GCC Manual

GCC(1)                                GNU                               GCC(1)


NAME
       gcc - GNU project C and C++ compiler


SYNOPSIS
       gcc [-c│-S│-E] [-std=standard]
           [-g] [-pg] [-Olevel]
           [-Wwarn...] [-pedantic]
           [-Idir...] [-Ldir...]
           [-Dmacro[=defn]...] [-Umacro]
           [-foption...] [-mmachine-option...]
           [-o outfile] infile...


       Only the most useful options are listed here; see below for the remainder.  g++ accepts mostly the same options
       as gcc.


DESCRIPTION
       When you invoke GCC, it normally does preprocessing, compilation, assembly and linking.  The "overall options"
       allow you to stop this process at an intermediate stage.  For example, the -c option says not to run the
       linker.  Then the output consists of object files output by the assembler.


       Other options are passed on to one stage of processing.  Some options control the preprocessor and others the
       compiler itself.  Yet other options control the assembler and linker; most of these are not documented here,
       since you rarely need to use any of them.


       Most of the command line options that you can use with GCC are useful for C programs; when an option is only
       useful with another language (usually C++), the explanation says so explicitly.  If the description for a par-
       ticular option does not mention a source language, you can use that option with all supported languages.


       The gcc program accepts options and file names as operands.  Many options have multi-letter names; therefore
       multiple single-letter options may not be grouped: -dr is very different from -d -r.


       You can mix options and other arguments.  For the most part, the order you use doesn’t matter.  Order does mat-
       ter when you use several options of the same kind; for example, if you specify -L more than once, the directo-
       ries are searched in the order specified.


       Many options have long names starting with -f or with -W---for example, -fstrength-reduce, -Wformat and so on.
       Most of these have both positive and negative forms; the negative form of -ffoo would be -fno-foo.  This manual
       documents only one of these two forms, whichever one is not the default.


OPTIONS
       Option Summary


       Here is a summary of all the options, grouped by type.  Explanations are in the following sections.


       Overall Options
           -c  -S  -E  -o file  -combine -pipe  -pass-exit-codes -x language  -v  -###  --help  --target-help  --ver-
           sion


       C Language Options
           -ansi  -std=standard  -fgnu89-inline -aux-info filename -fno-asm  -fno-builtin  -fno-builtin-function
           -fhosted  -ffreestanding  -fms-extensions -trigraphs  -no-integrated-cpp  -traditional  -traditional-cpp
           -fallow-single-precision  -fcond-mismatch -fsigned-bitfields  -fsigned-char -funsigned-bitfields  -fun-
           signed-char


       C++ Language Options
           -fabi-version=n  -fno-access-control  -fcheck-new -fconserve-space  -ffriend-injection  -fno-const-strings
           -fno-elide-constructors -fno-enforce-eh-specs -ffor-scope  -fno-for-scope  -fno-gnu-keywords
           -fno-implicit-templates -fno-implicit-inline-templates -fno-implement-inlines  -fms-extensions
           -fno-nonansi-builtins  -fno-operator-names -fno-optional-diags  -fpermissive -frepo  -fno-rtti  -fstats
           -ftemplate-depth-n -fno-threadsafe-statics -fuse-cxa-atexit  -fno-weak  -nostdinc++ -fno-default-inline
           -fvisibility-inlines-hidden -Wabi  -Wctor-dtor-privacy -Wnon-virtual-dtor  -Wreorder -Weffc++  -Wno-depre-
           cated  -Wstrict-null-sentinel -Wno-non-template-friend  -Wold-style-cast -Woverloaded-virtual
           -Wno-pmf-conversions -Wsign-promo


       Objective-C and Objective-C++ Language Options
           -fconstant-string-class=class-name -fgnu-runtime  -fnext-runtime -fno-nil-receivers -fobjc-call-cxx-cdtors
           -fobjc-direct-dispatch -fobjc-exceptions -fobjc-gc -freplace-objc-classes -fzero-link -gen-decls -Was-
           sign-intercept -Wno-protocol  -Wselector -Wstrict-selector-match -Wundeclared-selector


       Language Independent Options
           -fmessage-length=n -fdiagnostics-show-location=[once│every-line] -fdiagnostics-show-options


       Warning Options
           -fsyntax-only  -pedantic  -pedantic-errors -w  -Wextra  -Wall  -Waggregate-return -Wno-attributes
           -Wc++-compat -Wcast-align  -Wcast-qual  -Wchar-subscripts  -Wcomment -Wconversion  -Wno-deprecated-declara-
           tions -Wdisabled-optimization  -Wno-div-by-zero  -Wno-endif-labels -Werror  -Werror-implicit-function-dec-
           laration -Wfatal-errors  -Wfloat-equal  -Wformat  -Wformat=2 -Wno-format-extra-args -Wformat-nonliteral
           -Wformat-security  -Wformat-y2k -Wimplicit  -Wimplicit-function-declaration  -Wimplicit-int -Wimport
           -Wno-import  -Winit-self  -Winline -Wno-int-to-pointer-cast -Wno-invalid-offsetof  -Winvalid-pch
           -Wlarger-than-len  -Wunsafe-loop-optimizations  -Wlong-long -Wmain  -Wmissing-braces  -Wmissing-field-ini-
           tializers -Wmissing-format-attribute  -Wmissing-include-dirs -Wmissing-noreturn -Wno-multichar  -Wnonnull
           -Wpacked  -Wpadded -Wparentheses  -Wpointer-arith  -Wno-pointer-to-int-cast -Wredundant-decls -Wreturn-type
           -Wsequence-point  -Wshadow -Wsign-compare  -Wstack-protector -Wstrict-aliasing -Wstrict-aliasing=2 -Wswitch
           -Wswitch-default  -Wswitch-enum -Wsystem-headers  -Wtrigraphs  -Wundef  -Wuninitialized -Wunknown-pragmas
           -Wno-pragmas -Wunreachable-code -Wunused  -Wunused-function  -Wunused-label  -Wunused-parameter
           -Wunused-value  -Wunused-variable  -Wvariadic-macros -Wvolatile-register-var  -Wwrite-strings


       C-only Warning Options
           -Wbad-function-cast  -Wmissing-declarations -Wmissing-prototypes  -Wnested-externs  -Wold-style-definition
           -Wstrict-prototypes  -Wtraditional -Wdeclaration-after-statement -Wpointer-sign


       Debugging Options
           -dletters  -dumpspecs  -dumpmachine  -dumpversion -fdump-unnumbered  -fdump-translation-unit[-n]
           -fdump-class-hierarchy[-n] -fdump-ipa-all -fdump-ipa-cgraph -fdump-tree-all -fdump-tree-original[-n]
           -fdump-tree-optimized[-n] -fdump-tree-inlined[-n] -fdump-tree-cfg -fdump-tree-vcg -fdump-tree-alias
           -fdump-tree-ch -fdump-tree-ssa[-n] -fdump-tree-pre[-n] -fdump-tree-ccp[-n] -fdump-tree-dce[-n]
           -fdump-tree-gimple[-raw] -fdump-tree-mudflap[-n] -fdump-tree-dom[-n] -fdump-tree-dse[-n]
           -fdump-tree-phiopt[-n] -fdump-tree-forwprop[-n] -fdump-tree-copyrename[-n] -fdump-tree-nrv -fdump-tree-vect
           -fdump-tree-sink -fdump-tree-sra[-n] -fdump-tree-salias -fdump-tree-fre[-n] -fdump-tree-vrp[-n] -ftree-vec-
           torizer-verbose=n -fdump-tree-storeccp[-n] -feliminate-dwarf2-dups -feliminate-unused-debug-types -felimi-
           nate-unused-debug-symbols -fmem-report -fprofile-arcs -frandom-seed=string -fsched-verbose=n -ftest-cover-
           age  -ftime-report -fvar-tracking -g  -glevel  -gcoff -gdwarf-2 -ggdb  -gstabs  -gstabs+  -gvms  -gxcoff
           -gxcoff+ -p  -pg  -print-file-name=library  -print-libgcc-file-name -print-multi-directory
           -print-multi-lib  -print-multi-os-directory -print-prog-name=program  -print-search-dirs  -Q -save-temps
           -time


       Optimization Options
           -falign-functions[=n]  -falign-jumps[=n] -falign-labels[=n]  -falign-loops[=n] -fmudflap -fmudflapth -fmud-
           flapir -fbranch-probabilities -fprofile-values -fvpt -fbranch-target-load-optimize -fbranch-tar-
           get-load-optimize2 -fbtr-bb-exclusive -fcaller-saves  -fcprop-registers  -fcse-follow-jumps
           -fcse-skip-blocks  -fcx-limited-range  -fdata-sections -fdelayed-branch  -fdelete-null-pointer-checks
           -fearly-inlining -fexpensive-optimizations  -ffast-math  -ffloat-store -fforce-addr  -ffunction-sections
           -fgcse  -fgcse-lm  -fgcse-sm  -fgcse-las  -fgcse-after-reload -floop-optimize -fcrossjumping  -fif-conver-
           sion  -fif-conversion2 -finline-functions  -finline-functions-called-once -finline-limit=n
           -fkeep-inline-functions -fkeep-static-consts  -fmerge-constants  -fmerge-all-constants -fmodulo-sched
           -fno-branch-count-reg -fno-default-inline  -fno-defer-pop -floop-optimize2 -fmove-loop-invariants
           -fno-function-cse  -fno-guess-branch-probability -fno-inline  -fno-math-errno  -fno-peephole  -fno-peep-
           hole2 -funsafe-math-optimizations  -funsafe-loop-optimizations  -ffinite-math-only -fno-trapping-math
           -fno-zero-initialized-in-bss -fomit-frame-pointer  -foptimize-register-move -foptimize-sibling-calls
           -fprefetch-loop-arrays -fprofile-generate -fprofile-use -fregmove  -frename-registers -freorder-blocks
           -freorder-blocks-and-partition -freorder-functions -frerun-cse-after-loop  -frerun-loop-opt -frounding-math
           -fschedule-insns  -fschedule-insns2 -fno-sched-interblock  -fno-sched-spec  -fsched-spec-load
           -fsched-spec-load-dangerous -fsched-stalled-insns[=n] -fsched-stalled-insns-dep[=n]
           -fsched2-use-superblocks -fsched2-use-traces -freschedule-modulo-scheduled-loops -fsignaling-nans -fsin-
           gle-precision-constant -fstack-protector  -fstack-protector-all -fstrength-reduce  -fstrict-aliasing
           -ftracer  -fthread-jumps -funroll-all-loops  -funroll-loops  -fpeel-loops -fsplit-ivs-in-unroller
           -funswitch-loops -fvariable-expansion-in-unroller -ftree-pre  -ftree-ccp  -ftree-dce -ftree-loop-optimize
           -ftree-loop-linear -ftree-loop-im -ftree-loop-ivcanon -fivopts -ftree-dominator-opts -ftree-dse
           -ftree-copyrename -ftree-sink -ftree-ch -ftree-sra -ftree-ter -ftree-lrs -ftree-fre -ftree-vectorize
           -ftree-vect-loop-version -ftree-salias -fweb -ftree-copy-prop -ftree-store-ccp -ftree-store-copy-prop
           -ftree-vrp -funit-at-a-time -fwhole-program --param name=value -O  -O0  -O1  -O2  -O3  -Os


       Preprocessor Options
           -Aquestion=answer -A-question[=answer] -C  -dD  -dI  -dM  -dN -Dmacro[=defn]  -E  -H -idirafter dir
           -include file  -imacros file -iprefix file  -iwithprefix dir -iwithprefixbefore dir  -isystem dir -isysroot
           dir -M  -MM  -MF  -MG  -MP  -MQ  -MT  -nostdinc -P  -fworking-directory  -remap -trigraphs  -undef  -Umacro
           -Wp,option -Xpreprocessor option


       Assembler Option
           -Wa,option  -Xassembler option


       Linker Options
           object-file-name  -llibrary -nostartfiles  -nodefaultlibs  -nostdlib -pie -rdynamic -s  -static
           -static-libgcc  -shared  -shared-libgcc  -symbolic -Wl,option  -Xlinker option -u symbol


       Directory Options
           -Bprefix  -Idir  -iquotedir  -Ldir -specs=file  -I- --sysroot=dir


       Target Options
           -V version  -b machine


       Machine Dependent Options
           ARC Options -EB  -EL -mmangle-cpu  -mcpu=cpu  -mtext=text-section -mdata=data-section  -mrodata=readonly-
           data-section


           ARM Options -mapcs-frame  -mno-apcs-frame -mabi=name -mapcs-stack-check  -mno-apcs-stack-check -mapcs-float
           -mno-apcs-float -mapcs-reentrant  -mno-apcs-reentrant -msched-prolog  -mno-sched-prolog -mlittle-endian
           -mbig-endian  -mwords-little-endian -mfloat-abi=name  -msoft-float  -mhard-float  -mfpe -mthumb-interwork
           -mno-thumb-interwork -mcpu=name  -march=name  -mfpu=name -mstructure-size-boundary=n -mabort-on-noreturn
           -mlong-calls  -mno-long-calls -msingle-pic-base  -mno-single-pic-base -mpic-register=reg
           -mnop-fun-dllimport -mcirrus-fix-invalid-insns -mno-cirrus-fix-invalid-insns -mpoke-function-name -mthumb
           -marm -mtpcs-frame  -mtpcs-leaf-frame -mcaller-super-interworking  -mcallee-super-interworking -mtp=name


           AVR Options -mmcu=mcu  -msize  -minit-stack=n  -mno-interrupts -mcall-prologues  -mno-tablejump
           -mtiny-stack  -mint8


           Blackfin Options -momit-leaf-frame-pointer -mno-omit-leaf-frame-pointer -mspecld-anomaly
           -mno-specld-anomaly -mcsync-anomaly -mno-csync-anomaly -mlow-64k -mno-low64k -mid-shared-library
           -mno-id-shared-library -mshared-library-id=n -mlong-calls  -mno-long-calls


           CRIS Options -mcpu=cpu  -march=cpu  -mtune=cpu -mmax-stack-frame=n  -melinux-stacksize=n -metrax4
           -metrax100  -mpdebug  -mcc-init  -mno-side-effects -mstack-align  -mdata-align  -mconst-align -m32-bit
           -m16-bit  -m8-bit  -mno-prologue-epilogue  -mno-gotplt -melf  -maout  -melinux  -mlinux  -sim  -sim2
           -mmul-bug-workaround  -mno-mul-bug-workaround


