
6079 and also turns on the following optimization flags:
6080 -finline-functions -funswitch-loops -fpredictive-commoning
6081 -fgcse-after-reload -ftree-loop-vectorize -ftree-loop-distribution
6082 -ftree-loop-distribute-patterns -floop-interchange
6083 -floop-unroll-and-jam -fsplit-paths -ftree-slp-vectorize
6084 -fvect-cost-model -ftree-partial-pre -fpeel-loops -fipa-cp-clone
6085 -O0 Reduce compilation time and make debugging produce the expected
6087 results. This is the default.
6088 -Os Optimize for size. -Os enables all -O2 optimizations that do not
6090 typically increase code size. It also performs further
6091 optimizations designed to reduce code size.
6092 -Os disables the following optimization flags: -falign-functions
6094 -falign-jumps -falign-loops -falign-labels -freorder-blocks
6095 -freorder-blocks-algorithm=stc -freorder-blocks-and-partition
6096 -fprefetch-loop-arrays
6097 -Ofast
6099 Disregard strict standards compliance. -Ofast enables all -O3
6100 optimizations. It also enables optimizations that are not valid
6101 for all standard-compliant programs. It turns on -ffast-math and
6102 the Fortran-specific -fstack-arrays, unless -fmax-stack-var-size is
6103 specified, and -fno-protect-parens.
6104 -Og Optimize debugging experience. -Og enables optimizations that do
6106 not interfere with debugging. It should be the optimization level
6107 of choice for the standard edit-compile-debug cycle, offering a
6108 reasonable level of optimization while maintaining fast compilation
6109 and a good debugging experience.
6110 If you use multiple -O options, with or without level numbers, the last
6112 such option is the one that is effective.
6113 Options of the form -fflag specify machine-independent flags. Most
6115 flags have both positive and negative forms; the negative form of -ffoo
6116 is -fno-foo. In the table below, only one of the forms is listed---the
6117 one you typically use. You can figure out the other form by either
6118 removing no- or adding it.
6119 The following options control specific optimizations. They are either
6121 activated by -O options or are related to ones that are. You can use
6122 the following flags in the rare cases when "fine-tuning" of
6123 optimizations to be performed is desired.
6124 -fno-defer-pop
6126 Always pop the arguments to each function call as soon as that
6127 function returns. For machines that must pop arguments after a
6128 function call, the compiler normally lets arguments accumulate on
6129 the stack for several function calls and pops them all at once.
6130 Disabled at levels -O, -O2, -O3, -Os.
6132 -fforward-propagate
6134 Perform a forward propagation pass on RTL. The pass tries to
6135 combine two instructions and checks if the result can be
6136 simplified. If loop unrolling is active, two passes are performed
6137 and the second is scheduled after loop unrolling.
6138 This option is enabled by default at optimization levels -O, -O2,
6140 -O3, -Os.
6141 -ffp-contract=style
6143 -ffp-contract=off disables floating-point expression contraction.
6144 -ffp-contract=fast enables floating-point expression contraction
6145 such as forming of fused multiply-add operations if the target has
6146 native support for them. -ffp-contract=on enables floating-point
6147 expression contraction if allowed by the language standard. This
6148 is currently not implemented and treated equal to
6149 -ffp-contract=off.
6150 The default is -ffp-contract=fast.
6152 -fomit-frame-pointer
6154 Omit the frame pointer in functions that don't need one. This
6155 avoids the instructions to save, set up and restore the frame
6156 pointer; on many targets it also makes an extra register available.
6157 On some targets this flag has no effect because the standard
6159 calling sequence always uses a frame pointer, so it cannot be
6160 omitted.
6161 Note that -fno-omit-frame-pointer doesn't guarantee the frame
6163 pointer is used in all functions. Several targets always omit the
6164 frame pointer in leaf functions.
6165 Enabled by default at -O and higher.
6167 -foptimize-sibling-calls
6169 Optimize sibling and tail recursive calls.
6170 Enabled at levels -O2, -O3, -Os.
6172 -foptimize-strlen
6174 Optimize various standard C string functions (e.g. "strlen",
6175 "strchr" or "strcpy") and their "_FORTIFY_SOURCE" counterparts into
6176 faster alternatives.
6177 Enabled at levels -O2, -O3.
6179 -fno-inline
6181 Do not expand any functions inline apart from those marked with the
6182 "always_inline" attribute. This is the default when not
6183 optimizing.
6184 Single functions can be exempted from inlining by marking them with
6186 the "noinline" attribute.
6187 -finline-small-functions
6189 Integrate functions into their callers when their body is smaller
6190 than expected function call code (so overall size of program gets
6191 smaller). The compiler heuristically decides which functions are
6192 simple enough to be worth integrating in this way. This inlining
6193 applies to all functions, even those not declared inline.
6194 Enabled at levels -O2, -O3, -Os.
6196 -findirect-inlining
6198 Inline also indirect calls that are discovered to be known at
6199 compile time thanks to previous inlining. This option has any
6200 effect only when inlining itself is turned on by the
6201 -finline-functions or -finline-small-functions options.
6202 Enabled at levels -O2, -O3, -Os.
6204 -finline-functions
6206 Consider all functions for inlining, even if they are not declared
6207 inline. The compiler heuristically decides which functions are
6208 worth integrating in this way.
6209 If all calls to a given function are integrated, and the function
6211 is declared "static", then the function is normally not output as
6212 assembler code in its own right.
6213 Enabled at levels -O2, -O3, -Os.
6215 -finline-functions-called-once
6217 Consider all "static" functions called once for inlining into their
6218 caller even if they are not marked "inline". If a call to a given
6219 function is integrated, then the function is not output as
6220 assembler code in its own right.
6221 Enabled at levels -O1, -O2, -O3 and -Os.
6223 -fearly-inlining
6225 Inline functions marked by "always_inline" and functions whose body
6226 seems smaller than the function call overhead early before doing
6227 -fprofile-generate instrumentation and real inlining pass. Doing
6228 so makes profiling significantly cheaper and usually inlining
6229 faster on programs having large chains of nested wrapper functions.
6230 Enabled by default.
6232 -fipa-sra
6234 Perform interprocedural scalar replacement of aggregates, removal
6235 of unused parameters and replacement of parameters passed by
6236 reference by parameters passed by value.
6237 Enabled at levels -O2, -O3 and -Os.
6239 -finline-limit=n
6241 By default, GCC limits the size of functions that can be inlined.
6242 This flag allows coarse control of this limit. n is the size of
6243 functions that can be inlined in number of pseudo instructions.
6244 Inlining is actually controlled by a number of parameters, which
6246 may be specified individually by using --param name=value. The
6247 -finline-limit=n option sets some of these parameters as follows:
6248 max-inline-insns-single
6250 is set to n/2.
6251 max-inline-insns-auto
6253 is set to n/2.
6254 See below for a documentation of the individual parameters
6256 controlling inlining and for the defaults of these parameters.
6257 Note: there may be no value to -finline-limit that results in
6259 default behavior.
6260 Note: pseudo instruction represents, in this particular context, an
6262 abstract measurement of function's size. In no way does it
6263 represent a count of assembly instructions and as such its exact
6264 meaning might change from one release to an another.
6265 -fno-keep-inline-dllexport
6267 This is a more fine-grained version of -fkeep-inline-functions,
6268 which applies only to functions that are declared using the
6269 "dllexport" attribute or declspec.
6270 -fkeep-inline-functions
6272 In C, emit "static" functions that are declared "inline" into the
6273 object file, even if the function has been inlined into all of its
6274 callers. This switch does not affect functions using the "extern
6275 inline" extension in GNU C90. In C++, emit any and all inline
6276 functions into the object file.
6277 -fkeep-static-functions
6279 Emit "static" functions into the object file, even if the function
6280 is never used.
6281 -fkeep-static-consts
6283 Emit variables declared "static const" when optimization isn't
6284 turned on, even if the variables aren't referenced.
6285 GCC enables this option by default. If you want to force the
6287 compiler to check if a variable is referenced, regardless of
6288 whether or not optimization is turned on, use the
6289 -fno-keep-static-consts option.
6290 -fmerge-constants
6292 Attempt to merge identical constants (string constants and
6293 floating-point constants) across compilation units.
6294 This option is the default for optimized compilation if the
6296 assembler and linker support it. Use -fno-merge-constants to
6297 inhibit this behavior.
6298 Enabled at levels -O, -O2, -O3, -Os.
6300 -fmerge-all-constants
6302 Attempt to merge identical constants and identical variables.
6303 This option implies -fmerge-constants. In addition to
6305 -fmerge-constants this considers e.g. even constant initialized
6306 arrays or initialized constant variables with integral or floating-
6307 point types. Languages like C or C++ require each variable,
6308 including multiple instances of the same variable in recursive
6309 calls, to have distinct locations, so using this option results in
6310 non-conforming behavior.
6311 -fmodulo-sched
6313 Perform swing modulo scheduling immediately before the first
6314 scheduling pass. This pass looks at innermost loops and reorders
6315 their instructions by overlapping different iterations.
6316 -fmodulo-sched-allow-regmoves
6318 Perform more aggressive SMS-based modulo scheduling with register
6319 moves allowed. By setting this flag certain anti-dependences edges
6320 are deleted, which triggers the generation of reg-moves based on
6321 the life-range analysis. This option is effective only with
6322 -fmodulo-sched enabled.
6323 -fno-branch-count-reg
6325 Avoid running a pass scanning for opportunities to use "decrement
6326 and branch" instructions on a count register instead of generating
6327 sequences of instructions that decrement a register, compare it
6328 against zero, and then branch based upon the result. This option
6329 is only meaningful on architectures that support such instructions,
6330 which include x86, PowerPC, IA-64 and S/390. Note that the
6331 -fno-branch-count-reg option doesn't remove the decrement and
6332 branch instructions from the generated instruction stream
6333 introduced by other optimization passes.
6334 Enabled by default at -O1 and higher.
6336 The default is -fbranch-count-reg.
6338 -fno-function-cse
6340 Do not put function addresses in registers; make each instruction
6341 that calls a constant function contain the function's address
6342 explicitly.
6343 This option results in less efficient code, but some strange hacks
6345 that alter the assembler output may be confused by the
6346 optimizations performed when this option is not used.
6347 The default is -ffunction-cse
6349 -fno-zero-initialized-in-bss
6351 If the target supports a BSS section, GCC by default puts variables
6352 that are initialized to zero into BSS. This can save space in the
6353 resulting code.
6354 This option turns off this behavior because some programs
6356 explicitly rely on variables going to the data section---e.g., so
6357 that the resulting executable can find the beginning of that
6358 section and/or make assumptions based on that.
6359 The default is -fzero-initialized-in-bss.
6361 -fthread-jumps
6363 Perform optimizations that check to see if a jump branches to a
6364 location where another comparison subsumed by the first is found.
6365 If so, the first branch is redirected to either the destination of
6366 the second branch or a point immediately following it, depending on
6367 whether the condition is known to be true or false.
6368 Enabled at levels -O2, -O3, -Os.
6370 -fsplit-wide-types
6372 When using a type that occupies multiple registers, such as "long
6373 long" on a 32-bit system, split the registers apart and allocate
6374 them independently. This normally generates better code for those
6375 types, but may make debugging more difficult.
6376 Enabled at levels -O, -O2, -O3, -Os.
6378 -fcse-follow-jumps
6380 In common subexpression elimination (CSE), scan through jump
6381 instructions when the target of the jump is not reached by any
6382 other path. For example, when CSE encounters an "if" statement
6383 with an "else" clause, CSE follows the jump when the condition
6384 tested is false.
6385 Enabled at levels -O2, -O3, -Os.
6387 -fcse-skip-blocks
6389 This is similar to -fcse-follow-jumps, but causes CSE to follow
6390 jumps that conditionally skip over blocks. When CSE encounters a
6391 simple "if" statement with no else clause, -fcse-skip-blocks causes
6392 CSE to follow the jump around the body of the "if".
6393 Enabled at levels -O2, -O3, -Os.
6395 -frerun-cse-after-loop
6397 Re-run common subexpression elimination after loop optimizations
6398 are performed.
6399 Enabled at levels -O2, -O3, -Os.
6401 -fgcse
6403 Perform a global common subexpression elimination pass. This pass
6404 also performs global constant and copy propagation.
6405 Note: When compiling a program using computed gotos, a GCC
6407 extension, you may get better run-time performance if you disable
6408 the global common subexpression elimination pass by adding
6409 -fno-gcse to the command line.
6410 Enabled at levels -O2, -O3, -Os.
6412 -fgcse-lm
6414 When -fgcse-lm is enabled, global common subexpression elimination
6415 attempts to move loads that are only killed by stores into
6416 themselves. This allows a loop containing a load/store sequence to
6417 be changed to a load outside the loop, and a copy/store within the
6418 loop.
6419 Enabled by default when -fgcse is enabled.
6421 -fgcse-sm
6423 When -fgcse-sm is enabled, a store motion pass is run after global
6424 common subexpression elimination. This pass attempts to move
6425 stores out of loops. When used in conjunction with -fgcse-lm,
6426 loops containing a load/store sequence can be changed to a load
6427 before the loop and a store after the loop.
6428 Not enabled at any optimization level.
6430 -fgcse-las
6432 When -fgcse-las is enabled, the global common subexpression
6433 elimination pass eliminates redundant loads that come after stores
6434 to the same memory location (both partial and full redundancies).
6435 Not enabled at any optimization level.
6437 -fgcse-after-reload
6439 When -fgcse-after-reload is enabled, a redundant load elimination
6440 pass is performed after reload. The purpose of this pass is to
6441 clean up redundant spilling.
6442 -faggressive-loop-optimizations
6444 This option tells the loop optimizer to use language constraints to
6445 derive bounds for the number of iterations of a loop. This assumes
6446 that loop code does not invoke undefined behavior by for example
6447 causing signed integer overflows or out-of-bound array accesses.
6448 The bounds for the number of iterations of a loop are used to guide
6449 loop unrolling and peeling and loop exit test optimizations. This
6450 option is enabled by default.
6451 -funconstrained-commons
6453 This option tells the compiler that variables declared in common
6454 blocks (e.g. Fortran) may later be overridden with longer trailing
6455 arrays. This prevents certain optimizations that depend on knowing
6456 the array bounds.
6457 -fcrossjumping
6459 Perform cross-jumping transformation. This transformation unifies
6460 equivalent code and saves code size. The resulting code may or may
6461 not perform better than without cross-jumping.
6462 Enabled at levels -O2, -O3, -Os.
6464 -fauto-inc-dec
6466 Combine increments or decrements of addresses with memory accesses.
6467 This pass is always skipped on architectures that do not have
6468 instructions to support this. Enabled by default at -O and higher
6469 on architectures that support this.
6470 -fdce
6472 Perform dead code elimination (DCE) on RTL. Enabled by default at
6473 -O and higher.
6474 -fdse
6476 Perform dead store elimination (DSE) on RTL. Enabled by default at
6477 -O and higher.
6478 -fif-conversion
6480 Attempt to transform conditional jumps into branch-less
6481 equivalents. This includes use of conditional moves, min, max, set
6482 flags and abs instructions, and some tricks doable by standard
6483 arithmetics. The use of conditional execution on chips where it is
6484 available is controlled by -fif-conversion2.
6485 Enabled at levels -O, -O2, -O3, -Os.
6487 -fif-conversion2
6489 Use conditional execution (where available) to transform
6490 conditional jumps into branch-less equivalents.
6491 Enabled at levels -O, -O2, -O3, -Os.
6493 -fdeclone-ctor-dtor
6495 The C++ ABI requires multiple entry points for constructors and
6496 destructors: one for a base subobject, one for a complete object,
6497 and one for a virtual destructor that calls operator delete
6498 afterwards. For a hierarchy with virtual bases, the base and
6499 complete variants are clones, which means two copies of the
6500 function. With this option, the base and complete variants are
6501 changed to be thunks that call a common implementation.
6502 Enabled by -Os.
6504 -fdelete-null-pointer-checks
6506 Assume that programs cannot safely dereference null pointers, and
6507 that no code or data element resides at address zero. This option
6508 enables simple constant folding optimizations at all optimization
6509 levels. In addition, other optimization passes in GCC use this
6510 flag to control global dataflow analyses that eliminate useless
6511 checks for null pointers; these assume that a memory access to
6512 address zero always results in a trap, so that if a pointer is
6513 checked after it has already been dereferenced, it cannot be null.
6514 Note however that in some environments this assumption is not true.
6516 Use -fno-delete-null-pointer-checks to disable this optimization
6517 for programs that depend on that behavior.
6518 This option is enabled by default on most targets. On Nios II ELF,
6520 it defaults to off. On AVR, CR16, and MSP430, this option is
6521 completely disabled.
6522 Passes that use the dataflow information are enabled independently
6524 at different optimization levels.
6525 -fdevirtualize
6527 Attempt to convert calls to virtual functions to direct calls.
6528 This is done both within a procedure and interprocedurally as part
6529 of indirect inlining (-findirect-inlining) and interprocedural
6530 constant propagation (-fipa-cp). Enabled at levels -O2, -O3, -Os.
6531 -fdevirtualize-speculatively
6533 Attempt to convert calls to virtual functions to speculative direct
6534 calls. Based on the analysis of the type inheritance graph,
6535 determine for a given call the set of likely targets. If the set is
6536 small, preferably of size 1, change the call into a conditional
6537 deciding between direct and indirect calls. The speculative calls
6538 enable more optimizations, such as inlining. When they seem
6539 useless after further optimization, they are converted back into
6540 original form.
6541 -fdevirtualize-at-ltrans
6543 Stream extra information needed for aggressive devirtualization
6544 when running the link-time optimizer in local transformation mode.
6545 This option enables more devirtualization but significantly
6546 increases the size of streamed data. For this reason it is disabled
6547 by default.
6548 -fexpensive-optimizations
6550 Perform a number of minor optimizations that are relatively
6551 expensive.
6552 Enabled at levels -O2, -O3, -Os.
6554 -free
6556 Attempt to remove redundant extension instructions. This is
6557 especially helpful for the x86-64 architecture, which implicitly
6558 zero-extends in 64-bit registers after writing to their lower
6559 32-bit half.
6560 Enabled for Alpha, AArch64 and x86 at levels -O2, -O3, -Os.
6562 -fno-lifetime-dse
6564 In C++ the value of an object is only affected by changes within
6565 its lifetime: when the constructor begins, the object has an
6566 indeterminate value, and any changes during the lifetime of the
6567 object are dead when the object is destroyed. Normally dead store
6568 elimination will take advantage of this; if your code relies on the
6569 value of the object storage persisting beyond the lifetime of the
6570 object, you can use this flag to disable this optimization. To
6571 preserve stores before the constructor starts (e.g. because your
6572 operator new clears the object storage) but still treat the object
6573 as dead after the destructor you, can use -flifetime-dse=1. The
6574 default behavior can be explicitly selected with -flifetime-dse=2.
6575 -flifetime-dse=0 is equivalent to -fno-lifetime-dse.
6576 -flive-range-shrinkage
6578 Attempt to decrease register pressure through register live range
6579 shrinkage. This is helpful for fast processors with small or
6580 moderate size register sets.
6581 -fira-algorithm=algorithm
6583 Use the specified coloring algorithm for the integrated register
6584 allocator. The algorithm argument can be priority, which specifies
6585 Chow's priority coloring, or CB, which specifies Chaitin-Briggs
6586 coloring. Chaitin-Briggs coloring is not implemented for all
6587 architectures, but for those targets that do support it, it is the
6588 default because it generates better code.
6589 -fira-region=region
6591 Use specified regions for the integrated register allocator. The
6592 region argument should be one of the following:
6593 all Use all loops as register allocation regions. This can give
6595 the best results for machines with a small and/or irregular
6596 register set.
6597 mixed
6599 Use all loops except for loops with small register pressure as
6600 the regions. This value usually gives the best results in most
6601 cases and for most architectures, and is enabled by default
6602 when compiling with optimization for speed (-O, -O2, ...).
6603 one Use all functions as a single region. This typically results
6605 in the smallest code size, and is enabled by default for -Os or
6606 -O0.
6607 -fira-hoist-pressure
6609 Use IRA to evaluate register pressure in the code hoisting pass for
6610 decisions to hoist expressions. This option usually results in
6611 smaller code, but it can slow the compiler down.
6612 This option is enabled at level -Os for all targets.
6614 -fira-loop-pressure
6616 Use IRA to evaluate register pressure in loops for decisions to
6617 move loop invariants. This option usually results in generation of
6618 faster and smaller code on machines with large register files (>=
6619 32 registers), but it can slow the compiler down.
6620 This option is enabled at level -O3 for some targets.
6622 -fno-ira-share-save-slots
6624 Disable sharing of stack slots used for saving call-used hard
6625 registers living through a call. Each hard register gets a
6626 separate stack slot, and as a result function stack frames are
6627 larger.
6628 -fno-ira-share-spill-slots
6630 Disable sharing of stack slots allocated for pseudo-registers.
6631 Each pseudo-register that does not get a hard register gets a
6632 separate stack slot, and as a result function stack frames are
6633 larger.
6634 -flra-remat
6636 Enable CFG-sensitive rematerialization in LRA. Instead of loading
6637 values of spilled pseudos, LRA tries to rematerialize (recalculate)
6638 values if it is profitable.
6639 Enabled at levels -O2, -O3, -Os.
6641 -fdelayed-branch
6643 If supported for the target machine, attempt to reorder
6644 instructions to exploit instruction slots available after delayed
6645 branch instructions.
6646 Enabled at levels -O, -O2, -O3, -Os.
6648 -fschedule-insns
6650 If supported for the target machine, attempt to reorder
6651 instructions to eliminate execution stalls due to required data
6652 being unavailable. This helps machines that have slow floating
6653 point or memory load instructions by allowing other instructions to
6654 be issued until the result of the load or floating-point
6655 instruction is required.
6656 Enabled at levels -O2, -O3.
6658 -fschedule-insns2
6660 Similar to -fschedule-insns, but requests an additional pass of
6661 instruction scheduling after register allocation has been done.
6662 This is especially useful on machines with a relatively small
6663 number of registers and where memory load instructions take more
6664 than one cycle.
6665 Enabled at levels -O2, -O3, -Os.
6667 -fno-sched-interblock
6669 Don't schedule instructions across basic blocks. This is normally
6670 enabled by default when scheduling before register allocation, i.e.
6671 with -fschedule-insns or at -O2 or higher.
6672 -fno-sched-spec
6674 Don't allow speculative motion of non-load instructions. This is
6675 normally enabled by default when scheduling before register
6676 allocation, i.e. with -fschedule-insns or at -O2 or higher.
6677 -fsched-pressure
6679 Enable register pressure sensitive insn scheduling before register
6680 allocation. This only makes sense when scheduling before register
6681 allocation is enabled, i.e. with -fschedule-insns or at -O2 or
6682 higher. Usage of this option can improve the generated code and
6683 decrease its size by preventing register pressure increase above
6684 the number of available hard registers and subsequent spills in
6685 register allocation.
6686 -fsched-spec-load
6688 Allow speculative motion of some load instructions. This only
6689 makes sense when scheduling before register allocation, i.e. with
6690 -fschedule-insns or at -O2 or higher.
6691 -fsched-spec-load-dangerous
6693 Allow speculative motion of more load instructions. This only
6694 makes sense when scheduling before register allocation, i.e. with
6695 -fschedule-insns or at -O2 or higher.
6696 -fsched-stalled-insns
6698 -fsched-stalled-insns=n
6699 Define how many insns (if any) can be moved prematurely from the
6700 queue of stalled insns into the ready list during the second
6701 scheduling pass. -fno-sched-stalled-insns means that no insns are
6702 moved prematurely, -fsched-stalled-insns=0 means there is no limit
6703 on how many queued insns can be moved prematurely.
6704 -fsched-stalled-insns without a value is equivalent to
6705 -fsched-stalled-insns=1.
6706 -fsched-stalled-insns-dep
6708 -fsched-stalled-insns-dep=n
6709 Define how many insn groups (cycles) are examined for a dependency
6710 on a stalled insn that is a candidate for premature removal from
6711 the queue of stalled insns. This has an effect only during the
6712 second scheduling pass, and only if -fsched-stalled-insns is used.
6713 -fno-sched-stalled-insns-dep is equivalent to
6714 -fsched-stalled-insns-dep=0. -fsched-stalled-insns-dep without a
6715 value is equivalent to -fsched-stalled-insns-dep=1.
6716 -fsched2-use-superblocks
6718 When scheduling after register allocation, use superblock
6719 scheduling. This allows motion across basic block boundaries,
6720 resulting in faster schedules. This option is experimental, as not
6721 all machine descriptions used by GCC model the CPU closely enough
6722 to avoid unreliable results from the algorithm.
6723 This only makes sense when scheduling after register allocation,
6725 i.e. with -fschedule-insns2 or at -O2 or higher.
6726 -fsched-group-heuristic
6728 Enable the group heuristic in the scheduler. This heuristic favors
6729 the instruction that belongs to a schedule group. This is enabled
6730 by default when scheduling is enabled, i.e. with -fschedule-insns
6731 or -fschedule-insns2 or at -O2 or higher.
6732 -fsched-critical-path-heuristic
6734 Enable the critical-path heuristic in the scheduler. This
6735 heuristic favors instructions on the critical path. This is
6736 enabled by default when scheduling is enabled, i.e. with
6737 -fschedule-insns or -fschedule-insns2 or at -O2 or higher.
6738 -fsched-spec-insn-heuristic
6740 Enable the speculative instruction heuristic in the scheduler.
6741 This heuristic favors speculative instructions with greater
6742 dependency weakness. This is enabled by default when scheduling is
6743 enabled, i.e. with -fschedule-insns or -fschedule-insns2 or at -O2
6744 or higher.
6745 -fsched-rank-heuristic
6747 Enable the rank heuristic in the scheduler. This heuristic favors
6748 the instruction belonging to a basic block with greater size or
6749 frequency. This is enabled by default when scheduling is enabled,
6750 i.e. with -fschedule-insns or -fschedule-insns2 or at -O2 or
6751 higher.
6752 -fsched-last-insn-heuristic
6754 Enable the last-instruction heuristic in the scheduler. This
6755 heuristic favors the instruction that is less dependent on the last
6756 instruction scheduled. This is enabled by default when scheduling
6757 is enabled, i.e. with -fschedule-insns or -fschedule-insns2 or at
6758 -O2 or higher.
6759 -fsched-dep-count-heuristic
6761 Enable the dependent-count heuristic in the scheduler. This
6762 heuristic favors the instruction that has more instructions
6763 depending on it. This is enabled by default when scheduling is
6764 enabled, i.e. with -fschedule-insns or -fschedule-insns2 or at -O2
6765 or higher.
6766 -freschedule-modulo-scheduled-loops
6768 Modulo scheduling is performed before traditional scheduling. If a
6769 loop is modulo scheduled, later scheduling passes may change its
6770 schedule. Use this option to control that behavior.
6771 -fselective-scheduling
6773 Schedule instructions using selective scheduling algorithm.
6774 Selective scheduling runs instead of the first scheduler pass.
6775 -fselective-scheduling2
6777 Schedule instructions using selective scheduling algorithm.
6778 Selective scheduling runs instead of the second scheduler pass.
6779 -fsel-sched-pipelining
6781 Enable software pipelining of innermost loops during selective
6782 scheduling. This option has no effect unless one of
6783 -fselective-scheduling or -fselective-scheduling2 is turned on.
6784 -fsel-sched-pipelining-outer-loops
6786 When pipelining loops during selective scheduling, also pipeline
6787 outer loops. This option has no effect unless
6788 -fsel-sched-pipelining is turned on.
6789 -fsemantic-interposition
6791 Some object formats, like ELF, allow interposing of symbols by the
6792 dynamic linker. This means that for symbols exported from the DSO,
6793 the compiler cannot perform interprocedural propagation, inlining
6794 and other optimizations in anticipation that the function or
6795 variable in question may change. While this feature is useful, for
6796 example, to rewrite memory allocation functions by a debugging
6797 implementation, it is expensive in the terms of code quality. With
6798 -fno-semantic-interposition the compiler assumes that if
6799 interposition happens for functions the overwriting function will
6800 have precisely the same semantics (and side effects). Similarly if
6801 interposition happens for variables, the constructor of the
6802 variable will be the same. The flag has no effect for functions
6803 explicitly declared inline (where it is never allowed for
6804 interposition to change semantics) and for symbols explicitly
6805 declared weak.
6806 -fshrink-wrap
6808 Emit function prologues only before parts of the function that need
6809 it, rather than at the top of the function. This flag is enabled
6810 by default at -O and higher.
6811 -fshrink-wrap-separate
6813 Shrink-wrap separate parts of the prologue and epilogue separately,
6814 so that those parts are only executed when needed. This option is
6815 on by default, but has no effect unless -fshrink-wrap is also
6816 turned on and the target supports this.
6817 -fcaller-saves
6819 Enable allocation of values to registers that are clobbered by
6820 function calls, by emitting extra instructions to save and restore
6821 the registers around such calls. Such allocation is done only when
6822 it seems to result in better code.
6823 This option is always enabled by default on certain machines,
6825 usually those which have no call-preserved registers to use
6826 instead.
6827 Enabled at levels -O2, -O3, -Os.
6829 -fcombine-stack-adjustments
6831 Tracks stack adjustments (pushes and pops) and stack memory
6832 references and then tries to find ways to combine them.
6833 Enabled by default at -O1 and higher.
6835 -fipa-ra
6837 Use caller save registers for allocation if those registers are not
6838 used by any called function. In that case it is not necessary to
6839 save and restore them around calls. This is only possible if
6840 called functions are part of same compilation unit as current
6841 function and they are compiled before it.
6842 Enabled at levels -O2, -O3, -Os, however the option is disabled if
6844 generated code will be instrumented for profiling (-p, or -pg) or
6845 if callee's register usage cannot be known exactly (this happens on
6846 targets that do not expose prologues and epilogues in RTL).
6847 -fconserve-stack
6849 Attempt to minimize stack usage. The compiler attempts to use less
6850 stack space, even if that makes the program slower. This option
6851 implies setting the large-stack-frame parameter to 100 and the
6852 large-stack-frame-growth parameter to 400.
6853 -ftree-reassoc
6855 Perform reassociation on trees. This flag is enabled by default at
6856 -O and higher.
6857 -fcode-hoisting
6859 Perform code hoisting. Code hoisting tries to move the evaluation
6860 of expressions executed on all paths to the function exit as early
6861 as possible. This is especially useful as a code size
6862 optimization, but it often helps for code speed as well. This flag
6863 is enabled by default at -O2 and higher.
6864 -ftree-pre
6866 Perform partial redundancy elimination (PRE) on trees. This flag
6867 is enabled by default at -O2 and -O3.
6868 -ftree-partial-pre
6870 Make partial redundancy elimination (PRE) more aggressive. This
6871 flag is enabled by default at -O3.
6872 -ftree-forwprop
6874 Perform forward propagation on trees. This flag is enabled by
6875 default at -O and higher.
6876 -ftree-fre
6878 Perform full redundancy elimination (FRE) on trees. The difference
6879 between FRE and PRE is that FRE only considers expressions that are
6880 computed on all paths leading to the redundant computation. This
6881 analysis is faster than PRE, though it exposes fewer redundancies.
6882 This flag is enabled by default at -O and higher.
6883 -ftree-phiprop
6885 Perform hoisting of loads from conditional pointers on trees. This
6886 pass is enabled by default at -O and higher.
6887 -fhoist-adjacent-loads
6889 Speculatively hoist loads from both branches of an if-then-else if
6890 the loads are from adjacent locations in the same structure and the
6891 target architecture has a conditional move instruction. This flag
6892 is enabled by default at -O2 and higher.
6893 -ftree-copy-prop
6895 Perform copy propagation on trees. This pass eliminates
6896 unnecessary copy operations. This flag is enabled by default at -O
6897 and higher.
6898 -fipa-pure-const
6900 Discover which functions are pure or constant. Enabled by default
6901 at -O and higher.
6902 -fipa-reference
6904 Discover which static variables do not escape the compilation unit.
6905 Enabled by default at -O and higher.
6906 -fipa-pta
6908 Perform interprocedural pointer analysis and interprocedural
6909 modification and reference analysis. This option can cause
6910 excessive memory and compile-time usage on large compilation units.
6911 It is not enabled by default at any optimization level.
6912 -fipa-profile
6914 Perform interprocedural profile propagation. The functions called
6915 only from cold functions are marked as cold. Also functions
6916 executed once (such as "cold", "noreturn", static constructors or
6917 destructors) are identified. Cold functions and loop less parts of
6918 functions executed once are then optimized for size. Enabled by
6919 default at -O and higher.
6920 -fipa-cp
6922 Perform interprocedural constant propagation. This optimization
6923 analyzes the program to determine when values passed to functions
6924 are constants and then optimizes accordingly. This optimization
6925 can substantially increase performance if the application has
6926 constants passed to functions. This flag is enabled by default at
6927 -O2, -Os and -O3.
6928 -fipa-cp-clone
6930 Perform function cloning to make interprocedural constant
6931 propagation stronger. When enabled, interprocedural constant
6932 propagation performs function cloning when externally visible
6933 function can be called with constant arguments. Because this
6934 optimization can create multiple copies of functions, it may
6935 significantly increase code size (see --param
6936 ipcp-unit-growth=value). This flag is enabled by default at -O3.
6937 -fipa-bit-cp
6939 When enabled, perform interprocedural bitwise constant propagation.
6940 This flag is enabled by default at -O2. It requires that -fipa-cp
6941 is enabled.
6942 -fipa-vrp
6944 When enabled, perform interprocedural propagation of value ranges.
6945 This flag is enabled by default at -O2. It requires that -fipa-cp
6946 is enabled.
6947 -fipa-icf
6949 Perform Identical Code Folding for functions and read-only
6950 variables. The optimization reduces code size and may disturb
6951 unwind stacks by replacing a function by equivalent one with a
6952 different name. The optimization works more effectively with link-
6953 time optimization enabled.
6954 Nevertheless the behavior is similar to Gold Linker ICF
6956 optimization, GCC ICF works on different levels and thus the
6957 optimizations are not same - there are equivalences that are found
6958 only by GCC and equivalences found only by Gold.
6959 This flag is enabled by default at -O2 and -Os.
6961 -fisolate-erroneous-paths-dereference
6963 Detect paths that trigger erroneous or undefined behavior due to
6964 dereferencing a null pointer. Isolate those paths from the main
6965 control flow and turn the statement with erroneous or undefined
6966 behavior into a trap. This flag is enabled by default at -O2 and
6967 higher and depends on -fdelete-null-pointer-checks also being
6968 enabled.
6969 -fisolate-erroneous-paths-attribute
6971 Detect paths that trigger erroneous or undefined behavior due to a
6972 null value being used in a way forbidden by a "returns_nonnull" or
6973 "nonnull" attribute. Isolate those paths from the main control
6974 flow and turn the statement with erroneous or undefined behavior
6975 into a trap. This is not currently enabled, but may be enabled by
6976 -O2 in the future.
6977 -ftree-sink
6979 Perform forward store motion on trees. This flag is enabled by
6980 default at -O and higher.
6981 -ftree-bit-ccp
6983 Perform sparse conditional bit constant propagation on trees and
6984 propagate pointer alignment information. This pass only operates
6985 on local scalar variables and is enabled by default at -O and
6986 higher. It requires that -ftree-ccp is enabled.
6987 -ftree-ccp
6989 Perform sparse conditional constant propagation (CCP) on trees.
6990 This pass only operates on local scalar variables and is enabled by
6991 default at -O and higher.
6992 -fssa-backprop
6994 Propagate information about uses of a value up the definition chain
6995 in order to simplify the definitions. For example, this pass
6996 strips sign operations if the sign of a value never matters. The
6997 flag is enabled by default at -O and higher.
6998 -fssa-phiopt
7000 Perform pattern matching on SSA PHI nodes to optimize conditional
7001 code. This pass is enabled by default at -O and higher.
7002 -ftree-switch-conversion
7004 Perform conversion of simple initializations in a switch to
7005 initializations from a scalar array. This flag is enabled by
7006 default at -O2 and higher.
7007 -ftree-tail-merge
7009 Look for identical code sequences. When found, replace one with a
7010 jump to the other. This optimization is known as tail merging or
7011 cross jumping. This flag is enabled by default at -O2 and higher.
7012 The compilation time in this pass can be limited using max-tail-
7013 merge-comparisons parameter and max-tail-merge-iterations
7014 parameter.
7015 -ftree-dce
7017 Perform dead code elimination (DCE) on trees. This flag is enabled
7018 by default at -O and higher.
7019 -ftree-builtin-call-dce
7021 Perform conditional dead code elimination (DCE) for calls to built-
7022 in functions that may set "errno" but are otherwise free of side
7023 effects. This flag is enabled by default at -O2 and higher if -Os
7024 is not also specified.
7025 -ftree-dominator-opts
7027 Perform a variety of simple scalar cleanups (constant/copy
7028 propagation, redundancy elimination, range propagation and
7029 expression simplification) based on a dominator tree traversal.
7030 This also performs jump threading (to reduce jumps to jumps). This
7031 flag is enabled by default at -O and higher.
7032 -ftree-dse
7034 Perform dead store elimination (DSE) on trees. A dead store is a
7035 store into a memory location that is later overwritten by another
7036 store without any intervening loads. In this case the earlier
7037 store can be deleted. This flag is enabled by default at -O and
7038 higher.
7039 -ftree-ch
7041 Perform loop header copying on trees. This is beneficial since it
7042 increases effectiveness of code motion optimizations. It also
7043 saves one jump. This flag is enabled by default at -O and higher.
7044 It is not enabled for -Os, since it usually increases code size.
7045 -ftree-loop-optimize
7047 Perform loop optimizations on trees. This flag is enabled by
7048 default at -O and higher.
7049 -ftree-loop-linear
7051 -floop-strip-mine
7052 -floop-block
7053 Perform loop nest optimizations. Same as -floop-nest-optimize. To
7054 use this code transformation, GCC has to be configured with
7055 --with-isl to enable the Graphite loop transformation
7056 infrastructure.
7057 -fgraphite-identity
7059 Enable the identity transformation for graphite. For every SCoP we
7060 generate the polyhedral representation and transform it back to
7061 gimple. Using -fgraphite-identity we can check the costs or
7062 benefits of the GIMPLE -> GRAPHITE -> GIMPLE transformation. Some
7063 minimal optimizations are also performed by the code generator isl,
7064 like index splitting and dead code elimination in loops.
7065 -floop-nest-optimize
7067 Enable the isl based loop nest optimizer. This is a generic loop
7068 nest optimizer based on the Pluto optimization algorithms. It
7069 calculates a loop structure optimized for data-locality and
7070 parallelism. This option is experimental.
7071 -floop-parallelize-all
7073 Use the Graphite data dependence analysis to identify loops that
7074 can be parallelized. Parallelize all the loops that can be
7075 analyzed to not contain loop carried dependences without checking
7076 that it is profitable to parallelize the loops.
7077 -ftree-coalesce-vars
7079 While transforming the program out of the SSA representation,
7080 attempt to reduce copying by coalescing versions of different user-
7081 defined variables, instead of just compiler temporaries. This may
7082 severely limit the ability to debug an optimized program compiled
7083 with -fno-var-tracking-assignments. In the negated form, this flag
7084 prevents SSA coalescing of user variables. This option is enabled
7085 by default if optimization is enabled, and it does very little
7086 otherwise.
7087 -ftree-loop-if-convert
7089 Attempt to transform conditional jumps in the innermost loops to
7090 branch-less equivalents. The intent is to remove control-flow from
7091 the innermost loops in order to improve the ability of the
7092 vectorization pass to handle these loops. This is enabled by
7093 default if vectorization is enabled.
7094 -ftree-loop-distribution
7096 Perform loop distribution. This flag can improve cache performance
7097 on big loop bodies and allow further loop optimizations, like
7098 parallelization or vectorization, to take place. For example, the
7099 loop
7100 DO I = 1, N
7102 A(I) = B(I) + C
7103 D(I) = E(I) * F
7104 ENDDO
7105 is transformed to
7107 DO I = 1, N
7109 A(I) = B(I) + C
7110 ENDDO
7111 DO I = 1, N
7112 D(I) = E(I) * F
7113 ENDDO
7114 -ftree-loop-distribute-patterns
7116 Perform loop distribution of patterns that can be code generated
7117 with calls to a library. This flag is enabled by default at -O3.
7118 This pass distributes the initialization loops and generates a call
7120 to memset zero. For example, the loop
7121 DO I = 1, N
7123 A(I) = 0
7124 B(I) = A(I) + I
7125 ENDDO
7126 is transformed to
7128 DO I = 1, N
7130 A(I) = 0
7131 ENDDO
7132 DO I = 1, N
7133 B(I) = A(I) + I
7134 ENDDO
7135 and the initialization loop is transformed into a call to memset
7137 zero.
7138 -floop-interchange
7140 Perform loop interchange outside of graphite. This flag can
7141 improve cache performance on loop nest and allow further loop
7142 optimizations, like vectorization, to take place. For example, the
7143 loop
7144 for (int i = 0; i < N; i++)
7146 for (int j = 0; j < N; j++)
7147 for (int k = 0; k < N; k++)
7148 c[i][j] = c[i][j] + a[i][k]*b[k][j];
7149 is transformed to
7151 for (int i = 0; i < N; i++)
7153 for (int k = 0; k < N; k++)
7154 for (int j = 0; j < N; j++)
7155 c[i][j] = c[i][j] + a[i][k]*b[k][j];
7156 This flag is enabled by default at -O3.
7158 -floop-unroll-and-jam
7160 Apply unroll and jam transformations on feasible loops. In a loop
7161 nest this unrolls the outer loop by some factor and fuses the
7162 resulting multiple inner loops. This flag is enabled by default at
7163 -O3.
7164 -ftree-loop-im
7166 Perform loop invariant motion on trees. This pass moves only
7167 invariants that are hard to handle at RTL level (function calls,
7168 operations that expand to nontrivial sequences of insns). With
7169 -funswitch-loops it also moves operands of conditions that are
7170 invariant out of the loop, so that we can use just trivial
7171 invariantness analysis in loop unswitching. The pass also includes
7172 store motion.
7173 -ftree-loop-ivcanon
7175 Create a canonical counter for number of iterations in loops for
7176 which determining number of iterations requires complicated
7177 analysis. Later optimizations then may determine the number
7178 easily. Useful especially in connection with unrolling.
7179 -fivopts
7181 Perform induction variable optimizations (strength reduction,
7182 induction variable merging and induction variable elimination) on
7183 trees.
7184 -ftree-parallelize-loops=n
7186 Parallelize loops, i.e., split their iteration space to run in n
7187 threads. This is only possible for loops whose iterations are
7188 independent and can be arbitrarily reordered. The optimization is
7189 only profitable on multiprocessor machines, for loops that are CPU-
7190 intensive, rather than constrained e.g. by memory bandwidth. This
7191 option implies -pthread, and thus is only supported on targets that
7192 have support for -pthread.
7193 -ftree-pta
7195 Perform function-local points-to analysis on trees. This flag is
7196 enabled by default at -O and higher.
7197 -ftree-sra
7199 Perform scalar replacement of aggregates. This pass replaces
7200 structure references with scalars to prevent committing structures
7201 to memory too early. This flag is enabled by default at -O and
7202 higher.
7203 -fstore-merging
7205 Perform merging of narrow stores to consecutive memory addresses.
7206 This pass merges contiguous stores of immediate values narrower
7207 than a word into fewer wider stores to reduce the number of
7208 instructions. This is enabled by default at -O2 and higher as well
7209 as -Os.
7210 -ftree-ter
7212 Perform temporary expression replacement during the SSA->normal
7213 phase. Single use/single def temporaries are replaced at their use
7214 location with their defining expression. This results in non-
7215 GIMPLE code, but gives the expanders much more complex trees to
7216 work on resulting in better RTL generation. This is enabled by
7217 default at -O and higher.
7218 -ftree-slsr
7220 Perform straight-line strength reduction on trees. This recognizes
7221 related expressions involving multiplications and replaces them by
7222 less expensive calculations when possible. This is enabled by
7223 default at -O and higher.
7224 -ftree-vectorize
7226 Perform vectorization on trees. This flag enables
7227 -ftree-loop-vectorize and -ftree-slp-vectorize if not explicitly
7228 specified.
7229 -ftree-loop-vectorize
7231 Perform loop vectorization on trees. This flag is enabled by
7232 default at -O3 and when -ftree-vectorize is enabled.
7233 -ftree-slp-vectorize
7235 Perform basic block vectorization on trees. This flag is enabled by
7236 default at -O3 and when -ftree-vectorize is enabled.
7237 -fvect-cost-model=model
7239 Alter the cost model used for vectorization. The model argument
7240 should be one of unlimited, dynamic or cheap. With the unlimited
7241 model the vectorized code-path is assumed to be profitable while
7242 with the dynamic model a runtime check guards the vectorized code-
7243 path to enable it only for iteration counts that will likely
7244 execute faster than when executing the original scalar loop. The
7245 cheap model disables vectorization of loops where doing so would be
7246 cost prohibitive for example due to required runtime checks for
7247 data dependence or alignment but otherwise is equal to the dynamic
7248 model. The default cost model depends on other optimization flags
7249 and is either dynamic or cheap.
7250 -fsimd-cost-model=model
7252 Alter the cost model used for vectorization of loops marked with
7253 the OpenMP simd directive. The model argument should be one of
7254 unlimited, dynamic, cheap. All values of model have the same
7255 meaning as described in -fvect-cost-model and by default a cost
7256 model defined with -fvect-cost-model is used.
7257 -ftree-vrp
7259 Perform Value Range Propagation on trees. This is similar to the
7260 constant propagation pass, but instead of values, ranges of values
7261 are propagated. This allows the optimizers to remove unnecessary
7262 range checks like array bound checks and null pointer checks. This
7263 is enabled by default at -O2 and higher. Null pointer check
7264 elimination is only done if -fdelete-null-pointer-checks is
7265 enabled.
7266 -fsplit-paths
7268 Split paths leading to loop backedges. This can improve dead code
7269 elimination and common subexpression elimination. This is enabled
7270 by default at -O2 and above.
7271 -fsplit-ivs-in-unroller
7273 Enables expression of values of induction variables in later
7274 iterations of the unrolled loop using the value in the first
7275 iteration. This breaks long dependency chains, thus improving
7276 efficiency of the scheduling passes.
7277 A combination of -fweb and CSE is often sufficient to obtain the
7279 same effect. However, that is not reliable in cases where the loop
7280 body is more complicated than a single basic block. It also does
7281 not work at all on some architectures due to restrictions in the
7282 CSE pass.
7283 This optimization is enabled by default.
7285 -fvariable-expansion-in-unroller
7287 With this option, the compiler creates multiple copies of some
7288 local variables when unrolling a loop, which can result in superior
7289 code.
7290 -fpartial-inlining
7292 Inline parts of functions. This option has any effect only when
7293 inlining itself is turned on by the -finline-functions or
7294 -finline-small-functions options.
7295 Enabled at levels -O2, -O3, -Os.
7297 -fpredictive-commoning
7299 Perform predictive commoning optimization, i.e., reusing
7300 computations (especially memory loads and stores) performed in
7301 previous iterations of loops.
7302 This option is enabled at level -O3.
7304 -fprefetch-loop-arrays
7306 If supported by the target machine, generate instructions to
7307 prefetch memory to improve the performance of loops that access
7308 large arrays.
7309 This option may generate better or worse code; results are highly
7311 dependent on the structure of loops within the source code.
7312 Disabled at level -Os.
7314 -fno-printf-return-value
7316 Do not substitute constants for known return value of formatted
7317 output functions such as "sprintf", "snprintf", "vsprintf", and
7318 "vsnprintf" (but not "printf" of "fprintf"). This transformation
7319 allows GCC to optimize or even eliminate branches based on the
7320 known return value of these functions called with arguments that
7321 are either constant, or whose values are known to be in a range
7322 that makes determining the exact return value possible. For
7323 example, when -fprintf-return-value is in effect, both the branch
7324 and the body of the "if" statement (but not the call to "snprint")
7325 can be optimized away when "i" is a 32-bit or smaller integer
7326 because the return value is guaranteed to be at most 8.
7327 char buf[9];
7329 if (snprintf (buf, "%08x", i) >= sizeof buf)
7330 ...
7331 The -fprintf-return-value option relies on other optimizations and
7333 yields best results with -O2 and above. It works in tandem with
7334 the -Wformat-overflow and -Wformat-truncation options. The
7335 -fprintf-return-value option is enabled by default.
7336 -fno-peephole
7338 -fno-peephole2
7339 Disable any machine-specific peephole optimizations. The
7340 difference between -fno-peephole and -fno-peephole2 is in how they
7341 are implemented in the compiler; some targets use one, some use the
7342 other, a few use both.
7343 -fpeephole is enabled by default. -fpeephole2 enabled at levels
7345 -O2, -O3, -Os.
7346 -fno-guess-branch-probability
7348 Do not guess branch probabilities using heuristics.
7349 GCC uses heuristics to guess branch probabilities if they are not
7351 provided by profiling feedback (-fprofile-arcs). These heuristics
7352 are based on the control flow graph. If some branch probabilities
7353 are specified by "__builtin_expect", then the heuristics are used
7354 to guess branch probabilities for the rest of the control flow
7355 graph, taking the "__builtin_expect" info into account. The
7356 interactions between the heuristics and "__builtin_expect" can be
7357 complex, and in some cases, it may be useful to disable the
7358 heuristics so that the effects of "__builtin_expect" are easier to
7359 understand.
7360 The default is -fguess-branch-probability at levels -O, -O2, -O3,
7362 -Os.
7363 -freorder-blocks
7365 Reorder basic blocks in the compiled function in order to reduce
7366 number of taken branches and improve code locality.
7367 Enabled at levels -O, -O2, -O3, -Os.
7369 -freorder-blocks-algorithm=algorithm
7371 Use the specified algorithm for basic block reordering. The
7372 algorithm argument can be simple, which does not increase code size
7373 (except sometimes due to secondary effects like alignment), or stc,
7374 the "software trace cache" algorithm, which tries to put all often
7375 executed code together, minimizing the number of branches executed
7376 by making extra copies of code.
7377 The default is simple at levels -O, -Os, and stc at levels -O2,
7379 -O3.
7380 -freorder-blocks-and-partition
7382 In addition to reordering basic blocks in the compiled function, in
7383 order to reduce number of taken branches, partitions hot and cold
7384 basic blocks into separate sections of the assembly and .o files,
7385 to improve paging and cache locality performance.
7386 This optimization is automatically turned off in the presence of
7388 exception handling or unwind tables (on targets using
7389 setjump/longjump or target specific scheme), for linkonce sections,
7390 for functions with a user-defined section attribute and on any
7391 architecture that does not support named sections. When
7392 -fsplit-stack is used this option is not enabled by default (to
7393 avoid linker errors), but may be enabled explicitly (if using a
7394 working linker).
7395 Enabled for x86 at levels -O2, -O3, -Os.
7397 -freorder-functions
7399 Reorder functions in the object file in order to improve code
7400 locality. This is implemented by using special subsections
7401 ".text.hot" for most frequently executed functions and
7402 ".text.unlikely" for unlikely executed functions. Reordering is
7403 done by the linker so object file format must support named
7404 sections and linker must place them in a reasonable way.
7405 Also profile feedback must be available to make this option
7407 effective. See -fprofile-arcs for details.
7408 Enabled at levels -O2, -O3, -Os.
7410 -fstrict-aliasing
7412 Allow the compiler to assume the strictest aliasing rules
7413 applicable to the language being compiled. For C (and C++), this
7414 activates optimizations based on the type of expressions. In
7415 particular, an object of one type is assumed never to reside at the
7416 same address as an object of a different type, unless the types are
7417 almost the same. For example, an "unsigned int" can alias an
7418 "int", but not a "void*" or a "double". A character type may alias
7419 any other type.
7420 Pay special attention to code like this:
7422 union a_union {
7424 int i;
7425 double d;
7426 };
7427 int f() {
7429 union a_union t;
7430 t.d = 3.0;
7431 return t.i;
7432 }
7433 The practice of reading from a different union member than the one
7435 most recently written to (called "type-punning") is common. Even
7436 with -fstrict-aliasing, type-punning is allowed, provided the
7437 memory is accessed through the union type. So, the code above
7438 works as expected. However, this code might not:
7439 int f() {
7441 union a_union t;
7442 int* ip;
7443 t.d = 3.0;
7444 ip = &t.i;
7445 return *ip;
7446 }
7447 Similarly, access by taking the address, casting the resulting
7449 pointer and dereferencing the result has undefined behavior, even
7450 if the cast uses a union type, e.g.:
7451 int f() {
7453 double d = 3.0;
7454 return ((union a_union *) &d)->i;
7455 }
7456 The -fstrict-aliasing option is enabled at levels -O2, -O3, -Os.
7458 -falign-functions
7460 -falign-functions=n
7461 Align the start of functions to the next power-of-two greater than
7462 n, skipping up to n bytes. For instance, -falign-functions=32
7463 aligns functions to the next 32-byte boundary, but
7464 -falign-functions=24 aligns to the next 32-byte boundary only if
7465 this can be done by skipping 23 bytes or less.
7466 -fno-align-functions and -falign-functions=1 are equivalent and
7468 mean that functions are not aligned.
7469 Some assemblers only support this flag when n is a power of two; in
7471 that case, it is rounded up.
7472 If n is not specified or is zero, use a machine-dependent default.
7474 The maximum allowed n option value is 65536.
7475 Enabled at levels -O2, -O3.
7477 -flimit-function-alignment
7479 If this option is enabled, the compiler tries to avoid
7480 unnecessarily overaligning functions. It attempts to instruct the
7481 assembler to align by the amount specified by -falign-functions,
7482 but not to skip more bytes than the size of the function.
7483 -falign-labels
7485 -falign-labels=n
7486 Align all branch targets to a power-of-two boundary, skipping up to
7487 n bytes like -falign-functions. This option can easily make code
7488 slower, because it must insert dummy operations for when the branch
7489 target is reached in the usual flow of the code.
7490 -fno-align-labels and -falign-labels=1 are equivalent and mean that
7492 labels are not aligned.
7493 If -falign-loops or -falign-jumps are applicable and are greater
7495 than this value, then their values are used instead.
7496 If n is not specified or is zero, use a machine-dependent default
7498 which is very likely to be 1, meaning no alignment. The maximum
7499 allowed n option value is 65536.
7500 Enabled at levels -O2, -O3.
7502 -falign-loops
7504 -falign-loops=n
7505 Align loops to a power-of-two boundary, skipping up to n bytes like
7506 -falign-functions. If the loops are executed many times, this
7507 makes up for any execution of the dummy operations.
7508 -fno-align-loops and -falign-loops=1 are equivalent and mean that
7510 loops are not aligned. The maximum allowed n option value is
7511 65536.
7512 If n is not specified or is zero, use a machine-dependent default.
7514 Enabled at levels -O2, -O3.
7516 -falign-jumps
7518 -falign-jumps=n
7519 Align branch targets to a power-of-two boundary, for branch targets
7520 where the targets can only be reached by jumping, skipping up to n
7521 bytes like -falign-functions. In this case, no dummy operations
7522 need be executed.
7523 -fno-align-jumps and -falign-jumps=1 are equivalent and mean that
7525 loops are not aligned.
7526 If n is not specified or is zero, use a machine-dependent default.
7528 The maximum allowed n option value is 65536.
7529 Enabled at levels -O2, -O3.
7531 -funit-at-a-time
7533 This option is left for compatibility reasons. -funit-at-a-time has
7534 no effect, while -fno-unit-at-a-time implies -fno-toplevel-reorder
7535 and -fno-section-anchors.
7536 Enabled by default.
7538 -fno-toplevel-reorder
7540 Do not reorder top-level functions, variables, and "asm"
7541 statements. Output them in the same order that they appear in the
7542 input file. When this option is used, unreferenced static
7543 variables are not removed. This option is intended to support
7544 existing code that relies on a particular ordering. For new code,
7545 it is better to use attributes when possible.
7546 Enabled at level -O0. When disabled explicitly, it also implies
7548 -fno-section-anchors, which is otherwise enabled at -O0 on some
7549 targets.
7550 -fweb
7552 Constructs webs as commonly used for register allocation purposes
7553 and assign each web individual pseudo register. This allows the
7554 register allocation pass to operate on pseudos directly, but also
7555 strengthens several other optimization passes, such as CSE, loop
7556 optimizer and trivial dead code remover. It can, however, make
7557 debugging impossible, since variables no longer stay in a "home
7558 register".
7559 Enabled by default with -funroll-loops.
7561 -fwhole-program
7563 Assume that the current compilation unit represents the whole
7564 program being compiled. All public functions and variables with
7565 the exception of "main" and those merged by attribute
7566 "externally_visible" become static functions and in effect are
7567 optimized more aggressively by interprocedural optimizers.
7568 This option should not be used in combination with -flto. Instead
7570 relying on a linker plugin should provide safer and more precise
7571 information.
7572 -flto[=n]
7574 This option runs the standard link-time optimizer. When invoked
7575 with source code, it generates GIMPLE (one of GCC's internal
7576 representations) and writes it to special ELF sections in the
7577 object file. When the object files are linked together, all the
7578 function bodies are read from these ELF sections and instantiated
7579 as if they had been part of the same translation unit.
7580 To use the link-time optimizer, -flto and optimization options
7582 should be specified at compile time and during the final link. It
7583 is recommended that you compile all the files participating in the
7584 same link with the same options and also specify those options at
7585 link time. For example:
7586 gcc -c -O2 -flto foo.c
7588 gcc -c -O2 -flto bar.c
7589 gcc -o myprog -flto -O2 foo.o bar.o
7590 The first two invocations to GCC save a bytecode representation of
7592 GIMPLE into special ELF sections inside foo.o and bar.o. The final
7593 invocation reads the GIMPLE bytecode from foo.o and bar.o, merges
7594 the two files into a single internal image, and compiles the result
7595 as usual. Since both foo.o and bar.o are merged into a single
7596 image, this causes all the interprocedural analyses and
7597 optimizations in GCC to work across the two files as if they were a
7598 single one. This means, for example, that the inliner is able to
7599 inline functions in bar.o into functions in foo.o and vice-versa.
7600 Another (simpler) way to enable link-time optimization is:
7602 gcc -o myprog -flto -O2 foo.c bar.c
7604 The above generates bytecode for foo.c and bar.c, merges them
7606 together into a single GIMPLE representation and optimizes them as
7607 usual to produce myprog.
7608 The only important thing to keep in mind is that to enable link-
7610 time optimizations you need to use the GCC driver to perform the
7611 link step. GCC then automatically performs link-time optimization
7612 if any of the objects involved were compiled with the -flto
7613 command-line option. You generally should specify the optimization
7614 options to be used for link-time optimization though GCC tries to
7615 be clever at guessing an optimization level to use from the options
7616 used at compile time if you fail to specify one at link time. You
7617 can always override the automatic decision to do link-time
7618 optimization by passing -fno-lto to the link command.
7619 To make whole program optimization effective, it is necessary to
7621 make certain whole program assumptions. The compiler needs to know
7622 what functions and variables can be accessed by libraries and
7623 runtime outside of the link-time optimized unit. When supported by
7624 the linker, the linker plugin (see -fuse-linker-plugin) passes
7625 information to the compiler about used and externally visible
7626 symbols. When the linker plugin is not available, -fwhole-program
7627 should be used to allow the compiler to make these assumptions,
7628 which leads to more aggressive optimization decisions.
7629 When -fuse-linker-plugin is not enabled, when a file is compiled
7631 with -flto, the generated object file is larger than a regular
7632 object file because it contains GIMPLE bytecodes and the usual
7633 final code (see -ffat-lto-objects. This means that object files
7634 with LTO information can be linked as normal object files; if
7635 -fno-lto is passed to the linker, no interprocedural optimizations
7636 are applied. Note that when -fno-fat-lto-objects is enabled the
7637 compile stage is faster but you cannot perform a regular, non-LTO
7638 link on them.
7639 Additionally, the optimization flags used to compile individual
7641 files are not necessarily related to those used at link time. For
7642 instance,
7643 gcc -c -O0 -ffat-lto-objects -flto foo.c
7645 gcc -c -O0 -ffat-lto-objects -flto bar.c
7646 gcc -o myprog -O3 foo.o bar.o
7647 This produces individual object files with unoptimized assembler
7649 code, but the resulting binary myprog is optimized at -O3. If,
7650 instead, the final binary is generated with -fno-lto, then myprog
7651 is not optimized.
7652 When producing the final binary, GCC only applies link-time
7654 optimizations to those files that contain bytecode. Therefore, you
7655 can mix and match object files and libraries with GIMPLE bytecodes
7656 and final object code. GCC automatically selects which files to
7657 optimize in LTO mode and which files to link without further
7658 processing.
7659 There are some code generation flags preserved by GCC when
7661 generating bytecodes, as they need to be used during the final link
7662 stage. Generally options specified at link time override those
7663 specified at compile time.
7664 If you do not specify an optimization level option -O at link time,
7666 then GCC uses the highest optimization level used when compiling
7667 the object files.
7668 Currently, the following options and their settings are taken from
7670 the first object file that explicitly specifies them: -fPIC, -fpic,
7671 -fpie, -fcommon, -fexceptions, -fnon-call-exceptions, -fgnu-tm and
7672 all the -m target flags.
7673 Certain ABI-changing flags are required to match in all compilation
7675 units, and trying to override this at link time with a conflicting
7676 value is ignored. This includes options such as
7677 -freg-struct-return and -fpcc-struct-return.
7678 Other options such as -ffp-contract, -fno-strict-overflow, -fwrapv,
7680 -fno-trapv or -fno-strict-aliasing are passed through to the link
7681 stage and merged conservatively for conflicting translation units.
7682 Specifically -fno-strict-overflow, -fwrapv and -fno-trapv take
7683 precedence; and for example -ffp-contract=off takes precedence over
7684 -ffp-contract=fast. You can override them at link time.
7685 If LTO encounters objects with C linkage declared with incompatible
7687 types in separate translation units to be linked together
7688 (undefined behavior according to ISO C99 6.2.7), a non-fatal
7689 diagnostic may be issued. The behavior is still undefined at run
7690 time. Similar diagnostics may be raised for other languages.
7691 Another feature of LTO is that it is possible to apply
7693 interprocedural optimizations on files written in different
7694 languages:
7695 gcc -c -flto foo.c
7697 g++ -c -flto bar.cc
7698 gfortran -c -flto baz.f90
7699 g++ -o myprog -flto -O3 foo.o bar.o baz.o -lgfortran
7700 Notice that the final link is done with g++ to get the C++ runtime
7702 libraries and -lgfortran is added to get the Fortran runtime
7703 libraries. In general, when mixing languages in LTO mode, you
7704 should use the same link command options as when mixing languages
7705 in a regular (non-LTO) compilation.
7706 If object files containing GIMPLE bytecode are stored in a library
7708 archive, say libfoo.a, it is possible to extract and use them in an
7709 LTO link if you are using a linker with plugin support. To create
7710 static libraries suitable for LTO, use gcc-ar and gcc-ranlib
7711 instead of ar and ranlib; to show the symbols of object files with
7712 GIMPLE bytecode, use gcc-nm. Those commands require that ar,
7713 ranlib and nm have been compiled with plugin support. At link
7714 time, use the flag -fuse-linker-plugin to ensure that the library
7715 participates in the LTO optimization process:
7716 gcc -o myprog -O2 -flto -fuse-linker-plugin a.o b.o -lfoo
7718 With the linker plugin enabled, the linker extracts the needed
7720 GIMPLE files from libfoo.a and passes them on to the running GCC to
7721 make them part of the aggregated GIMPLE image to be optimized.
7722 If you are not using a linker with plugin support and/or do not
7724 enable the linker plugin, then the objects inside libfoo.a are
7725 extracted and linked as usual, but they do not participate in the
7726 LTO optimization process. In order to make a static library
7727 suitable for both LTO optimization and usual linkage, compile its
7728 object files with -flto -ffat-lto-objects.
7729 Link-time optimizations do not require the presence of the whole
7731 program to operate. If the program does not require any symbols to
7732 be exported, it is possible to combine -flto and -fwhole-program to
7733 allow the interprocedural optimizers to use more aggressive
7734 assumptions which may lead to improved optimization opportunities.
7735 Use of -fwhole-program is not needed when linker plugin is active
7736 (see -fuse-linker-plugin).
7737 The current implementation of LTO makes no attempt to generate
7739 bytecode that is portable between different types of hosts. The
7740 bytecode files are versioned and there is a strict version check,
7741 so bytecode files generated in one version of GCC do not work with
7742 an older or newer version of GCC.
7743 Link-time optimization does not work well with generation of
7745 debugging information on systems other than those using a
7746 combination of ELF and DWARF.
7747 If you specify the optional n, the optimization and code generation
7749 done at link time is executed in parallel using n parallel jobs by
7750 utilizing an installed make program. The environment variable MAKE
7751 may be used to override the program used. The default value for n
7752 is 1.
7753 You can also specify -flto=jobserver to use GNU make's job server
7755 mode to determine the number of parallel jobs. This is useful when
7756 the Makefile calling GCC is already executing in parallel. You
7757 must prepend a + to the command recipe in the parent Makefile for
7758 this to work. This option likely only works if MAKE is GNU make.
7759 -flto-partition=alg
7761 Specify the partitioning algorithm used by the link-time optimizer.
7762 The value is either 1to1 to specify a partitioning mirroring the
7763 original source files or balanced to specify partitioning into
7764 equally sized chunks (whenever possible) or max to create new
7765 partition for every symbol where possible. Specifying none as an
7766 algorithm disables partitioning and streaming completely. The
7767 default value is balanced. While 1to1 can be used as an workaround
7768 for various code ordering issues, the max partitioning is intended
7769 for internal testing only. The value one specifies that exactly
7770 one partition should be used while the value none bypasses
7771 partitioning and executes the link-time optimization step directly
7772 from the WPA phase.
7773 -flto-odr-type-merging
7775 Enable streaming of mangled types names of C++ types and their
7776 unification at link time. This increases size of LTO object files,
7777 but enables diagnostics about One Definition Rule violations.
7778 -flto-compression-level=n
7780 This option specifies the level of compression used for
7781 intermediate language written to LTO object files, and is only
7782 meaningful in conjunction with LTO mode (-flto). Valid values are
7783 0 (no compression) to 9 (maximum compression). Values outside this
7784 range are clamped to either 0 or 9. If the option is not given, a
7785 default balanced compression setting is used.
7786 -fuse-linker-plugin
7788 Enables the use of a linker plugin during link-time optimization.
7789 This option relies on plugin support in the linker, which is
7790 available in gold or in GNU ld 2.21 or newer.
7791 This option enables the extraction of object files with GIMPLE
7793 bytecode out of library archives. This improves the quality of
7794 optimization by exposing more code to the link-time optimizer.
7795 This information specifies what symbols can be accessed externally
7796 (by non-LTO object or during dynamic linking). Resulting code
7797 quality improvements on binaries (and shared libraries that use
7798 hidden visibility) are similar to -fwhole-program. See -flto for a
7799 description of the effect of this flag and how to use it.
7800 This option is enabled by default when LTO support in GCC is
7802 enabled and GCC was configured for use with a linker supporting
7803 plugins (GNU ld 2.21 or newer or gold).
7804 -ffat-lto-objects
7806 Fat LTO objects are object files that contain both the intermediate
7807 language and the object code. This makes them usable for both LTO
7808 linking and normal linking. This option is effective only when
7809 compiling with -flto and is ignored at link time.
7810 -fno-fat-lto-objects improves compilation time over plain LTO, but
7812 requires the complete toolchain to be aware of LTO. It requires a
7813 linker with linker plugin support for basic functionality.
7814 Additionally, nm, ar and ranlib need to support linker plugins to
7815 allow a full-featured build environment (capable of building static
7816 libraries etc). GCC provides the gcc-ar, gcc-nm, gcc-ranlib
7817 wrappers to pass the right options to these tools. With non fat LTO
7818 makefiles need to be modified to use them.
7819 Note that modern binutils provide plugin auto-load mechanism.
7821 Installing the linker plugin into $libdir/bfd-plugins has the same
7822 effect as usage of the command wrappers (gcc-ar, gcc-nm and gcc-
7823 ranlib).
7824 The default is -fno-fat-lto-objects on targets with linker plugin
7826 support.
7827 -fcompare-elim
7829 After register allocation and post-register allocation instruction
7830 splitting, identify arithmetic instructions that compute processor
7831 flags similar to a comparison operation based on that arithmetic.
7832 If possible, eliminate the explicit comparison operation.
7833 This pass only applies to certain targets that cannot explicitly
7835 represent the comparison operation before register allocation is
7836 complete.
7837 Enabled at levels -O, -O2, -O3, -Os.
7839 -fcprop-registers
7841 After register allocation and post-register allocation instruction
7842 splitting, perform a copy-propagation pass to try to reduce
7843 scheduling dependencies and occasionally eliminate the copy.
7844 Enabled at levels -O, -O2, -O3, -Os.
7846 -fprofile-correction
7848 Profiles collected using an instrumented binary for multi-threaded
7849 programs may be inconsistent due to missed counter updates. When
7850 this option is specified, GCC uses heuristics to correct or smooth
7851 out such inconsistencies. By default, GCC emits an error message
7852 when an inconsistent profile is detected.
7853 -fprofile-use
7855 -fprofile-use=path
7856 Enable profile feedback-directed optimizations, and the following
7857 optimizations which are generally profitable only with profile
7858 feedback available: -fbranch-probabilities, -fvpt, -funroll-loops,
7859 -fpeel-loops, -ftracer, -ftree-vectorize, and ftree-loop-
7860 distribute-patterns.
7861 Before you can use this option, you must first generate profiling
7863 information.
7864 By default, GCC emits an error message if the feedback profiles do
7866 not match the source code. This error can be turned into a warning
7867 by using -Wcoverage-mismatch. Note this may result in poorly
7868 optimized code.
7869 If path is specified, GCC looks at the path to find the profile
7871 feedback data files. See -fprofile-dir.
7872 -fauto-profile
7874 -fauto-profile=path
7875 Enable sampling-based feedback-directed optimizations, and the
7876 following optimizations which are generally profitable only with
7877 profile feedback available: -fbranch-probabilities, -fvpt,
7878 -funroll-loops, -fpeel-loops, -ftracer, -ftree-vectorize,
7879 -finline-functions, -fipa-cp, -fipa-cp-clone,
7880 -fpredictive-commoning, -funswitch-loops, -fgcse-after-reload, and
7881 -ftree-loop-distribute-patterns.
7882 path is the name of a file containing AutoFDO profile information.
7884 If omitted, it defaults to fbdata.afdo in the current directory.
7885 Producing an AutoFDO profile data file requires running your
7887 program with the perf utility on a supported GNU/Linux target
7888 system. For more information, see <https://perf.wiki.kernel.org/>.
7889 E.g.
7891 perf record -e br_inst_retired:near_taken -b -o perf.data \
7893 -- your_program
7894 Then use the create_gcov tool to convert the raw profile data to a
7896 format that can be used by GCC. You must also supply the
7897 unstripped binary for your program to this tool. See
7898 <https://github.com/google/autofdo>.
7899 E.g.
7901 create_gcov --binary=your_program.unstripped --profile=perf.data \
7903 --gcov=profile.afdo
7904 The following options control compiler behavior regarding floating-
7906 point arithmetic. These options trade off between speed and
7907 correctness. All must be specifically enabled.
7908 -ffloat-store
7910 Do not store floating-point variables in registers, and inhibit
7911 other options that might change whether a floating-point value is
7912 taken from a register or memory.
7913 This option prevents undesirable excess precision on machines such
7915 as the 68000 where the floating registers (of the 68881) keep more
7916 precision than a "double" is supposed to have. Similarly for the
7917 x86 architecture. For most programs, the excess precision does
7918 only good, but a few programs rely on the precise definition of
7919 IEEE floating point. Use -ffloat-store for such programs, after
7920 modifying them to store all pertinent intermediate computations
7921 into variables.
7922 -fexcess-precision=style
7924 This option allows further control over excess precision on
7925 machines where floating-point operations occur in a format with
7926 more precision or range than the IEEE standard and interchange
7927 floating-point types. By default, -fexcess-precision=fast is in
7928 effect; this means that operations may be carried out in a wider
7929 precision than the types specified in the source if that would
7930 result in faster code, and it is unpredictable when rounding to the
7931 types specified in the source code takes place. When compiling C,
7932 if -fexcess-precision=standard is specified then excess precision
7933 follows the rules specified in ISO C99; in particular, both casts
7934 and assignments cause values to be rounded to their semantic types
7935 (whereas -ffloat-store only affects assignments). This option is
7936 enabled by default for C if a strict conformance option such as
7937 -std=c99 is used. -ffast-math enables -fexcess-precision=fast by
7938 default regardless of whether a strict conformance option is used.
7939 -fexcess-precision=standard is not implemented for languages other
7941 than C. On the x86, it has no effect if -mfpmath=sse or
7942 -mfpmath=sse+387 is specified; in the former case, IEEE semantics
7943 apply without excess precision, and in the latter, rounding is
7944 unpredictable.
7945 -ffast-math
7947 Sets the options -fno-math-errno, -funsafe-math-optimizations,
7948 -ffinite-math-only, -fno-rounding-math, -fno-signaling-nans,
7949 -fcx-limited-range and -fexcess-precision=fast.
7950 This option causes the preprocessor macro "__FAST_MATH__" to be
7952 defined.
7953 This option is not turned on by any -O option besides -Ofast since
7955 it can result in incorrect output for programs that depend on an
7956 exact implementation of IEEE or ISO rules/specifications for math
7957 functions. It may, however, yield faster code for programs that do
7958 not require the guarantees of these specifications.
7959 -fno-math-errno
7961 Do not set "errno" after calling math functions that are executed
7962 with a single instruction, e.g., "sqrt". A program that relies on
7963 IEEE exceptions for math error handling may want to use this flag
7964 for speed while maintaining IEEE arithmetic compatibility.
7965 This option is not turned on by any -O option since it can result
7967 in incorrect output for programs that depend on an exact
7968 implementation of IEEE or ISO rules/specifications for math
7969 functions. It may, however, yield faster code for programs that do
7970 not require the guarantees of these specifications.
7971 The default is -fmath-errno.
7973 On Darwin systems, the math library never sets "errno". There is
7975 therefore no reason for the compiler to consider the possibility
7976 that it might, and -fno-math-errno is the default.
7977 -funsafe-math-optimizations
7979 Allow optimizations for floating-point arithmetic that (a) assume
7980 that arguments and results are valid and (b) may violate IEEE or
7981 ANSI standards. When used at link time, it may include libraries
7982 or startup files that change the default FPU control word or other
7983 similar optimizations.
7984 This option is not turned on by any -O option since it can result
7986 in incorrect output for programs that depend on an exact
7987 implementation of IEEE or ISO rules/specifications for math
7988 functions. It may, however, yield faster code for programs that do
7989 not require the guarantees of these specifications. Enables
7990 -fno-signed-zeros, -fno-trapping-math, -fassociative-math and
7991 -freciprocal-math.
7992 The default is -fno-unsafe-math-optimizations.
7994 -fassociative-math
7996 Allow re-association of operands in series of floating-point
7997 operations. This violates the ISO C and C++ language standard by
7998 possibly changing computation result. NOTE: re-ordering may change
7999 the sign of zero as well as ignore NaNs and inhibit or create
8000 underflow or overflow (and thus cannot be used on code that relies
8001 on rounding behavior like "(x + 2**52) - 2**52". May also reorder
8002 floating-point comparisons and thus may not be used when ordered
8003 comparisons are required. This option requires that both
8004 -fno-signed-zeros and -fno-trapping-math be in effect. Moreover,
8005 it doesn't make much sense with -frounding-math. For Fortran the
8006 option is automatically enabled when both -fno-signed-zeros and
8007 -fno-trapping-math are in effect.
8008 The default is -fno-associative-math.
8010 -freciprocal-math
8012 Allow the reciprocal of a value to be used instead of dividing by
8013 the value if this enables optimizations. For example "x / y" can
8014 be replaced with "x * (1/y)", which is useful if "(1/y)" is subject
8015 to common subexpression elimination. Note that this loses
8016 precision and increases the number of flops operating on the value.
8017 The default is -fno-reciprocal-math.
8019 -ffinite-math-only
8021 Allow optimizations for floating-point arithmetic that assume that
8022 arguments and results are not NaNs or +-Infs.
8023 This option is not turned on by any -O option since it can result
8025 in incorrect output for programs that depend on an exact
8026 implementation of IEEE or ISO rules/specifications for math
8027 functions. It may, however, yield faster code for programs that do
8028 not require the guarantees of these specifications.
8029 The default is -fno-finite-math-only.
8031 -fno-signed-zeros
8033 Allow optimizations for floating-point arithmetic that ignore the
8034 signedness of zero. IEEE arithmetic specifies the behavior of
8035 distinct +0.0 and -0.0 values, which then prohibits simplification
8036 of expressions such as x+0.0 or 0.0*x (even with
8037 -ffinite-math-only). This option implies that the sign of a zero
8038 result isn't significant.
8039 The default is -fsigned-zeros.
8041 -fno-trapping-math
8043 Compile code assuming that floating-point operations cannot
8044 generate user-visible traps. These traps include division by zero,
8045 overflow, underflow, inexact result and invalid operation. This
8046 option requires that -fno-signaling-nans be in effect. Setting
8047 this option may allow faster code if one relies on "non-stop" IEEE
8048 arithmetic, for example.
8049 This option should never be turned on by any -O option since it can
8051 result in incorrect output for programs that depend on an exact
8052 implementation of IEEE or ISO rules/specifications for math
8053 functions.
8054 The default is -ftrapping-math.
8056 -frounding-math
8058 Disable transformations and optimizations that assume default
8059 floating-point rounding behavior. This is round-to-zero for all
8060 floating point to integer conversions, and round-to-nearest for all
8061 other arithmetic truncations. This option should be specified for
8062 programs that change the FP rounding mode dynamically, or that may
8063 be executed with a non-default rounding mode. This option disables
8064 constant folding of floating-point expressions at compile time
8065 (which may be affected by rounding mode) and arithmetic
8066 transformations that are unsafe in the presence of sign-dependent
8067 rounding modes.
8068 The default is -fno-rounding-math.
8070 This option is experimental and does not currently guarantee to
8072 disable all GCC optimizations that are affected by rounding mode.
8073 Future versions of GCC may provide finer control of this setting
8074 using C99's "FENV_ACCESS" pragma. This command-line option will be
8075 used to specify the default state for "FENV_ACCESS".
8076 -fsignaling-nans
8078 Compile code assuming that IEEE signaling NaNs may generate user-
8079 visible traps during floating-point operations. Setting this
8080 option disables optimizations that may change the number of
8081 exceptions visible with signaling NaNs. This option implies
8082 -ftrapping-math.
8083 This option causes the preprocessor macro "__SUPPORT_SNAN__" to be
8085 defined.
8086 The default is -fno-signaling-nans.
8088 This option is experimental and does not currently guarantee to
8090 disable all GCC optimizations that affect signaling NaN behavior.
8091 -fno-fp-int-builtin-inexact
8093 Do not allow the built-in functions "ceil", "floor", "round" and
8094 "trunc", and their "float" and "long double" variants, to generate
8095 code that raises the "inexact" floating-point exception for
8096 noninteger arguments. ISO C99 and C11 allow these functions to
8097 raise the "inexact" exception, but ISO/IEC TS 18661-1:2014, the C
8098 bindings to IEEE 754-2008, does not allow these functions to do so.
8099 The default is -ffp-int-builtin-inexact, allowing the exception to
8101 be raised. This option does nothing unless -ftrapping-math is in
8102 effect.
8103 Even if -fno-fp-int-builtin-inexact is used, if the functions
8105 generate a call to a library function then the "inexact" exception
8106 may be raised if the library implementation does not follow TS
8107 18661.
8108 -fsingle-precision-constant
8110 Treat floating-point constants as single precision instead of
8111 implicitly converting them to double-precision constants.
8112 -fcx-limited-range
8114 When enabled, this option states that a range reduction step is not
8115 needed when performing complex division. Also, there is no
8116 checking whether the result of a complex multiplication or division
8117 is "NaN + I*NaN", with an attempt to rescue the situation in that
8118 case. The default is -fno-cx-limited-range, but is enabled by
8119 -ffast-math.
8120 This option controls the default setting of the ISO C99
8122 "CX_LIMITED_RANGE" pragma. Nevertheless, the option applies to all
8123 languages.
8124 -fcx-fortran-rules
8126 Complex multiplication and division follow Fortran rules. Range
8127 reduction is done as part of complex division, but there is no
8128 checking whether the result of a complex multiplication or division
8129 is "NaN + I*NaN", with an attempt to rescue the situation in that
8130 case.
8131 The default is -fno-cx-fortran-rules.
8133 The following options control optimizations that may improve
8135 performance, but are not enabled by any -O options. This section
8136 includes experimental options that may produce broken code.
8137 -fbranch-probabilities
8139 After running a program compiled with -fprofile-arcs, you can
8140 compile it a second time using -fbranch-probabilities, to improve
8141 optimizations based on the number of times each branch was taken.
8142 When a program compiled with -fprofile-arcs exits, it saves arc
8143 execution counts to a file called sourcename.gcda for each source
8144 file. The information in this data file is very dependent on the
8145 structure of the generated code, so you must use the same source
8146 code and the same optimization options for both compilations.
8147 With -fbranch-probabilities, GCC puts a REG_BR_PROB note on each
8149 JUMP_INSN and CALL_INSN. These can be used to improve
8150 optimization. Currently, they are only used in one place: in
8151 reorg.c, instead of guessing which path a branch is most likely to
8152 take, the REG_BR_PROB values are used to exactly determine which
8153 path is taken more often.
8154 -fprofile-values
8156 If combined with -fprofile-arcs, it adds code so that some data
8157 about values of expressions in the program is gathered.
8158 With -fbranch-probabilities, it reads back the data gathered from
8160 profiling values of expressions for usage in optimizations.
8161 Enabled with -fprofile-generate and -fprofile-use.
8163 -fprofile-reorder-functions
8165 Function reordering based on profile instrumentation collects first
8166 time of execution of a function and orders these functions in
8167 ascending order.
8168 Enabled with -fprofile-use.
8170 -fvpt
8172 If combined with -fprofile-arcs, this option instructs the compiler
8173 to add code to gather information about values of expressions.
8174 With -fbranch-probabilities, it reads back the data gathered and
8176 actually performs the optimizations based on them. Currently the
8177 optimizations include specialization of division operations using
8178 the knowledge about the value of the denominator.
8179 -frename-registers
8181 Attempt to avoid false dependencies in scheduled code by making use
8182 of registers left over after register allocation. This
8183 optimization most benefits processors with lots of registers.
8184 Depending on the debug information format adopted by the target,
8185 however, it can make debugging impossible, since variables no
8186 longer stay in a "home register".
8187 Enabled by default with -funroll-loops.
8189 -fschedule-fusion
8191 Performs a target dependent pass over the instruction stream to
8192 schedule instructions of same type together because target machine
8193 can execute them more efficiently if they are adjacent to each
8194 other in the instruction flow.
8195 Enabled at levels -O2, -O3, -Os.
8197 -ftracer
8199 Perform tail duplication to enlarge superblock size. This
8200 transformation simplifies the control flow of the function allowing
8201 other optimizations to do a better job.
8202 Enabled with -fprofile-use.
8204 -funroll-loops
8206 Unroll loops whose number of iterations can be determined at
8207 compile time or upon entry to the loop. -funroll-loops implies
8208 -frerun-cse-after-loop, -fweb and -frename-registers. It also
8209 turns on complete loop peeling (i.e. complete removal of loops with
8210 a small constant number of iterations). This option makes code
8211 larger, and may or may not make it run faster.
8212 Enabled with -fprofile-use.
8214 -funroll-all-loops
8216 Unroll all loops, even if their number of iterations is uncertain
8217 when the loop is entered. This usually makes programs run more
8218 slowly. -funroll-all-loops implies the same options as
8219 -funroll-loops.
8220 -fpeel-loops
8222 Peels loops for which there is enough information that they do not
8223 roll much (from profile feedback or static analysis). It also
8224 turns on complete loop peeling (i.e. complete removal of loops with
8225 small constant number of iterations).
8226 Enabled with -O3 and/or -fprofile-use.
8228 -fmove-loop-invariants
8230 Enables the loop invariant motion pass in the RTL loop optimizer.
8231 Enabled at level -O1
8232 -fsplit-loops
8234 Split a loop into two if it contains a condition that's always true
8235 for one side of the iteration space and false for the other.
8236 -funswitch-loops
8238 Move branches with loop invariant conditions out of the loop, with
8239 duplicates of the loop on both branches (modified according to
8240 result of the condition).
8241 -ffunction-sections
8243 -fdata-sections
8244 Place each function or data item into its own section in the output
8245 file if the target supports arbitrary sections. The name of the
8246 function or the name of the data item determines the section's name
8247 in the output file.
8248 Use these options on systems where the linker can perform
8250 optimizations to improve locality of reference in the instruction
8251 space. Most systems using the ELF object format have linkers with
8252 such optimizations. On AIX, the linker rearranges sections
8253 (CSECTs) based on the call graph. The performance impact varies.
8254 Together with a linker garbage collection (linker --gc-sections
8256 option) these options may lead to smaller statically-linked
8257 executables (after stripping).
8258 On ELF/DWARF systems these options do not degenerate the quality of
8260 the debug information. There could be issues with other object
8261 files/debug info formats.
8262 Only use these options when there are significant benefits from
8264 doing so. When you specify these options, the assembler and linker
8265 create larger object and executable files and are also slower.
8266 These options affect code generation. They prevent optimizations
8267 by the compiler and assembler using relative locations inside a
8268 translation unit since the locations are unknown until link time.
8269 An example of such an optimization is relaxing calls to short call
8270 instructions.
8271 -fbranch-target-load-optimize
8273 Perform branch target register load optimization before prologue /
8274 epilogue threading. The use of target registers can typically be
8275 exposed only during reload, thus hoisting loads out of loops and
8276 doing inter-block scheduling needs a separate optimization pass.
8277 -fbranch-target-load-optimize2
8279 Perform branch target register load optimization after prologue /
8280 epilogue threading.
8281 -fbtr-bb-exclusive
8283 When performing branch target register load optimization, don't
8284 reuse branch target registers within any basic block.
8285 -fstdarg-opt
8287 Optimize the prologue of variadic argument functions with respect
8288 to usage of those arguments.
8289 -fsection-anchors
8291 Try to reduce the number of symbolic address calculations by using
8292 shared "anchor" symbols to address nearby objects. This
8293 transformation can help to reduce the number of GOT entries and GOT
8294 accesses on some targets.
8295 For example, the implementation of the following function "foo":
8297 static int a, b, c;
8299 int foo (void) { return a + b + c; }
8300 usually calculates the addresses of all three variables, but if you
8302 compile it with -fsection-anchors, it accesses the variables from a
8303 common anchor point instead. The effect is similar to the
8304 following pseudocode (which isn't valid C):
8305 int foo (void)
8307 {
8308 register int *xr = &x;
8309 return xr[&a - &x] + xr[&b - &x] + xr[&c - &x];
8310 }
8311 Not all targets support this option.
8313 --param name=value
8315 In some places, GCC uses various constants to control the amount of
8316 optimization that is done. For example, GCC does not inline
8317 functions that contain more than a certain number of instructions.