           CRX Options -mmac -mpush-args


           Darwin Options -all_load  -allowable_client  -arch  -arch_errors_fatal -arch_only  -bind_at_load  -bundle
           -bundle_loader -client_name  -compatibility_version  -current_version -dead_strip -dependency-file
           -dylib_file  -dylinker_install_name -dynamic  -dynamiclib  -exported_symbols_list -filelist  -flat_names-
           pace  -force_cpusubtype_ALL -force_flat_namespace  -headerpad_max_install_names -image_base  -init
           -install_name  -keep_private_externs -multi_module  -multiply_defined  -multiply_defined_unused -noall_load
           -no_dead_strip_inits_and_terms -nofixprebinding -nomultidefs  -noprebind  -noseglinkedit -pagezero_size
           -prebind  -prebind_all_twolevel_modules -private_bundle  -read_only_relocs  -sectalign -sectobjectsymbols
           -whyload  -seg1addr -sectcreate  -sectobjectsymbols  -sectorder -segaddr -segs_read_only_addr
           -segs_read_write_addr -seg_addr_table  -seg_addr_table_filename  -seglinkedit -segprot
           -segs_read_only_addr  -segs_read_write_addr -single_module  -static  -sub_library  -sub_umbrella
           -twolevel_namespace  -umbrella  -undefined -unexported_symbols_list  -weak_reference_mismatches -what-
           sloaded -F -gused -gfull -mmacosx-version-min=version -mone-byte-bool


           DEC Alpha Options -mno-fp-regs  -msoft-float  -malpha-as  -mgas -mieee  -mieee-with-inexact  -mieee-confor-
           mant -mfp-trap-mode=mode  -mfp-rounding-mode=mode -mtrap-precision=mode  -mbuild-constants -mcpu=cpu-type
           -mtune=cpu-type -mbwx  -mmax  -mfix  -mcix -mfloat-vax  -mfloat-ieee -mexplicit-relocs  -msmall-data
           -mlarge-data -msmall-text  -mlarge-text -mmemory-latency=time


           DEC Alpha/VMS Options -mvms-return-codes


           FRV Options -mgpr-32  -mgpr-64  -mfpr-32  -mfpr-64 -mhard-float  -msoft-float -malloc-cc  -mfixed-cc
           -mdword  -mno-dword -mdouble  -mno-double -mmedia  -mno-media  -mmuladd  -mno-muladd -mfdpic  -minline-plt
           -mgprel-ro  -multilib-library-pic -mlinked-fp  -mlong-calls  -malign-labels -mlibrary-pic  -macc-4  -macc-8
           -mpack  -mno-pack  -mno-eflags  -mcond-move  -mno-cond-move -moptimize-membar -mno-optimize-membar -mscc
           -mno-scc  -mcond-exec  -mno-cond-exec -mvliw-branch  -mno-vliw-branch -mmulti-cond-exec
           -mno-multi-cond-exec  -mnested-cond-exec -mno-nested-cond-exec  -mtomcat-stats -mTLS -mtls -mcpu=cpu


           H8/300 Options -mrelax  -mh  -ms  -mn  -mint32  -malign-300


           HPPA Options -march=architecture-type -mbig-switch  -mdisable-fpregs  -mdisable-indexing -mfast-indi-
           rect-calls  -mgas  -mgnu-ld   -mhp-ld -mfixed-range=register-range -mjump-in-delay -mlinker-opt
           -mlong-calls -mlong-load-store  -mno-big-switch  -mno-disable-fpregs -mno-disable-indexing  -mno-fast-indi-
           rect-calls  -mno-gas -mno-jump-in-delay  -mno-long-load-store -mno-portable-runtime  -mno-soft-float
           -mno-space-regs  -msoft-float  -mpa-risc-1-0 -mpa-risc-1-1  -mpa-risc-2-0  -mportable-runtime -msched-
           ule=cpu-type  -mspace-regs  -msio  -mwsio -munix=unix-std  -nolibdld  -static  -threads


           i386 and x86-64 Options -mtune=cpu-type  -march=cpu-type -mfpmath=unit -masm=dialect  -mno-fancy-math-387
           -mno-fp-ret-in-387  -msoft-float  -msvr3-shlib -mno-wide-multiply  -mrtd  -malign-double -mpre-
           ferred-stack-boundary=num -mmmx  -msse  -msse2 -msse3 -mssse3 -msse4a -msse5 -m3dnow -mpopcnt -mabm
           -mthreads  -mno-align-stringops  -minline-all-stringops -mpush-args  -maccumulate-outgoing-args
           -m128bit-long-double -m96bit-long-double  -mregparm=num  -msseregparm -momit-leaf-frame-pointer
           -mno-red-zone -mno-tls-direct-seg-refs -mcmodel=code-model -m32  -m64 -mlarge-data-threshold=num
           -mfused-madd -mno-fused-madd


           IA-64 Options -mbig-endian  -mlittle-endian  -mgnu-as  -mgnu-ld  -mno-pic -mvolatile-asm-stop  -mregis-
           ter-names  -mno-sdata -mconstant-gp  -mauto-pic  -minline-float-divide-min-latency -min-
           line-float-divide-max-throughput -minline-int-divide-min-latency -minline-int-divide-max-throughput -min-
           line-sqrt-min-latency -minline-sqrt-max-throughput -mno-dwarf2-asm -mearly-stop-bits -mfixed-range=regis-
           ter-range -mtls-size=tls-size -mtune=cpu-type -mt -pthread -milp32 -mlp64


           M32R/D Options -m32r2 -m32rx -m32r -mdebug -malign-loops -mno-align-loops -missue-rate=number
           -mbranch-cost=number -mmodel=code-size-model-type -msdata=sdata-type -mno-flush-func -mflush-func=name
           -mno-flush-trap -mflush-trap=number -G num


           M32C Options -mcpu=cpu -msim -memregs=number


           M680x0 Options -m68000  -m68020  -m68020-40  -m68020-60  -m68030  -m68040 -m68060  -mcpu32  -m5200  -m68881
           -mbitfield  -mc68000  -mc68020 -mnobitfield  -mrtd  -mshort  -msoft-float  -mpcrel -malign-int
           -mstrict-align  -msep-data  -mno-sep-data -mshared-library-id=n  -mid-shared-library
           -mno-id-shared-library


           M68hc1x Options -m6811  -m6812  -m68hc11  -m68hc12   -m68hcs12 -mauto-incdec  -minmax  -mlong-calls
           -mshort -msoft-reg-count=count


           MCore Options -mhardlit  -mno-hardlit  -mdiv  -mno-div  -mrelax-immediates -mno-relax-immediates
           -mwide-bitfields  -mno-wide-bitfields -m4byte-functions  -mno-4byte-functions  -mcallgraph-data -mno-call-
           graph-data  -mslow-bytes  -mno-slow-bytes  -mno-lsim -mlittle-endian  -mbig-endian  -m210  -m340
           -mstack-increment


           MIPS Options -EL  -EB  -march=arch  -mtune=arch -mips1  -mips2  -mips3  -mips4  -mips32  -mips32r2  -mips64
           -mips16  -mno-mips16  -mabi=abi  -mabicalls  -mno-abicalls -mxgot  -mno-xgot  -mgp32  -mgp64  -mfp32
           -mfp64 -mhard-float  -msoft-float  -msingle-float  -mdouble-float -mdsp  -mpaired-single  -mips3d -mlong64
           -mlong32  -msym32  -mno-sym32 -Gnum  -membedded-data  -mno-embedded-data -muninit-const-in-rodata
           -mno-uninit-const-in-rodata -msplit-addresses  -mno-split-addresses -mexplicit-relocs  -mno-explicit-relocs
           -mcheck-zero-division  -mno-check-zero-division -mdivide-traps  -mdivide-breaks -mmemcpy  -mno-memcpy
           -mlong-calls  -mno-long-calls -mmad  -mno-mad  -mfused-madd  -mno-fused-madd  -nocpp -mfix-r4000
           -mno-fix-r4000  -mfix-r4400  -mno-fix-r4400 -mfix-vr4120  -mno-fix-vr4120  -mfix-vr4130 -mfix-sb1
           -mno-fix-sb1 -mflush-func=func  -mno-flush-func -mbranch-likely  -mno-branch-likely -mfp-exceptions
           -mno-fp-exceptions -mvr4130-align -mno-vr4130-align


           MMIX Options -mlibfuncs  -mno-libfuncs  -mepsilon  -mno-epsilon  -mabi=gnu -mabi=mmixware  -mzero-extend
           -mknuthdiv  -mtoplevel-symbols -melf  -mbranch-predict  -mno-branch-predict  -mbase-addresses
           -mno-base-addresses  -msingle-exit  -mno-single-exit


           MN10300 Options -mmult-bug  -mno-mult-bug -mam33  -mno-am33 -mam33-2  -mno-am33-2 -mreturn-pointer-on-d0
           -mno-crt0  -mrelax


           MT Options -mno-crt0 -mbacc -msim -march=cpu-type


           PDP-11 Options -mfpu  -msoft-float  -mac0  -mno-ac0  -m40  -m45  -m10 -mbcopy  -mbcopy-builtin  -mint32
           -mno-int16 -mint16  -mno-int32  -mfloat32  -mno-float64 -mfloat64  -mno-float32  -mabshi  -mno-abshi
           -mbranch-expensive  -mbranch-cheap -msplit  -mno-split  -munix-asm  -mdec-asm


           PowerPC Options See RS/6000 and PowerPC Options.


           RS/6000 and PowerPC Options -mcpu=cpu-type -mtune=cpu-type -mpower  -mno-power  -mpower2  -mno-power2
           -mpowerpc  -mpowerpc64  -mno-powerpc -maltivec  -mno-altivec -mpowerpc-gpopt  -mno-powerpc-gpopt -mpow-
           erpc-gfxopt  -mno-powerpc-gfxopt -mmfcrf  -mno-mfcrf  -mpopcntb  -mno-popcntb  -mfprnd  -mno-fprnd -mmfpgpr
           -mno-mfpgpr -mnew-mnemonics  -mold-mnemonics -mfull-toc   -mminimal-toc  -mno-fp-in-toc  -mno-sum-in-toc
           -m64  -m32  -mxl-compat  -mno-xl-compat  -mpe -malign-power  -malign-natural -msoft-float  -mhard-float
           -mmultiple  -mno-multiple -mstring  -mno-string  -mupdate  -mno-update -mfused-madd  -mno-fused-madd
           -mbit-align  -mno-bit-align -mstrict-align  -mno-strict-align  -mrelocatable -mno-relocatable  -mrelocat-
           able-lib  -mno-relocatable-lib -mtoc  -mno-toc  -mlittle  -mlittle-endian  -mbig  -mbig-endian -mdy-
           namic-no-pic  -maltivec  -mswdiv -mprioritize-restricted-insns=priority -msched-costly-dep=dependence_type
           -minsert-sched-nops=scheme -mcall-sysv  -mcall-netbsd -maix-struct-return  -msvr4-struct-return -mabi=abi-
           type -msecure-plt -mbss-plt -misel -mno-isel -misel=yes  -misel=no -mspe -mno-spe -mspe=yes  -mspe=no -mvr-
           save -mno-vrsave -mfloat-gprs=yes  -mfloat-gprs=no -mfloat-gprs=single -mfloat-gprs=double -mprototype
           -mno-prototype -msim  -mmvme  -mads  -myellowknife  -memb  -msdata -msdata=opt  -mvxworks  -mwindiss  -G
           num  -pthread


           S/390 and zSeries Options -mtune=cpu-type  -march=cpu-type -mhard-float  -msoft-float -mlong-double-64
           -mlong-double-128 -mbackchain  -mno-backchain -mpacked-stack  -mno-packed-stack -msmall-exec
           -mno-small-exec  -mmvcle -mno-mvcle -m64  -m31  -mdebug  -mno-debug  -mesa  -mzarch -mtpf-trace
           -mno-tpf-trace  -mfused-madd  -mno-fused-madd -mwarn-framesize  -mwarn-dynamicstack  -mstack-size
           -mstack-guard


           SH Options -m1  -m2  -m2e  -m3  -m3e -m4-nofpu  -m4-single-only  -m4-single  -m4 -m4a-nofpu -m4a-sin-
           gle-only -m4a-single -m4a -m4al -m5-64media  -m5-64media-nofpu -m5-32media  -m5-32media-nofpu -m5-compact
           -m5-compact-nofpu -mb  -ml  -mdalign  -mrelax -mbigtable  -mfmovd  -mhitachi -mrenesas -mno-renesas -mno-
           macsave -mieee  -misize  -mpadstruct  -mspace -mprefergot  -musermode -multcost=number -mdiv=strategy -mdi-
           vsi3_libfunc=name -madjust-unroll -mindexed-addressing -mgettrcost=number -mpt-fixed
            -minvalid-symbols


           SPARC Options -mcpu=cpu-type -mtune=cpu-type -mcmodel=code-model -m32  -m64  -mapp-regs  -mno-app-regs
           -mfaster-structs  -mno-faster-structs -mfpu  -mno-fpu  -mhard-float  -msoft-float -mhard-quad-float
           -msoft-quad-float -mimpure-text  -mno-impure-text  -mlittle-endian -mstack-bias  -mno-stack-bias
           -munaligned-doubles  -mno-unaligned-doubles -mv8plus  -mno-v8plus  -mvis  -mno-vis -threads -pthreads
           -pthread


           System V Options -Qy  -Qn  -YP,paths  -Ym,dir


           TMS320C3x/C4x Options -mcpu=cpu  -mbig  -msmall  -mregparm  -mmemparm -mfast-fix  -mmpyi  -mbk  -mti
           -mdp-isr-reload -mrpts=count  -mrptb  -mdb  -mloop-unsigned -mparallel-insns  -mparallel-mpy  -mpre-
           serve-float


           V850 Options -mlong-calls  -mno-long-calls  -mep  -mno-ep -mprolog-function  -mno-prolog-function  -mspace
           -mtda=n  -msda=n  -mzda=n -mapp-regs  -mno-app-regs -mdisable-callt  -mno-disable-callt -mv850e1 -mv850e
           -mv850  -mbig-switch


           VAX Options -mg  -mgnu  -munix


           x86-64 Options See i386 and x86-64 Options.


           Xstormy16 Options -msim


           Xtensa Options -mconst16 -mno-const16 -mfused-madd  -mno-fused-madd -mtext-section-literals  -mno-text-sec-
           tion-literals -mtarget-align  -mno-target-align -mlongcalls  -mno-longcalls


           zSeries Options See S/390 and zSeries Options.