8318 You can control some of these constants on the command line using
8319 the --param option.
8320 The names of specific parameters, and the meaning of the values,
8322 are tied to the internals of the compiler, and are subject to
8323 change without notice in future releases.
8324 In each case, the value is an integer. The allowable choices for
8326 name are:
8327 predictable-branch-outcome
8329 When branch is predicted to be taken with probability lower
8330 than this threshold (in percent), then it is considered well
8331 predictable. The default is 10.
8332 max-rtl-if-conversion-insns
8334 RTL if-conversion tries to remove conditional branches around a
8335 block and replace them with conditionally executed
8336 instructions. This parameter gives the maximum number of
8337 instructions in a block which should be considered for if-
8338 conversion. The default is 10, though the compiler will also
8339 use other heuristics to decide whether if-conversion is likely
8340 to be profitable.
8341 max-rtl-if-conversion-predictable-cost
8343 max-rtl-if-conversion-unpredictable-cost
8344 RTL if-conversion will try to remove conditional branches
8345 around a block and replace them with conditionally executed
8346 instructions. These parameters give the maximum permissible
8347 cost for the sequence that would be generated by if-conversion
8348 depending on whether the branch is statically determined to be
8349 predictable or not. The units for this parameter are the same
8350 as those for the GCC internal seq_cost metric. The compiler
8351 will try to provide a reasonable default for this parameter
8352 using the BRANCH_COST target macro.
8353 max-crossjump-edges
8355 The maximum number of incoming edges to consider for cross-
8356 jumping. The algorithm used by -fcrossjumping is O(N^2) in the
8357 number of edges incoming to each block. Increasing values mean
8358 more aggressive optimization, making the compilation time
8359 increase with probably small improvement in executable size.
8360 min-crossjump-insns
8362 The minimum number of instructions that must be matched at the
8363 end of two blocks before cross-jumping is performed on them.
8364 This value is ignored in the case where all instructions in the
8365 block being cross-jumped from are matched. The default value
8366 is 5.
8367 max-grow-copy-bb-insns
8369 The maximum code size expansion factor when copying basic
8370 blocks instead of jumping. The expansion is relative to a jump
8371 instruction. The default value is 8.
8372 max-goto-duplication-insns
8374 The maximum number of instructions to duplicate to a block that
8375 jumps to a computed goto. To avoid O(N^2) behavior in a number
8376 of passes, GCC factors computed gotos early in the compilation
8377 process, and unfactors them as late as possible. Only computed
8378 jumps at the end of a basic blocks with no more than max-goto-
8379 duplication-insns are unfactored. The default value is 8.
8380 max-delay-slot-insn-search
8382 The maximum number of instructions to consider when looking for
8383 an instruction to fill a delay slot. If more than this
8384 arbitrary number of instructions are searched, the time savings
8385 from filling the delay slot are minimal, so stop searching.
8386 Increasing values mean more aggressive optimization, making the
8387 compilation time increase with probably small improvement in
8388 execution time.
8389 max-delay-slot-live-search
8391 When trying to fill delay slots, the maximum number of
8392 instructions to consider when searching for a block with valid
8393 live register information. Increasing this arbitrarily chosen
8394 value means more aggressive optimization, increasing the
8395 compilation time. This parameter should be removed when the
8396 delay slot code is rewritten to maintain the control-flow
8397 graph.
8398 max-gcse-memory
8400 The approximate maximum amount of memory that can be allocated
8401 in order to perform the global common subexpression elimination
8402 optimization. If more memory than specified is required, the
8403 optimization is not done.
8404 max-gcse-insertion-ratio
8406 If the ratio of expression insertions to deletions is larger
8407 than this value for any expression, then RTL PRE inserts or
8408 removes the expression and thus leaves partially redundant
8409 computations in the instruction stream. The default value is
8410 20.
8411 max-pending-list-length
8413 The maximum number of pending dependencies scheduling allows
8414 before flushing the current state and starting over. Large
8415 functions with few branches or calls can create excessively
8416 large lists which needlessly consume memory and resources.
8417 max-modulo-backtrack-attempts
8419 The maximum number of backtrack attempts the scheduler should
8420 make when modulo scheduling a loop. Larger values can
8421 exponentially increase compilation time.
8422 max-inline-insns-single
8424 Several parameters control the tree inliner used in GCC. This
8425 number sets the maximum number of instructions (counted in
8426 GCC's internal representation) in a single function that the
8427 tree inliner considers for inlining. This only affects
8428 functions declared inline and methods implemented in a class
8429 declaration (C++). The default value is 400.
8430 max-inline-insns-auto
8432 When you use -finline-functions (included in -O3), a lot of
8433 functions that would otherwise not be considered for inlining
8434 by the compiler are investigated. To those functions, a
8435 different (more restrictive) limit compared to functions
8436 declared inline can be applied. The default value is 30.
8437 inline-min-speedup
8439 When estimated performance improvement of caller + callee
8440 runtime exceeds this threshold (in percent), the function can
8441 be inlined regardless of the limit on --param max-inline-insns-
8442 single and --param max-inline-insns-auto. The default value is
8443 15.
8444 large-function-insns
8446 The limit specifying really large functions. For functions
8447 larger than this limit after inlining, inlining is constrained
8448 by --param large-function-growth. This parameter is useful
8449 primarily to avoid extreme compilation time caused by non-
8450 linear algorithms used by the back end. The default value is
8451 2700.
8452 large-function-growth
8454 Specifies maximal growth of large function caused by inlining
8455 in percents. The default value is 100 which limits large
8456 function growth to 2.0 times the original size.
8457 large-unit-insns
8459 The limit specifying large translation unit. Growth caused by
8460 inlining of units larger than this limit is limited by --param
8461 inline-unit-growth. For small units this might be too tight.
8462 For example, consider a unit consisting of function A that is
8463 inline and B that just calls A three times. If B is small
8464 relative to A, the growth of unit is 300\% and yet such
8465 inlining is very sane. For very large units consisting of
8466 small inlineable functions, however, the overall unit growth
8467 limit is needed to avoid exponential explosion of code size.
8468 Thus for smaller units, the size is increased to --param large-
8469 unit-insns before applying --param inline-unit-growth. The
8470 default is 10000.
8471 inline-unit-growth
8473 Specifies maximal overall growth of the compilation unit caused
8474 by inlining. The default value is 20 which limits unit growth
8475 to 1.2 times the original size. Cold functions (either marked
8476 cold via an attribute or by profile feedback) are not accounted
8477 into the unit size.
8478 ipcp-unit-growth
8480 Specifies maximal overall growth of the compilation unit caused
8481 by interprocedural constant propagation. The default value is
8482 10 which limits unit growth to 1.1 times the original size.
8483 large-stack-frame
8485 The limit specifying large stack frames. While inlining the
8486 algorithm is trying to not grow past this limit too much. The
8487 default value is 256 bytes.
8488 large-stack-frame-growth
8490 Specifies maximal growth of large stack frames caused by
8491 inlining in percents. The default value is 1000 which limits
8492 large stack frame growth to 11 times the original size.
8493 max-inline-insns-recursive
8495 max-inline-insns-recursive-auto
8496 Specifies the maximum number of instructions an out-of-line
8497 copy of a self-recursive inline function can grow into by
8498 performing recursive inlining.
8499 --param max-inline-insns-recursive applies to functions
8501 declared inline. For functions not declared inline, recursive
8502 inlining happens only when -finline-functions (included in -O3)
8503 is enabled; --param max-inline-insns-recursive-auto applies
8504 instead. The default value is 450.
8505 max-inline-recursive-depth
8507 max-inline-recursive-depth-auto
8508 Specifies the maximum recursion depth used for recursive
8509 inlining.
8510 --param max-inline-recursive-depth applies to functions
8512 declared inline. For functions not declared inline, recursive
8513 inlining happens only when -finline-functions (included in -O3)
8514 is enabled; --param max-inline-recursive-depth-auto applies
8515 instead. The default value is 8.
8516 min-inline-recursive-probability
8518 Recursive inlining is profitable only for function having deep
8519 recursion in average and can hurt for function having little
8520 recursion depth by increasing the prologue size or complexity
8521 of function body to other optimizers.
8522 When profile feedback is available (see -fprofile-generate) the
8524 actual recursion depth can be guessed from the probability that
8525 function recurses via a given call expression. This parameter
8526 limits inlining only to call expressions whose probability
8527 exceeds the given threshold (in percents). The default value
8528 is 10.
8529 early-inlining-insns
8531 Specify growth that the early inliner can make. In effect it
8532 increases the amount of inlining for code having a large
8533 abstraction penalty. The default value is 14.
8534 max-early-inliner-iterations
8536 Limit of iterations of the early inliner. This basically
8537 bounds the number of nested indirect calls the early inliner
8538 can resolve. Deeper chains are still handled by late inlining.
8539 comdat-sharing-probability
8541 Probability (in percent) that C++ inline function with comdat
8542 visibility are shared across multiple compilation units. The
8543 default value is 20.
8544 profile-func-internal-id
8546 A parameter to control whether to use function internal id in
8547 profile database lookup. If the value is 0, the compiler uses
8548 an id that is based on function assembler name and filename,
8549 which makes old profile data more tolerant to source changes
8550 such as function reordering etc. The default value is 0.
8551 min-vect-loop-bound
8553 The minimum number of iterations under which loops are not
8554 vectorized when -ftree-vectorize is used. The number of
8555 iterations after vectorization needs to be greater than the
8556 value specified by this option to allow vectorization. The
8557 default value is 0.
8558 gcse-cost-distance-ratio
8560 Scaling factor in calculation of maximum distance an expression
8561 can be moved by GCSE optimizations. This is currently
8562 supported only in the code hoisting pass. The bigger the
8563 ratio, the more aggressive code hoisting is with simple
8564 expressions, i.e., the expressions that have cost less than
8565 gcse-unrestricted-cost. Specifying 0 disables hoisting of
8566 simple expressions. The default value is 10.
8567 gcse-unrestricted-cost
8569 Cost, roughly measured as the cost of a single typical machine
8570 instruction, at which GCSE optimizations do not constrain the
8571 distance an expression can travel. This is currently supported
8572 only in the code hoisting pass. The lesser the cost, the more
8573 aggressive code hoisting is. Specifying 0 allows all
8574 expressions to travel unrestricted distances. The default
8575 value is 3.
8576 max-hoist-depth
8578 The depth of search in the dominator tree for expressions to
8579 hoist. This is used to avoid quadratic behavior in hoisting
8580 algorithm. The value of 0 does not limit on the search, but
8581 may slow down compilation of huge functions. The default value
8582 is 30.
8583 max-tail-merge-comparisons
8585 The maximum amount of similar bbs to compare a bb with. This
8586 is used to avoid quadratic behavior in tree tail merging. The
8587 default value is 10.
8588 max-tail-merge-iterations
8590 The maximum amount of iterations of the pass over the function.
8591 This is used to limit compilation time in tree tail merging.
8592 The default value is 2.
8593 store-merging-allow-unaligned
8595 Allow the store merging pass to introduce unaligned stores if
8596 it is legal to do so. The default value is 1.
8597 max-stores-to-merge
8599 The maximum number of stores to attempt to merge into wider
8600 stores in the store merging pass. The minimum value is 2 and
8601 the default is 64.
8602 max-unrolled-insns
8604 The maximum number of instructions that a loop may have to be
8605 unrolled. If a loop is unrolled, this parameter also
8606 determines how many times the loop code is unrolled.
8607 max-average-unrolled-insns
8609 The maximum number of instructions biased by probabilities of
8610 their execution that a loop may have to be unrolled. If a loop
8611 is unrolled, this parameter also determines how many times the
8612 loop code is unrolled.
8613 max-unroll-times
8615 The maximum number of unrollings of a single loop.
8616 max-peeled-insns
8618 The maximum number of instructions that a loop may have to be
8619 peeled. If a loop is peeled, this parameter also determines
8620 how many times the loop code is peeled.
8621 max-peel-times
8623 The maximum number of peelings of a single loop.
8624 max-peel-branches
8626 The maximum number of branches on the hot path through the
8627 peeled sequence.
8628 max-completely-peeled-insns
8630 The maximum number of insns of a completely peeled loop.
8631 max-completely-peel-times
8633 The maximum number of iterations of a loop to be suitable for
8634 complete peeling.
8635 max-completely-peel-loop-nest-depth
8637 The maximum depth of a loop nest suitable for complete peeling.
8638 max-unswitch-insns
8640 The maximum number of insns of an unswitched loop.
8641 max-unswitch-level
8643 The maximum number of branches unswitched in a single loop.
8644 max-loop-headers-insns
8646 The maximum number of insns in loop header duplicated by the
8647 copy loop headers pass.
8648 lim-expensive
8650 The minimum cost of an expensive expression in the loop
8651 invariant motion.
8652 iv-consider-all-candidates-bound
8654 Bound on number of candidates for induction variables, below
8655 which all candidates are considered for each use in induction
8656 variable optimizations. If there are more candidates than
8657 this, only the most relevant ones are considered to avoid
8658 quadratic time complexity.
8659 iv-max-considered-uses
8661 The induction variable optimizations give up on loops that
8662 contain more induction variable uses.
8663 iv-always-prune-cand-set-bound
8665 If the number of candidates in the set is smaller than this
8666 value, always try to remove unnecessary ivs from the set when
8667 adding a new one.
8668 avg-loop-niter
8670 Average number of iterations of a loop.
8671 dse-max-object-size
8673 Maximum size (in bytes) of objects tracked bytewise by dead
8674 store elimination. Larger values may result in larger
8675 compilation times.
8676 scev-max-expr-size
8678 Bound on size of expressions used in the scalar evolutions
8679 analyzer. Large expressions slow the analyzer.
8680 scev-max-expr-complexity
8682 Bound on the complexity of the expressions in the scalar
8683 evolutions analyzer. Complex expressions slow the analyzer.
8684 max-tree-if-conversion-phi-args
8686 Maximum number of arguments in a PHI supported by TREE if
8687 conversion unless the loop is marked with simd pragma.
8688 vect-max-version-for-alignment-checks
8690 The maximum number of run-time checks that can be performed
8691 when doing loop versioning for alignment in the vectorizer.
8692 vect-max-version-for-alias-checks
8694 The maximum number of run-time checks that can be performed
8695 when doing loop versioning for alias in the vectorizer.
8696 vect-max-peeling-for-alignment
8698 The maximum number of loop peels to enhance access alignment
8699 for vectorizer. Value -1 means no limit.
8700 max-iterations-to-track
8702 The maximum number of iterations of a loop the brute-force
8703 algorithm for analysis of the number of iterations of the loop
8704 tries to evaluate.
8705 hot-bb-count-ws-permille
8707 A basic block profile count is considered hot if it contributes
8708 to the given permillage (i.e. 0...1000) of the entire profiled
8709 execution.
8710 hot-bb-frequency-fraction
8712 Select fraction of the entry block frequency of executions of
8713 basic block in function given basic block needs to have to be
8714 considered hot.
8715 max-predicted-iterations
8717 The maximum number of loop iterations we predict statically.
8718 This is useful in cases where a function contains a single loop
8719 with known bound and another loop with unknown bound. The
8720 known number of iterations is predicted correctly, while the
8721 unknown number of iterations average to roughly 10. This means
8722 that the loop without bounds appears artificially cold relative
8723 to the other one.
8724 builtin-expect-probability
8726 Control the probability of the expression having the specified
8727 value. This parameter takes a percentage (i.e. 0 ... 100) as
8728 input. The default probability of 90 is obtained empirically.
8729 align-threshold
8731 Select fraction of the maximal frequency of executions of a
8732 basic block in a function to align the basic block.
8733 align-loop-iterations
8735 A loop expected to iterate at least the selected number of
8736 iterations is aligned.
8737 tracer-dynamic-coverage
8739 tracer-dynamic-coverage-feedback
8740 This value is used to limit superblock formation once the given
8741 percentage of executed instructions is covered. This limits
8742 unnecessary code size expansion.
8743 The tracer-dynamic-coverage-feedback parameter is used only
8745 when profile feedback is available. The real profiles (as
8746 opposed to statically estimated ones) are much less balanced
8747 allowing the threshold to be larger value.
8748 tracer-max-code-growth
8750 Stop tail duplication once code growth has reached given
8751 percentage. This is a rather artificial limit, as most of the
8752 duplicates are eliminated later in cross jumping, so it may be
8753 set to much higher values than is the desired code growth.
8754 tracer-min-branch-ratio
8756 Stop reverse growth when the reverse probability of best edge
8757 is less than this threshold (in percent).
8758 tracer-min-branch-probability
8760 tracer-min-branch-probability-feedback
8761 Stop forward growth if the best edge has probability lower than
8762 this threshold.
8763 Similarly to tracer-dynamic-coverage two parameters are
8765 provided. tracer-min-branch-probability-feedback is used for
8766 compilation with profile feedback and tracer-min-branch-
8767 probability compilation without. The value for compilation
8768 with profile feedback needs to be more conservative (higher) in
8769 order to make tracer effective.
8770 stack-clash-protection-guard-size
8772 Specify the size of the operating system provided stack guard
8773 as 2 raised to num bytes. The default value is 12 (4096
8774 bytes). Acceptable values are between 12 and 30. Higher
8775 values may reduce the number of explicit probes, but a value
8776 larger than the operating system provided guard will leave code
8777 vulnerable to stack clash style attacks.
8778 stack-clash-protection-probe-interval
8780 Stack clash protection involves probing stack space as it is
8781 allocated. This param controls the maximum distance between
8782 probes into the stack as 2 raised to num bytes. Acceptable
8783 values are between 10 and 16 and defaults to 12. Higher values
8784 may reduce the number of explicit probes, but a value larger
8785 than the operating system provided guard will leave code
8786 vulnerable to stack clash style attacks.
8787 max-cse-path-length
8789 The maximum number of basic blocks on path that CSE considers.
8790 The default is 10.
8791 max-cse-insns
8793 The maximum number of instructions CSE processes before
8794 flushing. The default is 1000.
8795 ggc-min-expand
8797 GCC uses a garbage collector to manage its own memory
8798 allocation. This parameter specifies the minimum percentage by
8799 which the garbage collector's heap should be allowed to expand
8800 between collections. Tuning this may improve compilation
8801 speed; it has no effect on code generation.
8802 The default is 30% + 70% * (RAM/1GB) with an upper bound of
8804 100% when RAM >= 1GB. If "getrlimit" is available, the notion
8805 of "RAM" is the smallest of actual RAM and "RLIMIT_DATA" or
8806 "RLIMIT_AS". If GCC is not able to calculate RAM on a
8807 particular platform, the lower bound of 30% is used. Setting
8808 this parameter and ggc-min-heapsize to zero causes a full
8809 collection to occur at every opportunity. This is extremely
8810 slow, but can be useful for debugging.
8811 ggc-min-heapsize
8813 Minimum size of the garbage collector's heap before it begins
8814 bothering to collect garbage. The first collection occurs
8815 after the heap expands by ggc-min-expand% beyond ggc-min-
8816 heapsize. Again, tuning this may improve compilation speed,
8817 and has no effect on code generation.
8818 The default is the smaller of RAM/8, RLIMIT_RSS, or a limit
8820 that tries to ensure that RLIMIT_DATA or RLIMIT_AS are not
8821 exceeded, but with a lower bound of 4096 (four megabytes) and
8822 an upper bound of 131072 (128 megabytes). If GCC is not able
8823 to calculate RAM on a particular platform, the lower bound is
8824 used. Setting this parameter very large effectively disables
8825 garbage collection. Setting this parameter and ggc-min-expand
8826 to zero causes a full collection to occur at every opportunity.
8827 max-reload-search-insns
8829 The maximum number of instruction reload should look backward
8830 for equivalent register. Increasing values mean more
8831 aggressive optimization, making the compilation time increase
8832 with probably slightly better performance. The default value
8833 is 100.
8834 max-cselib-memory-locations
8836 The maximum number of memory locations cselib should take into
8837 account. Increasing values mean more aggressive optimization,
8838 making the compilation time increase with probably slightly
8839 better performance. The default value is 500.
8840 max-sched-ready-insns
8842 The maximum number of instructions ready to be issued the
8843 scheduler should consider at any given time during the first
8844 scheduling pass. Increasing values mean more thorough
8845 searches, making the compilation time increase with probably
8846 little benefit. The default value is 100.
8847 max-sched-region-blocks
8849 The maximum number of blocks in a region to be considered for
8850 interblock scheduling. The default value is 10.
8851 max-pipeline-region-blocks
8853 The maximum number of blocks in a region to be considered for
8854 pipelining in the selective scheduler. The default value is
8855 15.
8856 max-sched-region-insns
8858 The maximum number of insns in a region to be considered for
8859 interblock scheduling. The default value is 100.
8860 max-pipeline-region-insns
8862 The maximum number of insns in a region to be considered for
8863 pipelining in the selective scheduler. The default value is
8864 200.
8865 min-spec-prob
8867 The minimum probability (in percents) of reaching a source
8868 block for interblock speculative scheduling. The default value
8869 is 40.
8870 max-sched-extend-regions-iters
8872 The maximum number of iterations through CFG to extend regions.
8873 A value of 0 (the default) disables region extensions.
8874 max-sched-insn-conflict-delay
8876 The maximum conflict delay for an insn to be considered for
8877 speculative motion. The default value is 3.
8878 sched-spec-prob-cutoff
8880 The minimal probability of speculation success (in percents),
8881 so that speculative insns are scheduled. The default value is
8882 40.
8883 sched-state-edge-prob-cutoff
8885 The minimum probability an edge must have for the scheduler to
8886 save its state across it. The default value is 10.
8887 sched-mem-true-dep-cost
8889 Minimal distance (in CPU cycles) between store and load
8890 targeting same memory locations. The default value is 1.
8891 selsched-max-lookahead
8893 The maximum size of the lookahead window of selective
8894 scheduling. It is a depth of search for available
8895 instructions. The default value is 50.
8896 selsched-max-sched-times
8898 The maximum number of times that an instruction is scheduled
8899 during selective scheduling. This is the limit on the number
8900 of iterations through which the instruction may be pipelined.
8901 The default value is 2.
8902 selsched-insns-to-rename
8904 The maximum number of best instructions in the ready list that
8905 are considered for renaming in the selective scheduler. The
8906 default value is 2.
8907 sms-min-sc
8909 The minimum value of stage count that swing modulo scheduler
8910 generates. The default value is 2.
8911 max-last-value-rtl
8913 The maximum size measured as number of RTLs that can be
8914 recorded in an expression in combiner for a pseudo register as
8915 last known value of that register. The default is 10000.
8916 max-combine-insns
8918 The maximum number of instructions the RTL combiner tries to
8919 combine. The default value is 2 at -Og and 4 otherwise.
8920 integer-share-limit
8922 Small integer constants can use a shared data structure,
8923 reducing the compiler's memory usage and increasing its speed.
8924 This sets the maximum value of a shared integer constant. The
8925 default value is 256.
8926 ssp-buffer-size
8928 The minimum size of buffers (i.e. arrays) that receive stack
8929 smashing protection when -fstack-protection is used.
8930 min-size-for-stack-sharing
8932 The minimum size of variables taking part in stack slot sharing
8933 when not optimizing. The default value is 32.
8934 max-jump-thread-duplication-stmts
8936 Maximum number of statements allowed in a block that needs to
8937 be duplicated when threading jumps.
8938 max-fields-for-field-sensitive
8940 Maximum number of fields in a structure treated in a field
8941 sensitive manner during pointer analysis. The default is zero
8942 for -O0 and -O1, and 100 for -Os, -O2, and -O3.
8943 prefetch-latency
8945 Estimate on average number of instructions that are executed
8946 before prefetch finishes. The distance prefetched ahead is
8947 proportional to this constant. Increasing this number may also
8948 lead to less streams being prefetched (see simultaneous-
8949 prefetches).
8950 simultaneous-prefetches
8952 Maximum number of prefetches that can run at the same time.
8953 l1-cache-line-size
8955 The size of cache line in L1 cache, in bytes.
8956 l1-cache-size
8958 The size of L1 cache, in kilobytes.
8959 l2-cache-size
8961 The size of L2 cache, in kilobytes.
8962 loop-interchange-max-num-stmts
8964 The maximum number of stmts in a loop to be interchanged.
8965 loop-interchange-stride-ratio
8967 The minimum ratio between stride of two loops for interchange
8968 to be profitable.
8969 min-insn-to-prefetch-ratio
8971 The minimum ratio between the number of instructions and the
8972 number of prefetches to enable prefetching in a loop.
8973 prefetch-min-insn-to-mem-ratio
8975 The minimum ratio between the number of instructions and the
8976 number of memory references to enable prefetching in a loop.
8977 use-canonical-types
8979 Whether the compiler should use the "canonical" type system.
8980 By default, this should always be 1, which uses a more
8981 efficient internal mechanism for comparing types in C++ and
8982 Objective-C++. However, if bugs in the canonical type system
8983 are causing compilation failures, set this value to 0 to
8984 disable canonical types.
8985 switch-conversion-max-branch-ratio
8987 Switch initialization conversion refuses to create arrays that
8988 are bigger than switch-conversion-max-branch-ratio times the
8989 number of branches in the switch.
8990 max-partial-antic-length
8992 Maximum length of the partial antic set computed during the
8993 tree partial redundancy elimination optimization (-ftree-pre)
8994 when optimizing at -O3 and above. For some sorts of source
8995 code the enhanced partial redundancy elimination optimization
8996 can run away, consuming all of the memory available on the host
8997 machine. This parameter sets a limit on the length of the sets
8998 that are computed, which prevents the runaway behavior.
8999 Setting a value of 0 for this parameter allows an unlimited set
9000 length.
9001 sccvn-max-scc-size
9003 Maximum size of a strongly connected component (SCC) during
9004 SCCVN processing. If this limit is hit, SCCVN processing for
9005 the whole function is not done and optimizations depending on
9006 it are disabled. The default maximum SCC size is 10000.
9007 sccvn-max-alias-queries-per-access
9009 Maximum number of alias-oracle queries we perform when looking
9010 for redundancies for loads and stores. If this limit is hit
9011 the search is aborted and the load or store is not considered
9012 redundant. The number of queries is algorithmically limited to
9013 the number of stores on all paths from the load to the function
9014 entry. The default maximum number of queries is 1000.
9015 ira-max-loops-num
9017 IRA uses regional register allocation by default. If a
9018 function contains more loops than the number given by this
9019 parameter, only at most the given number of the most
9020 frequently-executed loops form regions for regional register
9021 allocation. The default value of the parameter is 100.
9022 ira-max-conflict-table-size
9024 Although IRA uses a sophisticated algorithm to compress the
9025 conflict table, the table can still require excessive amounts
9026 of memory for huge functions. If the conflict table for a
9027 function could be more than the size in MB given by this
9028 parameter, the register allocator instead uses a faster,
9029 simpler, and lower-quality algorithm that does not require
9030 building a pseudo-register conflict table. The default value
9031 of the parameter is 2000.
9032 ira-loop-reserved-regs
9034 IRA can be used to evaluate more accurate register pressure in
9035 loops for decisions to move loop invariants (see -O3). The
9036 number of available registers reserved for some other purposes
9037 is given by this parameter. The default value of the parameter
9038 is 2, which is the minimal number of registers needed by
9039 typical instructions. This value is the best found from
9040 numerous experiments.
9041 lra-inheritance-ebb-probability-cutoff
9043 LRA tries to reuse values reloaded in registers in subsequent
9044 insns. This optimization is called inheritance. EBB is used
9045 as a region to do this optimization. The parameter defines a
9046 minimal fall-through edge probability in percentage used to add
9047 BB to inheritance EBB in LRA. The default value of the
9048 parameter is 40. The value was chosen from numerous runs of
9049 SPEC2000 on x86-64.
9050 loop-invariant-max-bbs-in-loop
9052 Loop invariant motion can be very expensive, both in
9053 compilation time and in amount of needed compile-time memory,
9054 with very large loops. Loops with more basic blocks than this
9055 parameter won't have loop invariant motion optimization
9056 performed on them. The default value of the parameter is 1000
9057 for -O1 and 10000 for -O2 and above.
9058 loop-max-datarefs-for-datadeps
9060 Building data dependencies is expensive for very large loops.
9061 This parameter limits the number of data references in loops
9062 that are considered for data dependence analysis. These large
9063 loops are no handled by the optimizations using loop data
9064 dependencies. The default value is 1000.
9065 max-vartrack-size
9067 Sets a maximum number of hash table slots to use during
9068 variable tracking dataflow analysis of any function. If this
9069 limit is exceeded with variable tracking at assignments
9070 enabled, analysis for that function is retried without it,
9071 after removing all debug insns from the function. If the limit
9072 is exceeded even without debug insns, var tracking analysis is
9073 completely disabled for the function. Setting the parameter to
9074 zero makes it unlimited.
9075 max-vartrack-expr-depth
9077 Sets a maximum number of recursion levels when attempting to
9078 map variable names or debug temporaries to value expressions.
9079 This trades compilation time for more complete debug
9080 information. If this is set too low, value expressions that
9081 are available and could be represented in debug information may
9082 end up not being used; setting this higher may enable the
9083 compiler to find more complex debug expressions, but compile
9084 time and memory use may grow. The default is 12.
9085 max-debug-marker-count
9087 Sets a threshold on the number of debug markers (e.g. begin
9088 stmt markers) to avoid complexity explosion at inlining or
9089 expanding to RTL. If a function has more such gimple stmts
9090 than the set limit, such stmts will be dropped from the inlined
9091 copy of a function, and from its RTL expansion. The default is
9092 100000.
9093 min-nondebug-insn-uid
9095 Use uids starting at this parameter for nondebug insns. The
9096 range below the parameter is reserved exclusively for debug
9097 insns created by -fvar-tracking-assignments, but debug insns
9098 may get (non-overlapping) uids above it if the reserved range
9099 is exhausted.
9100 ipa-sra-ptr-growth-factor
9102 IPA-SRA replaces a pointer to an aggregate with one or more new
9103 parameters only when their cumulative size is less or equal to
9104 ipa-sra-ptr-growth-factor times the size of the original
9105 pointer parameter.
9106 sra-max-scalarization-size-Ospeed
9108 sra-max-scalarization-size-Osize
9109 The two Scalar Reduction of Aggregates passes (SRA and IPA-SRA)
9110 aim to replace scalar parts of aggregates with uses of
9111 independent scalar variables. These parameters control the
9112 maximum size, in storage units, of aggregate which is
9113 considered for replacement when compiling for speed (sra-max-
9114 scalarization-size-Ospeed) or size (sra-max-scalarization-size-
9115 Osize) respectively.
9116 tm-max-aggregate-size
9118 When making copies of thread-local variables in a transaction,
9119 this parameter specifies the size in bytes after which
9120 variables are saved with the logging functions as opposed to
9121 save/restore code sequence pairs. This option only applies
9122 when using -fgnu-tm.
9123 graphite-max-nb-scop-params
9125 To avoid exponential effects in the Graphite loop transforms,
9126 the number of parameters in a Static Control Part (SCoP) is
9127 bounded. The default value is 10 parameters, a value of zero
9128 can be used to lift the bound. A variable whose value is
9129 unknown at compilation time and defined outside a SCoP is a
9130 parameter of the SCoP.
9131 loop-block-tile-size
9133 Loop blocking or strip mining transforms, enabled with
9134 -floop-block or -floop-strip-mine, strip mine each loop in the
9135 loop nest by a given number of iterations. The strip length
9136 can be changed using the loop-block-tile-size parameter. The
9137 default value is 51 iterations.
9138 loop-unroll-jam-size
9140 Specify the unroll factor for the -floop-unroll-and-jam option.
9141 The default value is 4.
9142 loop-unroll-jam-depth
9144 Specify the dimension to be unrolled (counting from the most
9145 inner loop) for the -floop-unroll-and-jam. The default value
9146 is 2.
9147 ipa-cp-value-list-size
9149 IPA-CP attempts to track all possible values and types passed
9150 to a function's parameter in order to propagate them and
9151 perform devirtualization. ipa-cp-value-list-size is the
9152 maximum number of values and types it stores per one formal
9153 parameter of a function.
9154 ipa-cp-eval-threshold
9156 IPA-CP calculates its own score of cloning profitability
9157 heuristics and performs those cloning opportunities with scores
9158 that exceed ipa-cp-eval-threshold.
9159 ipa-cp-recursion-penalty
9161 Percentage penalty the recursive functions will receive when
9162 they are evaluated for cloning.
9163 ipa-cp-single-call-penalty
9165 Percentage penalty functions containing a single call to
9166 another function will receive when they are evaluated for
9167 cloning.
9168 ipa-max-agg-items
9170 IPA-CP is also capable to propagate a number of scalar values
9171 passed in an aggregate. ipa-max-agg-items controls the maximum
9172 number of such values per one parameter.
9173 ipa-cp-loop-hint-bonus
9175 When IPA-CP determines that a cloning candidate would make the
9176 number of iterations of a loop known, it adds a bonus of ipa-
9177 cp-loop-hint-bonus to the profitability score of the candidate.
9178 ipa-cp-array-index-hint-bonus
9180 When IPA-CP determines that a cloning candidate would make the
9181 index of an array access known, it adds a bonus of ipa-cp-
9182 array-index-hint-bonus to the profitability score of the
9183 candidate.
9184 ipa-max-aa-steps
9186 During its analysis of function bodies, IPA-CP employs alias
9187 analysis in order to track values pointed to by function
9188 parameters. In order not spend too much time analyzing huge
9189 functions, it gives up and consider all memory clobbered after
9190 examining ipa-max-aa-steps statements modifying memory.
9191 lto-partitions
9193 Specify desired number of partitions produced during WHOPR
9194 compilation. The number of partitions should exceed the number
9195 of CPUs used for compilation. The default value is 32.
9196 lto-min-partition
9198 Size of minimal partition for WHOPR (in estimated
9199 instructions). This prevents expenses of splitting very small
9200 programs into too many partitions.
9201 lto-max-partition
9203 Size of max partition for WHOPR (in estimated instructions).
9204 to provide an upper bound for individual size of partition.
9205 Meant to be used only with balanced partitioning.
9206 cxx-max-namespaces-for-diagnostic-help
9208 The maximum number of namespaces to consult for suggestions
9209 when C++ name lookup fails for an identifier. The default is
9210 1000.
9211 sink-frequency-threshold
9213 The maximum relative execution frequency (in percents) of the
9214 target block relative to a statement's original block to allow
9215 statement sinking of a statement. Larger numbers result in
9216 more aggressive statement sinking. The default value is 75. A
9217 small positive adjustment is applied for statements with memory
9218 operands as those are even more profitable so sink.
9219 max-stores-to-sink
9221 The maximum number of conditional store pairs that can be sunk.
9222 Set to 0 if either vectorization (-ftree-vectorize) or if-
9223 conversion (-ftree-loop-if-convert) is disabled. The default
9224 is 2.
9225 allow-store-data-races
9227 Allow optimizers to introduce new data races on stores. Set to
9228 1 to allow, otherwise to 0. This option is enabled by default
9229 at optimization level -Ofast.
9230 case-values-threshold
9232 The smallest number of different values for which it is best to
9233 use a jump-table instead of a tree of conditional branches. If
9234 the value is 0, use the default for the machine. The default
9235 is 0.
9236 tree-reassoc-width
9238 Set the maximum number of instructions executed in parallel in
9239 reassociated tree. This parameter overrides target dependent
9240 heuristics used by default if has non zero value.
9241 sched-pressure-algorithm
9243 Choose between the two available implementations of
9244 -fsched-pressure. Algorithm 1 is the original implementation
9245 and is the more likely to prevent instructions from being
9246 reordered. Algorithm 2 was designed to be a compromise between
9247 the relatively conservative approach taken by algorithm 1 and
9248 the rather aggressive approach taken by the default scheduler.
9249 It relies more heavily on having a regular register file and
9250 accurate register pressure classes. See haifa-sched.c in the
9251 GCC sources for more details.
9252 The default choice depends on the target.
9254 max-slsr-cand-scan
9256 Set the maximum number of existing candidates that are
9257 considered when seeking a basis for a new straight-line
9258 strength reduction candidate.
9259 asan-globals
9261 Enable buffer overflow detection for global objects. This kind
9262 of protection is enabled by default if you are using
9263 -fsanitize=address option. To disable global objects
9264 protection use --param asan-globals=0.
9265 asan-stack
9267 Enable buffer overflow detection for stack objects. This kind
9268 of protection is enabled by default when using
9269 -fsanitize=address. To disable stack protection use --param
9270 asan-stack=0 option.
9271 asan-instrument-reads
9273 Enable buffer overflow detection for memory reads. This kind
9274 of protection is enabled by default when using
9275 -fsanitize=address. To disable memory reads protection use
9276 --param asan-instrument-reads=0.
9277 asan-instrument-writes
9279 Enable buffer overflow detection for memory writes. This kind
9280 of protection is enabled by default when using
9281 -fsanitize=address. To disable memory writes protection use
9282 --param asan-instrument-writes=0 option.
9283 asan-memintrin
9285 Enable detection for built-in functions. This kind of
9286 protection is enabled by default when using -fsanitize=address.
9287 To disable built-in functions protection use --param
9288 asan-memintrin=0.
9289 asan-use-after-return
9291 Enable detection of use-after-return. This kind of protection
9292 is enabled by default when using the -fsanitize=address option.
9293 To disable it use --param asan-use-after-return=0.
9294 Note: By default the check is disabled at run time. To enable
9296 it, add "detect_stack_use_after_return=1" to the environment
9297 variable ASAN_OPTIONS.
9298 asan-instrumentation-with-call-threshold
9300 If number of memory accesses in function being instrumented is
9301 greater or equal to this number, use callbacks instead of
9302 inline checks. E.g. to disable inline code use --param
9303 asan-instrumentation-with-call-threshold=0.
9304 use-after-scope-direct-emission-threshold
9306 If the size of a local variable in bytes is smaller or equal to
9307 this number, directly poison (or unpoison) shadow memory
9308 instead of using run-time callbacks. The default value is 256.
9309 chkp-max-ctor-size
9311 Static constructors generated by Pointer Bounds Checker may
9312 become very large and significantly increase compile time at
9313 optimization level -O1 and higher. This parameter is a maximum
9314 number of statements in a single generated constructor.
9315 Default value is 5000.
9316 max-fsm-thread-path-insns
9318 Maximum number of instructions to copy when duplicating blocks
9319 on a finite state automaton jump thread path. The default is
9320 100.
9321 max-fsm-thread-length
9323 Maximum number of basic blocks on a finite state automaton jump
9324 thread path. The default is 10.
9325 max-fsm-thread-paths
9327 Maximum number of new jump thread paths to create for a finite
9328 state automaton. The default is 50.
9329 parloops-chunk-size
9331 Chunk size of omp schedule for loops parallelized by parloops.
9332 The default is 0.
9333 parloops-schedule
9335 Schedule type of omp schedule for loops parallelized by
9336 parloops (static, dynamic, guided, auto, runtime). The default
9337 is static.
9338 parloops-min-per-thread
9340 The minimum number of iterations per thread of an innermost
9341 parallelized loop for which the parallelized variant is
9342 prefered over the single threaded one. The default is 100.
9343 Note that for a parallelized loop nest the minimum number of
9344 iterations of the outermost loop per thread is two.
9345 max-ssa-name-query-depth
9347 Maximum depth of recursion when querying properties of SSA
9348 names in things like fold routines. One level of recursion
9349 corresponds to following a use-def chain.
9350 hsa-gen-debug-stores
9352 Enable emission of special debug stores within HSA kernels
9353 which are then read and reported by libgomp plugin. Generation
9354 of these stores is disabled by default, use --param
9355 hsa-gen-debug-stores=1 to enable it.
9356 max-speculative-devirt-maydefs
9358 The maximum number of may-defs we analyze when looking for a
9359 must-def specifying the dynamic type of an object that invokes
9360 a virtual call we may be able to devirtualize speculatively.
9361 max-vrp-switch-assertions
9363 The maximum number of assertions to add along the default edge
9364 of a switch statement during VRP. The default is 10.
9365 unroll-jam-min-percent
9367 The minimum percentage of memory references that must be
9368 optimized away for the unroll-and-jam transformation to be
9369 considered profitable.
9370 unroll-jam-max-unroll
9372 The maximum number of times the outer loop should be unrolled
9373 by the unroll-and-jam transformation.
9374 Program Instrumentation Options
9376 GCC supports a number of command-line options that control adding run-
9377 time instrumentation to the code it normally generates. For example,
9378 one purpose of instrumentation is collect profiling statistics for use
9379 in finding program hot spots, code coverage analysis, or profile-guided
9380 optimizations. Another class of program instrumentation is adding run-
9381 time checking to detect programming errors like invalid pointer
9382 dereferences or out-of-bounds array accesses, as well as deliberately
9383 hostile attacks such as stack smashing or C++ vtable hijacking. There
9384 is also a general hook which can be used to implement other forms of
9385 tracing or function-level instrumentation for debug or program analysis
9386 purposes.
9387 -p Generate extra code to write profile information suitable for the
9389 analysis program prof. You must use this option when compiling the
9390 source files you want data about, and you must also use it when
9391 linking.
9392 -pg Generate extra code to write profile information suitable for the
9394 analysis program gprof. You must use this option when compiling
9395 the source files you want data about, and you must also use it when
9396 linking.
9397 -fprofile-arcs
9399 Add code so that program flow arcs are instrumented. During
9400 execution the program records how many times each branch and call
9401 is executed and how many times it is taken or returns. On targets
9402 that support constructors with priority support, profiling properly
9403 handles constructors, destructors and C++ constructors (and
9404 destructors) of classes which are used as a type of a global
9405 variable.
9406 When the compiled program exits it saves this data to a file called
9408 auxname.gcda for each source file. The data may be used for
9409 profile-directed optimizations (-fbranch-probabilities), or for
9410 test coverage analysis (-ftest-coverage). Each object file's
9411 auxname is generated from the name of the output file, if
9412 explicitly specified and it is not the final executable, otherwise
9413 it is the basename of the source file. In both cases any suffix is
9414 removed (e.g. foo.gcda for input file dir/foo.c, or dir/foo.gcda
9415 for output file specified as -o dir/foo.o).
9416 --coverage
9418 This option is used to compile and link code instrumented for
9419 coverage analysis. The option is a synonym for -fprofile-arcs
9420 -ftest-coverage (when compiling) and -lgcov (when linking). See
9421 the documentation for those options for more details.
9422 * Compile the source files with -fprofile-arcs plus optimization
9424 and code generation options. For test coverage analysis, use
9425 the additional -ftest-coverage option. You do not need to
9426 profile every source file in a program.
9427 * Compile the source files additionally with -fprofile-abs-path
9429 to create absolute path names in the .gcno files. This allows
9430 gcov to find the correct sources in projects where compilations
9431 occur with different working directories.
9432 * Link your object files with -lgcov or -fprofile-arcs (the
9434 latter implies the former).
9435 * Run the program on a representative workload to generate the
9437 arc profile information. This may be repeated any number of
9438 times. You can run concurrent instances of your program, and
9439 provided that the file system supports locking, the data files
9440 will be correctly updated. Unless a strict ISO C dialect
9441 option is in effect, "fork" calls are detected and correctly
9442 handled without double counting.
9443 * For profile-directed optimizations, compile the source files
9445 again with the same optimization and code generation options
9446 plus -fbranch-probabilities.
9447 * For test coverage analysis, use gcov to produce human readable
9449 information from the .gcno and .gcda files. Refer to the gcov
9450 documentation for further information.
9451 With -fprofile-arcs, for each function of your program GCC creates
9453 a program flow graph, then finds a spanning tree for the graph.
9454 Only arcs that are not on the spanning tree have to be
9455 instrumented: the compiler adds code to count the number of times
9456 that these arcs are executed. When an arc is the only exit or only
9457 entrance to a block, the instrumentation code can be added to the
9458 block; otherwise, a new basic block must be created to hold the
9459 instrumentation code.
9460 -ftest-coverage
9462 Produce a notes file that the gcov code-coverage utility can use to
9463 show program coverage. Each source file's note file is called
9464 auxname.gcno. Refer to the -fprofile-arcs option above for a
9465 description of auxname and instructions on how to generate test
9466 coverage data. Coverage data matches the source files more closely
9467 if you do not optimize.
9468 -fprofile-abs-path
9470 Automatically convert relative source file names to absolute path
9471 names in the .gcno files. This allows gcov to find the correct
9472 sources in projects where compilations occur with different working
9473 directories.
9474 -fprofile-dir=path
9476 Set the directory to search for the profile data files in to path.
9477 This option affects only the profile data generated by
9478 -fprofile-generate, -ftest-coverage, -fprofile-arcs and used by
9479 -fprofile-use and -fbranch-probabilities and its related options.
9480 Both absolute and relative paths can be used. By default, GCC uses
9481 the current directory as path, thus the profile data file appears
9482 in the same directory as the object file.
9483 -fprofile-generate
9485 -fprofile-generate=path
9486 Enable options usually used for instrumenting application to
9487 produce profile useful for later recompilation with profile
9488 feedback based optimization. You must use -fprofile-generate both
9489 when compiling and when linking your program.
9490 The following options are enabled: -fprofile-arcs,
9492 -fprofile-values, -fvpt.
9493 If path is specified, GCC looks at the path to find the profile
9495 feedback data files. See -fprofile-dir.
9496 To optimize the program based on the collected profile information,
9498 use -fprofile-use.
9499 -fprofile-update=method
9501 Alter the update method for an application instrumented for profile
9502 feedback based optimization. The method argument should be one of
9503 single, atomic or prefer-atomic. The first one is useful for
9504 single-threaded applications, while the second one prevents profile
9505 corruption by emitting thread-safe code.
9506 Warning: When an application does not properly join all threads (or
9508 creates an detached thread), a profile file can be still corrupted.
9509 Using prefer-atomic would be transformed either to atomic, when
9511 supported by a target, or to single otherwise. The GCC driver
9512 automatically selects prefer-atomic when -pthread is present in the
9513 command line.
9514 -fsanitize=address
9516 Enable AddressSanitizer, a fast memory error detector. Memory
9517 access instructions are instrumented to detect out-of-bounds and
9518 use-after-free bugs. The option enables
9519 -fsanitize-address-use-after-scope. See
9520 <https://github.com/google/sanitizers/wiki/AddressSanitizer> for
9521 more details. The run-time behavior can be influenced using the
9522 ASAN_OPTIONS environment variable. When set to "help=1", the
9523 available options are shown at startup of the instrumented program.
9524 See
9525 <https://github.com/google/sanitizers/wiki/AddressSanitizerFlags#run-time-flags>
9526 for a list of supported options. The option cannot be combined
9527 with -fsanitize=thread and/or -fcheck-pointer-bounds.
9528 -fsanitize=kernel-address
9530 Enable AddressSanitizer for Linux kernel. See
9531 <https://github.com/google/kasan/wiki> for more details. The
9532 option cannot be combined with -fcheck-pointer-bounds.
9533 -fsanitize=pointer-compare
9535 Instrument comparison operation (<, <=, >, >=) with pointer
9536 operands. The option must be combined with either
9537 -fsanitize=kernel-address or -fsanitize=address The option cannot
9538 be combined with -fsanitize=thread and/or -fcheck-pointer-bounds.
9539 Note: By default the check is disabled at run time. To enable it,
9540 add "detect_invalid_pointer_pairs=2" to the environment variable
9541 ASAN_OPTIONS. Using "detect_invalid_pointer_pairs=1" detects
9542 invalid operation only when both pointers are non-null.
9543 -fsanitize=pointer-subtract
9545 Instrument subtraction with pointer operands. The option must be
9546 combined with either -fsanitize=kernel-address or
9547 -fsanitize=address The option cannot be combined with
9548 -fsanitize=thread and/or -fcheck-pointer-bounds. Note: By default
9549 the check is disabled at run time. To enable it, add
9550 "detect_invalid_pointer_pairs=2" to the environment variable
9551 ASAN_OPTIONS. Using "detect_invalid_pointer_pairs=1" detects
9552 invalid operation only when both pointers are non-null.
9553 -fsanitize=thread
9555 Enable ThreadSanitizer, a fast data race detector. Memory access
9556 instructions are instrumented to detect data race bugs. See
9557 <https://github.com/google/sanitizers/wiki#threadsanitizer> for
9558 more details. The run-time behavior can be influenced using the
9559 TSAN_OPTIONS environment variable; see
9560 <https://github.com/google/sanitizers/wiki/ThreadSanitizerFlags>
9561 for a list of supported options. The option cannot be combined
9562 with -fsanitize=address, -fsanitize=leak and/or
9563 -fcheck-pointer-bounds.
9564 Note that sanitized atomic builtins cannot throw exceptions when
9566 operating on invalid memory addresses with non-call exceptions
9567 (-fnon-call-exceptions).
9568 -fsanitize=leak
9570 Enable LeakSanitizer, a memory leak detector. This option only
9571 matters for linking of executables and the executable is linked
9572 against a library that overrides "malloc" and other allocator
9573 functions. See
9574 <https://github.com/google/sanitizers/wiki/AddressSanitizerLeakSanitizer>
9575 for more details. The run-time behavior can be influenced using
9576 the LSAN_OPTIONS environment variable. The option cannot be
9577 combined with -fsanitize=thread.
9578 -fsanitize=undefined
9580 Enable UndefinedBehaviorSanitizer, a fast undefined behavior
9581 detector. Various computations are instrumented to detect
9582 undefined behavior at runtime. Current suboptions are:
9583 -fsanitize=shift
9585 This option enables checking that the result of a shift
9586 operation is not undefined. Note that what exactly is
9587 considered undefined differs slightly between C and C++, as
9588 well as between ISO C90 and C99, etc. This option has two
9589 suboptions, -fsanitize=shift-base and
9590 -fsanitize=shift-exponent.
9591 -fsanitize=shift-exponent
9593 This option enables checking that the second argument of a
9594 shift operation is not negative and is smaller than the
9595 precision of the promoted first argument.
9596 -fsanitize=shift-base
9598 If the second argument of a shift operation is within range,
9599 check that the result of a shift operation is not undefined.
9600 Note that what exactly is considered undefined differs slightly
9601 between C and C++, as well as between ISO C90 and C99, etc.
9602 -fsanitize=integer-divide-by-zero
9604 Detect integer division by zero as well as "INT_MIN / -1"
9605 division.
9606 -fsanitize=unreachable
9608 With this option, the compiler turns the
9609 "__builtin_unreachable" call into a diagnostics message call
9610 instead. When reaching the "__builtin_unreachable" call, the
9611 behavior is undefined.
9612 -fsanitize=vla-bound
9614 This option instructs the compiler to check that the size of a
9615 variable length array is positive.
9616 -fsanitize=null
9618 This option enables pointer checking. Particularly, the
9619 application built with this option turned on will issue an
9620 error message when it tries to dereference a NULL pointer, or
9621 if a reference (possibly an rvalue reference) is bound to a
9622 NULL pointer, or if a method is invoked on an object pointed by
9623 a NULL pointer.
9624 -fsanitize=return
9626 This option enables return statement checking. Programs built
9627 with this option turned on will issue an error message when the
9628 end of a non-void function is reached without actually
9629 returning a value. This option works in C++ only.
9630 -fsanitize=signed-integer-overflow
9632 This option enables signed integer overflow checking. We check
9633 that the result of "+", "*", and both unary and binary "-" does
9634 not overflow in the signed arithmetics. Note, integer
9635 promotion rules must be taken into account. That is, the
9636 following is not an overflow:
9637 signed char a = SCHAR_MAX;
9639 a++;
9640 -fsanitize=bounds
9642 This option enables instrumentation of array bounds. Various
9643 out of bounds accesses are detected. Flexible array members,
9644 flexible array member-like arrays, and initializers of
9645 variables with static storage are not instrumented. The option
9646 cannot be combined with -fcheck-pointer-bounds.
9647 -fsanitize=bounds-strict
9649 This option enables strict instrumentation of array bounds.
9650 Most out of bounds accesses are detected, including flexible
9651 array members and flexible array member-like arrays.
9652 Initializers of variables with static storage are not
9653 instrumented. The option cannot be combined with
9654 -fcheck-pointer-bounds.
9655 -fsanitize=alignment
9657 This option enables checking of alignment of pointers when they
9658 are dereferenced, or when a reference is bound to
9659 insufficiently aligned target, or when a method or constructor
9660 is invoked on insufficiently aligned object.
9661 -fsanitize=object-size
9663 This option enables instrumentation of memory references using
9664 the "__builtin_object_size" function. Various out of bounds
9665 pointer accesses are detected.
9666 -fsanitize=float-divide-by-zero
9668 Detect floating-point division by zero. Unlike other similar
9669 options, -fsanitize=float-divide-by-zero is not enabled by
9670 -fsanitize=undefined, since floating-point division by zero can
9671 be a legitimate way of obtaining infinities and NaNs.
9672 -fsanitize=float-cast-overflow
9674 This option enables floating-point type to integer conversion
9675 checking. We check that the result of the conversion does not
9676 overflow. Unlike other similar options,
9677 -fsanitize=float-cast-overflow is not enabled by
9678 -fsanitize=undefined. This option does not work well with
9679 "FE_INVALID" exceptions enabled.
9680 -fsanitize=nonnull-attribute
9682 This option enables instrumentation of calls, checking whether
9683 null values are not passed to arguments marked as requiring a
9684 non-null value by the "nonnull" function attribute.
9685 -fsanitize=returns-nonnull-attribute
9687 This option enables instrumentation of return statements in
9688 functions marked with "returns_nonnull" function attribute, to
9689 detect returning of null values from such functions.
9690 -fsanitize=bool
9692 This option enables instrumentation of loads from bool. If a
9693 value other than 0/1 is loaded, a run-time error is issued.
9694 -fsanitize=enum
9696 This option enables instrumentation of loads from an enum type.
9697 If a value outside the range of values for the enum type is
9698 loaded, a run-time error is issued.
9699 -fsanitize=vptr
9701 This option enables instrumentation of C++ member function
9702 calls, member accesses and some conversions between pointers to
9703 base and derived classes, to verify the referenced object has
9704 the correct dynamic type.
9705 -fsanitize=pointer-overflow
9707 This option enables instrumentation of pointer arithmetics. If
9708 the pointer arithmetics overflows, a run-time error is issued.
9709 -fsanitize=builtin
9711 This option enables instrumentation of arguments to selected
9712 builtin functions. If an invalid value is passed to such
9713 arguments, a run-time error is issued. E.g. passing 0 as the
9714 argument to "__builtin_ctz" or "__builtin_clz" invokes
9715 undefined behavior and is diagnosed by this option.
9716 While -ftrapv causes traps for signed overflows to be emitted,
9718 -fsanitize=undefined gives a diagnostic message. This currently
9719 works only for the C family of languages.
9720 -fno-sanitize=all
9722 This option disables all previously enabled sanitizers.
9723 -fsanitize=all is not allowed, as some sanitizers cannot be used
9724 together.
9725 -fasan-shadow-offset=number
9727 This option forces GCC to use custom shadow offset in
9728 AddressSanitizer checks. It is useful for experimenting with
9729 different shadow memory layouts in Kernel AddressSanitizer.
9730 -fsanitize-sections=s1,s2,...
9732 Sanitize global variables in selected user-defined sections. si
9733 may contain wildcards.
9734 -fsanitize-recover[=opts]
9736 -fsanitize-recover= controls error recovery mode for sanitizers
9737 mentioned in comma-separated list of opts. Enabling this option
9738 for a sanitizer component causes it to attempt to continue running
9739 the program as if no error happened. This means multiple runtime
9740 errors can be reported in a single program run, and the exit code
9741 of the program may indicate success even when errors have been
9742 reported. The -fno-sanitize-recover= option can be used to alter
9743 this behavior: only the first detected error is reported and
9744 program then exits with a non-zero exit code.
9745 Currently this feature only works for -fsanitize=undefined (and its
9747 suboptions except for -fsanitize=unreachable and
9748 -fsanitize=return), -fsanitize=float-cast-overflow,
9749 -fsanitize=float-divide-by-zero, -fsanitize=bounds-strict,
9750 -fsanitize=kernel-address and -fsanitize=address. For these
9751 sanitizers error recovery is turned on by default, except
9752 -fsanitize=address, for which this feature is experimental.
9753 -fsanitize-recover=all and -fno-sanitize-recover=all is also
9754 accepted, the former enables recovery for all sanitizers that
9755 support it, the latter disables recovery for all sanitizers that
9756 support it.
9757 Even if a recovery mode is turned on the compiler side, it needs to
9759 be also enabled on the runtime library side, otherwise the failures
9760 are still fatal. The runtime library defaults to "halt_on_error=0"
9761 for ThreadSanitizer and UndefinedBehaviorSanitizer, while default
9762 value for AddressSanitizer is "halt_on_error=1". This can be
9763 overridden through setting the "halt_on_error" flag in the
9764 corresponding environment variable.
9765 Syntax without an explicit opts parameter is deprecated. It is
9767 equivalent to specifying an opts list of:
9768 undefined,float-cast-overflow,float-divide-by-zero,bounds-strict
9770 -fsanitize-address-use-after-scope
9772 Enable sanitization of local variables to detect use-after-scope
9773 bugs. The option sets -fstack-reuse to none.
9774 -fsanitize-undefined-trap-on-error
9776 The -fsanitize-undefined-trap-on-error option instructs the
9777 compiler to report undefined behavior using "__builtin_trap" rather
9778 than a "libubsan" library routine. The advantage of this is that
9779 the "libubsan" library is not needed and is not linked in, so this
9780 is usable even in freestanding environments.
9781 -fsanitize-coverage=trace-pc
9783 Enable coverage-guided fuzzing code instrumentation. Inserts a
9784 call to "__sanitizer_cov_trace_pc" into every basic block.
9785 -fsanitize-coverage=trace-cmp
9787 Enable dataflow guided fuzzing code instrumentation. Inserts a
9788 call to "__sanitizer_cov_trace_cmp1", "__sanitizer_cov_trace_cmp2",
9789 "__sanitizer_cov_trace_cmp4" or "__sanitizer_cov_trace_cmp8" for
9790 integral comparison with both operands variable or
9791 "__sanitizer_cov_trace_const_cmp1",
9792 "__sanitizer_cov_trace_const_cmp2",
9793 "__sanitizer_cov_trace_const_cmp4" or
9794 "__sanitizer_cov_trace_const_cmp8" for integral comparison with one
9795 operand constant, "__sanitizer_cov_trace_cmpf" or
9796 "__sanitizer_cov_trace_cmpd" for float or double comparisons and
9797 "__sanitizer_cov_trace_switch" for switch statements.
9798 -fbounds-check
9800 For front ends that support it, generate additional code to check
9801 that indices used to access arrays are within the declared range.
9802 This is currently only supported by the Fortran front end, where
9803 this option defaults to false.
9804 -fcheck-pointer-bounds
9806 Enable Pointer Bounds Checker instrumentation. Each memory
9807 reference is instrumented with checks of the pointer used for
9808 memory access against bounds associated with that pointer.
9809 Currently there is only an implementation for Intel MPX available,
9811 thus x86 GNU/Linux target and -mmpx are required to enable this
9812 feature. MPX-based instrumentation requires a runtime library to
9813 enable MPX in hardware and handle bounds violation signals. By
9814 default when -fcheck-pointer-bounds and -mmpx options are used to
9815 link a program, the GCC driver links against the libmpx and
9816 libmpxwrappers libraries. Bounds checking on calls to dynamic
9817 libraries requires a linker with -z bndplt support; if GCC was
9818 configured with a linker without support for this option (including
9819 the Gold linker and older versions of ld), a warning is given if
9820 you link with -mmpx without also specifying -static, since the
9821 overall effectiveness of the bounds checking protection is reduced.
9822 See also -static-libmpxwrappers.
9823 MPX-based instrumentation may be used for debugging and also may be
9825 included in production code to increase program security.
9826 Depending on usage, you may have different requirements for the
9827 runtime library. The current version of the MPX runtime library is
9828 more oriented for use as a debugging tool. MPX runtime library
9829 usage implies -lpthread. See also -static-libmpx. The runtime
9830 library behavior can be influenced using various CHKP_RT_*
9831 environment variables. See
9832 <https://gcc.gnu.org/wiki/Intel%20MPX%20support%20in%20the%20GCC%20compiler>
9833 for more details.
9834 Generated instrumentation may be controlled by various -fchkp-*
9836 options and by the "bnd_variable_size" structure field attribute
9837 and "bnd_legacy", and "bnd_instrument" function attributes. GCC
9838 also provides a number of built-in functions for controlling the
9839 Pointer Bounds Checker.
9840 -fchkp-check-incomplete-type
9842 Generate pointer bounds checks for variables with incomplete type.
9843 Enabled by default.
9844 -fchkp-narrow-bounds
9846 Controls bounds used by Pointer Bounds Checker for pointers to
9847 object fields. If narrowing is enabled then field bounds are used.
9848 Otherwise object bounds are used. See also
9849 -fchkp-narrow-to-innermost-array and
9850 -fchkp-first-field-has-own-bounds. Enabled by default.
9851 -fchkp-first-field-has-own-bounds
9853 Forces Pointer Bounds Checker to use narrowed bounds for the
9854 address of the first field in the structure. By default a pointer
9855 to the first field has the same bounds as a pointer to the whole
9856 structure.
9857 -fchkp-flexible-struct-trailing-arrays
9859 Forces Pointer Bounds Checker to treat all trailing arrays in
9860 structures as possibly flexible. By default only array fields with
9861 zero length or that are marked with attribute bnd_variable_size are
9862 treated as flexible.
9863 -fchkp-narrow-to-innermost-array
9865 Forces Pointer Bounds Checker to use bounds of the innermost arrays
9866 in case of nested static array access. By default this option is
9867 disabled and bounds of the outermost array are used.
9868 -fchkp-optimize
9870 Enables Pointer Bounds Checker optimizations. Enabled by default
9871 at optimization levels -O, -O2, -O3.
9872 -fchkp-use-fast-string-functions
9874 Enables use of *_nobnd versions of string functions (not copying
9875 bounds) by Pointer Bounds Checker. Disabled by default.
9876 -fchkp-use-nochk-string-functions
9878 Enables use of *_nochk versions of string functions (not checking
9879 bounds) by Pointer Bounds Checker. Disabled by default.
9880 -fchkp-use-static-bounds
9882 Allow Pointer Bounds Checker to generate static bounds holding
9883 bounds of static variables. Enabled by default.
9884 -fchkp-use-static-const-bounds
9886 Use statically-initialized bounds for constant bounds instead of
9887 generating them each time they are required. By default enabled
9888 when -fchkp-use-static-bounds is enabled.
9889 -fchkp-treat-zero-dynamic-size-as-infinite
9891 With this option, objects with incomplete type whose dynamically-
9892 obtained size is zero are treated as having infinite size instead
9893 by Pointer Bounds Checker. This option may be helpful if a program
9894 is linked with a library missing size information for some symbols.
9895 Disabled by default.
9896 -fchkp-check-read
9898 Instructs Pointer Bounds Checker to generate checks for all read
9899 accesses to memory. Enabled by default.
9900 -fchkp-check-write
9902 Instructs Pointer Bounds Checker to generate checks for all write
9903 accesses to memory. Enabled by default.
9904 -fchkp-store-bounds
9906 Instructs Pointer Bounds Checker to generate bounds stores for
9907 pointer writes. Enabled by default.
9908 -fchkp-instrument-calls
9910 Instructs Pointer Bounds Checker to pass pointer bounds to calls.
9911 Enabled by default.
9912 -fchkp-instrument-marked-only
9914 Instructs Pointer Bounds Checker to instrument only functions
9915 marked with the "bnd_instrument" attribute. Disabled by default.
9916 -fchkp-use-wrappers
9918 Allows Pointer Bounds Checker to replace calls to built-in
9919 functions with calls to wrapper functions. When
9920 -fchkp-use-wrappers is used to link a program, the GCC driver
9921 automatically links against libmpxwrappers. See also
9922 -static-libmpxwrappers. Enabled by default.
9923 -fcf-protection=[full|branch|return|none]
9925 Enable code instrumentation of control-flow transfers to increase
9926 program security by checking that target addresses of control-flow
9927 transfer instructions (such as indirect function call, function
9928 return, indirect jump) are valid. This prevents diverting the flow
9929 of control to an unexpected target. This is intended to protect
9930 against such threats as Return-oriented Programming (ROP), and
9931 similarly call/jmp-oriented programming (COP/JOP).
9932 The value "branch" tells the compiler to implement checking of
9934 validity of control-flow transfer at the point of indirect branch
9935 instructions, i.e. call/jmp instructions. The value "return"
9936 implements checking of validity at the point of returning from a
9937 function. The value "full" is an alias for specifying both
9938 "branch" and "return". The value "none" turns off instrumentation.
9939 The macro "__CET__" is defined when -fcf-protection is used. The
9941 first bit of "__CET__" is set to 1 for the value "branch" and the
9942 second bit of "__CET__" is set to 1 for the "return".
9943 You can also use the "nocf_check" attribute to identify which
9945 functions and calls should be skipped from instrumentation.
9946 Currently the x86 GNU/Linux target provides an implementation based
9948 on Intel Control-flow Enforcement Technology (CET).
9949 -fstack-protector
9951 Emit extra code to check for buffer overflows, such as stack
9952 smashing attacks. This is done by adding a guard variable to
9953 functions with vulnerable objects. This includes functions that
9954 call "alloca", and functions with buffers larger than 8 bytes. The
9955 guards are initialized when a function is entered and then checked
9956 when the function exits. If a guard check fails, an error message
9957 is printed and the program exits.
9958 -fstack-protector-all
9960 Like -fstack-protector except that all functions are protected.
9961 -fstack-protector-strong
9963 Like -fstack-protector but includes additional functions to be
9964 protected --- those that have local array definitions, or have
9965 references to local frame addresses.
9966 -fstack-protector-explicit
9968 Like -fstack-protector but only protects those functions which have
9969 the "stack_protect" attribute.
9970 -fstack-check
9972 Generate code to verify that you do not go beyond the boundary of
9973 the stack. You should specify this flag if you are running in an
9974 environment with multiple threads, but you only rarely need to
9975 specify it in a single-threaded environment since stack overflow is
9976 automatically detected on nearly all systems if there is only one
9977 stack.
9978 Note that this switch does not actually cause checking to be done;
9980 the operating system or the language runtime must do that. The
9981 switch causes generation of code to ensure that they see the stack
9982 being extended.
9983 You can additionally specify a string parameter: no means no
9985 checking, generic means force the use of old-style checking,
9986 specific means use the best checking method and is equivalent to
9987 bare -fstack-check.
9988 Old-style checking is a generic mechanism that requires no specific
9990 target support in the compiler but comes with the following
9991 drawbacks:
9992 1. Modified allocation strategy for large objects: they are always
9994 allocated dynamically if their size exceeds a fixed threshold.
9995 Note this may change the semantics of some code.
9996 2. Fixed limit on the size of the static frame of functions: when
9998 it is topped by a particular function, stack checking is not
9999 reliable and a warning is issued by the compiler.
10000 3. Inefficiency: because of both the modified allocation strategy
10002 and the generic implementation, code performance is hampered.
10003 Note that old-style stack checking is also the fallback method for
10005 specific if no target support has been added in the compiler.
10006 -fstack-check= is designed for Ada's needs to detect infinite
10008 recursion and stack overflows. specific is an excellent choice
10009 when compiling Ada code. It is not generally sufficient to protect
10010 against stack-clash attacks. To protect against those you want
10011 -fstack-clash-protection.
10012 -fstack-clash-protection
10014 Generate code to prevent stack clash style attacks. When this
10015 option is enabled, the compiler will only allocate one page of
10016 stack space at a time and each page is accessed immediately after
10017 allocation. Thus, it prevents allocations from jumping over any
10018 stack guard page provided by the operating system.
10019 Most targets do not fully support stack clash protection. However,
10021 on those targets -fstack-clash-protection will protect dynamic
10022 stack allocations. -fstack-clash-protection may also provide
10023 limited protection for static stack allocations if the target
10024 supports -fstack-check=specific.
10025 -fstack-limit-register=reg
10027 -fstack-limit-symbol=sym
10028 -fno-stack-limit
10029 Generate code to ensure that the stack does not grow beyond a
10030 certain value, either the value of a register or the address of a
10031 symbol. If a larger stack is required, a signal is raised at run
10032 time. For most targets, the signal is raised before the stack
10033 overruns the boundary, so it is possible to catch the signal
10034 without taking special precautions.
10035 For instance, if the stack starts at absolute address 0x80000000
10037 and grows downwards, you can use the flags
10038 -fstack-limit-symbol=__stack_limit and
10039 -Wl,--defsym,__stack_limit=0x7ffe0000 to enforce a stack limit of
10040 128KB. Note that this may only work with the GNU linker.
10041 You can locally override stack limit checking by using the
10043 "no_stack_limit" function attribute.
10044 -fsplit-stack
10046 Generate code to automatically split the stack before it overflows.
10047 The resulting program has a discontiguous stack which can only
10048 overflow if the program is unable to allocate any more memory.
10049 This is most useful when running threaded programs, as it is no
10050 longer necessary to calculate a good stack size to use for each
10051 thread. This is currently only implemented for the x86 targets
10052 running GNU/Linux.
10053 When code compiled with -fsplit-stack calls code compiled without
10055 -fsplit-stack, there may not be much stack space available for the
10056 latter code to run. If compiling all code, including library code,
10057 with -fsplit-stack is not an option, then the linker can fix up
10058 these calls so that the code compiled without -fsplit-stack always
10059 has a large stack. Support for this is implemented in the gold
10060 linker in GNU binutils release 2.21 and later.
10061 -fvtable-verify=[std|preinit|none]
10063 This option is only available when compiling C++ code. It turns on
10064 (or off, if using -fvtable-verify=none) the security feature that
10065 verifies at run time, for every virtual call, that the vtable
10066 pointer through which the call is made is valid for the type of the
10067 object, and has not been corrupted or overwritten. If an invalid
10068 vtable pointer is detected at run time, an error is reported and
10069 execution of the program is immediately halted.
10070 This option causes run-time data structures to be built at program
10072 startup, which are used for verifying the vtable pointers. The
10073 options std and preinit control the timing of when these data
10074 structures are built. In both cases the data structures are built
10075 before execution reaches "main". Using -fvtable-verify=std causes
10076 the data structures to be built after shared libraries have been
10077 loaded and initialized. -fvtable-verify=preinit causes them to be
10078 built before shared libraries have been loaded and initialized.
10079 If this option appears multiple times in the command line with
10081 different values specified, none takes highest priority over both
10082 std and preinit; preinit takes priority over std.
10083 -fvtv-debug
10085 When used in conjunction with -fvtable-verify=std or
10086 -fvtable-verify=preinit, causes debug versions of the runtime
10087 functions for the vtable verification feature to be called. This
10088 flag also causes the compiler to log information about which vtable
10089 pointers it finds for each class. This information is written to a
10090 file named vtv_set_ptr_data.log in the directory named by the
10091 environment variable VTV_LOGS_DIR if that is defined or the current
10092 working directory otherwise.
10093 Note: This feature appends data to the log file. If you want a
10095 fresh log file, be sure to delete any existing one.
10096 -fvtv-counts
10098 This is a debugging flag. When used in conjunction with
10099 -fvtable-verify=std or -fvtable-verify=preinit, this causes the
10100 compiler to keep track of the total number of virtual calls it
10101 encounters and the number of verifications it inserts. It also
10102 counts the number of calls to certain run-time library functions
10103 that it inserts and logs this information for each compilation
10104 unit. The compiler writes this information to a file named
10105 vtv_count_data.log in the directory named by the environment
10106 variable VTV_LOGS_DIR if that is defined or the current working
10107 directory otherwise. It also counts the size of the vtable pointer
10108 sets for each class, and writes this information to
10109 vtv_class_set_sizes.log in the same directory.
10110 Note: This feature appends data to the log files. To get fresh
10112 log files, be sure to delete any existing ones.
10113 -finstrument-functions
10115 Generate instrumentation calls for entry and exit to functions.
10116 Just after function entry and just before function exit, the
10117 following profiling functions are called with the address of the
10118 current function and its call site. (On some platforms,
10119 "__builtin_return_address" does not work beyond the current
10120 function, so the call site information may not be available to the
10121 profiling functions otherwise.)
10122 void __cyg_profile_func_enter (void *this_fn,
10124 void *call_site);
10125 void __cyg_profile_func_exit (void *this_fn,
10126 void *call_site);
10127 The first argument is the address of the start of the current
10129 function, which may be looked up exactly in the symbol table.
10130 This instrumentation is also done for functions expanded inline in
10132 other functions. The profiling calls indicate where, conceptually,
10133 the inline function is entered and exited. This means that
10134 addressable versions of such functions must be available. If all
10135 your uses of a function are expanded inline, this may mean an
10136 additional expansion of code size. If you use "extern inline" in
10137 your C code, an addressable version of such functions must be
10138 provided. (This is normally the case anyway, but if you get lucky
10139 and the optimizer always expands the functions inline, you might
10140 have gotten away without providing static copies.)
10141 A function may be given the attribute "no_instrument_function", in
10143 which case this instrumentation is not done. This can be used, for
10144 example, for the profiling functions listed above, high-priority
10145 interrupt routines, and any functions from which the profiling
10146 functions cannot safely be called (perhaps signal handlers, if the
10147 profiling routines generate output or allocate memory).
10148 -finstrument-functions-exclude-file-list=file,file,...
10150 Set the list of functions that are excluded from instrumentation
10151 (see the description of -finstrument-functions). If the file that
10152 contains a function definition matches with one of file, then that
10153 function is not instrumented. The match is done on substrings: if
10154 the file parameter is a substring of the file name, it is
10155 considered to be a match.
10156 For example:
10158 -finstrument-functions-exclude-file-list=/bits/stl,include/sys
10160 excludes any inline function defined in files whose pathnames
10162 contain /bits/stl or include/sys.
10163 If, for some reason, you want to include letter , in one of sym,
10165 write ,. For example,
10166 -finstrument-functions-exclude-file-list=',,tmp' (note the single
10167 quote surrounding the option).
10168 -finstrument-functions-exclude-function-list=sym,sym,...
10170 This is similar to -finstrument-functions-exclude-file-list, but
10171 this option sets the list of function names to be excluded from
10172 instrumentation. The function name to be matched is its user-
10173 visible name, such as "vector<int> blah(const vector<int> &)", not
10174 the internal mangled name (e.g., "_Z4blahRSt6vectorIiSaIiEE"). The
10175 match is done on substrings: if the sym parameter is a substring of
10176 the function name, it is considered to be a match. For C99 and C++
10177 extended identifiers, the function name must be given in UTF-8, not
10178 using universal character names.
10179 -fpatchable-function-entry=N[,M]
10181 Generate N NOPs right at the beginning of each function, with the
10182 function entry point before the Mth NOP. If M is omitted, it
10183 defaults to 0 so the function entry points to the address just at
10184 the first NOP. The NOP instructions reserve extra space which can
10185 be used to patch in any desired instrumentation at run time,
10186 provided that the code segment is writable. The amount of space is
10187 controllable indirectly via the number of NOPs; the NOP instruction
10188 used corresponds to the instruction emitted by the internal GCC
10189 back-end interface "gen_nop". This behavior is target-specific and
10190 may also depend on the architecture variant and/or other
10191 compilation options.
10192 For run-time identification, the starting addresses of these areas,
10194 which correspond to their respective function entries minus M, are
10195 additionally collected in the "__patchable_function_entries"
10196 section of the resulting binary.
10197 Note that the value of "__attribute__ ((patchable_function_entry
10199 (N,M)))" takes precedence over command-line option
10200 -fpatchable-function-entry=N,M. This can be used to increase the
10201 area size or to remove it completely on a single function. If
10202 "N=0", no pad location is recorded.
10203 The NOP instructions are inserted at---and maybe before, depending
10205 on M---the function entry address, even before the prologue.
10206 Options Controlling the Preprocessor
10208 These options control the C preprocessor, which is run on each C source
10209 file before actual compilation.
10210 If you use the -E option, nothing is done except preprocessing. Some
10212 of these options make sense only together with -E because they cause
10213 the preprocessor output to be unsuitable for actual compilation.
10214 In addition to the options listed here, there are a number of options
10216 to control search paths for include files documented in Directory
10217 Options. Options to control preprocessor diagnostics are listed in
10218 Warning Options.
10219 -D name
10221 Predefine name as a macro, with definition 1.
10222 -D name=definition
10224 The contents of definition are tokenized and processed as if they
10225 appeared during translation phase three in a #define directive. In
10226 particular, the definition is truncated by embedded newline
10227 characters.
10228 If you are invoking the preprocessor from a shell or shell-like
10230 program you may need to use the shell's quoting syntax to protect
10231 characters such as spaces that have a meaning in the shell syntax.
10232 If you wish to define a function-like macro on the command line,
10234 write its argument list with surrounding parentheses before the
10235 equals sign (if any). Parentheses are meaningful to most shells,
10236 so you should quote the option. With sh and csh,
10237 -D'name(args...)=definition' works.
10238 -D and -U options are processed in the order they are given on the
10240 command line. All -imacros file and -include file options are
10241 processed after all -D and -U options.
10242 -U name
10244 Cancel any previous definition of name, either built in or provided
10245 with a -D option.
10246 -include file
10248 Process file as if "#include "file"" appeared as the first line of
10249 the primary source file. However, the first directory searched for
10250 file is the preprocessor's working directory instead of the
10251 directory containing the main source file. If not found there, it
10252 is searched for in the remainder of the "#include "..."" search
10253 chain as normal.
10254 If multiple -include options are given, the files are included in
10256 the order they appear on the command line.
10257 -imacros file
10259 Exactly like -include, except that any output produced by scanning
10260 file is thrown away. Macros it defines remain defined. This
10261 allows you to acquire all the macros from a header without also
10262 processing its declarations.
10263 All files specified by -imacros are processed before all files
10265 specified by -include.
10266 -undef
10268 Do not predefine any system-specific or GCC-specific macros. The
10269 standard predefined macros remain defined.
10270 -pthread
10272 Define additional macros required for using the POSIX threads
10273 library. You should use this option consistently for both
10274 compilation and linking. This option is supported on GNU/Linux
10275 targets, most other Unix derivatives, and also on x86 Cygwin and
10276 MinGW targets.
10277 -M Instead of outputting the result of preprocessing, output a rule
10279 suitable for make describing the dependencies of the main source
10280 file. The preprocessor outputs one make rule containing the object
10281 file name for that source file, a colon, and the names of all the
10282 included files, including those coming from -include or -imacros
10283 command-line options.
10284 Unless specified explicitly (with -MT or -MQ), the object file name
10286 consists of the name of the source file with any suffix replaced
10287 with object file suffix and with any leading directory parts
10288 removed. If there are many included files then the rule is split
10289 into several lines using \-newline. The rule has no commands.
10290 This option does not suppress the preprocessor's debug output, such
10292 as -dM. To avoid mixing such debug output with the dependency
10293 rules you should explicitly specify the dependency output file with
10294 -MF, or use an environment variable like DEPENDENCIES_OUTPUT.
10295 Debug output is still sent to the regular output stream as normal.
10296 Passing -M to the driver implies -E, and suppresses warnings with
10298 an implicit -w.
10299 -MM Like -M but do not mention header files that are found in system
10301 header directories, nor header files that are included, directly or
10302 indirectly, from such a header.
10303 This implies that the choice of angle brackets or double quotes in
10305 an #include directive does not in itself determine whether that
10306 header appears in -MM dependency output.
10307 -MF file
10309 When used with -M or -MM, specifies a file to write the
10310 dependencies to. If no -MF switch is given the preprocessor sends
10311 the rules to the same place it would send preprocessed output.
10312 When used with the driver options -MD or -MMD, -MF overrides the
10314 default dependency output file.
10315 If file is -, then the dependencies are written to stdout.
10317 -MG In conjunction with an option such as -M requesting dependency
10319 generation, -MG assumes missing header files are generated files
10320 and adds them to the dependency list without raising an error. The
10321 dependency filename is taken directly from the "#include" directive
10322 without prepending any path. -MG also suppresses preprocessed
10323 output, as a missing header file renders this useless.
10324 This feature is used in automatic updating of makefiles.
10326 -MP This option instructs CPP to add a phony target for each dependency
10328 other than the main file, causing each to depend on nothing. These
10329 dummy rules work around errors make gives if you remove header
10330 files without updating the Makefile to match.
10331 This is typical output:
10333 test.o: test.c test.h
10335 test.h:
10337 -MT target
10339 Change the target of the rule emitted by dependency generation. By
10340 default CPP takes the name of the main input file, deletes any
10341 directory components and any file suffix such as .c, and appends
10342 the platform's usual object suffix. The result is the target.
10343 An -MT option sets the target to be exactly the string you specify.
10345 If you want multiple targets, you can specify them as a single
10346 argument to -MT, or use multiple -MT options.
10347 For example, -MT '$(objpfx)foo.o' might give
10349 $(objpfx)foo.o: foo.c
10351 -MQ target
10353 Same as -MT, but it quotes any characters which are special to
10354 Make. -MQ '$(objpfx)foo.o' gives
10355 $$(objpfx)foo.o: foo.c
10357 The default target is automatically quoted, as if it were given
10359 with -MQ.
10360 -MD -MD is equivalent to -M -MF file, except that -E is not implied.
10362 The driver determines file based on whether an -o option is given.
10363 If it is, the driver uses its argument but with a suffix of .d,
10364 otherwise it takes the name of the input file, removes any
10365 directory components and suffix, and applies a .d suffix.
10366 If -MD is used in conjunction with -E, any -o switch is understood
10368 to specify the dependency output file, but if used without -E, each
10369 -o is understood to specify a target object file.
10370 Since -E is not implied, -MD can be used to generate a dependency
10372 output file as a side effect of the compilation process.
10373 -MMD
10375 Like -MD except mention only user header files, not system header
10376 files.
10377 -fpreprocessed
10379 Indicate to the preprocessor that the input file has already been
10380 preprocessed. This suppresses things like macro expansion,
10381 trigraph conversion, escaped newline splicing, and processing of
10382 most directives. The preprocessor still recognizes and removes
10383 comments, so that you can pass a file preprocessed with -C to the
10384 compiler without problems. In this mode the integrated
10385 preprocessor is little more than a tokenizer for the front ends.
10386 -fpreprocessed is implicit if the input file has one of the
10388 extensions .i, .ii or .mi. These are the extensions that GCC uses
10389 for preprocessed files created by -save-temps.
10390 -fdirectives-only
10392 When preprocessing, handle directives, but do not expand macros.
10393 The option's behavior depends on the -E and -fpreprocessed options.
10395 With -E, preprocessing is limited to the handling of directives
10397 such as "#define", "#ifdef", and "#error". Other preprocessor
10398 operations, such as macro expansion and trigraph conversion are not
10399 performed. In addition, the -dD option is implicitly enabled.
10400 With -fpreprocessed, predefinition of command line and most builtin
10402 macros is disabled. Macros such as "__LINE__", which are
10403 contextually dependent, are handled normally. This enables
10404 compilation of files previously preprocessed with "-E
10405 -fdirectives-only".
10406 With both -E and -fpreprocessed, the rules for -fpreprocessed take
10408 precedence. This enables full preprocessing of files previously
10409 preprocessed with "-E -fdirectives-only".
10410 -fdollars-in-identifiers
10412 Accept $ in identifiers.
10413 -fextended-identifiers
10415 Accept universal character names in identifiers. This option is
10416 enabled by default for C99 (and later C standard versions) and C++.
10417 -fno-canonical-system-headers
10419 When preprocessing, do not shorten system header paths with
10420 canonicalization.
10421 -ftabstop=width
10423 Set the distance between tab stops. This helps the preprocessor
10424 report correct column numbers in warnings or errors, even if tabs
10425 appear on the line. If the value is less than 1 or greater than
10426 100, the option is ignored. The default is 8.
10427 -ftrack-macro-expansion[=level]
10429 Track locations of tokens across macro expansions. This allows the
10430 compiler to emit diagnostic about the current macro expansion stack
10431 when a compilation error occurs in a macro expansion. Using this
10432 option makes the preprocessor and the compiler consume more memory.
10433 The level parameter can be used to choose the level of precision of
10434 token location tracking thus decreasing the memory consumption if
10435 necessary. Value 0 of level de-activates this option. Value 1
10436 tracks tokens locations in a degraded mode for the sake of minimal
10437 memory overhead. In this mode all tokens resulting from the
10438 expansion of an argument of a function-like macro have the same
10439 location. Value 2 tracks tokens locations completely. This value is
10440 the most memory hungry. When this option is given no argument, the
10441 default parameter value is 2.
10442 Note that "-ftrack-macro-expansion=2" is activated by default.
10444 -fmacro-prefix-map=old=new
10446 When preprocessing files residing in directory old, expand the
10447 "__FILE__" and "__BASE_FILE__" macros as if the files resided in
10448 directory new instead. This can be used to change an absolute path
10449 to a relative path by using . for new which can result in more
10450 reproducible builds that are location independent. This option
10451 also affects "__builtin_FILE()" during compilation. See also
10452 -ffile-prefix-map.
10453 -fexec-charset=charset
10455 Set the execution character set, used for string and character
10456 constants. The default is UTF-8. charset can be any encoding
10457 supported by the system's "iconv" library routine.
10458 -fwide-exec-charset=charset
10460 Set the wide execution character set, used for wide string and
10461 character constants. The default is UTF-32 or UTF-16, whichever
10462 corresponds to the width of "wchar_t". As with -fexec-charset,
10463 charset can be any encoding supported by the system's "iconv"
10464 library routine; however, you will have problems with encodings
10465 that do not fit exactly in "wchar_t".
10466 -finput-charset=charset
10468 Set the input character set, used for translation from the
10469 character set of the input file to the source character set used by
10470 GCC. If the locale does not specify, or GCC cannot get this
10471 information from the locale, the default is UTF-8. This can be
10472 overridden by either the locale or this command-line option.
10473 Currently the command-line option takes precedence if there's a
10474 conflict. charset can be any encoding supported by the system's
10475 "iconv" library routine.
10476 -fpch-deps
10478 When using precompiled headers, this flag causes the dependency-
10479 output flags to also list the files from the precompiled header's
10480 dependencies. If not specified, only the precompiled header are
10481 listed and not the files that were used to create it, because those
10482 files are not consulted when a precompiled header is used.
10483 -fpch-preprocess
10485 This option allows use of a precompiled header together with -E.
10486 It inserts a special "#pragma", "#pragma GCC pch_preprocess
10487 "filename"" in the output to mark the place where the precompiled
10488 header was found, and its filename. When -fpreprocessed is in use,
10489 GCC recognizes this "#pragma" and loads the PCH.
10490 This option is off by default, because the resulting preprocessed
10492 output is only really suitable as input to GCC. It is switched on
10493 by -save-temps.
10494 You should not write this "#pragma" in your own code, but it is
10496 safe to edit the filename if the PCH file is available in a
10497 different location. The filename may be absolute or it may be
10498 relative to GCC's current directory.