       Code Generation Options
           -fcall-saved-reg  -fcall-used-reg -ffixed-reg  -fexceptions -fnon-call-exceptions  -funwind-tables -fasyn-
           chronous-unwind-tables -finhibit-size-directive  -finstrument-functions -fno-common  -fno-ident
           -fpcc-struct-return  -fpic  -fPIC -fpie -fPIE -fno-jump-tables -freg-struct-return  -fshared-data
           -fshort-enums -fshort-double  -fshort-wchar -fverbose-asm  -fpack-struct[=n]  -fstack-check
           -fstack-limit-register=reg  -fstack-limit-symbol=sym -fno-stack-limit  -fargument-alias  -fargument-noalias
           -fargument-noalias-global  -fleading-underscore -ftls-model=model -ftrapv  -fwrapv  -fbounds-check -fvisi-
           bility  -fopenmp


       Options Controlling the Kind of Output


       Compilation can involve up to four stages: preprocessing, compilation proper, assembly and linking, always in
       that order.  GCC is capable of preprocessing and compiling several files either into several assembler input
       files, or into one assembler input file; then each assembler input file produces an object file, and linking
       combines all the object files (those newly compiled, and those specified as input) into an executable file.


       For any given input file, the file name suffix determines what kind of compilation is done:


       file.c
           C source code which must be preprocessed.


       file.i
           C source code which should not be preprocessed.


       file.ii
           C++ source code which should not be preprocessed.


       file.m
           Objective-C source code.  Note that you must link with the libobjc library to make an Objective-C program
           work.


       file.mi
           Objective-C source code which should not be preprocessed.


       file.mm
       file.M
           Objective-C++ source code.  Note that you must link with the libobjc library to make an Objective-C++ pro-
           gram work.  Note that .M refers to a literal capital M.


       file.mii
           Objective-C++ source code which should not be preprocessed.


       file.h
           C, C++, Objective-C or Objective-C++ header file to be turned into a precompiled header.


       file.cc
       file.cp
       file.cxx
       file.cpp
       file.CPP
       file.c++
       file.C
           C++ source code which must be preprocessed.  Note that in .cxx, the last two letters must both be literally
           x.  Likewise, .C refers to a literal capital C.


       file.mm
       file.M
           Objective-C++ source code which must be preprocessed.


       file.mii
           Objective-C++ source code which should not be preprocessed.


       file.hh
       file.H
           C++ header file to be turned into a precompiled header.


       file.f
       file.for
       file.FOR
           Fixed form Fortran source code which should not be preprocessed.


       file.F
       file.fpp
       file.FPP
           Fixed form Fortran source code which must be preprocessed (with the traditional preprocessor).


       file.f90
       file.f95
           Free form Fortran source code which should not be preprocessed.


       file.F90
       file.F95
           Free form Fortran source code which must be preprocessed (with the traditional preprocessor).


       file.ads
           Ada source code file which contains a library unit declaration (a declaration of a package, subprogram, or
           generic, or a generic instantiation), or a library unit renaming declaration (a package, generic, or sub-
           program renaming declaration).  Such files are also called specs.


       file.adb
           Ada source code file containing a library unit body (a subprogram or package body).  Such files are also
           called bodies.


       file.s
           Assembler code.


       file.S
           Assembler code which must be preprocessed.


       other
           An object file to be fed straight into linking.  Any file name with no recognized suffix is treated this
           way.


       You can specify the input language explicitly with the -x option:


       -x language
           Specify explicitly the language for the following input files (rather than letting the compiler choose a
           default based on the file name suffix).  This option applies to all following input files until the next -x
           option.  Possible values for language are:


                   c  c-header  c-cpp-output
                   c++  c++-header  c++-cpp-output
                   objective-c  objective-c-header  objective-c-cpp-output
                   objective-c++ objective-c++-header objective-c++-cpp-output
                   assembler  assembler-with-cpp
                   ada
                   f95  f95-cpp-input
                   java
                   treelang


       -x none
           Turn off any specification of a language, so that subsequent files are handled according to their file name
           suffixes (as they are if -x has not been used at all).


       -pass-exit-codes
           Normally the gcc program will exit with the code of 1 if any phase of the compiler returns a non-success
           return code.  If you specify -pass-exit-codes, the gcc program will instead return with numerically highest
           error produced by any phase that returned an error indication.


       If you only want some of the stages of compilation, you can use -x (or filename suffixes) to tell gcc where to
       start, and one of the options -c, -S, or -E to say where gcc is to stop.  Note that some combinations (for
       example, -x cpp-output -E) instruct gcc to do nothing at all.


       -c  Compile or assemble the source files, but do not link.  The linking stage simply is not done.  The ultimate
           output is in the form of an object file for each source file.


           By default, the object file name for a source file is made by replacing the suffix .c, .i, .s, etc., with
           .o.


           Unrecognized input files, not requiring compilation or assembly, are ignored.


       -S  Stop after the stage of compilation proper; do not assemble.  The output is in the form of an assembler
           code file for each non-assembler input file specified.


           By default, the assembler file name for a source file is made by replacing the suffix .c, .i, etc., with
           .s.


           Input files that don’t require compilation are ignored.


       -E  Stop after the preprocessing stage; do not run the compiler proper.  The output is in the form of prepro-
           cessed source code, which is sent to the standard output.


           Input files which don’t require preprocessing are ignored.


       -o file
           Place output in file file.  This applies regardless to whatever sort of output is being produced, whether
           it be an executable file, an object file, an assembler file or preprocessed C code.


           If -o is not specified, the default is to put an executable file in a.out, the object file for source.suf-
           fix in source.o, its assembler file in source.s, a precompiled header file in source.suffix.gch, and all
           preprocessed C source on standard output.


       -v  Print (on standard error output) the commands executed to run the stages of compilation.  Also print the
           version number of the compiler driver program and of the preprocessor and the compiler proper.


       -###
           Like -v except the commands are not executed and all command arguments are quoted.  This is useful for
           shell scripts to capture the driver-generated command lines.


       -pipe
           Use pipes rather than temporary files for communication between the various stages of compilation.  This
           fails to work on some systems where the assembler is unable to read from a pipe; but the GNU assembler has
           no trouble.


       -combine
           If you are compiling multiple source files, this option tells the driver to pass all the source files to
           the compiler at once (for those languages for which the compiler can handle this).  This will allow inter-
           module analysis (IMA) to be performed by the compiler.  Currently the only language for which this is sup-
           ported is C.  If you pass source files for multiple languages to the driver, using this option, the driver
           will invoke the compiler(s) that support IMA once each, passing each compiler all the source files appro-
           priate for it.  For those languages that do not support IMA this option will be ignored, and the compiler
           will be invoked once for each source file in that language.  If you use this option in conjunction with
           -save-temps, the compiler will generate multiple pre-processed files (one for each source file), but only
           one (combined) .o or .s file.


       --help
           Print (on the standard output) a description of the command line options understood by gcc.  If the -v
           option is also specified then --help will also be passed on to the various processes invoked by gcc, so
           that they can display the command line options they accept.  If the -Wextra option is also specified then
           command line options which have no documentation associated with them will also be displayed.


       --target-help
           Print (on the standard output) a description of target specific command line options for each tool.


       --version
           Display the version number and copyrights of the invoked GCC.


       Compiling C++ Programs


       C++ source files conventionally use one of the suffixes .C, .cc, .cpp, .CPP, .c++, .cp, or .cxx; C++ header
       files often use .hh or .H; and preprocessed C++ files use the suffix .ii.  GCC recognizes files with these
       names and compiles them as C++ programs even if you call the compiler the same way as for compiling C programs
       (usually with the name gcc).


       However, C++ programs often require class libraries as well as a compiler that understands the C++ lan-
       guage---and under some circumstances, you might want to compile programs or header files from standard input,
       or otherwise without a suffix that flags them as C++ programs.  You might also like to precompile a C header
       file with a .h extension to be used in C++ compilations.  g++ is a program that calls GCC with the default lan-
       guage set to C++, and automatically specifies linking against the C++ library.  On many systems, g++ is also
       installed with the name c++.


       When you compile C++ programs, you may specify many of the same command-line options that you use for compiling
       programs in any language; or command-line options meaningful for C and related languages; or options that are
       meaningful only for C++ programs.


       Options Controlling C Dialect


       The following options control the dialect of C (or languages derived from C, such as C++, Objective-C and
       Objective-C++) that the compiler accepts:


       -ansi
           In C mode, support all ISO C90 programs.  In C++ mode, remove GNU extensions that conflict with ISO C++.


           This turns off certain features of GCC that are incompatible with ISO C90 (when compiling C code), or of
           standard C++ (when compiling C++ code), such as the "asm" and "typeof" keywords, and predefined macros such
           as "unix" and "vax" that identify the type of system you are using.  It also enables the undesirable and
           rarely used ISO trigraph feature.  For the C compiler, it disables recognition of C++ style // comments as
           well as the "inline" keyword.


           The alternate keywords "__asm__", "__extension__", "__inline__" and "__typeof__" continue to work despite
           -ansi.  You would not want to use them in an ISO C program, of course, but it is useful to put them in
           header files that might be included in compilations done with -ansi.  Alternate predefined macros such as
           "__unix__" and "__vax__" are also available, with or without -ansi.


           The -ansi option does not cause non-ISO programs to be rejected gratuitously.  For that, -pedantic is
           required in addition to -ansi.


           The macro "__STRICT_ANSI__" is predefined when the -ansi option is used.  Some header files may notice this
           macro and refrain from declaring certain functions or defining certain macros that the ISO standard doesn’t
           call for; this is to avoid interfering with any programs that might use these names for other things.


           Functions which would normally be built in but do not have semantics defined by ISO C (such as "alloca" and
           "ffs") are not built-in functions with -ansi is used.


       -std=
           Determine the language standard.  This option is currently only supported when compiling C or C++.  A value
           for this option must be provided; possible values are


           c89
           iso9899:1990
               ISO C90 (same as -ansi).


           iso9899:199409
               ISO C90 as modified in amendment 1.


           c99
           c9x
           iso9899:1999
           iso9899:199x
               ISO C99.  Note that this standard is not yet fully supported; see <http://gcc.gnu.org/gcc-4.1/c99sta-
               tus.html> for more information.  The names c9x and iso9899:199x are deprecated.


           gnu89
               Default, ISO C90 plus GNU extensions (including some C99 features).


           gnu99
           gnu9x
               ISO C99 plus GNU extensions.  When ISO C99 is fully implemented in GCC, this will become the default.
               The name gnu9x is deprecated.


           c++98
               The 1998 ISO C++ standard plus amendments.


           gnu++98
               The same as -std=c++98 plus GNU extensions.  This is the default for C++ code.


           Even when this option is not specified, you can still use some of the features of newer standards in so far
           as they do not conflict with previous C standards.  For example, you may use "__restrict__" even when
           -std=c99 is not specified.


           The -std options specifying some version of ISO C have the same effects as -ansi, except that features that
           were not in ISO C90 but are in the specified version (for example, // comments and the "inline" keyword in
           ISO C99) are not disabled.


       -fgnu89-inline
           The option -fgnu89-inline tells GCC to use the traditional GNU semantics for "inline" functions when in C99
           mode.
             Using this option is roughly equivalent to adding the "gnu_inline" function attribute to all inline func-
           tions.


           This option is accepted by GCC versions 4.1.3 and up.  In GCC versions prior to 4.3, C99 inline semantics
           are not supported, and thus this option is effectively assumed to be present regardless of whether or not
           it is specified; the only effect of specifying it explicitly is to disable warnings about using inline
           functions in C99 mode.  Likewise, the option -fno-gnu89-inline is not supported in versions of GCC before
           4.3.  It will be supported only in C99 or gnu99 mode, not in C89 or gnu89 mode.


           The preprocesor macros "__GNUC_GNU_INLINE__" and "__GNUC_STDC_INLINE__" may be used to check which seman-
           tics are in effect for "inline" functions.


       -aux-info filename
           Output to the given filename prototyped declarations for all functions declared and/or defined in a trans-
           lation unit, including those in header files.  This option is silently ignored in any language other than
           C.


           Besides declarations, the file indicates, in comments, the origin of each declaration (source file and
           line), whether the declaration was implicit, prototyped or unprototyped (I, N for new or O for old, respec-
           tively, in the first character after the line number and the colon), and whether it came from a declaration
           or a definition (C or F, respectively, in the following character).  In the case of function definitions, a
           K&R-style list of arguments followed by their declarations is also provided, inside comments, after the
           declaration.


       -fno-asm
           Do not recognize "asm", "inline" or "typeof" as a keyword, so that code can use these words as identifiers.
           You can use the keywords "__asm__", "__inline__" and "__typeof__" instead.  -ansi implies -fno-asm.


           In C++, this switch only affects the "typeof" keyword, since "asm" and "inline" are standard keywords.  You
           may want to use the -fno-gnu-keywords flag instead, which has the same effect.  In C99 mode (-std=c99 or
           -std=gnu99), this switch only affects the "asm" and "typeof" keywords, since "inline" is a standard keyword
           in ISO C99.


       -fno-builtin
       -fno-builtin-function
           Don’t recognize built-in functions that do not begin with __builtin_ as prefix.


           GCC normally generates special code to handle certain built-in functions more efficiently; for instance,
           calls to "alloca" may become single instructions that adjust the stack directly, and calls to "memcpy" may
           become inline copy loops.  The resulting code is often both smaller and faster, but since the function
           calls no longer appear as such, you cannot set a breakpoint on those calls, nor can you change the behavior
           of the functions by linking with a different library.  In addition, when a function is recognized as a
           built-in function, GCC may use information about that function to warn about problems with calls to that
           function, or to generate more efficient code, even if the resulting code still contains calls to that func-
           tion.  For example, warnings are given with -Wformat for bad calls to "printf", when "printf" is built in,
           and "strlen" is known not to modify global memory.


           With the -fno-builtin-function option only the built-in function function is disabled.  function must not
           begin with __builtin_.  If a function is named this is not built-in in this version of GCC, this option is
           ignored.  There is no corresponding -fbuiltin-function option; if you wish to enable built-in functions
           selectively when using -fno-builtin or -ffreestanding, you may define macros such as:


                   #define abs(n)          __builtin_abs ((n))
                   #define strcpy(d, s)    __builtin_strcpy ((d), (s))


       -fhosted
           Assert that compilation takes place in a hosted environment.  This implies -fbuiltin.  A hosted environment
           is one in which the entire standard library is available, and in which "main" has a return type of "int".
           Examples are nearly everything except a kernel.  This is equivalent to -fno-freestanding.


       -ffreestanding
           Assert that compilation takes place in a freestanding environment.  This implies -fno-builtin.  A free-
           standing environment is one in which the standard library may not exist, and program startup may not neces-
           sarily be at "main".  The most obvious example is an OS kernel.  This is equivalent to -fno-hosted.


       -fms-extensions
           Accept some non-standard constructs used in Microsoft header files.


           Some cases of unnamed fields in structures and unions are only accepted with this option.