10499 -fworking-directory
10501 Enable generation of linemarkers in the preprocessor output that
10502 let the compiler know the current working directory at the time of
10503 preprocessing. When this option is enabled, the preprocessor
10504 emits, after the initial linemarker, a second linemarker with the
10505 current working directory followed by two slashes. GCC uses this
10506 directory, when it's present in the preprocessed input, as the
10507 directory emitted as the current working directory in some
10508 debugging information formats. This option is implicitly enabled
10509 if debugging information is enabled, but this can be inhibited with
10510 the negated form -fno-working-directory. If the -P flag is present
10511 in the command line, this option has no effect, since no "#line"
10512 directives are emitted whatsoever.
10513 -A predicate=answer
10515 Make an assertion with the predicate predicate and answer answer.
10516 This form is preferred to the older form -A predicate(answer),
10517 which is still supported, because it does not use shell special
10518 characters.
10519 -A -predicate=answer
10521 Cancel an assertion with the predicate predicate and answer answer.
10522 -C Do not discard comments. All comments are passed through to the
10524 output file, except for comments in processed directives, which are
10525 deleted along with the directive.
10526 You should be prepared for side effects when using -C; it causes
10528 the preprocessor to treat comments as tokens in their own right.
10529 For example, comments appearing at the start of what would be a
10530 directive line have the effect of turning that line into an
10531 ordinary source line, since the first token on the line is no
10532 longer a #.
10533 -CC Do not discard comments, including during macro expansion. This is
10535 like -C, except that comments contained within macros are also
10536 passed through to the output file where the macro is expanded.
10537 In addition to the side effects of the -C option, the -CC option
10539 causes all C++-style comments inside a macro to be converted to
10540 C-style comments. This is to prevent later use of that macro from
10541 inadvertently commenting out the remainder of the source line.
10542 The -CC option is generally used to support lint comments.
10544 -P Inhibit generation of linemarkers in the output from the
10546 preprocessor. This might be useful when running the preprocessor
10547 on something that is not C code, and will be sent to a program
10548 which might be confused by the linemarkers.
10549 -traditional
10551 -traditional-cpp
10552 Try to imitate the behavior of pre-standard C preprocessors, as
10553 opposed to ISO C preprocessors. See the GNU CPP manual for
10554 details.
10555 Note that GCC does not otherwise attempt to emulate a pre-standard
10557 C compiler, and these options are only supported with the -E
10558 switch, or when invoking CPP explicitly.
10559 -trigraphs
10561 Support ISO C trigraphs. These are three-character sequences, all
10562 starting with ??, that are defined by ISO C to stand for single
10563 characters. For example, ??/ stands for \, so '??/n' is a
10564 character constant for a newline.
10565 The nine trigraphs and their replacements are
10567 Trigraph: ??( ??) ??< ??> ??= ??/ ??' ??! ??-
10569 Replacement: [ ] { } # \ ^ | ~
10570 By default, GCC ignores trigraphs, but in standard-conforming modes
10572 it converts them. See the -std and -ansi options.
10573 -remap
10575 Enable special code to work around file systems which only permit
10576 very short file names, such as MS-DOS.
10577 -H Print the name of each header file used, in addition to other
10579 normal activities. Each name is indented to show how deep in the
10580 #include stack it is. Precompiled header files are also printed,
10581 even if they are found to be invalid; an invalid precompiled header
10582 file is printed with ...x and a valid one with ...! .
10583 -dletters
10585 Says to make debugging dumps during compilation as specified by
10586 letters. The flags documented here are those relevant to the
10587 preprocessor. Other letters are interpreted by the compiler
10588 proper, or reserved for future versions of GCC, and so are silently
10589 ignored. If you specify letters whose behavior conflicts, the
10590 result is undefined.
10591 -dM Instead of the normal output, generate a list of #define
10593 directives for all the macros defined during the execution of
10594 the preprocessor, including predefined macros. This gives you
10595 a way of finding out what is predefined in your version of the
10596 preprocessor. Assuming you have no file foo.h, the command
10597 touch foo.h; cpp -dM foo.h
10599 shows all the predefined macros.
10601 If you use -dM without the -E option, -dM is interpreted as a
10603 synonym for -fdump-rtl-mach.
10604 -dD Like -dM except in two respects: it does not include the
10606 predefined macros, and it outputs both the #define directives
10607 and the result of preprocessing. Both kinds of output go to
10608 the standard output file.
10609 -dN Like -dD, but emit only the macro names, not their expansions.
10611 -dI Output #include directives in addition to the result of
10613 preprocessing.
10614 -dU Like -dD except that only macros that are expanded, or whose
10616 definedness is tested in preprocessor directives, are output;
10617 the output is delayed until the use or test of the macro; and
10618 #undef directives are also output for macros tested but
10619 undefined at the time.
10620 -fdebug-cpp
10622 This option is only useful for debugging GCC. When used from CPP
10623 or with -E, it dumps debugging information about location maps.
10624 Every token in the output is preceded by the dump of the map its
10625 location belongs to.
10626 When used from GCC without -E, this option has no effect.
10628 -Wp,option
10630 You can use -Wp,option to bypass the compiler driver and pass
10631 option directly through to the preprocessor. If option contains
10632 commas, it is split into multiple options at the commas. However,
10633 many options are modified, translated or interpreted by the
10634 compiler driver before being passed to the preprocessor, and -Wp
10635 forcibly bypasses this phase. The preprocessor's direct interface
10636 is undocumented and subject to change, so whenever possible you
10637 should avoid using -Wp and let the driver handle the options
10638 instead.
10639 -Xpreprocessor option
10641 Pass option as an option to the preprocessor. You can use this to
10642 supply system-specific preprocessor options that GCC does not
10643 recognize.
10644 If you want to pass an option that takes an argument, you must use
10646 -Xpreprocessor twice, once for the option and once for the
10647 argument.
10648 -no-integrated-cpp
10650 Perform preprocessing as a separate pass before compilation. By
10651 default, GCC performs preprocessing as an integrated part of input
10652 tokenization and parsing. If this option is provided, the
10653 appropriate language front end (cc1, cc1plus, or cc1obj for C, C++,
10654 and Objective-C, respectively) is instead invoked twice, once for
10655 preprocessing only and once for actual compilation of the
10656 preprocessed input. This option may be useful in conjunction with
10657 the -B or -wrapper options to specify an alternate preprocessor or
10658 perform additional processing of the program source between normal
10659 preprocessing and compilation.
10660 Passing Options to the Assembler
10662 You can pass options to the assembler.
10663 -Wa,option
10665 Pass option as an option to the assembler. If option contains
10666 commas, it is split into multiple options at the commas.
10667 -Xassembler option
10669 Pass option as an option to the assembler. You can use this to
10670 supply system-specific assembler options that GCC does not
10671 recognize.
10672 If you want to pass an option that takes an argument, you must use
10674 -Xassembler twice, once for the option and once for the argument.
10675 Options for Linking
10677 These options come into play when the compiler links object files into
10678 an executable output file. They are meaningless if the compiler is not
10679 doing a link step.
10680 object-file-name
10682 A file name that does not end in a special recognized suffix is
10683 considered to name an object file or library. (Object files are
10684 distinguished from libraries by the linker according to the file
10685 contents.) If linking is done, these object files are used as
10686 input to the linker.
10687 -c
10689 -S
10690 -E If any of these options is used, then the linker is not run, and
10691 object file names should not be used as arguments.
10692 -fuse-ld=bfd
10694 Use the bfd linker instead of the default linker.
10695 -fuse-ld=gold
10697 Use the gold linker instead of the default linker.
10698 -llibrary
10700 -l library
10701 Search the library named library when linking. (The second
10702 alternative with the library as a separate argument is only for
10703 POSIX compliance and is not recommended.)
10704 It makes a difference where in the command you write this option;
10706 the linker searches and processes libraries and object files in the
10707 order they are specified. Thus, foo.o -lz bar.o searches library z
10708 after file foo.o but before bar.o. If bar.o refers to functions in
10709 z, those functions may not be loaded.
10710 The linker searches a standard list of directories for the library,
10712 which is actually a file named liblibrary.a. The linker then uses
10713 this file as if it had been specified precisely by name.
10714 The directories searched include several standard system
10716 directories plus any that you specify with -L.
10717 Normally the files found this way are library files---archive files
10719 whose members are object files. The linker handles an archive file
10720 by scanning through it for members which define symbols that have
10721 so far been referenced but not defined. But if the file that is
10722 found is an ordinary object file, it is linked in the usual
10723 fashion. The only difference between using an -l option and
10724 specifying a file name is that -l surrounds library with lib and .a
10725 and searches several directories.
10726 -lobjc
10728 You need this special case of the -l option in order to link an
10729 Objective-C or Objective-C++ program.
10730 -nostartfiles
10732 Do not use the standard system startup files when linking. The
10733 standard system libraries are used normally, unless -nostdlib or
10734 -nodefaultlibs is used.
10735 -nodefaultlibs
10737 Do not use the standard system libraries when linking. Only the
10738 libraries you specify are passed to the linker, and options
10739 specifying linkage of the system libraries, such as -static-libgcc
10740 or -shared-libgcc, are ignored. The standard startup files are
10741 used normally, unless -nostartfiles is used.
10742 The compiler may generate calls to "memcmp", "memset", "memcpy" and
10744 "memmove". These entries are usually resolved by entries in libc.
10745 These entry points should be supplied through some other mechanism
10746 when this option is specified.
10747 -nostdlib
10749 Do not use the standard system startup files or libraries when
10750 linking. No startup files and only the libraries you specify are
10751 passed to the linker, and options specifying linkage of the system
10752 libraries, such as -static-libgcc or -shared-libgcc, are ignored.
10753 The compiler may generate calls to "memcmp", "memset", "memcpy" and
10755 "memmove". These entries are usually resolved by entries in libc.
10756 These entry points should be supplied through some other mechanism
10757 when this option is specified.
10758 One of the standard libraries bypassed by -nostdlib and
10760 -nodefaultlibs is libgcc.a, a library of internal subroutines which
10761 GCC uses to overcome shortcomings of particular machines, or
10762 special needs for some languages.
10763 In most cases, you need libgcc.a even when you want to avoid other
10765 standard libraries. In other words, when you specify -nostdlib or
10766 -nodefaultlibs you should usually specify -lgcc as well. This
10767 ensures that you have no unresolved references to internal GCC
10768 library subroutines. (An example of such an internal subroutine is
10769 "__main", used to ensure C++ constructors are called.)
10770 -pie
10772 Produce a dynamically linked position independent executable on
10773 targets that support it. For predictable results, you must also
10774 specify the same set of options used for compilation (-fpie, -fPIE,
10775 or model suboptions) when you specify this linker option.
10776 -no-pie
10778 Don't produce a dynamically linked position independent executable.
10779 -static-pie
10781 Produce a static position independent executable on targets that
10782 support it. A static position independent executable is similar to
10783 a static executable, but can be loaded at any address without a
10784 dynamic linker. For predictable results, you must also specify the
10785 same set of options used for compilation (-fpie, -fPIE, or model
10786 suboptions) when you specify this linker option.
10787 -pthread
10789 Link with the POSIX threads library. This option is supported on
10790 GNU/Linux targets, most other Unix derivatives, and also on x86
10791 Cygwin and MinGW targets. On some targets this option also sets
10792 flags for the preprocessor, so it should be used consistently for
10793 both compilation and linking.
10794 -rdynamic
10796 Pass the flag -export-dynamic to the ELF linker, on targets that
10797 support it. This instructs the linker to add all symbols, not only
10798 used ones, to the dynamic symbol table. This option is needed for
10799 some uses of "dlopen" or to allow obtaining backtraces from within
10800 a program.
10801 -s Remove all symbol table and relocation information from the
10803 executable.
10804 -static
10806 On systems that support dynamic linking, this overrides -pie and
10807 prevents linking with the shared libraries. On other systems, this
10808 option has no effect.
10809 -shared
10811 Produce a shared object which can then be linked with other objects
10812 to form an executable. Not all systems support this option. For
10813 predictable results, you must also specify the same set of options
10814 used for compilation (-fpic, -fPIC, or model suboptions) when you
10815 specify this linker option.[1]
10816 -shared-libgcc
10818 -static-libgcc
10819 On systems that provide libgcc as a shared library, these options
10820 force the use of either the shared or static version, respectively.
10821 If no shared version of libgcc was built when the compiler was
10822 configured, these options have no effect.
10823 There are several situations in which an application should use the
10825 shared libgcc instead of the static version. The most common of
10826 these is when the application wishes to throw and catch exceptions
10827 across different shared libraries. In that case, each of the
10828 libraries as well as the application itself should use the shared
10829 libgcc.
10830 Therefore, the G++ and driver automatically adds -shared-libgcc
10832 whenever you build a shared library or a main executable, because
10833 C++
10834 programs typically use exceptions, so this is the right thing to
10835 do.
10836 If, instead, you use the GCC driver to create shared libraries, you
10838 may find that they are not always linked with the shared libgcc.
10839 If GCC finds, at its configuration time, that you have a non-GNU
10840 linker or a GNU linker that does not support option --eh-frame-hdr,
10841 it links the shared version of libgcc into shared libraries by
10842 default. Otherwise, it takes advantage of the linker and optimizes
10843 away the linking with the shared version of libgcc, linking with
10844 the static version of libgcc by default. This allows exceptions to
10845 propagate through such shared libraries, without incurring
10846 relocation costs at library load time.
10847 However, if a library or main executable is supposed to throw or
10849 catch exceptions, you must link it using the G++ driver, as
10850 appropriate for the languages used in the program, or using the
10851 option -shared-libgcc, such that it is linked with the shared
10852 libgcc.
10853 -static-libasan
10855 When the -fsanitize=address option is used to link a program, the
10856 GCC driver automatically links against libasan. If libasan is
10857 available as a shared library, and the -static option is not used,
10858 then this links against the shared version of libasan. The
10859 -static-libasan option directs the GCC driver to link libasan
10860 statically, without necessarily linking other libraries statically.
10861 -static-libtsan
10863 When the -fsanitize=thread option is used to link a program, the
10864 GCC driver automatically links against libtsan. If libtsan is
10865 available as a shared library, and the -static option is not used,
10866 then this links against the shared version of libtsan. The
10867 -static-libtsan option directs the GCC driver to link libtsan
10868 statically, without necessarily linking other libraries statically.
10869 -static-liblsan
10871 When the -fsanitize=leak option is used to link a program, the GCC
10872 driver automatically links against liblsan. If liblsan is
10873 available as a shared library, and the -static option is not used,
10874 then this links against the shared version of liblsan. The
10875 -static-liblsan option directs the GCC driver to link liblsan
10876 statically, without necessarily linking other libraries statically.
10877 -static-libubsan
10879 When the -fsanitize=undefined option is used to link a program, the
10880 GCC driver automatically links against libubsan. If libubsan is
10881 available as a shared library, and the -static option is not used,
10882 then this links against the shared version of libubsan. The
10883 -static-libubsan option directs the GCC driver to link libubsan
10884 statically, without necessarily linking other libraries statically.
10885 -static-libmpx
10887 When the -fcheck-pointer bounds and -mmpx options are used to link
10888 a program, the GCC driver automatically links against libmpx. If
10889 libmpx is available as a shared library, and the -static option is
10890 not used, then this links against the shared version of libmpx.
10891 The -static-libmpx option directs the GCC driver to link libmpx
10892 statically, without necessarily linking other libraries statically.
10893 -static-libmpxwrappers
10895 When the -fcheck-pointer bounds and -mmpx options are used to link
10896 a program without also using -fno-chkp-use-wrappers, the GCC driver
10897 automatically links against libmpxwrappers. If libmpxwrappers is
10898 available as a shared library, and the -static option is not used,
10899 then this links against the shared version of libmpxwrappers. The
10900 -static-libmpxwrappers option directs the GCC driver to link
10901 libmpxwrappers statically, without necessarily linking other
10902 libraries statically.
10903 -static-libstdc++
10905 When the g++ program is used to link a C++ program, it normally
10906 automatically links against libstdc++. If libstdc++ is available
10907 as a shared library, and the -static option is not used, then this
10908 links against the shared version of libstdc++. That is normally
10909 fine. However, it is sometimes useful to freeze the version of
10910 libstdc++ used by the program without going all the way to a fully
10911 static link. The -static-libstdc++ option directs the g++ driver
10912 to link libstdc++ statically, without necessarily linking other
10913 libraries statically.
10914 -symbolic
10916 Bind references to global symbols when building a shared object.
10917 Warn about any unresolved references (unless overridden by the link
10918 editor option -Xlinker -z -Xlinker defs). Only a few systems
10919 support this option.
10920 -T script
10922 Use script as the linker script. This option is supported by most
10923 systems using the GNU linker. On some targets, such as bare-board
10924 targets without an operating system, the -T option may be required
10925 when linking to avoid references to undefined symbols.
10926 -Xlinker option
10928 Pass option as an option to the linker. You can use this to supply
10929 system-specific linker options that GCC does not recognize.
10930 If you want to pass an option that takes a separate argument, you
10932 must use -Xlinker twice, once for the option and once for the
10933 argument. For example, to pass -assert definitions, you must write
10934 -Xlinker -assert -Xlinker definitions. It does not work to write
10935 -Xlinker "-assert definitions", because this passes the entire
10936 string as a single argument, which is not what the linker expects.
10937 When using the GNU linker, it is usually more convenient to pass
10939 arguments to linker options using the option=value syntax than as
10940 separate arguments. For example, you can specify -Xlinker
10941 -Map=output.map rather than -Xlinker -Map -Xlinker output.map.
10942 Other linkers may not support this syntax for command-line options.
10943 -Wl,option
10945 Pass option as an option to the linker. If option contains commas,
10946 it is split into multiple options at the commas. You can use this
10947 syntax to pass an argument to the option. For example,
10948 -Wl,-Map,output.map passes -Map output.map to the linker. When
10949 using the GNU linker, you can also get the same effect with
10950 -Wl,-Map=output.map.
10951 -u symbol
10953 Pretend the symbol symbol is undefined, to force linking of library
10954 modules to define it. You can use -u multiple times with different
10955 symbols to force loading of additional library modules.
10956 -z keyword
10958 -z is passed directly on to the linker along with the keyword
10959 keyword. See the section in the documentation of your linker for
10960 permitted values and their meanings.
10961 Options for Directory Search
10963 These options specify directories to search for header files, for
10964 libraries and for parts of the compiler:
10965 -I dir
10967 -iquote dir
10968 -isystem dir
10969 -idirafter dir
10970 Add the directory dir to the list of directories to be searched for
10971 header files during preprocessing. If dir begins with = or
10972 $SYSROOT, then the = or $SYSROOT is replaced by the sysroot prefix;
10973 see --sysroot and -isysroot.
10974 Directories specified with -iquote apply only to the quote form of
10976 the directive, "#include "file"". Directories specified with -I,
10977 -isystem, or -idirafter apply to lookup for both the
10978 "#include "file"" and "#include <file>" directives.
10979 You can specify any number or combination of these options on the
10981 command line to search for header files in several directories.
10982 The lookup order is as follows:
10983 1. For the quote form of the include directive, the directory of
10985 the current file is searched first.
10986 2. For the quote form of the include directive, the directories
10988 specified by -iquote options are searched in left-to-right
10989 order, as they appear on the command line.
10990 3. Directories specified with -I options are scanned in left-to-
10992 right order.
10993 4. Directories specified with -isystem options are scanned in
10995 left-to-right order.
10996 5. Standard system directories are scanned.
10998 6. Directories specified with -idirafter options are scanned in
11000 left-to-right order.
11001 You can use -I to override a system header file, substituting your
11003 own version, since these directories are searched before the
11004 standard system header file directories. However, you should not
11005 use this option to add directories that contain vendor-supplied
11006 system header files; use -isystem for that.
11007 The -isystem and -idirafter options also mark the directory as a
11009 system directory, so that it gets the same special treatment that
11010 is applied to the standard system directories.
11011 If a standard system include directory, or a directory specified
11013 with -isystem, is also specified with -I, the -I option is ignored.
11014 The directory is still searched but as a system directory at its
11015 normal position in the system include chain. This is to ensure
11016 that GCC's procedure to fix buggy system headers and the ordering
11017 for the "#include_next" directive are not inadvertently changed.
11018 If you really need to change the search order for system
11019 directories, use the -nostdinc and/or -isystem options.
11020 -I- Split the include path. This option has been deprecated. Please
11022 use -iquote instead for -I directories before the -I- and remove
11023 the -I- option.
11024 Any directories specified with -I options before -I- are searched
11026 only for headers requested with "#include "file""; they are not
11027 searched for "#include <file>". If additional directories are
11028 specified with -I options after the -I-, those directories are
11029 searched for all #include directives.
11030 In addition, -I- inhibits the use of the directory of the current
11032 file directory as the first search directory for "#include "file"".
11033 There is no way to override this effect of -I-.
11034 -iprefix prefix
11036 Specify prefix as the prefix for subsequent -iwithprefix options.
11037 If the prefix represents a directory, you should include the final
11038 /.
11039 -iwithprefix dir
11041 -iwithprefixbefore dir
11042 Append dir to the prefix specified previously with -iprefix, and
11043 add the resulting directory to the include search path.
11044 -iwithprefixbefore puts it in the same place -I would; -iwithprefix
11045 puts it where -idirafter would.
11046 -isysroot dir
11048 This option is like the --sysroot option, but applies only to
11049 header files (except for Darwin targets, where it applies to both
11050 header files and libraries). See the --sysroot option for more
11051 information.
11052 -imultilib dir
11054 Use dir as a subdirectory of the directory containing target-
11055 specific C++ headers.
11056 -nostdinc
11058 Do not search the standard system directories for header files.
11059 Only the directories explicitly specified with -I, -iquote,
11060 -isystem, and/or -idirafter options (and the directory of the
11061 current file, if appropriate) are searched.
11062 -nostdinc++
11064 Do not search for header files in the C++-specific standard
11065 directories, but do still search the other standard directories.
11066 (This option is used when building the C++ library.)
11067 -iplugindir=dir
11069 Set the directory to search for plugins that are passed by
11070 -fplugin=name instead of -fplugin=path/name.so. This option is not
11071 meant to be used by the user, but only passed by the driver.
11072 -Ldir
11074 Add directory dir to the list of directories to be searched for -l.
11075 -Bprefix
11077 This option specifies where to find the executables, libraries,
11078 include files, and data files of the compiler itself.
11079 The compiler driver program runs one or more of the subprograms
11081 cpp, cc1, as and ld. It tries prefix as a prefix for each program
11082 it tries to run, both with and without machine/version/ for the
11083 corresponding target machine and compiler version.
11084 For each subprogram to be run, the compiler driver first tries the
11086 -B prefix, if any. If that name is not found, or if -B is not
11087 specified, the driver tries two standard prefixes, /usr/lib/gcc/
11088 and /usr/local/lib/gcc/. If neither of those results in a file
11089 name that is found, the unmodified program name is searched for
11090 using the directories specified in your PATH environment variable.
11091 The compiler checks to see if the path provided by -B refers to a
11093 directory, and if necessary it adds a directory separator character
11094 at the end of the path.
11095 -B prefixes that effectively specify directory names also apply to
11097 libraries in the linker, because the compiler translates these
11098 options into -L options for the linker. They also apply to include
11099 files in the preprocessor, because the compiler translates these
11100 options into -isystem options for the preprocessor. In this case,
11101 the compiler appends include to the prefix.
11102 The runtime support file libgcc.a can also be searched for using
11104 the -B prefix, if needed. If it is not found there, the two
11105 standard prefixes above are tried, and that is all. The file is
11106 left out of the link if it is not found by those means.
11107 Another way to specify a prefix much like the -B prefix is to use
11109 the environment variable GCC_EXEC_PREFIX.
11110 As a special kludge, if the path provided by -B is [dir/]stageN/,
11112 where N is a number in the range 0 to 9, then it is replaced by
11113 [dir/]include. This is to help with boot-strapping the compiler.
11114 -no-canonical-prefixes
11116 Do not expand any symbolic links, resolve references to /../ or
11117 /./, or make the path absolute when generating a relative prefix.
11118 --sysroot=dir
11120 Use dir as the logical root directory for headers and libraries.
11121 For example, if the compiler normally searches for headers in
11122 /usr/include and libraries in /usr/lib, it instead searches
11123 dir/usr/include and dir/usr/lib.
11124 If you use both this option and the -isysroot option, then the
11126 --sysroot option applies to libraries, but the -isysroot option
11127 applies to header files.
11128 The GNU linker (beginning with version 2.16) has the necessary
11130 support for this option. If your linker does not support this
11131 option, the header file aspect of --sysroot still works, but the
11132 library aspect does not.
11133 --no-sysroot-suffix
11135 For some targets, a suffix is added to the root directory specified
11136 with --sysroot, depending on the other options used, so that
11137 headers may for example be found in dir/suffix/usr/include instead
11138 of dir/usr/include. This option disables the addition of such a
11139 suffix.
11140 Options for Code Generation Conventions
11142 These machine-independent options control the interface conventions
11143 used in code generation.
11144 Most of them have both positive and negative forms; the negative form
11146 of -ffoo is -fno-foo. In the table below, only one of the forms is
11147 listed---the one that is not the default. You can figure out the other
11148 form by either removing no- or adding it.
11149 -fstack-reuse=reuse-level
11151 This option controls stack space reuse for user declared local/auto
11152 variables and compiler generated temporaries. reuse_level can be
11153 all, named_vars, or none. all enables stack reuse for all local
11154 variables and temporaries, named_vars enables the reuse only for
11155 user defined local variables with names, and none disables stack
11156 reuse completely. The default value is all. The option is needed
11157 when the program extends the lifetime of a scoped local variable or
11158 a compiler generated temporary beyond the end point defined by the
11159 language. When a lifetime of a variable ends, and if the variable
11160 lives in memory, the optimizing compiler has the freedom to reuse
11161 its stack space with other temporaries or scoped local variables
11162 whose live range does not overlap with it. Legacy code extending
11163 local lifetime is likely to break with the stack reuse
11164 optimization.
11165 For example,
11167 int *p;
11169 {
11170 int local1;
11171 p = &local1;
11173 local1 = 10;
11174 ....
11175 }
11176 {
11177 int local2;
11178 local2 = 20;
11179 ...
11180 }
11181 if (*p == 10) // out of scope use of local1
11183 {
11184 }
11186 Another example:
11188 struct A
11190 {
11191 A(int k) : i(k), j(k) { }
11192 int i;
11193 int j;
11194 };
11195 A *ap;
11197 void foo(const A& ar)
11199 {
11200 ap = &ar;
11201 }
11202 void bar()
11204 {
11205 foo(A(10)); // temp object's lifetime ends when foo returns
11206 {
11208 A a(20);
11209 ....
11210 }
11211 ap->i+= 10; // ap references out of scope temp whose space
11212 // is reused with a. What is the value of ap->i?
11213 }
11214 The lifetime of a compiler generated temporary is well defined by
11216 the C++ standard. When a lifetime of a temporary ends, and if the
11217 temporary lives in memory, the optimizing compiler has the freedom
11218 to reuse its stack space with other temporaries or scoped local
11219 variables whose live range does not overlap with it. However some
11220 of the legacy code relies on the behavior of older compilers in
11221 which temporaries' stack space is not reused, the aggressive stack
11222 reuse can lead to runtime errors. This option is used to control
11223 the temporary stack reuse optimization.
11224 -ftrapv
11226 This option generates traps for signed overflow on addition,
11227 subtraction, multiplication operations. The options -ftrapv and
11228 -fwrapv override each other, so using -ftrapv -fwrapv on the
11229 command-line results in -fwrapv being effective. Note that only
11230 active options override, so using -ftrapv -fwrapv -fno-wrapv on the
11231 command-line results in -ftrapv being effective.
11232 -fwrapv
11234 This option instructs the compiler to assume that signed arithmetic
11235 overflow of addition, subtraction and multiplication wraps around
11236 using twos-complement representation. This flag enables some
11237 optimizations and disables others. The options -ftrapv and -fwrapv
11238 override each other, so using -ftrapv -fwrapv on the command-line
11239 results in -fwrapv being effective. Note that only active options
11240 override, so using -ftrapv -fwrapv -fno-wrapv on the command-line
11241 results in -ftrapv being effective.
11242 -fwrapv-pointer
11244 This option instructs the compiler to assume that pointer
11245 arithmetic overflow on addition and subtraction wraps around using
11246 twos-complement representation. This flag disables some
11247 optimizations which assume pointer overflow is invalid.
11248 -fstrict-overflow
11250 This option implies -fno-wrapv -fno-wrapv-pointer and when negated
11251 implies -fwrapv -fwrapv-pointer.
11252 -fexceptions
11254 Enable exception handling. Generates extra code needed to
11255 propagate exceptions. For some targets, this implies GCC generates
11256 frame unwind information for all functions, which can produce
11257 significant data size overhead, although it does not affect
11258 execution. If you do not specify this option, GCC enables it by
11259 default for languages like C++ that normally require exception
11260 handling, and disables it for languages like C that do not normally
11261 require it. However, you may need to enable this option when
11262 compiling C code that needs to interoperate properly with exception
11263 handlers written in C++. You may also wish to disable this option
11264 if you are compiling older C++ programs that don't use exception
11265 handling.
11266 -fnon-call-exceptions
11268 Generate code that allows trapping instructions to throw
11269 exceptions. Note that this requires platform-specific runtime
11270 support that does not exist everywhere. Moreover, it only allows
11271 trapping instructions to throw exceptions, i.e. memory references
11272 or floating-point instructions. It does not allow exceptions to be
11273 thrown from arbitrary signal handlers such as "SIGALRM".
11274 -fdelete-dead-exceptions
11276 Consider that instructions that may throw exceptions but don't
11277 otherwise contribute to the execution of the program can be
11278 optimized away. This option is enabled by default for the Ada
11279 front end, as permitted by the Ada language specification.
11280 Optimization passes that cause dead exceptions to be removed are
11281 enabled independently at different optimization levels.
11282 -funwind-tables
11284 Similar to -fexceptions, except that it just generates any needed
11285 static data, but does not affect the generated code in any other
11286 way. You normally do not need to enable this option; instead, a
11287 language processor that needs this handling enables it on your
11288 behalf.
11289 -fasynchronous-unwind-tables
11291 Generate unwind table in DWARF format, if supported by target
11292 machine. The table is exact at each instruction boundary, so it
11293 can be used for stack unwinding from asynchronous events (such as
11294 debugger or garbage collector).
11295 -fno-gnu-unique
11297 On systems with recent GNU assembler and C library, the C++
11298 compiler uses the "STB_GNU_UNIQUE" binding to make sure that
11299 definitions of template static data members and static local
11300 variables in inline functions are unique even in the presence of
11301 "RTLD_LOCAL"; this is necessary to avoid problems with a library
11302 used by two different "RTLD_LOCAL" plugins depending on a
11303 definition in one of them and therefore disagreeing with the other
11304 one about the binding of the symbol. But this causes "dlclose" to
11305 be ignored for affected DSOs; if your program relies on
11306 reinitialization of a DSO via "dlclose" and "dlopen", you can use
11307 -fno-gnu-unique.
11308 -fpcc-struct-return
11310 Return "short" "struct" and "union" values in memory like longer
11311 ones, rather than in registers. This convention is less efficient,
11312 but it has the advantage of allowing intercallability between GCC-
11313 compiled files and files compiled with other compilers,
11314 particularly the Portable C Compiler (pcc).
11315 The precise convention for returning structures in memory depends
11317 on the target configuration macros.
11318 Short structures and unions are those whose size and alignment
11320 match that of some integer type.
11321 Warning: code compiled with the -fpcc-struct-return switch is not
11323 binary compatible with code compiled with the -freg-struct-return
11324 switch. Use it to conform to a non-default application binary
11325 interface.
11326 -freg-struct-return
11328 Return "struct" and "union" values in registers when possible.
11329 This is more efficient for small structures than
11330 -fpcc-struct-return.
11331 If you specify neither -fpcc-struct-return nor -freg-struct-return,
11333 GCC defaults to whichever convention is standard for the target.
11334 If there is no standard convention, GCC defaults to
11335 -fpcc-struct-return, except on targets where GCC is the principal
11336 compiler. In those cases, we can choose the standard, and we chose
11337 the more efficient register return alternative.
11338 Warning: code compiled with the -freg-struct-return switch is not
11340 binary compatible with code compiled with the -fpcc-struct-return
11341 switch. Use it to conform to a non-default application binary
11342 interface.
11343 -fshort-enums
11345 Allocate to an "enum" type only as many bytes as it needs for the
11346 declared range of possible values. Specifically, the "enum" type
11347 is equivalent to the smallest integer type that has enough room.
11348 Warning: the -fshort-enums switch causes GCC to generate code that
11350 is not binary compatible with code generated without that switch.
11351 Use it to conform to a non-default application binary interface.
11352 -fshort-wchar
11354 Override the underlying type for "wchar_t" to be "short unsigned
11355 int" instead of the default for the target. This option is useful
11356 for building programs to run under WINE.
11357 Warning: the -fshort-wchar switch causes GCC to generate code that
11359 is not binary compatible with code generated without that switch.
11360 Use it to conform to a non-default application binary interface.
11361 -fno-common
11363 In C code, this option controls the placement of global variables
11364 defined without an initializer, known as tentative definitions in
11365 the C standard. Tentative definitions are distinct from
11366 declarations of a variable with the "extern" keyword, which do not
11367 allocate storage.
11368 Unix C compilers have traditionally allocated storage for
11370 uninitialized global variables in a common block. This allows the
11371 linker to resolve all tentative definitions of the same variable in
11372 different compilation units to the same object, or to a non-
11373 tentative definition. This is the behavior specified by -fcommon,
11374 and is the default for GCC on most targets. On the other hand,
11375 this behavior is not required by ISO C, and on some targets may
11376 carry a speed or code size penalty on variable references.
11377 The -fno-common option specifies that the compiler should instead
11379 place uninitialized global variables in the data section of the
11380 object file. This inhibits the merging of tentative definitions by
11381 the linker so you get a multiple-definition error if the same
11382 variable is defined in more than one compilation unit. Compiling
11383 with -fno-common is useful on targets for which it provides better
11384 performance, or if you wish to verify that the program will work on
11385 other systems that always treat uninitialized variable definitions
11386 this way.
11387 -fno-ident
11389 Ignore the "#ident" directive.
11390 -finhibit-size-directive
11392 Don't output a ".size" assembler directive, or anything else that
11393 would cause trouble if the function is split in the middle, and the
11394 two halves are placed at locations far apart in memory. This
11395 option is used when compiling crtstuff.c; you should not need to
11396 use it for anything else.
11397 -fverbose-asm
11399 Put extra commentary information in the generated assembly code to
11400 make it more readable. This option is generally only of use to
11401 those who actually need to read the generated assembly code
11402 (perhaps while debugging the compiler itself).
11403 -fno-verbose-asm, the default, causes the extra information to be
11405 omitted and is useful when comparing two assembler files.
11406 The added comments include:
11408 * information on the compiler version and command-line options,
11410 * the source code lines associated with the assembly
11412 instructions, in the form FILENAME:LINENUMBER:CONTENT OF LINE,
11413 * hints on which high-level expressions correspond to the various
11415 assembly instruction operands.
11416 For example, given this C source file:
11418 int test (int n)
11420 {
11421 int i;
11422 int total = 0;
11423 for (i = 0; i < n; i++)
11425 total += i * i;
11426 return total;
11428 }
11429 compiling to (x86_64) assembly via -S and emitting the result
11431 direct to stdout via -o -
11432 gcc -S test.c -fverbose-asm -Os -o -
11434 gives output similar to this:
11436 .file "test.c"
11438 # GNU C11 (GCC) version 7.0.0 20160809 (experimental) (x86_64-pc-linux-gnu)
11439 [...snip...]
11440 # options passed:
11441 [...snip...]
11442 .text
11444 .globl test
11445 .type test, @function
11446 test:
11447 .LFB0:
11448 .cfi_startproc
11449 # test.c:4: int total = 0;
11450 xorl %eax, %eax # <retval>
11451 # test.c:6: for (i = 0; i < n; i++)
11452 xorl %edx, %edx # i
11453 .L2:
11454 # test.c:6: for (i = 0; i < n; i++)
11455 cmpl %edi, %edx # n, i
11456 jge .L5 #,
11457 # test.c:7: total += i * i;
11458 movl %edx, %ecx # i, tmp92
11459 imull %edx, %ecx # i, tmp92
11460 # test.c:6: for (i = 0; i < n; i++)
11461 incl %edx # i
11462 # test.c:7: total += i * i;
11463 addl %ecx, %eax # tmp92, <retval>
11464 jmp .L2 #
11465 .L5:
11466 # test.c:10: }
11467 ret
11468 .cfi_endproc
11469 .LFE0:
11470 .size test, .-test
11471 .ident "GCC: (GNU) 7.0.0 20160809 (experimental)"
11472 .section .note.GNU-stack,"",@progbits
11473 The comments are intended for humans rather than machines and hence
11475 the precise format of the comments is subject to change.
11476 -frecord-gcc-switches
11478 This switch causes the command line used to invoke the compiler to
11479 be recorded into the object file that is being created. This
11480 switch is only implemented on some targets and the exact format of
11481 the recording is target and binary file format dependent, but it
11482 usually takes the form of a section containing ASCII text. This
11483 switch is related to the -fverbose-asm switch, but that switch only
11484 records information in the assembler output file as comments, so it
11485 never reaches the object file. See also -grecord-gcc-switches for
11486 another way of storing compiler options into the object file.
11487 -fpic
11489 Generate position-independent code (PIC) suitable for use in a
11490 shared library, if supported for the target machine. Such code
11491 accesses all constant addresses through a global offset table
11492 (GOT). The dynamic loader resolves the GOT entries when the
11493 program starts (the dynamic loader is not part of GCC; it is part
11494 of the operating system). If the GOT size for the linked
11495 executable exceeds a machine-specific maximum size, you get an
11496 error message from the linker indicating that -fpic does not work;
11497 in that case, recompile with -fPIC instead. (These maximums are 8k
11498 on the SPARC, 28k on AArch64 and 32k on the m68k and RS/6000. The
11499 x86 has no such limit.)
11500 Position-independent code requires special support, and therefore
11502 works only on certain machines. For the x86, GCC supports PIC for
11503 System V but not for the Sun 386i. Code generated for the IBM
11504 RS/6000 is always position-independent.
11505 When this flag is set, the macros "__pic__" and "__PIC__" are
11507 defined to 1.
11508 -fPIC
11510 If supported for the target machine, emit position-independent
11511 code, suitable for dynamic linking and avoiding any limit on the
11512 size of the global offset table. This option makes a difference on
11513 AArch64, m68k, PowerPC and SPARC.
11514 Position-independent code requires special support, and therefore
11516 works only on certain machines.
11517 When this flag is set, the macros "__pic__" and "__PIC__" are
11519 defined to 2.
11520 -fpie
11522 -fPIE
11523 These options are similar to -fpic and -fPIC, but generated
11524 position independent code can be only linked into executables.
11525 Usually these options are used when -pie GCC option is used during
11526 linking.
11527 -fpie and -fPIE both define the macros "__pie__" and "__PIE__".
11529 The macros have the value 1 for -fpie and 2 for -fPIE.
11530 -fno-plt
11532 Do not use the PLT for external function calls in position-
11533 independent code. Instead, load the callee address at call sites
11534 from the GOT and branch to it. This leads to more efficient code
11535 by eliminating PLT stubs and exposing GOT loads to optimizations.
11536 On architectures such as 32-bit x86 where PLT stubs expect the GOT
11537 pointer in a specific register, this gives more register allocation
11538 freedom to the compiler. Lazy binding requires use of the PLT;
11539 with -fno-plt all external symbols are resolved at load time.
11540 Alternatively, the function attribute "noplt" can be used to avoid
11542 calls through the PLT for specific external functions.
11543 In position-dependent code, a few targets also convert calls to
11545 functions that are marked to not use the PLT to use the GOT
11546 instead.
11547 -fno-jump-tables
11549 Do not use jump tables for switch statements even where it would be
11550 more efficient than other code generation strategies. This option
11551 is of use in conjunction with -fpic or -fPIC for building code that
11552 forms part of a dynamic linker and cannot reference the address of
11553 a jump table. On some targets, jump tables do not require a GOT
11554 and this option is not needed.
11555 -ffixed-reg
11557 Treat the register named reg as a fixed register; generated code
11558 should never refer to it (except perhaps as a stack pointer, frame
11559 pointer or in some other fixed role).
11560 reg must be the name of a register. The register names accepted
11562 are machine-specific and are defined in the "REGISTER_NAMES" macro
11563 in the machine description macro file.
11564 This flag does not have a negative form, because it specifies a
11566 three-way choice.
11567 -fcall-used-reg
11569 Treat the register named reg as an allocable register that is
11570 clobbered by function calls. It may be allocated for temporaries
11571 or variables that do not live across a call. Functions compiled
11572 this way do not save and restore the register reg.
11573 It is an error to use this flag with the frame pointer or stack
11575 pointer. Use of this flag for other registers that have fixed
11576 pervasive roles in the machine's execution model produces
11577 disastrous results.
11578 This flag does not have a negative form, because it specifies a
11580 three-way choice.
11581 -fcall-saved-reg
11583 Treat the register named reg as an allocable register saved by
11584 functions. It may be allocated even for temporaries or variables
11585 that live across a call. Functions compiled this way save and
11586 restore the register reg if they use it.
11587 It is an error to use this flag with the frame pointer or stack
11589 pointer. Use of this flag for other registers that have fixed
11590 pervasive roles in the machine's execution model produces
11591 disastrous results.
11592 A different sort of disaster results from the use of this flag for
11594 a register in which function values may be returned.
11595 This flag does not have a negative form, because it specifies a
11597 three-way choice.
11598 -fpack-struct[=n]
11600 Without a value specified, pack all structure members together
11601 without holes. When a value is specified (which must be a small
11602 power of two), pack structure members according to this value,
11603 representing the maximum alignment (that is, objects with default
11604 alignment requirements larger than this are output potentially
11605 unaligned at the next fitting location.
11606 Warning: the -fpack-struct switch causes GCC to generate code that
11608 is not binary compatible with code generated without that switch.
11609 Additionally, it makes the code suboptimal. Use it to conform to a
11610 non-default application binary interface.
11611 -fleading-underscore
11613 This option and its counterpart, -fno-leading-underscore, forcibly
11614 change the way C symbols are represented in the object file. One
11615 use is to help link with legacy assembly code.
11616 Warning: the -fleading-underscore switch causes GCC to generate
11618 code that is not binary compatible with code generated without that
11619 switch. Use it to conform to a non-default application binary
11620 interface. Not all targets provide complete support for this
11621 switch.
11622 -ftls-model=model
11624 Alter the thread-local storage model to be used. The model
11625 argument should be one of global-dynamic, local-dynamic, initial-
11626 exec or local-exec. Note that the choice is subject to
11627 optimization: the compiler may use a more efficient model for
11628 symbols not visible outside of the translation unit, or if -fpic is
11629 not given on the command line.
11630 The default without -fpic is initial-exec; with -fpic the default
11632 is global-dynamic.
11633 -ftrampolines
11635 For targets that normally need trampolines for nested functions,
11636 always generate them instead of using descriptors. Otherwise, for
11637 targets that do not need them, like for example HP-PA or IA-64, do
11638 nothing.
11639 A trampoline is a small piece of code that is created at run time
11641 on the stack when the address of a nested function is taken, and is
11642 used to call the nested function indirectly. Therefore, it
11643 requires the stack to be made executable in order for the program
11644 to work properly.
11645 -fno-trampolines is enabled by default on a language by language
11647 basis to let the compiler avoid generating them, if it computes
11648 that this is safe, and replace them with descriptors. Descriptors
11649 are made up of data only, but the generated code must be prepared
11650 to deal with them. As of this writing, -fno-trampolines is enabled
11651 by default only for Ada.
11652 Moreover, code compiled with -ftrampolines and code compiled with
11654 -fno-trampolines are not binary compatible if nested functions are
11655 present. This option must therefore be used on a program-wide
11656 basis and be manipulated with extreme care.
11657 -fvisibility=[default|internal|hidden|protected]
11659 Set the default ELF image symbol visibility to the specified
11660 option---all symbols are marked with this unless overridden within
11661 the code. Using this feature can very substantially improve
11662 linking and load times of shared object libraries, produce more
11663 optimized code, provide near-perfect API export and prevent symbol
11664 clashes. It is strongly recommended that you use this in any
11665 shared objects you distribute.
11666 Despite the nomenclature, default always means public; i.e.,
11668 available to be linked against from outside the shared object.
11669 protected and internal are pretty useless in real-world usage so
11670 the only other commonly used option is hidden. The default if
11671 -fvisibility isn't specified is default, i.e., make every symbol
11672 public.
11673 A good explanation of the benefits offered by ensuring ELF symbols
11675 have the correct visibility is given by "How To Write Shared
11676 Libraries" by Ulrich Drepper (which can be found at
11677 <https://www.akkadia.org/drepper/>)---however a superior solution
11678 made possible by this option to marking things hidden when the
11679 default is public is to make the default hidden and mark things
11680 public. This is the norm with DLLs on Windows and with
11681 -fvisibility=hidden and "__attribute__ ((visibility("default")))"
11682 instead of "__declspec(dllexport)" you get almost identical
11683 semantics with identical syntax. This is a great boon to those
11684 working with cross-platform projects.
11685 For those adding visibility support to existing code, you may find
11687 "#pragma GCC visibility" of use. This works by you enclosing the
11688 declarations you wish to set visibility for with (for example)
11689 "#pragma GCC visibility push(hidden)" and "#pragma GCC visibility
11690 pop". Bear in mind that symbol visibility should be viewed as part
11691 of the API interface contract and thus all new code should always
11692 specify visibility when it is not the default; i.e., declarations
11693 only for use within the local DSO should always be marked
11694 explicitly as hidden as so to avoid PLT indirection
11695 overheads---making this abundantly clear also aids readability and
11696 self-documentation of the code. Note that due to ISO C++
11697 specification requirements, "operator new" and "operator delete"
11698 must always be of default visibility.
11699 Be aware that headers from outside your project, in particular
11701 system headers and headers from any other library you use, may not
11702 be expecting to be compiled with visibility other than the default.
11703 You may need to explicitly say "#pragma GCC visibility
11704 push(default)" before including any such headers.
11705 "extern" declarations are not affected by -fvisibility, so a lot of
11707 code can be recompiled with -fvisibility=hidden with no
11708 modifications. However, this means that calls to "extern"
11709 functions with no explicit visibility use the PLT, so it is more
11710 effective to use "__attribute ((visibility))" and/or "#pragma GCC
11711 visibility" to tell the compiler which "extern" declarations should
11712 be treated as hidden.
11713 Note that -fvisibility does affect C++ vague linkage entities. This
11715 means that, for instance, an exception class that is be thrown
11716 between DSOs must be explicitly marked with default visibility so
11717 that the type_info nodes are unified between the DSOs.
11718 An overview of these techniques, their benefits and how to use them
11720 is at <http://gcc.gnu.org/wiki/Visibility>.
11721 -fstrict-volatile-bitfields
11723 This option should be used if accesses to volatile bit-fields (or
11724 other structure fields, although the compiler usually honors those
11725 types anyway) should use a single access of the width of the
11726 field's type, aligned to a natural alignment if possible. For
11727 example, targets with memory-mapped peripheral registers might
11728 require all such accesses to be 16 bits wide; with this flag you
11729 can declare all peripheral bit-fields as "unsigned short" (assuming
11730 short is 16 bits on these targets) to force GCC to use 16-bit
11731 accesses instead of, perhaps, a more efficient 32-bit access.
11732 If this option is disabled, the compiler uses the most efficient
11734 instruction. In the previous example, that might be a 32-bit load
11735 instruction, even though that accesses bytes that do not contain
11736 any portion of the bit-field, or memory-mapped registers unrelated
11737 to the one being updated.
11738 In some cases, such as when the "packed" attribute is applied to a
11740 structure field, it may not be possible to access the field with a
11741 single read or write that is correctly aligned for the target
11742 machine. In this case GCC falls back to generating multiple
11743 accesses rather than code that will fault or truncate the result at
11744 run time.
11745 Note: Due to restrictions of the C/C++11 memory model, write
11747 accesses are not allowed to touch non bit-field members. It is
11748 therefore recommended to define all bits of the field's type as
11749 bit-field members.
11750 The default value of this option is determined by the application
11752 binary interface for the target processor.
11753 -fsync-libcalls
11755 This option controls whether any out-of-line instance of the
11756 "__sync" family of functions may be used to implement the C++11
11757 "__atomic" family of functions.
11758 The default value of this option is enabled, thus the only useful
11760 form of the option is -fno-sync-libcalls. This option is used in
11761 the implementation of the libatomic runtime library.
11762 GCC Developer Options
11764 This section describes command-line options that are primarily of
11765 interest to GCC developers, including options to support compiler
11766 testing and investigation of compiler bugs and compile-time performance
11767 problems. This includes options that produce debug dumps at various
11768 points in the compilation; that print statistics such as memory use and
11769 execution time; and that print information about GCC's configuration,
11770 such as where it searches for libraries. You should rarely need to use
11771 any of these options for ordinary compilation and linking tasks.
11772 -dletters
11774 -fdump-rtl-pass
11775 -fdump-rtl-pass=filename
11776 Says to make debugging dumps during compilation at times specified
11777 by letters. This is used for debugging the RTL-based passes of the
11778 compiler. The file names for most of the dumps are made by
11779 appending a pass number and a word to the dumpname, and the files
11780 are created in the directory of the output file. In case of
11781 =filename option, the dump is output on the given file instead of
11782 the pass numbered dump files. Note that the pass number is
11783 assigned as passes are registered into the pass manager. Most
11784 passes are registered in the order that they will execute and for
11785 these passes the number corresponds to the pass execution order.
11786 However, passes registered by plugins, passes specific to
11787 compilation targets, or passes that are otherwise registered after
11788 all the other passes are numbered higher than a pass named "final",
11789 even if they are executed earlier. dumpname is generated from the
11790 name of the output file if explicitly specified and not an
11791 executable, otherwise it is the basename of the source file.
11792 Some -dletters switches have different meaning when -E is used for
11794 preprocessing.
11795 Debug dumps can be enabled with a -fdump-rtl switch or some -d
11797 option letters. Here are the possible letters for use in pass and
11798 letters, and their meanings:
11799 -fdump-rtl-alignments
11801 Dump after branch alignments have been computed.
11802 -fdump-rtl-asmcons
11804 Dump after fixing rtl statements that have unsatisfied in/out
11805 constraints.
11806 -fdump-rtl-auto_inc_dec
11808 Dump after auto-inc-dec discovery. This pass is only run on
11809 architectures that have auto inc or auto dec instructions.
11810 -fdump-rtl-barriers
11812 Dump after cleaning up the barrier instructions.
11813 -fdump-rtl-bbpart
11815 Dump after partitioning hot and cold basic blocks.
11816 -fdump-rtl-bbro
11818 Dump after block reordering.
11819 -fdump-rtl-btl1
11821 -fdump-rtl-btl2
11822 -fdump-rtl-btl1 and -fdump-rtl-btl2 enable dumping after the
11823 two branch target load optimization passes.
11824 -fdump-rtl-bypass
11826 Dump after jump bypassing and control flow optimizations.
11827 -fdump-rtl-combine
11829 Dump after the RTL instruction combination pass.
11830 -fdump-rtl-compgotos
11832 Dump after duplicating the computed gotos.
11833 -fdump-rtl-ce1
11835 -fdump-rtl-ce2
11836 -fdump-rtl-ce3
11837 -fdump-rtl-ce1, -fdump-rtl-ce2, and -fdump-rtl-ce3 enable
11838 dumping after the three if conversion passes.
11839 -fdump-rtl-cprop_hardreg
11841 Dump after hard register copy propagation.
11842 -fdump-rtl-csa
11844 Dump after combining stack adjustments.
11845 -fdump-rtl-cse1
11847 -fdump-rtl-cse2
11848 -fdump-rtl-cse1 and -fdump-rtl-cse2 enable dumping after the
11849 two common subexpression elimination passes.
11850 -fdump-rtl-dce
11852 Dump after the standalone dead code elimination passes.
11853 -fdump-rtl-dbr
11855 Dump after delayed branch scheduling.
11856 -fdump-rtl-dce1
11858 -fdump-rtl-dce2
11859 -fdump-rtl-dce1 and -fdump-rtl-dce2 enable dumping after the
11860 two dead store elimination passes.
11861 -fdump-rtl-eh
11863 Dump after finalization of EH handling code.
11864 -fdump-rtl-eh_ranges
11866 Dump after conversion of EH handling range regions.
11867 -fdump-rtl-expand
11869 Dump after RTL generation.
11870 -fdump-rtl-fwprop1
11872 -fdump-rtl-fwprop2
11873 -fdump-rtl-fwprop1 and -fdump-rtl-fwprop2 enable dumping after
11874 the two forward propagation passes.
11875 -fdump-rtl-gcse1
11877 -fdump-rtl-gcse2
11878 -fdump-rtl-gcse1 and -fdump-rtl-gcse2 enable dumping after
11879 global common subexpression elimination.
11880 -fdump-rtl-init-regs
11882 Dump after the initialization of the registers.
11883 -fdump-rtl-initvals
11885 Dump after the computation of the initial value sets.
11886 -fdump-rtl-into_cfglayout
11888 Dump after converting to cfglayout mode.
11889 -fdump-rtl-ira
11891 Dump after iterated register allocation.
11892 -fdump-rtl-jump
11894 Dump after the second jump optimization.
11895 -fdump-rtl-loop2
11897 -fdump-rtl-loop2 enables dumping after the rtl loop
11898 optimization passes.
11899 -fdump-rtl-mach
11901 Dump after performing the machine dependent reorganization
11902 pass, if that pass exists.
11903 -fdump-rtl-mode_sw
11905 Dump after removing redundant mode switches.
11906 -fdump-rtl-rnreg
11908 Dump after register renumbering.
11909 -fdump-rtl-outof_cfglayout
11911 Dump after converting from cfglayout mode.
11912 -fdump-rtl-peephole2
11914 Dump after the peephole pass.
11915 -fdump-rtl-postreload
11917 Dump after post-reload optimizations.
11918 -fdump-rtl-pro_and_epilogue
11920 Dump after generating the function prologues and epilogues.
11921 -fdump-rtl-sched1
11923 -fdump-rtl-sched2
11924 -fdump-rtl-sched1 and -fdump-rtl-sched2 enable dumping after
11925 the basic block scheduling passes.
11926 -fdump-rtl-ree
11928 Dump after sign/zero extension elimination.
11929 -fdump-rtl-seqabstr
11931 Dump after common sequence discovery.
11932 -fdump-rtl-shorten
11934 Dump after shortening branches.
11935 -fdump-rtl-sibling
11937 Dump after sibling call optimizations.
11938 -fdump-rtl-split1
11940 -fdump-rtl-split2
11941 -fdump-rtl-split3
11942 -fdump-rtl-split4
11943 -fdump-rtl-split5
11944 These options enable dumping after five rounds of instruction
11945 splitting.
11946 -fdump-rtl-sms
11948 Dump after modulo scheduling. This pass is only run on some
11949 architectures.
11950 -fdump-rtl-stack
11952 Dump after conversion from GCC's "flat register file" registers
11953 to the x87's stack-like registers. This pass is only run on
11954 x86 variants.
11955 -fdump-rtl-subreg1
11957 -fdump-rtl-subreg2
11958 -fdump-rtl-subreg1 and -fdump-rtl-subreg2 enable dumping after
11959 the two subreg expansion passes.
11960 -fdump-rtl-unshare
11962 Dump after all rtl has been unshared.
11963 -fdump-rtl-vartrack
11965 Dump after variable tracking.
11966 -fdump-rtl-vregs
11968 Dump after converting virtual registers to hard registers.
11969 -fdump-rtl-web
11971 Dump after live range splitting.
11972 -fdump-rtl-regclass
11974 -fdump-rtl-subregs_of_mode_init
11975 -fdump-rtl-subregs_of_mode_finish
11976 -fdump-rtl-dfinit
11977 -fdump-rtl-dfinish
11978 These dumps are defined but always produce empty files.
11979 -da
11981 -fdump-rtl-all
11982 Produce all the dumps listed above.
11983 -dA Annotate the assembler output with miscellaneous debugging
11985 information.
11986 -dD Dump all macro definitions, at the end of preprocessing, in
11988 addition to normal output.
11989 -dH Produce a core dump whenever an error occurs.
11991 -dp Annotate the assembler output with a comment indicating which
11993 pattern and alternative is used. The length and cost of each
11994 instruction are also printed.
11995 -dP Dump the RTL in the assembler output as a comment before each
11997 instruction. Also turns on -dp annotation.
11998 -dx Just generate RTL for a function instead of compiling it.
12000 Usually used with -fdump-rtl-expand.
12001 -fdump-noaddr
12003 When doing debugging dumps, suppress address output. This makes it
12004 more feasible to use diff on debugging dumps for compiler
12005 invocations with different compiler binaries and/or different text
12006 / bss / data / heap / stack / dso start locations.
12007 -freport-bug
12009 Collect and dump debug information into a temporary file if an
12010 internal compiler error (ICE) occurs.
12011 -fdump-unnumbered
12013 When doing debugging dumps, suppress instruction numbers and
12014 address output. This makes it more feasible to use diff on
12015 debugging dumps for compiler invocations with different options, in
12016 particular with and without -g.
12017 -fdump-unnumbered-links
12019 When doing debugging dumps (see -d option above), suppress
12020 instruction numbers for the links to the previous and next
12021 instructions in a sequence.
12022 -fdump-ipa-switch
12024 Control the dumping at various stages of inter-procedural analysis
12025 language tree to a file. The file name is generated by appending a
12026 switch specific suffix to the source file name, and the file is
12027 created in the same directory as the output file. The following
12028 dumps are possible:
12029 all Enables all inter-procedural analysis dumps.
12031 cgraph
12033 Dumps information about call-graph optimization, unused
12034 function removal, and inlining decisions.
12035 inline
12037 Dump after function inlining.
12038 -fdump-lang-all
12040 -fdump-lang-switch
12041 -fdump-lang-switch-options
12042 -fdump-lang-switch-options=filename
12043 Control the dumping of language-specific information. The options
12044 and filename portions behave as described in the -fdump-tree
12045 option. The following switch values are accepted:
12046 all Enable all language-specific dumps.
12048 class
12050 Dump class hierarchy information. Virtual table information is
12051 emitted unless 'slim' is specified. This option is applicable
12052 to C++ only.
12053 raw Dump the raw internal tree data. This option is applicable to
12055 C++ only.
12056 -fdump-passes
12058 Print on stderr the list of optimization passes that are turned on
12059 and off by the current command-line options.
12060 -fdump-statistics-option
12062 Enable and control dumping of pass statistics in a separate file.
12063 The file name is generated by appending a suffix ending in
12064 .statistics to the source file name, and the file is created in the
12065 same directory as the output file. If the -option form is used,
12066 -stats causes counters to be summed over the whole compilation unit
12067 while -details dumps every event as the passes generate them. The
12068 default with no option is to sum counters for each function
12069 compiled.
12070 -fdump-tree-all
12072 -fdump-tree-switch
12073 -fdump-tree-switch-options
12074 -fdump-tree-switch-options=filename
12075 Control the dumping at various stages of processing the
12076 intermediate language tree to a file. The file name is generated
12077 by appending a switch-specific suffix to the source file name, and
12078 the file is created in the same directory as the output file. In
12079 case of =filename option, the dump is output on the given file
12080 instead of the auto named dump files. If the -options form is
12081 used, options is a list of - separated options which control the
12082 details of the dump. Not all options are applicable to all dumps;
12083 those that are not meaningful are ignored. The following options
12084 are available
12085 address
12087 Print the address of each node. Usually this is not meaningful
12088 as it changes according to the environment and source file.
12089 Its primary use is for tying up a dump file with a debug
12090 environment.
12091 asmname
12093 If "DECL_ASSEMBLER_NAME" has been set for a given decl, use
12094 that in the dump instead of "DECL_NAME". Its primary use is
12095 ease of use working backward from mangled names in the assembly
12096 file.
12097 slim
12099 When dumping front-end intermediate representations, inhibit
12100 dumping of members of a scope or body of a function merely
12101 because that scope has been reached. Only dump such items when
12102 they are directly reachable by some other path.
12103 When dumping pretty-printed trees, this option inhibits dumping
12105 the bodies of control structures.
12106 When dumping RTL, print the RTL in slim (condensed) form
12108 instead of the default LISP-like representation.
12109 raw Print a raw representation of the tree. By default, trees are
12111 pretty-printed into a C-like representation.
12112 details
12114 Enable more detailed dumps (not honored by every dump option).
12115 Also include information from the optimization passes.
12116 stats
12118 Enable dumping various statistics about the pass (not honored
12119 by every dump option).
12120 blocks
12122 Enable showing basic block boundaries (disabled in raw dumps).
12123 graph
12125 For each of the other indicated dump files (-fdump-rtl-pass),
12126 dump a representation of the control flow graph suitable for
12127 viewing with GraphViz to file.passid.pass.dot. Each function
12128 in the file is pretty-printed as a subgraph, so that GraphViz
12129 can render them all in a single plot.
12130 This option currently only works for RTL dumps, and the RTL is
12132 always dumped in slim form.
12133 vops
12135 Enable showing virtual operands for every statement.
12136 lineno
12138 Enable showing line numbers for statements.
12139 uid Enable showing the unique ID ("DECL_UID") for each variable.
12141 verbose
12143 Enable showing the tree dump for each statement.
12144 eh Enable showing the EH region number holding each statement.
12146 scev
12148 Enable showing scalar evolution analysis details.
12149 optimized
12151 Enable showing optimization information (only available in
12152 certain passes).
12153 missed
12155 Enable showing missed optimization information (only available
12156 in certain passes).
12157 note
12159 Enable other detailed optimization information (only available
12160 in certain passes).
12161 =filename
12163 Instead of an auto named dump file, output into the given file
12164 name. The file names stdout and stderr are treated specially
12165 and are considered already open standard streams. For example,
12166 gcc -O2 -ftree-vectorize -fdump-tree-vect-blocks=foo.dump
12168 -fdump-tree-pre=/dev/stderr file.c
12169 outputs vectorizer dump into foo.dump, while the PRE dump is
12171 output on to stderr. If two conflicting dump filenames are
12172 given for the same pass, then the latter option overrides the
12173 earlier one.
12174 all Turn on all options, except raw, slim, verbose and lineno.
12176 optall
12178 Turn on all optimization options, i.e., optimized, missed, and
12179 note.
12180 To determine what tree dumps are available or find the dump for a
12182 pass of interest follow the steps below.
12183 1. Invoke GCC with -fdump-passes and in the stderr output look for
12185 a code that corresponds to the pass you are interested in. For
12186 example, the codes "tree-evrp", "tree-vrp1", and "tree-vrp2"
12187 correspond to the three Value Range Propagation passes. The
12188 number at the end distinguishes distinct invocations of the
12189 same pass.
12190 2. To enable the creation of the dump file, append the pass code
12192 to the -fdump- option prefix and invoke GCC with it. For
12193 example, to enable the dump from the Early Value Range
12194 Propagation pass, invoke GCC with the -fdump-tree-evrp option.
12195 Optionally, you may specify the name of the dump file. If you
12196 don't specify one, GCC creates as described below.
12197 3. Find the pass dump in a file whose name is composed of three
12199 components separated by a period: the name of the source file
12200 GCC was invoked to compile, a numeric suffix indicating the
12201 pass number followed by the letter t for tree passes (and the
12202 letter r for RTL passes), and finally the pass code. For
12203 example, the Early VRP pass dump might be in a file named
12204 myfile.c.038t.evrp in the current working directory. Note that
12205 the numeric codes are not stable and may change from one
12206 version of GCC to another.
12207 -fopt-info
12209 -fopt-info-options
12210 -fopt-info-options=filename
12211 Controls optimization dumps from various optimization passes. If
12212 the -options form is used, options is a list of - separated option
12213 keywords to select the dump details and optimizations.
12214 The options can be divided into two groups: options describing the
12216 verbosity of the dump, and options describing which optimizations
12217 should be included. The options from both the groups can be freely
12218 mixed as they are non-overlapping. However, in case of any
12219 conflicts, the later options override the earlier options on the
12220 command line.
12221 The following options control the dump verbosity:
12223 optimized
12225 Print information when an optimization is successfully applied.
12226 It is up to a pass to decide which information is relevant. For
12227 example, the vectorizer passes print the source location of
12228 loops which are successfully vectorized.
12229 missed
12231 Print information about missed optimizations. Individual passes
12232 control which information to include in the output.
12233 note
12235 Print verbose information about optimizations, such as certain
12236 transformations, more detailed messages about decisions etc.
12237 all Print detailed optimization information. This includes
12239 optimized, missed, and note.
12240 One or more of the following option keywords can be used to
12242 describe a group of optimizations:
12243 ipa Enable dumps from all interprocedural optimizations.
12245 loop
12247 Enable dumps from all loop optimizations.
12248 inline
12250 Enable dumps from all inlining optimizations.
12251 omp Enable dumps from all OMP (Offloading and Multi Processing)
12253 optimizations.
12254 vec Enable dumps from all vectorization optimizations.
12256 optall
12258 Enable dumps from all optimizations. This is a superset of the
12259 optimization groups listed above.
12260 If options is omitted, it defaults to optimized-optall, which means
12262 to dump all info about successful optimizations from all the
12263 passes.
12264 If the filename is provided, then the dumps from all the applicable
12266 optimizations are concatenated into the filename. Otherwise the
12267 dump is output onto stderr. Though multiple -fopt-info options are
12268 accepted, only one of them can include a filename. If other
12269 filenames are provided then all but the first such option are
12270 ignored.
12271 Note that the output filename is overwritten in case of multiple
12273 translation units. If a combined output from multiple translation
12274 units is desired, stderr should be used instead.
12275 In the following example, the optimization info is output to
12277 stderr:
12278 gcc -O3 -fopt-info
12280 This example:
12282 gcc -O3 -fopt-info-missed=missed.all
12284 outputs missed optimization report from all the passes into
12286 missed.all, and this one:
12287 gcc -O2 -ftree-vectorize -fopt-info-vec-missed
12289 prints information about missed optimization opportunities from
12291 vectorization passes on stderr. Note that -fopt-info-vec-missed is
12292 equivalent to -fopt-info-missed-vec. The order of the optimization
12293 group names and message types listed after -fopt-info does not
12294 matter.
12295 As another example,
12297 gcc -O3 -fopt-info-inline-optimized-missed=inline.txt
12299 outputs information about missed optimizations as well as optimized
12301 locations from all the inlining passes into inline.txt.
12302 Finally, consider:
12304 gcc -fopt-info-vec-missed=vec.miss -fopt-info-loop-optimized=loop.opt
12306 Here the two output filenames vec.miss and loop.opt are in conflict
12308 since only one output file is allowed. In this case, only the first
12309 option takes effect and the subsequent options are ignored. Thus
12310 only vec.miss is produced which contains dumps from the vectorizer
12311 about missed opportunities.
12312 -fsched-verbose=n
12314 On targets that use instruction scheduling, this option controls
12315 the amount of debugging output the scheduler prints to the dump
12316 files.
12317 For n greater than zero, -fsched-verbose outputs the same
12319 information as -fdump-rtl-sched1 and -fdump-rtl-sched2. For n
12320 greater than one, it also output basic block probabilities,
12321 detailed ready list information and unit/insn info. For n greater
12322 than two, it includes RTL at abort point, control-flow and regions
12323 info. And for n over four, -fsched-verbose also includes
12324 dependence info.
12325 -fenable-kind-pass
12327 -fdisable-kind-pass=range-list
12328 This is a set of options that are used to explicitly disable/enable
12329 optimization passes. These options are intended for use for
12330 debugging GCC. Compiler users should use regular options for
12331 enabling/disabling passes instead.
12332 -fdisable-ipa-pass
12334 Disable IPA pass pass. pass is the pass name. If the same pass
12335 is statically invoked in the compiler multiple times, the pass
12336 name should be appended with a sequential number starting from
12337 1.
12338 -fdisable-rtl-pass
12340 -fdisable-rtl-pass=range-list
12341 Disable RTL pass pass. pass is the pass name. If the same
12342 pass is statically invoked in the compiler multiple times, the
12343 pass name should be appended with a sequential number starting
12344 from 1. range-list is a comma-separated list of function
12345 ranges or assembler names. Each range is a number pair
12346 separated by a colon. The range is inclusive in both ends. If
12347 the range is trivial, the number pair can be simplified as a
12348 single number. If the function's call graph node's uid falls
12349 within one of the specified ranges, the pass is disabled for
12350 that function. The uid is shown in the function header of a
12351 dump file, and the pass names can be dumped by using option
12352 -fdump-passes.
12353 -fdisable-tree-pass
12355 -fdisable-tree-pass=range-list
12356 Disable tree pass pass. See -fdisable-rtl for the description
12357 of option arguments.
12358 -fenable-ipa-pass
12360 Enable IPA pass pass. pass is the pass name. If the same pass
12361 is statically invoked in the compiler multiple times, the pass
12362 name should be appended with a sequential number starting from
12363 1.
12364 -fenable-rtl-pass
12366 -fenable-rtl-pass=range-list
12367 Enable RTL pass pass. See -fdisable-rtl for option argument
12368 description and examples.
12369 -fenable-tree-pass
12371 -fenable-tree-pass=range-list
12372 Enable tree pass pass. See -fdisable-rtl for the description
12373 of option arguments.
12374 Here are some examples showing uses of these options.
12376 # disable ccp1 for all functions
12378 -fdisable-tree-ccp1
12379 # disable complete unroll for function whose cgraph node uid is 1
12380 -fenable-tree-cunroll=1
12381 # disable gcse2 for functions at the following ranges [1,1],
12382 # [300,400], and [400,1000]
12383 # disable gcse2 for functions foo and foo2
12384 -fdisable-rtl-gcse2=foo,foo2
12385 # disable early inlining
12386 -fdisable-tree-einline
12387 # disable ipa inlining
12388 -fdisable-ipa-inline
12389 # enable tree full unroll
12390 -fenable-tree-unroll
12391 -fchecking
12393 -fchecking=n
12394 Enable internal consistency checking. The default depends on the
12395 compiler configuration. -fchecking=2 enables further internal
12396 consistency checking that might affect code generation.
12397 -frandom-seed=string
12399 This option provides a seed that GCC uses in place of random
12400 numbers in generating certain symbol names that have to be
12401 different in every compiled file. It is also used to place unique
12402 stamps in coverage data files and the object files that produce
12403 them. You can use the -frandom-seed option to produce reproducibly
12404 identical object files.
12405 The string can either be a number (decimal, octal or hex) or an
12407 arbitrary string (in which case it's converted to a number by
12408 computing CRC32).
12409 The string should be different for every file you compile.
12411 -save-temps
12413 -save-temps=cwd
12414 Store the usual "temporary" intermediate files permanently; place
12415 them in the current directory and name them based on the source
12416 file. Thus, compiling foo.c with -c -save-temps produces files
12417 foo.i and foo.s, as well as foo.o. This creates a preprocessed
12418 foo.i output file even though the compiler now normally uses an
12419 integrated preprocessor.
12420 When used in combination with the -x command-line option,
12422 -save-temps is sensible enough to avoid over writing an input
12423 source file with the same extension as an intermediate file. The
12424 corresponding intermediate file may be obtained by renaming the
12425 source file before using -save-temps.
12426 If you invoke GCC in parallel, compiling several different source
12428 files that share a common base name in different subdirectories or
12429 the same source file compiled for multiple output destinations, it
12430 is likely that the different parallel compilers will interfere with
12431 each other, and overwrite the temporary files. For instance:
12432 gcc -save-temps -o outdir1/foo.o indir1/foo.c&
12434 gcc -save-temps -o outdir2/foo.o indir2/foo.c&
12435 may result in foo.i and foo.o being written to simultaneously by
12437 both compilers.
12438 -save-temps=obj
12440 Store the usual "temporary" intermediate files permanently. If the
12441 -o option is used, the temporary files are based on the object
12442 file. If the -o option is not used, the -save-temps=obj switch
12443 behaves like -save-temps.
12444 For example:
12446 gcc -save-temps=obj -c foo.c
12448 gcc -save-temps=obj -c bar.c -o dir/xbar.o
12449 gcc -save-temps=obj foobar.c -o dir2/yfoobar
12450 creates foo.i, foo.s, dir/xbar.i, dir/xbar.s, dir2/yfoobar.i,
12452 dir2/yfoobar.s, and dir2/yfoobar.o.
12453 -time[=file]
12455 Report the CPU time taken by each subprocess in the compilation
12456 sequence. For C source files, this is the compiler proper and
12457 assembler (plus the linker if linking is done).
12458 Without the specification of an output file, the output looks like
12460 this:
12461 # cc1 0.12 0.01
12463 # as 0.00 0.01
12464 The first number on each line is the "user time", that is time
12466 spent executing the program itself. The second number is "system
12467 time", time spent executing operating system routines on behalf of
12468 the program. Both numbers are in seconds.
12469 With the specification of an output file, the output is appended to
12471 the named file, and it looks like this:
12472 0.12 0.01 cc1 <options>
12474 0.00 0.01 as <options>
12475 The "user time" and the "system time" are moved before the program
12477 name, and the options passed to the program are displayed, so that
12478 one can later tell what file was being compiled, and with which
12479 options.
12480 -fdump-final-insns[=file]
12482 Dump the final internal representation (RTL) to file. If the
12483 optional argument is omitted (or if file is "."), the name of the
12484 dump file is determined by appending ".gkd" to the compilation
12485 output file name.
12486 -fcompare-debug[=opts]
12488 If no error occurs during compilation, run the compiler a second
12489 time, adding opts and -fcompare-debug-second to the arguments
12490 passed to the second compilation. Dump the final internal
12491 representation in both compilations, and print an error if they
12492 differ.
12493 If the equal sign is omitted, the default -gtoggle is used.
12495 The environment variable GCC_COMPARE_DEBUG, if defined, non-empty
12497 and nonzero, implicitly enables -fcompare-debug. If
12498 GCC_COMPARE_DEBUG is defined to a string starting with a dash, then
12499 it is used for opts, otherwise the default -gtoggle is used.
12500 -fcompare-debug=, with the equal sign but without opts, is
12502 equivalent to -fno-compare-debug, which disables the dumping of the
12503 final representation and the second compilation, preventing even
12504 GCC_COMPARE_DEBUG from taking effect.
12505 To verify full coverage during -fcompare-debug testing, set
12507 GCC_COMPARE_DEBUG to say -fcompare-debug-not-overridden, which GCC
12508 rejects as an invalid option in any actual compilation (rather than
12509 preprocessing, assembly or linking). To get just a warning,
12510 setting GCC_COMPARE_DEBUG to -w%n-fcompare-debug not overridden
12511 will do.
12512 -fcompare-debug-second
12514 This option is implicitly passed to the compiler for the second
12515 compilation requested by -fcompare-debug, along with options to
12516 silence warnings, and omitting other options that would cause the
12517 compiler to produce output to files or to standard output as a side
12518 effect. Dump files and preserved temporary files are renamed so as
12519 to contain the ".gk" additional extension during the second
12520 compilation, to avoid overwriting those generated by the first.
12521 When this option is passed to the compiler driver, it causes the
12523 first compilation to be skipped, which makes it useful for little
12524 other than debugging the compiler proper.
12525 -gtoggle
12527 Turn off generation of debug info, if leaving out this option
12528 generates it, or turn it on at level 2 otherwise. The position of
12529 this argument in the command line does not matter; it takes effect
12530 after all other options are processed, and it does so only once, no
12531 matter how many times it is given. This is mainly intended to be
12532 used with -fcompare-debug.
12533 -fvar-tracking-assignments-toggle
12535 Toggle -fvar-tracking-assignments, in the same way that -gtoggle
12536 toggles -g.
12537 -Q Makes the compiler print out each function name as it is compiled,
12539 and print some statistics about each pass when it finishes.
12540 -ftime-report
12542 Makes the compiler print some statistics about the time consumed by
12543 each pass when it finishes.
12544 -ftime-report-details
12546 Record the time consumed by infrastructure parts separately for
12547 each pass.
12548 -fira-verbose=n
12550 Control the verbosity of the dump file for the integrated register
12551 allocator. The default value is 5. If the value n is greater or
12552 equal to 10, the dump output is sent to stderr using the same
12553 format as n minus 10.
12554 -flto-report
12556 Prints a report with internal details on the workings of the link-
12557 time optimizer. The contents of this report vary from version to
12558 version. It is meant to be useful to GCC developers when
12559 processing object files in LTO mode (via -flto).
12560 Disabled by default.
12562 -flto-report-wpa
12564 Like -flto-report, but only print for the WPA phase of Link Time
12565 Optimization.
12566 -fmem-report
12568 Makes the compiler print some statistics about permanent memory
12569 allocation when it finishes.
12570 -fmem-report-wpa
12572 Makes the compiler print some statistics about permanent memory
12573 allocation for the WPA phase only.
12574 -fpre-ipa-mem-report
12576 -fpost-ipa-mem-report
12577 Makes the compiler print some statistics about permanent memory
12578 allocation before or after interprocedural optimization.
12579 -fprofile-report
12581 Makes the compiler print some statistics about consistency of the
12582 (estimated) profile and effect of individual passes.
12583 -fstack-usage
12585 Makes the compiler output stack usage information for the program,
12586 on a per-function basis. The filename for the dump is made by
12587 appending .su to the auxname. auxname is generated from the name
12588 of the output file, if explicitly specified and it is not an
12589 executable, otherwise it is the basename of the source file. An
12590 entry is made up of three fields:
12591 * The name of the function.
12593 * A number of bytes.
12595 * One or more qualifiers: "static", "dynamic", "bounded".
12597 The qualifier "static" means that the function manipulates the
12599 stack statically: a fixed number of bytes are allocated for the
12600 frame on function entry and released on function exit; no stack
12601 adjustments are otherwise made in the function. The second field
12602 is this fixed number of bytes.
12603 The qualifier "dynamic" means that the function manipulates the
12605 stack dynamically: in addition to the static allocation described
12606 above, stack adjustments are made in the body of the function, for
12607 example to push/pop arguments around function calls. If the
12608 qualifier "bounded" is also present, the amount of these
12609 adjustments is bounded at compile time and the second field is an
12610 upper bound of the total amount of stack used by the function. If
12611 it is not present, the amount of these adjustments is not bounded
12612 at compile time and the second field only represents the bounded
12613 part.
12614 -fstats
12616 Emit statistics about front-end processing at the end of the
12617 compilation. This option is supported only by the C++ front end,
12618 and the information is generally only useful to the G++ development
12619 team.
12620 -fdbg-cnt-list
12622 Print the name and the counter upper bound for all debug counters.
12623 -fdbg-cnt=counter-value-list
12625 Set the internal debug counter upper bound. counter-value-list is
12626 a comma-separated list of name:value pairs which sets the upper
12627 bound of each debug counter name to value. All debug counters have
12628 the initial upper bound of "UINT_MAX"; thus "dbg_cnt" returns true
12629 always unless the upper bound is set by this option. For example,
12630 with -fdbg-cnt=dce:10,tail_call:0, "dbg_cnt(dce)" returns true only
12631 for first 10 invocations.
12632 -print-file-name=library
12634 Print the full absolute name of the library file library that would
12635 be used when linking---and don't do anything else. With this
12636 option, GCC does not compile or link anything; it just prints the
12637 file name.
12638 -print-multi-directory
12640 Print the directory name corresponding to the multilib selected by
12641 any other switches present in the command line. This directory is
12642 supposed to exist in GCC_EXEC_PREFIX.
12643 -print-multi-lib
12645 Print the mapping from multilib directory names to compiler
12646 switches that enable them. The directory name is separated from
12647 the switches by ;, and each switch starts with an @ instead of the
12648 -, without spaces between multiple switches. This is supposed to
12649 ease shell processing.
12650 -print-multi-os-directory
12652 Print the path to OS libraries for the selected multilib, relative
12653 to some lib subdirectory. If OS libraries are present in the lib
12654 subdirectory and no multilibs are used, this is usually just ., if
12655 OS libraries are present in libsuffix sibling directories this
12656 prints e.g. ../lib64, ../lib or ../lib32, or if OS libraries are
12657 present in lib/subdir subdirectories it prints e.g. amd64, sparcv9
12658 or ev6.
12659 -print-multiarch
12661 Print the path to OS libraries for the selected multiarch, relative
12662 to some lib subdirectory.
12663 -print-prog-name=program
12665 Like -print-file-name, but searches for a program such as cpp.
12666 -print-libgcc-file-name
12668 Same as -print-file-name=libgcc.a.
12669 This is useful when you use -nostdlib or -nodefaultlibs but you do
12671 want to link with libgcc.a. You can do:
12672 gcc -nostdlib <files>... `gcc -print-libgcc-file-name`
12674 -print-search-dirs
12676 Print the name of the configured installation directory and a list
12677 of program and library directories gcc searches---and don't do
12678 anything else.
12679 This is useful when gcc prints the error message installation
12681 problem, cannot exec cpp0: No such file or directory. To resolve
12682 this you either need to put cpp0 and the other compiler components
12683 where gcc expects to find them, or you can set the environment
12684 variable GCC_EXEC_PREFIX to the directory where you installed them.
12685 Don't forget the trailing /.
12686 -print-sysroot
12688 Print the target sysroot directory that is used during compilation.
12689 This is the target sysroot specified either at configure time or
12690 using the --sysroot option, possibly with an extra suffix that
12691 depends on compilation options. If no target sysroot is specified,
12692 the option prints nothing.
12693 -print-sysroot-headers-suffix
12695 Print the suffix added to the target sysroot when searching for
12696 headers, or give an error if the compiler is not configured with
12697 such a suffix---and don't do anything else.
12698 -dumpmachine
12700 Print the compiler's target machine (for example,
12701 i686-pc-linux-gnu)---and don't do anything else.
12702 -dumpversion
12704 Print the compiler version (for example, 3.0, 6.3.0 or 7)---and
12705 don't do anything else. This is the compiler version used in
12706 filesystem paths, specs, can be depending on how the compiler has
12707 been configured just a single number (major version), two numbers
12708 separated by dot (major and minor version) or three numbers
12709 separated by dots (major, minor and patchlevel version).
12710 -dumpfullversion
12712 Print the full compiler version, always 3 numbers separated by
12713 dots, major, minor and patchlevel version.
12714 -dumpspecs
12716 Print the compiler's built-in specs---and don't do anything else.
12717 (This is used when GCC itself is being built.)
12718 Machine-Dependent Options
12720 Each target machine supported by GCC can have its own options---for
12721 example, to allow you to compile for a particular processor variant or
12722 ABI, or to control optimizations specific to that machine. By
12723 convention, the names of machine-specific options start with -m.
12724 Some configurations of the compiler also support additional target-
12726 specific options, usually for compatibility with other compilers on the
12727 same platform.
12728 AArch64 Options
12730 These options are defined for AArch64 implementations:
12731 -mabi=name
12733 Generate code for the specified data model. Permissible values are
12734 ilp32 for SysV-like data model where int, long int and pointers are
12735 32 bits, and lp64 for SysV-like data model where int is 32 bits,
12736 but long int and pointers are 64 bits.
12737 The default depends on the specific target configuration. Note
12739 that the LP64 and ILP32 ABIs are not link-compatible; you must
12740 compile your entire program with the same ABI, and link with a
12741 compatible set of libraries.
12742 -mbig-endian
12744 Generate big-endian code. This is the default when GCC is
12745 configured for an aarch64_be-*-* target.
12746 -mgeneral-regs-only
12748 Generate code which uses only the general-purpose registers. This
12749 will prevent the compiler from using floating-point and Advanced
12750 SIMD registers but will not impose any restrictions on the
12751 assembler.
12752 -mlittle-endian
12754 Generate little-endian code. This is the default when GCC is
12755 configured for an aarch64-*-* but not an aarch64_be-*-* target.
12756 -mcmodel=tiny
12758 Generate code for the tiny code model. The program and its
12759 statically defined symbols must be within 1MB of each other.
12760 Programs can be statically or dynamically linked.
12761 -mcmodel=small
12763 Generate code for the small code model. The program and its
12764 statically defined symbols must be within 4GB of each other.
12765 Programs can be statically or dynamically linked. This is the
12766 default code model.
12767 -mcmodel=large
12769 Generate code for the large code model. This makes no assumptions
12770 about addresses and sizes of sections. Programs can be statically
12771 linked only.
12772 -mstrict-align
12774 Avoid generating memory accesses that may not be aligned on a
12775 natural object boundary as described in the architecture
12776 specification.
12777 -momit-leaf-frame-pointer
12779 -mno-omit-leaf-frame-pointer
12780 Omit or keep the frame pointer in leaf functions. The former
12781 behavior is the default.
12782 -mtls-dialect=desc
12784 Use TLS descriptors as the thread-local storage mechanism for
12785 dynamic accesses of TLS variables. This is the default.
12786 -mtls-dialect=traditional
12788 Use traditional TLS as the thread-local storage mechanism for
12789 dynamic accesses of TLS variables.
12790 -mtls-size=size
12792 Specify bit size of immediate TLS offsets. Valid values are 12,
12793 24, 32, 48. This option requires binutils 2.26 or newer.
12794 -mfix-cortex-a53-835769
12796 -mno-fix-cortex-a53-835769
12797 Enable or disable the workaround for the ARM Cortex-A53 erratum
12798 number 835769. This involves inserting a NOP instruction between
12799 memory instructions and 64-bit integer multiply-accumulate
12800 instructions.
12801 -mfix-cortex-a53-843419
12803 -mno-fix-cortex-a53-843419
12804 Enable or disable the workaround for the ARM Cortex-A53 erratum
12805 number 843419. This erratum workaround is made at link time and
12806 this will only pass the corresponding flag to the linker.
12807 -mlow-precision-recip-sqrt
12809 -mno-low-precision-recip-sqrt
12810 Enable or disable the reciprocal square root approximation. This
12811 option only has an effect if -ffast-math or
12812 -funsafe-math-optimizations is used as well. Enabling this reduces
12813 precision of reciprocal square root results to about 16 bits for
12814 single precision and to 32 bits for double precision.
12815 -mlow-precision-sqrt
12817 -mno-low-precision-sqrt
12818 Enable or disable the square root approximation. This option only
12819 has an effect if -ffast-math or -funsafe-math-optimizations is used
12820 as well. Enabling this reduces precision of square root results to
12821 about 16 bits for single precision and to 32 bits for double
12822 precision. If enabled, it implies -mlow-precision-recip-sqrt.
12823 -mlow-precision-div
12825 -mno-low-precision-div
12826 Enable or disable the division approximation. This option only has
12827 an effect if -ffast-math or -funsafe-math-optimizations is used as
12828 well. Enabling this reduces precision of division results to about
12829 16 bits for single precision and to 32 bits for double precision.
12830 -march=name
12832 Specify the name of the target architecture and, optionally, one or
12833 more feature modifiers. This option has the form
12834 -march=arch{+[no]feature}*.
12835 The permissible values for arch are armv8-a, armv8.1-a, armv8.2-a,
12837 armv8.3-a or armv8.4-a or native.
12838 The value armv8.4-a implies armv8.3-a and enables compiler support
12840 for the ARMv8.4-A architecture extensions.
12841 The value armv8.3-a implies armv8.2-a and enables compiler support
12843 for the ARMv8.3-A architecture extensions.