       -trigraphs
           Support ISO C trigraphs.  The -ansi option (and -std options for strict ISO C conformance) implies -tri-
           graphs.


       -no-integrated-cpp
           Performs a compilation in two passes: preprocessing and compiling.  This option allows a user supplied
           "cc1", "cc1plus", or "cc1obj" via the -B option.  The user supplied compilation step can then add in an
           additional preprocessing step after normal preprocessing but before compiling.  The default is to use the
           integrated cpp (internal cpp)


           The semantics of this option will change if "cc1", "cc1plus", and "cc1obj" are merged.


       -traditional
       -traditional-cpp
           Formerly, these options caused GCC to attempt to emulate a pre-standard C compiler.  They are now only sup-
           ported with the -E switch.  The preprocessor continues to support a pre-standard mode.  See the GNU CPP
           manual for details.


       -fcond-mismatch
           Allow conditional expressions with mismatched types in the second and third arguments.  The value of such
           an expression is void.  This option is not supported for C++.


       -funsigned-char
           Let the type "char" be unsigned, like "unsigned char".


           Each kind of machine has a default for what "char" should be.  It is either like "unsigned char" by default
           or like "signed char" by default.


           Ideally, a portable program should always use "signed char" or "unsigned char" when it depends on the
           signedness of an object.  But many programs have been written to use plain "char" and expect it to be
           signed, or expect it to be unsigned, depending on the machines they were written for.  This option, and its
           inverse, let you make such a program work with the opposite default.


           The type "char" is always a distinct type from each of "signed char" or "unsigned char", even though its
           behavior is always just like one of those two.


       -fsigned-char
           Let the type "char" be signed, like "signed char".


           Note that this is equivalent to -fno-unsigned-char, which is the negative form of -funsigned-char.  Like-
           wise, the option -fno-signed-char is equivalent to -funsigned-char.


       -fsigned-bitfields
       -funsigned-bitfields
       -fno-signed-bitfields
       -fno-unsigned-bitfields
           These options control whether a bit-field is signed or unsigned, when the declaration does not use either
           "signed" or "unsigned".  By default, such a bit-field is signed, because this is consistent: the basic
           integer types such as "int" are signed types.


       Options Controlling C++ Dialect


       This section describes the command-line options that are only meaningful for C++ programs; but you can also use
       most of the GNU compiler options regardless of what language your program is in.  For example, you might com-
       pile a file "firstClass.C" like this:


               g++ -g -frepo -O -c firstClass.C


       In this example, only -frepo is an option meant only for C++ programs; you can use the other options with any
       language supported by GCC.


       Here is a list of options that are only for compiling C++ programs:


       -fabi-version=n
           Use version n of the C++ ABI.  Version 2 is the version of the C++ ABI that first appeared in G++ 3.4.
           Version 1 is the version of the C++ ABI that first appeared in G++ 3.2.  Version 0 will always be the ver-
           sion that conforms most closely to the C++ ABI specification.  Therefore, the ABI obtained using version 0
           will change as ABI bugs are fixed.


           The default is version 2.


       -fno-access-control
           Turn off all access checking.  This switch is mainly useful for working around bugs in the access control
           code.


       -fcheck-new
           Check that the pointer returned by "operator new" is non-null before attempting to modify the storage allo-
           cated.  This check is normally unnecessary because the C++ standard specifies that "operator new" will only
           return 0 if it is declared throw(), in which case the compiler will always check the return value even
           without this option.  In all other cases, when "operator new" has a non-empty exception specification, mem-
           ory exhaustion is signalled by throwing "std::bad_alloc".  See also new (nothrow).


       -fconserve-space
           Put uninitialized or runtime-initialized global variables into the common segment, as C does.  This saves
           space in the executable at the cost of not diagnosing duplicate definitions.  If you compile with this flag
           and your program mysteriously crashes after "main()" has completed, you may have an object that is being
           destroyed twice because two definitions were merged.


           This option is no longer useful on most targets, now that support has been added for putting variables into
           BSS without making them common.


       -ffriend-injection
           Inject friend functions into the enclosing namespace, so that they are visible outside the scope of the
           class in which they are declared.  Friend functions were documented to work this way in the old Annotated
           C++ Reference Manual, and versions of G++ before 4.1 always worked that way.  However, in ISO C++ a friend
           function which is not declared in an enclosing scope can only be found using argument dependent lookup.
           This option causes friends to be injected as they were in earlier releases.


           This option is for compatibility, and may be removed in a future release of G++.


       -fno-const-strings
           Give string constants type "char *" instead of type "const char *".  By default, G++ uses type "const char
           *" as required by the standard.  Even if you use -fno-const-strings, you cannot actually modify the value
           of a string constant.


           This option might be removed in a future release of G++.  For maximum portability, you should structure
           your code so that it works with string constants that have type "const char *".


       -fno-elide-constructors
           The C++ standard allows an implementation to omit creating a temporary which is only used to initialize
           another object of the same type.  Specifying this option disables that optimization, and forces G++ to call
           the copy constructor in all cases.


       -fno-enforce-eh-specs
           Don’t generate code to check for violation of exception specifications at runtime.  This option violates
           the C++ standard, but may be useful for reducing code size in production builds, much like defining NDEBUG.
           This does not give user code permission to throw exceptions in violation of the exception specifications;
           the compiler will still optimize based on the specifications, so throwing an unexpected exception will
           result in undefined behavior.


       -ffor-scope
       -fno-for-scope
           If -ffor-scope is specified, the scope of variables declared in a for-init-statement is limited to the for
           loop itself, as specified by the C++ standard.  If -fno-for-scope is specified, the scope of variables
           declared in a for-init-statement extends to the end of the enclosing scope, as was the case in old versions
           of G++, and other (traditional) implementations of C++.


           The default if neither flag is given to follow the standard, but to allow and give a warning for old-style
           code that would otherwise be invalid, or have different behavior.


       -fno-gnu-keywords
           Do not recognize "typeof" as a keyword, so that code can use this word as an identifier.  You can use the
           keyword "__typeof__" instead.  -ansi implies -fno-gnu-keywords.


       -fno-implicit-templates
           Never emit code for non-inline templates which are instantiated implicitly (i.e. by use); only emit code
           for explicit instantiations.


       -fno-implicit-inline-templates
           Don’t emit code for implicit instantiations of inline templates, either.  The default is to handle inlines
           differently so that compiles with and without optimization will need the same set of explicit instantia-
           tions.


       -fno-implement-inlines
           To save space, do not emit out-of-line copies of inline functions controlled by #pragma implementation.
           This will cause linker errors if these functions are not inlined everywhere they are called.


       -fms-extensions
           Disable pedantic warnings about constructs used in MFC, such as implicit int and getting a pointer to mem-
           ber function via non-standard syntax.


       -fno-nonansi-builtins
           Disable built-in declarations of functions that are not mandated by ANSI/ISO C.  These include "ffs",
           "alloca", "_exit", "index", "bzero", "conjf", and other related functions.


       -fno-operator-names
           Do not treat the operator name keywords "and", "bitand", "bitor", "compl", "not", "or" and "xor" as syn-
           onyms as keywords.


       -fno-optional-diags
           Disable diagnostics that the standard says a compiler does not need to issue.  Currently, the only such
           diagnostic issued by G++ is the one for a name having multiple meanings within a class.


       -fpermissive
           Downgrade some diagnostics about nonconformant code from errors to warnings.  Thus, using -fpermissive will
           allow some nonconforming code to compile.


       -frepo
           Enable automatic template instantiation at link time.  This option also implies -fno-implicit-templates.


       -fno-rtti
           Disable generation of information about every class with virtual functions for use by the C++ runtime type
           identification features (dynamic_cast and typeid).  If you don’t use those parts of the language, you can
           save some space by using this flag.  Note that exception handling uses the same information, but it will
           generate it as needed.


       -fstats
           Emit statistics about front-end processing at the end of the compilation.  This information is generally
           only useful to the G++ development team.


       -ftemplate-depth-n
           Set the maximum instantiation depth for template classes to n.  A limit on the template instantiation depth
           is needed to detect endless recursions during template class instantiation.  ANSI/ISO C++ conforming pro-
           grams must not rely on a maximum depth greater than 17.


       -fno-threadsafe-statics
           Do not emit the extra code to use the routines specified in the C++ ABI for thread-safe initialization of
           local statics.  You can use this option to reduce code size slightly in code that doesn’t need to be
           thread-safe.


       -fuse-cxa-atexit
           Register destructors for objects with static storage duration with the "__cxa_atexit" function rather than
           the "atexit" function.  This option is required for fully standards-compliant handling of static destruc-
           tors, but will only work if your C library supports "__cxa_atexit".


       -fvisibility-inlines-hidden
           Causes all inlined methods to be marked with "__attribute__ ((visibility ("hidden")))" so that they do not
           appear in the export table of a DSO and do not require a PLT indirection when used within the DSO.
           Enabling this option can have a dramatic effect on load and link times of a DSO as it massively reduces the
           size of the dynamic export table when the library makes heavy use of templates.


           You may mark a method as having a visibility explicitly to negate the effect of the switch for that method.
           For example, if you do want to compare pointers to a particular inline method, or the method has local
           static data, you might mark it as having default visibility.


       -fno-weak
           Do not use weak symbol support, even if it is provided by the linker.  By default, G++ will use weak sym-
           bols if they are available.  This option exists only for testing, and should not be used by end-users; it
           will result in inferior code and has no benefits.  This option may be removed in a future release of G++.


       -nostdinc++
           Do not search for header files in the standard directories specific to C++, but do still search the other
           standard directories.  (This option is used when building the C++ library.)


       In addition, these optimization, warning, and code generation options have meanings only for C++ programs:


       -fno-default-inline
           Do not assume inline for functions defined inside a class scope.
             Note that these functions will have linkage like inline functions; they just won’t be inlined by default.


       -Wabi (C++ only)
           Warn when G++ generates code that is probably not compatible with the vendor-neutral C++ ABI.  Although an
           effort has been made to warn about all such cases, there are probably some cases that are not warned about,
           even though G++ is generating incompatible code.  There may also be cases where warnings are emitted even
           though the code that is generated will be compatible.


           You should rewrite your code to avoid these warnings if you are concerned about the fact that code gener-
           ated by G++ may not be binary compatible with code generated by other compilers.


           The known incompatibilities at this point include:


           *   Incorrect handling of tail-padding for bit-fields.  G++ may attempt to pack data into the same byte as
               a base class.  For example:


                       struct A { virtual void f(); int f1 : 1; };
                       struct B : public A { int f2 : 1; };


               In this case, G++ will place "B::f2" into the same byte as"A::f1"; other compilers will not.  You can
               avoid this problem by explicitly padding "A" so that its size is a multiple of the byte size on your
               platform; that will cause G++ and other compilers to layout "B" identically.


           *   Incorrect handling of tail-padding for virtual bases.  G++ does not use tail padding when laying out
               virtual bases.  For example:


                       struct A { virtual void f(); char c1; };
                       struct B { B(); char c2; };
                       struct C : public A, public virtual B {};


               In this case, G++ will not place "B" into the tail-padding for "A"; other compilers will.  You can
               avoid this problem by explicitly padding "A" so that its size is a multiple of its alignment (ignoring
               virtual base classes); that will cause G++ and other compilers to layout "C" identically.


           *   Incorrect handling of bit-fields with declared widths greater than that of their underlying types, when
               the bit-fields appear in a union.  For example:


                       union U { int i : 4096; };


               Assuming that an "int" does not have 4096 bits, G++ will make the union too small by the number of bits
               in an "int".


           *   Empty classes can be placed at incorrect offsets.  For example:


                       struct A {};


                       struct B {
                         A a;
                         virtual void f ();
                       };


                       struct C : public B, public A {};


               G++ will place the "A" base class of "C" at a nonzero offset; it should be placed at offset zero.  G++
               mistakenly believes that the "A" data member of "B" is already at offset zero.


           *   Names of template functions whose types involve "typename" or template template parameters can be man-
               gled incorrectly.


                       template <typename Q>
                       void f(typename Q::X) {}


                       template <template <typename> class Q>
                       void f(typename Q<int>::X) {}


               Instantiations of these templates may be mangled incorrectly.


       -Wctor-dtor-privacy (C++ only)
           Warn when a class seems unusable because all the constructors or destructors in that class are private, and
           it has neither friends nor public static member functions.


       -Wnon-virtual-dtor (C++ only)
           Warn when a class appears to be polymorphic, thereby requiring a virtual destructor, yet it declares a non-
           virtual one.  This warning is enabled by -Wall.


       -Wreorder (C++ only)
           Warn when the order of member initializers given in the code does not match the order in which they must be
           executed.  For instance:


                   struct A {
                     int i;
                     int j;
                     A(): j (0), i (1) { }
                   };


           The compiler will rearrange the member initializers for i and j to match the declaration order of the mem-
           bers, emitting a warning to that effect.  This warning is enabled by -Wall.


       The following -W... options are not affected by -Wall.


       -Weffc++ (C++ only)
           Warn about violations of the following style guidelines from Scott Meyers’ Effective C++ book:


           *   Item 11:  Define a copy constructor and an assignment operator for classes with dynamically allocated
               memory.


           *   Item 12:  Prefer initialization to assignment in constructors.


           *   Item 14:  Make destructors virtual in base classes.


           *   Item 15:  Have "operator=" return a reference to *this.


           *   Item 23:  Don’t try to return a reference when you must return an object.


           Also warn about violations of the following style guidelines from Scott Meyers’ More Effective C++ book:


           *   Item 6:  Distinguish between prefix and postfix forms of increment and decrement operators.


           *   Item 7:  Never overload "&&", "││", or ",".


           When selecting this option, be aware that the standard library headers do not obey all of these guidelines;
           use grep -v to filter out those warnings.


       -Wno-deprecated (C++ only)
           Do not warn about usage of deprecated features.


       -Wstrict-null-sentinel (C++ only)
           Warn also about the use of an uncasted "NULL" as sentinel.  When compiling only with GCC this is a valid
           sentinel, as "NULL" is defined to "__null".  Although it is a null pointer constant not a null pointer, it
           is guaranteed to of the same size as a pointer.  But this use is not portable across different compilers.


       -Wno-non-template-friend (C++ only)
           Disable warnings when non-templatized friend functions are declared within a template.  Since the advent of
           explicit template specification support in G++, if the name of the friend is an unqualified-id (i.e.,
           friend foo(int)), the C++ language specification demands that the friend declare or define an ordinary,
           nontemplate function.  (Section 14.5.3).  Before G++ implemented explicit specification, unqualified-ids
           could be interpreted as a particular specialization of a templatized function.  Because this non-conforming
           behavior is no longer the default behavior for G++, -Wnon-template-friend allows the compiler to check
           existing code for potential trouble spots and is on by default.  This new compiler behavior can be turned
           off with -Wno-non-template-friend which keeps the conformant compiler code but disables the helpful warn-
           ing.


       -Wold-style-cast (C++ only)
           Warn if an old-style (C-style) cast to a non-void type is used within a C++ program.  The new-style casts
           (dynamic_cast, static_cast, reinterpret_cast, and const_cast) are less vulnerable to unintended effects and
           much easier to search for.


       -Woverloaded-virtual (C++ only)
           Warn when a function declaration hides virtual functions from a base class.  For example, in:


                   struct A {
                     virtual void f();
                   };


                   struct B: public A {
                     void f(int);
                   };


           the "A" class version of "f" is hidden in "B", and code like:


                   B* b;
                   b->f();


           will fail to compile.


       -Wno-pmf-conversions (C++ only)
           Disable the diagnostic for converting a bound pointer to member function to a plain pointer.


       -Wsign-promo (C++ only)
           Warn when overload resolution chooses a promotion from unsigned or enumerated type to a signed type, over a
           conversion to an unsigned type of the same size.  Previous versions of G++ would try to preserve unsigned-
           ness, but the standard mandates the current behavior.


                   struct A {
                     operator int ();
                     A& operator = (int);
                   };


                   main ()
                   {
                     A a,b;
                     a = b;
                   }


           In this example, G++ will synthesize a default A& operator = (const A&);, while cfront will use the user-
           defined operator =.


       Options Controlling Objective-C and Objective-C++ Dialects


       (NOTE: This manual does not describe the Objective-C and Objective-C++ languages themselves.  See


       This section describes the command-line options that are only meaningful for Objective-C and Objective-C++ pro-
       grams, but you can also use most of the language-independent GNU compiler options.  For example, you might com-
       pile a file "some_class.m" like this:


               gcc -g -fgnu-runtime -O -c some_class.m


       In this example, -fgnu-runtime is an option meant only for Objective-C and Objective-C++ programs; you can use
       the other options with any language supported by GCC.


       Note that since Objective-C is an extension of the C language, Objective-C compilations may also use options
       specific to the C front-end (e.g., -Wtraditional).  Similarly, Objective-C++ compilations may use C++-specific
       options (e.g., -Wabi).


       Here is a list of options that are only for compiling Objective-C and Objective-C++ programs:


       -fconstant-string-class=class-name
           Use class-name as the name of the class to instantiate for each literal string specified with the syntax
           "@"..."".  The default class name is "NXConstantString" if the GNU runtime is being used, and "NSCon-
           stantString" if the NeXT runtime is being used (see below).  The -fconstant-cfstrings option, if also
           present, will override the -fconstant-string-class setting and cause "@"..."" literals to be laid out as
           constant CoreFoundation strings.


       -fgnu-runtime
           Generate object code compatible with the standard GNU Objective-C runtime.  This is the default for most
           types of systems.


       -fnext-runtime
           Generate output compatible with the NeXT runtime.  This is the default for NeXT-based systems, including
           Darwin and Mac OS X.  The macro "__NEXT_RUNTIME__" is predefined if (and only if) this option is used.


       -fno-nil-receivers
           Assume that all Objective-C message dispatches (e.g., "[receiver message:arg]") in this translation unit
           ensure that the receiver is not "nil".  This allows for more efficient entry points in the runtime to be
           used.  Currently, this option is only available in conjunction with the NeXT runtime on Mac OS X 10.3 and
           later.


       -fobjc-call-cxx-cdtors
           For each Objective-C class, check if any of its instance variables is a C++ object with a non-trivial
           default constructor.  If so, synthesize a special "- (id) .cxx_construct" instance method that will run
           non-trivial default constructors on any such instance variables, in order, and then return "self".  Simi-
           larly, check if any instance variable is a C++ object with a non-trivial destructor, and if so, synthesize
           a special "- (void) .cxx_destruct" method that will run all such default destructors, in reverse order.


           The "- (id) .cxx_construct" and/or "- (void) .cxx_destruct" methods thusly generated will only operate on
           instance variables declared in the current Objective-C class, and not those inherited from superclasses.
           It is the responsibility of the Objective-C runtime to invoke all such methods in an object’s inheritance
           hierarchy.  The "- (id) .cxx_construct" methods will be invoked by the runtime immediately after a new
           object instance is allocated; the "- (void) .cxx_destruct" methods will be invoked immediately before the
           runtime deallocates an object instance.


           As of this writing, only the NeXT runtime on Mac OS X 10.4 and later has support for invoking the "- (id)
           .cxx_construct" and "- (void) .cxx_destruct" methods.


       -fobjc-direct-dispatch
           Allow fast jumps to the message dispatcher.  On Darwin this is accomplished via the comm page.


       -fobjc-exceptions
           Enable syntactic support for structured exception handling in Objective-C, similar to what is offered by
           C++ and Java.  Currently, this option is only available in conjunction with the NeXT runtime on Mac OS X
           10.3 and later.


                     @try {
                       ...
                          @throw expr;
                       ...
                     }
                     @catch (AnObjCClass *exc) {
                       ...
                         @throw expr;
                       ...
                         @throw;
                       ...
                     }
                     @catch (AnotherClass *exc) {
                       ...
                     }
                     @catch (id allOthers) {
                       ...
                     }
                     @finally {
                       ...
                         @throw expr;
                       ...
                     }


           The @throw statement may appear anywhere in an Objective-C or Objective-C++ program; when used inside of a
           @catch block, the @throw may appear without an argument (as shown above), in which case the object caught
           by the @catch will be rethrown.


           Note that only (pointers to) Objective-C objects may be thrown and caught using this scheme.  When an
           object is thrown, it will be caught by the nearest @catch clause capable of handling objects of that type,
           analogously to how "catch" blocks work in C++ and Java.  A "@catch(id ...)" clause (as shown above) may
           also be provided to catch any and all Objective-C exceptions not caught by previous @catch clauses (if
           any).


           The @finally clause, if present, will be executed upon exit from the immediately preceding "@try ...
           @catch" section.  This will happen regardless of whether any exceptions are thrown, caught or rethrown
           inside the "@try ... @catch" section, analogously to the behavior of the "finally" clause in Java.


           There are several caveats to using the new exception mechanism:


           *   Although currently designed to be binary compatible with "NS_HANDLER"-style idioms provided by the
               "NSException" class, the new exceptions can only be used on Mac OS X 10.3 (Panther) and later systems,
               due to additional functionality needed in the (NeXT) Objective-C runtime.


           *   As mentioned above, the new exceptions do not support handling types other than Objective-C objects.
               Furthermore, when used from Objective-C++, the Objective-C exception model does not interoperate with
               C++ exceptions at this time.  This means you cannot @throw an exception from Objective-C and "catch" it
               in C++, or vice versa (i.e., "throw ... @catch").


           The -fobjc-exceptions switch also enables the use of synchronization blocks for thread-safe execution:


                     @synchronized (ObjCClass *guard) {
                       ...
                     }


           Upon entering the @synchronized block, a thread of execution shall first check whether a lock has been
           placed on the corresponding "guard" object by another thread.  If it has, the current thread shall wait
           until the other thread relinquishes its lock.  Once "guard" becomes available, the current thread will
           place its own lock on it, execute the code contained in the @synchronized block, and finally relinquish the
           lock (thereby making "guard" available to other threads).


           Unlike Java, Objective-C does not allow for entire methods to be marked @synchronized.  Note that throwing
           exceptions out of @synchronized blocks is allowed, and will cause the guarding object to be unlocked prop-
           erly.


       -fobjc-gc
           Enable garbage collection (GC) in Objective-C and Objective-C++ programs.


       -freplace-objc-classes
           Emit a special marker instructing ld(1) not to statically link in the resulting object file, and allow
           dyld(1) to load it in at run time instead.  This is used in conjunction with the Fix-and-Continue debugging
           mode, where the object file in question may be recompiled and dynamically reloaded in the course of program
           execution, without the need to restart the program itself.  Currently, Fix-and-Continue functionality is
           only available in conjunction with the NeXT runtime on Mac OS X 10.3 and later.


       -fzero-link
           When compiling for the NeXT runtime, the compiler ordinarily replaces calls to "objc_getClass("...")" (when
           the name of the class is known at compile time) with static class references that get initialized at load
           time, which improves run-time performance.  Specifying the -fzero-link flag suppresses this behavior and
           causes calls to "objc_getClass("...")"  to be retained.  This is useful in Zero-Link debugging mode, since
           it allows for individual class implementations to be modified during program execution.


       -gen-decls
           Dump interface declarations for all classes seen in the source file to a file named sourcename.decl.


       -Wassign-intercept
           Warn whenever an Objective-C assignment is being intercepted by the garbage collector.


       -Wno-protocol
           If a class is declared to implement a protocol, a warning is issued for every method in the protocol that
           is not implemented by the class.  The default behavior is to issue a warning for every method not explic-
           itly implemented in the class, even if a method implementation is inherited from the superclass.  If you
           use the -Wno-protocol option, then methods inherited from the superclass are considered to be implemented,
           and no warning is issued for them.


       -Wselector
           Warn if multiple methods of different types for the same selector are found during compilation.  The check
           is performed on the list of methods in the final stage of compilation.  Additionally, a check is performed
           for each selector appearing in a "@selector(...)"  expression, and a corresponding method for that selector
           has been found during compilation.  Because these checks scan the method table only at the end of compila-
           tion, these warnings are not produced if the final stage of compilation is not reached, for example because
           an error is found during compilation, or because the -fsyntax-only option is being used.


       -Wstrict-selector-match
           Warn if multiple methods with differing argument and/or return types are found for a given selector when
           attempting to send a message using this selector to a receiver of type "id" or "Class".  When this flag is
           off (which is the default behavior), the compiler will omit such warnings if any differences found are con-
           fined to types which share the same size and alignment.


       -Wundeclared-selector
           Warn if a "@selector(...)" expression referring to an undeclared selector is found.  A selector is consid-
           ered undeclared if no method with that name has been declared before the "@selector(...)" expression,
           either explicitly in an @interface or @protocol declaration, or implicitly in an @implementation section.
           This option always performs its checks as soon as a "@selector(...)" expression is found, while -Wselector
           only performs its checks in the final stage of compilation.  This also enforces the coding style convention
           that methods and selectors must be declared before being used.


       -print-objc-runtime-info
           Generate C header describing the largest structure that is passed by value, if any.


       Options to Control Diagnostic Messages Formatting


       Traditionally, diagnostic messages have been formatted irrespective of the output device’s aspect (e.g. its
       width, ...).  The options described below can be used to control the diagnostic messages formatting algorithm,
       e.g. how many characters per line, how often source location information should be reported.  Right now, only
       the C++ front end can honor these options.  However it is expected, in the near future, that the remaining
       front ends would be able to digest them correctly.


       -fmessage-length=n
           Try to format error messages so that they fit on lines of about n characters.  The default is 72 characters
           for g++ and 0 for the rest of the front ends supported by GCC.  If n is zero, then no line-wrapping will be
           done; each error message will appear on a single line.


       -fdiagnostics-show-location=once
           Only meaningful in line-wrapping mode.  Instructs the diagnostic messages reporter to emit once source
           location information; that is, in case the message is too long to fit on a single physical line and has to
           be wrapped, the source location won’t be emitted (as prefix) again, over and over, in subsequent continua-
           tion lines.  This is the default behavior.


       -fdiagnostics-show-location=every-line
           Only meaningful in line-wrapping mode.  Instructs the diagnostic messages reporter to emit the same source
           location information (as prefix) for physical lines that result from the process of breaking a message
           which is too long to fit on a single line.


       -fdiagnostics-show-options
           This option instructs the diagnostic machinery to add text to each diagnostic emitted, which indicates
           which command line option directly controls that diagnostic, when such an option is known to the diagnostic
           machinery.


       Options to Request or Suppress Warnings


       Warnings are diagnostic messages that report constructions which are not inherently erroneous but which are
       risky or suggest there may have been an error.


       You can request many specific warnings with options beginning -W, for example -Wimplicit to request warnings on
       implicit declarations.  Each of these specific warning options also has a negative form beginning -Wno- to turn
       off warnings; for example, -Wno-implicit.  This manual lists only one of the two forms, whichever is not the
       default.


       The following options control the amount and kinds of warnings produced by GCC; for further, language-specific
       options also refer to C++ Dialect Options and Objective-C and Objective-C++ Dialect Options.


       -fsyntax-only
           Check the code for syntax errors, but don’t do anything beyond that.


       -pedantic
           Issue all the warnings demanded by strict ISO C and ISO C++; reject all programs that use forbidden exten-
           sions, and some other programs that do not follow ISO C and ISO C++.  For ISO C, follows the version of the
           ISO C standard specified by any -std option used.


           Valid ISO C and ISO C++ programs should compile properly with or without this option (though a rare few
           will require -ansi or a -std option specifying the required version of ISO C).  However, without this
           option, certain GNU extensions and traditional C and C++ features are supported as well.  With this option,
           they are rejected.


           -pedantic does not cause warning messages for use of the alternate keywords whose names begin and end with
           __.  Pedantic warnings are also disabled in the expression that follows "__extension__".  However, only
           system header files should use these escape routes; application programs should avoid them.


           Some users try to use -pedantic to check programs for strict ISO C conformance.  They soon find that it
           does not do quite what they want: it finds some non-ISO practices, but not all---only those for which ISO C
           requires a diagnostic, and some others for which diagnostics have been added.


           A feature to report any failure to conform to ISO C might be useful in some instances, but would require
           considerable additional work and would be quite different from -pedantic.  We don’t have plans to support
           such a feature in the near future.


           Where the standard specified with -std represents a GNU extended dialect of C, such as gnu89 or gnu99,
           there is a corresponding base standard, the version of ISO C on which the GNU extended dialect is based.
           Warnings from -pedantic are given where they are required by the base standard.  (It would not make sense
           for such warnings to be given only for features not in the specified GNU C dialect, since by definition the
           GNU dialects of C include all features the compiler supports with the given option, and there would be
           nothing to warn about.)


       -pedantic-errors
           Like -pedantic, except that errors are produced rather than warnings.


       -w  Inhibit all warning messages.


       -Wno-import
           Inhibit warning messages about the use of #import.


       -Wchar-subscripts
           Warn if an array subscript has type "char".  This is a common cause of error, as programmers often forget
           that this type is signed on some machines.  This warning is enabled by -Wall.


       -Wcomment
           Warn whenever a comment-start sequence /* appears in a /* comment, or whenever a Backslash-Newline appears
           in a // comment.  This warning is enabled by -Wall.