12844 The value armv8.2-a implies armv8.1-a and enables compiler support
12846 for the ARMv8.2-A architecture extensions.
12847 The value armv8.1-a implies armv8-a and enables compiler support
12849 for the ARMv8.1-A architecture extension. In particular, it
12850 enables the +crc, +lse, and +rdma features.
12851 The value native is available on native AArch64 GNU/Linux and
12853 causes the compiler to pick the architecture of the host system.
12854 This option has no effect if the compiler is unable to recognize
12855 the architecture of the host system,
12856 The permissible values for feature are listed in the sub-section on
12858 aarch64-feature-modifiers,,-march and -mcpu Feature Modifiers.
12859 Where conflicting feature modifiers are specified, the right-most
12860 feature is used.
12861 GCC uses name to determine what kind of instructions it can emit
12863 when generating assembly code. If -march is specified without
12864 either of -mtune or -mcpu also being specified, the code is tuned
12865 to perform well across a range of target processors implementing
12866 the target architecture.
12867 -mtune=name
12869 Specify the name of the target processor for which GCC should tune
12870 the performance of the code. Permissible values for this option
12871 are: generic, cortex-a35, cortex-a53, cortex-a55, cortex-a57,
12872 cortex-a72, cortex-a73, cortex-a75, exynos-m1, falkor, qdf24xx,
12873 saphira, xgene1, vulcan, thunderx, thunderxt88, thunderxt88p1,
12874 thunderxt81, thunderxt83, thunderx2t99, cortex-a57.cortex-a53,
12875 cortex-a72.cortex-a53, cortex-a73.cortex-a35,
12876 cortex-a73.cortex-a53, cortex-a75.cortex-a55, native.
12877 The values cortex-a57.cortex-a53, cortex-a72.cortex-a53,
12879 cortex-a73.cortex-a35, cortex-a73.cortex-a53, cortex-a75.cortex-a55
12880 specify that GCC should tune for a big.LITTLE system.
12881 Additionally on native AArch64 GNU/Linux systems the value native
12883 tunes performance to the host system. This option has no effect if
12884 the compiler is unable to recognize the processor of the host
12885 system.
12886 Where none of -mtune=, -mcpu= or -march= are specified, the code is
12888 tuned to perform well across a range of target processors.
12889 This option cannot be suffixed by feature modifiers.
12891 -mcpu=name
12893 Specify the name of the target processor, optionally suffixed by
12894 one or more feature modifiers. This option has the form
12895 -mcpu=cpu{+[no]feature}*, where the permissible values for cpu are
12896 the same as those available for -mtune. The permissible values for
12897 feature are documented in the sub-section on
12898 aarch64-feature-modifiers,,-march and -mcpu Feature Modifiers.
12899 Where conflicting feature modifiers are specified, the right-most
12900 feature is used.
12901 GCC uses name to determine what kind of instructions it can emit
12903 when generating assembly code (as if by -march) and to determine
12904 the target processor for which to tune for performance (as if by
12905 -mtune). Where this option is used in conjunction with -march or
12906 -mtune, those options take precedence over the appropriate part of
12907 this option.
12908 -moverride=string
12910 Override tuning decisions made by the back-end in response to a
12911 -mtune= switch. The syntax, semantics, and accepted values for
12912 string in this option are not guaranteed to be consistent across
12913 releases.
12914 This option is only intended to be useful when developing GCC.
12916 -mverbose-cost-dump
12918 Enable verbose cost model dumping in the debug dump files. This
12919 option is provided for use in debugging the compiler.
12920 -mpc-relative-literal-loads
12922 -mno-pc-relative-literal-loads
12923 Enable or disable PC-relative literal loads. With this option
12924 literal pools are accessed using a single instruction and emitted
12925 after each function. This limits the maximum size of functions to
12926 1MB. This is enabled by default for -mcmodel=tiny.
12927 -msign-return-address=scope
12929 Select the function scope on which return address signing will be
12930 applied. Permissible values are none, which disables return
12931 address signing, non-leaf, which enables pointer signing for
12932 functions which are not leaf functions, and all, which enables
12933 pointer signing for all functions. The default value is none.
12934 -msve-vector-bits=bits
12936 Specify the number of bits in an SVE vector register. This option
12937 only has an effect when SVE is enabled.
12938 GCC supports two forms of SVE code generation: "vector-length
12940 agnostic" output that works with any size of vector register and
12941 "vector-length specific" output that only works when the vector
12942 registers are a particular size. Replacing bits with scalable
12943 selects vector-length agnostic output while replacing it with a
12944 number selects vector-length specific output. The possible lengths
12945 in the latter case are: 128, 256, 512, 1024 and 2048. scalable is
12946 the default.
12947 At present, -msve-vector-bits=128 produces the same output as
12949 -msve-vector-bits=scalable.
12950 -march and -mcpu Feature Modifiers
12952 Feature modifiers used with -march and -mcpu can be any of the
12954 following and their inverses nofeature:
12955 crc Enable CRC extension. This is on by default for -march=armv8.1-a.
12957 crypto
12959 Enable Crypto extension. This also enables Advanced SIMD and
12960 floating-point instructions.
12961 fp Enable floating-point instructions. This is on by default for all
12963 possible values for options -march and -mcpu.
12964 simd
12966 Enable Advanced SIMD instructions. This also enables floating-
12967 point instructions. This is on by default for all possible values
12968 for options -march and -mcpu.
12969 sve Enable Scalable Vector Extension instructions. This also enables
12971 Advanced SIMD and floating-point instructions.
12972 lse Enable Large System Extension instructions. This is on by default
12974 for -march=armv8.1-a.
12975 rdma
12977 Enable Round Double Multiply Accumulate instructions. This is on
12978 by default for -march=armv8.1-a.
12979 fp16
12981 Enable FP16 extension. This also enables floating-point
12982 instructions.
12983 fp16fml
12985 Enable FP16 fmla extension. This also enables FP16 extensions and
12986 floating-point instructions. This option is enabled by default for
12987 -march=armv8.4-a. Use of this option with architectures prior to
12988 Armv8.2-A is not supported.
12989 rcpc
12991 Enable the RcPc extension. This does not change code generation
12992 from GCC, but is passed on to the assembler, enabling inline asm
12993 statements to use instructions from the RcPc extension.
12994 dotprod
12996 Enable the Dot Product extension. This also enables Advanced SIMD
12997 instructions.
12998 aes Enable the Armv8-a aes and pmull crypto extension. This also
13000 enables Advanced SIMD instructions.
13001 sha2
13003 Enable the Armv8-a sha2 crypto extension. This also enables
13004 Advanced SIMD instructions.
13005 sha3
13007 Enable the sha512 and sha3 crypto extension. This also enables
13008 Advanced SIMD instructions. Use of this option with architectures
13009 prior to Armv8.2-A is not supported.
13010 sm4 Enable the sm3 and sm4 crypto extension. This also enables
13012 Advanced SIMD instructions. Use of this option with architectures
13013 prior to Armv8.2-A is not supported.
13014 Feature crypto implies aes, sha2, and simd, which implies fp.
13016 Conversely, nofp implies nosimd, which implies nocrypto, noaes and
13017 nosha2.
13018 Adapteva Epiphany Options
13020 These -m options are defined for Adapteva Epiphany:
13021 -mhalf-reg-file
13023 Don't allocate any register in the range "r32"..."r63". That
13024 allows code to run on hardware variants that lack these registers.
13025 -mprefer-short-insn-regs
13027 Preferentially allocate registers that allow short instruction
13028 generation. This can result in increased instruction count, so
13029 this may either reduce or increase overall code size.
13030 -mbranch-cost=num
13032 Set the cost of branches to roughly num "simple" instructions.
13033 This cost is only a heuristic and is not guaranteed to produce
13034 consistent results across releases.
13035 -mcmove
13037 Enable the generation of conditional moves.
13038 -mnops=num
13040 Emit num NOPs before every other generated instruction.
13041 -mno-soft-cmpsf
13043 For single-precision floating-point comparisons, emit an "fsub"
13044 instruction and test the flags. This is faster than a software
13045 comparison, but can get incorrect results in the presence of NaNs,
13046 or when two different small numbers are compared such that their
13047 difference is calculated as zero. The default is -msoft-cmpsf,
13048 which uses slower, but IEEE-compliant, software comparisons.
13049 -mstack-offset=num
13051 Set the offset between the top of the stack and the stack pointer.
13052 E.g., a value of 8 means that the eight bytes in the range
13053 "sp+0...sp+7" can be used by leaf functions without stack
13054 allocation. Values other than 8 or 16 are untested and unlikely to
13055 work. Note also that this option changes the ABI; compiling a
13056 program with a different stack offset than the libraries have been
13057 compiled with generally does not work. This option can be useful
13058 if you want to evaluate if a different stack offset would give you
13059 better code, but to actually use a different stack offset to build
13060 working programs, it is recommended to configure the toolchain with
13061 the appropriate --with-stack-offset=num option.
13062 -mno-round-nearest
13064 Make the scheduler assume that the rounding mode has been set to
13065 truncating. The default is -mround-nearest.
13066 -mlong-calls
13068 If not otherwise specified by an attribute, assume all calls might
13069 be beyond the offset range of the "b" / "bl" instructions, and
13070 therefore load the function address into a register before
13071 performing a (otherwise direct) call. This is the default.
13072 -mshort-calls
13074 If not otherwise specified by an attribute, assume all direct calls
13075 are in the range of the "b" / "bl" instructions, so use these
13076 instructions for direct calls. The default is -mlong-calls.
13077 -msmall16
13079 Assume addresses can be loaded as 16-bit unsigned values. This
13080 does not apply to function addresses for which -mlong-calls
13081 semantics are in effect.
13082 -mfp-mode=mode
13084 Set the prevailing mode of the floating-point unit. This
13085 determines the floating-point mode that is provided and expected at
13086 function call and return time. Making this mode match the mode you
13087 predominantly need at function start can make your programs smaller
13088 and faster by avoiding unnecessary mode switches.
13089 mode can be set to one the following values:
13091 caller
13093 Any mode at function entry is valid, and retained or restored
13094 when the function returns, and when it calls other functions.
13095 This mode is useful for compiling libraries or other
13096 compilation units you might want to incorporate into different
13097 programs with different prevailing FPU modes, and the
13098 convenience of being able to use a single object file outweighs
13099 the size and speed overhead for any extra mode switching that
13100 might be needed, compared with what would be needed with a more
13101 specific choice of prevailing FPU mode.
13102 truncate
13104 This is the mode used for floating-point calculations with
13105 truncating (i.e. round towards zero) rounding mode. That
13106 includes conversion from floating point to integer.
13107 round-nearest
13109 This is the mode used for floating-point calculations with
13110 round-to-nearest-or-even rounding mode.
13111 int This is the mode used to perform integer calculations in the
13113 FPU, e.g. integer multiply, or integer multiply-and-
13114 accumulate.
13115 The default is -mfp-mode=caller
13117 -mnosplit-lohi
13119 -mno-postinc
13120 -mno-postmodify
13121 Code generation tweaks that disable, respectively, splitting of
13122 32-bit loads, generation of post-increment addresses, and
13123 generation of post-modify addresses. The defaults are msplit-lohi,
13124 -mpost-inc, and -mpost-modify.
13125 -mnovect-double
13127 Change the preferred SIMD mode to SImode. The default is
13128 -mvect-double, which uses DImode as preferred SIMD mode.
13129 -max-vect-align=num
13131 The maximum alignment for SIMD vector mode types. num may be 4 or
13132 8. The default is 8. Note that this is an ABI change, even though
13133 many library function interfaces are unaffected if they don't use
13134 SIMD vector modes in places that affect size and/or alignment of
13135 relevant types.
13136 -msplit-vecmove-early
13138 Split vector moves into single word moves before reload. In theory
13139 this can give better register allocation, but so far the reverse
13140 seems to be generally the case.
13141 -m1reg-reg
13143 Specify a register to hold the constant -1, which makes loading
13144 small negative constants and certain bitmasks faster. Allowable
13145 values for reg are r43 and r63, which specify use of that register
13146 as a fixed register, and none, which means that no register is used
13147 for this purpose. The default is -m1reg-none.
13148 ARC Options
13150 The following options control the architecture variant for which code
13151 is being compiled:
13152 -mbarrel-shifter
13154 Generate instructions supported by barrel shifter. This is the
13155 default unless -mcpu=ARC601 or -mcpu=ARCEM is in effect.
13156 -mjli-always
13158 Force to call a function using jli_s instruction. This option is
13159 valid only for ARCv2 architecture.
13160 -mcpu=cpu
13162 Set architecture type, register usage, and instruction scheduling
13163 parameters for cpu. There are also shortcut alias options
13164 available for backward compatibility and convenience. Supported
13165 values for cpu are
13166 arc600
13168 Compile for ARC600. Aliases: -mA6, -mARC600.
13169 arc601
13171 Compile for ARC601. Alias: -mARC601.
13172 arc700
13174 Compile for ARC700. Aliases: -mA7, -mARC700. This is the
13175 default when configured with --with-cpu=arc700.
13176 arcem
13178 Compile for ARC EM.
13179 archs
13181 Compile for ARC HS.
13182 em Compile for ARC EM CPU with no hardware extensions.
13184 em4 Compile for ARC EM4 CPU.
13186 em4_dmips
13188 Compile for ARC EM4 DMIPS CPU.
13189 em4_fpus
13191 Compile for ARC EM4 DMIPS CPU with the single-precision
13192 floating-point extension.
13193 em4_fpuda
13195 Compile for ARC EM4 DMIPS CPU with single-precision floating-
13196 point and double assist instructions.
13197 hs Compile for ARC HS CPU with no hardware extensions except the
13199 atomic instructions.
13200 hs34
13202 Compile for ARC HS34 CPU.
13203 hs38
13205 Compile for ARC HS38 CPU.
13206 hs38_linux
13208 Compile for ARC HS38 CPU with all hardware extensions on.
13209 arc600_norm
13211 Compile for ARC 600 CPU with "norm" instructions enabled.
13212 arc600_mul32x16
13214 Compile for ARC 600 CPU with "norm" and 32x16-bit multiply
13215 instructions enabled.
13216 arc600_mul64
13218 Compile for ARC 600 CPU with "norm" and "mul64"-family
13219 instructions enabled.
13220 arc601_norm
13222 Compile for ARC 601 CPU with "norm" instructions enabled.
13223 arc601_mul32x16
13225 Compile for ARC 601 CPU with "norm" and 32x16-bit multiply
13226 instructions enabled.
13227 arc601_mul64
13229 Compile for ARC 601 CPU with "norm" and "mul64"-family
13230 instructions enabled.
13231 nps400
13233 Compile for ARC 700 on NPS400 chip.
13234 em_mini
13236 Compile for ARC EM minimalist configuration featuring reduced
13237 register set.
13238 -mdpfp
13240 -mdpfp-compact
13241 Generate double-precision FPX instructions, tuned for the compact
13242 implementation.
13243 -mdpfp-fast
13245 Generate double-precision FPX instructions, tuned for the fast
13246 implementation.
13247 -mno-dpfp-lrsr
13249 Disable "lr" and "sr" instructions from using FPX extension aux
13250 registers.
13251 -mea
13253 Generate extended arithmetic instructions. Currently only "divaw",
13254 "adds", "subs", and "sat16" are supported. This is always enabled
13255 for -mcpu=ARC700.
13256 -mno-mpy
13258 Do not generate "mpy"-family instructions for ARC700. This option
13259 is deprecated.
13260 -mmul32x16
13262 Generate 32x16-bit multiply and multiply-accumulate instructions.
13263 -mmul64
13265 Generate "mul64" and "mulu64" instructions. Only valid for
13266 -mcpu=ARC600.
13267 -mnorm
13269 Generate "norm" instructions. This is the default if -mcpu=ARC700
13270 is in effect.
13271 -mspfp
13273 -mspfp-compact
13274 Generate single-precision FPX instructions, tuned for the compact
13275 implementation.
13276 -mspfp-fast
13278 Generate single-precision FPX instructions, tuned for the fast
13279 implementation.
13280 -msimd
13282 Enable generation of ARC SIMD instructions via target-specific
13283 builtins. Only valid for -mcpu=ARC700.
13284 -msoft-float
13286 This option ignored; it is provided for compatibility purposes
13287 only. Software floating-point code is emitted by default, and this
13288 default can overridden by FPX options; -mspfp, -mspfp-compact, or
13289 -mspfp-fast for single precision, and -mdpfp, -mdpfp-compact, or
13290 -mdpfp-fast for double precision.
13291 -mswap
13293 Generate "swap" instructions.
13294 -matomic
13296 This enables use of the locked load/store conditional extension to
13297 implement atomic memory built-in functions. Not available for ARC
13298 6xx or ARC EM cores.
13299 -mdiv-rem
13301 Enable "div" and "rem" instructions for ARCv2 cores.
13302 -mcode-density
13304 Enable code density instructions for ARC EM. This option is on by
13305 default for ARC HS.
13306 -mll64
13308 Enable double load/store operations for ARC HS cores.
13309 -mtp-regno=regno
13311 Specify thread pointer register number.
13312 -mmpy-option=multo
13314 Compile ARCv2 code with a multiplier design option. You can
13315 specify the option using either a string or numeric value for
13316 multo. wlh1 is the default value. The recognized values are:
13317 0
13319 none
13320 No multiplier available.
13321 1
13323 w 16x16 multiplier, fully pipelined. The following instructions
13324 are enabled: "mpyw" and "mpyuw".
13325 2
13327 wlh1
13328 32x32 multiplier, fully pipelined (1 stage). The following
13329 instructions are additionally enabled: "mpy", "mpyu", "mpym",
13330 "mpymu", and "mpy_s".
13331 3
13333 wlh2
13334 32x32 multiplier, fully pipelined (2 stages). The following
13335 instructions are additionally enabled: "mpy", "mpyu", "mpym",
13336 "mpymu", and "mpy_s".
13337 4
13339 wlh3
13340 Two 16x16 multipliers, blocking, sequential. The following
13341 instructions are additionally enabled: "mpy", "mpyu", "mpym",
13342 "mpymu", and "mpy_s".
13343 5
13345 wlh4
13346 One 16x16 multiplier, blocking, sequential. The following
13347 instructions are additionally enabled: "mpy", "mpyu", "mpym",
13348 "mpymu", and "mpy_s".
13349 6
13351 wlh5
13352 One 32x4 multiplier, blocking, sequential. The following
13353 instructions are additionally enabled: "mpy", "mpyu", "mpym",
13354 "mpymu", and "mpy_s".
13355 7
13357 plus_dmpy
13358 ARC HS SIMD support.
13359 8
13361 plus_macd
13362 ARC HS SIMD support.
13363 9
13365 plus_qmacw
13366 ARC HS SIMD support.
13367 This option is only available for ARCv2 cores.
13369 -mfpu=fpu
13371 Enables support for specific floating-point hardware extensions for
13372 ARCv2 cores. Supported values for fpu are:
13373 fpus
13375 Enables support for single-precision floating-point hardware
13376 extensions.
13377 fpud
13379 Enables support for double-precision floating-point hardware
13380 extensions. The single-precision floating-point extension is
13381 also enabled. Not available for ARC EM.
13382 fpuda
13384 Enables support for double-precision floating-point hardware
13385 extensions using double-precision assist instructions. The
13386 single-precision floating-point extension is also enabled.
13387 This option is only available for ARC EM.
13388 fpuda_div
13390 Enables support for double-precision floating-point hardware
13391 extensions using double-precision assist instructions. The
13392 single-precision floating-point, square-root, and divide
13393 extensions are also enabled. This option is only available for
13394 ARC EM.
13395 fpuda_fma
13397 Enables support for double-precision floating-point hardware
13398 extensions using double-precision assist instructions. The
13399 single-precision floating-point and fused multiply and add
13400 hardware extensions are also enabled. This option is only
13401 available for ARC EM.
13402 fpuda_all
13404 Enables support for double-precision floating-point hardware
13405 extensions using double-precision assist instructions. All
13406 single-precision floating-point hardware extensions are also
13407 enabled. This option is only available for ARC EM.
13408 fpus_div
13410 Enables support for single-precision floating-point, square-
13411 root and divide hardware extensions.
13412 fpud_div
13414 Enables support for double-precision floating-point, square-
13415 root and divide hardware extensions. This option includes
13416 option fpus_div. Not available for ARC EM.
13417 fpus_fma
13419 Enables support for single-precision floating-point and fused
13420 multiply and add hardware extensions.
13421 fpud_fma
13423 Enables support for double-precision floating-point and fused
13424 multiply and add hardware extensions. This option includes
13425 option fpus_fma. Not available for ARC EM.
13426 fpus_all
13428 Enables support for all single-precision floating-point
13429 hardware extensions.
13430 fpud_all
13432 Enables support for all single- and double-precision floating-
13433 point hardware extensions. Not available for ARC EM.
13434 -mirq-ctrl-saved=register-range, blink, lp_count
13436 Specifies general-purposes registers that the processor
13437 automatically saves/restores on interrupt entry and exit.
13438 register-range is specified as two registers separated by a dash.
13439 The register range always starts with "r0", the upper limit is "fp"
13440 register. blink and lp_count are optional. This option is only
13441 valid for ARC EM and ARC HS cores.
13442 -mrgf-banked-regs=number
13444 Specifies the number of registers replicated in second register
13445 bank on entry to fast interrupt. Fast interrupts are interrupts
13446 with the highest priority level P0. These interrupts save only PC
13447 and STATUS32 registers to avoid memory transactions during
13448 interrupt entry and exit sequences. Use this option when you are
13449 using fast interrupts in an ARC V2 family processor. Permitted
13450 values are 4, 8, 16, and 32.
13451 -mlpc-width=width
13453 Specify the width of the "lp_count" register. Valid values for
13454 width are 8, 16, 20, 24, 28 and 32 bits. The default width is
13455 fixed to 32 bits. If the width is less than 32, the compiler does
13456 not attempt to transform loops in your program to use the zero-
13457 delay loop mechanism unless it is known that the "lp_count"
13458 register can hold the required loop-counter value. Depending on
13459 the width specified, the compiler and run-time library might
13460 continue to use the loop mechanism for various needs. This option
13461 defines macro "__ARC_LPC_WIDTH__" with the value of width.
13462 -mrf16
13464 This option instructs the compiler to generate code for a 16-entry
13465 register file. This option defines the "__ARC_RF16__" preprocessor
13466 macro.
13467 The following options are passed through to the assembler, and also
13469 define preprocessor macro symbols.
13470 -mdsp-packa
13472 Passed down to the assembler to enable the DSP Pack A extensions.
13473 Also sets the preprocessor symbol "__Xdsp_packa". This option is
13474 deprecated.
13475 -mdvbf
13477 Passed down to the assembler to enable the dual Viterbi butterfly
13478 extension. Also sets the preprocessor symbol "__Xdvbf". This
13479 option is deprecated.
13480 -mlock
13482 Passed down to the assembler to enable the locked load/store
13483 conditional extension. Also sets the preprocessor symbol
13484 "__Xlock".
13485 -mmac-d16
13487 Passed down to the assembler. Also sets the preprocessor symbol
13488 "__Xxmac_d16". This option is deprecated.
13489 -mmac-24
13491 Passed down to the assembler. Also sets the preprocessor symbol
13492 "__Xxmac_24". This option is deprecated.
13493 -mrtsc
13495 Passed down to the assembler to enable the 64-bit time-stamp
13496 counter extension instruction. Also sets the preprocessor symbol
13497 "__Xrtsc". This option is deprecated.
13498 -mswape
13500 Passed down to the assembler to enable the swap byte ordering
13501 extension instruction. Also sets the preprocessor symbol
13502 "__Xswape".
13503 -mtelephony
13505 Passed down to the assembler to enable dual- and single-operand
13506 instructions for telephony. Also sets the preprocessor symbol
13507 "__Xtelephony". This option is deprecated.
13508 -mxy
13510 Passed down to the assembler to enable the XY memory extension.
13511 Also sets the preprocessor symbol "__Xxy".
13512 The following options control how the assembly code is annotated:
13514 -misize
13516 Annotate assembler instructions with estimated addresses.
13517 -mannotate-align
13519 Explain what alignment considerations lead to the decision to make
13520 an instruction short or long.
13521 The following options are passed through to the linker:
13523 -marclinux
13525 Passed through to the linker, to specify use of the "arclinux"
13526 emulation. This option is enabled by default in tool chains built
13527 for "arc-linux-uclibc" and "arceb-linux-uclibc" targets when
13528 profiling is not requested.
13529 -marclinux_prof
13531 Passed through to the linker, to specify use of the "arclinux_prof"
13532 emulation. This option is enabled by default in tool chains built
13533 for "arc-linux-uclibc" and "arceb-linux-uclibc" targets when
13534 profiling is requested.
13535 The following options control the semantics of generated code:
13537 -mlong-calls
13539 Generate calls as register indirect calls, thus providing access to
13540 the full 32-bit address range.
13541 -mmedium-calls
13543 Don't use less than 25-bit addressing range for calls, which is the
13544 offset available for an unconditional branch-and-link instruction.
13545 Conditional execution of function calls is suppressed, to allow use
13546 of the 25-bit range, rather than the 21-bit range with conditional
13547 branch-and-link. This is the default for tool chains built for
13548 "arc-linux-uclibc" and "arceb-linux-uclibc" targets.
13549 -G num
13551 Put definitions of externally-visible data in a small data section
13552 if that data is no bigger than num bytes. The default value of num
13553 is 4 for any ARC configuration, or 8 when we have double load/store
13554 operations.
13555 -mno-sdata
13557 Do not generate sdata references. This is the default for tool
13558 chains built for "arc-linux-uclibc" and "arceb-linux-uclibc"
13559 targets.
13560 -mvolatile-cache
13562 Use ordinarily cached memory accesses for volatile references.
13563 This is the default.
13564 -mno-volatile-cache
13566 Enable cache bypass for volatile references.
13567 The following options fine tune code generation:
13569 -malign-call
13571 Do alignment optimizations for call instructions.
13572 -mauto-modify-reg
13574 Enable the use of pre/post modify with register displacement.
13575 -mbbit-peephole
13577 Enable bbit peephole2.
13578 -mno-brcc
13580 This option disables a target-specific pass in arc_reorg to
13581 generate compare-and-branch ("brcc") instructions. It has no
13582 effect on generation of these instructions driven by the combiner
13583 pass.
13584 -mcase-vector-pcrel
13586 Use PC-relative switch case tables to enable case table shortening.
13587 This is the default for -Os.
13588 -mcompact-casesi
13590 Enable compact "casesi" pattern. This is the default for -Os, and
13591 only available for ARCv1 cores.
13592 -mno-cond-exec
13594 Disable the ARCompact-specific pass to generate conditional
13595 execution instructions.
13596 Due to delay slot scheduling and interactions between operand
13598 numbers, literal sizes, instruction lengths, and the support for
13599 conditional execution, the target-independent pass to generate
13600 conditional execution is often lacking, so the ARC port has kept a
13601 special pass around that tries to find more conditional execution
13602 generation opportunities after register allocation, branch
13603 shortening, and delay slot scheduling have been done. This pass
13604 generally, but not always, improves performance and code size, at
13605 the cost of extra compilation time, which is why there is an option
13606 to switch it off. If you have a problem with call instructions
13607 exceeding their allowable offset range because they are
13608 conditionalized, you should consider using -mmedium-calls instead.
13609 -mearly-cbranchsi
13611 Enable pre-reload use of the "cbranchsi" pattern.
13612 -mexpand-adddi
13614 Expand "adddi3" and "subdi3" at RTL generation time into "add.f",
13615 "adc" etc. This option is deprecated.
13616 -mindexed-loads
13618 Enable the use of indexed loads. This can be problematic because
13619 some optimizers then assume that indexed stores exist, which is not
13620 the case.
13621 -mlra
13623 Enable Local Register Allocation. This is still experimental for
13624 ARC, so by default the compiler uses standard reload (i.e.
13625 -mno-lra).
13626 -mlra-priority-none
13628 Don't indicate any priority for target registers.
13629 -mlra-priority-compact
13631 Indicate target register priority for r0..r3 / r12..r15.
13632 -mlra-priority-noncompact
13634 Reduce target register priority for r0..r3 / r12..r15.
13635 -mno-millicode
13637 When optimizing for size (using -Os), prologues and epilogues that
13638 have to save or restore a large number of registers are often
13639 shortened by using call to a special function in libgcc; this is
13640 referred to as a millicode call. As these calls can pose
13641 performance issues, and/or cause linking issues when linking in a
13642 nonstandard way, this option is provided to turn off millicode call
13643 generation.
13644 -mmixed-code
13646 Tweak register allocation to help 16-bit instruction generation.
13647 This generally has the effect of decreasing the average instruction
13648 size while increasing the instruction count.
13649 -mq-class
13651 Enable q instruction alternatives. This is the default for -Os.
13652 -mRcq
13654 Enable Rcq constraint handling. Most short code generation depends
13655 on this. This is the default.
13656 -mRcw
13658 Enable Rcw constraint handling. Most ccfsm condexec mostly depends
13659 on this. This is the default.
13660 -msize-level=level
13662 Fine-tune size optimization with regards to instruction lengths and
13663 alignment. The recognized values for level are:
13664 0 No size optimization. This level is deprecated and treated
13666 like 1.
13667 1 Short instructions are used opportunistically.
13669 2 In addition, alignment of loops and of code after barriers are
13671 dropped.
13672 3 In addition, optional data alignment is dropped, and the option
13674 Os is enabled.
13675 This defaults to 3 when -Os is in effect. Otherwise, the behavior
13677 when this is not set is equivalent to level 1.
13678 -mtune=cpu
13680 Set instruction scheduling parameters for cpu, overriding any
13681 implied by -mcpu=.
13682 Supported values for cpu are
13684 ARC600
13686 Tune for ARC600 CPU.
13687 ARC601
13689 Tune for ARC601 CPU.
13690 ARC700
13692 Tune for ARC700 CPU with standard multiplier block.
13693 ARC700-xmac
13695 Tune for ARC700 CPU with XMAC block.
13696 ARC725D
13698 Tune for ARC725D CPU.
13699 ARC750D
13701 Tune for ARC750D CPU.
13702 -mmultcost=num
13704 Cost to assume for a multiply instruction, with 4 being equal to a
13705 normal instruction.
13706 -munalign-prob-threshold=probability
13708 Set probability threshold for unaligning branches. When tuning for
13709 ARC700 and optimizing for speed, branches without filled delay slot
13710 are preferably emitted unaligned and long, unless profiling
13711 indicates that the probability for the branch to be taken is below
13712 probability. The default is (REG_BR_PROB_BASE/2), i.e. 5000.
13713 The following options are maintained for backward compatibility, but
13715 are now deprecated and will be removed in a future release:
13716 -margonaut
13718 Obsolete FPX.
13719 -mbig-endian
13721 -EB Compile code for big-endian targets. Use of these options is now
13722 deprecated. Big-endian code is supported by configuring GCC to
13723 build "arceb-elf32" and "arceb-linux-uclibc" targets, for which big
13724 endian is the default.
13725 -mlittle-endian
13727 -EL Compile code for little-endian targets. Use of these options is
13728 now deprecated. Little-endian code is supported by configuring GCC
13729 to build "arc-elf32" and "arc-linux-uclibc" targets, for which
13730 little endian is the default.
13731 -mbarrel_shifter
13733 Replaced by -mbarrel-shifter.
13734 -mdpfp_compact
13736 Replaced by -mdpfp-compact.
13737 -mdpfp_fast
13739 Replaced by -mdpfp-fast.
13740 -mdsp_packa
13742 Replaced by -mdsp-packa.
13743 -mEA
13745 Replaced by -mea.
13746 -mmac_24
13748 Replaced by -mmac-24.
13749 -mmac_d16
13751 Replaced by -mmac-d16.
13752 -mspfp_compact
13754 Replaced by -mspfp-compact.
13755 -mspfp_fast
13757 Replaced by -mspfp-fast.
13758 -mtune=cpu
13760 Values arc600, arc601, arc700 and arc700-xmac for cpu are replaced
13761 by ARC600, ARC601, ARC700 and ARC700-xmac respectively.
13762 -multcost=num
13764 Replaced by -mmultcost.
13765 ARM Options
13767 These -m options are defined for the ARM port:
13768 -mabi=name
13770 Generate code for the specified ABI. Permissible values are: apcs-
13771 gnu, atpcs, aapcs, aapcs-linux and iwmmxt.
13772 -mapcs-frame
13774 Generate a stack frame that is compliant with the ARM Procedure
13775 Call Standard for all functions, even if this is not strictly
13776 necessary for correct execution of the code. Specifying
13777 -fomit-frame-pointer with this option causes the stack frames not
13778 to be generated for leaf functions. The default is
13779 -mno-apcs-frame. This option is deprecated.
13780 -mapcs
13782 This is a synonym for -mapcs-frame and is deprecated.
13783 -mthumb-interwork
13785 Generate code that supports calling between the ARM and Thumb
13786 instruction sets. Without this option, on pre-v5 architectures,
13787 the two instruction sets cannot be reliably used inside one
13788 program. The default is -mno-thumb-interwork, since slightly
13789 larger code is generated when -mthumb-interwork is specified. In
13790 AAPCS configurations this option is meaningless.
13791 -mno-sched-prolog
13793 Prevent the reordering of instructions in the function prologue, or
13794 the merging of those instruction with the instructions in the
13795 function's body. This means that all functions start with a
13796 recognizable set of instructions (or in fact one of a choice from a
13797 small set of different function prologues), and this information
13798 can be used to locate the start of functions inside an executable
13799 piece of code. The default is -msched-prolog.
13800 -mfloat-abi=name
13802 Specifies which floating-point ABI to use. Permissible values are:
13803 soft, softfp and hard.
13804 Specifying soft causes GCC to generate output containing library
13806 calls for floating-point operations. softfp allows the generation
13807 of code using hardware floating-point instructions, but still uses
13808 the soft-float calling conventions. hard allows generation of
13809 floating-point instructions and uses FPU-specific calling
13810 conventions.
13811 The default depends on the specific target configuration. Note
13813 that the hard-float and soft-float ABIs are not link-compatible;
13814 you must compile your entire program with the same ABI, and link
13815 with a compatible set of libraries.
13816 -mlittle-endian
13818 Generate code for a processor running in little-endian mode. This
13819 is the default for all standard configurations.
13820 -mbig-endian
13822 Generate code for a processor running in big-endian mode; the
13823 default is to compile code for a little-endian processor.
13824 -mbe8
13826 -mbe32
13827 When linking a big-endian image select between BE8 and BE32
13828 formats. The option has no effect for little-endian images and is
13829 ignored. The default is dependent on the selected target
13830 architecture. For ARMv6 and later architectures the default is
13831 BE8, for older architectures the default is BE32. BE32 format has
13832 been deprecated by ARM.
13833 -march=name[+extension...]
13835 This specifies the name of the target ARM architecture. GCC uses
13836 this name to determine what kind of instructions it can emit when
13837 generating assembly code. This option can be used in conjunction
13838 with or instead of the -mcpu= option.
13839 Permissible names are: armv4t, armv5t, armv5te, armv6, armv6j,
13841 armv6k, armv6kz, armv6t2, armv6z, armv6zk, armv7, armv7-a, armv7ve,
13842 armv8-a, armv8.1-a, armv8.2-a, armv8.3-a, armv8.4-a, armv7-r,
13843 armv8-r, armv6-m, armv6s-m, armv7-m, armv7e-m, armv8-m.base,
13844 armv8-m.main, iwmmxt and iwmmxt2.
13845 Additionally, the following architectures, which lack support for
13847 the Thumb execution state, are recognized but support is
13848 deprecated: armv2, armv2a, armv3, armv3m, armv4, armv5 and armv5e.
13849 Many of the architectures support extensions. These can be added
13851 by appending +extension to the architecture name. Extension
13852 options are processed in order and capabilities accumulate. An
13853 extension will also enable any necessary base extensions upon which
13854 it depends. For example, the +crypto extension will always enable
13855 the +simd extension. The exception to the additive construction is
13856 for extensions that are prefixed with +no...: these extensions
13857 disable the specified option and any other extensions that may
13858 depend on the presence of that extension.
13859 For example, -march=armv7-a+simd+nofp+vfpv4 is equivalent to
13861 writing -march=armv7-a+vfpv4 since the +simd option is entirely
13862 disabled by the +nofp option that follows it.
13863 Most extension names are generically named, but have an effect that
13865 is dependent upon the architecture to which it is applied. For
13866 example, the +simd option can be applied to both armv7-a and
13867 armv8-a architectures, but will enable the original ARMv7-A
13868 Advanced SIMD (Neon) extensions for armv7-a and the ARMv8-A variant
13869 for armv8-a.
13870 The table below lists the supported extensions for each
13872 architecture. Architectures not mentioned do not support any
13873 extensions.
13874 armv5e
13876 armv5te
13877 armv6
13878 armv6j
13879 armv6k
13880 armv6kz
13881 armv6t2
13882 armv6z
13883 armv6zk
13884 +fp The VFPv2 floating-point instructions. The extension
13885 +vfpv2 can be used as an alias for this extension.
13886 +nofp
13888 Disable the floating-point instructions.
13889 armv7
13891 The common subset of the ARMv7-A, ARMv7-R and ARMv7-M
13892 architectures.
13893 +fp The VFPv3 floating-point instructions, with 16 double-
13895 precision registers. The extension +vfpv3-d16 can be used
13896 as an alias for this extension. Note that floating-point
13897 is not supported by the base ARMv7-M architecture, but is
13898 compatible with both the ARMv7-A and ARMv7-R architectures.
13899 +nofp
13901 Disable the floating-point instructions.
13902 armv7-a
13904 +fp The VFPv3 floating-point instructions, with 16 double-
13905 precision registers. The extension +vfpv3-d16 can be used
13906 as an alias for this extension.
13907 +simd
13909 The Advanced SIMD (Neon) v1 and the VFPv3 floating-point
13910 instructions. The extensions +neon and +neon-vfpv3 can be
13911 used as aliases for this extension.
13912 +vfpv3
13914 The VFPv3 floating-point instructions, with 32 double-
13915 precision registers.
13916 +vfpv3-d16-fp16
13918 The VFPv3 floating-point instructions, with 16 double-
13919 precision registers and the half-precision floating-point
13920 conversion operations.
13921 +vfpv3-fp16
13923 The VFPv3 floating-point instructions, with 32 double-
13924 precision registers and the half-precision floating-point
13925 conversion operations.
13926 +vfpv4-d16
13928 The VFPv4 floating-point instructions, with 16 double-
13929 precision registers.
13930 +vfpv4
13932 The VFPv4 floating-point instructions, with 32 double-
13933 precision registers.
13934 +neon-fp16
13936 The Advanced SIMD (Neon) v1 and the VFPv3 floating-point
13937 instructions, with the half-precision floating-point
13938 conversion operations.
13939 +neon-vfpv4
13941 The Advanced SIMD (Neon) v2 and the VFPv4 floating-point
13942 instructions.
13943 +nosimd
13945 Disable the Advanced SIMD instructions (does not disable
13946 floating point).
13947 +nofp
13949 Disable the floating-point and Advanced SIMD instructions.
13950 armv7ve
13952 The extended version of the ARMv7-A architecture with support
13953 for virtualization.
13954 +fp The VFPv4 floating-point instructions, with 16 double-
13956 precision registers. The extension +vfpv4-d16 can be used
13957 as an alias for this extension.
13958 +simd
13960 The Advanced SIMD (Neon) v2 and the VFPv4 floating-point
13961 instructions. The extension +neon-vfpv4 can be used as an
13962 alias for this extension.
13963 +vfpv3-d16
13965 The VFPv3 floating-point instructions, with 16 double-
13966 precision registers.
13967 +vfpv3
13969 The VFPv3 floating-point instructions, with 32 double-
13970 precision registers.
13971 +vfpv3-d16-fp16
13973 The VFPv3 floating-point instructions, with 16 double-
13974 precision registers and the half-precision floating-point
13975 conversion operations.
13976 +vfpv3-fp16
13978 The VFPv3 floating-point instructions, with 32 double-
13979 precision registers and the half-precision floating-point
13980 conversion operations.
13981 +vfpv4-d16
13983 The VFPv4 floating-point instructions, with 16 double-
13984 precision registers.
13985 +vfpv4
13987 The VFPv4 floating-point instructions, with 32 double-
13988 precision registers.
13989 +neon
13991 The Advanced SIMD (Neon) v1 and the VFPv3 floating-point
13992 instructions. The extension +neon-vfpv3 can be used as an
13993 alias for this extension.
13994 +neon-fp16
13996 The Advanced SIMD (Neon) v1 and the VFPv3 floating-point
13997 instructions, with the half-precision floating-point
13998 conversion operations.
13999 +nosimd
14001 Disable the Advanced SIMD instructions (does not disable
14002 floating point).
14003 +nofp
14005 Disable the floating-point and Advanced SIMD instructions.
14006 armv8-a
14008 +crc
14009 The Cyclic Redundancy Check (CRC) instructions.
14010 +simd
14012 The ARMv8-A Advanced SIMD and floating-point instructions.
14013 +crypto
14015 The cryptographic instructions.
14016 +nocrypto
14018 Disable the cryptographic instructions.
14019 +nofp
14021 Disable the floating-point, Advanced SIMD and cryptographic
14022 instructions.
14023 armv8.1-a
14025 +simd
14026 The ARMv8.1-A Advanced SIMD and floating-point
14027 instructions.
14028 +crypto
14030 The cryptographic instructions. This also enables the
14031 Advanced SIMD and floating-point instructions.
14032 +nocrypto
14034 Disable the cryptographic instructions.
14035 +nofp
14037 Disable the floating-point, Advanced SIMD and cryptographic
14038 instructions.
14039 armv8.2-a
14041 armv8.3-a
14042 +fp16
14043 The half-precision floating-point data processing
14044 instructions. This also enables the Advanced SIMD and
14045 floating-point instructions.
14046 +fp16fml
14048 The half-precision floating-point fmla extension. This
14049 also enables the half-precision floating-point extension
14050 and Advanced SIMD and floating-point instructions.
14051 +simd
14053 The ARMv8.1-A Advanced SIMD and floating-point
14054 instructions.
14055 +crypto
14057 The cryptographic instructions. This also enables the
14058 Advanced SIMD and floating-point instructions.
14059 +dotprod
14061 Enable the Dot Product extension. This also enables
14062 Advanced SIMD instructions.
14063 +nocrypto
14065 Disable the cryptographic extension.
14066 +nofp
14068 Disable the floating-point, Advanced SIMD and cryptographic
14069 instructions.
14070 armv8.4-a
14072 +fp16
14073 The half-precision floating-point data processing
14074 instructions. This also enables the Advanced SIMD and
14075 floating-point instructions as well as the Dot Product
14076 extension and the half-precision floating-point fmla
14077 extension.
14078 +simd
14080 The ARMv8.3-A Advanced SIMD and floating-point instructions
14081 as well as the Dot Product extension.
14082 +crypto
14084 The cryptographic instructions. This also enables the
14085 Advanced SIMD and floating-point instructions as well as
14086 the Dot Product extension.
14087 +nocrypto
14089 Disable the cryptographic extension.
14090 +nofp
14092 Disable the floating-point, Advanced SIMD and cryptographic
14093 instructions.
14094 armv7-r
14096 +fp.sp
14097 The single-precision VFPv3 floating-point instructions.
14098 The extension +vfpv3xd can be used as an alias for this
14099 extension.
14100 +fp The VFPv3 floating-point instructions with 16 double-
14102 precision registers. The extension +vfpv3-d16 can be used
14103 as an alias for this extension.
14104 +nofp
14106 Disable the floating-point extension.
14107 +idiv
14109 The ARM-state integer division instructions.
14110 +noidiv
14112 Disable the ARM-state integer division extension.
14113 armv7e-m
14115 +fp The single-precision VFPv4 floating-point instructions.
14116 +fpv5
14118 The single-precision FPv5 floating-point instructions.
14119 +fp.dp
14121 The single- and double-precision FPv5 floating-point
14122 instructions.
14123 +nofp
14125 Disable the floating-point extensions.
14126 armv8-m.main
14128 +dsp
14129 The DSP instructions.
14130 +nodsp
14132 Disable the DSP extension.
14133 +fp The single-precision floating-point instructions.
14135 +fp.dp
14137 The single- and double-precision floating-point
14138 instructions.
14139 +nofp
14141 Disable the floating-point extension.
14142 armv8-r
14144 +crc
14145 The Cyclic Redundancy Check (CRC) instructions.
14146 +fp.sp
14148 The single-precision FPv5 floating-point instructions.
14149 +simd
14151 The ARMv8-A Advanced SIMD and floating-point instructions.
14152 +crypto
14154 The cryptographic instructions.
14155 +nocrypto
14157 Disable the cryptographic instructions.
14158 +nofp
14160 Disable the floating-point, Advanced SIMD and cryptographic
14161 instructions.
14162 -march=native causes the compiler to auto-detect the architecture
14164 of the build computer. At present, this feature is only supported
14165 on GNU/Linux, and not all architectures are recognized. If the
14166 auto-detect is unsuccessful the option has no effect.
14167 -mtune=name
14169 This option specifies the name of the target ARM processor for
14170 which GCC should tune the performance of the code. For some ARM
14171 implementations better performance can be obtained by using this
14172 option. Permissible names are: arm2, arm250, arm3, arm6, arm60,
14173 arm600, arm610, arm620, arm7, arm7m, arm7d, arm7dm, arm7di,
14174 arm7dmi, arm70, arm700, arm700i, arm710, arm710c, arm7100, arm720,
14175 arm7500, arm7500fe, arm7tdmi, arm7tdmi-s, arm710t, arm720t,
14176 arm740t, strongarm, strongarm110, strongarm1100, strongarm1110,
14177 arm8, arm810, arm9, arm9e, arm920, arm920t, arm922t, arm946e-s,
14178 arm966e-s, arm968e-s, arm926ej-s, arm940t, arm9tdmi, arm10tdmi,
14179 arm1020t, arm1026ej-s, arm10e, arm1020e, arm1022e, arm1136j-s,
14180 arm1136jf-s, mpcore, mpcorenovfp, arm1156t2-s, arm1156t2f-s,
14181 arm1176jz-s, arm1176jzf-s, generic-armv7-a, cortex-a5, cortex-a7,
14182 cortex-a8, cortex-a9, cortex-a12, cortex-a15, cortex-a17,
14183 cortex-a32, cortex-a35, cortex-a53, cortex-a55, cortex-a57,
14184 cortex-a72, cortex-a73, cortex-a75, cortex-r4, cortex-r4f,
14185 cortex-r5, cortex-r7, cortex-r8, cortex-r52, cortex-m33,
14186 cortex-m23, cortex-m7, cortex-m4, cortex-m3, cortex-m1, cortex-m0,
14187 cortex-m0plus, cortex-m1.small-multiply, cortex-m0.small-multiply,
14188 cortex-m0plus.small-multiply, exynos-m1, marvell-pj4, xscale,
14189 iwmmxt, iwmmxt2, ep9312, fa526, fa626, fa606te, fa626te, fmp626,
14190 fa726te, xgene1.
14191 Additionally, this option can specify that GCC should tune the
14193 performance of the code for a big.LITTLE system. Permissible names
14194 are: cortex-a15.cortex-a7, cortex-a17.cortex-a7,
14195 cortex-a57.cortex-a53, cortex-a72.cortex-a53,
14196 cortex-a72.cortex-a35, cortex-a73.cortex-a53,
14197 cortex-a75.cortex-a55.
14198 -mtune=generic-arch specifies that GCC should tune the performance
14200 for a blend of processors within architecture arch. The aim is to
14201 generate code that run well on the current most popular processors,
14202 balancing between optimizations that benefit some CPUs in the
14203 range, and avoiding performance pitfalls of other CPUs. The
14204 effects of this option may change in future GCC versions as CPU
14205 models come and go.
14206 -mtune permits the same extension options as -mcpu, but the
14208 extension options do not affect the tuning of the generated code.
14209 -mtune=native causes the compiler to auto-detect the CPU of the
14211 build computer. At present, this feature is only supported on
14212 GNU/Linux, and not all architectures are recognized. If the auto-
14213 detect is unsuccessful the option has no effect.
14214 -mcpu=name[+extension...]
14216 This specifies the name of the target ARM processor. GCC uses this
14217 name to derive the name of the target ARM architecture (as if
14218 specified by -march) and the ARM processor type for which to tune
14219 for performance (as if specified by -mtune). Where this option is
14220 used in conjunction with -march or -mtune, those options take
14221 precedence over the appropriate part of this option.
14222 Many of the supported CPUs implement optional architectural
14224 extensions. Where this is so the architectural extensions are
14225 normally enabled by default. If implementations that lack the
14226 extension exist, then the extension syntax can be used to disable
14227 those extensions that have been omitted. For floating-point and
14228 Advanced SIMD (Neon) instructions, the settings of the options
14229 -mfloat-abi and -mfpu must also be considered: floating-point and
14230 Advanced SIMD instructions will only be used if -mfloat-abi is not
14231 set to soft; and any setting of -mfpu other than auto will override
14232 the available floating-point and SIMD extension instructions.
14233 For example, cortex-a9 can be found in three major configurations:
14235 integer only, with just a floating-point unit or with floating-
14236 point and Advanced SIMD. The default is to enable all the
14237 instructions, but the extensions +nosimd and +nofp can be used to
14238 disable just the SIMD or both the SIMD and floating-point
14239 instructions respectively.
14240 Permissible names for this option are the same as those for -mtune.
14242 The following extension options are common to the listed CPUs:
14244 +nodsp
14246 Disable the DSP instructions on cortex-m33.
14247 +nofp
14249 Disables the floating-point instructions on arm9e, arm946e-s,
14250 arm966e-s, arm968e-s, arm10e, arm1020e, arm1022e, arm926ej-s,
14251 arm1026ej-s, cortex-r5, cortex-r7, cortex-r8, cortex-m4,
14252 cortex-m7 and cortex-m33. Disables the floating-point and SIMD
14253 instructions on generic-armv7-a, cortex-a5, cortex-a7,
14254 cortex-a8, cortex-a9, cortex-a12, cortex-a15, cortex-a17,
14255 cortex-a15.cortex-a7, cortex-a17.cortex-a7, cortex-a32,
14256 cortex-a35, cortex-a53 and cortex-a55.
14257 +nofp.dp
14259 Disables the double-precision component of the floating-point
14260 instructions on cortex-r5, cortex-r52 and cortex-m7.
14261 +nosimd
14263 Disables the SIMD (but not floating-point) instructions on
14264 generic-armv7-a, cortex-a5, cortex-a7 and cortex-a9.
14265 +crypto
14267 Enables the cryptographic instructions on cortex-a32,
14268 cortex-a35, cortex-a53, cortex-a55, cortex-a57, cortex-a72,
14269 cortex-a73, cortex-a75, exynos-m1, xgene1,
14270 cortex-a57.cortex-a53, cortex-a72.cortex-a53,
14271 cortex-a73.cortex-a35, cortex-a73.cortex-a53 and
14272 cortex-a75.cortex-a55.
14273 Additionally the generic-armv7-a pseudo target defaults to VFPv3
14275 with 16 double-precision registers. It supports the following
14276 extension options: vfpv3-d16, vfpv3, vfpv3-d16-fp16, vfpv3-fp16,
14277 vfpv4-d16, vfpv4, neon, neon-vfpv3, neon-fp16, neon-vfpv4. The
14278 meanings are the same as for the extensions to -march=armv7-a.
14279 -mcpu=generic-arch is also permissible, and is equivalent to
14281 -march=arch -mtune=generic-arch. See -mtune for more information.
14282 -mcpu=native causes the compiler to auto-detect the CPU of the
14284 build computer. At present, this feature is only supported on
14285 GNU/Linux, and not all architectures are recognized. If the auto-
14286 detect is unsuccessful the option has no effect.
14287 -mfpu=name
14289 This specifies what floating-point hardware (or hardware emulation)
14290 is available on the target. Permissible names are: auto, vfpv2,
14291 vfpv3, vfpv3-fp16, vfpv3-d16, vfpv3-d16-fp16, vfpv3xd,
14292 vfpv3xd-fp16, neon-vfpv3, neon-fp16, vfpv4, vfpv4-d16, fpv4-sp-d16,
14293 neon-vfpv4, fpv5-d16, fpv5-sp-d16, fp-armv8, neon-fp-armv8 and
14294 crypto-neon-fp-armv8. Note that neon is an alias for neon-vfpv3
14295 and vfp is an alias for vfpv2.
14296 The setting auto is the default and is special. It causes the
14298 compiler to select the floating-point and Advanced SIMD
14299 instructions based on the settings of -mcpu and -march.
14300 If the selected floating-point hardware includes the NEON extension
14302 (e.g. -mfpu=neon), note that floating-point operations are not
14303 generated by GCC's auto-vectorization pass unless
14304 -funsafe-math-optimizations is also specified. This is because
14305 NEON hardware does not fully implement the IEEE 754 standard for
14306 floating-point arithmetic (in particular denormal values are
14307 treated as zero), so the use of NEON instructions may lead to a
14308 loss of precision.
14309 You can also set the fpu name at function level by using the
14311 "target("fpu=")" function attributes or pragmas.
14312 -mfp16-format=name
14314 Specify the format of the "__fp16" half-precision floating-point
14315 type. Permissible names are none, ieee, and alternative; the
14316 default is none, in which case the "__fp16" type is not defined.
14317 -mstructure-size-boundary=n
14319 The sizes of all structures and unions are rounded up to a multiple
14320 of the number of bits set by this option. Permissible values are
14321 8, 32 and 64. The default value varies for different toolchains.
14322 For the COFF targeted toolchain the default value is 8. A value of
14323 64 is only allowed if the underlying ABI supports it.
14324 Specifying a larger number can produce faster, more efficient code,
14326 but can also increase the size of the program. Different values
14327 are potentially incompatible. Code compiled with one value cannot
14328 necessarily expect to work with code or libraries compiled with
14329 another value, if they exchange information using structures or
14330 unions.
14331 This option is deprecated.
14333 -mabort-on-noreturn
14335 Generate a call to the function "abort" at the end of a "noreturn"
14336 function. It is executed if the function tries to return.
14337 -mlong-calls
14339 -mno-long-calls
14340 Tells the compiler to perform function calls by first loading the
14341 address of the function into a register and then performing a
14342 subroutine call on this register. This switch is needed if the
14343 target function lies outside of the 64-megabyte addressing range of
14344 the offset-based version of subroutine call instruction.
14345 Even if this switch is enabled, not all function calls are turned
14347 into long calls. The heuristic is that static functions, functions
14348 that have the "short_call" attribute, functions that are inside the
14349 scope of a "#pragma no_long_calls" directive, and functions whose
14350 definitions have already been compiled within the current
14351 compilation unit are not turned into long calls. The exceptions to
14352 this rule are that weak function definitions, functions with the
14353 "long_call" attribute or the "section" attribute, and functions
14354 that are within the scope of a "#pragma long_calls" directive are
14355 always turned into long calls.
14356 This feature is not enabled by default. Specifying -mno-long-calls
14358 restores the default behavior, as does placing the function calls
14359 within the scope of a "#pragma long_calls_off" directive. Note
14360 these switches have no effect on how the compiler generates code to
14361 handle function calls via function pointers.
14362 -msingle-pic-base
14364 Treat the register used for PIC addressing as read-only, rather
14365 than loading it in the prologue for each function. The runtime
14366 system is responsible for initializing this register with an
14367 appropriate value before execution begins.
14368 -mpic-register=reg
14370 Specify the register to be used for PIC addressing. For standard
14371 PIC base case, the default is any suitable register determined by
14372 compiler. For single PIC base case, the default is R9 if target is
14373 EABI based or stack-checking is enabled, otherwise the default is
14374 R10.
14375 -mpic-data-is-text-relative
14377 Assume that the displacement between the text and data segments is
14378 fixed at static link time. This permits using PC-relative
14379 addressing operations to access data known to be in the data
14380 segment. For non-VxWorks RTP targets, this option is enabled by
14381 default. When disabled on such targets, it will enable
14382 -msingle-pic-base by default.
14383 -mpoke-function-name
14385 Write the name of each function into the text section, directly
14386 preceding the function prologue. The generated code is similar to
14387 this:
14388 t0
14390 .ascii "arm_poke_function_name", 0
14391 .align
14392 t1
14393 .word 0xff000000 + (t1 - t0)
14394 arm_poke_function_name
14395 mov ip, sp
14396 stmfd sp!, {fp, ip, lr, pc}
14397 sub fp, ip, #4
14398 When performing a stack backtrace, code can inspect the value of
14400 "pc" stored at "fp + 0". If the trace function then looks at
14401 location "pc - 12" and the top 8 bits are set, then we know that
14402 there is a function name embedded immediately preceding this
14403 location and has length "((pc[-3]) & 0xff000000)".
14404 -mthumb
14406 -marm
14407 Select between generating code that executes in ARM and Thumb
14408 states. The default for most configurations is to generate code
14409 that executes in ARM state, but the default can be changed by
14410 configuring GCC with the --with-mode=state configure option.
14411 You can also override the ARM and Thumb mode for each function by
14413 using the "target("thumb")" and "target("arm")" function attributes
14414 or pragmas.
14415 -mflip-thumb
14417 Switch ARM/Thumb modes on alternating functions. This option is
14418 provided for regression testing of mixed Thumb/ARM code generation,
14419 and is not intended for ordinary use in compiling code.
14420 -mtpcs-frame
14422 Generate a stack frame that is compliant with the Thumb Procedure
14423 Call Standard for all non-leaf functions. (A leaf function is one
14424 that does not call any other functions.) The default is
14425 -mno-tpcs-frame.
14426 -mtpcs-leaf-frame
14428 Generate a stack frame that is compliant with the Thumb Procedure
14429 Call Standard for all leaf functions. (A leaf function is one that
14430 does not call any other functions.) The default is
14431 -mno-apcs-leaf-frame.
14432 -mcallee-super-interworking
14434 Gives all externally visible functions in the file being compiled
14435 an ARM instruction set header which switches to Thumb mode before
14436 executing the rest of the function. This allows these functions to
14437 be called from non-interworking code. This option is not valid in
14438 AAPCS configurations because interworking is enabled by default.
14439 -mcaller-super-interworking
14441 Allows calls via function pointers (including virtual functions) to
14442 execute correctly regardless of whether the target code has been
14443 compiled for interworking or not. There is a small overhead in the
14444 cost of executing a function pointer if this option is enabled.
14445 This option is not valid in AAPCS configurations because
14446 interworking is enabled by default.
14447 -mtp=name
14449 Specify the access model for the thread local storage pointer. The
14450 valid models are soft, which generates calls to "__aeabi_read_tp",
14451 cp15, which fetches the thread pointer from "cp15" directly
14452 (supported in the arm6k architecture), and auto, which uses the
14453 best available method for the selected processor. The default
14454 setting is auto.
14455 -mtls-dialect=dialect
14457 Specify the dialect to use for accessing thread local storage. Two
14458 dialects are supported---gnu and gnu2. The gnu dialect selects the
14459 original GNU scheme for supporting local and global dynamic TLS
14460 models. The gnu2 dialect selects the GNU descriptor scheme, which
14461 provides better performance for shared libraries. The GNU
14462 descriptor scheme is compatible with the original scheme, but does
14463 require new assembler, linker and library support. Initial and
14464 local exec TLS models are unaffected by this option and always use
14465 the original scheme.
14466 -mword-relocations
14468 Only generate absolute relocations on word-sized values (i.e.
14469 R_ARM_ABS32). This is enabled by default on targets (uClinux,
14470 SymbianOS) where the runtime loader imposes this restriction, and
14471 when -fpic or -fPIC is specified.
14472 -mfix-cortex-m3-ldrd
14474 Some Cortex-M3 cores can cause data corruption when "ldrd"
14475 instructions with overlapping destination and base registers are
14476 used. This option avoids generating these instructions. This
14477 option is enabled by default when -mcpu=cortex-m3 is specified.
14478 -munaligned-access
14480 -mno-unaligned-access
14481 Enables (or disables) reading and writing of 16- and 32- bit values
14482 from addresses that are not 16- or 32- bit aligned. By default
14483 unaligned access is disabled for all pre-ARMv6, all ARMv6-M and for
14484 ARMv8-M Baseline architectures, and enabled for all other
14485 architectures. If unaligned access is not enabled then words in
14486 packed data structures are accessed a byte at a time.
14487 The ARM attribute "Tag_CPU_unaligned_access" is set in the
14489 generated object file to either true or false, depending upon the
14490 setting of this option. If unaligned access is enabled then the
14491 preprocessor symbol "__ARM_FEATURE_UNALIGNED" is also defined.
14492 -mneon-for-64bits
14494 Enables using Neon to handle scalar 64-bits operations. This is
14495 disabled by default since the cost of moving data from core
14496 registers to Neon is high.
14497 -mslow-flash-data
14499 Assume loading data from flash is slower than fetching instruction.
14500 Therefore literal load is minimized for better performance. This
14501 option is only supported when compiling for ARMv7 M-profile and off
14502 by default.
14503 -masm-syntax-unified
14505 Assume inline assembler is using unified asm syntax. The default
14506 is currently off which implies divided syntax. This option has no
14507 impact on Thumb2. However, this may change in future releases of
14508 GCC. Divided syntax should be considered deprecated.
14509 -mrestrict-it
14511 Restricts generation of IT blocks to conform to the rules of
14512 ARMv8-A. IT blocks can only contain a single 16-bit instruction
14513 from a select set of instructions. This option is on by default for
14514 ARMv8-A Thumb mode.
14515 -mprint-tune-info
14517 Print CPU tuning information as comment in assembler file. This is
14518 an option used only for regression testing of the compiler and not
14519 intended for ordinary use in compiling code. This option is
14520 disabled by default.
14521 -mverbose-cost-dump
14523 Enable verbose cost model dumping in the debug dump files. This
14524 option is provided for use in debugging the compiler.
14525 -mpure-code
14527 Do not allow constant data to be placed in code sections.
14528 Additionally, when compiling for ELF object format give all text
14529 sections the ELF processor-specific section attribute
14530 "SHF_ARM_PURECODE". This option is only available when generating
14531 non-pic code for M-profile targets with the MOVT instruction.
14532 -mcmse
14534 Generate secure code as per the "ARMv8-M Security Extensions:
14535 Requirements on Development Tools Engineering Specification", which
14536 can be found on
14537 <http://infocenter.arm.com/help/topic/com.arm.doc.ecm0359818/ECM0359818_armv8m_security_extensions_reqs_on_dev_tools_1_0.pdf>.
14538 AVR Options
14540 These options are defined for AVR implementations:
14541 -mmcu=mcu
14543 Specify Atmel AVR instruction set architectures (ISA) or MCU type.
14544 The default for this option is@tie{}avr2.
14546 GCC supports the following AVR devices and ISAs:
14548 "avr2"
14550 "Classic" devices with up to 8@tie{}KiB of program memory.
14551 mcu@tie{}= "attiny22", "attiny26", "at90c8534", "at90s2313",
14552 "at90s2323", "at90s2333", "at90s2343", "at90s4414",
14553 "at90s4433", "at90s4434", "at90s8515", "at90s8535".
14554 "avr25"
14556 "Classic" devices with up to 8@tie{}KiB of program memory and
14557 with the "MOVW" instruction. mcu@tie{}= "ata5272", "ata6616c",
14558 "attiny13", "attiny13a", "attiny2313", "attiny2313a",
14559 "attiny24", "attiny24a", "attiny25", "attiny261", "attiny261a",
14560 "attiny43u", "attiny4313", "attiny44", "attiny44a",
14561 "attiny441", "attiny45", "attiny461", "attiny461a", "attiny48",
14562 "attiny828", "attiny84", "attiny84a", "attiny841", "attiny85",
14563 "attiny861", "attiny861a", "attiny87", "attiny88", "at86rf401".
14564 "avr3"
14566 "Classic" devices with 16@tie{}KiB up to 64@tie{}KiB of
14567 program memory. mcu@tie{}= "at43usb355", "at76c711".
14568 "avr31"
14570 "Classic" devices with 128@tie{}KiB of program memory.
14571 mcu@tie{}= "atmega103", "at43usb320".
14572 "avr35"
14574 "Classic" devices with 16@tie{}KiB up to 64@tie{}KiB of program
14575 memory and with the "MOVW" instruction. mcu@tie{}= "ata5505",
14576 "ata6617c", "ata664251", "atmega16u2", "atmega32u2",
14577 "atmega8u2", "attiny1634", "attiny167", "at90usb162",
14578 "at90usb82".
14579 "avr4"
14581 "Enhanced" devices with up to 8@tie{}KiB of program memory.
14582 mcu@tie{}= "ata6285", "ata6286", "ata6289", "ata6612c",
14583 "atmega48", "atmega48a", "atmega48p", "atmega48pa",
14584 "atmega48pb", "atmega8", "atmega8a", "atmega8hva",
14585 "atmega8515", "atmega8535", "atmega88", "atmega88a",
14586 "atmega88p", "atmega88pa", "atmega88pb", "at90pwm1",
14587 "at90pwm2", "at90pwm2b", "at90pwm3", "at90pwm3b", "at90pwm81".
14588 "avr5"
14590 "Enhanced" devices with 16@tie{}KiB up to 64@tie{}KiB of
14591 program memory. mcu@tie{}= "ata5702m322", "ata5782",
14592 "ata5790", "ata5790n", "ata5791", "ata5795", "ata5831",
14593 "ata6613c", "ata6614q", "ata8210", "ata8510", "atmega16",
14594 "atmega16a", "atmega16hva", "atmega16hva2", "atmega16hvb",
14595 "atmega16hvbrevb", "atmega16m1", "atmega16u4", "atmega161",
14596 "atmega162", "atmega163", "atmega164a", "atmega164p",
14597 "atmega164pa", "atmega165", "atmega165a", "atmega165p",
14598 "atmega165pa", "atmega168", "atmega168a", "atmega168p",
14599 "atmega168pa", "atmega168pb", "atmega169", "atmega169a",
14600 "atmega169p", "atmega169pa", "atmega32", "atmega32a",
14601 "atmega32c1", "atmega32hvb", "atmega32hvbrevb", "atmega32m1",
14602 "atmega32u4", "atmega32u6", "atmega323", "atmega324a",
14603 "atmega324p", "atmega324pa", "atmega325", "atmega325a",
14604 "atmega325p", "atmega325pa", "atmega3250", "atmega3250a",
14605 "atmega3250p", "atmega3250pa", "atmega328", "atmega328p",
14606 "atmega328pb", "atmega329", "atmega329a", "atmega329p",
14607 "atmega329pa", "atmega3290", "atmega3290a", "atmega3290p",
14608 "atmega3290pa", "atmega406", "atmega64", "atmega64a",
14609 "atmega64c1", "atmega64hve", "atmega64hve2", "atmega64m1",
14610 "atmega64rfr2", "atmega640", "atmega644", "atmega644a",
14611 "atmega644p", "atmega644pa", "atmega644rfr2", "atmega645",
14612 "atmega645a", "atmega645p", "atmega6450", "atmega6450a",
14613 "atmega6450p", "atmega649", "atmega649a", "atmega649p",
14614 "atmega6490", "atmega6490a", "atmega6490p", "at90can32",
14615 "at90can64", "at90pwm161", "at90pwm216", "at90pwm316",
14616 "at90scr100", "at90usb646", "at90usb647", "at94k", "m3000".
14617 "avr51"
14619 "Enhanced" devices with 128@tie{}KiB of program memory.
14620 mcu@tie{}= "atmega128", "atmega128a", "atmega128rfa1",
14621 "atmega128rfr2", "atmega1280", "atmega1281", "atmega1284",
14622 "atmega1284p", "atmega1284rfr2", "at90can128", "at90usb1286",
14623 "at90usb1287".
14624 "avr6"
14626 "Enhanced" devices with 3-byte PC, i.e. with more than
14627 128@tie{}KiB of program memory. mcu@tie{}= "atmega256rfr2",
14628 "atmega2560", "atmega2561", "atmega2564rfr2".
14629 "avrxmega2"
14631 "XMEGA" devices with more than 8@tie{}KiB and up to 64@tie{}KiB
14632 of program memory. mcu@tie{}= "atxmega16a4", "atxmega16a4u",
14633 "atxmega16c4", "atxmega16d4", "atxmega16e5", "atxmega32a4",
14634 "atxmega32a4u", "atxmega32c3", "atxmega32c4", "atxmega32d3",
14635 "atxmega32d4", "atxmega32e5", "atxmega8e5".
14636 "avrxmega3"
14638 "XMEGA" devices with up to 64@tie{}KiB of combined program
14639 memory and RAM, and with program memory visible in the RAM
14640 address space. mcu@tie{}= "attiny1614", "attiny1616",
14641 "attiny1617", "attiny212", "attiny214", "attiny3214",
14642 "attiny3216", "attiny3217", "attiny412", "attiny414",
14643 "attiny416", "attiny417", "attiny814", "attiny816",
14644 "attiny817".
14645 "avrxmega4"
14647 "XMEGA" devices with more than 64@tie{}KiB and up to
14648 128@tie{}KiB of program memory. mcu@tie{}= "atxmega64a3",
14649 "atxmega64a3u", "atxmega64a4u", "atxmega64b1", "atxmega64b3",
14650 "atxmega64c3", "atxmega64d3", "atxmega64d4".
14651 "avrxmega5"
14653 "XMEGA" devices with more than 64@tie{}KiB and up to
14654 128@tie{}KiB of program memory and more than 64@tie{}KiB of
14655 RAM. mcu@tie{}= "atxmega64a1", "atxmega64a1u".
14656 "avrxmega6"
14658 "XMEGA" devices with more than 128@tie{}KiB of program memory.
14659 mcu@tie{}= "atxmega128a3", "atxmega128a3u", "atxmega128b1",
14660 "atxmega128b3", "atxmega128c3", "atxmega128d3", "atxmega128d4",
14661 "atxmega192a3", "atxmega192a3u", "atxmega192c3",
14662 "atxmega192d3", "atxmega256a3", "atxmega256a3b",
14663 "atxmega256a3bu", "atxmega256a3u", "atxmega256c3",
14664 "atxmega256d3", "atxmega384c3", "atxmega384d3".
14665 "avrxmega7"
14667 "XMEGA" devices with more than 128@tie{}KiB of program memory
14668 and more than 64@tie{}KiB of RAM. mcu@tie{}= "atxmega128a1",
14669 "atxmega128a1u", "atxmega128a4u".
14670 "avrtiny"
14672 "TINY" Tiny core devices with 512@tie{}B up to 4@tie{}KiB of
14673 program memory. mcu@tie{}= "attiny10", "attiny20", "attiny4",
14674 "attiny40", "attiny5", "attiny9".
14675 "avr1"
14677 This ISA is implemented by the minimal AVR core and supported
14678 for assembler only. mcu@tie{}= "attiny11", "attiny12",
14679 "attiny15", "attiny28", "at90s1200".
14680 -mabsdata
14682 Assume that all data in static storage can be accessed by LDS / STS
14683 instructions. This option has only an effect on reduced Tiny
14684 devices like ATtiny40. See also the "absdata" AVR Variable
14685 Attributes,variable attribute.
14686 -maccumulate-args
14688 Accumulate outgoing function arguments and acquire/release the
14689 needed stack space for outgoing function arguments once in function
14690 prologue/epilogue. Without this option, outgoing arguments are
14691 pushed before calling a function and popped afterwards.
14692 Popping the arguments after the function call can be expensive on
14694 AVR so that accumulating the stack space might lead to smaller
14695 executables because arguments need not be removed from the stack
14696 after such a function call.
14697 This option can lead to reduced code size for functions that
14699 perform several calls to functions that get their arguments on the
14700 stack like calls to printf-like functions.
14701 -mbranch-cost=cost
14703 Set the branch costs for conditional branch instructions to cost.
14704 Reasonable values for cost are small, non-negative integers. The
14705 default branch cost is 0.
14706 -mcall-prologues
14708 Functions prologues/epilogues are expanded as calls to appropriate
14709 subroutines. Code size is smaller.
14710 -mgas-isr-prologues
14712 Interrupt service routines (ISRs) may use the "__gcc_isr" pseudo
14713 instruction supported by GNU Binutils. If this option is on, the
14714 feature can still be disabled for individual ISRs by means of the
14715 AVR Function Attributes,,"no_gccisr" function attribute. This
14716 feature is activated per default if optimization is on (but not
14717 with -Og, @pxref{Optimize Options}), and if GNU Binutils support
14718 PR21683 ("https://sourceware.org/PR21683").
14719 -mint8
14721 Assume "int" to be 8-bit integer. This affects the sizes of all
14722 types: a "char" is 1 byte, an "int" is 1 byte, a "long" is 2 bytes,
14723 and "long long" is 4 bytes. Please note that this option does not
14724 conform to the C standards, but it results in smaller code size.
14725 -mmain-is-OS_task
14727 Do not save registers in "main". The effect is the same like
14728 attaching attribute AVR Function Attributes,,"OS_task" to "main".
14729 It is activated per default if optimization is on.
14730 -mn-flash=num
14732 Assume that the flash memory has a size of num times 64@tie{}KiB.
14733 -mno-interrupts
14735 Generated code is not compatible with hardware interrupts. Code
14736 size is smaller.
14737 -mrelax
14739 Try to replace "CALL" resp. "JMP" instruction by the shorter
14740 "RCALL" resp. "RJMP" instruction if applicable. Setting -mrelax
14741 just adds the --mlink-relax option to the assembler's command line
14742 and the --relax option to the linker's command line.
14743 Jump relaxing is performed by the linker because jump offsets are
14745 not known before code is located. Therefore, the assembler code
14746 generated by the compiler is the same, but the instructions in the
14747 executable may differ from instructions in the assembler code.
14748 Relaxing must be turned on if linker stubs are needed, see the
14750 section on "EIND" and linker stubs below.
14751 -mrmw
14753 Assume that the device supports the Read-Modify-Write instructions
14754 "XCH", "LAC", "LAS" and "LAT".
14755 -mshort-calls
14757 Assume that "RJMP" and "RCALL" can target the whole program memory.
14758 This option is used internally for multilib selection. It is not
14760 an optimization option, and you don't need to set it by hand.
14761 -msp8
14763 Treat the stack pointer register as an 8-bit register, i.e. assume
14764 the high byte of the stack pointer is zero. In general, you don't
14765 need to set this option by hand.
14766 This option is used internally by the compiler to select and build
14768 multilibs for architectures "avr2" and "avr25". These
14769 architectures mix devices with and without "SPH". For any setting
14770 other than -mmcu=avr2 or -mmcu=avr25 the compiler driver adds or
14771 removes this option from the compiler proper's command line,
14772 because the compiler then knows if the device or architecture has
14773 an 8-bit stack pointer and thus no "SPH" register or not.
14774 -mstrict-X
14776 Use address register "X" in a way proposed by the hardware. This
14777 means that "X" is only used in indirect, post-increment or pre-
14778 decrement addressing.
14779 Without this option, the "X" register may be used in the same way
14781 as "Y" or "Z" which then is emulated by additional instructions.
14782 For example, loading a value with "X+const" addressing with a small
14783 non-negative "const < 64" to a register Rn is performed as
14784 adiw r26, const ; X += const
14786 ld <Rn>, X ; <Rn> = *X
14787 sbiw r26, const ; X -= const
14788 -mtiny-stack
14790 Only change the lower 8@tie{}bits of the stack pointer.
14791 -mfract-convert-truncate
14793 Allow to use truncation instead of rounding towards zero for
14794 fractional fixed-point types.
14795 -nodevicelib
14797 Don't link against AVR-LibC's device specific library "lib<mcu>.a".
14798 -Waddr-space-convert
14800 Warn about conversions between address spaces in the case where the
14801 resulting address space is not contained in the incoming address
14802 space.
14803 -Wmisspelled-isr
14805 Warn if the ISR is misspelled, i.e. without __vector prefix.
14806 Enabled by default.
14807 "EIND" and Devices with More Than 128 Ki Bytes of Flash
14809 Pointers in the implementation are 16@tie{}bits wide. The address of a
14811 function or label is represented as word address so that indirect jumps
14812 and calls can target any code address in the range of 64@tie{}Ki words.
14813 In order to facilitate indirect jump on devices with more than
14815 128@tie{}Ki bytes of program memory space, there is a special function
14816 register called "EIND" that serves as most significant part of the
14817 target address when "EICALL" or "EIJMP" instructions are used.
14818 Indirect jumps and calls on these devices are handled as follows by the
14820 compiler and are subject to some limitations:
14821 * The compiler never sets "EIND".
14823 * The compiler uses "EIND" implicitly in "EICALL"/"EIJMP"
14825 instructions or might read "EIND" directly in order to emulate an
14826 indirect call/jump by means of a "RET" instruction.
14827 * The compiler assumes that "EIND" never changes during the startup
14829 code or during the application. In particular, "EIND" is not
14830 saved/restored in function or interrupt service routine
14831 prologue/epilogue.
14832 * For indirect calls to functions and computed goto, the linker
14834 generates stubs. Stubs are jump pads sometimes also called
14835 trampolines. Thus, the indirect call/jump jumps to such a stub.
14836 The stub contains a direct jump to the desired address.
14837 * Linker relaxation must be turned on so that the linker generates
14839 the stubs correctly in all situations. See the compiler option
14840 -mrelax and the linker option --relax. There are corner cases
14841 where the linker is supposed to generate stubs but aborts without
14842 relaxation and without a helpful error message.
14843 * The default linker script is arranged for code with "EIND = 0". If
14845 code is supposed to work for a setup with "EIND != 0", a custom
14846 linker script has to be used in order to place the sections whose
14847 name start with ".trampolines" into the segment where "EIND" points
14848 to.
14849 * The startup code from libgcc never sets "EIND". Notice that
14851 startup code is a blend of code from libgcc and AVR-LibC. For the
14852 impact of AVR-LibC on "EIND", see the AVR-LibC user manual
14853 ("http://nongnu.org/avr-libc/user-manual/").
14854 * It is legitimate for user-specific startup code to set up "EIND"
14856 early, for example by means of initialization code located in
14857 section ".init3". Such code runs prior to general startup code that
14858 initializes RAM and calls constructors, but after the bit of
14859 startup code from AVR-LibC that sets "EIND" to the segment where
14860 the vector table is located.
14861 #include <avr/io.h>
14863 static void
14865 __attribute__((section(".init3"),naked,used,no_instrument_function))
14866 init3_set_eind (void)
14867 {
14868 __asm volatile ("ldi r24,pm_hh8(__trampolines_start)\n\t"
14869 "out %i0,r24" :: "n" (&EIND) : "r24","memory");
14870 }
14871 The "__trampolines_start" symbol is defined in the linker script.
14873 * Stubs are generated automatically by the linker if the following
14875 two conditions are met:
14876 -<The address of a label is taken by means of the "gs" modifier>
14878 (short for generate stubs) like so:
14879 LDI r24, lo8(gs(<func>))
14881 LDI r25, hi8(gs(<func>))
14882 -<The final location of that label is in a code segment>
14884 outside the segment where the stubs are located.
14885 * The compiler emits such "gs" modifiers for code labels in the
14887 following situations:
14888 -<Taking address of a function or code label.>
14890 -<Computed goto.>
14891 -<If prologue-save function is used, see -mcall-prologues>
14892 command-line option.
14893 -<Switch/case dispatch tables. If you do not want such dispatch>
14895 tables you can specify the -fno-jump-tables command-line
14896 option.
14897 -<C and C++ constructors/destructors called during
14899 startup/shutdown.>
14900 -<If the tools hit a "gs()" modifier explained above.>
14901 * Jumping to non-symbolic addresses like so is not supported:
14902 int main (void)
14904 {
14905 /* Call function at word address 0x2 */
14906 return ((int(*)(void)) 0x2)();
14907 }
14908 Instead, a stub has to be set up, i.e. the function has to be
14910 called through a symbol ("func_4" in the example):
14911 int main (void)
14913 {
14914 extern int func_4 (void);
14915 /* Call function at byte address 0x4 */
14917 return func_4();
14918 }
14919 and the application be linked with -Wl,--defsym,func_4=0x4.
14921 Alternatively, "func_4" can be defined in the linker script.
14922 Handling of the "RAMPD", "RAMPX", "RAMPY" and "RAMPZ" Special Function
14924 Registers
14925 Some AVR devices support memories larger than the 64@tie{}KiB range
14927 that can be accessed with 16-bit pointers. To access memory locations
14928 outside this 64@tie{}KiB range, the content of a "RAMP" register is
14929 used as high part of the address: The "X", "Y", "Z" address register is
14930 concatenated with the "RAMPX", "RAMPY", "RAMPZ" special function
14931 register, respectively, to get a wide address. Similarly, "RAMPD" is
14932 used together with direct addressing.
14933 * The startup code initializes the "RAMP" special function registers
14935 with zero.
14936 * If a AVR Named Address Spaces,named address space other than
14938 generic or "__flash" is used, then "RAMPZ" is set as needed before
14939 the operation.
14940 * If the device supports RAM larger than 64@tie{}KiB and the compiler
14942 needs to change "RAMPZ" to accomplish an operation, "RAMPZ" is
14943 reset to zero after the operation.
14944 * If the device comes with a specific "RAMP" register, the ISR
14946 prologue/epilogue saves/restores that SFR and initializes it with
14947 zero in case the ISR code might (implicitly) use it.
14948 * RAM larger than 64@tie{}KiB is not supported by GCC for AVR
14950 targets. If you use inline assembler to read from locations
14951 outside the 16-bit address range and change one of the "RAMP"
14952 registers, you must reset it to zero after the access.
14953 AVR Built-in Macros
14955 GCC defines several built-in macros so that the user code can test for
14957 the presence or absence of features. Almost any of the following
14958 built-in macros are deduced from device capabilities and thus triggered
14959 by the -mmcu= command-line option.
14960 For even more AVR-specific built-in macros see AVR Named Address Spaces
14962 and AVR Built-in Functions.
14963 "__AVR_ARCH__"
14965 Build-in macro that resolves to a decimal number that identifies
14966 the architecture and depends on the -mmcu=mcu option. Possible
14967 values are:
14968 2, 25, 3, 31, 35, 4, 5, 51, 6
14970 for mcu="avr2", "avr25", "avr3", "avr31", "avr35", "avr4", "avr5",
14972 "avr51", "avr6",
14973 respectively and
14975 100, 102, 103, 104, 105, 106, 107
14977 for mcu="avrtiny", "avrxmega2", "avrxmega3", "avrxmega4",
14979 "avrxmega5", "avrxmega6", "avrxmega7", respectively. If mcu
14980 specifies a device, this built-in macro is set accordingly. For
14981 example, with -mmcu=atmega8 the macro is defined to 4.
14982 "__AVR_Device__"
14984 Setting -mmcu=device defines this built-in macro which reflects the
14985 device's name. For example, -mmcu=atmega8 defines the built-in
14986 macro "__AVR_ATmega8__", -mmcu=attiny261a defines
14987 "__AVR_ATtiny261A__", etc.
14988 The built-in macros' names follow the scheme "__AVR_Device__" where
14990 Device is the device name as from the AVR user manual. The
14991 difference between Device in the built-in macro and device in
14992 -mmcu=device is that the latter is always lowercase.
14993 If device is not a device but only a core architecture like avr51,
14995 this macro is not defined.
14996 "__AVR_DEVICE_NAME__"
14998 Setting -mmcu=device defines this built-in macro to the device's
14999 name. For example, with -mmcu=atmega8 the macro is defined to
15000 "atmega8".
15001 If device is not a device but only a core architecture like avr51,
15003 this macro is not defined.
15004 "__AVR_XMEGA__"
15006 The device / architecture belongs to the XMEGA family of devices.
15007 "__AVR_HAVE_ELPM__"
15009 The device has the "ELPM" instruction.
15010 "__AVR_HAVE_ELPMX__"
15012 The device has the "ELPM Rn,Z" and "ELPM Rn,Z+" instructions.
15013 "__AVR_HAVE_MOVW__"
15015 The device has the "MOVW" instruction to perform 16-bit register-
15016 register moves.
15017 "__AVR_HAVE_LPMX__"
15019 The device has the "LPM Rn,Z" and "LPM Rn,Z+" instructions.
15020 "__AVR_HAVE_MUL__"
15022 The device has a hardware multiplier.
15023 "__AVR_HAVE_JMP_CALL__"
15025 The device has the "JMP" and "CALL" instructions. This is the case
15026 for devices with more than 8@tie{}KiB of program memory.
15027 "__AVR_HAVE_EIJMP_EICALL__"
15029 "__AVR_3_BYTE_PC__"
15030 The device has the "EIJMP" and "EICALL" instructions. This is the
15031 case for devices with more than 128@tie{}KiB of program memory.
15032 This also means that the program counter (PC) is 3@tie{}bytes wide.
15033 "__AVR_2_BYTE_PC__"
15035 The program counter (PC) is 2@tie{}bytes wide. This is the case for
15036 devices with up to 128@tie{}KiB of program memory.
15037 "__AVR_HAVE_8BIT_SP__"
15039 "__AVR_HAVE_16BIT_SP__"
15040 The stack pointer (SP) register is treated as 8-bit respectively
15041 16-bit register by the compiler. The definition of these macros is
15042 affected by -mtiny-stack.
15043 "__AVR_HAVE_SPH__"
15045 "__AVR_SP8__"
15046 The device has the SPH (high part of stack pointer) special
15047 function register or has an 8-bit stack pointer, respectively. The
15048 definition of these macros is affected by -mmcu= and in the cases
15049 of -mmcu=avr2 and -mmcu=avr25 also by -msp8.
15050 "__AVR_HAVE_RAMPD__"
15052 "__AVR_HAVE_RAMPX__"
15053 "__AVR_HAVE_RAMPY__"
15054 "__AVR_HAVE_RAMPZ__"
15055 The device has the "RAMPD", "RAMPX", "RAMPY", "RAMPZ" special
15056 function register, respectively.
15057 "__NO_INTERRUPTS__"
15059 This macro reflects the -mno-interrupts command-line option.
15060 "__AVR_ERRATA_SKIP__"
15062 "__AVR_ERRATA_SKIP_JMP_CALL__"
15063 Some AVR devices (AT90S8515, ATmega103) must not skip 32-bit
15064 instructions because of a hardware erratum. Skip instructions are
15065 "SBRS", "SBRC", "SBIS", "SBIC" and "CPSE". The second macro is
15066 only defined if "__AVR_HAVE_JMP_CALL__" is also set.
15067 "__AVR_ISA_RMW__"
15069 The device has Read-Modify-Write instructions (XCH, LAC, LAS and
15070 LAT).
15071 "__AVR_SFR_OFFSET__=offset"
15073 Instructions that can address I/O special function registers
15074 directly like "IN", "OUT", "SBI", etc. may use a different address
15075 as if addressed by an instruction to access RAM like "LD" or "STS".
15076 This offset depends on the device architecture and has to be
15077 subtracted from the RAM address in order to get the respective
15078 I/O@tie{}address.
15079 "__AVR_SHORT_CALLS__"
15081 The -mshort-calls command line option is set.
15082 "__AVR_PM_BASE_ADDRESS__=addr"
15084 Some devices support reading from flash memory by means of "LD*"
15085 instructions. The flash memory is seen in the data address space
15086 at an offset of "__AVR_PM_BASE_ADDRESS__". If this macro is not
15087 defined, this feature is not available. If defined, the address
15088 space is linear and there is no need to put ".rodata" into RAM.