       -Wfatal-errors
           This option causes the compiler to abort compilation on the first error occurred rather than trying to keep
           going and printing further error messages.


       -Wformat
           Check calls to "printf" and "scanf", etc., to make sure that the arguments supplied have types appropriate
           to the format string specified, and that the conversions specified in the format string make sense.  This
           includes standard functions, and others specified by format attributes, in the "printf", "scanf", "strf-
           time" and "strfmon" (an X/Open extension, not in the C standard) families (or other target-specific fami-
           lies).  Which functions are checked without format attributes having been specified depends on the standard
           version selected, and such checks of functions without the attribute specified are disabled by -ffreestand-
           ing or -fno-builtin.


           The formats are checked against the format features supported by GNU libc version 2.2.  These include all
           ISO C90 and C99 features, as well as features from the Single Unix Specification and some BSD and GNU
           extensions.  Other library implementations may not support all these features; GCC does not support warning
           about features that go beyond a particular library’s limitations.  However, if -pedantic is used with
           -Wformat, warnings will be given about format features not in the selected standard version (but not for
           "strfmon" formats, since those are not in any version of the C standard).


           Since -Wformat also checks for null format arguments for several functions, -Wformat also implies -Wnon-
           null.


           -Wformat is included in -Wall.  For more control over some aspects of format checking, the options -Wfor-
           mat-y2k, -Wno-format-extra-args, -Wno-format-zero-length, -Wformat-nonliteral, -Wformat-security, and
           -Wformat=2 are available, but are not included in -Wall.


       -Wformat-y2k
           If -Wformat is specified, also warn about "strftime" formats which may yield only a two-digit year.


       -Wno-format-extra-args
           If -Wformat is specified, do not warn about excess arguments to a "printf" or "scanf" format function.  The
           C standard specifies that such arguments are ignored.


           Where the unused arguments lie between used arguments that are specified with $ operand number specifica-
           tions, normally warnings are still given, since the implementation could not know what type to pass to
           "va_arg" to skip the unused arguments.  However, in the case of "scanf" formats, this option will suppress
           the warning if the unused arguments are all pointers, since the Single Unix Specification says that such
           unused arguments are allowed.


       -Wno-format-zero-length
           If -Wformat is specified, do not warn about zero-length formats.  The C standard specifies that zero-length
           formats are allowed.


       -Wformat-nonliteral
           If -Wformat is specified, also warn if the format string is not a string literal and so cannot be checked,
           unless the format function takes its format arguments as a "va_list".


       -Wformat-security
           If -Wformat is specified, also warn about uses of format functions that represent possible security prob-
           lems.  At present, this warns about calls to "printf" and "scanf" functions where the format string is not
           a string literal and there are no format arguments, as in "printf (foo);".  This may be a security hole if
           the format string came from untrusted input and contains %n.  (This is currently a subset of what -Wfor-
           mat-nonliteral warns about, but in future warnings may be added to -Wformat-security that are not included
           in -Wformat-nonliteral.)


       -Wformat=2
           Enable -Wformat plus format checks not included in -Wformat.  Currently equivalent to -Wformat -Wfor-
           mat-nonliteral -Wformat-security -Wformat-y2k.


       -Wnonnull
           Warn about passing a null pointer for arguments marked as requiring a non-null value by the "nonnull" func-
           tion attribute.


           -Wnonnull is included in -Wall and -Wformat.  It can be disabled with the -Wno-nonnull option.


       -Winit-self (C, C++, Objective-C and Objective-C++ only)
           Warn about uninitialized variables which are initialized with themselves.  Note this option can only be
           used with the -Wuninitialized option, which in turn only works with -O1 and above.


           For example, GCC will warn about "i" being uninitialized in the following snippet only when -Winit-self has
           been specified:


                   int f()
                   {
                     int i = i;
                     return i;
                   }


       -Wimplicit-int
           Warn when a declaration does not specify a type.  This warning is enabled by -Wall.


       -Wimplicit-function-declaration
       -Werror-implicit-function-declaration
           Give a warning (or error) whenever a function is used before being declared.  The form
           -Wno-error-implicit-function-declaration is not supported.  This warning is enabled by -Wall (as a warning,
           not an error).


       -Wimplicit
           Same as -Wimplicit-int and -Wimplicit-function-declaration.  This warning is enabled by -Wall.


       -Wmain
           Warn if the type of main is suspicious.  main should be a function with external linkage, returning int,
           taking either zero arguments, two, or three arguments of appropriate types.  This warning is enabled by
           -Wall.


       -Wmissing-braces
           Warn if an aggregate or union initializer is not fully bracketed.  In the following example, the initial-
           izer for a is not fully bracketed, but that for b is fully bracketed.


                   int a[2][2] = { 0, 1, 2, 3 };
                   int b[2][2] = { { 0, 1 }, { 2, 3 } };


           This warning is enabled by -Wall.


       -Wmissing-include-dirs (C, C++, Objective-C and Objective-C++ only)
           Warn if a user-supplied include directory does not exist.


       -Wparentheses
           Warn if parentheses are omitted in certain contexts, such as when there is an assignment in a context where
           a truth value is expected, or when operators are nested whose precedence people often get confused about.
           Only the warning for an assignment used as a truth value is supported when compiling C++; the other warn-
           ings are only supported when compiling C.


           Also warn if a comparison like x<=y<=z appears; this is equivalent to (x<=y ? 1 : 0) <= z, which is a dif-
           ferent interpretation from that of ordinary mathematical notation.


           Also warn about constructions where there may be confusion to which "if" statement an "else" branch
           belongs.  Here is an example of such a case:


                   {
                     if (a)
                       if (b)
                         foo ();
                     else
                       bar ();
                   }


           In C, every "else" branch belongs to the innermost possible "if" statement, which in this example is "if
           (b)".  This is often not what the programmer expected, as illustrated in the above example by indentation
           the programmer chose.  When there is the potential for this confusion, GCC will issue a warning when this
           flag is specified.  To eliminate the warning, add explicit braces around the innermost "if" statement so
           there is no way the "else" could belong to the enclosing "if".  The resulting code would look like this:


                   {
                     if (a)
                       {
                         if (b)
                           foo ();
                         else
                           bar ();
                       }
                   }


           This warning is enabled by -Wall.


       -Wsequence-point
           Warn about code that may have undefined semantics because of violations of sequence point rules in the C
           standard.


           The C standard defines the order in which expressions in a C program are evaluated in terms of sequence
           points, which represent a partial ordering between the execution of parts of the program: those executed
           before the sequence point, and those executed after it.  These occur after the evaluation of a full expres-
           sion (one which is not part of a larger expression), after the evaluation of the first operand of a "&&",
           "││", "? :" or "," (comma) operator, before a function is called (but after the evaluation of its arguments
           and the expression denoting the called function), and in certain other places.  Other than as expressed by
           the sequence point rules, the order of evaluation of subexpressions of an expression is not specified.  All
           these rules describe only a partial order rather than a total order, since, for example, if two functions
           are called within one expression with no sequence point between them, the order in which the functions are
           called is not specified.  However, the standards committee have ruled that function calls do not overlap.


           It is not specified when between sequence points modifications to the values of objects take effect.  Pro-
           grams whose behavior depends on this have undefined behavior; the C standard specifies that "Between the
           previous and next sequence point an object shall have its stored value modified at most once by the evalua-
           tion of an expression.  Furthermore, the prior value shall be read only to determine the value to be
           stored.".  If a program breaks these rules, the results on any particular implementation are entirely
           unpredictable.


           Examples of code with undefined behavior are "a = a++;", "a[n] = b[n++]" and "a[i++] = i;".  Some more com-
           plicated cases are not diagnosed by this option, and it may give an occasional false positive result, but
           in general it has been found fairly effective at detecting this sort of problem in programs.


           The present implementation of this option only works for C programs.  A future implementation may also work
           for C++ programs.


           The C standard is worded confusingly, therefore there is some debate over the precise meaning of the
           sequence point rules in subtle cases.  Links to discussions of the problem, including proposed formal defi-
           nitions, may be found on the GCC readings page, at <http://gcc.gnu.org/readings.html>.


           This warning is enabled by -Wall.


       -Wreturn-type
           Warn whenever a function is defined with a return-type that defaults to "int".  Also warn about any
           "return" statement with no return-value in a function whose return-type is not "void".


           For C, also warn if the return type of a function has a type qualifier such as "const".  Such a type quali-
           fier has no effect, since the value returned by a function is not an lvalue.  ISO C prohibits qualified
           "void" return types on function definitions, so such return types always receive a warning even without
           this option.


           For C++, a function without return type always produces a diagnostic message, even when -Wno-return-type is
           specified.  The only exceptions are main and functions defined in system headers.


           This warning is enabled by -Wall.


       -Wswitch
           Warn whenever a "switch" statement has an index of enumerated type and lacks a "case" for one or more of
           the named codes of that enumeration.  (The presence of a "default" label prevents this warning.)  "case"
           labels outside the enumeration range also provoke warnings when this option is used.  This warning is
           enabled by -Wall.


       -Wswitch-default
           Warn whenever a "switch" statement does not have a "default" case.


       -Wswitch-enum
           Warn whenever a "switch" statement has an index of enumerated type and lacks a "case" for one or more of
           the named codes of that enumeration.  "case" labels outside the enumeration range also provoke warnings
           when this option is used.


       -Wtrigraphs
           Warn if any trigraphs are encountered that might change the meaning of the program (trigraphs within com-
           ments are not warned about).  This warning is enabled by -Wall.


       -Wunused-function
           Warn whenever a static function is declared but not defined or a non-inline static function is unused.
           This warning is enabled by -Wall.


       -Wunused-label
           Warn whenever a label is declared but not used.  This warning is enabled by -Wall.


           To suppress this warning use the unused attribute.


       -Wunused-parameter
           Warn whenever a function parameter is unused aside from its declaration.


           To suppress this warning use the unused attribute.


       -Wunused-variable
           Warn whenever a local variable or non-constant static variable is unused aside from its declaration.  This
           warning is enabled by -Wall.


           To suppress this warning use the unused attribute.


       -Wunused-value
           Warn whenever a statement computes a result that is explicitly not used.  This warning is enabled by -Wall.


           To suppress this warning cast the expression to void.


       -Wunused
           All the above -Wunused options combined.


           In order to get a warning about an unused function parameter, you must either specify -Wextra -Wunused
           (note that -Wall implies -Wunused), or separately specify -Wunused-parameter.


       -Wuninitialized
           Warn if an automatic variable is used without first being initialized or if a variable may be clobbered by
           a "setjmp" call.


           These warnings are possible only in optimizing compilation, because they require data flow information that
           is computed only when optimizing.  If you don’t specify -O, you simply won’t get these warnings.


           If you want to warn about code which uses the uninitialized value of the variable in its own initializer,
           use the -Winit-self option.


           These warnings occur for individual uninitialized or clobbered elements of structure, union or array vari-
           ables as well as for variables which are uninitialized or clobbered as a whole.  They do not occur for
           variables or elements declared "volatile".  Because these warnings depend on optimization, the exact vari-
           ables or elements for which there are warnings will depend on the precise optimization options and version
           of GCC used.


           Note that there may be no warning about a variable that is used only to compute a value that itself is
           never used, because such computations may be deleted by data flow analysis before the warnings are printed.


           These warnings are made optional because GCC is not smart enough to see all the reasons why the code might
           be correct despite appearing to have an error.  Here is one example of how this can happen:


                   {
                     int x;
                     switch (y)
                       {
                       case 1: x = 1;
                         break;
                       case 2: x = 4;
                         break;
                       case 3: x = 5;
                       }
                     foo (x);
                   }


           If the value of "y" is always 1, 2 or 3, then "x" is always initialized, but GCC doesn’t know this.  Here
           is another common case:


                   {
                     int save_y;
                     if (change_y) save_y = y, y = new_y;
                     ...
                     if (change_y) y = save_y;
                   }


           This has no bug because "save_y" is used only if it is set.


           This option also warns when a non-volatile automatic variable might be changed by a call to "longjmp".
           These warnings as well are possible only in optimizing compilation.


           The compiler sees only the calls to "setjmp".  It cannot know where "longjmp" will be called; in fact, a
           signal handler could call it at any point in the code.  As a result, you may get a warning even when there
           is in fact no problem because "longjmp" cannot in fact be called at the place which would cause a problem.


           Some spurious warnings can be avoided if you declare all the functions you use that never return as "nore-
           turn".


           This warning is enabled by -Wall.


       -Wunknown-pragmas
           Warn when a #pragma directive is encountered which is not understood by GCC.  If this command line option
           is used, warnings will even be issued for unknown pragmas in system header files.  This is not the case if
           the warnings were only enabled by the -Wall command line option.


       -Wno-pragmas
           Do not warn about misuses of pragmas, such as incorrect parameters, invalid syntax, or conflicts between
           pragmas.  See also -Wunknown-pragmas.


       -Wstrict-aliasing
           This option is only active when -fstrict-aliasing is active.  It warns about code which might break the
           strict aliasing rules that the compiler is using for optimization.  The warning does not catch all cases,
           but does attempt to catch the more common pitfalls.  It is included in -Wall.


       -Wstrict-aliasing=2
           This option is only active when -fstrict-aliasing is active.  It warns about code which might break the
           strict aliasing rules that the compiler is using for optimization.  This warning catches more cases than
           -Wstrict-aliasing, but it will also give a warning for some ambiguous cases that are safe.


       -Wall
           All of the above -W options combined.  This enables all the warnings about constructions that some users
           consider questionable, and that are easy to avoid (or modify to prevent the warning), even in conjunction
           with macros.  This also enables some language-specific warnings described in C++ Dialect Options and Objec-
           tive-C and Objective-C++ Dialect Options.


       The following -W... options are not implied by -Wall.  Some of them warn about constructions that users gener-
       ally do not consider questionable, but which occasionally you might wish to check for; others warn about con-
       structions that are necessary or hard to avoid in some cases, and there is no simple way to modify the code to
       suppress the warning.


       -Wextra
           (This option used to be called -W.  The older name is still supported, but the newer name is more descrip-
           tive.)  Print extra warning messages for these events:


           *   A function can return either with or without a value.  (Falling off the end of the function body is
               considered returning without a value.)  For example, this function would evoke such a warning:


                       foo (a)
                       {
                         if (a > 0)
                           return a;
                       }


           *   An expression-statement or the left-hand side of a comma expression contains no side effects.  To sup-
               press the warning, cast the unused expression to void.  For example, an expression such as x[i,j] will
               cause a warning, but x[(void)i,j] will not.


           *   An unsigned value is compared against zero with < or >=.


           *   Storage-class specifiers like "static" are not the first things in a declaration.  According to the C
               Standard, this usage is obsolescent.


           *   If -Wall or -Wunused is also specified, warn about unused arguments.


           *   A comparison between signed and unsigned values could produce an incorrect result when the signed value
               is converted to unsigned.  (But don’t warn if -Wno-sign-compare is also specified.)


           *   An aggregate has an initializer which does not initialize all members.  This warning can be indepen-
               dently controlled by -Wmissing-field-initializers.


           *   A function parameter is declared without a type specifier in K&R-style functions:


                       void foo(bar) { }


           *   An empty body occurs in an if or else statement.


           *   A pointer is compared against integer zero with <, <=, >, or >=.


           *   A variable might be changed by longjmp or vfork.


           *   Any of several floating-point events that often indicate errors, such as overflow, underflow, loss of
               precision, etc.


           *<(C++ only)>
               An enumerator and a non-enumerator both appear in a conditional expression.


           *<(C++ only)>
               A non-static reference or non-static const member appears in a class without constructors.


           *<(C++ only)>
               Ambiguous virtual bases.


           *<(C++ only)>
               Subscripting an array which has been declared register.


           *<(C++ only)>
               Taking the address of a variable which has been declared register.


           *<(C++ only)>
               A base class is not initialized in a derived class’ copy constructor.


       -Wno-div-by-zero
           Do not warn about compile-time integer division by zero.  Floating point division by zero is not warned
           about, as it can be a legitimate way of obtaining infinities and NaNs.


       -Wsystem-headers
           Print warning messages for constructs found in system header files.  Warnings from system headers are nor-
           mally suppressed, on the assumption that they usually do not indicate real problems and would only make the
           compiler output harder to read.  Using this command line option tells GCC to emit warnings from system
           headers as if they occurred in user code.  However, note that using -Wall in conjunction with this option
           will not warn about unknown pragmas in system headers---for that, -Wunknown-pragmas must also be used.


       -Wfloat-equal
           Warn if floating point values are used in equality comparisons.


           The idea behind this is that sometimes it is convenient (for the programmer) to consider floating-point
           values as approximations to infinitely precise real numbers.  If you are doing this, then you need to com-
           pute (by analyzing the code, or in some other way) the maximum or likely maximum error that the computation
           introduces, and allow for it when performing comparisons (and when producing output, but that’s a different
           problem).  In particular, instead of testing for equality, you would check to see whether the two values
           have ranges that overlap; and this is done with the relational operators, so equality comparisons are prob-
           ably mistaken.


       -Wtraditional (C only)
           Warn about certain constructs that behave differently in traditional and ISO C.  Also warn about ISO C con-
           structs that have no traditional C equivalent, and/or problematic constructs which should be avoided.


           *   Macro parameters that appear within string literals in the macro body.  In traditional C macro replace-
               ment takes place within string literals, but does not in ISO C.


           *   In traditional C, some preprocessor directives did not exist.  Traditional preprocessors would only
               consider a line to be a directive if the # appeared in column 1 on the line.  Therefore -Wtraditional
               warns about directives that traditional C understands but would ignore because the # does not appear as
               the first character on the line.  It also suggests you hide directives like #pragma not understood by
               traditional C by indenting them.  Some traditional implementations would not recognize #elif, so it
               suggests avoiding it altogether.


           *   A function-like macro that appears without arguments.


           *   The unary plus operator.


           *   The U integer constant suffix, or the F or L floating point constant suffixes.  (Traditional C does
               support the L suffix on integer constants.)  Note, these suffixes appear in macros defined in the sys-
               tem headers of most modern systems, e.g. the _MIN/_MAX macros in "<limits.h>".  Use of these macros in
               user code might normally lead to spurious warnings, however GCC’s integrated preprocessor has enough
               context to avoid warning in these cases.


           *   A function declared external in one block and then used after the end of the block.


           *   A "switch" statement has an operand of type "long".


           *   A non-"static" function declaration follows a "static" one.  This construct is not accepted by some
               traditional C compilers.


           *   The ISO type of an integer constant has a different width or signedness from its traditional type.
               This warning is only issued if the base of the constant is ten.  I.e. hexadecimal or octal values,
               which typically represent bit patterns, are not warned about.


           *   Usage of ISO string concatenation is detected.


           *   Initialization of automatic aggregates.


           *   Identifier conflicts with labels.  Traditional C lacks a separate namespace for labels.


           *   Initialization of unions.  If the initializer is zero, the warning is omitted.  This is done under the
               assumption that the zero initializer in user code appears conditioned on e.g. "__STDC__" to avoid miss-
               ing initializer warnings and relies on default initialization to zero in the traditional C case.


           *   Conversions by prototypes between fixed/floating point values and vice versa.  The absence of these
               prototypes when compiling with traditional C would cause serious problems.  This is a subset of the
               possible conversion warnings, for the full set use -Wconversion.


           *   Use of ISO C style function definitions.  This warning intentionally is not issued for prototype decla-
               rations or variadic functions because these ISO C features will appear in your code when using libib-
               erty’s traditional C compatibility macros, "PARAMS" and "VPARAMS".  This warning is also bypassed for
               nested functions because that feature is already a GCC extension and thus not relevant to traditional C
               compatibility.


       -Wdeclaration-after-statement (C only)
           Warn when a declaration is found after a statement in a block.  This construct, known from C++, was intro-
           duced with ISO C99 and is by default allowed in GCC.  It is not supported by ISO C90 and was not supported
           by GCC versions before GCC 3.0.


       -Wundef
           Warn if an undefined identifier is evaluated in an #if directive.


       -Wno-endif-labels
           Do not warn whenever an #else or an #endif are followed by text.


       -Wshadow
           Warn whenever a local variable shadows another local variable, parameter or global variable or whenever a
           built-in function is shadowed.


       -Wlarger-than-len
           Warn whenever an object of larger than len bytes is defined.


       -Wunsafe-loop-optimizations
           Warn if the loop cannot be optimized because the compiler could not assume anything on the bounds of the
           loop indices.  With -funsafe-loop-optimizations warn if the compiler made such assumptions.


       -Wpointer-arith
           Warn about anything that depends on the "size of" a function type or of "void".  GNU C assigns these types
           a size of 1, for convenience in calculations with "void *" pointers and pointers to functions.


       -Wbad-function-cast (C only)
           Warn whenever a function call is cast to a non-matching type.  For example, warn if "int malloc()" is cast
           to "anything *".


       -Wc++-compat
           Warn about ISO C constructs that are outside of the common subset of ISO C and ISO C++, e.g. request for
           implicit conversion from "void *" to a pointer to non-"void" type.


       -Wcast-qual
           Warn whenever a pointer is cast so as to remove a type qualifier from the target type.  For example, warn
           if a "const char *" is cast to an ordinary "char *".


       -Wcast-align
           Warn whenever a pointer is cast such that the required alignment of the target is increased.  For example,
           warn if a "char *" is cast to an "int *" on machines where integers can only be accessed at two- or four-
           byte boundaries.


       -Wwrite-strings
           When compiling C, give string constants the type "const char[length]" so that copying the address of one
           into a non-"const" "char *" pointer will get a warning; when compiling C++, warn about the deprecated con-
           version from string constants to "char *".  These warnings will help you find at compile time code that can
           try to write into a string constant, but only if you have been very careful about using "const" in declara-
           tions and prototypes.  Otherwise, it will just be a nuisance; this is why we did not make -Wall request
           these warnings.


       -Wconversion
           Warn if a prototype causes a type conversion that is different from what would happen to the same argument
           in the absence of a prototype.  This includes conversions of fixed point to floating and vice versa, and
           conversions changing the width or signedness of a fixed point argument except when the same as the default
           promotion.


           Also, warn if a negative integer constant expression is implicitly converted to an unsigned type.  For
           example, warn about the assignment "x = -1" if "x" is unsigned.  But do not warn about explicit casts like
           "(unsigned) -1".


       -Wsign-compare
           Warn when a comparison between signed and unsigned values could produce an incorrect result when the signed
           value is converted to unsigned.  This warning is also enabled by -Wextra; to get the other warnings of
           -Wextra without this warning, use -Wextra -Wno-sign-compare.


       -Waggregate-return
           Warn if any functions that return structures or unions are defined or called.  (In languages where you can
           return an array, this also elicits a warning.)


       -Wno-attributes
           Do not warn if an unexpected "__attribute__" is used, such as unrecognized attributes, function attributes
           applied to variables, etc.  This will not stop errors for incorrect use of supported attributes.


       -Wstrict-prototypes (C only)
           Warn if a function is declared or defined without specifying the argument types.  (An old-style function
           definition is permitted without a warning if preceded by a declaration which specifies the argument types.)


       -Wold-style-definition (C only)
           Warn if an old-style function definition is used.  A warning is given even if there is a previous proto-
           type.


       -Wmissing-prototypes (C only)
           Warn if a global function is defined without a previous prototype declaration.  This warning is issued even
           if the definition itself provides a prototype.  The aim is to detect global functions that fail to be
           declared in header files.


       -Wmissing-declarations (C only)
           Warn if a global function is defined without a previous declaration.  Do so even if the definition itself
           provides a prototype.  Use this option to detect global functions that are not declared in header files.


       -Wmissing-field-initializers
           Warn if a structure’s initializer has some fields missing.  For example, the following code would cause
           such a warning, because "x.h" is implicitly zero:


                   struct s { int f, g, h; };
                   struct s x = { 3, 4 };


           This option does not warn about designated initializers, so the following modification would not trigger a
           warning:


                   struct s { int f, g, h; };
                   struct s x = { .f = 3, .g = 4 };


           This warning is included in -Wextra.  To get other -Wextra warnings without this one, use -Wextra
           -Wno-missing-field-initializers.


       -Wmissing-noreturn
           Warn about functions which might be candidates for attribute "noreturn".  Note these are only possible can-
           didates, not absolute ones.  Care should be taken to manually verify functions actually do not ever return
           before adding the "noreturn" attribute, otherwise subtle code generation bugs could be introduced.  You
           will not get a warning for "main" in hosted C environments.


       -Wmissing-format-attribute
           Warn about function pointers which might be candidates for "format" attributes.  Note these are only possi-
           ble candidates, not absolute ones.  GCC will guess that function pointers with "format" attributes that are
           used in assignment, initialization, parameter passing or return statements should have a corresponding
           "format" attribute in the resulting type.  I.e. the left-hand side of the assignment or initialization, the
           type of the parameter variable, or the return type of the containing function respectively should also have
           a "format" attribute to avoid the warning.


           GCC will also warn about function definitions which might be candidates for "format" attributes.  Again,
           these are only possible candidates.  GCC will guess that "format" attributes might be appropriate for any
           function that calls a function like "vprintf" or "vscanf", but this might not always be the case, and some
           functions for which "format" attributes are appropriate may not be detected.


       -Wno-multichar
           Do not warn if a multicharacter constant (’FOOF’) is used.  Usually they indicate a typo in the user’s
           code, as they have implementation-defined values, and should not be used in portable code.


       -Wnormalized=<none│id│nfc│nfkc>
           In ISO C and ISO C++, two identifiers are different if they are different sequences of characters.  How-
           ever, sometimes when characters outside the basic ASCII character set are used, you can have two different
           character sequences that look the same.  To avoid confusion, the ISO 10646 standard sets out some normal-
           ization rules which when applied ensure that two sequences that look the same are turned into the same
           sequence.  GCC can warn you if you are using identifiers which have not been normalized; this option con-
           trols that warning.


           There are four levels of warning that GCC supports.  The default is -Wnormalized=nfc, which warns about any
           identifier which is not in the ISO 10646 "C" normalized form, NFC.  NFC is the recommended form for most
           uses.


           Unfortunately, there are some characters which ISO C and ISO C++ allow in identifiers that when turned into
           NFC aren’t allowable as identifiers.  That is, there’s no way to use these symbols in portable ISO C or C++
           and have all your identifiers in NFC.  -Wnormalized=id suppresses the warning for these characters.  It is
           hoped that future versions of the standards involved will correct this, which is why this option is not the
           default.


           You can switch the warning off for all characters by writing -Wnormalized=none.  You would only want to do
           this if you were using some other normalization scheme (like "D"), because otherwise you can easily create
           bugs that are literally impossible to see.


           Some characters in ISO 10646 have distinct meanings but look identical in some fonts or display methodolo-
           gies, especially once formatting has been applied.  For instance "\u207F", "SUPERSCRIPT LATIN SMALL LETTER
           N", will display just like a regular "n" which has been placed in a superscript.  ISO 10646 defines the
           NFKC normalisation scheme to convert all these into a standard form as well, and GCC will warn if your code
           is not in NFKC if you use -Wnormalized=nfkc.  This warning is comparable to warning about every identifier
           that contains the letter O because it might be confused with the digit 0, and so is not the default, but
           may be useful as a local coding convention if the programming environment is unable to be fixed to display
           these characters distinctly.


       -Wno-deprecated-declarations
           Do not warn about uses of functions, variables, and types marked as deprecated by using the "deprecated"
           attribute.  (@pxref{Function Attributes}, @pxref{Variable Attributes}, @pxref{Type Attributes}.)


       -Wpacked
           Warn if a structure is given the packed attribute, but the packed attribute has no effect on the layout or
           size of the structure.  Such structures may be mis-aligned for little benefit.  For instance, in this code,
           the variable "f.x" in "struct bar" will be misaligned even though "struct bar" does not itself have the
           packed attribute:


                   struct foo {
                     int x;
                     char a, b, c, d;
                   } __attribute__((packed));
                   struct bar {
                     char z;
                     struct foo f;
                   };


       -Wpadded
           Warn if padding is included in a structure, either to align an element of the structure or to align the
           whole structure.  Sometimes when this happens it is possible to rearrange the fields of the structure to
           reduce the padding and so make the structure smaller.


       -Wredundant-decls
           Warn if anything is declared more than once in the same scope, even in cases where multiple declaration is
           valid and changes nothing.


       -Wnested-externs (C only)
           Warn if an "extern" declaration is encountered within a function.


       -Wunreachable-code
           Warn if the compiler detects that code will never be executed.


           This option is intended to warn when the compiler detects that at least a whole line of source code will
           never be executed, because some condition is never satisfied or because it is after a procedure that never
           returns.


           It is possible for this option to produce a warning even though there are circumstances under which part of
           the affected line can be executed, so care should be taken when removing apparently-unreachable code.


           For instance, when a function is inlined, a warning may mean that the line is unreachable in only one
           inlined copy of the function.


           This option is not made part of -Wall because in a debugging version of a program there is often substan-
           tial code which checks correct functioning of the program and is, hopefully, unreachable because the pro-
           gram does work.  Another common use of unreachable code is to provide behavior which is selectable at com-
           pile-time.