15089 This is handled by the default linker description file, and is
15090 currently available for "avrtiny" and "avrxmega3". Even more
15091 convenient, there is no need to use address spaces like "__flash"
15092 or features like attribute "progmem" and "pgm_read_*".
15093 "__WITH_AVRLIBC__"
15095 The compiler is configured to be used together with AVR-Libc. See
15096 the --with-avrlibc configure option.
15097 Blackfin Options
15099 -mcpu=cpu[-sirevision]
15100 Specifies the name of the target Blackfin processor. Currently,
15101 cpu can be one of bf512, bf514, bf516, bf518, bf522, bf523, bf524,
15102 bf525, bf526, bf527, bf531, bf532, bf533, bf534, bf536, bf537,
15103 bf538, bf539, bf542, bf544, bf547, bf548, bf549, bf542m, bf544m,
15104 bf547m, bf548m, bf549m, bf561, bf592.
15105 The optional sirevision specifies the silicon revision of the
15107 target Blackfin processor. Any workarounds available for the
15108 targeted silicon revision are enabled. If sirevision is none, no
15109 workarounds are enabled. If sirevision is any, all workarounds for
15110 the targeted processor are enabled. The "__SILICON_REVISION__"
15111 macro is defined to two hexadecimal digits representing the major
15112 and minor numbers in the silicon revision. If sirevision is none,
15113 the "__SILICON_REVISION__" is not defined. If sirevision is any,
15114 the "__SILICON_REVISION__" is defined to be 0xffff. If this
15115 optional sirevision is not used, GCC assumes the latest known
15116 silicon revision of the targeted Blackfin processor.
15117 GCC defines a preprocessor macro for the specified cpu. For the
15119 bfin-elf toolchain, this option causes the hardware BSP provided by
15120 libgloss to be linked in if -msim is not given.
15121 Without this option, bf532 is used as the processor by default.
15123 Note that support for bf561 is incomplete. For bf561, only the
15125 preprocessor macro is defined.
15126 -msim
15128 Specifies that the program will be run on the simulator. This
15129 causes the simulator BSP provided by libgloss to be linked in.
15130 This option has effect only for bfin-elf toolchain. Certain other
15131 options, such as -mid-shared-library and -mfdpic, imply -msim.
15132 -momit-leaf-frame-pointer
15134 Don't keep the frame pointer in a register for leaf functions.
15135 This avoids the instructions to save, set up and restore frame
15136 pointers and makes an extra register available in leaf functions.
15137 -mspecld-anomaly
15139 When enabled, the compiler ensures that the generated code does not
15140 contain speculative loads after jump instructions. If this option
15141 is used, "__WORKAROUND_SPECULATIVE_LOADS" is defined.
15142 -mno-specld-anomaly
15144 Don't generate extra code to prevent speculative loads from
15145 occurring.
15146 -mcsync-anomaly
15148 When enabled, the compiler ensures that the generated code does not
15149 contain CSYNC or SSYNC instructions too soon after conditional
15150 branches. If this option is used, "__WORKAROUND_SPECULATIVE_SYNCS"
15151 is defined.
15152 -mno-csync-anomaly
15154 Don't generate extra code to prevent CSYNC or SSYNC instructions
15155 from occurring too soon after a conditional branch.
15156 -mlow-64k
15158 When enabled, the compiler is free to take advantage of the
15159 knowledge that the entire program fits into the low 64k of memory.
15160 -mno-low-64k
15162 Assume that the program is arbitrarily large. This is the default.
15163 -mstack-check-l1
15165 Do stack checking using information placed into L1 scratchpad
15166 memory by the uClinux kernel.
15167 -mid-shared-library
15169 Generate code that supports shared libraries via the library ID
15170 method. This allows for execute in place and shared libraries in
15171 an environment without virtual memory management. This option
15172 implies -fPIC. With a bfin-elf target, this option implies -msim.
15173 -mno-id-shared-library
15175 Generate code that doesn't assume ID-based shared libraries are
15176 being used. This is the default.
15177 -mleaf-id-shared-library
15179 Generate code that supports shared libraries via the library ID
15180 method, but assumes that this library or executable won't link
15181 against any other ID shared libraries. That allows the compiler to
15182 use faster code for jumps and calls.
15183 -mno-leaf-id-shared-library
15185 Do not assume that the code being compiled won't link against any
15186 ID shared libraries. Slower code is generated for jump and call
15187 insns.
15188 -mshared-library-id=n
15190 Specifies the identification number of the ID-based shared library
15191 being compiled. Specifying a value of 0 generates more compact
15192 code; specifying other values forces the allocation of that number
15193 to the current library but is no more space- or time-efficient than
15194 omitting this option.
15195 -msep-data
15197 Generate code that allows the data segment to be located in a
15198 different area of memory from the text segment. This allows for
15199 execute in place in an environment without virtual memory
15200 management by eliminating relocations against the text section.
15201 -mno-sep-data
15203 Generate code that assumes that the data segment follows the text
15204 segment. This is the default.
15205 -mlong-calls
15207 -mno-long-calls
15208 Tells the compiler to perform function calls by first loading the
15209 address of the function into a register and then performing a
15210 subroutine call on this register. This switch is needed if the
15211 target function lies outside of the 24-bit addressing range of the
15212 offset-based version of subroutine call instruction.
15213 This feature is not enabled by default. Specifying -mno-long-calls
15215 restores the default behavior. Note these switches have no effect
15216 on how the compiler generates code to handle function calls via
15217 function pointers.
15218 -mfast-fp
15220 Link with the fast floating-point library. This library relaxes
15221 some of the IEEE floating-point standard's rules for checking
15222 inputs against Not-a-Number (NAN), in the interest of performance.
15223 -minline-plt
15225 Enable inlining of PLT entries in function calls to functions that
15226 are not known to bind locally. It has no effect without -mfdpic.
15227 -mmulticore
15229 Build a standalone application for multicore Blackfin processors.
15230 This option causes proper start files and link scripts supporting
15231 multicore to be used, and defines the macro "__BFIN_MULTICORE". It
15232 can only be used with -mcpu=bf561[-sirevision].
15233 This option can be used with -mcorea or -mcoreb, which selects the
15235 one-application-per-core programming model. Without -mcorea or
15236 -mcoreb, the single-application/dual-core programming model is
15237 used. In this model, the main function of Core B should be named as
15238 "coreb_main".
15239 If this option is not used, the single-core application programming
15241 model is used.
15242 -mcorea
15244 Build a standalone application for Core A of BF561 when using the
15245 one-application-per-core programming model. Proper start files and
15246 link scripts are used to support Core A, and the macro
15247 "__BFIN_COREA" is defined. This option can only be used in
15248 conjunction with -mmulticore.
15249 -mcoreb
15251 Build a standalone application for Core B of BF561 when using the
15252 one-application-per-core programming model. Proper start files and
15253 link scripts are used to support Core B, and the macro
15254 "__BFIN_COREB" is defined. When this option is used, "coreb_main"
15255 should be used instead of "main". This option can only be used in
15256 conjunction with -mmulticore.
15257 -msdram
15259 Build a standalone application for SDRAM. Proper start files and
15260 link scripts are used to put the application into SDRAM, and the
15261 macro "__BFIN_SDRAM" is defined. The loader should initialize
15262 SDRAM before loading the application.
15263 -micplb
15265 Assume that ICPLBs are enabled at run time. This has an effect on
15266 certain anomaly workarounds. For Linux targets, the default is to
15267 assume ICPLBs are enabled; for standalone applications the default
15268 is off.
15269 C6X Options
15271 -march=name
15272 This specifies the name of the target architecture. GCC uses this
15273 name to determine what kind of instructions it can emit when
15274 generating assembly code. Permissible names are: c62x, c64x,
15275 c64x+, c67x, c67x+, c674x.
15276 -mbig-endian
15278 Generate code for a big-endian target.
15279 -mlittle-endian
15281 Generate code for a little-endian target. This is the default.
15282 -msim
15284 Choose startup files and linker script suitable for the simulator.
15285 -msdata=default
15287 Put small global and static data in the ".neardata" section, which
15288 is pointed to by register "B14". Put small uninitialized global
15289 and static data in the ".bss" section, which is adjacent to the
15290 ".neardata" section. Put small read-only data into the ".rodata"
15291 section. The corresponding sections used for large pieces of data
15292 are ".fardata", ".far" and ".const".
15293 -msdata=all
15295 Put all data, not just small objects, into the sections reserved
15296 for small data, and use addressing relative to the "B14" register
15297 to access them.
15298 -msdata=none
15300 Make no use of the sections reserved for small data, and use
15301 absolute addresses to access all data. Put all initialized global
15302 and static data in the ".fardata" section, and all uninitialized
15303 data in the ".far" section. Put all constant data into the
15304 ".const" section.
15305 CRIS Options
15307 These options are defined specifically for the CRIS ports.
15308 -march=architecture-type
15310 -mcpu=architecture-type
15311 Generate code for the specified architecture. The choices for
15312 architecture-type are v3, v8 and v10 for respectively ETRAX 4,
15313 ETRAX 100, and ETRAX 100 LX. Default is v0 except for cris-axis-
15314 linux-gnu, where the default is v10.
15315 -mtune=architecture-type
15317 Tune to architecture-type everything applicable about the generated
15318 code, except for the ABI and the set of available instructions.
15319 The choices for architecture-type are the same as for
15320 -march=architecture-type.
15321 -mmax-stack-frame=n
15323 Warn when the stack frame of a function exceeds n bytes.
15324 -metrax4
15326 -metrax100
15327 The options -metrax4 and -metrax100 are synonyms for -march=v3 and
15328 -march=v8 respectively.
15329 -mmul-bug-workaround
15331 -mno-mul-bug-workaround
15332 Work around a bug in the "muls" and "mulu" instructions for CPU
15333 models where it applies. This option is active by default.
15334 -mpdebug
15336 Enable CRIS-specific verbose debug-related information in the
15337 assembly code. This option also has the effect of turning off the
15338 #NO_APP formatted-code indicator to the assembler at the beginning
15339 of the assembly file.
15340 -mcc-init
15342 Do not use condition-code results from previous instruction; always
15343 emit compare and test instructions before use of condition codes.
15344 -mno-side-effects
15346 Do not emit instructions with side effects in addressing modes
15347 other than post-increment.
15348 -mstack-align
15350 -mno-stack-align
15351 -mdata-align
15352 -mno-data-align
15353 -mconst-align
15354 -mno-const-align
15355 These options (no- options) arrange (eliminate arrangements) for
15356 the stack frame, individual data and constants to be aligned for
15357 the maximum single data access size for the chosen CPU model. The
15358 default is to arrange for 32-bit alignment. ABI details such as
15359 structure layout are not affected by these options.
15360 -m32-bit
15362 -m16-bit
15363 -m8-bit
15364 Similar to the stack- data- and const-align options above, these
15365 options arrange for stack frame, writable data and constants to all
15366 be 32-bit, 16-bit or 8-bit aligned. The default is 32-bit
15367 alignment.
15368 -mno-prologue-epilogue
15370 -mprologue-epilogue
15371 With -mno-prologue-epilogue, the normal function prologue and
15372 epilogue which set up the stack frame are omitted and no return
15373 instructions or return sequences are generated in the code. Use
15374 this option only together with visual inspection of the compiled
15375 code: no warnings or errors are generated when call-saved registers
15376 must be saved, or storage for local variables needs to be
15377 allocated.
15378 -mno-gotplt
15380 -mgotplt
15381 With -fpic and -fPIC, don't generate (do generate) instruction
15382 sequences that load addresses for functions from the PLT part of
15383 the GOT rather than (traditional on other architectures) calls to
15384 the PLT. The default is -mgotplt.
15385 -melf
15387 Legacy no-op option only recognized with the cris-axis-elf and
15388 cris-axis-linux-gnu targets.
15389 -mlinux
15391 Legacy no-op option only recognized with the cris-axis-linux-gnu
15392 target.
15393 -sim
15395 This option, recognized for the cris-axis-elf, arranges to link
15396 with input-output functions from a simulator library. Code,
15397 initialized data and zero-initialized data are allocated
15398 consecutively.
15399 -sim2
15401 Like -sim, but pass linker options to locate initialized data at
15402 0x40000000 and zero-initialized data at 0x80000000.
15403 CR16 Options
15405 These options are defined specifically for the CR16 ports.
15406 -mmac
15408 Enable the use of multiply-accumulate instructions. Disabled by
15409 default.
15410 -mcr16cplus
15412 -mcr16c
15413 Generate code for CR16C or CR16C+ architecture. CR16C+ architecture
15414 is default.
15415 -msim
15417 Links the library libsim.a which is in compatible with simulator.
15418 Applicable to ELF compiler only.
15419 -mint32
15421 Choose integer type as 32-bit wide.
15422 -mbit-ops
15424 Generates "sbit"/"cbit" instructions for bit manipulations.
15425 -mdata-model=model
15427 Choose a data model. The choices for model are near, far or medium.
15428 medium is default. However, far is not valid with -mcr16c, as the
15429 CR16C architecture does not support the far data model.
15430 Darwin Options
15432 These options are defined for all architectures running the Darwin
15433 operating system.
15434 FSF GCC on Darwin does not create "fat" object files; it creates an
15436 object file for the single architecture that GCC was built to target.
15437 Apple's GCC on Darwin does create "fat" files if multiple -arch options
15438 are used; it does so by running the compiler or linker multiple times
15439 and joining the results together with lipo.
15440 The subtype of the file created (like ppc7400 or ppc970 or i686) is
15442 determined by the flags that specify the ISA that GCC is targeting,
15443 like -mcpu or -march. The -force_cpusubtype_ALL option can be used to
15444 override this.
15445 The Darwin tools vary in their behavior when presented with an ISA
15447 mismatch. The assembler, as, only permits instructions to be used that
15448 are valid for the subtype of the file it is generating, so you cannot
15449 put 64-bit instructions in a ppc750 object file. The linker for shared
15450 libraries, /usr/bin/libtool, fails and prints an error if asked to
15451 create a shared library with a less restrictive subtype than its input
15452 files (for instance, trying to put a ppc970 object file in a ppc7400
15453 library). The linker for executables, ld, quietly gives the executable
15454 the most restrictive subtype of any of its input files.
15455 -Fdir
15457 Add the framework directory dir to the head of the list of
15458 directories to be searched for header files. These directories are
15459 interleaved with those specified by -I options and are scanned in a
15460 left-to-right order.
15461 A framework directory is a directory with frameworks in it. A
15463 framework is a directory with a Headers and/or PrivateHeaders
15464 directory contained directly in it that ends in .framework. The
15465 name of a framework is the name of this directory excluding the
15466 .framework. Headers associated with the framework are found in one
15467 of those two directories, with Headers being searched first. A
15468 subframework is a framework directory that is in a framework's
15469 Frameworks directory. Includes of subframework headers can only
15470 appear in a header of a framework that contains the subframework,
15471 or in a sibling subframework header. Two subframeworks are
15472 siblings if they occur in the same framework. A subframework
15473 should not have the same name as a framework; a warning is issued
15474 if this is violated. Currently a subframework cannot have
15475 subframeworks; in the future, the mechanism may be extended to
15476 support this. The standard frameworks can be found in
15477 /System/Library/Frameworks and /Library/Frameworks. An example
15478 include looks like "#include <Framework/header.h>", where Framework
15479 denotes the name of the framework and header.h is found in the
15480 PrivateHeaders or Headers directory.
15481 -iframeworkdir
15483 Like -F except the directory is a treated as a system directory.
15484 The main difference between this -iframework and -F is that with
15485 -iframework the compiler does not warn about constructs contained
15486 within header files found via dir. This option is valid only for
15487 the C family of languages.
15488 -gused
15490 Emit debugging information for symbols that are used. For stabs
15491 debugging format, this enables -feliminate-unused-debug-symbols.
15492 This is by default ON.
15493 -gfull
15495 Emit debugging information for all symbols and types.
15496 -mmacosx-version-min=version
15498 The earliest version of MacOS X that this executable will run on is
15499 version. Typical values of version include 10.1, 10.2, and 10.3.9.
15500 If the compiler was built to use the system's headers by default,
15502 then the default for this option is the system version on which the
15503 compiler is running, otherwise the default is to make choices that
15504 are compatible with as many systems and code bases as possible.
15505 -mkernel
15507 Enable kernel development mode. The -mkernel option sets -static,
15508 -fno-common, -fno-use-cxa-atexit, -fno-exceptions,
15509 -fno-non-call-exceptions, -fapple-kext, -fno-weak and -fno-rtti
15510 where applicable. This mode also sets -mno-altivec, -msoft-float,
15511 -fno-builtin and -mlong-branch for PowerPC targets.
15512 -mone-byte-bool
15514 Override the defaults for "bool" so that "sizeof(bool)==1". By
15515 default "sizeof(bool)" is 4 when compiling for Darwin/PowerPC and 1
15516 when compiling for Darwin/x86, so this option has no effect on x86.
15517 Warning: The -mone-byte-bool switch causes GCC to generate code
15519 that is not binary compatible with code generated without that
15520 switch. Using this switch may require recompiling all other
15521 modules in a program, including system libraries. Use this switch
15522 to conform to a non-default data model.
15523 -mfix-and-continue
15525 -ffix-and-continue
15526 -findirect-data
15527 Generate code suitable for fast turnaround development, such as to
15528 allow GDB to dynamically load .o files into already-running
15529 programs. -findirect-data and -ffix-and-continue are provided for
15530 backwards compatibility.
15531 -all_load
15533 Loads all members of static archive libraries. See man ld(1) for
15534 more information.
15535 -arch_errors_fatal
15537 Cause the errors having to do with files that have the wrong
15538 architecture to be fatal.
15539 -bind_at_load
15541 Causes the output file to be marked such that the dynamic linker
15542 will bind all undefined references when the file is loaded or
15543 launched.
15544 -bundle
15546 Produce a Mach-o bundle format file. See man ld(1) for more
15547 information.
15548 -bundle_loader executable
15550 This option specifies the executable that will load the build
15551 output file being linked. See man ld(1) for more information.
15552 -dynamiclib
15554 When passed this option, GCC produces a dynamic library instead of
15555 an executable when linking, using the Darwin libtool command.
15556 -force_cpusubtype_ALL
15558 This causes GCC's output file to have the ALL subtype, instead of
15559 one controlled by the -mcpu or -march option.
15560 -allowable_client client_name
15562 -client_name
15563 -compatibility_version
15564 -current_version
15565 -dead_strip
15566 -dependency-file
15567 -dylib_file
15568 -dylinker_install_name
15569 -dynamic
15570 -exported_symbols_list
15571 -filelist
15572 -flat_namespace
15573 -force_flat_namespace
15574 -headerpad_max_install_names
15575 -image_base
15576 -init
15577 -install_name
15578 -keep_private_externs
15579 -multi_module
15580 -multiply_defined
15581 -multiply_defined_unused
15582 -noall_load
15583 -no_dead_strip_inits_and_terms
15584 -nofixprebinding
15585 -nomultidefs
15586 -noprebind
15587 -noseglinkedit
15588 -pagezero_size
15589 -prebind
15590 -prebind_all_twolevel_modules
15591 -private_bundle
15592 -read_only_relocs
15593 -sectalign
15594 -sectobjectsymbols
15595 -whyload
15596 -seg1addr
15597 -sectcreate
15598 -sectobjectsymbols
15599 -sectorder
15600 -segaddr
15601 -segs_read_only_addr
15602 -segs_read_write_addr
15603 -seg_addr_table
15604 -seg_addr_table_filename
15605 -seglinkedit
15606 -segprot
15607 -segs_read_only_addr
15608 -segs_read_write_addr
15609 -single_module
15610 -static
15611 -sub_library
15612 -sub_umbrella
15613 -twolevel_namespace
15614 -umbrella
15615 -undefined
15616 -unexported_symbols_list
15617 -weak_reference_mismatches
15618 -whatsloaded
15619 These options are passed to the Darwin linker. The Darwin linker
15620 man page describes them in detail.
15621 DEC Alpha Options
15623 These -m options are defined for the DEC Alpha implementations:
15624 -mno-soft-float
15626 -msoft-float
15627 Use (do not use) the hardware floating-point instructions for
15628 floating-point operations. When -msoft-float is specified,
15629 functions in libgcc.a are used to perform floating-point
15630 operations. Unless they are replaced by routines that emulate the
15631 floating-point operations, or compiled in such a way as to call
15632 such emulations routines, these routines issue floating-point
15633 operations. If you are compiling for an Alpha without floating-
15634 point operations, you must ensure that the library is built so as
15635 not to call them.
15636 Note that Alpha implementations without floating-point operations
15638 are required to have floating-point registers.
15639 -mfp-reg
15641 -mno-fp-regs
15642 Generate code that uses (does not use) the floating-point register
15643 set. -mno-fp-regs implies -msoft-float. If the floating-point
15644 register set is not used, floating-point operands are passed in
15645 integer registers as if they were integers and floating-point
15646 results are passed in $0 instead of $f0. This is a non-standard
15647 calling sequence, so any function with a floating-point argument or
15648 return value called by code compiled with -mno-fp-regs must also be
15649 compiled with that option.
15650 A typical use of this option is building a kernel that does not
15652 use, and hence need not save and restore, any floating-point
15653 registers.
15654 -mieee
15656 The Alpha architecture implements floating-point hardware optimized
15657 for maximum performance. It is mostly compliant with the IEEE
15658 floating-point standard. However, for full compliance, software
15659 assistance is required. This option generates code fully IEEE-
15660 compliant code except that the inexact-flag is not maintained (see
15661 below). If this option is turned on, the preprocessor macro
15662 "_IEEE_FP" is defined during compilation. The resulting code is
15663 less efficient but is able to correctly support denormalized
15664 numbers and exceptional IEEE values such as not-a-number and
15665 plus/minus infinity. Other Alpha compilers call this option
15666 -ieee_with_no_inexact.
15667 -mieee-with-inexact
15669 This is like -mieee except the generated code also maintains the
15670 IEEE inexact-flag. Turning on this option causes the generated
15671 code to implement fully-compliant IEEE math. In addition to
15672 "_IEEE_FP", "_IEEE_FP_EXACT" is defined as a preprocessor macro.
15673 On some Alpha implementations the resulting code may execute
15674 significantly slower than the code generated by default. Since
15675 there is very little code that depends on the inexact-flag, you
15676 should normally not specify this option. Other Alpha compilers
15677 call this option -ieee_with_inexact.
15678 -mfp-trap-mode=trap-mode
15680 This option controls what floating-point related traps are enabled.
15681 Other Alpha compilers call this option -fptm trap-mode. The trap
15682 mode can be set to one of four values:
15683 n This is the default (normal) setting. The only traps that are
15685 enabled are the ones that cannot be disabled in software (e.g.,
15686 division by zero trap).
15687 u In addition to the traps enabled by n, underflow traps are
15689 enabled as well.
15690 su Like u, but the instructions are marked to be safe for software
15692 completion (see Alpha architecture manual for details).
15693 sui Like su, but inexact traps are enabled as well.
15695 -mfp-rounding-mode=rounding-mode
15697 Selects the IEEE rounding mode. Other Alpha compilers call this
15698 option -fprm rounding-mode. The rounding-mode can be one of:
15699 n Normal IEEE rounding mode. Floating-point numbers are rounded
15701 towards the nearest machine number or towards the even machine
15702 number in case of a tie.
15703 m Round towards minus infinity.
15705 c Chopped rounding mode. Floating-point numbers are rounded
15707 towards zero.
15708 d Dynamic rounding mode. A field in the floating-point control
15710 register (fpcr, see Alpha architecture reference manual)
15711 controls the rounding mode in effect. The C library
15712 initializes this register for rounding towards plus infinity.
15713 Thus, unless your program modifies the fpcr, d corresponds to
15714 round towards plus infinity.
15715 -mtrap-precision=trap-precision
15717 In the Alpha architecture, floating-point traps are imprecise.
15718 This means without software assistance it is impossible to recover
15719 from a floating trap and program execution normally needs to be
15720 terminated. GCC can generate code that can assist operating system
15721 trap handlers in determining the exact location that caused a
15722 floating-point trap. Depending on the requirements of an
15723 application, different levels of precisions can be selected:
15724 p Program precision. This option is the default and means a trap
15726 handler can only identify which program caused a floating-point
15727 exception.
15728 f Function precision. The trap handler can determine the
15730 function that caused a floating-point exception.
15731 i Instruction precision. The trap handler can determine the
15733 exact instruction that caused a floating-point exception.
15734 Other Alpha compilers provide the equivalent options called
15736 -scope_safe and -resumption_safe.
15737 -mieee-conformant
15739 This option marks the generated code as IEEE conformant. You must
15740 not use this option unless you also specify -mtrap-precision=i and
15741 either -mfp-trap-mode=su or -mfp-trap-mode=sui. Its only effect is
15742 to emit the line .eflag 48 in the function prologue of the
15743 generated assembly file.
15744 -mbuild-constants
15746 Normally GCC examines a 32- or 64-bit integer constant to see if it
15747 can construct it from smaller constants in two or three
15748 instructions. If it cannot, it outputs the constant as a literal
15749 and generates code to load it from the data segment at run time.
15750 Use this option to require GCC to construct all integer constants
15752 using code, even if it takes more instructions (the maximum is
15753 six).
15754 You typically use this option to build a shared library dynamic
15756 loader. Itself a shared library, it must relocate itself in memory
15757 before it can find the variables and constants in its own data
15758 segment.
15759 -mbwx
15761 -mno-bwx
15762 -mcix
15763 -mno-cix
15764 -mfix
15765 -mno-fix
15766 -mmax
15767 -mno-max
15768 Indicate whether GCC should generate code to use the optional BWX,
15769 CIX, FIX and MAX instruction sets. The default is to use the
15770 instruction sets supported by the CPU type specified via -mcpu=
15771 option or that of the CPU on which GCC was built if none is
15772 specified.
15773 -mfloat-vax
15775 -mfloat-ieee
15776 Generate code that uses (does not use) VAX F and G floating-point
15777 arithmetic instead of IEEE single and double precision.
15778 -mexplicit-relocs
15780 -mno-explicit-relocs
15781 Older Alpha assemblers provided no way to generate symbol
15782 relocations except via assembler macros. Use of these macros does
15783 not allow optimal instruction scheduling. GNU binutils as of
15784 version 2.12 supports a new syntax that allows the compiler to
15785 explicitly mark which relocations should apply to which
15786 instructions. This option is mostly useful for debugging, as GCC
15787 detects the capabilities of the assembler when it is built and sets
15788 the default accordingly.
15789 -msmall-data
15791 -mlarge-data
15792 When -mexplicit-relocs is in effect, static data is accessed via
15793 gp-relative relocations. When -msmall-data is used, objects 8
15794 bytes long or smaller are placed in a small data area (the ".sdata"
15795 and ".sbss" sections) and are accessed via 16-bit relocations off
15796 of the $gp register. This limits the size of the small data area
15797 to 64KB, but allows the variables to be directly accessed via a
15798 single instruction.
15799 The default is -mlarge-data. With this option the data area is
15801 limited to just below 2GB. Programs that require more than 2GB of
15802 data must use "malloc" or "mmap" to allocate the data in the heap
15803 instead of in the program's data segment.
15804 When generating code for shared libraries, -fpic implies
15806 -msmall-data and -fPIC implies -mlarge-data.
15807 -msmall-text
15809 -mlarge-text
15810 When -msmall-text is used, the compiler assumes that the code of
15811 the entire program (or shared library) fits in 4MB, and is thus
15812 reachable with a branch instruction. When -msmall-data is used,
15813 the compiler can assume that all local symbols share the same $gp
15814 value, and thus reduce the number of instructions required for a
15815 function call from 4 to 1.
15816 The default is -mlarge-text.
15818 -mcpu=cpu_type
15820 Set the instruction set and instruction scheduling parameters for
15821 machine type cpu_type. You can specify either the EV style name or
15822 the corresponding chip number. GCC supports scheduling parameters
15823 for the EV4, EV5 and EV6 family of processors and chooses the
15824 default values for the instruction set from the processor you
15825 specify. If you do not specify a processor type, GCC defaults to
15826 the processor on which the compiler was built.
15827 Supported values for cpu_type are
15829 ev4
15831 ev45
15832 21064
15833 Schedules as an EV4 and has no instruction set extensions.
15834 ev5
15836 21164
15837 Schedules as an EV5 and has no instruction set extensions.
15838 ev56
15840 21164a
15841 Schedules as an EV5 and supports the BWX extension.
15842 pca56
15844 21164pc
15845 21164PC
15846 Schedules as an EV5 and supports the BWX and MAX extensions.
15847 ev6
15849 21264
15850 Schedules as an EV6 and supports the BWX, FIX, and MAX
15851 extensions.
15852 ev67
15854 21264a
15855 Schedules as an EV6 and supports the BWX, CIX, FIX, and MAX
15856 extensions.
15857 Native toolchains also support the value native, which selects the
15859 best architecture option for the host processor. -mcpu=native has
15860 no effect if GCC does not recognize the processor.
15861 -mtune=cpu_type
15863 Set only the instruction scheduling parameters for machine type
15864 cpu_type. The instruction set is not changed.
15865 Native toolchains also support the value native, which selects the
15867 best architecture option for the host processor. -mtune=native has
15868 no effect if GCC does not recognize the processor.
15869 -mmemory-latency=time
15871 Sets the latency the scheduler should assume for typical memory
15872 references as seen by the application. This number is highly
15873 dependent on the memory access patterns used by the application and
15874 the size of the external cache on the machine.
15875 Valid options for time are
15877 number
15879 A decimal number representing clock cycles.
15880 L1
15882 L2
15883 L3
15884 main
15885 The compiler contains estimates of the number of clock cycles
15886 for "typical" EV4 & EV5 hardware for the Level 1, 2 & 3 caches
15887 (also called Dcache, Scache, and Bcache), as well as to main
15888 memory. Note that L3 is only valid for EV5.
15889 FR30 Options
15891 These options are defined specifically for the FR30 port.
15892 -msmall-model
15894 Use the small address space model. This can produce smaller code,
15895 but it does assume that all symbolic values and addresses fit into
15896 a 20-bit range.
15897 -mno-lsim
15899 Assume that runtime support has been provided and so there is no
15900 need to include the simulator library (libsim.a) on the linker
15901 command line.
15902 FT32 Options
15904 These options are defined specifically for the FT32 port.
15905 -msim
15907 Specifies that the program will be run on the simulator. This
15908 causes an alternate runtime startup and library to be linked. You
15909 must not use this option when generating programs that will run on
15910 real hardware; you must provide your own runtime library for
15911 whatever I/O functions are needed.
15912 -mlra
15914 Enable Local Register Allocation. This is still experimental for
15915 FT32, so by default the compiler uses standard reload.
15916 -mnodiv
15918 Do not use div and mod instructions.
15919 -mft32b
15921 Enable use of the extended instructions of the FT32B processor.
15922 -mcompress
15924 Compress all code using the Ft32B code compression scheme.
15925 -mnopm
15927 Do not generate code that reads program memory.
15928 FRV Options
15930 -mgpr-32
15931 Only use the first 32 general-purpose registers.
15932 -mgpr-64
15934 Use all 64 general-purpose registers.
15935 -mfpr-32
15937 Use only the first 32 floating-point registers.
15938 -mfpr-64
15940 Use all 64 floating-point registers.
15941 -mhard-float
15943 Use hardware instructions for floating-point operations.
15944 -msoft-float
15946 Use library routines for floating-point operations.
15947 -malloc-cc
15949 Dynamically allocate condition code registers.
15950 -mfixed-cc
15952 Do not try to dynamically allocate condition code registers, only
15953 use "icc0" and "fcc0".
15954 -mdword
15956 Change ABI to use double word insns.
15957 -mno-dword
15959 Do not use double word instructions.
15960 -mdouble
15962 Use floating-point double instructions.
15963 -mno-double
15965 Do not use floating-point double instructions.
15966 -mmedia
15968 Use media instructions.
15969 -mno-media
15971 Do not use media instructions.
15972 -mmuladd
15974 Use multiply and add/subtract instructions.
15975 -mno-muladd
15977 Do not use multiply and add/subtract instructions.
15978 -mfdpic
15980 Select the FDPIC ABI, which uses function descriptors to represent
15981 pointers to functions. Without any PIC/PIE-related options, it
15982 implies -fPIE. With -fpic or -fpie, it assumes GOT entries and
15983 small data are within a 12-bit range from the GOT base address;
15984 with -fPIC or -fPIE, GOT offsets are computed with 32 bits. With a
15985 bfin-elf target, this option implies -msim.
15986 -minline-plt
15988 Enable inlining of PLT entries in function calls to functions that
15989 are not known to bind locally. It has no effect without -mfdpic.
15990 It's enabled by default if optimizing for speed and compiling for
15991 shared libraries (i.e., -fPIC or -fpic), or when an optimization
15992 option such as -O3 or above is present in the command line.
15993 -mTLS
15995 Assume a large TLS segment when generating thread-local code.
15996 -mtls
15998 Do not assume a large TLS segment when generating thread-local
15999 code.
16000 -mgprel-ro
16002 Enable the use of "GPREL" relocations in the FDPIC ABI for data
16003 that is known to be in read-only sections. It's enabled by
16004 default, except for -fpic or -fpie: even though it may help make
16005 the global offset table smaller, it trades 1 instruction for 4.
16006 With -fPIC or -fPIE, it trades 3 instructions for 4, one of which
16007 may be shared by multiple symbols, and it avoids the need for a GOT
16008 entry for the referenced symbol, so it's more likely to be a win.
16009 If it is not, -mno-gprel-ro can be used to disable it.
16010 -multilib-library-pic
16012 Link with the (library, not FD) pic libraries. It's implied by
16013 -mlibrary-pic, as well as by -fPIC and -fpic without -mfdpic. You
16014 should never have to use it explicitly.
16015 -mlinked-fp
16017 Follow the EABI requirement of always creating a frame pointer
16018 whenever a stack frame is allocated. This option is enabled by
16019 default and can be disabled with -mno-linked-fp.
16020 -mlong-calls
16022 Use indirect addressing to call functions outside the current
16023 compilation unit. This allows the functions to be placed anywhere
16024 within the 32-bit address space.
16025 -malign-labels
16027 Try to align labels to an 8-byte boundary by inserting NOPs into
16028 the previous packet. This option only has an effect when VLIW
16029 packing is enabled. It doesn't create new packets; it merely adds
16030 NOPs to existing ones.
16031 -mlibrary-pic
16033 Generate position-independent EABI code.
16034 -macc-4
16036 Use only the first four media accumulator registers.
16037 -macc-8
16039 Use all eight media accumulator registers.
16040 -mpack
16042 Pack VLIW instructions.
16043 -mno-pack
16045 Do not pack VLIW instructions.
16046 -mno-eflags
16048 Do not mark ABI switches in e_flags.
16049 -mcond-move
16051 Enable the use of conditional-move instructions (default).
16052 This switch is mainly for debugging the compiler and will likely be
16054 removed in a future version.
16055 -mno-cond-move
16057 Disable the use of conditional-move instructions.
16058 This switch is mainly for debugging the compiler and will likely be
16060 removed in a future version.
16061 -mscc
16063 Enable the use of conditional set instructions (default).
16064 This switch is mainly for debugging the compiler and will likely be
16066 removed in a future version.
16067 -mno-scc
16069 Disable the use of conditional set instructions.
16070 This switch is mainly for debugging the compiler and will likely be
16072 removed in a future version.
16073 -mcond-exec
16075 Enable the use of conditional execution (default).
16076 This switch is mainly for debugging the compiler and will likely be
16078 removed in a future version.
16079 -mno-cond-exec
16081 Disable the use of conditional execution.
16082 This switch is mainly for debugging the compiler and will likely be
16084 removed in a future version.
16085 -mvliw-branch
16087 Run a pass to pack branches into VLIW instructions (default).
16088 This switch is mainly for debugging the compiler and will likely be
16090 removed in a future version.
16091 -mno-vliw-branch
16093 Do not run a pass to pack branches into VLIW instructions.
16094 This switch is mainly for debugging the compiler and will likely be
16096 removed in a future version.
16097 -mmulti-cond-exec
16099 Enable optimization of "&&" and "||" in conditional execution
16100 (default).
16101 This switch is mainly for debugging the compiler and will likely be
16103 removed in a future version.
16104 -mno-multi-cond-exec
16106 Disable optimization of "&&" and "||" in conditional execution.
16107 This switch is mainly for debugging the compiler and will likely be
16109 removed in a future version.
16110 -mnested-cond-exec
16112 Enable nested conditional execution optimizations (default).
16113 This switch is mainly for debugging the compiler and will likely be
16115 removed in a future version.
16116 -mno-nested-cond-exec
16118 Disable nested conditional execution optimizations.
16119 This switch is mainly for debugging the compiler and will likely be
16121 removed in a future version.
16122 -moptimize-membar
16124 This switch removes redundant "membar" instructions from the
16125 compiler-generated code. It is enabled by default.
16126 -mno-optimize-membar
16128 This switch disables the automatic removal of redundant "membar"
16129 instructions from the generated code.
16130 -mtomcat-stats
16132 Cause gas to print out tomcat statistics.
16133 -mcpu=cpu
16135 Select the processor type for which to generate code. Possible
16136 values are frv, fr550, tomcat, fr500, fr450, fr405, fr400, fr300
16137 and simple.
16138 GNU/Linux Options
16140 These -m options are defined for GNU/Linux targets:
16141 -mglibc
16143 Use the GNU C library. This is the default except on
16144 *-*-linux-*uclibc*, *-*-linux-*musl* and *-*-linux-*android*
16145 targets.
16146 -muclibc
16148 Use uClibc C library. This is the default on *-*-linux-*uclibc*
16149 targets.
16150 -mmusl
16152 Use the musl C library. This is the default on *-*-linux-*musl*
16153 targets.
16154 -mbionic
16156 Use Bionic C library. This is the default on *-*-linux-*android*
16157 targets.
16158 -mandroid
16160 Compile code compatible with Android platform. This is the default
16161 on *-*-linux-*android* targets.
16162 When compiling, this option enables -mbionic, -fPIC,
16164 -fno-exceptions and -fno-rtti by default. When linking, this
16165 option makes the GCC driver pass Android-specific options to the
16166 linker. Finally, this option causes the preprocessor macro
16167 "__ANDROID__" to be defined.
16168 -tno-android-cc
16170 Disable compilation effects of -mandroid, i.e., do not enable
16171 -mbionic, -fPIC, -fno-exceptions and -fno-rtti by default.
16172 -tno-android-ld
16174 Disable linking effects of -mandroid, i.e., pass standard Linux
16175 linking options to the linker.
16176 H8/300 Options
16178 These -m options are defined for the H8/300 implementations:
16179 -mrelax
16181 Shorten some address references at link time, when possible; uses
16182 the linker option -relax.
16183 -mh Generate code for the H8/300H.
16185 -ms Generate code for the H8S.
16187 -mn Generate code for the H8S and H8/300H in the normal mode. This
16189 switch must be used either with -mh or -ms.
16190 -ms2600
16192 Generate code for the H8S/2600. This switch must be used with -ms.
16193 -mexr
16195 Extended registers are stored on stack before execution of function
16196 with monitor attribute. Default option is -mexr. This option is
16197 valid only for H8S targets.
16198 -mno-exr
16200 Extended registers are not stored on stack before execution of
16201 function with monitor attribute. Default option is -mno-exr. This
16202 option is valid only for H8S targets.
16203 -mint32
16205 Make "int" data 32 bits by default.
16206 -malign-300
16208 On the H8/300H and H8S, use the same alignment rules as for the
16209 H8/300. The default for the H8/300H and H8S is to align longs and
16210 floats on 4-byte boundaries. -malign-300 causes them to be aligned
16211 on 2-byte boundaries. This option has no effect on the H8/300.
16212 HPPA Options
16214 These -m options are defined for the HPPA family of computers:
16215 -march=architecture-type
16217 Generate code for the specified architecture. The choices for
16218 architecture-type are 1.0 for PA 1.0, 1.1 for PA 1.1, and 2.0 for
16219 PA 2.0 processors. Refer to /usr/lib/sched.models on an HP-UX
16220 system to determine the proper architecture option for your
16221 machine. Code compiled for lower numbered architectures runs on
16222 higher numbered architectures, but not the other way around.
16223 -mpa-risc-1-0
16225 -mpa-risc-1-1
16226 -mpa-risc-2-0
16227 Synonyms for -march=1.0, -march=1.1, and -march=2.0 respectively.
16228 -mcaller-copies
16230 The caller copies function arguments passed by hidden reference.
16231 This option should be used with care as it is not compatible with
16232 the default 32-bit runtime. However, only aggregates larger than
16233 eight bytes are passed by hidden reference and the option provides
16234 better compatibility with OpenMP.
16235 -mjump-in-delay
16237 This option is ignored and provided for compatibility purposes
16238 only.
16239 -mdisable-fpregs
16241 Prevent floating-point registers from being used in any manner.
16242 This is necessary for compiling kernels that perform lazy context
16243 switching of floating-point registers. If you use this option and
16244 attempt to perform floating-point operations, the compiler aborts.
16245 -mdisable-indexing
16247 Prevent the compiler from using indexing address modes. This
16248 avoids some rather obscure problems when compiling MIG generated
16249 code under MACH.
16250 -mno-space-regs
16252 Generate code that assumes the target has no space registers. This
16253 allows GCC to generate faster indirect calls and use unscaled index
16254 address modes.
16255 Such code is suitable for level 0 PA systems and kernels.
16257 -mfast-indirect-calls
16259 Generate code that assumes calls never cross space boundaries.
16260 This allows GCC to emit code that performs faster indirect calls.
16261 This option does not work in the presence of shared libraries or
16263 nested functions.
16264 -mfixed-range=register-range
16266 Generate code treating the given register range as fixed registers.
16267 A fixed register is one that the register allocator cannot use.
16268 This is useful when compiling kernel code. A register range is
16269 specified as two registers separated by a dash. Multiple register
16270 ranges can be specified separated by a comma.
16271 -mlong-load-store
16273 Generate 3-instruction load and store sequences as sometimes
16274 required by the HP-UX 10 linker. This is equivalent to the +k
16275 option to the HP compilers.
16276 -mportable-runtime
16278 Use the portable calling conventions proposed by HP for ELF
16279 systems.
16280 -mgas
16282 Enable the use of assembler directives only GAS understands.
16283 -mschedule=cpu-type
16285 Schedule code according to the constraints for the machine type
16286 cpu-type. The choices for cpu-type are 700 7100, 7100LC, 7200,
16287 7300 and 8000. Refer to /usr/lib/sched.models on an HP-UX system
16288 to determine the proper scheduling option for your machine. The
16289 default scheduling is 8000.
16290 -mlinker-opt
16292 Enable the optimization pass in the HP-UX linker. Note this makes
16293 symbolic debugging impossible. It also triggers a bug in the HP-UX
16294 8 and HP-UX 9 linkers in which they give bogus error messages when
16295 linking some programs.
16296 -msoft-float
16298 Generate output containing library calls for floating point.
16299 Warning: the requisite libraries are not available for all HPPA
16300 targets. Normally the facilities of the machine's usual C compiler
16301 are used, but this cannot be done directly in cross-compilation.
16302 You must make your own arrangements to provide suitable library
16303 functions for cross-compilation.
16304 -msoft-float changes the calling convention in the output file;
16306 therefore, it is only useful if you compile all of a program with
16307 this option. In particular, you need to compile libgcc.a, the
16308 library that comes with GCC, with -msoft-float in order for this to
16309 work.
16310 -msio
16312 Generate the predefine, "_SIO", for server IO. The default is
16313 -mwsio. This generates the predefines, "__hp9000s700",
16314 "__hp9000s700__" and "_WSIO", for workstation IO. These options
16315 are available under HP-UX and HI-UX.
16316 -mgnu-ld
16318 Use options specific to GNU ld. This passes -shared to ld when
16319 building a shared library. It is the default when GCC is
16320 configured, explicitly or implicitly, with the GNU linker. This
16321 option does not affect which ld is called; it only changes what
16322 parameters are passed to that ld. The ld that is called is
16323 determined by the --with-ld configure option, GCC's program search
16324 path, and finally by the user's PATH. The linker used by GCC can
16325 be printed using which `gcc -print-prog-name=ld`. This option is
16326 only available on the 64-bit HP-UX GCC, i.e. configured with
16327 hppa*64*-*-hpux*.
16328 -mhp-ld
16330 Use options specific to HP ld. This passes -b to ld when building
16331 a shared library and passes +Accept TypeMismatch to ld on all
16332 links. It is the default when GCC is configured, explicitly or
16333 implicitly, with the HP linker. This option does not affect which
16334 ld is called; it only changes what parameters are passed to that
16335 ld. The ld that is called is determined by the --with-ld configure
16336 option, GCC's program search path, and finally by the user's PATH.
16337 The linker used by GCC can be printed using which `gcc
16338 -print-prog-name=ld`. This option is only available on the 64-bit
16339 HP-UX GCC, i.e. configured with hppa*64*-*-hpux*.
16340 -mlong-calls
16342 Generate code that uses long call sequences. This ensures that a
16343 call is always able to reach linker generated stubs. The default
16344 is to generate long calls only when the distance from the call site
16345 to the beginning of the function or translation unit, as the case
16346 may be, exceeds a predefined limit set by the branch type being
16347 used. The limits for normal calls are 7,600,000 and 240,000 bytes,
16348 respectively for the PA 2.0 and PA 1.X architectures. Sibcalls are
16349 always limited at 240,000 bytes.
16350 Distances are measured from the beginning of functions when using
16352 the -ffunction-sections option, or when using the -mgas and
16353 -mno-portable-runtime options together under HP-UX with the SOM
16354 linker.
16355 It is normally not desirable to use this option as it degrades
16357 performance. However, it may be useful in large applications,
16358 particularly when partial linking is used to build the application.
16359 The types of long calls used depends on the capabilities of the
16361 assembler and linker, and the type of code being generated. The
16362 impact on systems that support long absolute calls, and long pic
16363 symbol-difference or pc-relative calls should be relatively small.
16364 However, an indirect call is used on 32-bit ELF systems in pic code
16365 and it is quite long.
16366 -munix=unix-std
16368 Generate compiler predefines and select a startfile for the
16369 specified UNIX standard. The choices for unix-std are 93, 95 and
16370 98. 93 is supported on all HP-UX versions. 95 is available on HP-
16371 UX 10.10 and later. 98 is available on HP-UX 11.11 and later. The
16372 default values are 93 for HP-UX 10.00, 95 for HP-UX 10.10 though to
16373 11.00, and 98 for HP-UX 11.11 and later.
16374 -munix=93 provides the same predefines as GCC 3.3 and 3.4.
16376 -munix=95 provides additional predefines for "XOPEN_UNIX" and
16377 "_XOPEN_SOURCE_EXTENDED", and the startfile unix95.o. -munix=98
16378 provides additional predefines for "_XOPEN_UNIX",
16379 "_XOPEN_SOURCE_EXTENDED", "_INCLUDE__STDC_A1_SOURCE" and
16380 "_INCLUDE_XOPEN_SOURCE_500", and the startfile unix98.o.
16381 It is important to note that this option changes the interfaces for
16383 various library routines. It also affects the operational behavior
16384 of the C library. Thus, extreme care is needed in using this
16385 option.
16386 Library code that is intended to operate with more than one UNIX
16388 standard must test, set and restore the variable
16389 "__xpg4_extended_mask" as appropriate. Most GNU software doesn't
16390 provide this capability.
16391 -nolibdld
16393 Suppress the generation of link options to search libdld.sl when
16394 the -static option is specified on HP-UX 10 and later.
16395 -static
16397 The HP-UX implementation of setlocale in libc has a dependency on
16398 libdld.sl. There isn't an archive version of libdld.sl. Thus,
16399 when the -static option is specified, special link options are
16400 needed to resolve this dependency.
16401 On HP-UX 10 and later, the GCC driver adds the necessary options to
16403 link with libdld.sl when the -static option is specified. This
16404 causes the resulting binary to be dynamic. On the 64-bit port, the
16405 linkers generate dynamic binaries by default in any case. The
16406 -nolibdld option can be used to prevent the GCC driver from adding
16407 these link options.
16408 -threads
16410 Add support for multithreading with the dce thread library under
16411 HP-UX. This option sets flags for both the preprocessor and
16412 linker.
16413 IA-64 Options
16415 These are the -m options defined for the Intel IA-64 architecture.
16416 -mbig-endian
16418 Generate code for a big-endian target. This is the default for HP-
16419 UX.
16420 -mlittle-endian
16422 Generate code for a little-endian target. This is the default for
16423 AIX5 and GNU/Linux.
16424 -mgnu-as
16426 -mno-gnu-as
16427 Generate (or don't) code for the GNU assembler. This is the
16428 default.
16429 -mgnu-ld
16431 -mno-gnu-ld
16432 Generate (or don't) code for the GNU linker. This is the default.
16433 -mno-pic
16435 Generate code that does not use a global pointer register. The
16436 result is not position independent code, and violates the IA-64
16437 ABI.
16438 -mvolatile-asm-stop
16440 -mno-volatile-asm-stop
16441 Generate (or don't) a stop bit immediately before and after
16442 volatile asm statements.
16443 -mregister-names
16445 -mno-register-names
16446 Generate (or don't) in, loc, and out register names for the stacked
16447 registers. This may make assembler output more readable.
16448 -mno-sdata
16450 -msdata
16451 Disable (or enable) optimizations that use the small data section.
16452 This may be useful for working around optimizer bugs.
16453 -mconstant-gp
16455 Generate code that uses a single constant global pointer value.
16456 This is useful when compiling kernel code.
16457 -mauto-pic
16459 Generate code that is self-relocatable. This implies
16460 -mconstant-gp. This is useful when compiling firmware code.
16461 -minline-float-divide-min-latency
16463 Generate code for inline divides of floating-point values using the
16464 minimum latency algorithm.
16465 -minline-float-divide-max-throughput
16467 Generate code for inline divides of floating-point values using the
16468 maximum throughput algorithm.
16469 -mno-inline-float-divide
16471 Do not generate inline code for divides of floating-point values.
16472 -minline-int-divide-min-latency
16474 Generate code for inline divides of integer values using the
16475 minimum latency algorithm.
16476 -minline-int-divide-max-throughput
16478 Generate code for inline divides of integer values using the
16479 maximum throughput algorithm.
16480 -mno-inline-int-divide
16482 Do not generate inline code for divides of integer values.
16483 -minline-sqrt-min-latency
16485 Generate code for inline square roots using the minimum latency
16486 algorithm.
16487 -minline-sqrt-max-throughput
16489 Generate code for inline square roots using the maximum throughput
16490 algorithm.
16491 -mno-inline-sqrt
16493 Do not generate inline code for "sqrt".
16494 -mfused-madd
16496 -mno-fused-madd
16497 Do (don't) generate code that uses the fused multiply/add or
16498 multiply/subtract instructions. The default is to use these
16499 instructions.
16500 -mno-dwarf2-asm
16502 -mdwarf2-asm
16503 Don't (or do) generate assembler code for the DWARF line number
16504 debugging info. This may be useful when not using the GNU
16505 assembler.
16506 -mearly-stop-bits
16508 -mno-early-stop-bits
16509 Allow stop bits to be placed earlier than immediately preceding the
16510 instruction that triggered the stop bit. This can improve
16511 instruction scheduling, but does not always do so.
16512 -mfixed-range=register-range
16514 Generate code treating the given register range as fixed registers.
16515 A fixed register is one that the register allocator cannot use.
16516 This is useful when compiling kernel code. A register range is
16517 specified as two registers separated by a dash. Multiple register
16518 ranges can be specified separated by a comma.
16519 -mtls-size=tls-size
16521 Specify bit size of immediate TLS offsets. Valid values are 14,
16522 22, and 64.
16523 -mtune=cpu-type
16525 Tune the instruction scheduling for a particular CPU, Valid values
16526 are itanium, itanium1, merced, itanium2, and mckinley.
16527 -milp32
16529 -mlp64
16530 Generate code for a 32-bit or 64-bit environment. The 32-bit
16531 environment sets int, long and pointer to 32 bits. The 64-bit
16532 environment sets int to 32 bits and long and pointer to 64 bits.
16533 These are HP-UX specific flags.
16534 -mno-sched-br-data-spec
16536 -msched-br-data-spec
16537 (Dis/En)able data speculative scheduling before reload. This
16538 results in generation of "ld.a" instructions and the corresponding
16539 check instructions ("ld.c" / "chk.a"). The default setting is
16540 disabled.
16541 -msched-ar-data-spec
16543 -mno-sched-ar-data-spec
16544 (En/Dis)able data speculative scheduling after reload. This
16545 results in generation of "ld.a" instructions and the corresponding
16546 check instructions ("ld.c" / "chk.a"). The default setting is
16547 enabled.
16548 -mno-sched-control-spec
16550 -msched-control-spec
16551 (Dis/En)able control speculative scheduling. This feature is
16552 available only during region scheduling (i.e. before reload). This
16553 results in generation of the "ld.s" instructions and the
16554 corresponding check instructions "chk.s". The default setting is
16555 disabled.
16556 -msched-br-in-data-spec
16558 -mno-sched-br-in-data-spec
16559 (En/Dis)able speculative scheduling of the instructions that are
16560 dependent on the data speculative loads before reload. This is
16561 effective only with -msched-br-data-spec enabled. The default
16562 setting is enabled.
16563 -msched-ar-in-data-spec
16565 -mno-sched-ar-in-data-spec
16566 (En/Dis)able speculative scheduling of the instructions that are
16567 dependent on the data speculative loads after reload. This is
16568 effective only with -msched-ar-data-spec enabled. The default
16569 setting is enabled.
16570 -msched-in-control-spec
16572 -mno-sched-in-control-spec
16573 (En/Dis)able speculative scheduling of the instructions that are
16574 dependent on the control speculative loads. This is effective only
16575 with -msched-control-spec enabled. The default setting is enabled.
16576 -mno-sched-prefer-non-data-spec-insns
16578 -msched-prefer-non-data-spec-insns
16579 If enabled, data-speculative instructions are chosen for schedule
16580 only if there are no other choices at the moment. This makes the
16581 use of the data speculation much more conservative. The default
16582 setting is disabled.
16583 -mno-sched-prefer-non-control-spec-insns
16585 -msched-prefer-non-control-spec-insns
16586 If enabled, control-speculative instructions are chosen for
16587 schedule only if there are no other choices at the moment. This
16588 makes the use of the control speculation much more conservative.
16589 The default setting is disabled.
16590 -mno-sched-count-spec-in-critical-path
16592 -msched-count-spec-in-critical-path
16593 If enabled, speculative dependencies are considered during
16594 computation of the instructions priorities. This makes the use of
16595 the speculation a bit more conservative. The default setting is
16596 disabled.
16597 -msched-spec-ldc
16599 Use a simple data speculation check. This option is on by default.
16600 -msched-control-spec-ldc
16602 Use a simple check for control speculation. This option is on by
16603 default.
16604 -msched-stop-bits-after-every-cycle
16606 Place a stop bit after every cycle when scheduling. This option is
16607 on by default.
16608 -msched-fp-mem-deps-zero-cost
16610 Assume that floating-point stores and loads are not likely to cause
16611 a conflict when placed into the same instruction group. This
16612 option is disabled by default.
16613 -msel-sched-dont-check-control-spec
16615 Generate checks for control speculation in selective scheduling.
16616 This flag is disabled by default.
16617 -msched-max-memory-insns=max-insns
16619 Limit on the number of memory insns per instruction group, giving
16620 lower priority to subsequent memory insns attempting to schedule in
16621 the same instruction group. Frequently useful to prevent cache bank
16622 conflicts. The default value is 1.
16623 -msched-max-memory-insns-hard-limit
16625 Makes the limit specified by msched-max-memory-insns a hard limit,
16626 disallowing more than that number in an instruction group.
16627 Otherwise, the limit is "soft", meaning that non-memory operations
16628 are preferred when the limit is reached, but memory operations may
16629 still be scheduled.
16630 LM32 Options
16632 These -m options are defined for the LatticeMico32 architecture:
16633 -mbarrel-shift-enabled
16635 Enable barrel-shift instructions.
16636 -mdivide-enabled
16638 Enable divide and modulus instructions.
16639 -mmultiply-enabled
16641 Enable multiply instructions.
16642 -msign-extend-enabled
16644 Enable sign extend instructions.
16645 -muser-enabled
16647 Enable user-defined instructions.
16648 M32C Options
16650 -mcpu=name
16651 Select the CPU for which code is generated. name may be one of r8c
16652 for the R8C/Tiny series, m16c for the M16C (up to /60) series,
16653 m32cm for the M16C/80 series, or m32c for the M32C/80 series.
16654 -msim
16656 Specifies that the program will be run on the simulator. This
16657 causes an alternate runtime library to be linked in which supports,
16658 for example, file I/O. You must not use this option when
16659 generating programs that will run on real hardware; you must
16660 provide your own runtime library for whatever I/O functions are
16661 needed.
16662 -memregs=number
16664 Specifies the number of memory-based pseudo-registers GCC uses
16665 during code generation. These pseudo-registers are used like real
16666 registers, so there is a tradeoff between GCC's ability to fit the
16667 code into available registers, and the performance penalty of using
16668 memory instead of registers. Note that all modules in a program
16669 must be compiled with the same value for this option. Because of
16670 that, you must not use this option with GCC's default runtime
16671 libraries.
16672 M32R/D Options
16674 These -m options are defined for Renesas M32R/D architectures:
16675 -m32r2
16677 Generate code for the M32R/2.
16678 -m32rx
16680 Generate code for the M32R/X.
16681 -m32r
16683 Generate code for the M32R. This is the default.
16684 -mmodel=small
16686 Assume all objects live in the lower 16MB of memory (so that their
16687 addresses can be loaded with the "ld24" instruction), and assume
16688 all subroutines are reachable with the "bl" instruction. This is
16689 the default.
16690 The addressability of a particular object can be set with the
16692 "model" attribute.
16693 -mmodel=medium
16695 Assume objects may be anywhere in the 32-bit address space (the
16696 compiler generates "seth/add3" instructions to load their
16697 addresses), and assume all subroutines are reachable with the "bl"
16698 instruction.
16699 -mmodel=large
16701 Assume objects may be anywhere in the 32-bit address space (the
16702 compiler generates "seth/add3" instructions to load their
16703 addresses), and assume subroutines may not be reachable with the
16704 "bl" instruction (the compiler generates the much slower
16705 "seth/add3/jl" instruction sequence).
16706 -msdata=none
16708 Disable use of the small data area. Variables are put into one of
16709 ".data", ".bss", or ".rodata" (unless the "section" attribute has
16710 been specified). This is the default.
16711 The small data area consists of sections ".sdata" and ".sbss".
16713 Objects may be explicitly put in the small data area with the
16714 "section" attribute using one of these sections.
16715 -msdata=sdata
16717 Put small global and static data in the small data area, but do not
16718 generate special code to reference them.
16719 -msdata=use
16721 Put small global and static data in the small data area, and
16722 generate special instructions to reference them.
16723 -G num
16725 Put global and static objects less than or equal to num bytes into
16726 the small data or BSS sections instead of the normal data or BSS
16727 sections. The default value of num is 8. The -msdata option must
16728 be set to one of sdata or use for this option to have any effect.
16729 All modules should be compiled with the same -G num value.
16731 Compiling with different values of num may or may not work; if it
16732 doesn't the linker gives an error message---incorrect code is not
16733 generated.
16734 -mdebug
16736 Makes the M32R-specific code in the compiler display some
16737 statistics that might help in debugging programs.
16738 -malign-loops
16740 Align all loops to a 32-byte boundary.
16741 -mno-align-loops
16743 Do not enforce a 32-byte alignment for loops. This is the default.
16744 -missue-rate=number
16746 Issue number instructions per cycle. number can only be 1 or 2.
16747 -mbranch-cost=number
16749 number can only be 1 or 2. If it is 1 then branches are preferred
16750 over conditional code, if it is 2, then the opposite applies.
16751 -mflush-trap=number
16753 Specifies the trap number to use to flush the cache. The default
16754 is 12. Valid numbers are between 0 and 15 inclusive.
16755 -mno-flush-trap
16757 Specifies that the cache cannot be flushed by using a trap.
16758 -mflush-func=name
16760 Specifies the name of the operating system function to call to
16761 flush the cache. The default is _flush_cache, but a function call
16762 is only used if a trap is not available.
16763 -mno-flush-func
16765 Indicates that there is no OS function for flushing the cache.
16766 M680x0 Options
16768 These are the -m options defined for M680x0 and ColdFire processors.
16769 The default settings depend on which architecture was selected when the
16770 compiler was configured; the defaults for the most common choices are
16771 given below.
16772 -march=arch
16774 Generate code for a specific M680x0 or ColdFire instruction set
16775 architecture. Permissible values of arch for M680x0 architectures
16776 are: 68000, 68010, 68020, 68030, 68040, 68060 and cpu32. ColdFire
16777 architectures are selected according to Freescale's ISA
16778 classification and the permissible values are: isaa, isaaplus, isab
16779 and isac.
16780 GCC defines a macro "__mcfarch__" whenever it is generating code
16782 for a ColdFire target. The arch in this macro is one of the -march
16783 arguments given above.
16784 When used together, -march and -mtune select code that runs on a
16786 family of similar processors but that is optimized for a particular
16787 microarchitecture.
16788 -mcpu=cpu
16790 Generate code for a specific M680x0 or ColdFire processor. The
16791 M680x0 cpus are: 68000, 68010, 68020, 68030, 68040, 68060, 68302,
16792 68332 and cpu32. The ColdFire cpus are given by the table below,
16793 which also classifies the CPUs into families:
16794 Family : -mcpu arguments
16796 51 : 51 51ac 51ag 51cn 51em 51je 51jf 51jg 51jm 51mm 51qe 51qm
16797 5206 : 5202 5204 5206
16798 5206e : 5206e
16799 5208 : 5207 5208
16800 5211a : 5210a 5211a
16801 5213 : 5211 5212 5213
16802 5216 : 5214 5216
16803 52235 : 52230 52231 52232 52233 52234 52235
16804 5225 : 5224 5225
16805 52259 : 52252 52254 52255 52256 52258 52259
16806 5235 : 5232 5233 5234 5235 523x
16807 5249 : 5249
16808 5250 : 5250
16809 5271 : 5270 5271
16810 5272 : 5272
16811 5275 : 5274 5275
16812 5282 : 5280 5281 5282 528x
16813 53017 : 53011 53012 53013 53014 53015 53016 53017
16814 5307 : 5307
16815 5329 : 5327 5328 5329 532x
16816 5373 : 5372 5373 537x
16817 5407 : 5407
16818 5475 : 5470 5471 5472 5473 5474 5475 547x 5480 5481 5482 5483 5484
16819 5485
16820 -mcpu=cpu overrides -march=arch if arch is compatible with cpu.
16822 Other combinations of -mcpu and -march are rejected.
16823 GCC defines the macro "__mcf_cpu_cpu" when ColdFire target cpu is
16825 selected. It also defines "__mcf_family_family", where the value
16826 of family is given by the table above.
16827 -mtune=tune
16829 Tune the code for a particular microarchitecture within the
16830 constraints set by -march and -mcpu. The M680x0 microarchitectures
16831 are: 68000, 68010, 68020, 68030, 68040, 68060 and cpu32. The
16832 ColdFire microarchitectures are: cfv1, cfv2, cfv3, cfv4 and cfv4e.
16833 You can also use -mtune=68020-40 for code that needs to run
16835 relatively well on 68020, 68030 and 68040 targets. -mtune=68020-60
16836 is similar but includes 68060 targets as well. These two options
16837 select the same tuning decisions as -m68020-40 and -m68020-60
16838 respectively.
16839 GCC defines the macros "__mcarch" and "__mcarch__" when tuning for
16841 680x0 architecture arch. It also defines "mcarch" unless either
16842 -ansi or a non-GNU -std option is used. If GCC is tuning for a
16843 range of architectures, as selected by -mtune=68020-40 or
16844 -mtune=68020-60, it defines the macros for every architecture in
16845 the range.
16846 GCC also defines the macro "__muarch__" when tuning for ColdFire
16848 microarchitecture uarch, where uarch is one of the arguments given
16849 above.
16850 -m68000
16852 -mc68000
16853 Generate output for a 68000. This is the default when the compiler
16854 is configured for 68000-based systems. It is equivalent to
16855 -march=68000.
16856 Use this option for microcontrollers with a 68000 or EC000 core,
16858 including the 68008, 68302, 68306, 68307, 68322, 68328 and 68356.
16859 -m68010
16861 Generate output for a 68010. This is the default when the compiler
16862 is configured for 68010-based systems. It is equivalent to
16863 -march=68010.
16864 -m68020
16866 -mc68020
16867 Generate output for a 68020. This is the default when the compiler
16868 is configured for 68020-based systems. It is equivalent to
16869 -march=68020.
16870 -m68030
16872 Generate output for a 68030. This is the default when the compiler
16873 is configured for 68030-based systems. It is equivalent to
16874 -march=68030.
16875 -m68040
16877 Generate output for a 68040. This is the default when the compiler
16878 is configured for 68040-based systems. It is equivalent to
16879 -march=68040.
16880 This option inhibits the use of 68881/68882 instructions that have
16882 to be emulated by software on the 68040. Use this option if your
16883 68040 does not have code to emulate those instructions.
16884 -m68060
16886 Generate output for a 68060. This is the default when the compiler
16887 is configured for 68060-based systems. It is equivalent to
16888 -march=68060.
16889 This option inhibits the use of 68020 and 68881/68882 instructions
16891 that have to be emulated by software on the 68060. Use this option
16892 if your 68060 does not have code to emulate those instructions.
16893 -mcpu32
16895 Generate output for a CPU32. This is the default when the compiler
16896 is configured for CPU32-based systems. It is equivalent to
16897 -march=cpu32.
16898 Use this option for microcontrollers with a CPU32 or CPU32+ core,
16900 including the 68330, 68331, 68332, 68333, 68334, 68336, 68340,
16901 68341, 68349 and 68360.
16902 -m5200
16904 Generate output for a 520X ColdFire CPU. This is the default when
16905 the compiler is configured for 520X-based systems. It is
16906 equivalent to -mcpu=5206, and is now deprecated in favor of that
16907 option.
16908 Use this option for microcontroller with a 5200 core, including the
16910 MCF5202, MCF5203, MCF5204 and MCF5206.
16911 -m5206e
16913 Generate output for a 5206e ColdFire CPU. The option is now
16914 deprecated in favor of the equivalent -mcpu=5206e.
16915 -m528x
16917 Generate output for a member of the ColdFire 528X family. The
16918 option is now deprecated in favor of the equivalent -mcpu=528x.
16919 -m5307
16921 Generate output for a ColdFire 5307 CPU. The option is now
16922 deprecated in favor of the equivalent -mcpu=5307.
16923 -m5407
16925 Generate output for a ColdFire 5407 CPU. The option is now
16926 deprecated in favor of the equivalent -mcpu=5407.
16927 -mcfv4e
16929 Generate output for a ColdFire V4e family CPU (e.g. 547x/548x).
16930 This includes use of hardware floating-point instructions. The
16931 option is equivalent to -mcpu=547x, and is now deprecated in favor
16932 of that option.
16933 -m68020-40
16935 Generate output for a 68040, without using any of the new
16936 instructions. This results in code that can run relatively
16937 efficiently on either a 68020/68881 or a 68030 or a 68040. The
16938 generated code does use the 68881 instructions that are emulated on
16939 the 68040.
16940 The option is equivalent to -march=68020 -mtune=68020-40.
16942 -m68020-60
16944 Generate output for a 68060, without using any of the new
16945 instructions. This results in code that can run relatively
16946 efficiently on either a 68020/68881 or a 68030 or a 68040. The
16947 generated code does use the 68881 instructions that are emulated on
16948 the 68060.
16949 The option is equivalent to -march=68020 -mtune=68020-60.
16951 -mhard-float
16953 -m68881
16954 Generate floating-point instructions. This is the default for
16955 68020 and above, and for ColdFire devices that have an FPU. It
16956 defines the macro "__HAVE_68881__" on M680x0 targets and
16957 "__mcffpu__" on ColdFire targets.
16958 -msoft-float
16960 Do not generate floating-point instructions; use library calls
16961 instead. This is the default for 68000, 68010, and 68832 targets.
16962 It is also the default for ColdFire devices that have no FPU.
16963 -mdiv
16965 -mno-div
16966 Generate (do not generate) ColdFire hardware divide and remainder
16967 instructions. If -march is used without -mcpu, the default is "on"
16968 for ColdFire architectures and "off" for M680x0 architectures.
16969 Otherwise, the default is taken from the target CPU (either the
16970 default CPU, or the one specified by -mcpu). For example, the
16971 default is "off" for -mcpu=5206 and "on" for -mcpu=5206e.
16972 GCC defines the macro "__mcfhwdiv__" when this option is enabled.
16974 -mshort
16976 Consider type "int" to be 16 bits wide, like "short int".
16977 Additionally, parameters passed on the stack are also aligned to a
16978 16-bit boundary even on targets whose API mandates promotion to
16979 32-bit.
16980 -mno-short
16982 Do not consider type "int" to be 16 bits wide. This is the
16983 default.
16984 -mnobitfield
16986 -mno-bitfield
16987 Do not use the bit-field instructions. The -m68000, -mcpu32 and
16988 -m5200 options imply -mnobitfield.
16989 -mbitfield
16991 Do use the bit-field instructions. The -m68020 option implies
16992 -mbitfield. This is the default if you use a configuration
16993 designed for a 68020.
16994 -mrtd
16996 Use a different function-calling convention, in which functions
16997 that take a fixed number of arguments return with the "rtd"
16998 instruction, which pops their arguments while returning. This
16999 saves one instruction in the caller since there is no need to pop
17000 the arguments there.
17001 This calling convention is incompatible with the one normally used
17003 on Unix, so you cannot use it if you need to call libraries
17004 compiled with the Unix compiler.
17005 Also, you must provide function prototypes for all functions that
17007 take variable numbers of arguments (including "printf"); otherwise
17008 incorrect code is generated for calls to those functions.
17009 In addition, seriously incorrect code results if you call a
17011 function with too many arguments. (Normally, extra arguments are
17012 harmlessly ignored.)
17013 The "rtd" instruction is supported by the 68010, 68020, 68030,
17015 68040, 68060 and CPU32 processors, but not by the 68000 or 5200.
17016 -mno-rtd
17018 Do not use the calling conventions selected by -mrtd. This is the
17019 default.
17020 -malign-int
17022 -mno-align-int
17023 Control whether GCC aligns "int", "long", "long long", "float",
17024 "double", and "long double" variables on a 32-bit boundary
17025 (-malign-int) or a 16-bit boundary (-mno-align-int). Aligning
17026 variables on 32-bit boundaries produces code that runs somewhat
17027 faster on processors with 32-bit busses at the expense of more
17028 memory.
17029 Warning: if you use the -malign-int switch, GCC aligns structures
17031 containing the above types differently than most published
17032 application binary interface specifications for the m68k.
17033 -mpcrel
17035 Use the pc-relative addressing mode of the 68000 directly, instead
17036 of using a global offset table. At present, this option implies
17037 -fpic, allowing at most a 16-bit offset for pc-relative addressing.
17038 -fPIC is not presently supported with -mpcrel, though this could be
17039 supported for 68020 and higher processors.
17040 -mno-strict-align
17042 -mstrict-align
17043 Do not (do) assume that unaligned memory references are handled by
17044 the system.
17045 -msep-data
17047 Generate code that allows the data segment to be located in a
17048 different area of memory from the text segment. This allows for
17049 execute-in-place in an environment without virtual memory
17050 management. This option implies -fPIC.
17051 -mno-sep-data
17053 Generate code that assumes that the data segment follows the text
17054 segment. This is the default.
17055 -mid-shared-library
17057 Generate code that supports shared libraries via the library ID
17058 method. This allows for execute-in-place and shared libraries in
17059 an environment without virtual memory management. This option
17060 implies -fPIC.
17061 -mno-id-shared-library
17063 Generate code that doesn't assume ID-based shared libraries are
17064 being used. This is the default.
17065 -mshared-library-id=n
17067 Specifies the identification number of the ID-based shared library
17068 being compiled. Specifying a value of 0 generates more compact
17069 code; specifying other values forces the allocation of that number
17070 to the current library, but is no more space- or time-efficient
17071 than omitting this option.
17072 -mxgot
17074 -mno-xgot
17075 When generating position-independent code for ColdFire, generate
17076 code that works if the GOT has more than 8192 entries. This code
17077 is larger and slower than code generated without this option. On
17078 M680x0 processors, this option is not needed; -fPIC suffices.
17079 GCC normally uses a single instruction to load values from the GOT.
17081 While this is relatively efficient, it only works if the GOT is
17082 smaller than about 64k. Anything larger causes the linker to
17083 report an error such as:
17084 relocation truncated to fit: R_68K_GOT16O foobar
17086 If this happens, you should recompile your code with -mxgot. It
17088 should then work with very large GOTs. However, code generated
17089 with -mxgot is less efficient, since it takes 4 instructions to
17090 fetch the value of a global symbol.
17091 Note that some linkers, including newer versions of the GNU linker,
17093 can create multiple GOTs and sort GOT entries. If you have such a
17094 linker, you should only need to use -mxgot when compiling a single
17095 object file that accesses more than 8192 GOT entries. Very few do.
17096 These options have no effect unless GCC is generating position-
17098 independent code.
17099 -mlong-jump-table-offsets
17101 Use 32-bit offsets in "switch" tables. The default is to use
17102 16-bit offsets.
17103 MCore Options
17105 These are the -m options defined for the Motorola M*Core processors.
17106 -mhardlit
17108 -mno-hardlit
17109 Inline constants into the code stream if it can be done in two
17110 instructions or less.
17111 -mdiv
17113 -mno-div
17114 Use the divide instruction. (Enabled by default).
17115 -mrelax-immediate
17117 -mno-relax-immediate
17118 Allow arbitrary-sized immediates in bit operations.
17119 -mwide-bitfields
17121 -mno-wide-bitfields
17122 Always treat bit-fields as "int"-sized.
17123 -m4byte-functions
17125 -mno-4byte-functions
17126 Force all functions to be aligned to a 4-byte boundary.
17127 -mcallgraph-data
17129 -mno-callgraph-data
17130 Emit callgraph information.
17131 -mslow-bytes
17133 -mno-slow-bytes
17134 Prefer word access when reading byte quantities.
17135 -mlittle-endian
17137 -mbig-endian
17138 Generate code for a little-endian target.
17139 -m210
17141 -m340
17142 Generate code for the 210 processor.
17143 -mno-lsim
17145 Assume that runtime support has been provided and so omit the
17146 simulator library (libsim.a) from the linker command line.
17147 -mstack-increment=size
17149 Set the maximum amount for a single stack increment operation.
17150 Large values can increase the speed of programs that contain
17151 functions that need a large amount of stack space, but they can
17152 also trigger a segmentation fault if the stack is extended too
17153 much. The default value is 0x1000.
17154 MeP Options
17156 -mabsdiff
17157 Enables the "abs" instruction, which is the absolute difference
17158 between two registers.
17159 -mall-opts
17161 Enables all the optional instructions---average, multiply, divide,
17162 bit operations, leading zero, absolute difference, min/max, clip,
17163 and saturation.
17164 -maverage
17166 Enables the "ave" instruction, which computes the average of two
17167 registers.
17168 -mbased=n
17170 Variables of size n bytes or smaller are placed in the ".based"
17171 section by default. Based variables use the $tp register as a base
17172 register, and there is a 128-byte limit to the ".based" section.
17173 -mbitops
17175 Enables the bit operation instructions---bit test ("btstm"), set
17176 ("bsetm"), clear ("bclrm"), invert ("bnotm"), and test-and-set
17177 ("tas").
17178 -mc=name
17180 Selects which section constant data is placed in. name may be
17181 tiny, near, or far.
17182 -mclip
17184 Enables the "clip" instruction. Note that -mclip is not useful
17185 unless you also provide -mminmax.
17186 -mconfig=name
17188 Selects one of the built-in core configurations. Each MeP chip has
17189 one or more modules in it; each module has a core CPU and a variety
17190 of coprocessors, optional instructions, and peripherals. The
17191 "MeP-Integrator" tool, not part of GCC, provides these
17192 configurations through this option; using this option is the same
17193 as using all the corresponding command-line options. The default
17194 configuration is default.
17195 -mcop
17197 Enables the coprocessor instructions. By default, this is a 32-bit
17198 coprocessor. Note that the coprocessor is normally enabled via the
17199 -mconfig= option.
17200 -mcop32
17202 Enables the 32-bit coprocessor's instructions.
17203 -mcop64
17205 Enables the 64-bit coprocessor's instructions.
17206 -mivc2
17208 Enables IVC2 scheduling. IVC2 is a 64-bit VLIW coprocessor.
17209 -mdc
17211 Causes constant variables to be placed in the ".near" section.
17212 -mdiv
17214 Enables the "div" and "divu" instructions.
17215 -meb
17217 Generate big-endian code.
17218 -mel
17220 Generate little-endian code.
17221 -mio-volatile
17223 Tells the compiler that any variable marked with the "io" attribute
17224 is to be considered volatile.
17225 -ml Causes variables to be assigned to the ".far" section by default.
17227 -mleadz
17229 Enables the "leadz" (leading zero) instruction.
17230 -mm Causes variables to be assigned to the ".near" section by default.
17232 -mminmax
17234 Enables the "min" and "max" instructions.
17235 -mmult
17237 Enables the multiplication and multiply-accumulate instructions.
17238 -mno-opts
17240 Disables all the optional instructions enabled by -mall-opts.
17241 -mrepeat
17243 Enables the "repeat" and "erepeat" instructions, used for low-
17244 overhead looping.
17245 -ms Causes all variables to default to the ".tiny" section. Note that
17247 there is a 65536-byte limit to this section. Accesses to these
17248 variables use the %gp base register.
17249 -msatur
17251 Enables the saturation instructions. Note that the compiler does
17252 not currently generate these itself, but this option is included
17253 for compatibility with other tools, like "as".
17254 -msdram
17256 Link the SDRAM-based runtime instead of the default ROM-based
17257 runtime.
17258 -msim
17260 Link the simulator run-time libraries.
17261 -msimnovec
17263 Link the simulator runtime libraries, excluding built-in support
17264 for reset and exception vectors and tables.
17265 -mtf
17267 Causes all functions to default to the ".far" section. Without
17268 this option, functions default to the ".near" section.
17269 -mtiny=n
17271 Variables that are n bytes or smaller are allocated to the ".tiny"
17272 section. These variables use the $gp base register. The default
17273 for this option is 4, but note that there's a 65536-byte limit to
17274 the ".tiny" section.
17275 MicroBlaze Options
17277 -msoft-float
17278 Use software emulation for floating point (default).
17279 -mhard-float
17281 Use hardware floating-point instructions.
17282 -mmemcpy
17284 Do not optimize block moves, use "memcpy".
17285 -mno-clearbss
17287 This option is deprecated. Use -fno-zero-initialized-in-bss
17288 instead.
17289 -mcpu=cpu-type
17291 Use features of, and schedule code for, the given CPU. Supported
17292 values are in the format vX.YY.Z, where X is a major version, YY is
17293 the minor version, and Z is compatibility code. Example values are
17294 v3.00.a, v4.00.b, v5.00.a, v5.00.b, v6.00.a.
17295 -mxl-soft-mul
17297 Use software multiply emulation (default).
17298 -mxl-soft-div
17300 Use software emulation for divides (default).
17301 -mxl-barrel-shift
17303 Use the hardware barrel shifter.
17304 -mxl-pattern-compare
17306 Use pattern compare instructions.
17307 -msmall-divides
17309 Use table lookup optimization for small signed integer divisions.
17310 -mxl-stack-check
17312 This option is deprecated. Use -fstack-check instead.
17313 -mxl-gp-opt
17315 Use GP-relative ".sdata"/".sbss" sections.
17316 -mxl-multiply-high
17318 Use multiply high instructions for high part of 32x32 multiply.
17319 -mxl-float-convert
17321 Use hardware floating-point conversion instructions.
17322 -mxl-float-sqrt
17324 Use hardware floating-point square root instruction.
17325 -mbig-endian
17327 Generate code for a big-endian target.
17328 -mlittle-endian
17330 Generate code for a little-endian target.
17331 -mxl-reorder
17333 Use reorder instructions (swap and byte reversed load/store).
17334 -mxl-mode-app-model
17336 Select application model app-model. Valid models are
17337 executable
17339 normal executable (default), uses startup code crt0.o.
17340 xmdstub
17342 for use with Xilinx Microprocessor Debugger (XMD) based
17343 software intrusive debug agent called xmdstub. This uses
17344 startup file crt1.o and sets the start address of the program
17345 to 0x800.
17346 bootstrap
17348 for applications that are loaded using a bootloader. This
17349 model uses startup file crt2.o which does not contain a
17350 processor reset vector handler. This is suitable for
17351 transferring control on a processor reset to the bootloader
17352 rather than the application.
17353 novectors
17355 for applications that do not require any of the MicroBlaze
17356 vectors. This option may be useful for applications running
17357 within a monitoring application. This model uses crt3.o as a
17358 startup file.
17359 Option -xl-mode-app-model is a deprecated alias for -mxl-mode-app-
17361 model.
17362 MIPS Options
17364 -EB Generate big-endian code.
17365 -EL Generate little-endian code. This is the default for mips*el-*-*
17367 configurations.
17368 -march=arch
17370 Generate code that runs on arch, which can be the name of a generic
17371 MIPS ISA, or the name of a particular processor. The ISA names
17372 are: mips1, mips2, mips3, mips4, mips32, mips32r2, mips32r3,
17373 mips32r5, mips32r6, mips64, mips64r2, mips64r3, mips64r5 and
17374 mips64r6. The processor names are: 4kc, 4km, 4kp, 4ksc, 4kec,
17375 4kem, 4kep, 4ksd, 5kc, 5kf, 20kc, 24kc, 24kf2_1, 24kf1_1, 24kec,
17376 24kef2_1, 24kef1_1, 34kc, 34kf2_1, 34kf1_1, 34kn, 74kc, 74kf2_1,
17377 74kf1_1, 74kf3_2, 1004kc, 1004kf2_1, 1004kf1_1, i6400, interaptiv,
17378 loongson2e, loongson2f, loongson3a, m4k, m14k, m14kc, m14ke,
17379 m14kec, m5100, m5101, octeon, octeon+, octeon2, octeon3, orion,
17380 p5600, r2000, r3000, r3900, r4000, r4400, r4600, r4650, r4700,
17381 r6000, r8000, rm7000, rm9000, r10000, r12000, r14000, r16000, sb1,
17382 sr71000, vr4100, vr4111, vr4120, vr4130, vr4300, vr5000, vr5400,
17383 vr5500, xlr and xlp. The special value from-abi selects the most
17384 compatible architecture for the selected ABI (that is, mips1 for
17385 32-bit ABIs and mips3 for 64-bit ABIs).
17386 The native Linux/GNU toolchain also supports the value native,
17388 which selects the best architecture option for the host processor.
17389 -march=native has no effect if GCC does not recognize the
17390 processor.
17391 In processor names, a final 000 can be abbreviated as k (for
17393 example, -march=r2k). Prefixes are optional, and vr may be written
17394 r.
17395 Names of the form nf2_1 refer to processors with FPUs clocked at
17397 half the rate of the core, names of the form nf1_1 refer to
17398 processors with FPUs clocked at the same rate as the core, and
17399 names of the form nf3_2 refer to processors with FPUs clocked a
17400 ratio of 3:2 with respect to the core. For compatibility reasons,
17401 nf is accepted as a synonym for nf2_1 while nx and bfx are accepted
17402 as synonyms for nf1_1.
17403 GCC defines two macros based on the value of this option. The
17405 first is "_MIPS_ARCH", which gives the name of target architecture,
17406 as a string. The second has the form "_MIPS_ARCH_foo", where foo
17407 is the capitalized value of "_MIPS_ARCH". For example,
17408 -march=r2000 sets "_MIPS_ARCH" to "r2000" and defines the macro
17409 "_MIPS_ARCH_R2000".
17410 Note that the "_MIPS_ARCH" macro uses the processor names given
17412 above. In other words, it has the full prefix and does not
17413 abbreviate 000 as k. In the case of from-abi, the macro names the
17414 resolved architecture (either "mips1" or "mips3"). It names the
17415 default architecture when no -march option is given.
17416 -mtune=arch
17418 Optimize for arch. Among other things, this option controls the
17419 way instructions are scheduled, and the perceived cost of
17420 arithmetic operations. The list of arch values is the same as for
17421 -march.
17422 When this option is not used, GCC optimizes for the processor
17424 specified by -march. By using -march and -mtune together, it is
17425 possible to generate code that runs on a family of processors, but
17426 optimize the code for one particular member of that family.
17427 -mtune defines the macros "_MIPS_TUNE" and "_MIPS_TUNE_foo", which
17429 work in the same way as the -march ones described above.
17430 -mips1
17432 Equivalent to -march=mips1.
17433 -mips2
17435 Equivalent to -march=mips2.
17436 -mips3
17438 Equivalent to -march=mips3.
17439 -mips4
17441 Equivalent to -march=mips4.
17442 -mips32
17444 Equivalent to -march=mips32.
17445 -mips32r3
17447 Equivalent to -march=mips32r3.
17448 -mips32r5
17450 Equivalent to -march=mips32r5.
17451 -mips32r6
17453 Equivalent to -march=mips32r6.
17454 -mips64
17456 Equivalent to -march=mips64.
17457 -mips64r2
17459 Equivalent to -march=mips64r2.
17460 -mips64r3
17462 Equivalent to -march=mips64r3.
17463 -mips64r5
17465 Equivalent to -march=mips64r5.
17466 -mips64r6
17468 Equivalent to -march=mips64r6.
17469 -mips16
17471 -mno-mips16
17472 Generate (do not generate) MIPS16 code. If GCC is targeting a
17473 MIPS32 or MIPS64 architecture, it makes use of the MIPS16e ASE.
17474 MIPS16 code generation can also be controlled on a per-function
17476 basis by means of "mips16" and "nomips16" attributes.
17477 -mflip-mips16
17479 Generate MIPS16 code on alternating functions. This option is
17480 provided for regression testing of mixed MIPS16/non-MIPS16 code
17481 generation, and is not intended for ordinary use in compiling user
17482 code.
17483 -minterlink-compressed
17485 -mno-interlink-compressed
17486 Require (do not require) that code using the standard
17487 (uncompressed) MIPS ISA be link-compatible with MIPS16 and
17488 microMIPS code, and vice versa.
17489 For example, code using the standard ISA encoding cannot jump
17491 directly to MIPS16 or microMIPS code; it must either use a call or
17492 an indirect jump. -minterlink-compressed therefore disables direct
17493 jumps unless GCC knows that the target of the jump is not
17494 compressed.
17495 -minterlink-mips16
17497 -mno-interlink-mips16
17498 Aliases of -minterlink-compressed and -mno-interlink-compressed.
17499 These options predate the microMIPS ASE and are retained for
17500 backwards compatibility.
17501 -mabi=32
17503 -mabi=o64
17504 -mabi=n32
17505 -mabi=64
17506 -mabi=eabi
17507 Generate code for the given ABI.
17508 Note that the EABI has a 32-bit and a 64-bit variant. GCC normally
17510 generates 64-bit code when you select a 64-bit architecture, but
17511 you can use -mgp32 to get 32-bit code instead.
17512 For information about the O64 ABI, see
17514 <http://gcc.gnu.org/projects/mipso64-abi.html>.
17515 GCC supports a variant of the o32 ABI in which floating-point
17517 registers are 64 rather than 32 bits wide. You can select this
17518 combination with -mabi=32 -mfp64. This ABI relies on the "mthc1"
17519 and "mfhc1" instructions and is therefore only supported for
17520 MIPS32R2, MIPS32R3 and MIPS32R5 processors.
17521 The register assignments for arguments and return values remain the
17523 same, but each scalar value is passed in a single 64-bit register
17524 rather than a pair of 32-bit registers. For example, scalar
17525 floating-point values are returned in $f0 only, not a $f0/$f1 pair.
17526 The set of call-saved registers also remains the same in that the
17527 even-numbered double-precision registers are saved.
17528 Two additional variants of the o32 ABI are supported to enable a
17530 transition from 32-bit to 64-bit registers. These are FPXX
17531 (-mfpxx) and FP64A (-mfp64 -mno-odd-spreg). The FPXX extension
17532 mandates that all code must execute correctly when run using 32-bit
17533 or 64-bit registers. The code can be interlinked with either FP32
17534 or FP64, but not both. The FP64A extension is similar to the FP64
17535 extension but forbids the use of odd-numbered single-precision
17536 registers. This can be used in conjunction with the "FRE" mode of
17537 FPUs in MIPS32R5 processors and allows both FP32 and FP64A code to
17538 interlink and run in the same process without changing FPU modes.
17539 -mabicalls
17541 -mno-abicalls
17542 Generate (do not generate) code that is suitable for SVR4-style
17543 dynamic objects. -mabicalls is the default for SVR4-based systems.
17544 -mshared
17546 -mno-shared
17547 Generate (do not generate) code that is fully position-independent,
17548 and that can therefore be linked into shared libraries. This
17549 option only affects -mabicalls.
17550 All -mabicalls code has traditionally been position-independent,
17552 regardless of options like -fPIC and -fpic. However, as an
17553 extension, the GNU toolchain allows executables to use absolute
17554 accesses for locally-binding symbols. It can also use shorter GP
17555 initialization sequences and generate direct calls to locally-
17556 defined functions. This mode is selected by -mno-shared.
17557 -mno-shared depends on binutils 2.16 or higher and generates
17559 objects that can only be linked by the GNU linker. However, the
17560 option does not affect the ABI of the final executable; it only
17561 affects the ABI of relocatable objects. Using -mno-shared
17562 generally makes executables both smaller and quicker.
17563 -mshared is the default.
17565 -mplt
17567 -mno-plt
17568 Assume (do not assume) that the static and dynamic linkers support
17569 PLTs and copy relocations. This option only affects -mno-shared
17570 -mabicalls. For the n64 ABI, this option has no effect without
17571 -msym32.
17572 You can make -mplt the default by configuring GCC with
17574 --with-mips-plt. The default is -mno-plt otherwise.
17575 -mxgot
17577 -mno-xgot
17578 Lift (do not lift) the usual restrictions on the size of the global
17579 offset table.
17580 GCC normally uses a single instruction to load values from the GOT.
17582 While this is relatively efficient, it only works if the GOT is
17583 smaller than about 64k. Anything larger causes the linker to
17584 report an error such as:
17585 relocation truncated to fit: R_MIPS_GOT16 foobar
17587 If this happens, you should recompile your code with -mxgot. This
17589 works with very large GOTs, although the code is also less
17590 efficient, since it takes three instructions to fetch the value of
17591 a global symbol.
17592 Note that some linkers can create multiple GOTs. If you have such
17594 a linker, you should only need to use -mxgot when a single object
17595 file accesses more than 64k's worth of GOT entries. Very few do.
17596 These options have no effect unless GCC is generating position
17598 independent code.
17599 -mgp32
17601 Assume that general-purpose registers are 32 bits wide.
17602 -mgp64
17604 Assume that general-purpose registers are 64 bits wide.
17605 -mfp32
17607 Assume that floating-point registers are 32 bits wide.
17608 -mfp64
17610 Assume that floating-point registers are 64 bits wide.
17611 -mfpxx
17613 Do not assume the width of floating-point registers.
17614 -mhard-float
17616 Use floating-point coprocessor instructions.
17617 -msoft-float
17619 Do not use floating-point coprocessor instructions. Implement
17620 floating-point calculations using library calls instead.
17621 -mno-float
17623 Equivalent to -msoft-float, but additionally asserts that the
17624 program being compiled does not perform any floating-point
17625 operations. This option is presently supported only by some bare-
17626 metal MIPS configurations, where it may select a special set of
17627 libraries that lack all floating-point support (including, for
17628 example, the floating-point "printf" formats). If code compiled
17629 with -mno-float accidentally contains floating-point operations, it
17630 is likely to suffer a link-time or run-time failure.
17631 -msingle-float
17633 Assume that the floating-point coprocessor only supports single-
17634 precision operations.
17635 -mdouble-float
17637 Assume that the floating-point coprocessor supports double-
17638 precision operations. This is the default.
17639 -modd-spreg
17641 -mno-odd-spreg
17642 Enable the use of odd-numbered single-precision floating-point
17643 registers for the o32 ABI. This is the default for processors that
17644 are known to support these registers. When using the o32 FPXX ABI,
17645 -mno-odd-spreg is set by default.
17646 -mabs=2008
17648 -mabs=legacy
17649 These options control the treatment of the special not-a-number
17650 (NaN) IEEE 754 floating-point data with the "abs.fmt" and "neg.fmt"
17651 machine instructions.
17652 By default or when -mabs=legacy is used the legacy treatment is
17654 selected. In this case these instructions are considered
17655 arithmetic and avoided where correct operation is required and the
17656 input operand might be a NaN. A longer sequence of instructions
17657 that manipulate the sign bit of floating-point datum manually is
17658 used instead unless the -ffinite-math-only option has also been
17659 specified.
17660 The -mabs=2008 option selects the IEEE 754-2008 treatment. In this
17662 case these instructions are considered non-arithmetic and therefore
17663 operating correctly in all cases, including in particular where the
17664 input operand is a NaN. These instructions are therefore always
17665 used for the respective operations.
17666 -mnan=2008
17668 -mnan=legacy
17669 These options control the encoding of the special not-a-number
17670 (NaN) IEEE 754 floating-point data.
17671 The -mnan=legacy option selects the legacy encoding. In this case
17673 quiet NaNs (qNaNs) are denoted by the first bit of their trailing
17674 significand field being 0, whereas signaling NaNs (sNaNs) are
17675 denoted by the first bit of their trailing significand field being
17676 1.
17677 The -mnan=2008 option selects the IEEE 754-2008 encoding. In this
17679 case qNaNs are denoted by the first bit of their trailing
17680 significand field being 1, whereas sNaNs are denoted by the first
17681 bit of their trailing significand field being 0.
17682 The default is -mnan=legacy unless GCC has been configured with
17684 --with-nan=2008.
17685 -mllsc
17687 -mno-llsc
17688 Use (do not use) ll, sc, and sync instructions to implement atomic
17689 memory built-in functions. When neither option is specified, GCC
17690 uses the instructions if the target architecture supports them.
17691 -mllsc is useful if the runtime environment can emulate the
17693 instructions and -mno-llsc can be useful when compiling for
17694 nonstandard ISAs. You can make either option the default by
17695 configuring GCC with --with-llsc and --without-llsc respectively.
17696 --with-llsc is the default for some configurations; see the
17697 installation documentation for details.
17698 -mdsp
17700 -mno-dsp
17701 Use (do not use) revision 1 of the MIPS DSP ASE.
17702 This option defines the preprocessor macro "__mips_dsp". It also
17703 defines "__mips_dsp_rev" to 1.
17704 -mdspr2
17706 -mno-dspr2
17707 Use (do not use) revision 2 of the MIPS DSP ASE.
17708 This option defines the preprocessor macros "__mips_dsp" and
17709 "__mips_dspr2". It also defines "__mips_dsp_rev" to 2.
17710 -msmartmips
17712 -mno-smartmips
17713 Use (do not use) the MIPS SmartMIPS ASE.
17714 -mpaired-single
17716 -mno-paired-single
17717 Use (do not use) paired-single floating-point instructions.
17718 This option requires hardware floating-point support to be
17719 enabled.
17720 -mdmx
17722 -mno-mdmx
17723 Use (do not use) MIPS Digital Media Extension instructions. This
17724 option can only be used when generating 64-bit code and requires
17725 hardware floating-point support to be enabled.
17726 -mips3d
17728 -mno-mips3d
17729 Use (do not use) the MIPS-3D ASE. The option -mips3d implies
17730 -mpaired-single.
17731 -mmicromips
17733 -mno-micromips
17734 Generate (do not generate) microMIPS code.
17735 MicroMIPS code generation can also be controlled on a per-function
17737 basis by means of "micromips" and "nomicromips" attributes.
17738 -mmt
17740 -mno-mt
17741 Use (do not use) MT Multithreading instructions.
17742 -mmcu
17744 -mno-mcu
17745 Use (do not use) the MIPS MCU ASE instructions.
17746 -meva
17748 -mno-eva
17749 Use (do not use) the MIPS Enhanced Virtual Addressing instructions.
17750 -mvirt
17752 -mno-virt
17753 Use (do not use) the MIPS Virtualization (VZ) instructions.
17754 -mxpa
17756 -mno-xpa
17757 Use (do not use) the MIPS eXtended Physical Address (XPA)
17758 instructions.
17759 -mlong64
17761 Force "long" types to be 64 bits wide. See -mlong32 for an
17762 explanation of the default and the way that the pointer size is
17763 determined.
17764 -mlong32
17766 Force "long", "int", and pointer types to be 32 bits wide.
17767 The default size of "int"s, "long"s and pointers depends on the
17769 ABI. All the supported ABIs use 32-bit "int"s. The n64 ABI uses
17770 64-bit "long"s, as does the 64-bit EABI; the others use 32-bit
17771 "long"s. Pointers are the same size as "long"s, or the same size
17772 as integer registers, whichever is smaller.
17773 -msym32
17775 -mno-sym32
17776 Assume (do not assume) that all symbols have 32-bit values,
17777 regardless of the selected ABI. This option is useful in
17778 combination with -mabi=64 and -mno-abicalls because it allows GCC
17779 to generate shorter and faster references to symbolic addresses.
17780 -G num
17782 Put definitions of externally-visible data in a small data section
17783 if that data is no bigger than num bytes. GCC can then generate
17784 more efficient accesses to the data; see -mgpopt for details.
17785 The default -G option depends on the configuration.
17787 -mlocal-sdata
17789 -mno-local-sdata
17790 Extend (do not extend) the -G behavior to local data too, such as
17791 to static variables in C. -mlocal-sdata is the default for all
17792 configurations.
17793 If the linker complains that an application is using too much small
17795 data, you might want to try rebuilding the less performance-
17796 critical parts with -mno-local-sdata. You might also want to build
17797 large libraries with -mno-local-sdata, so that the libraries leave
17798 more room for the main program.
17799 -mextern-sdata
17801 -mno-extern-sdata
17802 Assume (do not assume) that externally-defined data is in a small
17803 data section if the size of that data is within the -G limit.
17804 -mextern-sdata is the default for all configurations.
17805 If you compile a module Mod with -mextern-sdata -G num -mgpopt, and
17807 Mod references a variable Var that is no bigger than num bytes, you
17808 must make sure that Var is placed in a small data section. If Var
17809 is defined by another module, you must either compile that module
17810 with a high-enough -G setting or attach a "section" attribute to
17811 Var's definition. If Var is common, you must link the application
17812 with a high-enough -G setting.
17813 The easiest way of satisfying these restrictions is to compile and
17815 link every module with the same -G option. However, you may wish
17816 to build a library that supports several different small data
17817 limits. You can do this by compiling the library with the highest
17818 supported -G setting and additionally using -mno-extern-sdata to
17819 stop the library from making assumptions about externally-defined
17820 data.
17821 -mgpopt
17823 -mno-gpopt
17824 Use (do not use) GP-relative accesses for symbols that are known to
17825 be in a small data section; see -G, -mlocal-sdata and
17826 -mextern-sdata. -mgpopt is the default for all configurations.
17827 -mno-gpopt is useful for cases where the $gp register might not
17829 hold the value of "_gp". For example, if the code is part of a
17830 library that might be used in a boot monitor, programs that call
17831 boot monitor routines pass an unknown value in $gp. (In such
17832 situations, the boot monitor itself is usually compiled with -G0.)
17833 -mno-gpopt implies -mno-local-sdata and -mno-extern-sdata.
17835 -membedded-data
17837 -mno-embedded-data
17838 Allocate variables to the read-only data section first if possible,
17839 then next in the small data section if possible, otherwise in data.
17840 This gives slightly slower code than the default, but reduces the
17841 amount of RAM required when executing, and thus may be preferred
17842 for some embedded systems.
17843 -muninit-const-in-rodata
17845 -mno-uninit-const-in-rodata
17846 Put uninitialized "const" variables in the read-only data section.
17847 This option is only meaningful in conjunction with -membedded-data.
17848 -mcode-readable=setting
17850 Specify whether GCC may generate code that reads from executable
17851 sections. There are three possible settings:
17852 -mcode-readable=yes
17854 Instructions may freely access executable sections. This is
17855 the default setting.
17856 -mcode-readable=pcrel
17858 MIPS16 PC-relative load instructions can access executable
17859 sections, but other instructions must not do so. This option
17860 is useful on 4KSc and 4KSd processors when the code TLBs have
17861 the Read Inhibit bit set. It is also useful on processors that
17862 can be configured to have a dual instruction/data SRAM
17863 interface and that, like the M4K, automatically redirect PC-
17864 relative loads to the instruction RAM.
17865 -mcode-readable=no
17867 Instructions must not access executable sections. This option
17868 can be useful on targets that are configured to have a dual
17869 instruction/data SRAM interface but that (unlike the M4K) do
17870 not automatically redirect PC-relative loads to the instruction
17871 RAM.
17872 -msplit-addresses
17874 -mno-split-addresses
17875 Enable (disable) use of the "%hi()" and "%lo()" assembler
17876 relocation operators. This option has been superseded by
17877 -mexplicit-relocs but is retained for backwards compatibility.
17878 -mexplicit-relocs
17880 -mno-explicit-relocs
17881 Use (do not use) assembler relocation operators when dealing with
17882 symbolic addresses. The alternative, selected by
17883 -mno-explicit-relocs, is to use assembler macros instead.
17884 -mexplicit-relocs is the default if GCC was configured to use an
17886 assembler that supports relocation operators.
17887 -mcheck-zero-division
17889 -mno-check-zero-division
17890 Trap (do not trap) on integer division by zero.
17891 The default is -mcheck-zero-division.
17893 -mdivide-traps
17895 -mdivide-breaks
17896 MIPS systems check for division by zero by generating either a
17897 conditional trap or a break instruction. Using traps results in
17898 smaller code, but is only supported on MIPS II and later. Also,
17899 some versions of the Linux kernel have a bug that prevents trap
17900 from generating the proper signal ("SIGFPE"). Use -mdivide-traps
17901 to allow conditional traps on architectures that support them and
17902 -mdivide-breaks to force the use of breaks.
17903 The default is usually -mdivide-traps, but this can be overridden
17905 at configure time using --with-divide=breaks. Divide-by-zero
17906 checks can be completely disabled using -mno-check-zero-division.
17907 -mload-store-pairs
17909 -mno-load-store-pairs
17910 Enable (disable) an optimization that pairs consecutive load or
17911 store instructions to enable load/store bonding. This option is
17912 enabled by default but only takes effect when the selected
17913 architecture is known to support bonding.
17914 -mmemcpy
17916 -mno-memcpy
17917 Force (do not force) the use of "memcpy" for non-trivial block
17918 moves. The default is -mno-memcpy, which allows GCC to inline most
17919 constant-sized copies.
17920 -mlong-calls
17922 -mno-long-calls
17923 Disable (do not disable) use of the "jal" instruction. Calling
17924 functions using "jal" is more efficient but requires the caller and
17925 callee to be in the same 256 megabyte segment.
17926 This option has no effect on abicalls code. The default is
17928 -mno-long-calls.
17929 -mmad
17931 -mno-mad
17932 Enable (disable) use of the "mad", "madu" and "mul" instructions,
17933 as provided by the R4650 ISA.
17934 -mimadd
17936 -mno-imadd
17937 Enable (disable) use of the "madd" and "msub" integer instructions.
17938 The default is -mimadd on architectures that support "madd" and
17939 "msub" except for the 74k architecture where it was found to
17940 generate slower code.
17941 -mfused-madd
17943 -mno-fused-madd
17944 Enable (disable) use of the floating-point multiply-accumulate
17945 instructions, when they are available. The default is
17946 -mfused-madd.
17947 On the R8000 CPU when multiply-accumulate instructions are used,
17949 the intermediate product is calculated to infinite precision and is
17950 not subject to the FCSR Flush to Zero bit. This may be undesirable
17951 in some circumstances. On other processors the result is
17952 numerically identical to the equivalent computation using separate
17953 multiply, add, subtract and negate instructions.
17954 -nocpp
17956 Tell the MIPS assembler to not run its preprocessor over user
17957 assembler files (with a .s suffix) when assembling them.
17958 -mfix-24k
17960 -mno-fix-24k
17961 Work around the 24K E48 (lost data on stores during refill) errata.
17962 The workarounds are implemented by the assembler rather than by
17963 GCC.
17964 -mfix-r4000
17966 -mno-fix-r4000
17967 Work around certain R4000 CPU errata:
17968 - A double-word or a variable shift may give an incorrect result
17970 if executed immediately after starting an integer division.
17971 - A double-word or a variable shift may give an incorrect result
17973 if executed while an integer multiplication is in progress.
17974 - An integer division may give an incorrect result if started in
17976 a delay slot of a taken branch or a jump.
17977 -mfix-r4400
17979 -mno-fix-r4400
17980 Work around certain R4400 CPU errata:
17981 - A double-word or a variable shift may give an incorrect result
17983 if executed immediately after starting an integer division.
17984 -mfix-r10000
17986 -mno-fix-r10000
17987 Work around certain R10000 errata:
17988 - "ll"/"sc" sequences may not behave atomically on revisions
17990 prior to 3.0. They may deadlock on revisions 2.6 and earlier.
17991 This option can only be used if the target architecture supports
17993 branch-likely instructions. -mfix-r10000 is the default when
17994 -march=r10000 is used; -mno-fix-r10000 is the default otherwise.
17995 -mfix-rm7000
17997 -mno-fix-rm7000
17998 Work around the RM7000 "dmult"/"dmultu" errata. The workarounds
17999 are implemented by the assembler rather than by GCC.
18000 -mfix-vr4120
18002 -mno-fix-vr4120
18003 Work around certain VR4120 errata:
18004 - "dmultu" does not always produce the correct result.
18006 - "div" and "ddiv" do not always produce the correct result if
18008 one of the operands is negative.
18009 The workarounds for the division errata rely on special functions
18011 in libgcc.a. At present, these functions are only provided by the
18012 "mips64vr*-elf" configurations.
18013 Other VR4120 errata require a NOP to be inserted between certain
18015 pairs of instructions. These errata are handled by the assembler,
18016 not by GCC itself.
18017 -mfix-vr4130
18019 Work around the VR4130 "mflo"/"mfhi" errata. The workarounds are
18020 implemented by the assembler rather than by GCC, although GCC
18021 avoids using "mflo" and "mfhi" if the VR4130 "macc", "macchi",
18022 "dmacc" and "dmacchi" instructions are available instead.
18023 -mfix-sb1
18025 -mno-fix-sb1
18026 Work around certain SB-1 CPU core errata. (This flag currently
18027 works around the SB-1 revision 2 "F1" and "F2" floating-point
18028 errata.)
18029 -mr10k-cache-barrier=setting
18031 Specify whether GCC should insert cache barriers to avoid the side
18032 effects of speculation on R10K processors.
18033 In common with many processors, the R10K tries to predict the
18035 outcome of a conditional branch and speculatively executes
18036 instructions from the "taken" branch. It later aborts these
18037 instructions if the predicted outcome is wrong. However, on the
18038 R10K, even aborted instructions can have side effects.
18039 This problem only affects kernel stores and, depending on the
18041 system, kernel loads. As an example, a speculatively-executed
18042 store may load the target memory into cache and mark the cache line
18043 as dirty, even if the store itself is later aborted. If a DMA
18044 operation writes to the same area of memory before the "dirty" line
18045 is flushed, the cached data overwrites the DMA-ed data. See the
18046 R10K processor manual for a full description, including other
18047 potential problems.
18048 One workaround is to insert cache barrier instructions before every
18050 memory access that might be speculatively executed and that might
18051 have side effects even if aborted. -mr10k-cache-barrier=setting
18052 controls GCC's implementation of this workaround. It assumes that
18053 aborted accesses to any byte in the following regions does not have
18054 side effects:
18055 1. the memory occupied by the current function's stack frame;
18057 2. the memory occupied by an incoming stack argument;
18059 3. the memory occupied by an object with a link-time-constant
18061 address.
18062 It is the kernel's responsibility to ensure that speculative
18064 accesses to these regions are indeed safe.
18065 If the input program contains a function declaration such as:
18067 void foo (void);
18069 then the implementation of "foo" must allow "j foo" and "jal foo"
18071 to be executed speculatively. GCC honors this restriction for
18072 functions it compiles itself. It expects non-GCC functions (such
18073 as hand-written assembly code) to do the same.
18074 The option has three forms:
18076 -mr10k-cache-barrier=load-store
18078 Insert a cache barrier before a load or store that might be
18079 speculatively executed and that might have side effects even if
18080 aborted.
18081 -mr10k-cache-barrier=store
18083 Insert a cache barrier before a store that might be
18084 speculatively executed and that might have side effects even if
18085 aborted.
18086 -mr10k-cache-barrier=none
18088 Disable the insertion of cache barriers. This is the default
18089 setting.
18090 -mflush-func=func
18092 -mno-flush-func
18093 Specifies the function to call to flush the I and D caches, or to
18094 not call any such function. If called, the function must take the
18095 same arguments as the common "_flush_func", that is, the address of
18096 the memory range for which the cache is being flushed, the size of
18097 the memory range, and the number 3 (to flush both caches). The
18098 default depends on the target GCC was configured for, but commonly
18099 is either "_flush_func" or "__cpu_flush".
18100 mbranch-cost=num
18102 Set the cost of branches to roughly num "simple" instructions.
18103 This cost is only a heuristic and is not guaranteed to produce
18104 consistent results across releases. A zero cost redundantly
18105 selects the default, which is based on the -mtune setting.
18106 -mbranch-likely
18108 -mno-branch-likely
18109 Enable or disable use of Branch Likely instructions, regardless of
18110 the default for the selected architecture. By default, Branch
18111 Likely instructions may be generated if they are supported by the
18112 selected architecture. An exception is for the MIPS32 and MIPS64
18113 architectures and processors that implement those architectures;
18114 for those, Branch Likely instructions are not be generated by
18115 default because the MIPS32 and MIPS64 architectures specifically
18116 deprecate their use.
18117 -mcompact-branches=never
18119 -mcompact-branches=optimal
18120 -mcompact-branches=always
18121 These options control which form of branches will be generated.
18122 The default is -mcompact-branches=optimal.
18123 The -mcompact-branches=never option ensures that compact branch
18125 instructions will never be generated.
18126 The -mcompact-branches=always option ensures that a compact branch
18128 instruction will be generated if available. If a compact branch
18129 instruction is not available, a delay slot form of the branch will
18130 be used instead.
18131 This option is supported from MIPS Release 6 onwards.
18133 The -mcompact-branches=optimal option will cause a delay slot
18135 branch to be used if one is available in the current ISA and the
18136 delay slot is successfully filled. If the delay slot is not
18137 filled, a compact branch will be chosen if one is available.
18138 -mfp-exceptions
18140 -mno-fp-exceptions
18141 Specifies whether FP exceptions are enabled. This affects how FP
18142 instructions are scheduled for some processors. The default is
18143 that FP exceptions are enabled.
18144 For instance, on the SB-1, if FP exceptions are disabled, and we
18146 are emitting 64-bit code, then we can use both FP pipes.
18147 Otherwise, we can only use one FP pipe.
18148 -mvr4130-align
18150 -mno-vr4130-align
18151 The VR4130 pipeline is two-way superscalar, but can only issue two
18152 instructions together if the first one is 8-byte aligned. When
18153 this option is enabled, GCC aligns pairs of instructions that it
18154 thinks should execute in parallel.
18155 This option only has an effect when optimizing for the VR4130. It
18157 normally makes code faster, but at the expense of making it bigger.
18158 It is enabled by default at optimization level -O3.
18159 -msynci
18161 -mno-synci
18162 Enable (disable) generation of "synci" instructions on
18163 architectures that support it. The "synci" instructions (if
18164 enabled) are generated when "__builtin___clear_cache" is compiled.
18165 This option defaults to -mno-synci, but the default can be
18167 overridden by configuring GCC with --with-synci.
18168 When compiling code for single processor systems, it is generally
18170 safe to use "synci". However, on many multi-core (SMP) systems, it
18171 does not invalidate the instruction caches on all cores and may
18172 lead to undefined behavior.
18173 -mrelax-pic-calls
18175 -mno-relax-pic-calls
18176 Try to turn PIC calls that are normally dispatched via register $25
18177 into direct calls. This is only possible if the linker can resolve
18178 the destination at link time and if the destination is within range
18179 for a direct call.
18180 -mrelax-pic-calls is the default if GCC was configured to use an
18182 assembler and a linker that support the ".reloc" assembly directive
18183 and -mexplicit-relocs is in effect. With -mno-explicit-relocs,
18184 this optimization can be performed by the assembler and the linker
18185 alone without help from the compiler.
18186 -mmcount-ra-address
18188 -mno-mcount-ra-address
18189 Emit (do not emit) code that allows "_mcount" to modify the calling
18190 function's return address. When enabled, this option extends the
18191 usual "_mcount" interface with a new ra-address parameter, which
18192 has type "intptr_t *" and is passed in register $12. "_mcount" can
18193 then modify the return address by doing both of the following:
18194 * Returning the new address in register $31.
18196 * Storing the new address in "*ra-address", if ra-address is
18198 nonnull.
18199 The default is -mno-mcount-ra-address.
18201 -mframe-header-opt
18203 -mno-frame-header-opt
18204 Enable (disable) frame header optimization in the o32 ABI. When
18205 using the o32 ABI, calling functions will allocate 16 bytes on the
18206 stack for the called function to write out register arguments.
18207 When enabled, this optimization will suppress the allocation of the
18208 frame header if it can be determined that it is unused.
18209 This optimization is off by default at all optimization levels.
18211 -mlxc1-sxc1
18213 -mno-lxc1-sxc1
18214 When applicable, enable (disable) the generation of "lwxc1",
18215 "swxc1", "ldxc1", "sdxc1" instructions. Enabled by default.
18216 -mmadd4
18218 -mno-madd4
18219 When applicable, enable (disable) the generation of 4-operand
18220 "madd.s", "madd.d" and related instructions. Enabled by default.
18221 MMIX Options
18223 These options are defined for the MMIX:
18224 -mlibfuncs
18226 -mno-libfuncs
18227 Specify that intrinsic library functions are being compiled,
18228 passing all values in registers, no matter the size.
18229 -mepsilon
18231 -mno-epsilon
18232 Generate floating-point comparison instructions that compare with
18233 respect to the "rE" epsilon register.
18234 -mabi=mmixware
18236 -mabi=gnu
18237 Generate code that passes function parameters and return values
18238 that (in the called function) are seen as registers $0 and up, as
18239 opposed to the GNU ABI which uses global registers $231 and up.
18240 -mzero-extend
18242 -mno-zero-extend
18243 When reading data from memory in sizes shorter than 64 bits, use
18244 (do not use) zero-extending load instructions by default, rather
18245 than sign-extending ones.
18246 -mknuthdiv
18248 -mno-knuthdiv
18249 Make the result of a division yielding a remainder have the same
18250 sign as the divisor. With the default, -mno-knuthdiv, the sign of
18251 the remainder follows the sign of the dividend. Both methods are
18252 arithmetically valid, the latter being almost exclusively used.
18253 -mtoplevel-symbols
18255 -mno-toplevel-symbols
18256 Prepend (do not prepend) a : to all global symbols, so the assembly
18257 code can be used with the "PREFIX" assembly directive.
18258 -melf
18260 Generate an executable in the ELF format, rather than the default
18261 mmo format used by the mmix simulator.
18262 -mbranch-predict
18264 -mno-branch-predict
18265 Use (do not use) the probable-branch instructions, when static
18266 branch prediction indicates a probable branch.
18267 -mbase-addresses
18269 -mno-base-addresses
18270 Generate (do not generate) code that uses base addresses. Using a
18271 base address automatically generates a request (handled by the
18272 assembler and the linker) for a constant to be set up in a global
18273 register. The register is used for one or more base address
18274 requests within the range 0 to 255 from the value held in the
18275 register. The generally leads to short and fast code, but the
18276 number of different data items that can be addressed is limited.
18277 This means that a program that uses lots of static data may require
18278 -mno-base-addresses.
18279 -msingle-exit
18281 -mno-single-exit
18282 Force (do not force) generated code to have a single exit point in
18283 each function.
18284 MN10300 Options
18286 These -m options are defined for Matsushita MN10300 architectures:
18287 -mmult-bug
18289 Generate code to avoid bugs in the multiply instructions for the
18290 MN10300 processors. This is the default.
18291 -mno-mult-bug
18293 Do not generate code to avoid bugs in the multiply instructions for
18294 the MN10300 processors.
18295 -mam33
18297 Generate code using features specific to the AM33 processor.
18298 -mno-am33
18300 Do not generate code using features specific to the AM33 processor.
18301 This is the default.
18302 -mam33-2
18304 Generate code using features specific to the AM33/2.0 processor.
18305 -mam34
18307 Generate code using features specific to the AM34 processor.
18308 -mtune=cpu-type
18310 Use the timing characteristics of the indicated CPU type when
18311 scheduling instructions. This does not change the targeted
18312 processor type. The CPU type must be one of mn10300, am33, am33-2
18313 or am34.
18314 -mreturn-pointer-on-d0
18316 When generating a function that returns a pointer, return the
18317 pointer in both "a0" and "d0". Otherwise, the pointer is returned
18318 only in "a0", and attempts to call such functions without a
18319 prototype result in errors. Note that this option is on by
18320 default; use -mno-return-pointer-on-d0 to disable it.
18321 -mno-crt0
18323 Do not link in the C run-time initialization object file.
18324 -mrelax
18326 Indicate to the linker that it should perform a relaxation
18327 optimization pass to shorten branches, calls and absolute memory
18328 addresses. This option only has an effect when used on the command
18329 line for the final link step.
18330 This option makes symbolic debugging impossible.
18332 -mliw
18334 Allow the compiler to generate Long Instruction Word instructions
18335 if the target is the AM33 or later. This is the default. This
18336 option defines the preprocessor macro "__LIW__".
18337 -mnoliw
18339 Do not allow the compiler to generate Long Instruction Word
18340 instructions. This option defines the preprocessor macro
18341 "__NO_LIW__".
18342 -msetlb
18344 Allow the compiler to generate the SETLB and Lcc instructions if
18345 the target is the AM33 or later. This is the default. This option
18346 defines the preprocessor macro "__SETLB__".
18347 -mnosetlb
18349 Do not allow the compiler to generate SETLB or Lcc instructions.
18350 This option defines the preprocessor macro "__NO_SETLB__".
18351 Moxie Options
18353 -meb
18354 Generate big-endian code. This is the default for moxie-*-*
18355 configurations.
18356 -mel
18358 Generate little-endian code.
18359 -mmul.x
18361 Generate mul.x and umul.x instructions. This is the default for
18362 moxiebox-*-* configurations.
18363 -mno-crt0
18365 Do not link in the C run-time initialization object file.
18366 MSP430 Options
18368 These options are defined for the MSP430:
18369 -masm-hex
18371 Force assembly output to always use hex constants. Normally such
18372 constants are signed decimals, but this option is available for
18373 testsuite and/or aesthetic purposes.
18374 -mmcu=
18376 Select the MCU to target. This is used to create a C preprocessor
18377 symbol based upon the MCU name, converted to upper case and pre-
18378 and post-fixed with __. This in turn is used by the msp430.h
18379 header file to select an MCU-specific supplementary header file.
18380 The option also sets the ISA to use. If the MCU name is one that
18382 is known to only support the 430 ISA then that is selected,
18383 otherwise the 430X ISA is selected. A generic MCU name of msp430
18384 can also be used to select the 430 ISA. Similarly the generic
18385 msp430x MCU name selects the 430X ISA.
18386 In addition an MCU-specific linker script is added to the linker
18388 command line. The script's name is the name of the MCU with .ld
18389 appended. Thus specifying -mmcu=xxx on the gcc command line
18390 defines the C preprocessor symbol "__XXX__" and cause the linker to
18391 search for a script called xxx.ld.
18392 This option is also passed on to the assembler.
18394 -mwarn-mcu
18396 -mno-warn-mcu
18397 This option enables or disables warnings about conflicts between
18398 the MCU name specified by the -mmcu option and the ISA set by the
18399 -mcpu option and/or the hardware multiply support set by the
18400 -mhwmult option. It also toggles warnings about unrecognized MCU
18401 names. This option is on by default.
18402 -mcpu=
18404 Specifies the ISA to use. Accepted values are msp430, msp430x and
18405 msp430xv2. This option is deprecated. The -mmcu= option should be
18406 used to select the ISA.
18407 -msim
18409 Link to the simulator runtime libraries and linker script.
18410 Overrides any scripts that would be selected by the -mmcu= option.
18411 -mlarge
18413 Use large-model addressing (20-bit pointers, 32-bit "size_t").
18414 -msmall
18416 Use small-model addressing (16-bit pointers, 16-bit "size_t").
18417 -mrelax
18419 This option is passed to the assembler and linker, and allows the
18420 linker to perform certain optimizations that cannot be done until
18421 the final link.
18422 mhwmult=
18424 Describes the type of hardware multiply supported by the target.
18425 Accepted values are none for no hardware multiply, 16bit for the
18426 original 16-bit-only multiply supported by early MCUs. 32bit for
18427 the 16/32-bit multiply supported by later MCUs and f5series for the
18428 16/32-bit multiply supported by F5-series MCUs. A value of auto
18429 can also be given. This tells GCC to deduce the hardware multiply
18430 support based upon the MCU name provided by the -mmcu option. If
18431 no -mmcu option is specified or if the MCU name is not recognized
18432 then no hardware multiply support is assumed. "auto" is the
18433 default setting.
18434 Hardware multiplies are normally performed by calling a library
18436 routine. This saves space in the generated code. When compiling
18437 at -O3 or higher however the hardware multiplier is invoked inline.
18438 This makes for bigger, but faster code.
18439 The hardware multiply routines disable interrupts whilst running
18441 and restore the previous interrupt state when they finish. This
18442 makes them safe to use inside interrupt handlers as well as in
18443 normal code.
18444 -minrt
18446 Enable the use of a minimum runtime environment - no static
18447 initializers or constructors. This is intended for memory-
18448 constrained devices. The compiler includes special symbols in some
18449 objects that tell the linker and runtime which code fragments are
18450 required.
18451 -mcode-region=
18453 -mdata-region=
18454 These options tell the compiler where to place functions and data
18455 that do not have one of the "lower", "upper", "either" or "section"
18456 attributes. Possible values are "lower", "upper", "either" or
18457 "any". The first three behave like the corresponding attribute.
18458 The fourth possible value - "any" - is the default. It leaves
18459 placement entirely up to the linker script and how it assigns the
18460 standard sections (".text", ".data", etc) to the memory regions.
18461 -msilicon-errata=
18463 This option passes on a request to assembler to enable the fixes
18464 for the named silicon errata.
18465 -msilicon-errata-warn=
18467 This option passes on a request to the assembler to enable warning
18468 messages when a silicon errata might need to be applied.
18469 NDS32 Options
18471 These options are defined for NDS32 implementations:
18472 -mbig-endian
18474 Generate code in big-endian mode.
18475 -mlittle-endian
18477 Generate code in little-endian mode.
18478 -mreduced-regs
18480 Use reduced-set registers for register allocation.
18481 -mfull-regs
18483 Use full-set registers for register allocation.
18484 -mcmov
18486 Generate conditional move instructions.
18487 -mno-cmov
18489 Do not generate conditional move instructions.
18490 -mext-perf
18492 Generate performance extension instructions.
18493 -mno-ext-perf
18495 Do not generate performance extension instructions.
18496 -mext-perf2
18498 Generate performance extension 2 instructions.
18499 -mno-ext-perf2
18501 Do not generate performance extension 2 instructions.
18502 -mext-string
18504 Generate string extension instructions.
18505 -mno-ext-string
18507 Do not generate string extension instructions.
18508 -mv3push
18510 Generate v3 push25/pop25 instructions.
18511 -mno-v3push
18513 Do not generate v3 push25/pop25 instructions.
18514 -m16-bit
18516 Generate 16-bit instructions.
18517 -mno-16-bit
18519 Do not generate 16-bit instructions.
18520 -misr-vector-size=num
18522 Specify the size of each interrupt vector, which must be 4 or 16.
18523 -mcache-block-size=num
18525 Specify the size of each cache block, which must be a power of 2
18526 between 4 and 512.
18527 -march=arch
18529 Specify the name of the target architecture.
18530 -mcmodel=code-model
18532 Set the code model to one of
18533 small
18535 All the data and read-only data segments must be within 512KB
18536 addressing space. The text segment must be within 16MB
18537 addressing space.
18538 medium
18540 The data segment must be within 512KB while the read-only data
18541 segment can be within 4GB addressing space. The text segment
18542 should be still within 16MB addressing space.
18543 large
18545 All the text and data segments can be within 4GB addressing
18546 space.
18547 -mctor-dtor
18549 Enable constructor/destructor feature.
18550 -mrelax
18552 Guide linker to relax instructions.
18553 Nios II Options
18555 These are the options defined for the Altera Nios II processor.
18556 -G num
18558 Put global and static objects less than or equal to num bytes into
18559 the small data or BSS sections instead of the normal data or BSS
18560 sections. The default value of num is 8.
18561 -mgpopt=option
18563 -mgpopt
18564 -mno-gpopt
18565 Generate (do not generate) GP-relative accesses. The following
18566 option names are recognized:
18567 none
18569 Do not generate GP-relative accesses.
18570 local
18572 Generate GP-relative accesses for small data objects that are
18573 not external, weak, or uninitialized common symbols. Also use
18574 GP-relative addressing for objects that have been explicitly
18575 placed in a small data section via a "section" attribute.
18576 global
18578 As for local, but also generate GP-relative accesses for small
18579 data objects that are external, weak, or common. If you use
18580 this option, you must ensure that all parts of your program
18581 (including libraries) are compiled with the same -G setting.
18582 data
18584 Generate GP-relative accesses for all data objects in the
18585 program. If you use this option, the entire data and BSS
18586 segments of your program must fit in 64K of memory and you must
18587 use an appropriate linker script to allocate them within the
18588 addressable range of the global pointer.
18589 all Generate GP-relative addresses for function pointers as well as
18591 data pointers. If you use this option, the entire text, data,
18592 and BSS segments of your program must fit in 64K of memory and
18593 you must use an appropriate linker script to allocate them
18594 within the addressable range of the global pointer.
18595 -mgpopt is equivalent to -mgpopt=local, and -mno-gpopt is
18597 equivalent to -mgpopt=none.
18598 The default is -mgpopt except when -fpic or -fPIC is specified to
18600 generate position-independent code. Note that the Nios II ABI does
18601 not permit GP-relative accesses from shared libraries.
18602 You may need to specify -mno-gpopt explicitly when building
18604 programs that include large amounts of small data, including large
18605 GOT data sections. In this case, the 16-bit offset for GP-relative
18606 addressing may not be large enough to allow access to the entire
18607 small data section.
18608 -mgprel-sec=regexp
18610 This option specifies additional section names that can be accessed
18611 via GP-relative addressing. It is most useful in conjunction with
18612 "section" attributes on variable declarations and a custom linker
18613 script. The regexp is a POSIX Extended Regular Expression.
18614 This option does not affect the behavior of the -G option, and the
18616 specified sections are in addition to the standard ".sdata" and
18617 ".sbss" small-data sections that are recognized by -mgpopt.
18618 -mr0rel-sec=regexp
18620 This option specifies names of sections that can be accessed via a
18621 16-bit offset from "r0"; that is, in the low 32K or high 32K of the
18622 32-bit address space. It is most useful in conjunction with
18623 "section" attributes on variable declarations and a custom linker
18624 script. The regexp is a POSIX Extended Regular Expression.
18625 In contrast to the use of GP-relative addressing for small data,
18627 zero-based addressing is never generated by default and there are
18628 no conventional section names used in standard linker scripts for
18629 sections in the low or high areas of memory.
18630 -mel
18632 -meb
18633 Generate little-endian (default) or big-endian (experimental) code,
18634 respectively.
18635 -march=arch
18637 This specifies the name of the target Nios II architecture. GCC
18638 uses this name to determine what kind of instructions it can emit
18639 when generating assembly code. Permissible names are: r1, r2.
18640 The preprocessor macro "__nios2_arch__" is available to programs,
18642 with value 1 or 2, indicating the targeted ISA level.
18643 -mbypass-cache
18645 -mno-bypass-cache
18646 Force all load and store instructions to always bypass cache by
18647 using I/O variants of the instructions. The default is not to
18648 bypass the cache.
18649 -mno-cache-volatile
18651 -mcache-volatile
18652 Volatile memory access bypass the cache using the I/O variants of
18653 the load and store instructions. The default is not to bypass the
18654 cache.
18655 -mno-fast-sw-div
18657 -mfast-sw-div
18658 Do not use table-based fast divide for small numbers. The default
18659 is to use the fast divide at -O3 and above.
18660 -mno-hw-mul
18662 -mhw-mul
18663 -mno-hw-mulx
18664 -mhw-mulx
18665 -mno-hw-div
18666 -mhw-div
18667 Enable or disable emitting "mul", "mulx" and "div" family of
18668 instructions by the compiler. The default is to emit "mul" and not
18669 emit "div" and "mulx".
18670 -mbmx
18672 -mno-bmx
18673 -mcdx
18674 -mno-cdx
18675 Enable or disable generation of Nios II R2 BMX (bit manipulation)
18676 and CDX (code density) instructions. Enabling these instructions
18677 also requires -march=r2. Since these instructions are optional
18678 extensions to the R2 architecture, the default is not to emit them.
18679 -mcustom-insn=N
18681 -mno-custom-insn
18682 Each -mcustom-insn=N option enables use of a custom instruction
18683 with encoding N when generating code that uses insn. For example,
18684 -mcustom-fadds=253 generates custom instruction 253 for single-
18685 precision floating-point add operations instead of the default
18686 behavior of using a library call.
18687 The following values of insn are supported. Except as otherwise
18689 noted, floating-point operations are expected to be implemented
18690 with normal IEEE 754 semantics and correspond directly to the C
18691 operators or the equivalent GCC built-in functions.
18692 Single-precision floating point:
18694 fadds, fsubs, fdivs, fmuls
18696 Binary arithmetic operations.
18697 fnegs
18699 Unary negation.
18700 fabss
18702 Unary absolute value.
18703 fcmpeqs, fcmpges, fcmpgts, fcmples, fcmplts, fcmpnes
18705 Comparison operations.
18706 fmins, fmaxs
18708 Floating-point minimum and maximum. These instructions are
18709 only generated if -ffinite-math-only is specified.
18710 fsqrts
18712 Unary square root operation.
18713 fcoss, fsins, ftans, fatans, fexps, flogs
18715 Floating-point trigonometric and exponential functions. These
18716 instructions are only generated if -funsafe-math-optimizations
18717 is also specified.
18718 Double-precision floating point:
18720 faddd, fsubd, fdivd, fmuld
18722 Binary arithmetic operations.
18723 fnegd
18725 Unary negation.
18726 fabsd
18728 Unary absolute value.
18729 fcmpeqd, fcmpged, fcmpgtd, fcmpled, fcmpltd, fcmpned
18731 Comparison operations.
18732 fmind, fmaxd
18734 Double-precision minimum and maximum. These instructions are
18735 only generated if -ffinite-math-only is specified.
18736 fsqrtd
18738 Unary square root operation.
18739 fcosd, fsind, ftand, fatand, fexpd, flogd
18741 Double-precision trigonometric and exponential functions.
18742 These instructions are only generated if
18743 -funsafe-math-optimizations is also specified.
18744 Conversions:
18746 fextsd
18748 Conversion from single precision to double precision.
18749 ftruncds
18751 Conversion from double precision to single precision.
18752 fixsi, fixsu, fixdi, fixdu
18754 Conversion from floating point to signed or unsigned integer
18755 types, with truncation towards zero.
18756 round
18758 Conversion from single-precision floating point to signed
18759 integer, rounding to the nearest integer and ties away from
18760 zero. This corresponds to the "__builtin_lroundf" function
18761 when -fno-math-errno is used.
18762 floatis, floatus, floatid, floatud
18764 Conversion from signed or unsigned integer types to floating-
18765 point types.
18766 In addition, all of the following transfer instructions for
18768 internal registers X and Y must be provided to use any of the
18769 double-precision floating-point instructions. Custom instructions
18770 taking two double-precision source operands expect the first
18771 operand in the 64-bit register X. The other operand (or only
18772 operand of a unary operation) is given to the custom arithmetic
18773 instruction with the least significant half in source register src1
18774 and the most significant half in src2. A custom instruction that
18775 returns a double-precision result returns the most significant 32
18776 bits in the destination register and the other half in 32-bit
18777 register Y. GCC automatically generates the necessary code
18778 sequences to write register X and/or read register Y when double-
18779 precision floating-point instructions are used.
18780 fwrx
18782 Write src1 into the least significant half of X and src2 into
18783 the most significant half of X.
18784 fwry
18786 Write src1 into Y.
18787 frdxhi, frdxlo
18789 Read the most or least (respectively) significant half of X and
18790 store it in dest.
18791 frdy
18793 Read the value of Y and store it into dest.
18794 Note that you can gain more local control over generation of Nios
18796 II custom instructions by using the "target("custom-insn=N")" and
18797 "target("no-custom-insn")" function attributes or pragmas.
18798 -mcustom-fpu-cfg=name
18800 This option enables a predefined, named set of custom instruction
18801 encodings (see -mcustom-insn above). Currently, the following sets
18802 are defined:
18803 -mcustom-fpu-cfg=60-1 is equivalent to: -mcustom-fmuls=252
18805 -mcustom-fadds=253 -mcustom-fsubs=254 -fsingle-precision-constant
18806 -mcustom-fpu-cfg=60-2 is equivalent to: -mcustom-fmuls=252
18808 -mcustom-fadds=253 -mcustom-fsubs=254 -mcustom-fdivs=255
18809 -fsingle-precision-constant
18810 -mcustom-fpu-cfg=72-3 is equivalent to: -mcustom-floatus=243
18812 -mcustom-fixsi=244 -mcustom-floatis=245 -mcustom-fcmpgts=246
18813 -mcustom-fcmples=249 -mcustom-fcmpeqs=250 -mcustom-fcmpnes=251
18814 -mcustom-fmuls=252 -mcustom-fadds=253 -mcustom-fsubs=254
18815 -mcustom-fdivs=255 -fsingle-precision-constant
18816 Custom instruction assignments given by individual -mcustom-insn=
18818 options override those given by -mcustom-fpu-cfg=, regardless of
18819 the order of the options on the command line.
18820 Note that you can gain more local control over selection of a FPU
18822 configuration by using the "target("custom-fpu-cfg=name")" function
18823 attribute or pragma.
18824 These additional -m options are available for the Altera Nios II ELF
18826 (bare-metal) target:
18827 -mhal
18829 Link with HAL BSP. This suppresses linking with the GCC-provided C
18830 runtime startup and termination code, and is typically used in
18831 conjunction with -msys-crt0= to specify the location of the
18832 alternate startup code provided by the HAL BSP.
18833 -msmallc
18835 Link with a limited version of the C library, -lsmallc, rather than
18836 Newlib.
18837 -msys-crt0=startfile
18839 startfile is the file name of the startfile (crt0) to use when
18840 linking. This option is only useful in conjunction with -mhal.
18841 -msys-lib=systemlib
18843 systemlib is the library name of the library that provides low-
18844 level system calls required by the C library, e.g. "read" and
18845 "write". This option is typically used to link with a library
18846 provided by a HAL BSP.
18847 Nvidia PTX Options
18849 These options are defined for Nvidia PTX:
18850 -m32
18852 -m64
18853 Generate code for 32-bit or 64-bit ABI.
18854 -mmainkernel
18856 Link in code for a __main kernel. This is for stand-alone instead
18857 of offloading execution.
18858 -moptimize
18860 Apply partitioned execution optimizations. This is the default
18861 when any level of optimization is selected.
18862 -msoft-stack
18864 Generate code that does not use ".local" memory directly for stack
18865 storage. Instead, a per-warp stack pointer is maintained
18866 explicitly. This enables variable-length stack allocation (with
18867 variable-length arrays or "alloca"), and when global memory is used
18868 for underlying storage, makes it possible to access automatic
18869 variables from other threads, or with atomic instructions. This
18870 code generation variant is used for OpenMP offloading, but the
18871 option is exposed on its own for the purpose of testing the
18872 compiler; to generate code suitable for linking into programs using
18873 OpenMP offloading, use option -mgomp.
18874 -muniform-simt
18876 Switch to code generation variant that allows to execute all
18877 threads in each warp, while maintaining memory state and side
18878 effects as if only one thread in each warp was active outside of
18879 OpenMP SIMD regions. All atomic operations and calls to runtime
18880 (malloc, free, vprintf) are conditionally executed (iff current
18881 lane index equals the master lane index), and the register being
18882 assigned is copied via a shuffle instruction from the master lane.
18883 Outside of SIMD regions lane 0 is the master; inside, each thread
18884 sees itself as the master. Shared memory array "int __nvptx_uni[]"
18885 stores all-zeros or all-ones bitmasks for each warp, indicating
18886 current mode (0 outside of SIMD regions). Each thread can bitwise-
18887 and the bitmask at position "tid.y" with current lane index to
18888 compute the master lane index.
18889 -mgomp
18891 Generate code for use in OpenMP offloading: enables -msoft-stack
18892 and -muniform-simt options, and selects corresponding multilib
18893 variant.
18894 PDP-11 Options
18896 These options are defined for the PDP-11:
18897 -mfpu
18899 Use hardware FPP floating point. This is the default. (FIS
18900 floating point on the PDP-11/40 is not supported.)
18901 -msoft-float
18903 Do not use hardware floating point.
18904 -mac0
18906 Return floating-point results in ac0 (fr0 in Unix assembler
18907 syntax).
18908 -mno-ac0
18910 Return floating-point results in memory. This is the default.
18911 -m40
18913 Generate code for a PDP-11/40.
18914 -m45
18916 Generate code for a PDP-11/45. This is the default.
18917 -m10
18919 Generate code for a PDP-11/10.
18920 -mbcopy-builtin
18922 Use inline "movmemhi" patterns for copying memory. This is the
18923 default.
18924 -mbcopy
18926 Do not use inline "movmemhi" patterns for copying memory.
18927 -mint16
18929 -mno-int32
18930 Use 16-bit "int". This is the default.
18931 -mint32
18933 -mno-int16
18934 Use 32-bit "int".
18935 -mfloat64
18937 -mno-float32
18938 Use 64-bit "float". This is the default.
18939 -mfloat32
18941 -mno-float64
18942 Use 32-bit "float".
18943 -mabshi
18945 Use "abshi2" pattern. This is the default.
18946 -mno-abshi
18948 Do not use "abshi2" pattern.
18949 -mbranch-expensive
18951 Pretend that branches are expensive. This is for experimenting
18952 with code generation only.
18953 -mbranch-cheap
18955 Do not pretend that branches are expensive. This is the default.
18956 -munix-asm
18958 Use Unix assembler syntax. This is the default when configured for
18959 pdp11-*-bsd.
18960 -mdec-asm
18962 Use DEC assembler syntax. This is the default when configured for
18963 any PDP-11 target other than pdp11-*-bsd.
18964 picoChip Options
18966 These -m options are defined for picoChip implementations:
18967 -mae=ae_type
18969 Set the instruction set, register set, and instruction scheduling
18970 parameters for array element type ae_type. Supported values for
18971 ae_type are ANY, MUL, and MAC.
18972 -mae=ANY selects a completely generic AE type. Code generated with
18974 this option runs on any of the other AE types. The code is not as
18975 efficient as it would be if compiled for a specific AE type, and
18976 some types of operation (e.g., multiplication) do not work properly
18977 on all types of AE.
18978 -mae=MUL selects a MUL AE type. This is the most useful AE type
18980 for compiled code, and is the default.
18981 -mae=MAC selects a DSP-style MAC AE. Code compiled with this
18983 option may suffer from poor performance of byte (char)
18984 manipulation, since the DSP AE does not provide hardware support
18985 for byte load/stores.
18986 -msymbol-as-address
18988 Enable the compiler to directly use a symbol name as an address in
18989 a load/store instruction, without first loading it into a register.
18990 Typically, the use of this option generates larger programs, which
18991 run faster than when the option isn't used. However, the results
18992 vary from program to program, so it is left as a user option,
18993 rather than being permanently enabled.
18994 -mno-inefficient-warnings
18996 Disables warnings about the generation of inefficient code. These
18997 warnings can be generated, for example, when compiling code that
18998 performs byte-level memory operations on the MAC AE type. The MAC
18999 AE has no hardware support for byte-level memory operations, so all
19000 byte load/stores must be synthesized from word load/store
19001 operations. This is inefficient and a warning is generated to
19002 indicate that you should rewrite the code to avoid byte operations,
19003 or to target an AE type that has the necessary hardware support.
19004 This option disables these warnings.
19005 PowerPC Options
19007 These are listed under
19008 PowerPC SPE Options
19010 These -m options are defined for PowerPC SPE:
19011 -mmfcrf
19013 -mno-mfcrf
19014 -mpopcntb
19015 -mno-popcntb
19016 You use these options to specify which instructions are available
19017 on the processor you are using. The default value of these options
19018 is determined when configuring GCC. Specifying the -mcpu=cpu_type
19019 overrides the specification of these options. We recommend you use
19020 the -mcpu=cpu_type option rather than the options listed above.
19021 The -mmfcrf option allows GCC to generate the move from condition
19023 register field instruction implemented on the POWER4 processor and
19024 other processors that support the PowerPC V2.01 architecture. The
19025 -mpopcntb option allows GCC to generate the popcount and double-
19026 precision FP reciprocal estimate instruction implemented on the
19027 POWER5 processor and other processors that support the PowerPC
19028 V2.02 architecture.
19029 -mcpu=cpu_type
19031 Set architecture type, register usage, and instruction scheduling
19032 parameters for machine type cpu_type. Supported values for
19033 cpu_type are 8540, 8548, and native.
19034 -mcpu=powerpc specifies pure 32-bit PowerPC (either endian), with
19036 an appropriate, generic processor model assumed for scheduling
19037 purposes.
19038 Specifying native as cpu type detects and selects the architecture
19040 option that corresponds to the host processor of the system
19041 performing the compilation. -mcpu=native has no effect if GCC does
19042 not recognize the processor.
19043 The other options specify a specific processor. Code generated
19045 under those options runs best on that processor, and may not run at
19046 all on others.
19047 The -mcpu options automatically enable or disable the following
19049 options:
19050 -mhard-float -mmfcrf -mmultiple -mpopcntb -mpopcntd
19052 -msingle-float -mdouble-float -mfloat128
19053 The particular options set for any particular CPU varies between
19055 compiler versions, depending on what setting seems to produce
19056 optimal code for that CPU; it doesn't necessarily reflect the
19057 actual hardware's capabilities. If you wish to set an individual
19058 option to a particular value, you may specify it after the -mcpu
19059 option, like -mcpu=8548.
19060 -mtune=cpu_type
19062 Set the instruction scheduling parameters for machine type
19063 cpu_type, but do not set the architecture type or register usage,
19064 as -mcpu=cpu_type does. The same values for cpu_type are used for
19065 -mtune as for -mcpu. If both are specified, the code generated
19066 uses the architecture and registers set by -mcpu, but the
19067 scheduling parameters set by -mtune.
19068 -msecure-plt
19070 Generate code that allows ld and ld.so to build executables and
19071 shared libraries with non-executable ".plt" and ".got" sections.
19072 This is a PowerPC 32-bit SYSV ABI option.
19073 -mbss-plt
19075 Generate code that uses a BSS ".plt" section that ld.so fills in,
19076 and requires ".plt" and ".got" sections that are both writable and
19077 executable. This is a PowerPC 32-bit SYSV ABI option.
19078 -misel
19080 -mno-isel
19081 This switch enables or disables the generation of ISEL
19082 instructions.
19083 -misel=yes/no
19085 This switch has been deprecated. Use -misel and -mno-isel instead.
19086 -mspe
19088 -mno-spe
19089 This switch enables or disables the generation of SPE simd
19090 instructions.
19091 -mspe=yes/no
19093 This option has been deprecated. Use -mspe and -mno-spe instead.
19094 -mfloat128
19096 -mno-float128
19097 Enable/disable the __float128 keyword for IEEE 128-bit floating
19098 point and use either software emulation for IEEE 128-bit floating
19099 point or hardware instructions.
19100 -mfloat-gprs=yes/single/double/no
19102 -mfloat-gprs
19103 This switch enables or disables the generation of floating-point
19104 operations on the general-purpose registers for architectures that
19105 support it.
19106 The argument yes or single enables the use of single-precision
19108 floating-point operations.
19109 The argument double enables the use of single and double-precision
19111 floating-point operations.
19112 The argument no disables floating-point operations on the general-
19114 purpose registers.
19115 This option is currently only available on the MPC854x.
19117 -mfull-toc
19119 -mno-fp-in-toc
19120 -mno-sum-in-toc
19121 -mminimal-toc
19122 Modify generation of the TOC (Table Of Contents), which is created
19123 for every executable file. The -mfull-toc option is selected by
19124 default. In that case, GCC allocates at least one TOC entry for
19125 each unique non-automatic variable reference in your program. GCC
19126 also places floating-point constants in the TOC. However, only
19127 16,384 entries are available in the TOC.
19128 If you receive a linker error message that saying you have
19130 overflowed the available TOC space, you can reduce the amount of
19131 TOC space used with the -mno-fp-in-toc and -mno-sum-in-toc options.
19132 -mno-fp-in-toc prevents GCC from putting floating-point constants
19133 in the TOC and -mno-sum-in-toc forces GCC to generate code to
19134 calculate the sum of an address and a constant at run time instead
19135 of putting that sum into the TOC. You may specify one or both of
19136 these options. Each causes GCC to produce very slightly slower and
19137 larger code at the expense of conserving TOC space.
19138 If you still run out of space in the TOC even when you specify both
19140 of these options, specify -mminimal-toc instead. This option
19141 causes GCC to make only one TOC entry for every file. When you
19142 specify this option, GCC produces code that is slower and larger
19143 but which uses extremely little TOC space. You may wish to use
19144 this option only on files that contain less frequently-executed
19145 code.
19146 -maix32
19148 Disables the 64-bit ABI. GCC defaults to -maix32.
19149 -mxl-compat
19151 -mno-xl-compat
19152 Produce code that conforms more closely to IBM XL compiler
19153 semantics when using AIX-compatible ABI. Pass floating-point
19154 arguments to prototyped functions beyond the register save area
19155 (RSA) on the stack in addition to argument FPRs. Do not assume
19156 that most significant double in 128-bit long double value is
19157 properly rounded when comparing values and converting to double.
19158 Use XL symbol names for long double support routines.
19159 The AIX calling convention was extended but not initially
19161 documented to handle an obscure K&R C case of calling a function
19162 that takes the address of its arguments with fewer arguments than
19163 declared. IBM XL compilers access floating-point arguments that do
19164 not fit in the RSA from the stack when a subroutine is compiled
19165 without optimization. Because always storing floating-point
19166 arguments on the stack is inefficient and rarely needed, this
19167 option is not enabled by default and only is necessary when calling
19168 subroutines compiled by IBM XL compilers without optimization.
19169 -malign-natural
19171 -malign-power
19172 On AIX, 32-bit Darwin, and 64-bit PowerPC GNU/Linux, the option
19173 -malign-natural overrides the ABI-defined alignment of larger
19174 types, such as floating-point doubles, on their natural size-based
19175 boundary. The option -malign-power instructs GCC to follow the
19176 ABI-specified alignment rules. GCC defaults to the standard
19177 alignment defined in the ABI.
19178 On 64-bit Darwin, natural alignment is the default, and
19180 -malign-power is not supported.
19181 -msoft-float
19183 -mhard-float
19184 Generate code that does not use (uses) the floating-point register
19185 set. Software floating-point emulation is provided if you use the
19186 -msoft-float option, and pass the option to GCC when linking.
19187 -msingle-float
19189 -mdouble-float
19190 Generate code for single- or double-precision floating-point
19191 operations. -mdouble-float implies -msingle-float.
19192 -mmultiple
19194 -mno-multiple
19195 Generate code that uses (does not use) the load multiple word
19196 instructions and the store multiple word instructions. These
19197 instructions are generated by default on POWER systems, and not
19198 generated on PowerPC systems. Do not use -mmultiple on little-
19199 endian PowerPC systems, since those instructions do not work when
19200 the processor is in little-endian mode. The exceptions are PPC740
19201 and PPC750 which permit these instructions in little-endian mode.
19202 -mupdate
19204 -mno-update
19205 Generate code that uses (does not use) the load or store
19206 instructions that update the base register to the address of the
19207 calculated memory location. These instructions are generated by
19208 default. If you use -mno-update, there is a small window between
19209 the time that the stack pointer is updated and the address of the
19210 previous frame is stored, which means code that walks the stack
19211 frame across interrupts or signals may get corrupted data.
19212 -mavoid-indexed-addresses
19214 -mno-avoid-indexed-addresses
19215 Generate code that tries to avoid (not avoid) the use of indexed
19216 load or store instructions. These instructions can incur a
19217 performance penalty on Power6 processors in certain situations,
19218 such as when stepping through large arrays that cross a 16M
19219 boundary. This option is enabled by default when targeting Power6
19220 and disabled otherwise.
19221 -mfused-madd
19223 -mno-fused-madd
19224 Generate code that uses (does not use) the floating-point multiply
19225 and accumulate instructions. These instructions are generated by
19226 default if hardware floating point is used. The machine-dependent
19227 -mfused-madd option is now mapped to the machine-independent
19228 -ffp-contract=fast option, and -mno-fused-madd is mapped to
19229 -ffp-contract=off.
19230 -mno-strict-align
19232 -mstrict-align
19233 On System V.4 and embedded PowerPC systems do not (do) assume that
19234 unaligned memory references are handled by the system.
19235 -mrelocatable
19237 -mno-relocatable
19238 Generate code that allows (does not allow) a static executable to
19239 be relocated to a different address at run time. A simple embedded
19240 PowerPC system loader should relocate the entire contents of
19241 ".got2" and 4-byte locations listed in the ".fixup" section, a
19242 table of 32-bit addresses generated by this option. For this to
19243 work, all objects linked together must be compiled with
19244 -mrelocatable or -mrelocatable-lib. -mrelocatable code aligns the
19245 stack to an 8-byte boundary.
19246 -mrelocatable-lib
19248 -mno-relocatable-lib
19249 Like -mrelocatable, -mrelocatable-lib generates a ".fixup" section
19250 to allow static executables to be relocated at run time, but
19251 -mrelocatable-lib does not use the smaller stack alignment of
19252 -mrelocatable. Objects compiled with -mrelocatable-lib may be
19253 linked with objects compiled with any combination of the
19254 -mrelocatable options.
19255 -mno-toc
19257 -mtoc
19258 On System V.4 and embedded PowerPC systems do not (do) assume that
19259 register 2 contains a pointer to a global area pointing to the
19260 addresses used in the program.
19261 -mlittle
19263 -mlittle-endian
19264 On System V.4 and embedded PowerPC systems compile code for the
19265 processor in little-endian mode. The -mlittle-endian option is the
19266 same as -mlittle.
19267 -mbig
19269 -mbig-endian
19270 On System V.4 and embedded PowerPC systems compile code for the
19271 processor in big-endian mode. The -mbig-endian option is the same
19272 as -mbig.
19273 -mdynamic-no-pic
19275 On Darwin and Mac OS X systems, compile code so that it is not
19276 relocatable, but that its external references are relocatable. The
19277 resulting code is suitable for applications, but not shared
19278 libraries.
19279 -msingle-pic-base
19281 Treat the register used for PIC addressing as read-only, rather
19282 than loading it in the prologue for each function. The runtime
19283 system is responsible for initializing this register with an
19284 appropriate value before execution begins.
19285 -mprioritize-restricted-insns=priority
19287 This option controls the priority that is assigned to dispatch-slot
19288 restricted instructions during the second scheduling pass. The
19289 argument priority takes the value 0, 1, or 2 to assign no, highest,
19290 or second-highest (respectively) priority to dispatch-slot
19291 restricted instructions.
19292 -msched-costly-dep=dependence_type
19294 This option controls which dependences are considered costly by the
19295 target during instruction scheduling. The argument dependence_type
19296 takes one of the following values:
19297 no No dependence is costly.
19299 all All dependences are costly.
19301 true_store_to_load
19303 A true dependence from store to load is costly.
19304 store_to_load
19306 Any dependence from store to load is costly.
19307 number
19309 Any dependence for which the latency is greater than or equal
19310 to number is costly.
19311 -minsert-sched-nops=scheme
19313 This option controls which NOP insertion scheme is used during the
19314 second scheduling pass. The argument scheme takes one of the
19315 following values:
19316 no Don't insert NOPs.
19318 pad Pad with NOPs any dispatch group that has vacant issue slots,
19320 according to the scheduler's grouping.
19321 regroup_exact
19323 Insert NOPs to force costly dependent insns into separate
19324 groups. Insert exactly as many NOPs as needed to force an insn
19325 to a new group, according to the estimated processor grouping.
19326 number
19328 Insert NOPs to force costly dependent insns into separate
19329 groups. Insert number NOPs to force an insn to a new group.
19330 -mcall-sysv
19332 On System V.4 and embedded PowerPC systems compile code using
19333 calling conventions that adhere to the March 1995 draft of the
19334 System V Application Binary Interface, PowerPC processor
19335 supplement. This is the default unless you configured GCC using
19336 powerpc-*-eabiaix.
19337 -mcall-sysv-eabi
19339 -mcall-eabi
19340 Specify both -mcall-sysv and -meabi options.
19341 -mcall-sysv-noeabi
19343 Specify both -mcall-sysv and -mno-eabi options.
19344 -mcall-aixdesc
19346 On System V.4 and embedded PowerPC systems compile code for the AIX
19347 operating system.
19348 -mcall-linux
19350 On System V.4 and embedded PowerPC systems compile code for the
19351 Linux-based GNU system.
19352 -mcall-freebsd
19354 On System V.4 and embedded PowerPC systems compile code for the
19355 FreeBSD operating system.
19356 -mcall-netbsd
19358 On System V.4 and embedded PowerPC systems compile code for the
19359 NetBSD operating system.
19360 -mcall-openbsd
19362 On System V.4 and embedded PowerPC systems compile code for the
19363 OpenBSD operating system.
19364 -maix-struct-return
19366 Return all structures in memory (as specified by the AIX ABI).
19367 -msvr4-struct-return
19369 Return structures smaller than 8 bytes in registers (as specified
19370 by the SVR4 ABI).
19371 -mabi=abi-type
19373 Extend the current ABI with a particular extension, or remove such
19374 extension. Valid values are altivec, no-altivec, spe, no-spe,
19375 ibmlongdouble, ieeelongdouble, elfv1, elfv2.
19376 -mabi=spe
19378 Extend the current ABI with SPE ABI extensions. This does not
19379 change the default ABI, instead it adds the SPE ABI extensions to
19380 the current ABI.
19381 -mabi=no-spe
19383 Disable Book-E SPE ABI extensions for the current ABI.
19384 -mabi=ibmlongdouble
19386 Change the current ABI to use IBM extended-precision long double.
19387 This is not likely to work if your system defaults to using IEEE
19388 extended-precision long double. If you change the long double type
19389 from IEEE extended-precision, the compiler will issue a warning
19390 unless you use the -Wno-psabi option.
19391 -mabi=ieeelongdouble
19393 Change the current ABI to use IEEE extended-precision long double.
19394 This is not likely to work if your system defaults to using IBM
19395 extended-precision long double. If you change the long double type
19396 from IBM extended-precision, the compiler will issue a warning
19397 unless you use the -Wno-psabi option.
19398 -mabi=elfv1
19400 Change the current ABI to use the ELFv1 ABI. This is the default
19401 ABI for big-endian PowerPC 64-bit Linux. Overriding the default
19402 ABI requires special system support and is likely to fail in
19403 spectacular ways.
19404 -mabi=elfv2
19406 Change the current ABI to use the ELFv2 ABI. This is the default
19407 ABI for little-endian PowerPC 64-bit Linux. Overriding the default
19408 ABI requires special system support and is likely to fail in
19409 spectacular ways.
19410 -mgnu-attribute
19412 -mno-gnu-attribute
19413 Emit .gnu_attribute assembly directives to set tag/value pairs in a
19414 .gnu.attributes section that specify ABI variations in function
19415 parameters or return values.
19416 -mprototype
19418 -mno-prototype
19419 On System V.4 and embedded PowerPC systems assume that all calls to
19420 variable argument functions are properly prototyped. Otherwise,
19421 the compiler must insert an instruction before every non-prototyped
19422 call to set or clear bit 6 of the condition code register ("CR") to
19423 indicate whether floating-point values are passed in the floating-
19424 point registers in case the function takes variable arguments.
19425 With -mprototype, only calls to prototyped variable argument
19426 functions set or clear the bit.
19427 -msim
19429 On embedded PowerPC systems, assume that the startup module is
19430 called sim-crt0.o and that the standard C libraries are libsim.a
19431 and libc.a. This is the default for powerpc-*-eabisim
19432 configurations.
19433 -mmvme
19435 On embedded PowerPC systems, assume that the startup module is
19436 called crt0.o and the standard C libraries are libmvme.a and
19437 libc.a.
19438 -mads
19440 On embedded PowerPC systems, assume that the startup module is
19441 called crt0.o and the standard C libraries are libads.a and libc.a.
19442 -myellowknife
19444 On embedded PowerPC systems, assume that the startup module is
19445 called crt0.o and the standard C libraries are libyk.a and libc.a.
19446 -mvxworks
19448 On System V.4 and embedded PowerPC systems, specify that you are
19449 compiling for a VxWorks system.
19450 -memb
19452 On embedded PowerPC systems, set the "PPC_EMB" bit in the ELF flags
19453 header to indicate that eabi extended relocations are used.
19454 -meabi
19456 -mno-eabi
19457 On System V.4 and embedded PowerPC systems do (do not) adhere to
19458 the Embedded Applications Binary Interface (EABI), which is a set
19459 of modifications to the System V.4 specifications. Selecting
19460 -meabi means that the stack is aligned to an 8-byte boundary, a
19461 function "__eabi" is called from "main" to set up the EABI
19462 environment, and the -msdata option can use both "r2" and "r13" to
19463 point to two separate small data areas. Selecting -mno-eabi means
19464 that the stack is aligned to a 16-byte boundary, no EABI
19465 initialization function is called from "main", and the -msdata
19466 option only uses "r13" to point to a single small data area. The
19467 -meabi option is on by default if you configured GCC using one of
19468 the powerpc*-*-eabi* options.
19469 -msdata=eabi
19471 On System V.4 and embedded PowerPC systems, put small initialized
19472 "const" global and static data in the ".sdata2" section, which is
19473 pointed to by register "r2". Put small initialized non-"const"
19474 global and static data in the ".sdata" section, which is pointed to
19475 by register "r13". Put small uninitialized global and static data
19476 in the ".sbss" section, which is adjacent to the ".sdata" section.
19477 The -msdata=eabi option is incompatible with the -mrelocatable
19478 option. The -msdata=eabi option also sets the -memb option.
19479 -msdata=sysv
19481 On System V.4 and embedded PowerPC systems, put small global and
19482 static data in the ".sdata" section, which is pointed to by
19483 register "r13". Put small uninitialized global and static data in
19484 the ".sbss" section, which is adjacent to the ".sdata" section.
19485 The -msdata=sysv option is incompatible with the -mrelocatable
19486 option.
19487 -msdata=default
19489 -msdata
19490 On System V.4 and embedded PowerPC systems, if -meabi is used,
19491 compile code the same as -msdata=eabi, otherwise compile code the
19492 same as -msdata=sysv.
19493 -msdata=data
19495 On System V.4 and embedded PowerPC systems, put small global data
19496 in the ".sdata" section. Put small uninitialized global data in
19497 the ".sbss" section. Do not use register "r13" to address small
19498 data however. This is the default behavior unless other -msdata
19499 options are used.
19500 -msdata=none
19502 -mno-sdata
19503 On embedded PowerPC systems, put all initialized global and static
19504 data in the ".data" section, and all uninitialized data in the
19505 ".bss" section.
19506 -mblock-move-inline-limit=num
19508 Inline all block moves (such as calls to "memcpy" or structure
19509 copies) less than or equal to num bytes. The minimum value for num
19510 is 32 bytes on 32-bit targets and 64 bytes on 64-bit targets. The
19511 default value is target-specific.
19512 -G num
19514 On embedded PowerPC systems, put global and static items less than
19515 or equal to num bytes into the small data or BSS sections instead
19516 of the normal data or BSS section. By default, num is 8. The -G
19517 num switch is also passed to the linker. All modules should be
19518 compiled with the same -G num value.
19519 -mregnames
19521 -mno-regnames
19522 On System V.4 and embedded PowerPC systems do (do not) emit
19523 register names in the assembly language output using symbolic
19524 forms.
19525 -mlongcall
19527 -mno-longcall
19528 By default assume that all calls are far away so that a longer and
19529 more expensive calling sequence is required. This is required for
19530 calls farther than 32 megabytes (33,554,432 bytes) from the current
19531 location. A short call is generated if the compiler knows the call
19532 cannot be that far away. This setting can be overridden by the
19533 "shortcall" function attribute, or by "#pragma longcall(0)".
19534 Some linkers are capable of detecting out-of-range calls and
19536 generating glue code on the fly. On these systems, long calls are
19537 unnecessary and generate slower code. As of this writing, the AIX
19538 linker can do this, as can the GNU linker for PowerPC/64. It is
19539 planned to add this feature to the GNU linker for 32-bit PowerPC
19540 systems as well.
19541 In the future, GCC may ignore all longcall specifications when the
19543 linker is known to generate glue.
19544 -mtls-markers
19546 -mno-tls-markers
19547 Mark (do not mark) calls to "__tls_get_addr" with a relocation
19548 specifying the function argument. The relocation allows the linker
19549 to reliably associate function call with argument setup
19550 instructions for TLS optimization, which in turn allows GCC to
19551 better schedule the sequence.
19552 -mrecip
19554 -mno-recip
19555 This option enables use of the reciprocal estimate and reciprocal
19556 square root estimate instructions with additional Newton-Raphson
19557 steps to increase precision instead of doing a divide or square
19558 root and divide for floating-point arguments. You should use the
19559 -ffast-math option when using -mrecip (or at least
19560 -funsafe-math-optimizations, -ffinite-math-only, -freciprocal-math
19561 and -fno-trapping-math). Note that while the throughput of the
19562 sequence is generally higher than the throughput of the non-
19563 reciprocal instruction, the precision of the sequence can be
19564 decreased by up to 2 ulp (i.e. the inverse of 1.0 equals
19565 0.99999994) for reciprocal square roots.
19566 -mrecip=opt
19568 This option controls which reciprocal estimate instructions may be
19569 used. opt is a comma-separated list of options, which may be
19570 preceded by a "!" to invert the option:
19571 all Enable all estimate instructions.
19573 default
19575 Enable the default instructions, equivalent to -mrecip.
19576 none
19578 Disable all estimate instructions, equivalent to -mno-recip.
19579 div Enable the reciprocal approximation instructions for both
19581 single and double precision.
19582 divf
19584 Enable the single-precision reciprocal approximation
19585 instructions.
19586 divd
19588 Enable the double-precision reciprocal approximation
19589 instructions.
19590 rsqrt
19592 Enable the reciprocal square root approximation instructions
19593 for both single and double precision.
19594 rsqrtf
19596 Enable the single-precision reciprocal square root
19597 approximation instructions.
19598 rsqrtd
19600 Enable the double-precision reciprocal square root
19601 approximation instructions.
19602 So, for example, -mrecip=all,!rsqrtd enables all of the reciprocal
19604 estimate instructions, except for the "FRSQRTE", "XSRSQRTEDP", and
19605 "XVRSQRTEDP" instructions which handle the double-precision
19606 reciprocal square root calculations.
19607 -mrecip-precision
19609 -mno-recip-precision
19610 Assume (do not assume) that the reciprocal estimate instructions
19611 provide higher-precision estimates than is mandated by the PowerPC
19612 ABI. Selecting -mcpu=power6, -mcpu=power7 or -mcpu=power8
19613 automatically selects -mrecip-precision. The double-precision
19614 square root estimate instructions are not generated by default on
19615 low-precision machines, since they do not provide an estimate that
19616 converges after three steps.
19617 -mpointers-to-nested-functions
19619 -mno-pointers-to-nested-functions
19620 Generate (do not generate) code to load up the static chain
19621 register ("r11") when calling through a pointer on AIX and 64-bit
19622 Linux systems where a function pointer points to a 3-word
19623 descriptor giving the function address, TOC value to be loaded in
19624 register "r2", and static chain value to be loaded in register
19625 "r11". The -mpointers-to-nested-functions is on by default. You
19626 cannot call through pointers to nested functions or pointers to
19627 functions compiled in other languages that use the static chain if
19628 you use -mno-pointers-to-nested-functions.
19629 -msave-toc-indirect
19631 -mno-save-toc-indirect
19632 Generate (do not generate) code to save the TOC value in the
19633 reserved stack location in the function prologue if the function
19634 calls through a pointer on AIX and 64-bit Linux systems. If the
19635 TOC value is not saved in the prologue, it is saved just before the
19636 call through the pointer. The -mno-save-toc-indirect option is the
19637 default.
19638 -mcompat-align-parm
19640 -mno-compat-align-parm
19641 Generate (do not generate) code to pass structure parameters with a
19642 maximum alignment of 64 bits, for compatibility with older versions
19643 of GCC.
19644 Older versions of GCC (prior to 4.9.0) incorrectly did not align a
19646 structure parameter on a 128-bit boundary when that structure
19647 contained a member requiring 128-bit alignment. This is corrected
19648 in more recent versions of GCC. This option may be used to
19649 generate code that is compatible with functions compiled with older
19650 versions of GCC.
19651 The -mno-compat-align-parm option is the default.
19653 -mstack-protector-guard=guard
19655 -mstack-protector-guard-reg=reg
19656 -mstack-protector-guard-offset=offset
19657 -mstack-protector-guard-symbol=symbol
19658 Generate stack protection code using canary at guard. Supported
19659 locations are global for global canary or tls for per-thread canary
19660 in the TLS block (the default with GNU libc version 2.4 or later).
19661 With the latter choice the options -mstack-protector-guard-reg=reg
19663 and -mstack-protector-guard-offset=offset furthermore specify which
19664 register to use as base register for reading the canary, and from
19665 what offset from that base register. The default for those is as
19666 specified in the relevant ABI.
19667 -mstack-protector-guard-symbol=symbol overrides the offset with a
19668 symbol reference to a canary in the TLS block.
19669 RISC-V Options
19671 These command-line options are defined for RISC-V targets:
19672 -mbranch-cost=n
19674 Set the cost of branches to roughly n instructions.
19675 -mplt
19677 -mno-plt
19678 When generating PIC code, do or don't allow the use of PLTs.
19679 Ignored for non-PIC. The default is -mplt.
19680 -mabi=ABI-string
19682 Specify integer and floating-point calling convention. ABI-string
19683 contains two parts: the size of integer types and the registers
19684 used for floating-point types. For example -march=rv64ifd
19685 -mabi=lp64d means that long and pointers are 64-bit (implicitly
19686 defining int to be 32-bit), and that floating-point values up to 64
19687 bits wide are passed in F registers. Contrast this with
19688 -march=rv64ifd -mabi=lp64f, which still allows the compiler to
19689 generate code that uses the F and D extensions but only allows
19690 floating-point values up to 32 bits long to be passed in registers;
19691 or -march=rv64ifd -mabi=lp64, in which no floating-point arguments
19692 will be passed in registers.
19693 The default for this argument is system dependent, users who want a
19695 specific calling convention should specify one explicitly. The
19696 valid calling conventions are: ilp32, ilp32f, ilp32d, lp64, lp64f,
19697 and lp64d. Some calling conventions are impossible to implement on
19698 some ISAs: for example, -march=rv32if -mabi=ilp32d is invalid
19699 because the ABI requires 64-bit values be passed in F registers,
19700 but F registers are only 32 bits wide.
19701 -mfdiv
19703 -mno-fdiv
19704 Do or don't use hardware floating-point divide and square root
19705 instructions. This requires the F or D extensions for floating-
19706 point registers. The default is to use them if the specified
19707 architecture has these instructions.
19708 -mdiv
19710 -mno-div
19711 Do or don't use hardware instructions for integer division. This
19712 requires the M extension. The default is to use them if the
19713 specified architecture has these instructions.
19714 -march=ISA-string
19716 Generate code for given RISC-V ISA (e.g. rv64im). ISA strings must
19717 be lower-case. Examples include rv64i, rv32g, and rv32imaf.
19718 -mtune=processor-string
19720 Optimize the output for the given processor, specified by
19721 microarchitecture name.
19722 -mpreferred-stack-boundary=num
19724 Attempt to keep the stack boundary aligned to a 2 raised to num
19725 byte boundary. If -mpreferred-stack-boundary is not specified, the
19726 default is 4 (16 bytes or 128-bits).
19727 Warning: If you use this switch, then you must build all modules
19729 with the same value, including any libraries. This includes the
19730 system libraries and startup modules.
19731 -msmall-data-limit=n
19733 Put global and static data smaller than n bytes into a special
19734 section (on some targets).
19735 -msave-restore
19737 -mno-save-restore
19738 Do or don't use smaller but slower prologue and epilogue code that
19739 uses library function calls. The default is to use fast inline
19740 prologues and epilogues.
19741 -mstrict-align
19743 -mno-strict-align
19744 Do not or do generate unaligned memory accesses. The default is
19745 set depending on whether the processor we are optimizing for
19746 supports fast unaligned access or not.
19747 -mcmodel=medlow
19749 Generate code for the medium-low code model. The program and its
19750 statically defined symbols must lie within a single 2 GiB address
19751 range and must lie between absolute addresses -2 GiB and +2 GiB.
19752 Programs can be statically or dynamically linked. This is the
19753 default code model.
19754 -mcmodel=medany
19756 Generate code for the medium-any code model. The program and its
19757 statically defined symbols must be within any single 2 GiB address
19758 range. Programs can be statically or dynamically linked.
19759 -mexplicit-relocs
19761 -mno-exlicit-relocs
19762 Use or do not use assembler relocation operators when dealing with
19763 symbolic addresses. The alternative is to use assembler macros
19764 instead, which may limit optimization.
19765 -mrelax
19767 -mno-relax
19768 Take advantage of linker relaxations to reduce the number of
19769 instructions required to materialize symbol addresses. The default
19770 is to take advantage of linker relaxations.
19771 RL78 Options
19773 -msim
19774 Links in additional target libraries to support operation within a
19775 simulator.
19776 -mmul=none
19778 -mmul=g10
19779 -mmul=g13
19780 -mmul=g14
19781 -mmul=rl78
19782 Specifies the type of hardware multiplication and division support
19783 to be used. The simplest is "none", which uses software for both
19784 multiplication and division. This is the default. The "g13" value
19785 is for the hardware multiply/divide peripheral found on the
19786 RL78/G13 (S2 core) targets. The "g14" value selects the use of the
19787 multiplication and division instructions supported by the RL78/G14
19788 (S3 core) parts. The value "rl78" is an alias for "g14" and the
19789 value "mg10" is an alias for "none".
19790 In addition a C preprocessor macro is defined, based upon the
19792 setting of this option. Possible values are: "__RL78_MUL_NONE__",
19793 "__RL78_MUL_G13__" or "__RL78_MUL_G14__".
19794 -mcpu=g10
19796 -mcpu=g13
19797 -mcpu=g14
19798 -mcpu=rl78
19799 Specifies the RL78 core to target. The default is the G14 core,
19800 also known as an S3 core or just RL78. The G13 or S2 core does not
19801 have multiply or divide instructions, instead it uses a hardware
19802 peripheral for these operations. The G10 or S1 core does not have
19803 register banks, so it uses a different calling convention.
19804 If this option is set it also selects the type of hardware multiply
19806 support to use, unless this is overridden by an explicit -mmul=none
19807 option on the command line. Thus specifying -mcpu=g13 enables the
19808 use of the G13 hardware multiply peripheral and specifying
19809 -mcpu=g10 disables the use of hardware multiplications altogether.
19810 Note, although the RL78/G14 core is the default target, specifying
19812 -mcpu=g14 or -mcpu=rl78 on the command line does change the
19813 behavior of the toolchain since it also enables G14 hardware
19814 multiply support. If these options are not specified on the
19815 command line then software multiplication routines will be used
19816 even though the code targets the RL78 core. This is for backwards
19817 compatibility with older toolchains which did not have hardware
19818 multiply and divide support.
19819 In addition a C preprocessor macro is defined, based upon the
19821 setting of this option. Possible values are: "__RL78_G10__",
19822 "__RL78_G13__" or "__RL78_G14__".
19823 -mg10
19825 -mg13
19826 -mg14
19827 -mrl78
19828 These are aliases for the corresponding -mcpu= option. They are
19829 provided for backwards compatibility.
19830 -mallregs
19832 Allow the compiler to use all of the available registers. By
19833 default registers "r24..r31" are reserved for use in interrupt
19834 handlers. With this option enabled these registers can be used in
19835 ordinary functions as well.
19836 -m64bit-doubles
19838 -m32bit-doubles
19839 Make the "double" data type be 64 bits (-m64bit-doubles) or 32 bits
19840 (-m32bit-doubles) in size. The default is -m32bit-doubles.
19841 -msave-mduc-in-interrupts
19843 -mno-save-mduc-in-interrupts
19844 Specifies that interrupt handler functions should preserve the MDUC
19845 registers. This is only necessary if normal code might use the
19846 MDUC registers, for example because it performs multiplication and
19847 division operations. The default is to ignore the MDUC registers
19848 as this makes the interrupt handlers faster. The target option
19849 -mg13 needs to be passed for this to work as this feature is only
19850 available on the G13 target (S2 core). The MDUC registers will
19851 only be saved if the interrupt handler performs a multiplication or
19852 division operation or it calls another function.
19853 IBM RS/6000 and PowerPC Options
19855 These -m options are defined for the IBM RS/6000 and PowerPC:
19856 -mpowerpc-gpopt
19858 -mno-powerpc-gpopt
19859 -mpowerpc-gfxopt
19860 -mno-powerpc-gfxopt
19861 -mpowerpc64
19862 -mno-powerpc64
19863 -mmfcrf
19864 -mno-mfcrf
19865 -mpopcntb
19866 -mno-popcntb
19867 -mpopcntd
19868 -mno-popcntd
19869 -mfprnd
19870 -mno-fprnd
19871 -mcmpb
19872 -mno-cmpb
19873 -mmfpgpr
19874 -mno-mfpgpr
19875 -mhard-dfp
19876 -mno-hard-dfp
19877 You use these options to specify which instructions are available
19878 on the processor you are using. The default value of these options
19879 is determined when configuring GCC. Specifying the -mcpu=cpu_type
19880 overrides the specification of these options. We recommend you use
19881 the -mcpu=cpu_type option rather than the options listed above.
19882 Specifying -mpowerpc-gpopt allows GCC to use the optional PowerPC
19884 architecture instructions in the General Purpose group, including
19885 floating-point square root. Specifying -mpowerpc-gfxopt allows GCC
19886 to use the optional PowerPC architecture instructions in the
19887 Graphics group, including floating-point select.
19888 The -mmfcrf option allows GCC to generate the move from condition
19890 register field instruction implemented on the POWER4 processor and
19891 other processors that support the PowerPC V2.01 architecture. The
19892 -mpopcntb option allows GCC to generate the popcount and double-
19893 precision FP reciprocal estimate instruction implemented on the
19894 POWER5 processor and other processors that support the PowerPC
19895 V2.02 architecture. The -mpopcntd option allows GCC to generate
19896 the popcount instruction implemented on the POWER7 processor and
19897 other processors that support the PowerPC V2.06 architecture. The
19898 -mfprnd option allows GCC to generate the FP round to integer
19899 instructions implemented on the POWER5+ processor and other
19900 processors that support the PowerPC V2.03 architecture. The -mcmpb
19901 option allows GCC to generate the compare bytes instruction
19902 implemented on the POWER6 processor and other processors that
19903 support the PowerPC V2.05 architecture. The -mmfpgpr option allows
19904 GCC to generate the FP move to/from general-purpose register
19905 instructions implemented on the POWER6X processor and other
19906 processors that support the extended PowerPC V2.05 architecture.
19907 The -mhard-dfp option allows GCC to generate the decimal floating-
19908 point instructions implemented on some POWER processors.
19909 The -mpowerpc64 option allows GCC to generate the additional 64-bit
19911 instructions that are found in the full PowerPC64 architecture and
19912 to treat GPRs as 64-bit, doubleword quantities. GCC defaults to
19913 -mno-powerpc64.
19914 -mcpu=cpu_type
19916 Set architecture type, register usage, and instruction scheduling
19917 parameters for machine type cpu_type. Supported values for
19918 cpu_type are 401, 403, 405, 405fp, 440, 440fp, 464, 464fp, 476,
19919 476fp, 505, 601, 602, 603, 603e, 604, 604e, 620, 630, 740, 7400,
19920 7450, 750, 801, 821, 823, 860, 970, 8540, a2, e300c2, e300c3,
19921 e500mc, e500mc64, e5500, e6500, ec603e, G3, G4, G5, titan, power3,
19922 power4, power5, power5+, power6, power6x, power7, power8, power9,
19923 powerpc, powerpc64, powerpc64le, rs64, and native.
19924 -mcpu=powerpc, -mcpu=powerpc64, and -mcpu=powerpc64le specify pure
19926 32-bit PowerPC (either endian), 64-bit big endian PowerPC and
19927 64-bit little endian PowerPC architecture machine types, with an
19928 appropriate, generic processor model assumed for scheduling
19929 purposes.
19930 Specifying native as cpu type detects and selects the architecture
19932 option that corresponds to the host processor of the system
19933 performing the compilation. -mcpu=native has no effect if GCC does
19934 not recognize the processor.
19935 The other options specify a specific processor. Code generated
19937 under those options runs best on that processor, and may not run at
19938 all on others.
19939 The -mcpu options automatically enable or disable the following
19941 options:
19942 -maltivec -mfprnd -mhard-float -mmfcrf -mmultiple -mpopcntb
19944 -mpopcntd -mpowerpc64 -mpowerpc-gpopt -mpowerpc-gfxopt
19945 -msingle-float -mdouble-float -msimple-fpu -mmulhw -mdlmzb
19946 -mmfpgpr -mvsx -mcrypto -mhtm -mpower8-fusion -mpower8-vector
19947 -mquad-memory -mquad-memory-atomic -mfloat128 -mfloat128-hardware
19948 The particular options set for any particular CPU varies between
19950 compiler versions, depending on what setting seems to produce
19951 optimal code for that CPU; it doesn't necessarily reflect the
19952 actual hardware's capabilities. If you wish to set an individual
19953 option to a particular value, you may specify it after the -mcpu
19954 option, like -mcpu=970 -mno-altivec.
19955 On AIX, the -maltivec and -mpowerpc64 options are not enabled or
19957 disabled by the -mcpu option at present because AIX does not have
19958 full support for these options. You may still enable or disable
19959 them individually if you're sure it'll work in your environment.
19960 -mtune=cpu_type
19962 Set the instruction scheduling parameters for machine type
19963 cpu_type, but do not set the architecture type or register usage,
19964 as -mcpu=cpu_type does. The same values for cpu_type are used for
19965 -mtune as for -mcpu. If both are specified, the code generated
19966 uses the architecture and registers set by -mcpu, but the
19967 scheduling parameters set by -mtune.
19968 -mcmodel=small
19970 Generate PowerPC64 code for the small model: The TOC is limited to
19971 64k.
19972 -mcmodel=medium
19974 Generate PowerPC64 code for the medium model: The TOC and other
19975 static data may be up to a total of 4G in size. This is the
19976 default for 64-bit Linux.
19977 -mcmodel=large
19979 Generate PowerPC64 code for the large model: The TOC may be up to
19980 4G in size. Other data and code is only limited by the 64-bit
19981 address space.
19982 -maltivec
19984 -mno-altivec
19985 Generate code that uses (does not use) AltiVec instructions, and
19986 also enable the use of built-in functions that allow more direct
19987 access to the AltiVec instruction set. You may also need to set
19988 -mabi=altivec to adjust the current ABI with AltiVec ABI
19989 enhancements.
19990 When -maltivec is used, rather than -maltivec=le or -maltivec=be,
19992 the element order for AltiVec intrinsics such as "vec_splat",
19993 "vec_extract", and "vec_insert" match array element order
19994 corresponding to the endianness of the target. That is, element
19995 zero identifies the leftmost element in a vector register when
19996 targeting a big-endian platform, and identifies the rightmost
19997 element in a vector register when targeting a little-endian
19998 platform.
19999 -maltivec=be
20001 Generate AltiVec instructions using big-endian element order,
20002 regardless of whether the target is big- or little-endian. This is
20003 the default when targeting a big-endian platform. Using this
20004 option is currently deprecated. Support for this feature will be
20005 removed in GCC 9.
20006 The element order is used to interpret element numbers in AltiVec
20008 intrinsics such as "vec_splat", "vec_extract", and "vec_insert".
20009 By default, these match array element order corresponding to the
20010 endianness for the target.
20011 -maltivec=le
20013 Generate AltiVec instructions using little-endian element order,
20014 regardless of whether the target is big- or little-endian. This is
20015 the default when targeting a little-endian platform. This option
20016 is currently ignored when targeting a big-endian platform.
20017 The element order is used to interpret element numbers in AltiVec
20019 intrinsics such as "vec_splat", "vec_extract", and "vec_insert".
20020 By default, these match array element order corresponding to the
20021 endianness for the target.
20022 -mvrsave
20024 -mno-vrsave
20025 Generate VRSAVE instructions when generating AltiVec code.
20026 -msecure-plt
20028 Generate code that allows ld and ld.so to build executables and
20029 shared libraries with non-executable ".plt" and ".got" sections.
20030 This is a PowerPC 32-bit SYSV ABI option.
20031 -mbss-plt
20033 Generate code that uses a BSS ".plt" section that ld.so fills in,
20034 and requires ".plt" and ".got" sections that are both writable and
20035 executable. This is a PowerPC 32-bit SYSV ABI option.
20036 -misel
20038 -mno-isel
20039 This switch enables or disables the generation of ISEL
20040 instructions.
20041 -misel=yes/no
20043 This switch has been deprecated. Use -misel and -mno-isel instead.
20044 -mpaired
20046 -mno-paired
20047 This switch enables or disables the generation of PAIRED simd
20048 instructions.
20049 -mvsx
20051 -mno-vsx
20052 Generate code that uses (does not use) vector/scalar (VSX)
20053 instructions, and also enable the use of built-in functions that
20054 allow more direct access to the VSX instruction set.
20055 -mcrypto
20057 -mno-crypto
20058 Enable the use (disable) of the built-in functions that allow
20059 direct access to the cryptographic instructions that were added in
20060 version 2.07 of the PowerPC ISA.
20061 -mhtm
20063 -mno-htm
20064 Enable (disable) the use of the built-in functions that allow
20065 direct access to the Hardware Transactional Memory (HTM)
20066 instructions that were added in version 2.07 of the PowerPC ISA.
20067 -mpower8-fusion
20069 -mno-power8-fusion
20070 Generate code that keeps (does not keeps) some integer operations
20071 adjacent so that the instructions can be fused together on power8
20072 and later processors.
20073 -mpower8-vector
20075 -mno-power8-vector
20076 Generate code that uses (does not use) the vector and scalar
20077 instructions that were added in version 2.07 of the PowerPC ISA.
20078 Also enable the use of built-in functions that allow more direct
20079 access to the vector instructions.
20080 -mquad-memory
20082 -mno-quad-memory
20083 Generate code that uses (does not use) the non-atomic quad word
20084 memory instructions. The -mquad-memory option requires use of
20085 64-bit mode.
20086 -mquad-memory-atomic
20088 -mno-quad-memory-atomic
20089 Generate code that uses (does not use) the atomic quad word memory
20090 instructions. The -mquad-memory-atomic option requires use of
20091 64-bit mode.
20092 -mfloat128
20094 -mno-float128
20095 Enable/disable the __float128 keyword for IEEE 128-bit floating
20096 point and use either software emulation for IEEE 128-bit floating
20097 point or hardware instructions.
20098 The VSX instruction set (-mvsx, -mcpu=power7, -mcpu=power8), or
20100 -mcpu=power9 must be enabled to use the IEEE 128-bit floating point
20101 support. The IEEE 128-bit floating point support only works on
20102 PowerPC Linux systems.
20103 The default for -mfloat128 is enabled on PowerPC Linux systems
20105 using the VSX instruction set, and disabled on other systems.
20106 If you use the ISA 3.0 instruction set (-mpower9-vector or
20108 -mcpu=power9) on a 64-bit system, the IEEE 128-bit floating point
20109 support will also enable the generation of ISA 3.0 IEEE 128-bit
20110 floating point instructions. Otherwise, if you do not specify to
20111 generate ISA 3.0 instructions or you are targeting a 32-bit big
20112 endian system, IEEE 128-bit floating point will be done with
20113 software emulation.
20114 -mfloat128-hardware
20116 -mno-float128-hardware
20117 Enable/disable using ISA 3.0 hardware instructions to support the
20118 __float128 data type.
20119 The default for -mfloat128-hardware is enabled on PowerPC Linux
20121 systems using the ISA 3.0 instruction set, and disabled on other
20122 systems.
20123 -m32
20125 -m64
20126 Generate code for 32-bit or 64-bit environments of Darwin and SVR4
20127 targets (including GNU/Linux). The 32-bit environment sets int,
20128 long and pointer to 32 bits and generates code that runs on any
20129 PowerPC variant. The 64-bit environment sets int to 32 bits and
20130 long and pointer to 64 bits, and generates code for PowerPC64, as
20131 for -mpowerpc64.
20132 -mfull-toc
20134 -mno-fp-in-toc
20135 -mno-sum-in-toc
20136 -mminimal-toc
20137 Modify generation of the TOC (Table Of Contents), which is created
20138 for every executable file. The -mfull-toc option is selected by
20139 default. In that case, GCC allocates at least one TOC entry for
20140 each unique non-automatic variable reference in your program. GCC
20141 also places floating-point constants in the TOC. However, only
20142 16,384 entries are available in the TOC.
20143 If you receive a linker error message that saying you have
20145 overflowed the available TOC space, you can reduce the amount of
20146 TOC space used with the -mno-fp-in-toc and -mno-sum-in-toc options.
20147 -mno-fp-in-toc prevents GCC from putting floating-point constants
20148 in the TOC and -mno-sum-in-toc forces GCC to generate code to
20149 calculate the sum of an address and a constant at run time instead
20150 of putting that sum into the TOC. You may specify one or both of
20151 these options. Each causes GCC to produce very slightly slower and
20152 larger code at the expense of conserving TOC space.
20153 If you still run out of space in the TOC even when you specify both
20155 of these options, specify -mminimal-toc instead. This option
20156 causes GCC to make only one TOC entry for every file. When you
20157 specify this option, GCC produces code that is slower and larger
20158 but which uses extremely little TOC space. You may wish to use
20159 this option only on files that contain less frequently-executed
20160 code.
20161 -maix64
20163 -maix32
20164 Enable 64-bit AIX ABI and calling convention: 64-bit pointers,
20165 64-bit "long" type, and the infrastructure needed to support them.
20166 Specifying -maix64 implies -mpowerpc64, while -maix32 disables the
20167 64-bit ABI and implies -mno-powerpc64. GCC defaults to -maix32.
20168 -mxl-compat
20170 -mno-xl-compat
20171 Produce code that conforms more closely to IBM XL compiler
20172 semantics when using AIX-compatible ABI. Pass floating-point
20173 arguments to prototyped functions beyond the register save area
20174 (RSA) on the stack in addition to argument FPRs. Do not assume
20175 that most significant double in 128-bit long double value is
20176 properly rounded when comparing values and converting to double.
20177 Use XL symbol names for long double support routines.
20178 The AIX calling convention was extended but not initially
20180 documented to handle an obscure K&R C case of calling a function
20181 that takes the address of its arguments with fewer arguments than
20182 declared. IBM XL compilers access floating-point arguments that do
20183 not fit in the RSA from the stack when a subroutine is compiled
20184 without optimization. Because always storing floating-point
20185 arguments on the stack is inefficient and rarely needed, this
20186 option is not enabled by default and only is necessary when calling
20187 subroutines compiled by IBM XL compilers without optimization.
20188 -mpe
20190 Support IBM RS/6000 SP Parallel Environment (PE). Link an
20191 application written to use message passing with special startup
20192 code to enable the application to run. The system must have PE
20193 installed in the standard location (/usr/lpp/ppe.poe/), or the
20194 specs file must be overridden with the -specs= option to specify
20195 the appropriate directory location. The Parallel Environment does
20196 not support threads, so the -mpe option and the -pthread option are
20197 incompatible.
20198 -malign-natural
20200 -malign-power
20201 On AIX, 32-bit Darwin, and 64-bit PowerPC GNU/Linux, the option
20202 -malign-natural overrides the ABI-defined alignment of larger
20203 types, such as floating-point doubles, on their natural size-based
20204 boundary. The option -malign-power instructs GCC to follow the
20205 ABI-specified alignment rules. GCC defaults to the standard
20206 alignment defined in the ABI.
20207 On 64-bit Darwin, natural alignment is the default, and
20209 -malign-power is not supported.
20210 -msoft-float
20212 -mhard-float
20213 Generate code that does not use (uses) the floating-point register
20214 set. Software floating-point emulation is provided if you use the
20215 -msoft-float option, and pass the option to GCC when linking.
20216 -msingle-float
20218 -mdouble-float
20219 Generate code for single- or double-precision floating-point
20220 operations. -mdouble-float implies -msingle-float.
20221 -msimple-fpu
20223 Do not generate "sqrt" and "div" instructions for hardware
20224 floating-point unit.
20225 -mfpu=name
20227 Specify type of floating-point unit. Valid values for name are
20228 sp_lite (equivalent to -msingle-float -msimple-fpu), dp_lite
20229 (equivalent to -mdouble-float -msimple-fpu), sp_full (equivalent to
20230 -msingle-float), and dp_full (equivalent to -mdouble-float).
20231 -mxilinx-fpu
20233 Perform optimizations for the floating-point unit on Xilinx PPC
20234 405/440.
20235 -mmultiple
20237 -mno-multiple
20238 Generate code that uses (does not use) the load multiple word
20239 instructions and the store multiple word instructions. These
20240 instructions are generated by default on POWER systems, and not
20241 generated on PowerPC systems. Do not use -mmultiple on little-
20242 endian PowerPC systems, since those instructions do not work when
20243 the processor is in little-endian mode. The exceptions are PPC740
20244 and PPC750 which permit these instructions in little-endian mode.
20245 -mupdate
20247 -mno-update
20248 Generate code that uses (does not use) the load or store
20249 instructions that update the base register to the address of the
20250 calculated memory location. These instructions are generated by
20251 default. If you use -mno-update, there is a small window between
20252 the time that the stack pointer is updated and the address of the
20253 previous frame is stored, which means code that walks the stack
20254 frame across interrupts or signals may get corrupted data.
20255 -mavoid-indexed-addresses
20257 -mno-avoid-indexed-addresses
20258 Generate code that tries to avoid (not avoid) the use of indexed
20259 load or store instructions. These instructions can incur a
20260 performance penalty on Power6 processors in certain situations,
20261 such as when stepping through large arrays that cross a 16M
20262 boundary. This option is enabled by default when targeting Power6
20263 and disabled otherwise.
20264 -mfused-madd
20266 -mno-fused-madd
20267 Generate code that uses (does not use) the floating-point multiply
20268 and accumulate instructions. These instructions are generated by
20269 default if hardware floating point is used. The machine-dependent
20270 -mfused-madd option is now mapped to the machine-independent
20271 -ffp-contract=fast option, and -mno-fused-madd is mapped to
20272 -ffp-contract=off.
20273 -mmulhw
20275 -mno-mulhw
20276 Generate code that uses (does not use) the half-word multiply and
20277 multiply-accumulate instructions on the IBM 405, 440, 464 and 476
20278 processors. These instructions are generated by default when
20279 targeting those processors.
20280 -mdlmzb
20282 -mno-dlmzb
20283 Generate code that uses (does not use) the string-search dlmzb
20284 instruction on the IBM 405, 440, 464 and 476 processors. This
20285 instruction is generated by default when targeting those
20286 processors.
20287 -mno-bit-align
20289 -mbit-align
20290 On System V.4 and embedded PowerPC systems do not (do) force
20291 structures and unions that contain bit-fields to be aligned to the
20292 base type of the bit-field.
20293 For example, by default a structure containing nothing but 8
20295 "unsigned" bit-fields of length 1 is aligned to a 4-byte boundary
20296 and has a size of 4 bytes. By using -mno-bit-align, the structure
20297 is aligned to a 1-byte boundary and is 1 byte in size.
20298 -mno-strict-align
20300 -mstrict-align
20301 On System V.4 and embedded PowerPC systems do not (do) assume that
20302 unaligned memory references are handled by the system.
20303 -mrelocatable
20305 -mno-relocatable
20306 Generate code that allows (does not allow) a static executable to
20307 be relocated to a different address at run time. A simple embedded
20308 PowerPC system loader should relocate the entire contents of
20309 ".got2" and 4-byte locations listed in the ".fixup" section, a
20310 table of 32-bit addresses generated by this option. For this to
20311 work, all objects linked together must be compiled with
20312 -mrelocatable or -mrelocatable-lib. -mrelocatable code aligns the
20313 stack to an 8-byte boundary.
20314 -mrelocatable-lib
20316 -mno-relocatable-lib
20317 Like -mrelocatable, -mrelocatable-lib generates a ".fixup" section
20318 to allow static executables to be relocated at run time, but
20319 -mrelocatable-lib does not use the smaller stack alignment of
20320 -mrelocatable. Objects compiled with -mrelocatable-lib may be
20321 linked with objects compiled with any combination of the
20322 -mrelocatable options.
20323 -mno-toc
20325 -mtoc
20326 On System V.4 and embedded PowerPC systems do not (do) assume that
20327 register 2 contains a pointer to a global area pointing to the
20328 addresses used in the program.
20329 -mlittle
20331 -mlittle-endian
20332 On System V.4 and embedded PowerPC systems compile code for the
20333 processor in little-endian mode. The -mlittle-endian option is the
20334 same as -mlittle.
20335 -mbig
20337 -mbig-endian
20338 On System V.4 and embedded PowerPC systems compile code for the
20339 processor in big-endian mode. The -mbig-endian option is the same
20340 as -mbig.
20341 -mdynamic-no-pic
20343 On Darwin and Mac OS X systems, compile code so that it is not
20344 relocatable, but that its external references are relocatable. The
20345 resulting code is suitable for applications, but not shared
20346 libraries.
20347 -msingle-pic-base
20349 Treat the register used for PIC addressing as read-only, rather
20350 than loading it in the prologue for each function. The runtime
20351 system is responsible for initializing this register with an
20352 appropriate value before execution begins.
20353 -mprioritize-restricted-insns=priority
20355 This option controls the priority that is assigned to dispatch-slot
20356 restricted instructions during the second scheduling pass. The
20357 argument priority takes the value 0, 1, or 2 to assign no, highest,
20358 or second-highest (respectively) priority to dispatch-slot
20359 restricted instructions.
20360 -msched-costly-dep=dependence_type
20362 This option controls which dependences are considered costly by the
20363 target during instruction scheduling. The argument dependence_type
20364 takes one of the following values:
20365 no No dependence is costly.
20367 all All dependences are costly.
20369 true_store_to_load
20371 A true dependence from store to load is costly.
20372 store_to_load
20374 Any dependence from store to load is costly.
20375 number
20377 Any dependence for which the latency is greater than or equal
20378 to number is costly.
20379 -minsert-sched-nops=scheme
20381 This option controls which NOP insertion scheme is used during the
20382 second scheduling pass. The argument scheme takes one of the
20383 following values:
20384 no Don't insert NOPs.
20386 pad Pad with NOPs any dispatch group that has vacant issue slots,
20388 according to the scheduler's grouping.
20389 regroup_exact
20391 Insert NOPs to force costly dependent insns into separate
20392 groups. Insert exactly as many NOPs as needed to force an insn
20393 to a new group, according to the estimated processor grouping.
20394 number
20396 Insert NOPs to force costly dependent insns into separate
20397 groups. Insert number NOPs to force an insn to a new group.
20398 -mcall-sysv
20400 On System V.4 and embedded PowerPC systems compile code using
20401 calling conventions that adhere to the March 1995 draft of the
20402 System V Application Binary Interface, PowerPC processor
20403 supplement. This is the default unless you configured GCC using
20404 powerpc-*-eabiaix.
20405 -mcall-sysv-eabi
20407 -mcall-eabi
20408 Specify both -mcall-sysv and -meabi options.
20409 -mcall-sysv-noeabi
20411 Specify both -mcall-sysv and -mno-eabi options.
20412 -mcall-aixdesc
20414 On System V.4 and embedded PowerPC systems compile code for the AIX
20415 operating system.
20416 -mcall-linux
20418 On System V.4 and embedded PowerPC systems compile code for the
20419 Linux-based GNU system.
20420 -mcall-freebsd
20422 On System V.4 and embedded PowerPC systems compile code for the
20423 FreeBSD operating system.
20424 -mcall-netbsd
20426 On System V.4 and embedded PowerPC systems compile code for the
20427 NetBSD operating system.
20428 -mcall-openbsd
20430 On System V.4 and embedded PowerPC systems compile code for the
20431 OpenBSD operating system.
20432 -mtraceback=traceback_type
20434 Select the type of traceback table. Valid values for traceback_type
20435 are full, part, and no.
20436 -maix-struct-return
20438 Return all structures in memory (as specified by the AIX ABI).
20439 -msvr4-struct-return
20441 Return structures smaller than 8 bytes in registers (as specified
20442 by the SVR4 ABI).
20443 -mabi=abi-type
20445 Extend the current ABI with a particular extension, or remove such
20446 extension. Valid values are altivec, no-altivec, spe, no-spe,
20447 ibmlongdouble, ieeelongdouble, elfv1, elfv2.
20448 -mabi=ibmlongdouble
20450 Change the current ABI to use IBM extended-precision long double.
20451 This is not likely to work if your system defaults to using IEEE
20452 extended-precision long double. If you change the long double type
20453 from IEEE extended-precision, the compiler will issue a warning
20454 unless you use the -Wno-psabi option.
20455 -mabi=ieeelongdouble
20457 Change the current ABI to use IEEE extended-precision long double.
20458 This is not likely to work if your system defaults to using IBM
20459 extended-precision long double. If you change the long double type
20460 from IBM extended-precision, the compiler will issue a warning
20461 unless you use the -Wno-psabi option.
20462 -mabi=elfv1
20464 Change the current ABI to use the ELFv1 ABI. This is the default
20465 ABI for big-endian PowerPC 64-bit Linux. Overriding the default
20466 ABI requires special system support and is likely to fail in
20467 spectacular ways.
20468 -mabi=elfv2
20470 Change the current ABI to use the ELFv2 ABI. This is the default
20471 ABI for little-endian PowerPC 64-bit Linux. Overriding the default
20472 ABI requires special system support and is likely to fail in
20473 spectacular ways.
20474 -mgnu-attribute
20476 -mno-gnu-attribute
20477 Emit .gnu_attribute assembly directives to set tag/value pairs in a
20478 .gnu.attributes section that specify ABI variations in function
20479 parameters or return values.
20480 -mprototype
20482 -mno-prototype
20483 On System V.4 and embedded PowerPC systems assume that all calls to
20484 variable argument functions are properly prototyped. Otherwise,
20485 the compiler must insert an instruction before every non-prototyped
20486 call to set or clear bit 6 of the condition code register ("CR") to
20487 indicate whether floating-point values are passed in the floating-
20488 point registers in case the function takes variable arguments.
20489 With -mprototype, only calls to prototyped variable argument
20490 functions set or clear the bit.
20491 -msim
20493 On embedded PowerPC systems, assume that the startup module is
20494 called sim-crt0.o and that the standard C libraries are libsim.a
20495 and libc.a. This is the default for powerpc-*-eabisim
20496 configurations.
20497 -mmvme
20499 On embedded PowerPC systems, assume that the startup module is
20500 called crt0.o and the standard C libraries are libmvme.a and
20501 libc.a.
20502 -mads
20504 On embedded PowerPC systems, assume that the startup module is
20505 called crt0.o and the standard C libraries are libads.a and libc.a.
20506 -myellowknife
20508 On embedded PowerPC systems, assume that the startup module is
20509 called crt0.o and the standard C libraries are libyk.a and libc.a.
20510 -mvxworks
20512 On System V.4 and embedded PowerPC systems, specify that you are
20513 compiling for a VxWorks system.
20514 -memb
20516 On embedded PowerPC systems, set the "PPC_EMB" bit in the ELF flags
20517 header to indicate that eabi extended relocations are used.
20518 -meabi
20520 -mno-eabi
20521 On System V.4 and embedded PowerPC systems do (do not) adhere to
20522 the Embedded Applications Binary Interface (EABI), which is a set
20523 of modifications to the System V.4 specifications. Selecting
20524 -meabi means that the stack is aligned to an 8-byte boundary, a
20525 function "__eabi" is called from "main" to set up the EABI
20526 environment, and the -msdata option can use both "r2" and "r13" to
20527 point to two separate small data areas. Selecting -mno-eabi means
20528 that the stack is aligned to a 16-byte boundary, no EABI
20529 initialization function is called from "main", and the -msdata
20530 option only uses "r13" to point to a single small data area. The
20531 -meabi option is on by default if you configured GCC using one of
20532 the powerpc*-*-eabi* options.
20533 -msdata=eabi
20535 On System V.4 and embedded PowerPC systems, put small initialized
20536 "const" global and static data in the ".sdata2" section, which is
20537 pointed to by register "r2". Put small initialized non-"const"
20538 global and static data in the ".sdata" section, which is pointed to
20539 by register "r13". Put small uninitialized global and static data
20540 in the ".sbss" section, which is adjacent to the ".sdata" section.
20541 The -msdata=eabi option is incompatible with the -mrelocatable
20542 option. The -msdata=eabi option also sets the -memb option.
20543 -msdata=sysv
20545 On System V.4 and embedded PowerPC systems, put small global and
20546 static data in the ".sdata" section, which is pointed to by
20547 register "r13". Put small uninitialized global and static data in
20548 the ".sbss" section, which is adjacent to the ".sdata" section.
20549 The -msdata=sysv option is incompatible with the -mrelocatable
20550 option.
20551 -msdata=default
20553 -msdata
20554 On System V.4 and embedded PowerPC systems, if -meabi is used,
20555 compile code the same as -msdata=eabi, otherwise compile code the
20556 same as -msdata=sysv.
20557 -msdata=data
20559 On System V.4 and embedded PowerPC systems, put small global data
20560 in the ".sdata" section. Put small uninitialized global data in
20561 the ".sbss" section. Do not use register "r13" to address small
20562 data however. This is the default behavior unless other -msdata
20563 options are used.
20564 -msdata=none
20566 -mno-sdata
20567 On embedded PowerPC systems, put all initialized global and static
20568 data in the ".data" section, and all uninitialized data in the
20569 ".bss" section.
20570 -mreadonly-in-sdata
20572 -mreadonly-in-sdata
20573 Put read-only objects in the ".sdata" section as well. This is the
20574 default.
20575 -mblock-move-inline-limit=num
20577 Inline all block moves (such as calls to "memcpy" or structure
20578 copies) less than or equal to num bytes. The minimum value for num
20579 is 32 bytes on 32-bit targets and 64 bytes on 64-bit targets. The
20580 default value is target-specific.
20581 -mblock-compare-inline-limit=num
20583 Generate non-looping inline code for all block compares (such as
20584 calls to "memcmp" or structure compares) less than or equal to num
20585 bytes. If num is 0, all inline expansion (non-loop and loop) of
20586 block compare is disabled. The default value is target-specific.
20587 -mblock-compare-inline-loop-limit=num
20589 Generate an inline expansion using loop code for all block compares
20590 that are less than or equal to num bytes, but greater than the
20591 limit for non-loop inline block compare expansion. If the block
20592 length is not constant, at most num bytes will be compared before
20593 "memcmp" is called to compare the remainder of the block. The
20594 default value is target-specific.
20595 -mstring-compare-inline-limit=num
20597 Compare at most num string bytes with inline code. If the
20598 difference or end of string is not found at the end of the inline
20599 compare a call to "strcmp" or "strncmp" will take care of the rest
20600 of the comparison. The default is 64 bytes.
20601 -G num
20603 On embedded PowerPC systems, put global and static items less than
20604 or equal to num bytes into the small data or BSS sections instead
20605 of the normal data or BSS section. By default, num is 8. The -G
20606 num switch is also passed to the linker. All modules should be
20607 compiled with the same -G num value.
20608 -mregnames
20610 -mno-regnames
20611 On System V.4 and embedded PowerPC systems do (do not) emit
20612 register names in the assembly language output using symbolic
20613 forms.
20614 -mlongcall
20616 -mno-longcall
20617 By default assume that all calls are far away so that a longer and
20618 more expensive calling sequence is required. This is required for
20619 calls farther than 32 megabytes (33,554,432 bytes) from the current
20620 location. A short call is generated if the compiler knows the call
20621 cannot be that far away. This setting can be overridden by the
20622 "shortcall" function attribute, or by "#pragma longcall(0)".
20623 Some linkers are capable of detecting out-of-range calls and
20625 generating glue code on the fly. On these systems, long calls are
20626 unnecessary and generate slower code. As of this writing, the AIX
20627 linker can do this, as can the GNU linker for PowerPC/64. It is
20628 planned to add this feature to the GNU linker for 32-bit PowerPC
20629 systems as well.
20630 On Darwin/PPC systems, "#pragma longcall" generates "jbsr callee,
20632 L42", plus a branch island (glue code). The two target addresses
20633 represent the callee and the branch island. The Darwin/PPC linker
20634 prefers the first address and generates a "bl callee" if the PPC
20635 "bl" instruction reaches the callee directly; otherwise, the linker
20636 generates "bl L42" to call the branch island. The branch island is
20637 appended to the body of the calling function; it computes the full
20638 32-bit address of the callee and jumps to it.
20639 On Mach-O (Darwin) systems, this option directs the compiler emit
20641 to the glue for every direct call, and the Darwin linker decides
20642 whether to use or discard it.
20643 In the future, GCC may ignore all longcall specifications when the
20645 linker is known to generate glue.
20646 -mtls-markers
20648 -mno-tls-markers
20649 Mark (do not mark) calls to "__tls_get_addr" with a relocation
20650 specifying the function argument. The relocation allows the linker
20651 to reliably associate function call with argument setup
20652 instructions for TLS optimization, which in turn allows GCC to
20653 better schedule the sequence.
20654 -mrecip
20656 -mno-recip
20657 This option enables use of the reciprocal estimate and reciprocal
20658 square root estimate instructions with additional Newton-Raphson
20659 steps to increase precision instead of doing a divide or square
20660 root and divide for floating-point arguments. You should use the
20661 -ffast-math option when using -mrecip (or at least
20662 -funsafe-math-optimizations, -ffinite-math-only, -freciprocal-math
20663 and -fno-trapping-math). Note that while the throughput of the
20664 sequence is generally higher than the throughput of the non-
20665 reciprocal instruction, the precision of the sequence can be
20666 decreased by up to 2 ulp (i.e. the inverse of 1.0 equals
20667 0.99999994) for reciprocal square roots.
20668 -mrecip=opt
20670 This option controls which reciprocal estimate instructions may be
20671 used. opt is a comma-separated list of options, which may be
20672 preceded by a "!" to invert the option:
20673 all Enable all estimate instructions.
20675 default
20677 Enable the default instructions, equivalent to -mrecip.
20678 none
20680 Disable all estimate instructions, equivalent to -mno-recip.
20681 div Enable the reciprocal approximation instructions for both
20683 single and double precision.
20684 divf
20686 Enable the single-precision reciprocal approximation
20687 instructions.
20688 divd
20690 Enable the double-precision reciprocal approximation
20691 instructions.
20692 rsqrt
20694 Enable the reciprocal square root approximation instructions
20695 for both single and double precision.
20696 rsqrtf
20698 Enable the single-precision reciprocal square root
20699 approximation instructions.
20700 rsqrtd
20702 Enable the double-precision reciprocal square root
20703 approximation instructions.
20704 So, for example, -mrecip=all,!rsqrtd enables all of the reciprocal
20706 estimate instructions, except for the "FRSQRTE", "XSRSQRTEDP", and
20707 "XVRSQRTEDP" instructions which handle the double-precision
20708 reciprocal square root calculations.
20709 -mrecip-precision
20711 -mno-recip-precision
20712 Assume (do not assume) that the reciprocal estimate instructions
20713 provide higher-precision estimates than is mandated by the PowerPC
20714 ABI. Selecting -mcpu=power6, -mcpu=power7 or -mcpu=power8
20715 automatically selects -mrecip-precision. The double-precision
20716 square root estimate instructions are not generated by default on
20717 low-precision machines, since they do not provide an estimate that
20718 converges after three steps.
20719 -mveclibabi=type
20721 Specifies the ABI type to use for vectorizing intrinsics using an
20722 external library. The only type supported at present is mass,
20723 which specifies to use IBM's Mathematical Acceleration Subsystem
20724 (MASS) libraries for vectorizing intrinsics using external
20725 libraries. GCC currently emits calls to "acosd2", "acosf4",
20726 "acoshd2", "acoshf4", "asind2", "asinf4", "asinhd2", "asinhf4",
20727 "atan2d2", "atan2f4", "atand2", "atanf4", "atanhd2", "atanhf4",
20728 "cbrtd2", "cbrtf4", "cosd2", "cosf4", "coshd2", "coshf4", "erfcd2",
20729 "erfcf4", "erfd2", "erff4", "exp2d2", "exp2f4", "expd2", "expf4",
20730 "expm1d2", "expm1f4", "hypotd2", "hypotf4", "lgammad2", "lgammaf4",
20731 "log10d2", "log10f4", "log1pd2", "log1pf4", "log2d2", "log2f4",
20732 "logd2", "logf4", "powd2", "powf4", "sind2", "sinf4", "sinhd2",
20733 "sinhf4", "sqrtd2", "sqrtf4", "tand2", "tanf4", "tanhd2", and
20734 "tanhf4" when generating code for power7. Both -ftree-vectorize
20735 and -funsafe-math-optimizations must also be enabled. The MASS
20736 libraries must be specified at link time.
20737 -mfriz
20739 -mno-friz
20740 Generate (do not generate) the "friz" instruction when the
20741 -funsafe-math-optimizations option is used to optimize rounding of
20742 floating-point values to 64-bit integer and back to floating point.
20743 The "friz" instruction does not return the same value if the
20744 floating-point number is too large to fit in an integer.
20745 -mpointers-to-nested-functions
20747 -mno-pointers-to-nested-functions
20748 Generate (do not generate) code to load up the static chain
20749 register ("r11") when calling through a pointer on AIX and 64-bit
20750 Linux systems where a function pointer points to a 3-word
20751 descriptor giving the function address, TOC value to be loaded in
20752 register "r2", and static chain value to be loaded in register
20753 "r11". The -mpointers-to-nested-functions is on by default. You
20754 cannot call through pointers to nested functions or pointers to
20755 functions compiled in other languages that use the static chain if
20756 you use -mno-pointers-to-nested-functions.
20757 -msave-toc-indirect
20759 -mno-save-toc-indirect
20760 Generate (do not generate) code to save the TOC value in the
20761 reserved stack location in the function prologue if the function
20762 calls through a pointer on AIX and 64-bit Linux systems. If the
20763 TOC value is not saved in the prologue, it is saved just before the
20764 call through the pointer. The -mno-save-toc-indirect option is the
20765 default.
20766 -mcompat-align-parm
20768 -mno-compat-align-parm
20769 Generate (do not generate) code to pass structure parameters with a
20770 maximum alignment of 64 bits, for compatibility with older versions
20771 of GCC.
20772 Older versions of GCC (prior to 4.9.0) incorrectly did not align a
20774 structure parameter on a 128-bit boundary when that structure
20775 contained a member requiring 128-bit alignment. This is corrected
20776 in more recent versions of GCC. This option may be used to
20777 generate code that is compatible with functions compiled with older
20778 versions of GCC.
20779 The -mno-compat-align-parm option is the default.
20781 -mstack-protector-guard=guard
20783 -mstack-protector-guard-reg=reg
20784 -mstack-protector-guard-offset=offset
20785 -mstack-protector-guard-symbol=symbol
20786 Generate stack protection code using canary at guard. Supported
20787 locations are global for global canary or tls for per-thread canary
20788 in the TLS block (the default with GNU libc version 2.4 or later).
20789 With the latter choice the options -mstack-protector-guard-reg=reg
20791 and -mstack-protector-guard-offset=offset furthermore specify which
20792 register to use as base register for reading the canary, and from
20793 what offset from that base register. The default for those is as
20794 specified in the relevant ABI.
20795 -mstack-protector-guard-symbol=symbol overrides the offset with a
20796 symbol reference to a canary in the TLS block.
20797 RX Options
20799 These command-line options are defined for RX targets:
20800 -m64bit-doubles
20802 -m32bit-doubles
20803 Make the "double" data type be 64 bits (-m64bit-doubles) or 32 bits
20804 (-m32bit-doubles) in size. The default is -m32bit-doubles. Note
20805 RX floating-point hardware only works on 32-bit values, which is
20806 why the default is -m32bit-doubles.
20807 -fpu
20809 -nofpu
20810 Enables (-fpu) or disables (-nofpu) the use of RX floating-point
20811 hardware. The default is enabled for the RX600 series and disabled
20812 for the RX200 series.
20813 Floating-point instructions are only generated for 32-bit floating-
20815 point values, however, so the FPU hardware is not used for doubles
20816 if the -m64bit-doubles option is used.
20817 Note If the -fpu option is enabled then -funsafe-math-optimizations
20819 is also enabled automatically. This is because the RX FPU
20820 instructions are themselves unsafe.
20821 -mcpu=name
20823 Selects the type of RX CPU to be targeted. Currently three types
20824 are supported, the generic RX600 and RX200 series hardware and the
20825 specific RX610 CPU. The default is RX600.
20826 The only difference between RX600 and RX610 is that the RX610 does
20828 not support the "MVTIPL" instruction.
20829 The RX200 series does not have a hardware floating-point unit and
20831 so -nofpu is enabled by default when this type is selected.
20832 -mbig-endian-data
20834 -mlittle-endian-data
20835 Store data (but not code) in the big-endian format. The default is
20836 -mlittle-endian-data, i.e. to store data in the little-endian
20837 format.
20838 -msmall-data-limit=N
20840 Specifies the maximum size in bytes of global and static variables
20841 which can be placed into the small data area. Using the small data
20842 area can lead to smaller and faster code, but the size of area is
20843 limited and it is up to the programmer to ensure that the area does
20844 not overflow. Also when the small data area is used one of the
20845 RX's registers (usually "r13") is reserved for use pointing to this
20846 area, so it is no longer available for use by the compiler. This
20847 could result in slower and/or larger code if variables are pushed
20848 onto the stack instead of being held in this register.
20849 Note, common variables (variables that have not been initialized)
20851 and constants are not placed into the small data area as they are
20852 assigned to other sections in the output executable.
20853 The default value is zero, which disables this feature. Note, this
20855 feature is not enabled by default with higher optimization levels
20856 (-O2 etc) because of the potentially detrimental effects of
20857 reserving a register. It is up to the programmer to experiment and
20858 discover whether this feature is of benefit to their program. See
20859 the description of the -mpid option for a description of how the
20860 actual register to hold the small data area pointer is chosen.
20861 -msim
20863 -mno-sim
20864 Use the simulator runtime. The default is to use the libgloss
20865 board-specific runtime.
20866 -mas100-syntax
20868 -mno-as100-syntax
20869 When generating assembler output use a syntax that is compatible
20870 with Renesas's AS100 assembler. This syntax can also be handled by
20871 the GAS assembler, but it has some restrictions so it is not
20872 generated by default.
20873 -mmax-constant-size=N
20875 Specifies the maximum size, in bytes, of a constant that can be
20876 used as an operand in a RX instruction. Although the RX
20877 instruction set does allow constants of up to 4 bytes in length to
20878 be used in instructions, a longer value equates to a longer
20879 instruction. Thus in some circumstances it can be beneficial to
20880 restrict the size of constants that are used in instructions.
20881 Constants that are too big are instead placed into a constant pool
20882 and referenced via register indirection.
20883 The value N can be between 0 and 4. A value of 0 (the default) or
20885 4 means that constants of any size are allowed.
20886 -mrelax
20888 Enable linker relaxation. Linker relaxation is a process whereby
20889 the linker attempts to reduce the size of a program by finding
20890 shorter versions of various instructions. Disabled by default.
20891 -mint-register=N
20893 Specify the number of registers to reserve for fast interrupt
20894 handler functions. The value N can be between 0 and 4. A value of
20895 1 means that register "r13" is reserved for the exclusive use of
20896 fast interrupt handlers. A value of 2 reserves "r13" and "r12". A
20897 value of 3 reserves "r13", "r12" and "r11", and a value of 4
20898 reserves "r13" through "r10". A value of 0, the default, does not
20899 reserve any registers.
20900 -msave-acc-in-interrupts
20902 Specifies that interrupt handler functions should preserve the
20903 accumulator register. This is only necessary if normal code might
20904 use the accumulator register, for example because it performs
20905 64-bit multiplications. The default is to ignore the accumulator
20906 as this makes the interrupt handlers faster.
20907 -mpid
20909 -mno-pid
20910 Enables the generation of position independent data. When enabled
20911 any access to constant data is done via an offset from a base
20912 address held in a register. This allows the location of constant
20913 data to be determined at run time without requiring the executable
20914 to be relocated, which is a benefit to embedded applications with
20915 tight memory constraints. Data that can be modified is not
20916 affected by this option.
20917 Note, using this feature reserves a register, usually "r13", for
20919 the constant data base address. This can result in slower and/or
20920 larger code, especially in complicated functions.
20921 The actual register chosen to hold the constant data base address
20923 depends upon whether the -msmall-data-limit and/or the
20924 -mint-register command-line options are enabled. Starting with
20925 register "r13" and proceeding downwards, registers are allocated
20926 first to satisfy the requirements of -mint-register, then -mpid and
20927 finally -msmall-data-limit. Thus it is possible for the small data
20928 area register to be "r8" if both -mint-register=4 and -mpid are
20929 specified on the command line.
20930 By default this feature is not enabled. The default can be
20932 restored via the -mno-pid command-line option.
20933 -mno-warn-multiple-fast-interrupts
20935 -mwarn-multiple-fast-interrupts
20936 Prevents GCC from issuing a warning message if it finds more than
20937 one fast interrupt handler when it is compiling a file. The
20938 default is to issue a warning for each extra fast interrupt handler
20939 found, as the RX only supports one such interrupt.
20940 -mallow-string-insns
20942 -mno-allow-string-insns
20943 Enables or disables the use of the string manipulation instructions
20944 "SMOVF", "SCMPU", "SMOVB", "SMOVU", "SUNTIL" "SWHILE" and also the
20945 "RMPA" instruction. These instructions may prefetch data, which is
20946 not safe to do if accessing an I/O register. (See section 12.2.7
20947 of the RX62N Group User's Manual for more information).
20948 The default is to allow these instructions, but it is not possible
20950 for GCC to reliably detect all circumstances where a string
20951 instruction might be used to access an I/O register, so their use
20952 cannot be disabled automatically. Instead it is reliant upon the
20953 programmer to use the -mno-allow-string-insns option if their
20954 program accesses I/O space.
20955 When the instructions are enabled GCC defines the C preprocessor
20957 symbol "__RX_ALLOW_STRING_INSNS__", otherwise it defines the symbol
20958 "__RX_DISALLOW_STRING_INSNS__".
20959 -mjsr
20961 -mno-jsr
20962 Use only (or not only) "JSR" instructions to access functions.
20963 This option can be used when code size exceeds the range of "BSR"
20964 instructions. Note that -mno-jsr does not mean to not use "JSR"
20965 but instead means that any type of branch may be used.
20966 Note: The generic GCC command-line option -ffixed-reg has special
20968 significance to the RX port when used with the "interrupt" function
20969 attribute. This attribute indicates a function intended to process
20970 fast interrupts. GCC ensures that it only uses the registers "r10",
20971 "r11", "r12" and/or "r13" and only provided that the normal use of the
20972 corresponding registers have been restricted via the -ffixed-reg or
20973 -mint-register command-line options.
20974 S/390 and zSeries Options
20976 These are the -m options defined for the S/390 and zSeries
20977 architecture.
20978 -mhard-float
20980 -msoft-float
20981 Use (do not use) the hardware floating-point instructions and
20982 registers for floating-point operations. When -msoft-float is
20983 specified, functions in libgcc.a are used to perform floating-point
20984 operations. When -mhard-float is specified, the compiler generates
20985 IEEE floating-point instructions. This is the default.
20986 -mhard-dfp
20988 -mno-hard-dfp
20989 Use (do not use) the hardware decimal-floating-point instructions
20990 for decimal-floating-point operations. When -mno-hard-dfp is
20991 specified, functions in libgcc.a are used to perform decimal-
20992 floating-point operations. When -mhard-dfp is specified, the
20993 compiler generates decimal-floating-point hardware instructions.
20994 This is the default for -march=z9-ec or higher.
20995 -mlong-double-64
20997 -mlong-double-128
20998 These switches control the size of "long double" type. A size of 64
20999 bits makes the "long double" type equivalent to the "double" type.
21000 This is the default.
21001 -mbackchain
21003 -mno-backchain
21004 Store (do not store) the address of the caller's frame as backchain
21005 pointer into the callee's stack frame. A backchain may be needed
21006 to allow debugging using tools that do not understand DWARF call
21007 frame information. When -mno-packed-stack is in effect, the
21008 backchain pointer is stored at the bottom of the stack frame; when
21009 -mpacked-stack is in effect, the backchain is placed into the
21010 topmost word of the 96/160 byte register save area.
21011 In general, code compiled with -mbackchain is call-compatible with
21013 code compiled with -mmo-backchain; however, use of the backchain
21014 for debugging purposes usually requires that the whole binary is
21015 built with -mbackchain. Note that the combination of -mbackchain,
21016 -mpacked-stack and -mhard-float is not supported. In order to
21017 build a linux kernel use -msoft-float.
21018 The default is to not maintain the backchain.
21020 -mpacked-stack
21022 -mno-packed-stack
21023 Use (do not use) the packed stack layout. When -mno-packed-stack
21024 is specified, the compiler uses the all fields of the 96/160 byte
21025 register save area only for their default purpose; unused fields
21026 still take up stack space. When -mpacked-stack is specified,
21027 register save slots are densely packed at the top of the register
21028 save area; unused space is reused for other purposes, allowing for
21029 more efficient use of the available stack space. However, when
21030 -mbackchain is also in effect, the topmost word of the save area is
21031 always used to store the backchain, and the return address register
21032 is always saved two words below the backchain.
21033 As long as the stack frame backchain is not used, code generated
21035 with -mpacked-stack is call-compatible with code generated with
21036 -mno-packed-stack. Note that some non-FSF releases of GCC 2.95 for
21037 S/390 or zSeries generated code that uses the stack frame backchain
21038 at run time, not just for debugging purposes. Such code is not
21039 call-compatible with code compiled with -mpacked-stack. Also, note
21040 that the combination of -mbackchain, -mpacked-stack and
21041 -mhard-float is not supported. In order to build a linux kernel
21042 use -msoft-float.
21043 The default is to not use the packed stack layout.
21045 -msmall-exec
21047 -mno-small-exec
21048 Generate (or do not generate) code using the "bras" instruction to
21049 do subroutine calls. This only works reliably if the total
21050 executable size does not exceed 64k. The default is to use the
21051 "basr" instruction instead, which does not have this limitation.
21052 -m64
21054 -m31
21055 When -m31 is specified, generate code compliant to the GNU/Linux
21056 for S/390 ABI. When -m64 is specified, generate code compliant to
21057 the GNU/Linux for zSeries ABI. This allows GCC in particular to
21058 generate 64-bit instructions. For the s390 targets, the default is
21059 -m31, while the s390x targets default to -m64.
21060 -mzarch
21062 -mesa
21063 When -mzarch is specified, generate code using the instructions
21064 available on z/Architecture. When -mesa is specified, generate
21065 code using the instructions available on ESA/390. Note that -mesa
21066 is not possible with -m64. When generating code compliant to the
21067 GNU/Linux for S/390 ABI, the default is -mesa. When generating
21068 code compliant to the GNU/Linux for zSeries ABI, the default is
21069 -mzarch.
21070 -mhtm
21072 -mno-htm
21073 The -mhtm option enables a set of builtins making use of
21074 instructions available with the transactional execution facility
21075 introduced with the IBM zEnterprise EC12 machine generation S/390
21076 System z Built-in Functions. -mhtm is enabled by default when
21077 using -march=zEC12.
21078 -mvx
21080 -mno-vx
21081 When -mvx is specified, generate code using the instructions
21082 available with the vector extension facility introduced with the
21083 IBM z13 machine generation. This option changes the ABI for some
21084 vector type values with regard to alignment and calling
21085 conventions. In case vector type values are being used in an ABI-
21086 relevant context a GAS .gnu_attribute command will be added to mark
21087 the resulting binary with the ABI used. -mvx is enabled by default
21088 when using -march=z13.
21089 -mzvector
21091 -mno-zvector
21092 The -mzvector option enables vector language extensions and
21093 builtins using instructions available with the vector extension
21094 facility introduced with the IBM z13 machine generation. This
21095 option adds support for vector to be used as a keyword to define
21096 vector type variables and arguments. vector is only available when
21097 GNU extensions are enabled. It will not be expanded when
21098 requesting strict standard compliance e.g. with -std=c99. In
21099 addition to the GCC low-level builtins -mzvector enables a set of
21100 builtins added for compatibility with AltiVec-style implementations
21101 like Power and Cell. In order to make use of these builtins the
21102 header file vecintrin.h needs to be included. -mzvector is
21103 disabled by default.
21104 -mmvcle
21106 -mno-mvcle
21107 Generate (or do not generate) code using the "mvcle" instruction to
21108 perform block moves. When -mno-mvcle is specified, use a "mvc"
21109 loop instead. This is the default unless optimizing for size.
21110 -mdebug
21112 -mno-debug
21113 Print (or do not print) additional debug information when
21114 compiling. The default is to not print debug information.
21115 -march=cpu-type
21117 Generate code that runs on cpu-type, which is the name of a system
21118 representing a certain processor type. Possible values for cpu-
21119 type are z900/arch5, z990/arch6, z9-109, z9-ec/arch7, z10/arch8,
21120 z196/arch9, zEC12, z13/arch11, and native.
21121 The default is -march=z900. g5/arch3 and g6 are deprecated and
21123 will be removed with future releases.
21124 Specifying native as cpu type can be used to select the best
21126 architecture option for the host processor. -march=native has no
21127 effect if GCC does not recognize the processor.
21128 -mtune=cpu-type
21130 Tune to cpu-type everything applicable about the generated code,
21131 except for the ABI and the set of available instructions. The list
21132 of cpu-type values is the same as for -march. The default is the
21133 value used for -march.
21134 -mtpf-trace
21136 -mno-tpf-trace
21137 Generate code that adds (does not add) in TPF OS specific branches
21138 to trace routines in the operating system. This option is off by
21139 default, even when compiling for the TPF OS.
21140 -mfused-madd
21142 -mno-fused-madd
21143 Generate code that uses (does not use) the floating-point multiply
21144 and accumulate instructions. These instructions are generated by
21145 default if hardware floating point is used.
21146 -mwarn-framesize=framesize
21148 Emit a warning if the current function exceeds the given frame
21149 size. Because this is a compile-time check it doesn't need to be a
21150 real problem when the program runs. It is intended to identify
21151 functions that most probably cause a stack overflow. It is useful
21152 to be used in an environment with limited stack size e.g. the linux
21153 kernel.
21154 -mwarn-dynamicstack
21156 Emit a warning if the function calls "alloca" or uses dynamically-
21157 sized arrays. This is generally a bad idea with a limited stack
21158 size.
21159 -mstack-guard=stack-guard
21161 -mstack-size=stack-size
21162 If these options are provided the S/390 back end emits additional
21163 instructions in the function prologue that trigger a trap if the
21164 stack size is stack-guard bytes above the stack-size (remember that
21165 the stack on S/390 grows downward). If the stack-guard option is
21166 omitted the smallest power of 2 larger than the frame size of the
21167 compiled function is chosen. These options are intended to be used
21168 to help debugging stack overflow problems. The additionally
21169 emitted code causes only little overhead and hence can also be used
21170 in production-like systems without greater performance degradation.
21171 The given values have to be exact powers of 2 and stack-size has to
21172 be greater than stack-guard without exceeding 64k. In order to be
21173 efficient the extra code makes the assumption that the stack starts
21174 at an address aligned to the value given by stack-size. The stack-
21175 guard option can only be used in conjunction with stack-size.
21176 -mhotpatch=pre-halfwords,post-halfwords
21178 If the hotpatch option is enabled, a "hot-patching" function
21179 prologue is generated for all functions in the compilation unit.
21180 The funtion label is prepended with the given number of two-byte
21181 NOP instructions (pre-halfwords, maximum 1000000). After the
21182 label, 2 * post-halfwords bytes are appended, using the largest NOP
21183 like instructions the architecture allows (maximum 1000000).
21184 If both arguments are zero, hotpatching is disabled.
21186 This option can be overridden for individual functions with the
21188 "hotpatch" attribute.
21189 Score Options
21191 These options are defined for Score implementations:
21192 -meb
21194 Compile code for big-endian mode. This is the default.
21195 -mel
21197 Compile code for little-endian mode.
21198 -mnhwloop
21200 Disable generation of "bcnz" instructions.
21201 -muls
21203 Enable generation of unaligned load and store instructions.
21204 -mmac
21206 Enable the use of multiply-accumulate instructions. Disabled by
21207 default.
21208 -mscore5
21210 Specify the SCORE5 as the target architecture.
21211 -mscore5u
21213 Specify the SCORE5U of the target architecture.
21214 -mscore7
21216 Specify the SCORE7 as the target architecture. This is the default.
21217 -mscore7d
21219 Specify the SCORE7D as the target architecture.
21220 SH Options
21222 These -m options are defined for the SH implementations:
21223 -m1 Generate code for the SH1.
21225 -m2 Generate code for the SH2.
21227 -m2e
21229 Generate code for the SH2e.
21230 -m2a-nofpu
21232 Generate code for the SH2a without FPU, or for a SH2a-FPU in such a
21233 way that the floating-point unit is not used.
21234 -m2a-single-only
21236 Generate code for the SH2a-FPU, in such a way that no double-
21237 precision floating-point operations are used.
21238 -m2a-single
21240 Generate code for the SH2a-FPU assuming the floating-point unit is
21241 in single-precision mode by default.
21242 -m2a
21244 Generate code for the SH2a-FPU assuming the floating-point unit is
21245 in double-precision mode by default.
21246 -m3 Generate code for the SH3.
21248 -m3e
21250 Generate code for the SH3e.
21251 -m4-nofpu
21253 Generate code for the SH4 without a floating-point unit.
21254 -m4-single-only
21256 Generate code for the SH4 with a floating-point unit that only
21257 supports single-precision arithmetic.
21258 -m4-single
21260 Generate code for the SH4 assuming the floating-point unit is in
21261 single-precision mode by default.
21262 -m4 Generate code for the SH4.
21264 -m4-100
21266 Generate code for SH4-100.
21267 -m4-100-nofpu
21269 Generate code for SH4-100 in such a way that the floating-point
21270 unit is not used.
21271 -m4-100-single
21273 Generate code for SH4-100 assuming the floating-point unit is in
21274 single-precision mode by default.
21275 -m4-100-single-only
21277 Generate code for SH4-100 in such a way that no double-precision
21278 floating-point operations are used.
21279 -m4-200
21281 Generate code for SH4-200.
21282 -m4-200-nofpu
21284 Generate code for SH4-200 without in such a way that the floating-
21285 point unit is not used.
21286 -m4-200-single
21288 Generate code for SH4-200 assuming the floating-point unit is in
21289 single-precision mode by default.
21290 -m4-200-single-only
21292 Generate code for SH4-200 in such a way that no double-precision
21293 floating-point operations are used.
21294 -m4-300
21296 Generate code for SH4-300.
21297 -m4-300-nofpu
21299 Generate code for SH4-300 without in such a way that the floating-
21300 point unit is not used.
21301 -m4-300-single
21303 Generate code for SH4-300 in such a way that no double-precision
21304 floating-point operations are used.
21305 -m4-300-single-only
21307 Generate code for SH4-300 in such a way that no double-precision
21308 floating-point operations are used.
21309 -m4-340
21311 Generate code for SH4-340 (no MMU, no FPU).
21312 -m4-500
21314 Generate code for SH4-500 (no FPU). Passes -isa=sh4-nofpu to the
21315 assembler.
21316 -m4a-nofpu
21318 Generate code for the SH4al-dsp, or for a SH4a in such a way that
21319 the floating-point unit is not used.
21320 -m4a-single-only
21322 Generate code for the SH4a, in such a way that no double-precision
21323 floating-point operations are used.
21324 -m4a-single
21326 Generate code for the SH4a assuming the floating-point unit is in
21327 single-precision mode by default.
21328 -m4a
21330 Generate code for the SH4a.
21331 -m4al
21333 Same as -m4a-nofpu, except that it implicitly passes -dsp to the
21334 assembler. GCC doesn't generate any DSP instructions at the
21335 moment.
21336 -mb Compile code for the processor in big-endian mode.
21338 -ml Compile code for the processor in little-endian mode.
21340 -mdalign
21342 Align doubles at 64-bit boundaries. Note that this changes the
21343 calling conventions, and thus some functions from the standard C
21344 library do not work unless you recompile it first with -mdalign.
21345 -mrelax
21347 Shorten some address references at link time, when possible; uses
21348 the linker option -relax.
21349 -mbigtable
21351 Use 32-bit offsets in "switch" tables. The default is to use
21352 16-bit offsets.
21353 -mbitops
21355 Enable the use of bit manipulation instructions on SH2A.
21356 -mfmovd
21358 Enable the use of the instruction "fmovd". Check -mdalign for
21359 alignment constraints.
21360 -mrenesas
21362 Comply with the calling conventions defined by Renesas.
21363 -mno-renesas
21365 Comply with the calling conventions defined for GCC before the
21366 Renesas conventions were available. This option is the default for
21367 all targets of the SH toolchain.
21368 -mnomacsave
21370 Mark the "MAC" register as call-clobbered, even if -mrenesas is
21371 given.
21372 -mieee
21374 -mno-ieee
21375 Control the IEEE compliance of floating-point comparisons, which
21376 affects the handling of cases where the result of a comparison is
21377 unordered. By default -mieee is implicitly enabled. If
21378 -ffinite-math-only is enabled -mno-ieee is implicitly set, which
21379 results in faster floating-point greater-equal and less-equal
21380 comparisons. The implicit settings can be overridden by specifying
21381 either -mieee or -mno-ieee.
21382 -minline-ic_invalidate
21384 Inline code to invalidate instruction cache entries after setting
21385 up nested function trampolines. This option has no effect if
21386 -musermode is in effect and the selected code generation option
21387 (e.g. -m4) does not allow the use of the "icbi" instruction. If
21388 the selected code generation option does not allow the use of the
21389 "icbi" instruction, and -musermode is not in effect, the inlined
21390 code manipulates the instruction cache address array directly with
21391 an associative write. This not only requires privileged mode at
21392 run time, but it also fails if the cache line had been mapped via
21393 the TLB and has become unmapped.
21394 -misize
21396 Dump instruction size and location in the assembly code.
21397 -mpadstruct
21399 This option is deprecated. It pads structures to multiple of 4
21400 bytes, which is incompatible with the SH ABI.
21401 -matomic-model=model
21403 Sets the model of atomic operations and additional parameters as a
21404 comma separated list. For details on the atomic built-in functions
21405 see __atomic Builtins. The following models and parameters are
21406 supported:
21407 none
21409 Disable compiler generated atomic sequences and emit library
21410 calls for atomic operations. This is the default if the target
21411 is not "sh*-*-linux*".
21412 soft-gusa
21414 Generate GNU/Linux compatible gUSA software atomic sequences
21415 for the atomic built-in functions. The generated atomic
21416 sequences require additional support from the
21417 interrupt/exception handling code of the system and are only
21418 suitable for SH3* and SH4* single-core systems. This option is
21419 enabled by default when the target is "sh*-*-linux*" and SH3*
21420 or SH4*. When the target is SH4A, this option also partially
21421 utilizes the hardware atomic instructions "movli.l" and
21422 "movco.l" to create more efficient code, unless strict is
21423 specified.
21424 soft-tcb
21426 Generate software atomic sequences that use a variable in the
21427 thread control block. This is a variation of the gUSA
21428 sequences which can also be used on SH1* and SH2* targets. The
21429 generated atomic sequences require additional support from the
21430 interrupt/exception handling code of the system and are only
21431 suitable for single-core systems. When using this model, the
21432 gbr-offset= parameter has to be specified as well.
21433 soft-imask
21435 Generate software atomic sequences that temporarily disable
21436 interrupts by setting "SR.IMASK = 1111". This model works only
21437 when the program runs in privileged mode and is only suitable
21438 for single-core systems. Additional support from the
21439 interrupt/exception handling code of the system is not
21440 required. This model is enabled by default when the target is
21441 "sh*-*-linux*" and SH1* or SH2*.
21442 hard-llcs
21444 Generate hardware atomic sequences using the "movli.l" and
21445 "movco.l" instructions only. This is only available on SH4A
21446 and is suitable for multi-core systems. Since the hardware
21447 instructions support only 32 bit atomic variables access to 8
21448 or 16 bit variables is emulated with 32 bit accesses. Code
21449 compiled with this option is also compatible with other
21450 software atomic model interrupt/exception handling systems if
21451 executed on an SH4A system. Additional support from the
21452 interrupt/exception handling code of the system is not required
21453 for this model.
21454 gbr-offset=
21456 This parameter specifies the offset in bytes of the variable in
21457 the thread control block structure that should be used by the
21458 generated atomic sequences when the soft-tcb model has been
21459 selected. For other models this parameter is ignored. The
21460 specified value must be an integer multiple of four and in the
21461 range 0-1020.
21462 strict
21464 This parameter prevents mixed usage of multiple atomic models,
21465 even if they are compatible, and makes the compiler generate
21466 atomic sequences of the specified model only.
21467 -mtas
21469 Generate the "tas.b" opcode for "__atomic_test_and_set". Notice
21470 that depending on the particular hardware and software
21471 configuration this can degrade overall performance due to the
21472 operand cache line flushes that are implied by the "tas.b"
21473 instruction. On multi-core SH4A processors the "tas.b" instruction
21474 must be used with caution since it can result in data corruption
21475 for certain cache configurations.
21476 -mprefergot
21478 When generating position-independent code, emit function calls
21479 using the Global Offset Table instead of the Procedure Linkage
21480 Table.
21481 -musermode
21483 -mno-usermode
21484 Don't allow (allow) the compiler generating privileged mode code.
21485 Specifying -musermode also implies -mno-inline-ic_invalidate if the
21486 inlined code would not work in user mode. -musermode is the
21487 default when the target is "sh*-*-linux*". If the target is SH1*
21488 or SH2* -musermode has no effect, since there is no user mode.
21489 -multcost=number
21491 Set the cost to assume for a multiply insn.
21492 -mdiv=strategy
21494 Set the division strategy to be used for integer division
21495 operations. strategy can be one of:
21496 call-div1
21498 Calls a library function that uses the single-step division
21499 instruction "div1" to perform the operation. Division by zero
21500 calculates an unspecified result and does not trap. This is
21501 the default except for SH4, SH2A and SHcompact.
21502 call-fp
21504 Calls a library function that performs the operation in double
21505 precision floating point. Division by zero causes a floating-
21506 point exception. This is the default for SHcompact with FPU.
21507 Specifying this for targets that do not have a double precision
21508 FPU defaults to "call-div1".
21509 call-table
21511 Calls a library function that uses a lookup table for small
21512 divisors and the "div1" instruction with case distinction for
21513 larger divisors. Division by zero calculates an unspecified
21514 result and does not trap. This is the default for SH4.
21515 Specifying this for targets that do not have dynamic shift
21516 instructions defaults to "call-div1".
21517 When a division strategy has not been specified the default
21519 strategy is selected based on the current target. For SH2A the
21520 default strategy is to use the "divs" and "divu" instructions
21521 instead of library function calls.
21522 -maccumulate-outgoing-args
21524 Reserve space once for outgoing arguments in the function prologue
21525 rather than around each call. Generally beneficial for performance
21526 and size. Also needed for unwinding to avoid changing the stack
21527 frame around conditional code.
21528 -mdivsi3_libfunc=name
21530 Set the name of the library function used for 32-bit signed
21531 division to name. This only affects the name used in the call
21532 division strategies, and the compiler still expects the same sets
21533 of input/output/clobbered registers as if this option were not
21534 present.
21535 -mfixed-range=register-range
21537 Generate code treating the given register range as fixed registers.
21538 A fixed register is one that the register allocator can not use.
21539 This is useful when compiling kernel code. A register range is
21540 specified as two registers separated by a dash. Multiple register
21541 ranges can be specified separated by a comma.
21542 -mbranch-cost=num
21544 Assume num to be the cost for a branch instruction. Higher numbers
21545 make the compiler try to generate more branch-free code if
21546 possible. If not specified the value is selected depending on the
21547 processor type that is being compiled for.
21548 -mzdcbranch
21550 -mno-zdcbranch
21551 Assume (do not assume) that zero displacement conditional branch
21552 instructions "bt" and "bf" are fast. If -mzdcbranch is specified,
21553 the compiler prefers zero displacement branch code sequences. This
21554 is enabled by default when generating code for SH4 and SH4A. It
21555 can be explicitly disabled by specifying -mno-zdcbranch.
21556 -mcbranch-force-delay-slot
21558 Force the usage of delay slots for conditional branches, which
21559 stuffs the delay slot with a "nop" if a suitable instruction cannot
21560 be found. By default this option is disabled. It can be enabled
21561 to work around hardware bugs as found in the original SH7055.
21562 -mfused-madd
21564 -mno-fused-madd
21565 Generate code that uses (does not use) the floating-point multiply
21566 and accumulate instructions. These instructions are generated by
21567 default if hardware floating point is used. The machine-dependent
21568 -mfused-madd option is now mapped to the machine-independent
21569 -ffp-contract=fast option, and -mno-fused-madd is mapped to
21570 -ffp-contract=off.
21571 -mfsca
21573 -mno-fsca
21574 Allow or disallow the compiler to emit the "fsca" instruction for
21575 sine and cosine approximations. The option -mfsca must be used in
21576 combination with -funsafe-math-optimizations. It is enabled by
21577 default when generating code for SH4A. Using -mno-fsca disables
21578 sine and cosine approximations even if -funsafe-math-optimizations
21579 is in effect.
21580 -mfsrra
21582 -mno-fsrra
21583 Allow or disallow the compiler to emit the "fsrra" instruction for
21584 reciprocal square root approximations. The option -mfsrra must be
21585 used in combination with -funsafe-math-optimizations and
21586 -ffinite-math-only. It is enabled by default when generating code
21587 for SH4A. Using -mno-fsrra disables reciprocal square root
21588 approximations even if -funsafe-math-optimizations and
21589 -ffinite-math-only are in effect.
21590 -mpretend-cmove
21592 Prefer zero-displacement conditional branches for conditional move
21593 instruction patterns. This can result in faster code on the SH4
21594 processor.
21595 -mfdpic
21597 Generate code using the FDPIC ABI.
21598 Solaris 2 Options
21600 These -m options are supported on Solaris 2:
21601 -mclear-hwcap
21603 -mclear-hwcap tells the compiler to remove the hardware
21604 capabilities generated by the Solaris assembler. This is only
21605 necessary when object files use ISA extensions not supported by the
21606 current machine, but check at runtime whether or not to use them.
21607 -mimpure-text
21609 -mimpure-text, used in addition to -shared, tells the compiler to
21610 not pass -z text to the linker when linking a shared object. Using
21611 this option, you can link position-dependent code into a shared
21612 object.
21613 -mimpure-text suppresses the "relocations remain against
21615 allocatable but non-writable sections" linker error message.
21616 However, the necessary relocations trigger copy-on-write, and the
21617 shared object is not actually shared across processes. Instead of
21618 using -mimpure-text, you should compile all source code with -fpic
21619 or -fPIC.
21620 These switches are supported in addition to the above on Solaris 2:
21622 -pthreads
21624 This is a synonym for -pthread.
21625 SPARC Options
21627 These -m options are supported on the SPARC:
21628 -mno-app-regs
21630 -mapp-regs
21631 Specify -mapp-regs to generate output using the global registers 2
21632 through 4, which the SPARC SVR4 ABI reserves for applications.
21633 Like the global register 1, each global register 2 through 4 is
21634 then treated as an allocable register that is clobbered by function
21635 calls. This is the default.
21636 To be fully SVR4 ABI-compliant at the cost of some performance
21638 loss, specify -mno-app-regs. You should compile libraries and
21639 system software with this option.
21640 -mflat
21642 -mno-flat
21643 With -mflat, the compiler does not generate save/restore
21644 instructions and uses a "flat" or single register window model.
21645 This model is compatible with the regular register window model.
21646 The local registers and the input registers (0--5) are still
21647 treated as "call-saved" registers and are saved on the stack as
21648 needed.
21649 With -mno-flat (the default), the compiler generates save/restore
21651 instructions (except for leaf functions). This is the normal
21652 operating mode.
21653 -mfpu
21655 -mhard-float
21656 Generate output containing floating-point instructions. This is
21657 the default.
21658 -mno-fpu
21660 -msoft-float
21661 Generate output containing library calls for floating point.
21662 Warning: the requisite libraries are not available for all SPARC
21663 targets. Normally the facilities of the machine's usual C compiler
21664 are used, but this cannot be done directly in cross-compilation.
21665 You must make your own arrangements to provide suitable library
21666 functions for cross-compilation. The embedded targets sparc-*-aout
21667 and sparclite-*-* do provide software floating-point support.
21668 -msoft-float changes the calling convention in the output file;
21670 therefore, it is only useful if you compile all of a program with
21671 this option. In particular, you need to compile libgcc.a, the
21672 library that comes with GCC, with -msoft-float in order for this to
21673 work.
21674 -mhard-quad-float
21676 Generate output containing quad-word (long double) floating-point
21677 instructions.
21678 -msoft-quad-float
21680 Generate output containing library calls for quad-word (long
21681 double) floating-point instructions. The functions called are
21682 those specified in the SPARC ABI. This is the default.
21683 As of this writing, there are no SPARC implementations that have
21685 hardware support for the quad-word floating-point instructions.
21686 They all invoke a trap handler for one of these instructions, and
21687 then the trap handler emulates the effect of the instruction.
21688 Because of the trap handler overhead, this is much slower than
21689 calling the ABI library routines. Thus the -msoft-quad-float
21690 option is the default.
21691 -mno-unaligned-doubles
21693 -munaligned-doubles
21694 Assume that doubles have 8-byte alignment. This is the default.
21695 With -munaligned-doubles, GCC assumes that doubles have 8-byte
21697 alignment only if they are contained in another type, or if they
21698 have an absolute address. Otherwise, it assumes they have 4-byte
21699 alignment. Specifying this option avoids some rare compatibility
21700 problems with code generated by other compilers. It is not the
21701 default because it results in a performance loss, especially for
21702 floating-point code.
21703 -muser-mode
21705 -mno-user-mode
21706 Do not generate code that can only run in supervisor mode. This is
21707 relevant only for the "casa" instruction emitted for the LEON3
21708 processor. This is the default.
21709 -mfaster-structs
21711 -mno-faster-structs
21712 With -mfaster-structs, the compiler assumes that structures should
21713 have 8-byte alignment. This enables the use of pairs of "ldd" and
21714 "std" instructions for copies in structure assignment, in place of
21715 twice as many "ld" and "st" pairs. However, the use of this
21716 changed alignment directly violates the SPARC ABI. Thus, it's
21717 intended only for use on targets where the developer acknowledges
21718 that their resulting code is not directly in line with the rules of
21719 the ABI.
21720 -mstd-struct-return
21722 -mno-std-struct-return
21723 With -mstd-struct-return, the compiler generates checking code in
21724 functions returning structures or unions to detect size mismatches
21725 between the two sides of function calls, as per the 32-bit ABI.
21726 The default is -mno-std-struct-return. This option has no effect
21728 in 64-bit mode.
21729 -mlra
21731 -mno-lra
21732 Enable Local Register Allocation. This is the default for SPARC
21733 since GCC 7 so -mno-lra needs to be passed to get old Reload.
21734 -mcpu=cpu_type
21736 Set the instruction set, register set, and instruction scheduling
21737 parameters for machine type cpu_type. Supported values for
21738 cpu_type are v7, cypress, v8, supersparc, hypersparc, leon, leon3,
21739 leon3v7, sparclite, f930, f934, sparclite86x, sparclet, tsc701, v9,
21740 ultrasparc, ultrasparc3, niagara, niagara2, niagara3, niagara4,
21741 niagara7 and m8.
21742 Native Solaris and GNU/Linux toolchains also support the value
21744 native, which selects the best architecture option for the host
21745 processor. -mcpu=native has no effect if GCC does not recognize
21746 the processor.
21747 Default instruction scheduling parameters are used for values that
21749 select an architecture and not an implementation. These are v7,
21750 v8, sparclite, sparclet, v9.
21751 Here is a list of each supported architecture and their supported
21753 implementations.
21754 v7 cypress, leon3v7
21756 v8 supersparc, hypersparc, leon, leon3
21758 sparclite
21760 f930, f934, sparclite86x
21761 sparclet
21763 tsc701
21764 v9 ultrasparc, ultrasparc3, niagara, niagara2, niagara3, niagara4,
21766 niagara7, m8
21767 By default (unless configured otherwise), GCC generates code for
21769 the V7 variant of the SPARC architecture. With -mcpu=cypress, the
21770 compiler additionally optimizes it for the Cypress CY7C602 chip, as
21771 used in the SPARCStation/SPARCServer 3xx series. This is also
21772 appropriate for the older SPARCStation 1, 2, IPX etc.
21773 With -mcpu=v8, GCC generates code for the V8 variant of the SPARC
21775 architecture. The only difference from V7 code is that the
21776 compiler emits the integer multiply and integer divide instructions
21777 which exist in SPARC-V8 but not in SPARC-V7. With
21778 -mcpu=supersparc, the compiler additionally optimizes it for the
21779 SuperSPARC chip, as used in the SPARCStation 10, 1000 and 2000
21780 series.
21781 With -mcpu=sparclite, GCC generates code for the SPARClite variant
21783 of the SPARC architecture. This adds the integer multiply, integer
21784 divide step and scan ("ffs") instructions which exist in SPARClite
21785 but not in SPARC-V7. With -mcpu=f930, the compiler additionally
21786 optimizes it for the Fujitsu MB86930 chip, which is the original
21787 SPARClite, with no FPU. With -mcpu=f934, the compiler additionally
21788 optimizes it for the Fujitsu MB86934 chip, which is the more recent
21789 SPARClite with FPU.
21790 With -mcpu=sparclet, GCC generates code for the SPARClet variant of
21792 the SPARC architecture. This adds the integer multiply,
21793 multiply/accumulate, integer divide step and scan ("ffs")
21794 instructions which exist in SPARClet but not in SPARC-V7. With
21795 -mcpu=tsc701, the compiler additionally optimizes it for the TEMIC
21796 SPARClet chip.
21797 With -mcpu=v9, GCC generates code for the V9 variant of the SPARC
21799 architecture. This adds 64-bit integer and floating-point move
21800 instructions, 3 additional floating-point condition code registers
21801 and conditional move instructions. With -mcpu=ultrasparc, the
21802 compiler additionally optimizes it for the Sun UltraSPARC I/II/IIi
21803 chips. With -mcpu=ultrasparc3, the compiler additionally optimizes
21804 it for the Sun UltraSPARC III/III+/IIIi/IIIi+/IV/IV+ chips. With
21805 -mcpu=niagara, the compiler additionally optimizes it for Sun
21806 UltraSPARC T1 chips. With -mcpu=niagara2, the compiler
21807 additionally optimizes it for Sun UltraSPARC T2 chips. With
21808 -mcpu=niagara3, the compiler additionally optimizes it for Sun
21809 UltraSPARC T3 chips. With -mcpu=niagara4, the compiler
21810 additionally optimizes it for Sun UltraSPARC T4 chips. With
21811 -mcpu=niagara7, the compiler additionally optimizes it for Oracle
21812 SPARC M7 chips. With -mcpu=m8, the compiler additionally optimizes
21813 it for Oracle M8 chips.
21814 -mtune=cpu_type
21816 Set the instruction scheduling parameters for machine type
21817 cpu_type, but do not set the instruction set or register set that
21818 the option -mcpu=cpu_type does.
21819 The same values for -mcpu=cpu_type can be used for -mtune=cpu_type,
21821 but the only useful values are those that select a particular CPU
21822 implementation. Those are cypress, supersparc, hypersparc, leon,
21823 leon3, leon3v7, f930, f934, sparclite86x, tsc701, ultrasparc,
21824 ultrasparc3, niagara, niagara2, niagara3, niagara4, niagara7 and
21825 m8. With native Solaris and GNU/Linux toolchains, native can also
21826 be used.
21827 -mv8plus
21829 -mno-v8plus
21830 With -mv8plus, GCC generates code for the SPARC-V8+ ABI. The
21831 difference from the V8 ABI is that the global and out registers are
21832 considered 64 bits wide. This is enabled by default on Solaris in
21833 32-bit mode for all SPARC-V9 processors.
21834 -mvis
21836 -mno-vis
21837 With -mvis, GCC generates code that takes advantage of the
21838 UltraSPARC Visual Instruction Set extensions. The default is
21839 -mno-vis.
21840 -mvis2
21842 -mno-vis2
21843 With -mvis2, GCC generates code that takes advantage of version 2.0
21844 of the UltraSPARC Visual Instruction Set extensions. The default
21845 is -mvis2 when targeting a cpu that supports such instructions,
21846 such as UltraSPARC-III and later. Setting -mvis2 also sets -mvis.
21847 -mvis3
21849 -mno-vis3
21850 With -mvis3, GCC generates code that takes advantage of version 3.0
21851 of the UltraSPARC Visual Instruction Set extensions. The default
21852 is -mvis3 when targeting a cpu that supports such instructions,
21853 such as niagara-3 and later. Setting -mvis3 also sets -mvis2 and
21854 -mvis.
21855 -mvis4
21857 -mno-vis4
21858 With -mvis4, GCC generates code that takes advantage of version 4.0
21859 of the UltraSPARC Visual Instruction Set extensions. The default
21860 is -mvis4 when targeting a cpu that supports such instructions,
21861 such as niagara-7 and later. Setting -mvis4 also sets -mvis3,
21862 -mvis2 and -mvis.
21863 -mvis4b
21865 -mno-vis4b
21866 With -mvis4b, GCC generates code that takes advantage of version
21867 4.0 of the UltraSPARC Visual Instruction Set extensions, plus the
21868 additional VIS instructions introduced in the Oracle SPARC
21869 Architecture 2017. The default is -mvis4b when targeting a cpu
21870 that supports such instructions, such as m8 and later. Setting
21871 -mvis4b also sets -mvis4, -mvis3, -mvis2 and -mvis.
21872 -mcbcond
21874 -mno-cbcond
21875 With -mcbcond, GCC generates code that takes advantage of the
21876 UltraSPARC Compare-and-Branch-on-Condition instructions. The
21877 default is -mcbcond when targeting a CPU that supports such
21878 instructions, such as Niagara-4 and later.
21879 -mfmaf
21881 -mno-fmaf
21882 With -mfmaf, GCC generates code that takes advantage of the
21883 UltraSPARC Fused Multiply-Add Floating-point instructions. The
21884 default is -mfmaf when targeting a CPU that supports such
21885 instructions, such as Niagara-3 and later.
21886 -mfsmuld
21888 -mno-fsmuld
21889 With -mfsmuld, GCC generates code that takes advantage of the
21890 Floating-point Multiply Single to Double (FsMULd) instruction. The
21891 default is -mfsmuld when targeting a CPU supporting the
21892 architecture versions V8 or V9 with FPU except -mcpu=leon.
21893 -mpopc
21895 -mno-popc
21896 With -mpopc, GCC generates code that takes advantage of the
21897 UltraSPARC Population Count instruction. The default is -mpopc
21898 when targeting a CPU that supports such an instruction, such as
21899 Niagara-2 and later.
21900 -msubxc
21902 -mno-subxc
21903 With -msubxc, GCC generates code that takes advantage of the
21904 UltraSPARC Subtract-Extended-with-Carry instruction. The default
21905 is -msubxc when targeting a CPU that supports such an instruction,
21906 such as Niagara-7 and later.
21907 -mfix-at697f
21909 Enable the documented workaround for the single erratum of the
21910 Atmel AT697F processor (which corresponds to erratum #13 of the
21911 AT697E processor).
21912 -mfix-ut699
21914 Enable the documented workarounds for the floating-point errata and
21915 the data cache nullify errata of the UT699 processor.
21916 -mfix-ut700
21918 Enable the documented workaround for the back-to-back store errata
21919 of the UT699E/UT700 processor.
21920 -mfix-gr712rc
21922 Enable the documented workaround for the back-to-back store errata
21923 of the GR712RC processor.
21924 These -m options are supported in addition to the above on SPARC-V9
21926 processors in 64-bit environments:
21927 -m32
21929 -m64
21930 Generate code for a 32-bit or 64-bit environment. The 32-bit
21931 environment sets int, long and pointer to 32 bits. The 64-bit
21932 environment sets int to 32 bits and long and pointer to 64 bits.
21933 -mcmodel=which
21935 Set the code model to one of
21936 medlow
21938 The Medium/Low code model: 64-bit addresses, programs must be
21939 linked in the low 32 bits of memory. Programs can be
21940 statically or dynamically linked.
21941 medmid
21943 The Medium/Middle code model: 64-bit addresses, programs must
21944 be linked in the low 44 bits of memory, the text and data
21945 segments must be less than 2GB in size and the data segment
21946 must be located within 2GB of the text segment.
21947 medany
21949 The Medium/Anywhere code model: 64-bit addresses, programs may
21950 be linked anywhere in memory, the text and data segments must
21951 be less than 2GB in size and the data segment must be located
21952 within 2GB of the text segment.
21953 embmedany
21955 The Medium/Anywhere code model for embedded systems: 64-bit
21956 addresses, the text and data segments must be less than 2GB in
21957 size, both starting anywhere in memory (determined at link
21958 time). The global register %g4 points to the base of the data
21959 segment. Programs are statically linked and PIC is not
21960 supported.
21961 -mmemory-model=mem-model
21963 Set the memory model in force on the processor to one of
21964 default
21966 The default memory model for the processor and operating
21967 system.
21968 rmo Relaxed Memory Order
21970 pso Partial Store Order
21972 tso Total Store Order
21974 sc Sequential Consistency
21976 These memory models are formally defined in Appendix D of the
21978 SPARC-V9 architecture manual, as set in the processor's "PSTATE.MM"
21979 field.
21980 -mstack-bias
21982 -mno-stack-bias
21983 With -mstack-bias, GCC assumes that the stack pointer, and frame
21984 pointer if present, are offset by -2047 which must be added back
21985 when making stack frame references. This is the default in 64-bit
21986 mode. Otherwise, assume no such offset is present.
21987 SPU Options
21989 These -m options are supported on the SPU:
21990 -mwarn-reloc
21992 -merror-reloc
21993 The loader for SPU does not handle dynamic relocations. By
21994 default, GCC gives an error when it generates code that requires a
21995 dynamic relocation. -mno-error-reloc disables the error,
21996 -mwarn-reloc generates a warning instead.
21997 -msafe-dma
21999 -munsafe-dma
22000 Instructions that initiate or test completion of DMA must not be
22001 reordered with respect to loads and stores of the memory that is
22002 being accessed. With -munsafe-dma you must use the "volatile"
22003 keyword to protect memory accesses, but that can lead to
22004 inefficient code in places where the memory is known to not change.
22005 Rather than mark the memory as volatile, you can use -msafe-dma to
22006 tell the compiler to treat the DMA instructions as potentially
22007 affecting all memory.
22008 -mbranch-hints
22010 By default, GCC generates a branch hint instruction to avoid
22011 pipeline stalls for always-taken or probably-taken branches. A
22012 hint is not generated closer than 8 instructions away from its
22013 branch. There is little reason to disable them, except for
22014 debugging purposes, or to make an object a little bit smaller.
22015 -msmall-mem
22017 -mlarge-mem
22018 By default, GCC generates code assuming that addresses are never
22019 larger than 18 bits. With -mlarge-mem code is generated that
22020 assumes a full 32-bit address.
22021 -mstdmain
22023 By default, GCC links against startup code that assumes the SPU-
22024 style main function interface (which has an unconventional
22025 parameter list). With -mstdmain, GCC links your program against
22026 startup code that assumes a C99-style interface to "main",
22027 including a local copy of "argv" strings.
22028 -mfixed-range=register-range
22030 Generate code treating the given register range as fixed registers.
22031 A fixed register is one that the register allocator cannot use.
22032 This is useful when compiling kernel code. A register range is
22033 specified as two registers separated by a dash. Multiple register
22034 ranges can be specified separated by a comma.
22035 -mea32
22037 -mea64
22038 Compile code assuming that pointers to the PPU address space
22039 accessed via the "__ea" named address space qualifier are either 32
22040 or 64 bits wide. The default is 32 bits. As this is an ABI-
22041 changing option, all object code in an executable must be compiled
22042 with the same setting.
22043 -maddress-space-conversion
22045 -mno-address-space-conversion
22046 Allow/disallow treating the "__ea" address space as superset of the
22047 generic address space. This enables explicit type casts between
22048 "__ea" and generic pointer as well as implicit conversions of
22049 generic pointers to "__ea" pointers. The default is to allow
22050 address space pointer conversions.
22051 -mcache-size=cache-size
22053 This option controls the version of libgcc that the compiler links
22054 to an executable and selects a software-managed cache for accessing
22055 variables in the "__ea" address space with a particular cache size.
22056 Possible options for cache-size are 8, 16, 32, 64 and 128. The
22057 default cache size is 64KB.
22058 -matomic-updates
22060 -mno-atomic-updates
22061 This option controls the version of libgcc that the compiler links
22062 to an executable and selects whether atomic updates to the
22063 software-managed cache of PPU-side variables are used. If you use
22064 atomic updates, changes to a PPU variable from SPU code using the
22065 "__ea" named address space qualifier do not interfere with changes
22066 to other PPU variables residing in the same cache line from PPU
22067 code. If you do not use atomic updates, such interference may
22068 occur; however, writing back cache lines is more efficient. The
22069 default behavior is to use atomic updates.
22070 -mdual-nops
22072 -mdual-nops=n
22073 By default, GCC inserts NOPs to increase dual issue when it expects
22074 it to increase performance. n can be a value from 0 to 10. A
22075 smaller n inserts fewer NOPs. 10 is the default, 0 is the same as
22076 -mno-dual-nops. Disabled with -Os.
22077 -mhint-max-nops=n
22079 Maximum number of NOPs to insert for a branch hint. A branch hint
22080 must be at least 8 instructions away from the branch it is
22081 affecting. GCC inserts up to n NOPs to enforce this, otherwise it
22082 does not generate the branch hint.
22083 -mhint-max-distance=n
22085 The encoding of the branch hint instruction limits the hint to be
22086 within 256 instructions of the branch it is affecting. By default,
22087 GCC makes sure it is within 125.
22088 -msafe-hints
22090 Work around a hardware bug that causes the SPU to stall
22091 indefinitely. By default, GCC inserts the "hbrp" instruction to
22092 make sure this stall won't happen.
22093 Options for System V
22095 These additional options are available on System V Release 4 for
22096 compatibility with other compilers on those systems:
22097 -G Create a shared object. It is recommended that -symbolic or
22099 -shared be used instead.
22100 -Qy Identify the versions of each tool used by the compiler, in a
22102 ".ident" assembler directive in the output.
22103 -Qn Refrain from adding ".ident" directives to the output file (this is
22105 the default).
22106 -YP,dirs
22108 Search the directories dirs, and no others, for libraries specified
22109 with -l.
22110 -Ym,dir
22112 Look in the directory dir to find the M4 preprocessor. The
22113 assembler uses this option.
22114 TILE-Gx Options
22116 These -m options are supported on the TILE-Gx:
22117 -mcmodel=small
22119 Generate code for the small model. The distance for direct calls
22120 is limited to 500M in either direction. PC-relative addresses are
22121 32 bits. Absolute addresses support the full address range.
22122 -mcmodel=large
22124 Generate code for the large model. There is no limitation on call
22125 distance, pc-relative addresses, or absolute addresses.
22126 -mcpu=name
22128 Selects the type of CPU to be targeted. Currently the only
22129 supported type is tilegx.
22130 -m32
22132 -m64
22133 Generate code for a 32-bit or 64-bit environment. The 32-bit
22134 environment sets int, long, and pointer to 32 bits. The 64-bit
22135 environment sets int to 32 bits and long and pointer to 64 bits.
22136 -mbig-endian
22138 -mlittle-endian
22139 Generate code in big/little endian mode, respectively.
22140 TILEPro Options
22142 These -m options are supported on the TILEPro:
22143 -mcpu=name
22145 Selects the type of CPU to be targeted. Currently the only
22146 supported type is tilepro.
22147 -m32
22149 Generate code for a 32-bit environment, which sets int, long, and
22150 pointer to 32 bits. This is the only supported behavior so the
22151 flag is essentially ignored.
22152 V850 Options
22154 These -m options are defined for V850 implementations:
22155 -mlong-calls
22157 -mno-long-calls
22158 Treat all calls as being far away (near). If calls are assumed to
22159 be far away, the compiler always loads the function's address into
22160 a register, and calls indirect through the pointer.
22161 -mno-ep
22163 -mep
22164 Do not optimize (do optimize) basic blocks that use the same index
22165 pointer 4 or more times to copy pointer into the "ep" register, and
22166 use the shorter "sld" and "sst" instructions. The -mep option is
22167 on by default if you optimize.
22168 -mno-prolog-function
22170 -mprolog-function
22171 Do not use (do use) external functions to save and restore
22172 registers at the prologue and epilogue of a function. The external
22173 functions are slower, but use less code space if more than one
22174 function saves the same number of registers. The -mprolog-function
22175 option is on by default if you optimize.
22176 -mspace
22178 Try to make the code as small as possible. At present, this just
22179 turns on the -mep and -mprolog-function options.
22180 -mtda=n
22182 Put static or global variables whose size is n bytes or less into
22183 the tiny data area that register "ep" points to. The tiny data
22184 area can hold up to 256 bytes in total (128 bytes for byte
22185 references).
22186 -msda=n
22188 Put static or global variables whose size is n bytes or less into
22189 the small data area that register "gp" points to. The small data
22190 area can hold up to 64 kilobytes.
22191 -mzda=n
22193 Put static or global variables whose size is n bytes or less into
22194 the first 32 kilobytes of memory.
22195 -mv850
22197 Specify that the target processor is the V850.
22198 -mv850e3v5
22200 Specify that the target processor is the V850E3V5. The
22201 preprocessor constant "__v850e3v5__" is defined if this option is
22202 used.
22203 -mv850e2v4
22205 Specify that the target processor is the V850E3V5. This is an
22206 alias for the -mv850e3v5 option.
22207 -mv850e2v3
22209 Specify that the target processor is the V850E2V3. The
22210 preprocessor constant "__v850e2v3__" is defined if this option is
22211 used.
22212 -mv850e2
22214 Specify that the target processor is the V850E2. The preprocessor
22215 constant "__v850e2__" is defined if this option is used.
22216 -mv850e1
22218 Specify that the target processor is the V850E1. The preprocessor
22219 constants "__v850e1__" and "__v850e__" are defined if this option
22220 is used.
22221 -mv850es
22223 Specify that the target processor is the V850ES. This is an alias
22224 for the -mv850e1 option.
22225 -mv850e
22227 Specify that the target processor is the V850E. The preprocessor
22228 constant "__v850e__" is defined if this option is used.
22229 If neither -mv850 nor -mv850e nor -mv850e1 nor -mv850e2 nor
22231 -mv850e2v3 nor -mv850e3v5 are defined then a default target
22232 processor is chosen and the relevant __v850*__ preprocessor
22233 constant is defined.
22234 The preprocessor constants "__v850" and "__v851__" are always
22236 defined, regardless of which processor variant is the target.
22237 -mdisable-callt
22239 -mno-disable-callt
22240 This option suppresses generation of the "CALLT" instruction for
22241 the v850e, v850e1, v850e2, v850e2v3 and v850e3v5 flavors of the
22242 v850 architecture.
22243 This option is enabled by default when the RH850 ABI is in use (see
22245 -mrh850-abi), and disabled by default when the GCC ABI is in use.
22246 If "CALLT" instructions are being generated then the C preprocessor
22247 symbol "__V850_CALLT__" is defined.
22248 -mrelax
22250 -mno-relax
22251 Pass on (or do not pass on) the -mrelax command-line option to the
22252 assembler.
22253 -mlong-jumps
22255 -mno-long-jumps
22256 Disable (or re-enable) the generation of PC-relative jump
22257 instructions.
22258 -msoft-float
22260 -mhard-float
22261 Disable (or re-enable) the generation of hardware floating point
22262 instructions. This option is only significant when the target
22263 architecture is V850E2V3 or higher. If hardware floating point
22264 instructions are being generated then the C preprocessor symbol
22265 "__FPU_OK__" is defined, otherwise the symbol "__NO_FPU__" is
22266 defined.
22267 -mloop
22269 Enables the use of the e3v5 LOOP instruction. The use of this
22270 instruction is not enabled by default when the e3v5 architecture is
22271 selected because its use is still experimental.
22272 -mrh850-abi
22274 -mghs
22275 Enables support for the RH850 version of the V850 ABI. This is the
22276 default. With this version of the ABI the following rules apply:
22277 * Integer sized structures and unions are returned via a memory
22279 pointer rather than a register.
22280 * Large structures and unions (more than 8 bytes in size) are
22282 passed by value.
22283 * Functions are aligned to 16-bit boundaries.
22285 * The -m8byte-align command-line option is supported.
22287 * The -mdisable-callt command-line option is enabled by default.
22289 The -mno-disable-callt command-line option is not supported.
22290 When this version of the ABI is enabled the C preprocessor symbol
22292 "__V850_RH850_ABI__" is defined.
22293 -mgcc-abi
22295 Enables support for the old GCC version of the V850 ABI. With this
22296 version of the ABI the following rules apply:
22297 * Integer sized structures and unions are returned in register
22299 "r10".
22300 * Large structures and unions (more than 8 bytes in size) are
22302 passed by reference.
22303 * Functions are aligned to 32-bit boundaries, unless optimizing
22305 for size.
22306 * The -m8byte-align command-line option is not supported.
22308 * The -mdisable-callt command-line option is supported but not
22310 enabled by default.
22311 When this version of the ABI is enabled the C preprocessor symbol
22313 "__V850_GCC_ABI__" is defined.
22314 -m8byte-align
22316 -mno-8byte-align
22317 Enables support for "double" and "long long" types to be aligned on
22318 8-byte boundaries. The default is to restrict the alignment of all
22319 objects to at most 4-bytes. When -m8byte-align is in effect the C
22320 preprocessor symbol "__V850_8BYTE_ALIGN__" is defined.
22321 -mbig-switch
22323 Generate code suitable for big switch tables. Use this option only
22324 if the assembler/linker complain about out of range branches within
22325 a switch table.
22326 -mapp-regs
22328 This option causes r2 and r5 to be used in the code generated by
22329 the compiler. This setting is the default.
22330 -mno-app-regs
22332 This option causes r2 and r5 to be treated as fixed registers.
22333 VAX Options
22335 These -m options are defined for the VAX:
22336 -munix
22338 Do not output certain jump instructions ("aobleq" and so on) that
22339 the Unix assembler for the VAX cannot handle across long ranges.
22340 -mgnu
22342 Do output those jump instructions, on the assumption that the GNU
22343 assembler is being used.
22344 -mg Output code for G-format floating-point numbers instead of
22346 D-format.
22347 Visium Options
22349 -mdebug
22350 A program which performs file I/O and is destined to run on an MCM
22351 target should be linked with this option. It causes the libraries
22352 libc.a and libdebug.a to be linked. The program should be run on
22353 the target under the control of the GDB remote debugging stub.
22354 -msim
22356 A program which performs file I/O and is destined to run on the
22357 simulator should be linked with option. This causes libraries
22358 libc.a and libsim.a to be linked.
22359 -mfpu
22361 -mhard-float
22362 Generate code containing floating-point instructions. This is the
22363 default.
22364 -mno-fpu
22366 -msoft-float
22367 Generate code containing library calls for floating-point.
22368 -msoft-float changes the calling convention in the output file;
22370 therefore, it is only useful if you compile all of a program with
22371 this option. In particular, you need to compile libgcc.a, the
22372 library that comes with GCC, with -msoft-float in order for this to
22373 work.
22374 -mcpu=cpu_type
22376 Set the instruction set, register set, and instruction scheduling
22377 parameters for machine type cpu_type. Supported values for
22378 cpu_type are mcm, gr5 and gr6.
22379 mcm is a synonym of gr5 present for backward compatibility.
22381 By default (unless configured otherwise), GCC generates code for
22383 the GR5 variant of the Visium architecture.
22384 With -mcpu=gr6, GCC generates code for the GR6 variant of the
22386 Visium architecture. The only difference from GR5 code is that the
22387 compiler will generate block move instructions.
22388 -mtune=cpu_type
22390 Set the instruction scheduling parameters for machine type
22391 cpu_type, but do not set the instruction set or register set that
22392 the option -mcpu=cpu_type would.
22393 -msv-mode
22395 Generate code for the supervisor mode, where there are no
22396 restrictions on the access to general registers. This is the
22397 default.
22398 -muser-mode
22400 Generate code for the user mode, where the access to some general
22401 registers is forbidden: on the GR5, registers r24 to r31 cannot be
22402 accessed in this mode; on the GR6, only registers r29 to r31 are
22403 affected.
22404 VMS Options
22406 These -m options are defined for the VMS implementations:
22407 -mvms-return-codes
22409 Return VMS condition codes from "main". The default is to return
22410 POSIX-style condition (e.g. error) codes.
22411 -mdebug-main=prefix
22413 Flag the first routine whose name starts with prefix as the main
22414 routine for the debugger.
22415 -mmalloc64
22417 Default to 64-bit memory allocation routines.
22418 -mpointer-size=size
22420 Set the default size of pointers. Possible options for size are 32
22421 or short for 32 bit pointers, 64 or long for 64 bit pointers, and
22422 no for supporting only 32 bit pointers. The later option disables
22423 "pragma pointer_size".
22424 VxWorks Options
22426 The options in this section are defined for all VxWorks targets.
22427 Options specific to the target hardware are listed with the other
22428 options for that target.
22429 -mrtp
22431 GCC can generate code for both VxWorks kernels and real time
22432 processes (RTPs). This option switches from the former to the
22433 latter. It also defines the preprocessor macro "__RTP__".
22434 -non-static
22436 Link an RTP executable against shared libraries rather than static
22437 libraries. The options -static and -shared can also be used for
22438 RTPs; -static is the default.
22439 -Bstatic
22441 -Bdynamic
22442 These options are passed down to the linker. They are defined for
22443 compatibility with Diab.
22444 -Xbind-lazy
22446 Enable lazy binding of function calls. This option is equivalent
22447 to -Wl,-z,now and is defined for compatibility with Diab.
22448 -Xbind-now
22450 Disable lazy binding of function calls. This option is the default
22451 and is defined for compatibility with Diab.
22452 x86 Options
22454 These -m options are defined for the x86 family of computers.
22455 -march=cpu-type
22457 Generate instructions for the machine type cpu-type. In contrast
22458 to -mtune=cpu-type, which merely tunes the generated code for the
22459 specified cpu-type, -march=cpu-type allows GCC to generate code
22460 that may not run at all on processors other than the one indicated.
22461 Specifying -march=cpu-type implies -mtune=cpu-type.
22462 The choices for cpu-type are:
22464 native
22466 This selects the CPU to generate code for at compilation time
22467 by determining the processor type of the compiling machine.
22468 Using -march=native enables all instruction subsets supported
22469 by the local machine (hence the result might not run on
22470 different machines). Using -mtune=native produces code
22471 optimized for the local machine under the constraints of the
22472 selected instruction set.
22473 x86-64
22475 A generic CPU with 64-bit extensions.
22476 i386
22478 Original Intel i386 CPU.
22479 i486
22481 Intel i486 CPU. (No scheduling is implemented for this chip.)
22482 i586
22484 pentium
22485 Intel Pentium CPU with no MMX support.
22486 lakemont
22488 Intel Lakemont MCU, based on Intel Pentium CPU.
22489 pentium-mmx
22491 Intel Pentium MMX CPU, based on Pentium core with MMX
22492 instruction set support.
22493 pentiumpro
22495 Intel Pentium Pro CPU.
22496 i686
22498 When used with -march, the Pentium Pro instruction set is used,
22499 so the code runs on all i686 family chips. When used with
22500 -mtune, it has the same meaning as generic.
22501 pentium2
22503 Intel Pentium II CPU, based on Pentium Pro core with MMX
22504 instruction set support.
22505 pentium3
22507 pentium3m
22508 Intel Pentium III CPU, based on Pentium Pro core with MMX and
22509 SSE instruction set support.
22510 pentium-m
22512 Intel Pentium M; low-power version of Intel Pentium III CPU
22513 with MMX, SSE and SSE2 instruction set support. Used by
22514 Centrino notebooks.
22515 pentium4
22517 pentium4m
22518 Intel Pentium 4 CPU with MMX, SSE and SSE2 instruction set
22519 support.
22520 prescott
22522 Improved version of Intel Pentium 4 CPU with MMX, SSE, SSE2 and
22523 SSE3 instruction set support.
22524 nocona
22526 Improved version of Intel Pentium 4 CPU with 64-bit extensions,
22527 MMX, SSE, SSE2 and SSE3 instruction set support.
22528 core2
22530 Intel Core 2 CPU with 64-bit extensions, MMX, SSE, SSE2, SSE3
22531 and SSSE3 instruction set support.
22532 nehalem
22534 Intel Nehalem CPU with 64-bit extensions, MMX, SSE, SSE2, SSE3,
22535 SSSE3, SSE4.1, SSE4.2 and POPCNT instruction set support.
22536 westmere
22538 Intel Westmere CPU with 64-bit extensions, MMX, SSE, SSE2,
22539 SSE3, SSSE3, SSE4.1, SSE4.2, POPCNT, AES and PCLMUL instruction
22540 set support.
22541 sandybridge
22543 Intel Sandy Bridge CPU with 64-bit extensions, MMX, SSE, SSE2,
22544 SSE3, SSSE3, SSE4.1, SSE4.2, POPCNT, AVX, AES and PCLMUL
22545 instruction set support.
22546 ivybridge
22548 Intel Ivy Bridge CPU with 64-bit extensions, MMX, SSE, SSE2,
22549 SSE3, SSSE3, SSE4.1, SSE4.2, POPCNT, AVX, AES, PCLMUL,
22550 FSGSBASE, RDRND and F16C instruction set support.
22551 haswell
22553 Intel Haswell CPU with 64-bit extensions, MOVBE, MMX, SSE,
22554 SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, POPCNT, AVX, AVX2, AES,
22555 PCLMUL, FSGSBASE, RDRND, FMA, BMI, BMI2 and F16C instruction
22556 set support.
22557 broadwell
22559 Intel Broadwell CPU with 64-bit extensions, MOVBE, MMX, SSE,
22560 SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, POPCNT, AVX, AVX2, AES,
22561 PCLMUL, FSGSBASE, RDRND, FMA, BMI, BMI2, F16C, RDSEED, ADCX and
22562 PREFETCHW instruction set support.
22563 skylake
22565 Intel Skylake CPU with 64-bit extensions, MOVBE, MMX, SSE,
22566 SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, POPCNT, AVX, AVX2, AES,
22567 PCLMUL, FSGSBASE, RDRND, FMA, BMI, BMI2, F16C, RDSEED, ADCX,
22568 PREFETCHW, CLFLUSHOPT, XSAVEC and XSAVES instruction set
22569 support.
22570 bonnell
22572 Intel Bonnell CPU with 64-bit extensions, MOVBE, MMX, SSE,
22573 SSE2, SSE3 and SSSE3 instruction set support.
22574 silvermont
22576 Intel Silvermont CPU with 64-bit extensions, MOVBE, MMX, SSE,
22577 SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, POPCNT, AES, PCLMUL and
22578 RDRND instruction set support.
22579 knl Intel Knight's Landing CPU with 64-bit extensions, MOVBE, MMX,
22581 SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, POPCNT, AVX, AVX2, AES,
22582 PCLMUL, FSGSBASE, RDRND, FMA, BMI, BMI2, F16C, RDSEED, ADCX,
22583 PREFETCHW, AVX512F, AVX512PF, AVX512ER and AVX512CD instruction
22584 set support.
22585 knm Intel Knights Mill CPU with 64-bit extensions, MOVBE, MMX, SSE,
22587 SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, POPCNT, AVX, AVX2, AES,
22588 PCLMUL, FSGSBASE, RDRND, FMA, BMI, BMI2, F16C, RDSEED, ADCX,
22589 PREFETCHW, AVX512F, AVX512PF, AVX512ER, AVX512CD, AVX5124VNNIW,
22590 AVX5124FMAPS and AVX512VPOPCNTDQ instruction set support.
22591 skylake-avx512
22593 Intel Skylake Server CPU with 64-bit extensions, MOVBE, MMX,
22594 SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, POPCNT, PKU, AVX, AVX2,
22595 AES, PCLMUL, FSGSBASE, RDRND, FMA, BMI, BMI2, F16C, RDSEED,
22596 ADCX, PREFETCHW, CLFLUSHOPT, XSAVEC, XSAVES, AVX512F, CLWB,
22597 AVX512VL, AVX512BW, AVX512DQ and AVX512CD instruction set
22598 support.
22599 cannonlake
22601 Intel Cannonlake Server CPU with 64-bit extensions, MOVBE, MMX,
22602 SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, POPCNT, PKU, AVX, AVX2,
22603 AES, PCLMUL, FSGSBASE, RDRND, FMA, BMI, BMI2, F16C, RDSEED,
22604 ADCX, PREFETCHW, CLFLUSHOPT, XSAVEC, XSAVES, AVX512F, AVX512VL,
22605 AVX512BW, AVX512DQ, AVX512CD, AVX512VBMI, AVX512IFMA, SHA and
22606 UMIP instruction set support.
22607 icelake-client
22609 Intel Icelake Client CPU with 64-bit extensions, MOVBE, MMX,
22610 SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, POPCNT, PKU, AVX, AVX2,
22611 AES, PCLMUL, FSGSBASE, RDRND, FMA, BMI, BMI2, F16C, RDSEED,
22612 ADCX, PREFETCHW, CLFLUSHOPT, XSAVEC, XSAVES, AVX512F, AVX512VL,
22613 AVX512BW, AVX512DQ, AVX512CD, AVX512VBMI, AVX512IFMA, SHA,
22614 CLWB, UMIP, RDPID, GFNI, AVX512VBMI2, AVX512VPOPCNTDQ,
22615 AVX512BITALG, AVX512VNNI, VPCLMULQDQ, VAES instruction set
22616 support.
22617 icelake-server
22619 Intel Icelake Server CPU with 64-bit extensions, MOVBE, MMX,
22620 SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, POPCNT, PKU, AVX, AVX2,
22621 AES, PCLMUL, FSGSBASE, RDRND, FMA, BMI, BMI2, F16C, RDSEED,
22622 ADCX, PREFETCHW, CLFLUSHOPT, XSAVEC, XSAVES, AVX512F, AVX512VL,
22623 AVX512BW, AVX512DQ, AVX512CD, AVX512VBMI, AVX512IFMA, SHA,
22624 CLWB, UMIP, RDPID, GFNI, AVX512VBMI2, AVX512VPOPCNTDQ,
22625 AVX512BITALG, AVX512VNNI, VPCLMULQDQ, VAES, PCONFIG and
22626 WBNOINVD instruction set support.
22627 k6 AMD K6 CPU with MMX instruction set support.
22629 k6-2
22631 k6-3
22632 Improved versions of AMD K6 CPU with MMX and 3DNow! instruction
22633 set support.
22634 athlon
22636 athlon-tbird
22637 AMD Athlon CPU with MMX, 3dNOW!, enhanced 3DNow! and SSE
22638 prefetch instructions support.
22639 athlon-4
22641 athlon-xp
22642 athlon-mp
22643 Improved AMD Athlon CPU with MMX, 3DNow!, enhanced 3DNow! and
22644 full SSE instruction set support.
22645 k8
22647 opteron
22648 athlon64
22649 athlon-fx
22650 Processors based on the AMD K8 core with x86-64 instruction set
22651 support, including the AMD Opteron, Athlon 64, and Athlon 64 FX
22652 processors. (This supersets MMX, SSE, SSE2, 3DNow!, enhanced
22653 3DNow! and 64-bit instruction set extensions.)
22654 k8-sse3
22656 opteron-sse3
22657 athlon64-sse3
22658 Improved versions of AMD K8 cores with SSE3 instruction set
22659 support.
22660 amdfam10
22662 barcelona
22663 CPUs based on AMD Family 10h cores with x86-64 instruction set
22664 support. (This supersets MMX, SSE, SSE2, SSE3, SSE4A, 3DNow!,
22665 enhanced 3DNow!, ABM and 64-bit instruction set extensions.)
22666 bdver1
22668 CPUs based on AMD Family 15h cores with x86-64 instruction set
22669 support. (This supersets FMA4, AVX, XOP, LWP, AES, PCL_MUL,
22670 CX16, MMX, SSE, SSE2, SSE3, SSE4A, SSSE3, SSE4.1, SSE4.2, ABM
22671 and 64-bit instruction set extensions.)
22672 bdver2
22674 AMD Family 15h core based CPUs with x86-64 instruction set
22675 support. (This supersets BMI, TBM, F16C, FMA, FMA4, AVX, XOP,
22676 LWP, AES, PCL_MUL, CX16, MMX, SSE, SSE2, SSE3, SSE4A, SSSE3,
22677 SSE4.1, SSE4.2, ABM and 64-bit instruction set extensions.)
22678 bdver3
22680 AMD Family 15h core based CPUs with x86-64 instruction set
22681 support. (This supersets BMI, TBM, F16C, FMA, FMA4, FSGSBASE,
22682 AVX, XOP, LWP, AES, PCL_MUL, CX16, MMX, SSE, SSE2, SSE3, SSE4A,
22683 SSSE3, SSE4.1, SSE4.2, ABM and 64-bit instruction set
22684 extensions.
22685 bdver4
22687 AMD Family 15h core based CPUs with x86-64 instruction set
22688 support. (This supersets BMI, BMI2, TBM, F16C, FMA, FMA4,
22689 FSGSBASE, AVX, AVX2, XOP, LWP, AES, PCL_MUL, CX16, MOVBE, MMX,
22690 SSE, SSE2, SSE3, SSE4A, SSSE3, SSE4.1, SSE4.2, ABM and 64-bit
22691 instruction set extensions.
22692 znver1
22694 AMD Family 17h core based CPUs with x86-64 instruction set
22695 support. (This supersets BMI, BMI2, F16C, FMA, FSGSBASE, AVX,
22696 AVX2, ADCX, RDSEED, MWAITX, SHA, CLZERO, AES, PCL_MUL, CX16,
22697 MOVBE, MMX, SSE, SSE2, SSE3, SSE4A, SSSE3, SSE4.1, SSE4.2, ABM,
22698 XSAVEC, XSAVES, CLFLUSHOPT, POPCNT, and 64-bit instruction set
22699 extensions.
22700 btver1
22702 CPUs based on AMD Family 14h cores with x86-64 instruction set
22703 support. (This supersets MMX, SSE, SSE2, SSE3, SSSE3, SSE4A,
22704 CX16, ABM and 64-bit instruction set extensions.)
22705 btver2
22707 CPUs based on AMD Family 16h cores with x86-64 instruction set
22708 support. This includes MOVBE, F16C, BMI, AVX, PCL_MUL, AES,
22709 SSE4.2, SSE4.1, CX16, ABM, SSE4A, SSSE3, SSE3, SSE2, SSE, MMX
22710 and 64-bit instruction set extensions.
22711 winchip-c6
22713 IDT WinChip C6 CPU, dealt in same way as i486 with additional
22714 MMX instruction set support.
22715 winchip2
22717 IDT WinChip 2 CPU, dealt in same way as i486 with additional
22718 MMX and 3DNow! instruction set support.
22719 c3 VIA C3 CPU with MMX and 3DNow! instruction set support. (No
22721 scheduling is implemented for this chip.)
22722 c3-2
22724 VIA C3-2 (Nehemiah/C5XL) CPU with MMX and SSE instruction set
22725 support. (No scheduling is implemented for this chip.)
22726 c7 VIA C7 (Esther) CPU with MMX, SSE, SSE2 and SSE3 instruction
22728 set support. (No scheduling is implemented for this chip.)
22729 samuel-2
22731 VIA Eden Samuel 2 CPU with MMX and 3DNow! instruction set
22732 support. (No scheduling is implemented for this chip.)
22733 nehemiah
22735 VIA Eden Nehemiah CPU with MMX and SSE instruction set support.
22736 (No scheduling is implemented for this chip.)
22737 esther
22739 VIA Eden Esther CPU with MMX, SSE, SSE2 and SSE3 instruction
22740 set support. (No scheduling is implemented for this chip.)
22741 eden-x2
22743 VIA Eden X2 CPU with x86-64, MMX, SSE, SSE2 and SSE3
22744 instruction set support. (No scheduling is implemented for
22745 this chip.)
22746 eden-x4
22748 VIA Eden X4 CPU with x86-64, MMX, SSE, SSE2, SSE3, SSSE3,
22749 SSE4.1, SSE4.2, AVX and AVX2 instruction set support. (No
22750 scheduling is implemented for this chip.)
22751 nano
22753 Generic VIA Nano CPU with x86-64, MMX, SSE, SSE2, SSE3 and
22754 SSSE3 instruction set support. (No scheduling is implemented
22755 for this chip.)
22756 nano-1000
22758 VIA Nano 1xxx CPU with x86-64, MMX, SSE, SSE2, SSE3 and SSSE3
22759 instruction set support. (No scheduling is implemented for
22760 this chip.)
22761 nano-2000
22763 VIA Nano 2xxx CPU with x86-64, MMX, SSE, SSE2, SSE3 and SSSE3
22764 instruction set support. (No scheduling is implemented for
22765 this chip.)
22766 nano-3000
22768 VIA Nano 3xxx CPU with x86-64, MMX, SSE, SSE2, SSE3, SSSE3 and
22769 SSE4.1 instruction set support. (No scheduling is implemented
22770 for this chip.)
22771 nano-x2
22773 VIA Nano Dual Core CPU with x86-64, MMX, SSE, SSE2, SSE3, SSSE3
22774 and SSE4.1 instruction set support. (No scheduling is
22775 implemented for this chip.)
22776 nano-x4
22778 VIA Nano Quad Core CPU with x86-64, MMX, SSE, SSE2, SSE3, SSSE3
22779 and SSE4.1 instruction set support. (No scheduling is
22780 implemented for this chip.)
22781 geode
22783 AMD Geode embedded processor with MMX and 3DNow! instruction
22784 set support.
22785 -mtune=cpu-type
22787 Tune to cpu-type everything applicable about the generated code,
22788 except for the ABI and the set of available instructions. While
22789 picking a specific cpu-type schedules things appropriately for that
22790 particular chip, the compiler does not generate any code that
22791 cannot run on the default machine type unless you use a -march=cpu-
22792 type option. For example, if GCC is configured for
22793 i686-pc-linux-gnu then -mtune=pentium4 generates code that is tuned
22794 for Pentium 4 but still runs on i686 machines.
22795 The choices for cpu-type are the same as for -march. In addition,
22797 -mtune supports 2 extra choices for cpu-type:
22798 generic
22800 Produce code optimized for the most common IA32/AMD64/EM64T
22801 processors. If you know the CPU on which your code will run,
22802 then you should use the corresponding -mtune or -march option
22803 instead of -mtune=generic. But, if you do not know exactly
22804 what CPU users of your application will have, then you should
22805 use this option.
22806 As new processors are deployed in the marketplace, the behavior
22808 of this option will change. Therefore, if you upgrade to a
22809 newer version of GCC, code generation controlled by this option
22810 will change to reflect the processors that are most common at
22811 the time that version of GCC is released.
22812 There is no -march=generic option because -march indicates the
22814 instruction set the compiler can use, and there is no generic
22815 instruction set applicable to all processors. In contrast,
22816 -mtune indicates the processor (or, in this case, collection of
22817 processors) for which the code is optimized.
22818 intel
22820 Produce code optimized for the most current Intel processors,
22821 which are Haswell and Silvermont for this version of GCC. If
22822 you know the CPU on which your code will run, then you should
22823 use the corresponding -mtune or -march option instead of
22824 -mtune=intel. But, if you want your application performs
22825 better on both Haswell and Silvermont, then you should use this
22826 option.
22827 As new Intel processors are deployed in the marketplace, the
22829 behavior of this option will change. Therefore, if you upgrade
22830 to a newer version of GCC, code generation controlled by this
22831 option will change to reflect the most current Intel processors
22832 at the time that version of GCC is released.
22833 There is no -march=intel option because -march indicates the
22835 instruction set the compiler can use, and there is no common
22836 instruction set applicable to all processors. In contrast,
22837 -mtune indicates the processor (or, in this case, collection of
22838 processors) for which the code is optimized.
22839 -mcpu=cpu-type
22841 A deprecated synonym for -mtune.
22842 -mfpmath=unit
22844 Generate floating-point arithmetic for selected unit unit. The
22845 choices for unit are:
22846 387 Use the standard 387 floating-point coprocessor present on the
22848 majority of chips and emulated otherwise. Code compiled with
22849 this option runs almost everywhere. The temporary results are
22850 computed in 80-bit precision instead of the precision specified
22851 by the type, resulting in slightly different results compared
22852 to most of other chips. See -ffloat-store for more detailed
22853 description.
22854 This is the default choice for non-Darwin x86-32 targets.
22856 sse Use scalar floating-point instructions present in the SSE
22858 instruction set. This instruction set is supported by Pentium
22859 III and newer chips, and in the AMD line by Athlon-4, Athlon XP
22860 and Athlon MP chips. The earlier version of the SSE
22861 instruction set supports only single-precision arithmetic, thus
22862 the double and extended-precision arithmetic are still done
22863 using 387. A later version, present only in Pentium 4 and AMD
22864 x86-64 chips, supports double-precision arithmetic too.
22865 For the x86-32 compiler, you must use -march=cpu-type, -msse or
22867 -msse2 switches to enable SSE extensions and make this option
22868 effective. For the x86-64 compiler, these extensions are
22869 enabled by default.
22870 The resulting code should be considerably faster in the
22872 majority of cases and avoid the numerical instability problems
22873 of 387 code, but may break some existing code that expects
22874 temporaries to be 80 bits.
22875 This is the default choice for the x86-64 compiler, Darwin
22877 x86-32 targets, and the default choice for x86-32 targets with
22878 the SSE2 instruction set when -ffast-math is enabled.
22879 sse,387
22881 sse+387
22882 both
22883 Attempt to utilize both instruction sets at once. This
22884 effectively doubles the amount of available registers, and on
22885 chips with separate execution units for 387 and SSE the
22886 execution resources too. Use this option with care, as it is
22887 still experimental, because the GCC register allocator does not
22888 model separate functional units well, resulting in unstable
22889 performance.
22890 -masm=dialect
22892 Output assembly instructions using selected dialect. Also affects
22893 which dialect is used for basic "asm" and extended "asm". Supported
22894 choices (in dialect order) are att or intel. The default is att.
22895 Darwin does not support intel.
22896 -mieee-fp
22898 -mno-ieee-fp
22899 Control whether or not the compiler uses IEEE floating-point
22900 comparisons. These correctly handle the case where the result of a
22901 comparison is unordered.
22902 -m80387
22904 -mhard-float
22905 Generate output containing 80387 instructions for floating point.
22906 -mno-80387
22908 -msoft-float
22909 Generate output containing library calls for floating point.
22910 Warning: the requisite libraries are not part of GCC. Normally the
22912 facilities of the machine's usual C compiler are used, but this
22913 cannot be done directly in cross-compilation. You must make your
22914 own arrangements to provide suitable library functions for cross-
22915 compilation.
22916 On machines where a function returns floating-point results in the
22918 80387 register stack, some floating-point opcodes may be emitted
22919 even if -msoft-float is used.
22920 -mno-fp-ret-in-387
22922 Do not use the FPU registers for return values of functions.
22923 The usual calling convention has functions return values of types
22925 "float" and "double" in an FPU register, even if there is no FPU.
22926 The idea is that the operating system should emulate an FPU.
22927 The option -mno-fp-ret-in-387 causes such values to be returned in
22929 ordinary CPU registers instead.
22930 -mno-fancy-math-387
22932 Some 387 emulators do not support the "sin", "cos" and "sqrt"
22933 instructions for the 387. Specify this option to avoid generating
22934 those instructions. This option is the default on OpenBSD and
22935 NetBSD. This option is overridden when -march indicates that the
22936 target CPU always has an FPU and so the instruction does not need
22937 emulation. These instructions are not generated unless you also
22938 use the -funsafe-math-optimizations switch.
22939 -malign-double
22941 -mno-align-double
22942 Control whether GCC aligns "double", "long double", and "long long"
22943 variables on a two-word boundary or a one-word boundary. Aligning
22944 "double" variables on a two-word boundary produces code that runs
22945 somewhat faster on a Pentium at the expense of more memory.
22946 On x86-64, -malign-double is enabled by default.
22948 Warning: if you use the -malign-double switch, structures
22950 containing the above types are aligned differently than the
22951 published application binary interface specifications for the
22952 x86-32 and are not binary compatible with structures in code
22953 compiled without that switch.
22954 -m96bit-long-double
22956 -m128bit-long-double
22957 These switches control the size of "long double" type. The x86-32
22958 application binary interface specifies the size to be 96 bits, so
22959 -m96bit-long-double is the default in 32-bit mode.
22960 Modern architectures (Pentium and newer) prefer "long double" to be
22962 aligned to an 8- or 16-byte boundary. In arrays or structures
22963 conforming to the ABI, this is not possible. So specifying
22964 -m128bit-long-double aligns "long double" to a 16-byte boundary by
22965 padding the "long double" with an additional 32-bit zero.
22966 In the x86-64 compiler, -m128bit-long-double is the default choice
22968 as its ABI specifies that "long double" is aligned on 16-byte
22969 boundary.
22970 Notice that neither of these options enable any extra precision
22972 over the x87 standard of 80 bits for a "long double".
22973 Warning: if you override the default value for your target ABI,
22975 this changes the size of structures and arrays containing "long
22976 double" variables, as well as modifying the function calling
22977 convention for functions taking "long double". Hence they are not
22978 binary-compatible with code compiled without that switch.
22979 -mlong-double-64
22981 -mlong-double-80
22982 -mlong-double-128
22983 These switches control the size of "long double" type. A size of 64
22984 bits makes the "long double" type equivalent to the "double" type.
22985 This is the default for 32-bit Bionic C library. A size of 128
22986 bits makes the "long double" type equivalent to the "__float128"
22987 type. This is the default for 64-bit Bionic C library.
22988 Warning: if you override the default value for your target ABI,
22990 this changes the size of structures and arrays containing "long
22991 double" variables, as well as modifying the function calling
22992 convention for functions taking "long double". Hence they are not
22993 binary-compatible with code compiled without that switch.
22994 -malign-data=type
22996 Control how GCC aligns variables. Supported values for type are
22997 compat uses increased alignment value compatible uses GCC 4.8 and
22998 earlier, abi uses alignment value as specified by the psABI, and
22999 cacheline uses increased alignment value to match the cache line
23000 size. compat is the default.
23001 -mlarge-data-threshold=threshold
23003 When -mcmodel=medium is specified, data objects larger than
23004 threshold are placed in the large data section. This value must be
23005 the same across all objects linked into the binary, and defaults to
23006 65535.
23007 -mrtd
23009 Use a different function-calling convention, in which functions
23010 that take a fixed number of arguments return with the "ret num"
23011 instruction, which pops their arguments while returning. This
23012 saves one instruction in the caller since there is no need to pop
23013 the arguments there.
23014 You can specify that an individual function is called with this
23016 calling sequence with the function attribute "stdcall". You can
23017 also override the -mrtd option by using the function attribute
23018 "cdecl".
23019 Warning: this calling convention is incompatible with the one
23021 normally used on Unix, so you cannot use it if you need to call
23022 libraries compiled with the Unix compiler.
23023 Also, you must provide function prototypes for all functions that
23025 take variable numbers of arguments (including "printf"); otherwise
23026 incorrect code is generated for calls to those functions.
23027 In addition, seriously incorrect code results if you call a
23029 function with too many arguments. (Normally, extra arguments are
23030 harmlessly ignored.)
23031 -mregparm=num
23033 Control how many registers are used to pass integer arguments. By
23034 default, no registers are used to pass arguments, and at most 3
23035 registers can be used. You can control this behavior for a
23036 specific function by using the function attribute "regparm".
23037 Warning: if you use this switch, and num is nonzero, then you must
23039 build all modules with the same value, including any libraries.
23040 This includes the system libraries and startup modules.
23041 -msseregparm
23043 Use SSE register passing conventions for float and double arguments
23044 and return values. You can control this behavior for a specific
23045 function by using the function attribute "sseregparm".
23046 Warning: if you use this switch then you must build all modules
23048 with the same value, including any libraries. This includes the
23049 system libraries and startup modules.
23050 -mvect8-ret-in-mem
23052 Return 8-byte vectors in memory instead of MMX registers. This is
23053 the default on Solaris@tie{}8 and 9 and VxWorks to match the ABI of
23054 the Sun Studio compilers until version 12. Later compiler versions
23055 (starting with Studio 12 Update@tie{}1) follow the ABI used by
23056 other x86 targets, which is the default on Solaris@tie{}10 and
23057 later. Only use this option if you need to remain compatible with
23058 existing code produced by those previous compiler versions or older
23059 versions of GCC.
23060 -mpc32
23062 -mpc64
23063 -mpc80
23064 Set 80387 floating-point precision to 32, 64 or 80 bits. When
23065 -mpc32 is specified, the significands of results of floating-point
23066 operations are rounded to 24 bits (single precision); -mpc64 rounds
23067 the significands of results of floating-point operations to 53 bits
23068 (double precision) and -mpc80 rounds the significands of results of
23069 floating-point operations to 64 bits (extended double precision),
23070 which is the default. When this option is used, floating-point
23071 operations in higher precisions are not available to the programmer
23072 without setting the FPU control word explicitly.
23073 Setting the rounding of floating-point operations to less than the
23075 default 80 bits can speed some programs by 2% or more. Note that
23076 some mathematical libraries assume that extended-precision (80-bit)
23077 floating-point operations are enabled by default; routines in such
23078 libraries could suffer significant loss of accuracy, typically
23079 through so-called "catastrophic cancellation", when this option is
23080 used to set the precision to less than extended precision.
23081 -mstackrealign
23083 Realign the stack at entry. On the x86, the -mstackrealign option
23084 generates an alternate prologue and epilogue that realigns the run-
23085 time stack if necessary. This supports mixing legacy codes that
23086 keep 4-byte stack alignment with modern codes that keep 16-byte
23087 stack alignment for SSE compatibility. See also the attribute
23088 "force_align_arg_pointer", applicable to individual functions.
23089 -mpreferred-stack-boundary=num
23091 Attempt to keep the stack boundary aligned to a 2 raised to num
23092 byte boundary. If -mpreferred-stack-boundary is not specified, the
23093 default is 4 (16 bytes or 128 bits).
23094 Warning: When generating code for the x86-64 architecture with SSE
23096 extensions disabled, -mpreferred-stack-boundary=3 can be used to
23097 keep the stack boundary aligned to 8 byte boundary. Since x86-64
23098 ABI require 16 byte stack alignment, this is ABI incompatible and
23099 intended to be used in controlled environment where stack space is
23100 important limitation. This option leads to wrong code when
23101 functions compiled with 16 byte stack alignment (such as functions
23102 from a standard library) are called with misaligned stack. In this
23103 case, SSE instructions may lead to misaligned memory access traps.
23104 In addition, variable arguments are handled incorrectly for 16 byte
23105 aligned objects (including x87 long double and __int128), leading
23106 to wrong results. You must build all modules with
23107 -mpreferred-stack-boundary=3, including any libraries. This
23108 includes the system libraries and startup modules.
23109 -mincoming-stack-boundary=num
23111 Assume the incoming stack is aligned to a 2 raised to num byte
23112 boundary. If -mincoming-stack-boundary is not specified, the one
23113 specified by -mpreferred-stack-boundary is used.
23114 On Pentium and Pentium Pro, "double" and "long double" values
23116 should be aligned to an 8-byte boundary (see -malign-double) or
23117 suffer significant run time performance penalties. On Pentium III,
23118 the Streaming SIMD Extension (SSE) data type "__m128" may not work
23119 properly if it is not 16-byte aligned.
23120 To ensure proper alignment of this values on the stack, the stack
23122 boundary must be as aligned as that required by any value stored on
23123 the stack. Further, every function must be generated such that it
23124 keeps the stack aligned. Thus calling a function compiled with a
23125 higher preferred stack boundary from a function compiled with a
23126 lower preferred stack boundary most likely misaligns the stack. It
23127 is recommended that libraries that use callbacks always use the
23128 default setting.
23129 This extra alignment does consume extra stack space, and generally
23131 increases code size. Code that is sensitive to stack space usage,
23132 such as embedded systems and operating system kernels, may want to
23133 reduce the preferred alignment to -mpreferred-stack-boundary=2.
23134 -mmmx
23136 -msse
23137 -msse2
23138 -msse3
23139 -mssse3
23140 -msse4
23141 -msse4a
23142 -msse4.1
23143 -msse4.2
23144 -mavx
23145 -mavx2
23146 -mavx512f
23147 -mavx512pf
23148 -mavx512er
23149 -mavx512cd
23150 -mavx512vl
23151 -mavx512bw
23152 -mavx512dq
23153 -mavx512ifma
23154 -mavx512vbmi
23155 -msha
23156 -maes
23157 -mpclmul
23158 -mclflushopt
23159 -mfsgsbase
23160 -mrdrnd
23161 -mf16c
23162 -mfma
23163 -mpconfig
23164 -mwbnoinvd
23165 -mfma4
23166 -mprefetchwt1
23167 -mxop
23168 -mlwp
23169 -m3dnow
23170 -m3dnowa
23171 -mpopcnt
23172 -mabm
23173 -mbmi
23174 -mbmi2
23175 -mlzcnt
23176 -mfxsr
23177 -mxsave
23178 -mxsaveopt
23179 -mxsavec
23180 -mxsaves
23181 -mrtm
23182 -mtbm
23183 -mmpx
23184 -mmwaitx
23185 -mclzero
23186 -mpku
23187 -mavx512vbmi2
23188 -mgfni
23189 -mvaes
23190 -mvpclmulqdq
23191 -mavx512bitalg
23192 -mmovdiri
23193 -mmovdir64b
23194 -mavx512vpopcntdq
23195 These switches enable the use of instructions in the MMX, SSE,
23196 SSE2, SSE3, SSSE3, SSE4.1, AVX, AVX2, AVX512F, AVX512PF, AVX512ER,
23197 AVX512CD, SHA, AES, PCLMUL, FSGSBASE, RDRND, F16C, FMA, SSE4A,
23198 FMA4, XOP, LWP, ABM, AVX512VL, AVX512BW, AVX512DQ, AVX512IFMA,
23199 AVX512VBMI, BMI, BMI2, VAES, FXSR, XSAVE, XSAVEOPT, LZCNT, RTM,
23200 MPX, MWAITX, PKU, IBT, SHSTK, AVX512VBMI2, GFNI, VPCLMULQDQ,
23201 AVX512BITALG, MOVDIRI, MOVDIR64B, AVX512VPOPCNTDQ3DNow! or enhanced
23202 3DNow! extended instruction sets. Each has a corresponding -mno-
23203 option to disable use of these instructions.
23204 These extensions are also available as built-in functions: see x86
23206 Built-in Functions, for details of the functions enabled and
23207 disabled by these switches.
23208 To generate SSE/SSE2 instructions automatically from floating-point
23210 code (as opposed to 387 instructions), see -mfpmath=sse.
23211 GCC depresses SSEx instructions when -mavx is used. Instead, it
23213 generates new AVX instructions or AVX equivalence for all SSEx
23214 instructions when needed.
23215 These options enable GCC to use these extended instructions in
23217 generated code, even without -mfpmath=sse. Applications that
23218 perform run-time CPU detection must compile separate files for each
23219 supported architecture, using the appropriate flags. In
23220 particular, the file containing the CPU detection code should be
23221 compiled without these options.
23222 -mdump-tune-features
23224 This option instructs GCC to dump the names of the x86 performance
23225 tuning features and default settings. The names can be used in
23226 -mtune-ctrl=feature-list.
23227 -mtune-ctrl=feature-list
23229 This option is used to do fine grain control of x86 code generation
23230 features. feature-list is a comma separated list of feature names.
23231 See also -mdump-tune-features. When specified, the feature is
23232 turned on if it is not preceded with ^, otherwise, it is turned
23233 off. -mtune-ctrl=feature-list is intended to be used by GCC
23234 developers. Using it may lead to code paths not covered by testing
23235 and can potentially result in compiler ICEs or runtime errors.
23236 -mno-default
23238 This option instructs GCC to turn off all tunable features. See
23239 also -mtune-ctrl=feature-list and -mdump-tune-features.
23240 -mcld
23242 This option instructs GCC to emit a "cld" instruction in the
23243 prologue of functions that use string instructions. String
23244 instructions depend on the DF flag to select between autoincrement
23245 or autodecrement mode. While the ABI specifies the DF flag to be
23246 cleared on function entry, some operating systems violate this
23247 specification by not clearing the DF flag in their exception
23248 dispatchers. The exception handler can be invoked with the DF flag
23249 set, which leads to wrong direction mode when string instructions
23250 are used. This option can be enabled by default on 32-bit x86
23251 targets by configuring GCC with the --enable-cld configure option.
23252 Generation of "cld" instructions can be suppressed with the
23253 -mno-cld compiler option in this case.
23254 -mvzeroupper
23256 This option instructs GCC to emit a "vzeroupper" instruction before
23257 a transfer of control flow out of the function to minimize the AVX
23258 to SSE transition penalty as well as remove unnecessary "zeroupper"
23259 intrinsics.
23260 -mprefer-avx128
23262 This option instructs GCC to use 128-bit AVX instructions instead
23263 of 256-bit AVX instructions in the auto-vectorizer.
23264 -mprefer-vector-width=opt
23266 This option instructs GCC to use opt-bit vector width in
23267 instructions instead of default on the selected platform.
23268 none
23270 No extra limitations applied to GCC other than defined by the
23271 selected platform.
23272 128 Prefer 128-bit vector width for instructions.
23274 256 Prefer 256-bit vector width for instructions.
23276 512 Prefer 512-bit vector width for instructions.
23278 -mcx16
23280 This option enables GCC to generate "CMPXCHG16B" instructions in
23281 64-bit code to implement compare-and-exchange operations on 16-byte
23282 aligned 128-bit objects. This is useful for atomic updates of data
23283 structures exceeding one machine word in size. The compiler uses
23284 this instruction to implement __sync Builtins. However, for
23285 __atomic Builtins operating on 128-bit integers, a library call is
23286 always used.
23287 -msahf
23289 This option enables generation of "SAHF" instructions in 64-bit
23290 code. Early Intel Pentium 4 CPUs with Intel 64 support, prior to
23291 the introduction of Pentium 4 G1 step in December 2005, lacked the
23292 "LAHF" and "SAHF" instructions which are supported by AMD64. These
23293 are load and store instructions, respectively, for certain status
23294 flags. In 64-bit mode, the "SAHF" instruction is used to optimize
23295 "fmod", "drem", and "remainder" built-in functions; see Other
23296 Builtins for details.
23297 -mmovbe
23299 This option enables use of the "movbe" instruction to implement
23300 "__builtin_bswap32" and "__builtin_bswap64".
23301 -mshstk
23303 The -mshstk option enables shadow stack built-in functions from x86
23304 Control-flow Enforcement Technology (CET).
23305 -mcrc32
23307 This option enables built-in functions "__builtin_ia32_crc32qi",
23308 "__builtin_ia32_crc32hi", "__builtin_ia32_crc32si" and
23309 "__builtin_ia32_crc32di" to generate the "crc32" machine
23310 instruction.
23311 -mrecip
23313 This option enables use of "RCPSS" and "RSQRTSS" instructions (and
23314 their vectorized variants "RCPPS" and "RSQRTPS") with an additional
23315 Newton-Raphson step to increase precision instead of "DIVSS" and
23316 "SQRTSS" (and their vectorized variants) for single-precision
23317 floating-point arguments. These instructions are generated only
23318 when -funsafe-math-optimizations is enabled together with
23319 -ffinite-math-only and -fno-trapping-math. Note that while the
23320 throughput of the sequence is higher than the throughput of the
23321 non-reciprocal instruction, the precision of the sequence can be
23322 decreased by up to 2 ulp (i.e. the inverse of 1.0 equals
23323 0.99999994).
23324 Note that GCC implements "1.0f/sqrtf(x)" in terms of "RSQRTSS" (or
23326 "RSQRTPS") already with -ffast-math (or the above option
23327 combination), and doesn't need -mrecip.
23328 Also note that GCC emits the above sequence with additional Newton-
23330 Raphson step for vectorized single-float division and vectorized
23331 "sqrtf(x)" already with -ffast-math (or the above option
23332 combination), and doesn't need -mrecip.
23333 -mrecip=opt
23335 This option controls which reciprocal estimate instructions may be
23336 used. opt is a comma-separated list of options, which may be
23337 preceded by a ! to invert the option:
23338 all Enable all estimate instructions.
23340 default
23342 Enable the default instructions, equivalent to -mrecip.
23343 none
23345 Disable all estimate instructions, equivalent to -mno-recip.
23346 div Enable the approximation for scalar division.
23348 vec-div
23350 Enable the approximation for vectorized division.
23351 sqrt
23353 Enable the approximation for scalar square root.
23354 vec-sqrt
23356 Enable the approximation for vectorized square root.
23357 So, for example, -mrecip=all,!sqrt enables all of the reciprocal
23359 approximations, except for square root.
23360 -mveclibabi=type
23362 Specifies the ABI type to use for vectorizing intrinsics using an
23363 external library. Supported values for type are svml for the Intel
23364 short vector math library and acml for the AMD math core library.
23365 To use this option, both -ftree-vectorize and
23366 -funsafe-math-optimizations have to be enabled, and an SVML or ACML
23367 ABI-compatible library must be specified at link time.
23368 GCC currently emits calls to "vmldExp2", "vmldLn2", "vmldLog102",
23370 "vmldPow2", "vmldTanh2", "vmldTan2", "vmldAtan2", "vmldAtanh2",
23371 "vmldCbrt2", "vmldSinh2", "vmldSin2", "vmldAsinh2", "vmldAsin2",
23372 "vmldCosh2", "vmldCos2", "vmldAcosh2", "vmldAcos2", "vmlsExp4",
23373 "vmlsLn4", "vmlsLog104", "vmlsPow4", "vmlsTanh4", "vmlsTan4",
23374 "vmlsAtan4", "vmlsAtanh4", "vmlsCbrt4", "vmlsSinh4", "vmlsSin4",
23375 "vmlsAsinh4", "vmlsAsin4", "vmlsCosh4", "vmlsCos4", "vmlsAcosh4"
23376 and "vmlsAcos4" for corresponding function type when
23377 -mveclibabi=svml is used, and "__vrd2_sin", "__vrd2_cos",
23378 "__vrd2_exp", "__vrd2_log", "__vrd2_log2", "__vrd2_log10",
23379 "__vrs4_sinf", "__vrs4_cosf", "__vrs4_expf", "__vrs4_logf",
23380 "__vrs4_log2f", "__vrs4_log10f" and "__vrs4_powf" for the
23381 corresponding function type when -mveclibabi=acml is used.
23382 -mabi=name
23384 Generate code for the specified calling convention. Permissible
23385 values are sysv for the ABI used on GNU/Linux and other systems,
23386 and ms for the Microsoft ABI. The default is to use the Microsoft
23387 ABI when targeting Microsoft Windows and the SysV ABI on all other
23388 systems. You can control this behavior for specific functions by
23389 using the function attributes "ms_abi" and "sysv_abi".
23390 -mforce-indirect-call
23392 Force all calls to functions to be indirect. This is useful when
23393 using Intel Processor Trace where it generates more precise timing
23394 information for function calls.
23395 -mcall-ms2sysv-xlogues
23397 Due to differences in 64-bit ABIs, any Microsoft ABI function that
23398 calls a System V ABI function must consider RSI, RDI and XMM6-15 as
23399 clobbered. By default, the code for saving and restoring these
23400 registers is emitted inline, resulting in fairly lengthy prologues
23401 and epilogues. Using -mcall-ms2sysv-xlogues emits prologues and
23402 epilogues that use stubs in the static portion of libgcc to perform
23403 these saves and restores, thus reducing function size at the cost
23404 of a few extra instructions.
23405 -mtls-dialect=type
23407 Generate code to access thread-local storage using the gnu or gnu2
23408 conventions. gnu is the conservative default; gnu2 is more
23409 efficient, but it may add compile- and run-time requirements that
23410 cannot be satisfied on all systems.
23411 -mpush-args
23413 -mno-push-args
23414 Use PUSH operations to store outgoing parameters. This method is
23415 shorter and usually equally fast as method using SUB/MOV operations
23416 and is enabled by default. In some cases disabling it may improve
23417 performance because of improved scheduling and reduced
23418 dependencies.
23419 -maccumulate-outgoing-args
23421 If enabled, the maximum amount of space required for outgoing
23422 arguments is computed in the function prologue. This is faster on
23423 most modern CPUs because of reduced dependencies, improved
23424 scheduling and reduced stack usage when the preferred stack
23425 boundary is not equal to 2. The drawback is a notable increase in
23426 code size. This switch implies -mno-push-args.
23427 -mthreads
23429 Support thread-safe exception handling on MinGW. Programs that
23430 rely on thread-safe exception handling must compile and link all
23431 code with the -mthreads option. When compiling, -mthreads defines
23432 -D_MT; when linking, it links in a special thread helper library
23433 -lmingwthrd which cleans up per-thread exception-handling data.
23434 -mms-bitfields
23436 -mno-ms-bitfields
23437 Enable/disable bit-field layout compatible with the native
23438 Microsoft Windows compiler.
23439 If "packed" is used on a structure, or if bit-fields are used, it
23441 may be that the Microsoft ABI lays out the structure differently
23442 than the way GCC normally does. Particularly when moving packed
23443 data between functions compiled with GCC and the native Microsoft
23444 compiler (either via function call or as data in a file), it may be
23445 necessary to access either format.
23446 This option is enabled by default for Microsoft Windows targets.
23448 This behavior can also be controlled locally by use of variable or
23449 type attributes. For more information, see x86 Variable Attributes
23450 and x86 Type Attributes.
23451 The Microsoft structure layout algorithm is fairly simple with the
23453 exception of the bit-field packing. The padding and alignment of
23454 members of structures and whether a bit-field can straddle a
23455 storage-unit boundary are determine by these rules:
23456 1. Structure members are stored sequentially in the order in which
23458 they are
23459 declared: the first member has the lowest memory address and
23460 the last member the highest.
23461 2. Every data object has an alignment requirement. The alignment
23463 requirement
23464 for all data except structures, unions, and arrays is either
23465 the size of the object or the current packing size (specified
23466 with either the "aligned" attribute or the "pack" pragma),
23467 whichever is less. For structures, unions, and arrays, the
23468 alignment requirement is the largest alignment requirement of
23469 its members. Every object is allocated an offset so that:
23470 offset % alignment_requirement == 0
23472 3. Adjacent bit-fields are packed into the same 1-, 2-, or 4-byte
23474 allocation
23475 unit if the integral types are the same size and if the next
23476 bit-field fits into the current allocation unit without
23477 crossing the boundary imposed by the common alignment
23478 requirements of the bit-fields.
23479 MSVC interprets zero-length bit-fields in the following ways:
23481 1. If a zero-length bit-field is inserted between two bit-fields
23483 that
23484 are normally coalesced, the bit-fields are not coalesced.
23485 For example:
23487 struct
23489 {
23490 unsigned long bf_1 : 12;
23491 unsigned long : 0;
23492 unsigned long bf_2 : 12;
23493 } t1;
23494 The size of "t1" is 8 bytes with the zero-length bit-field. If
23496 the zero-length bit-field were removed, "t1"'s size would be 4
23497 bytes.
23498 2. If a zero-length bit-field is inserted after a bit-field, "foo",
23500 and the
23501 alignment of the zero-length bit-field is greater than the
23502 member that follows it, "bar", "bar" is aligned as the type of
23503 the zero-length bit-field.
23504 For example:
23506 struct
23508 {
23509 char foo : 4;
23510 short : 0;
23511 char bar;
23512 } t2;
23513 struct
23515 {
23516 char foo : 4;
23517 short : 0;
23518 double bar;
23519 } t3;
23520 For "t2", "bar" is placed at offset 2, rather than offset 1.
23522 Accordingly, the size of "t2" is 4. For "t3", the zero-length
23523 bit-field does not affect the alignment of "bar" or, as a
23524 result, the size of the structure.
23525 Taking this into account, it is important to note the
23527 following:
23528 1. If a zero-length bit-field follows a normal bit-field, the
23530 type of the
23531 zero-length bit-field may affect the alignment of the
23532 structure as whole. For example, "t2" has a size of 4
23533 bytes, since the zero-length bit-field follows a normal
23534 bit-field, and is of type short.
23535 2. Even if a zero-length bit-field is not followed by a normal
23537 bit-field, it may
23538 still affect the alignment of the structure:
23539 struct
23541 {
23542 char foo : 6;
23543 long : 0;
23544 } t4;
23545 Here, "t4" takes up 4 bytes.
23547 3. Zero-length bit-fields following non-bit-field members are
23549 ignored:
23550 struct
23551 {
23552 char foo;
23553 long : 0;
23554 char bar;
23555 } t5;
23556 Here, "t5" takes up 2 bytes.
23558 -mno-align-stringops
23560 Do not align the destination of inlined string operations. This
23561 switch reduces code size and improves performance in case the
23562 destination is already aligned, but GCC doesn't know about it.
23563 -minline-all-stringops
23565 By default GCC inlines string operations only when the destination
23566 is known to be aligned to least a 4-byte boundary. This enables
23567 more inlining and increases code size, but may improve performance
23568 of code that depends on fast "memcpy", "strlen", and "memset" for
23569 short lengths.
23570 -minline-stringops-dynamically
23572 For string operations of unknown size, use run-time checks with
23573 inline code for small blocks and a library call for large blocks.
23574 -mstringop-strategy=alg
23576 Override the internal decision heuristic for the particular
23577 algorithm to use for inlining string operations. The allowed
23578 values for alg are:
23579 rep_byte
23581 rep_4byte
23582 rep_8byte
23583 Expand using i386 "rep" prefix of the specified size.
23584 byte_loop
23586 loop
23587 unrolled_loop
23588 Expand into an inline loop.
23589 libcall
23591 Always use a library call.
23592 -mmemcpy-strategy=strategy
23594 Override the internal decision heuristic to decide if
23595 "__builtin_memcpy" should be inlined and what inline algorithm to
23596 use when the expected size of the copy operation is known. strategy
23597 is a comma-separated list of alg:max_size:dest_align triplets. alg
23598 is specified in -mstringop-strategy, max_size specifies the max
23599 byte size with which inline algorithm alg is allowed. For the last
23600 triplet, the max_size must be "-1". The max_size of the triplets in
23601 the list must be specified in increasing order. The minimal byte
23602 size for alg is 0 for the first triplet and "max_size + 1" of the
23603 preceding range.
23604 -mmemset-strategy=strategy
23606 The option is similar to -mmemcpy-strategy= except that it is to
23607 control "__builtin_memset" expansion.
23608 -momit-leaf-frame-pointer
23610 Don't keep the frame pointer in a register for leaf functions.
23611 This avoids the instructions to save, set up, and restore frame
23612 pointers and makes an extra register available in leaf functions.
23613 The option -fomit-leaf-frame-pointer removes the frame pointer for
23614 leaf functions, which might make debugging harder.
23615 -mtls-direct-seg-refs
23617 -mno-tls-direct-seg-refs
23618 Controls whether TLS variables may be accessed with offsets from
23619 the TLS segment register (%gs for 32-bit, %fs for 64-bit), or
23620 whether the thread base pointer must be added. Whether or not this
23621 is valid depends on the operating system, and whether it maps the
23622 segment to cover the entire TLS area.
23623 For systems that use the GNU C Library, the default is on.
23625 -msse2avx
23627 -mno-sse2avx
23628 Specify that the assembler should encode SSE instructions with VEX
23629 prefix. The option -mavx turns this on by default.
23630 -mfentry
23632 -mno-fentry
23633 If profiling is active (-pg), put the profiling counter call before
23634 the prologue. Note: On x86 architectures the attribute
23635 "ms_hook_prologue" isn't possible at the moment for -mfentry and
23636 -pg.
23637 -mrecord-mcount
23639 -mno-record-mcount
23640 If profiling is active (-pg), generate a __mcount_loc section that
23641 contains pointers to each profiling call. This is useful for
23642 automatically patching and out calls.
23643 -mnop-mcount
23645 -mno-nop-mcount
23646 If profiling is active (-pg), generate the calls to the profiling
23647 functions as NOPs. This is useful when they should be patched in
23648 later dynamically. This is likely only useful together with
23649 -mrecord-mcount.
23650 -mskip-rax-setup
23652 -mno-skip-rax-setup
23653 When generating code for the x86-64 architecture with SSE
23654 extensions disabled, -mskip-rax-setup can be used to skip setting
23655 up RAX register when there are no variable arguments passed in
23656 vector registers.
23657 Warning: Since RAX register is used to avoid unnecessarily saving
23659 vector registers on stack when passing variable arguments, the
23660 impacts of this option are callees may waste some stack space,
23661 misbehave or jump to a random location. GCC 4.4 or newer don't
23662 have those issues, regardless the RAX register value.
23663 -m8bit-idiv
23665 -mno-8bit-idiv
23666 On some processors, like Intel Atom, 8-bit unsigned integer divide
23667 is much faster than 32-bit/64-bit integer divide. This option
23668 generates a run-time check. If both dividend and divisor are
23669 within range of 0 to 255, 8-bit unsigned integer divide is used
23670 instead of 32-bit/64-bit integer divide.
23671 -mavx256-split-unaligned-load
23673 -mavx256-split-unaligned-store
23674 Split 32-byte AVX unaligned load and store.
23675 -mstack-protector-guard=guard
23677 -mstack-protector-guard-reg=reg
23678 -mstack-protector-guard-offset=offset
23679 Generate stack protection code using canary at guard. Supported
23680 locations are global for global canary or tls for per-thread canary
23681 in the TLS block (the default). This option has effect only when
23682 -fstack-protector or -fstack-protector-all is specified.
23683 With the latter choice the options -mstack-protector-guard-reg=reg
23685 and -mstack-protector-guard-offset=offset furthermore specify which
23686 segment register (%fs or %gs) to use as base register for reading
23687 the canary, and from what offset from that base register. The
23688 default for those is as specified in the relevant ABI.
23689 -mmitigate-rop
23691 Try to avoid generating code sequences that contain unintended
23692 return opcodes, to mitigate against certain forms of attack. At the
23693 moment, this option is limited in what it can do and should not be
23694 relied on to provide serious protection.
23695 -mgeneral-regs-only
23697 Generate code that uses only the general-purpose registers. This
23698 prevents the compiler from using floating-point, vector, mask and
23699 bound registers.
23700 -mindirect-branch=choice
23702 Convert indirect call and jump with choice. The default is keep,
23703 which keeps indirect call and jump unmodified. thunk converts
23704 indirect call and jump to call and return thunk. thunk-inline
23705 converts indirect call and jump to inlined call and return thunk.
23706 thunk-extern converts indirect call and jump to external call and
23707 return thunk provided in a separate object file. You can control
23708 this behavior for a specific function by using the function
23709 attribute "indirect_branch".
23710 Note that -mcmodel=large is incompatible with
23712 -mindirect-branch=thunk and -mindirect-branch=thunk-extern since
23713 the thunk function may not be reachable in the large code model.
23714 Note that -mindirect-branch=thunk-extern is incompatible with
23716 -fcf-protection=branch and -fcheck-pointer-bounds since the
23717 external thunk can not be modified to disable control-flow check.
23718 -mfunction-return=choice
23720 Convert function return with choice. The default is keep, which
23721 keeps function return unmodified. thunk converts function return
23722 to call and return thunk. thunk-inline converts function return to
23723 inlined call and return thunk. thunk-extern converts function
23724 return to external call and return thunk provided in a separate
23725 object file. You can control this behavior for a specific function
23726 by using the function attribute "function_return".
23727 Note that -mcmodel=large is incompatible with
23729 -mfunction-return=thunk and -mfunction-return=thunk-extern since
23730 the thunk function may not be reachable in the large code model.
23731 -mindirect-branch-register
23733 Force indirect call and jump via register.
23734 These -m switches are supported in addition to the above on x86-64
23736 processors in 64-bit environments.
23737 -m32
23739 -m64
23740 -mx32
23741 -m16
23742 -miamcu
23743 Generate code for a 16-bit, 32-bit or 64-bit environment. The -m32
23744 option sets "int", "long", and pointer types to 32 bits, and
23745 generates code that runs on any i386 system.
23746 The -m64 option sets "int" to 32 bits and "long" and pointer types
23748 to 64 bits, and generates code for the x86-64 architecture. For
23749 Darwin only the -m64 option also turns off the -fno-pic and
23750 -mdynamic-no-pic options.
23751 The -mx32 option sets "int", "long", and pointer types to 32 bits,
23753 and generates code for the x86-64 architecture.
23754 The -m16 option is the same as -m32, except for that it outputs the
23756 ".code16gcc" assembly directive at the beginning of the assembly
23757 output so that the binary can run in 16-bit mode.
23758 The -miamcu option generates code which conforms to Intel MCU
23760 psABI. It requires the -m32 option to be turned on.
23761 -mno-red-zone
23763 Do not use a so-called "red zone" for x86-64 code. The red zone is
23764 mandated by the x86-64 ABI; it is a 128-byte area beyond the
23765 location of the stack pointer that is not modified by signal or
23766 interrupt handlers and therefore can be used for temporary data
23767 without adjusting the stack pointer. The flag -mno-red-zone
23768 disables this red zone.
23769 -mcmodel=small
23771 Generate code for the small code model: the program and its symbols
23772 must be linked in the lower 2 GB of the address space. Pointers
23773 are 64 bits. Programs can be statically or dynamically linked.
23774 This is the default code model.
23775 -mcmodel=kernel
23777 Generate code for the kernel code model. The kernel runs in the
23778 negative 2 GB of the address space. This model has to be used for
23779 Linux kernel code.
23780 -mcmodel=medium
23782 Generate code for the medium model: the program is linked in the
23783 lower 2 GB of the address space. Small symbols are also placed
23784 there. Symbols with sizes larger than -mlarge-data-threshold are
23785 put into large data or BSS sections and can be located above 2GB.
23786 Programs can be statically or dynamically linked.
23787 -mcmodel=large
23789 Generate code for the large model. This model makes no assumptions
23790 about addresses and sizes of sections.
23791 -maddress-mode=long
23793 Generate code for long address mode. This is only supported for
23794 64-bit and x32 environments. It is the default address mode for
23795 64-bit environments.
23796 -maddress-mode=short
23798 Generate code for short address mode. This is only supported for
23799 32-bit and x32 environments. It is the default address mode for
23800 32-bit and x32 environments.
23801 x86 Windows Options
23803 These additional options are available for Microsoft Windows targets:
23804 -mconsole
23806 This option specifies that a console application is to be
23807 generated, by instructing the linker to set the PE header subsystem
23808 type required for console applications. This option is available
23809 for Cygwin and MinGW targets and is enabled by default on those
23810 targets.
23811 -mdll
23813 This option is available for Cygwin and MinGW targets. It
23814 specifies that a DLL---a dynamic link library---is to be generated,
23815 enabling the selection of the required runtime startup object and
23816 entry point.
23817 -mnop-fun-dllimport
23819 This option is available for Cygwin and MinGW targets. It
23820 specifies that the "dllimport" attribute should be ignored.
23821 -mthread
23823 This option is available for MinGW targets. It specifies that
23824 MinGW-specific thread support is to be used.
23825 -municode
23827 This option is available for MinGW-w64 targets. It causes the
23828 "UNICODE" preprocessor macro to be predefined, and chooses Unicode-
23829 capable runtime startup code.
23830 -mwin32
23832 This option is available for Cygwin and MinGW targets. It
23833 specifies that the typical Microsoft Windows predefined macros are
23834 to be set in the pre-processor, but does not influence the choice
23835 of runtime library/startup code.
23836 -mwindows
23838 This option is available for Cygwin and MinGW targets. It
23839 specifies that a GUI application is to be generated by instructing
23840 the linker to set the PE header subsystem type appropriately.
23841 -fno-set-stack-executable
23843 This option is available for MinGW targets. It specifies that the
23844 executable flag for the stack used by nested functions isn't set.
23845 This is necessary for binaries running in kernel mode of Microsoft
23846 Windows, as there the User32 API, which is used to set executable
23847 privileges, isn't available.
23848 -fwritable-relocated-rdata
23850 This option is available for MinGW and Cygwin targets. It
23851 specifies that relocated-data in read-only section is put into the
23852 ".data" section. This is a necessary for older runtimes not
23853 supporting modification of ".rdata" sections for pseudo-relocation.
23854 -mpe-aligned-commons
23856 This option is available for Cygwin and MinGW targets. It
23857 specifies that the GNU extension to the PE file format that permits
23858 the correct alignment of COMMON variables should be used when
23859 generating code. It is enabled by default if GCC detects that the
23860 target assembler found during configuration supports the feature.
23861 See also under x86 Options for standard options.
23863 Xstormy16 Options
23865 These options are defined for Xstormy16:
23866 -msim
23868 Choose startup files and linker script suitable for the simulator.
23869 Xtensa Options
23871 These options are supported for Xtensa targets:
23872 -mconst16
23874 -mno-const16
23875 Enable or disable use of "CONST16" instructions for loading
23876 constant values. The "CONST16" instruction is currently not a
23877 standard option from Tensilica. When enabled, "CONST16"
23878 instructions are always used in place of the standard "L32R"
23879 instructions. The use of "CONST16" is enabled by default only if
23880 the "L32R" instruction is not available.
23881 -mfused-madd
23883 -mno-fused-madd
23884 Enable or disable use of fused multiply/add and multiply/subtract
23885 instructions in the floating-point option. This has no effect if
23886 the floating-point option is not also enabled. Disabling fused
23887 multiply/add and multiply/subtract instructions forces the compiler
23888 to use separate instructions for the multiply and add/subtract
23889 operations. This may be desirable in some cases where strict IEEE
23890 754-compliant results are required: the fused multiply add/subtract
23891 instructions do not round the intermediate result, thereby
23892 producing results with more bits of precision than specified by the
23893 IEEE standard. Disabling fused multiply add/subtract instructions
23894 also ensures that the program output is not sensitive to the
23895 compiler's ability to combine multiply and add/subtract operations.
23896 -mserialize-volatile
23898 -mno-serialize-volatile
23899 When this option is enabled, GCC inserts "MEMW" instructions before
23900 "volatile" memory references to guarantee sequential consistency.
23901 The default is -mserialize-volatile. Use -mno-serialize-volatile
23902 to omit the "MEMW" instructions.
23903 -mforce-no-pic
23905 For targets, like GNU/Linux, where all user-mode Xtensa code must
23906 be position-independent code (PIC), this option disables PIC for
23907 compiling kernel code.
23908 -mtext-section-literals
23910 -mno-text-section-literals
23911 These options control the treatment of literal pools. The default
23912 is -mno-text-section-literals, which places literals in a separate
23913 section in the output file. This allows the literal pool to be
23914 placed in a data RAM/ROM, and it also allows the linker to combine
23915 literal pools from separate object files to remove redundant
23916 literals and improve code size. With -mtext-section-literals, the
23917 literals are interspersed in the text section in order to keep them
23918 as close as possible to their references. This may be necessary
23919 for large assembly files. Literals for each function are placed
23920 right before that function.
23921 -mauto-litpools
23923 -mno-auto-litpools
23924 These options control the treatment of literal pools. The default
23925 is -mno-auto-litpools, which places literals in a separate section
23926 in the output file unless -mtext-section-literals is used. With
23927 -mauto-litpools the literals are interspersed in the text section
23928 by the assembler. Compiler does not produce explicit ".literal"
23929 directives and loads literals into registers with "MOVI"
23930 instructions instead of "L32R" to let the assembler do relaxation
23931 and place literals as necessary. This option allows assembler to
23932 create several literal pools per function and assemble very big
23933 functions, which may not be possible with -mtext-section-literals.
23934 -mtarget-align
23936 -mno-target-align
23937 When this option is enabled, GCC instructs the assembler to
23938 automatically align instructions to reduce branch penalties at the
23939 expense of some code density. The assembler attempts to widen
23940 density instructions to align branch targets and the instructions
23941 following call instructions. If there are not enough preceding
23942 safe density instructions to align a target, no widening is
23943 performed. The default is -mtarget-align. These options do not
23944 affect the treatment of auto-aligned instructions like "LOOP",
23945 which the assembler always aligns, either by widening density
23946 instructions or by inserting NOP instructions.
23947 -mlongcalls
23949 -mno-longcalls
23950 When this option is enabled, GCC instructs the assembler to
23951 translate direct calls to indirect calls unless it can determine
23952 that the target of a direct call is in the range allowed by the
23953 call instruction. This translation typically occurs for calls to
23954 functions in other source files. Specifically, the assembler
23955 translates a direct "CALL" instruction into an "L32R" followed by a
23956 "CALLX" instruction. The default is -mno-longcalls. This option
23957 should be used in programs where the call target can potentially be
23958 out of range. This option is implemented in the assembler, not the
23959 compiler, so the assembly code generated by GCC still shows direct
23960 call instructions---look at the disassembled object code to see the
23961 actual instructions. Note that the assembler uses an indirect call
23962 for every cross-file call, not just those that really are out of
23963 range.
23964 zSeries Options
23966 These are listed under
23967
23969 This section describes several environment variables that affect how
23970 GCC operates. Some of them work by specifying directories or prefixes
23971 to use when searching for various kinds of files. Some are used to
23972 specify other aspects of the compilation environment.
23973 Note that you can also specify places to search using options such as
23975 -B, -I and -L. These take precedence over places specified using
23976 environment variables, which in turn take precedence over those
23977 specified by the configuration of GCC.
23978 LANG
23980 LC_CTYPE
23981 LC_MESSAGES
23982 LC_ALL
23983 These environment variables control the way that GCC uses
23984 localization information which allows GCC to work with different
23985 national conventions. GCC inspects the locale categories LC_CTYPE
23986 and LC_MESSAGES if it has been configured to do so. These locale
23987 categories can be set to any value supported by your installation.
23988 A typical value is en_GB.UTF-8 for English in the United Kingdom
23989 encoded in UTF-8.
23990 The LC_CTYPE environment variable specifies character
23992 classification. GCC uses it to determine the character boundaries
23993 in a string; this is needed for some multibyte encodings that
23994 contain quote and escape characters that are otherwise interpreted
23995 as a string end or escape.
23996 The LC_MESSAGES environment variable specifies the language to use
23998 in diagnostic messages.
23999 If the LC_ALL environment variable is set, it overrides the value
24001 of LC_CTYPE and LC_MESSAGES; otherwise, LC_CTYPE and LC_MESSAGES
24002 default to the value of the LANG environment variable. If none of
24003 these variables are set, GCC defaults to traditional C English
24004 behavior.
24005 TMPDIR
24007 If TMPDIR is set, it specifies the directory to use for temporary
24008 files. GCC uses temporary files to hold the output of one stage of
24009 compilation which is to be used as input to the next stage: for
24010 example, the output of the preprocessor, which is the input to the
24011 compiler proper.
24012 GCC_COMPARE_DEBUG
24014 Setting GCC_COMPARE_DEBUG is nearly equivalent to passing
24015 -fcompare-debug to the compiler driver. See the documentation of
24016 this option for more details.
24017 GCC_EXEC_PREFIX
24019 If GCC_EXEC_PREFIX is set, it specifies a prefix to use in the
24020 names of the subprograms executed by the compiler. No slash is
24021 added when this prefix is combined with the name of a subprogram,
24022 but you can specify a prefix that ends with a slash if you wish.
24023 If GCC_EXEC_PREFIX is not set, GCC attempts to figure out an
24025 appropriate prefix to use based on the pathname it is invoked with.
24026 If GCC cannot find the subprogram using the specified prefix, it
24028 tries looking in the usual places for the subprogram.
24029 The default value of GCC_EXEC_PREFIX is prefix/lib/gcc/ where
24031 prefix is the prefix to the installed compiler. In many cases
24032 prefix is the value of "prefix" when you ran the configure script.
24033 Other prefixes specified with -B take precedence over this prefix.
24035 This prefix is also used for finding files such as crt0.o that are
24037 used for linking.
24038 In addition, the prefix is used in an unusual way in finding the
24040 directories to search for header files. For each of the standard
24041 directories whose name normally begins with /usr/local/lib/gcc
24042 (more precisely, with the value of GCC_INCLUDE_DIR), GCC tries
24043 replacing that beginning with the specified prefix to produce an
24044 alternate directory name. Thus, with -Bfoo/, GCC searches foo/bar
24045 just before it searches the standard directory /usr/local/lib/bar.
24046 If a standard directory begins with the configured prefix then the
24047 value of prefix is replaced by GCC_EXEC_PREFIX when looking for
24048 header files.
24049 COMPILER_PATH
24051 The value of COMPILER_PATH is a colon-separated list of
24052 directories, much like PATH. GCC tries the directories thus
24053 specified when searching for subprograms, if it cannot find the
24054 subprograms using GCC_EXEC_PREFIX.
24055 LIBRARY_PATH
24057 The value of LIBRARY_PATH is a colon-separated list of directories,
24058 much like PATH. When configured as a native compiler, GCC tries
24059 the directories thus specified when searching for special linker
24060 files, if it cannot find them using GCC_EXEC_PREFIX. Linking using
24061 GCC also uses these directories when searching for ordinary
24062 libraries for the -l option (but directories specified with -L come
24063 first).
24064 LANG
24066 This variable is used to pass locale information to the compiler.
24067 One way in which this information is used is to determine the
24068 character set to be used when character literals, string literals
24069 and comments are parsed in C and C++. When the compiler is
24070 configured to allow multibyte characters, the following values for
24071 LANG are recognized:
24072 C-JIS
24074 Recognize JIS characters.
24075 C-SJIS
24077 Recognize SJIS characters.
24078 C-EUCJP
24080 Recognize EUCJP characters.
24081 If LANG is not defined, or if it has some other value, then the
24083 compiler uses "mblen" and "mbtowc" as defined by the default locale
24084 to recognize and translate multibyte characters.
24085 Some additional environment variables affect the behavior of the
24087 preprocessor.
24088 CPATH
24090 C_INCLUDE_PATH
24091 CPLUS_INCLUDE_PATH
24092 OBJC_INCLUDE_PATH
24093 Each variable's value is a list of directories separated by a
24094 special character, much like PATH, in which to look for header
24095 files. The special character, "PATH_SEPARATOR", is target-
24096 dependent and determined at GCC build time. For Microsoft Windows-
24097 based targets it is a semicolon, and for almost all other targets
24098 it is a colon.
24099 CPATH specifies a list of directories to be searched as if
24101 specified with -I, but after any paths given with -I options on the
24102 command line. This environment variable is used regardless of
24103 which language is being preprocessed.
24104 The remaining environment variables apply only when preprocessing
24106 the particular language indicated. Each specifies a list of
24107 directories to be searched as if specified with -isystem, but after
24108 any paths given with -isystem options on the command line.
24109 In all these variables, an empty element instructs the compiler to
24111 search its current working directory. Empty elements can appear at
24112 the beginning or end of a path. For instance, if the value of
24113 CPATH is ":/special/include", that has the same effect as
24114 -I. -I/special/include.
24115 DEPENDENCIES_OUTPUT
24117 If this variable is set, its value specifies how to output
24118 dependencies for Make based on the non-system header files
24119 processed by the compiler. System header files are ignored in the
24120 dependency output.
24121 The value of DEPENDENCIES_OUTPUT can be just a file name, in which
24123 case the Make rules are written to that file, guessing the target
24124 name from the source file name. Or the value can have the form
24125 file target, in which case the rules are written to file file using
24126 target as the target name.
24127 In other words, this environment variable is equivalent to
24129 combining the options -MM and -MF, with an optional -MT switch too.
24130 SUNPRO_DEPENDENCIES
24132 This variable is the same as DEPENDENCIES_OUTPUT (see above),
24133 except that system header files are not ignored, so it implies -M
24134 rather than -MM. However, the dependence on the main input file is
24135 omitted.
24136 SOURCE_DATE_EPOCH
24138 If this variable is set, its value specifies a UNIX timestamp to be
24139 used in replacement of the current date and time in the "__DATE__"
24140 and "__TIME__" macros, so that the embedded timestamps become
24141 reproducible.
24142 The value of SOURCE_DATE_EPOCH must be a UNIX timestamp, defined as
24144 the number of seconds (excluding leap seconds) since 01 Jan 1970
24145 00:00:00 represented in ASCII; identical to the output of
24146 @command{date +%s} on GNU/Linux and other systems that support the
24147 %s extension in the "date" command.
24148 The value should be a known timestamp such as the last modification
24150 time of the source or package and it should be set by the build
24151 process.
24152
24154 For instructions on reporting bugs, see
24155 <http://bugzilla.redhat.com/bugzilla>.
24156
24158 1. On some systems, gcc -shared needs to build supplementary stub code
24159 for constructors to work. On multi-libbed systems, gcc -shared
24160 must select the correct support libraries to link against. Failing
24161 to supply the correct flags may lead to subtle defects. Supplying
24162 them in cases where they are not necessary is innocuous.
24163
24165 gpl(7), gfdl(7), fsf-funding(7), cpp(1), gcov(1), as(1), ld(1), gdb(1),
24166 dbx(1) and the Info entries for gcc, cpp, as, ld, binutils and gdb.
24167
24169 See the Info entry for gcc, or
24170 <http://gcc.gnu.org/onlinedocs/gcc/Contributors.html>, for contributors
24171 to GCC.
24172
24174 Copyright (c) 1988-2018 Free Software Foundation, Inc.
24175 Permission is granted to copy, distribute and/or modify this document
24177 under the terms of the GNU Free Documentation License, Version 1.3 or
24178 any later version published by the Free Software Foundation; with the
24179 Invariant Sections being "GNU General Public License" and "Funding Free
24180 Software", the Front-Cover texts being (a) (see below), and with the
24181 Back-Cover Texts being (b) (see below). A copy of the license is
24182 included in the gfdl(7) man page.
24183 (a) The FSF's Front-Cover Text is:
24185 A GNU Manual
24187 (b) The FSF's Back-Cover Text is:
24189 You have freedom to copy and modify this GNU Manual, like GNU
24191 software. Copies published by the Free Software Foundation raise
24192 funds for GNU development.
24193gcc-8 2018-09-05 GCC(1)