       -Winline
           Warn if a function can not be inlined and it was declared as inline.  Even with this option, the compiler
           will not warn about failures to inline functions declared in system headers.


           The compiler uses a variety of heuristics to determine whether or not to inline a function.  For example,
           the compiler takes into account the size of the function being inlined and the amount of inlining that has
           already been done in the current function.  Therefore, seemingly insignificant changes in the source pro-
           gram can cause the warnings produced by -Winline to appear or disappear.


       -Wno-invalid-offsetof (C++ only)
           Suppress warnings from applying the offsetof macro to a non-POD type.  According to the 1998 ISO C++ stan-
           dard, applying offsetof to a non-POD type is undefined.  In existing C++ implementations, however, offsetof
           typically gives meaningful results even when applied to certain kinds of non-POD types. (Such as a simple
           struct that fails to be a POD type only by virtue of having a constructor.)  This flag is for users who are
           aware that they are writing nonportable code and who have deliberately chosen to ignore the warning about
           it.


           The restrictions on offsetof may be relaxed in a future version of the C++ standard.


       -Wno-int-to-pointer-cast (C only)
           Suppress warnings from casts to pointer type of an integer of a different size.


       -Wno-pointer-to-int-cast (C only)
           Suppress warnings from casts from a pointer to an integer type of a different size.


       -Winvalid-pch
           Warn if a precompiled header is found in the search path but can’t be used.


       -Wlong-long
           Warn if long long type is used.  This is default.  To inhibit the warning messages, use -Wno-long-long.
           Flags -Wlong-long and -Wno-long-long are taken into account only when -pedantic flag is used.


       -Wvariadic-macros
           Warn if variadic macros are used in pedantic ISO C90 mode, or the GNU alternate syntax when in pedantic ISO
           C99 mode.  This is default.  To inhibit the warning messages, use -Wno-variadic-macros.


       -Wvolatile-register-var
           Warn if a register variable is declared volatile.  The volatile modifier does not inhibit all optimizations
           that may eliminate reads and/or writes to register variables.


       -Wdisabled-optimization
           Warn if a requested optimization pass is disabled.  This warning does not generally indicate that there is
           anything wrong with your code; it merely indicates that GCC’s optimizers were unable to handle the code
           effectively.  Often, the problem is that your code is too big or too complex; GCC will refuse to optimize
           programs when the optimization itself is likely to take inordinate amounts of time.


       -Wpointer-sign
           Warn for pointer argument passing or assignment with different signedness.  This option is only supported
           for C and Objective-C.  It is implied by -Wall and by -pedantic, which can be disabled with
           -Wno-pointer-sign.


       -Werror
           Make all warnings into errors.


       -Wstack-protector
           This option is only active when -fstack-protector is active.  It warns about functions that will not be
           protected against stack smashing.


       Options for Debugging Your Program or GCC


       GCC has various special options that are used for debugging either your program or GCC:


       -g  Produce debugging information in the operating system’s native format (stabs, COFF, XCOFF, or DWARF 2).
           GDB can work with this debugging information.


           On most systems that use stabs format, -g enables use of extra debugging information that only GDB can use;
           this extra information makes debugging work better in GDB but will probably make other debuggers crash or
           refuse to read the program.  If you want to control for certain whether to generate the extra information,
           use -gstabs+, -gstabs, -gxcoff+, -gxcoff, or -gvms (see below).


           GCC allows you to use -g with -O.  The shortcuts taken by optimized code may occasionally produce surpris-
           ing results: some variables you declared may not exist at all; flow of control may briefly move where you
           did not expect it; some statements may not be executed because they compute constant results or their val-
           ues were already at hand; some statements may execute in different places because they were moved out of
           loops.


           Nevertheless it proves possible to debug optimized output.  This makes it reasonable to use the optimizer
           for programs that might have bugs.


           The following options are useful when GCC is generated with the capability for more than one debugging for-
           mat.


       -ggdb
           Produce debugging information for use by GDB.  This means to use the most expressive format available
           (DWARF 2, stabs, or the native format if neither of those are supported), including GDB extensions if at
           all possible.


       -gstabs
           Produce debugging information in stabs format (if that is supported), without GDB extensions.  This is the
           format used by DBX on most BSD systems.  On MIPS, Alpha and System V Release 4 systems this option produces
           stabs debugging output which is not understood by DBX or SDB.  On System V Release 4 systems this option
           requires the GNU assembler.


       -feliminate-unused-debug-symbols
           Produce debugging information in stabs format (if that is supported), for only symbols that are actually
           used.


       -gstabs+
           Produce debugging information in stabs format (if that is supported), using GNU extensions understood only
           by the GNU debugger (GDB).  The use of these extensions is likely to make other debuggers crash or refuse
           to read the program.


       -gcoff
           Produce debugging information in COFF format (if that is supported).  This is the format used by SDB on
           most System V systems prior to System V Release 4.


       -gxcoff
           Produce debugging information in XCOFF format (if that is supported).  This is the format used by the DBX
           debugger on IBM RS/6000 systems.


       -gxcoff+
           Produce debugging information in XCOFF format (if that is supported), using GNU extensions understood only
           by the GNU debugger (GDB).  The use of these extensions is likely to make other debuggers crash or refuse
           to read the program, and may cause assemblers other than the GNU assembler (GAS) to fail with an error.


       -gdwarf-2
           Produce debugging information in DWARF version 2 format (if that is supported).  This is the format used by
           DBX on IRIX 6.  With this option, GCC uses features of DWARF version 3 when they are useful; version 3 is
           upward compatible with version 2, but may still cause problems for older debuggers.


       -gvms
           Produce debugging information in VMS debug format (if that is supported).  This is the format used by DEBUG
           on VMS systems.


       -glevel
       -ggdblevel
       -gstabslevel
       -gcofflevel
       -gxcofflevel
       -gvmslevel
           Request debugging information and also use level to specify how much information.  The default level is 2.


           Level 1 produces minimal information, enough for making backtraces in parts of the program that you don’t
           plan to debug.  This includes descriptions of functions and external variables, but no information about
           local variables and no line numbers.


           Level 3 includes extra information, such as all the macro definitions present in the program.  Some debug-
           gers support macro expansion when you use -g3.


           -gdwarf-2 does not accept a concatenated debug level, because GCC used to support an option -gdwarf that
           meant to generate debug information in version 1 of the DWARF format (which is very different from version
           2), and it would have been too confusing.  That debug format is long obsolete, but the option cannot be
           changed now.  Instead use an additional -glevel option to change the debug level for DWARF2.


       -feliminate-dwarf2-dups
           Compress DWARF2 debugging information by eliminating duplicated information about each symbol.  This option
           only makes sense when generating DWARF2 debugging information with -gdwarf-2.


       -p  Generate extra code to write profile information suitable for the analysis program prof.  You must use this
           option when compiling the source files you want data about, and you must also use it when linking.


       -pg Generate extra code to write profile information suitable for the analysis program gprof.  You must use
           this option when compiling the source files you want data about, and you must also use it when linking.


       -Q  Makes the compiler print out each function name as it is compiled, and print some statistics about each
           pass when it finishes.


       -ftime-report
           Makes the compiler print some statistics about the time consumed by each pass when it finishes.


       -fmem-report
           Makes the compiler print some statistics about permanent memory allocation when it finishes.


       -fprofile-arcs
           Add code so that program flow arcs are instrumented.  During execution the program records how many times
           each branch and call is executed and how many times it is taken or returns.  When the compiled program
           exits it saves this data to a file called auxname.gcda for each source file.  The data may be used for pro-
           file-directed optimizations (-fbranch-probabilities), or for test coverage analysis (-ftest-coverage).
           Each object file’s auxname is generated from the name of the output file, if explicitly specified and it is
           not the final executable, otherwise it is the basename of the source file.  In both cases any suffix is
           removed (e.g. foo.gcda for input file dir/foo.c, or dir/foo.gcda for output file specified as -o
           dir/foo.o).


       --coverage
           This option is used to compile and link code instrumented for coverage analysis.  The option is a synonym
           for -fprofile-arcs -ftest-coverage (when compiling) and -lgcov (when linking).  See the documentation for
           those options for more details.


           @bullet
               Compile the source files with -fprofile-arcs plus optimization and code generation options.  For test
               coverage analysis, use the additional -ftest-coverage option.  You do not need to profile every source
               file in a program.


           @cvmmfu
               Link your object files with -lgcov or -fprofile-arcs (the latter implies the former).


           @dwnngv
               Run the program on a representative workload to generate the arc profile information.  This may be
               repeated any number of times.  You can run concurrent instances of your program, and provided that the
               file system supports locking, the data files will be correctly updated.  Also "fork" calls are detected
               and correctly handled (double counting will not happen).


           @exoohw
               For profile-directed optimizations, compile the source files again with the same optimization and code
               generation options plus -fbranch-probabilities.


           @fyppix
               For test coverage analysis, use gcov to produce human readable information from the .gcno and .gcda
               files.  Refer to the gcov documentation for further information.


           With -fprofile-arcs, for each function of your program GCC creates a program flow graph, then finds a span-
           ning tree for the graph.  Only arcs that are not on the spanning tree have to be instrumented: the compiler
           adds code to count the number of times that these arcs are executed.  When an arc is the only exit or only
           entrance to a block, the instrumentation code can be added to the block; otherwise, a new basic block must
           be created to hold the instrumentation code.


       -ftest-coverage
           Produce a notes file that the gcov code-coverage utility can use to show program coverage.  Each source
           file’s note file is called auxname.gcno.  Refer to the -fprofile-arcs option above for a description of
           auxname and instructions on how to generate test coverage data.  Coverage data will match the source files
           more closely, if you do not optimize.


       -dletters
       -fdump-rtl-pass
           Says to make debugging dumps during compilation at times specified by letters.    This is used for debug-
           ging the RTL-based passes of the compiler.  The file names for most of the dumps are made by appending a
           pass number and a word to the dumpname.  dumpname is generated from the name of the output file, if explic-
           itly specified and it is not an executable, otherwise it is the basename of the source file.


           Most debug dumps can be enabled either passing a letter to the -d option, or with a long -fdump-rtl switch;
           here are the possible letters for use in letters and pass, and their meanings:


           -dA Annotate the assembler output with miscellaneous debugging information.


           -db
           -fdump-rtl-bp
               Dump after computing branch probabilities, to file.09.bp.


           -dB
           -fdump-rtl-bbro
               Dump after block reordering, to file.30.bbro.


           -dc
           -fdump-rtl-combine
               Dump after instruction combination, to the file file.17.combine.


           -dC
           -fdump-rtl-ce1
           -fdump-rtl-ce2
               -dC and -fdump-rtl-ce1 enable dumping after the first if conversion, to the file file.11.ce1.  -dC and
               -fdump-rtl-ce2 enable dumping after the second if conversion, to the file file.18.ce2.


           -dd
           -fdump-rtl-btl
           -fdump-rtl-dbr
               -dd and -fdump-rtl-btl enable dumping after branch target load optimization, to file.31.btl.  -dd and
               -fdump-rtl-dbr enable dumping after delayed branch scheduling, to file.36.dbr.


           -dD Dump all macro definitions, at the end of preprocessing, in addition to normal output.


           -dE
           -fdump-rtl-ce3
               Dump after the third if conversion, to file.28.ce3.


           -df
           -fdump-rtl-cfg
           -fdump-rtl-life
               -df and -fdump-rtl-cfg enable dumping after control and data flow analysis, to file.08.cfg.  -df and
               -fdump-rtl-cfg enable dumping dump after life analysis, to file.16.life.


           -dg
           -fdump-rtl-greg
               Dump after global register allocation, to file.23.greg.


           -dG
           -fdump-rtl-gcse
           -fdump-rtl-bypass
               -dG and -fdump-rtl-gcse enable dumping after GCSE, to file.05.gcse.  -dG and -fdump-rtl-bypass enable
               dumping after jump bypassing and control flow optimizations, to file.07.bypass.


           -dh
           -fdump-rtl-eh
               Dump after finalization of EH handling code, to file.02.eh.


           -di
           -fdump-rtl-sibling
               Dump after sibling call optimizations, to file.01.sibling.


           -dj
           -fdump-rtl-jump
               Dump after the first jump optimization, to file.03.jump.


           -dk
           -fdump-rtl-stack
               Dump after conversion from registers to stack, to file.33.stack.


           -dl
           -fdump-rtl-lreg
               Dump after local register allocation, to file.22.lreg.


           -dL
           -fdump-rtl-loop
           -fdump-rtl-loop2
               -dL and -fdump-rtl-loop enable dumping after the first loop optimization pass, to file.06.loop.  -dL
               and -fdump-rtl-loop2 enable dumping after the second pass, to file.13.loop2.


           -dm
           -fdump-rtl-sms
               Dump after modulo scheduling, to file.20.sms.


           -dM
           -fdump-rtl-mach
               Dump after performing the machine dependent reorganization pass, to file.35.mach.


           -dn
           -fdump-rtl-rnreg
               Dump after register renumbering, to file.29.rnreg.


           -dN
           -fdump-rtl-regmove
               Dump after the register move pass, to file.19.regmove.


           -do
           -fdump-rtl-postreload
               Dump after post-reload optimizations, to file.24.postreload.


           -dr
           -fdump-rtl-expand
               Dump after RTL generation, to file.00.expand.


           -dR
           -fdump-rtl-sched2
               Dump after the second scheduling pass, to file.32.sched2.


           -ds
           -fdump-rtl-cse
               Dump after CSE (including the jump optimization that sometimes follows CSE), to file.04.cse.


           -dS
           -fdump-rtl-sched
               Dump after the first scheduling pass, to file.21.sched.


           -dt
           -fdump-rtl-cse2
               Dump after the second CSE pass (including the jump optimization that sometimes follows CSE), to
               file.15.cse2.


           -dT
           -fdump-rtl-tracer
               Dump after running tracer, to file.12.tracer.


           -dV
           -fdump-rtl-vpt
           -fdump-rtl-vartrack
               -dV and -fdump-rtl-vpt enable dumping after the value profile transformations, to file.10.vpt.  -dV and
               -fdump-rtl-vartrack enable dumping after variable tracking, to file.34.vartrack.


           -dw
           -fdump-rtl-flow2
               Dump after the second flow pass, to file.26.flow2.


           -dz
           -fdump-rtl-peephole2
               Dump after the peephole pass, to file.27.peephole2.


           -dZ
           -fdump-rtl-web
               Dump after live range splitting, to file.14.web.


           -da
           -fdump-rtl-all
               Produce all the dumps listed above.


           -dH Produce a core dump whenever an error occurs.


           -dm Print statistics on memory usage, at the end of the run, to standard error.


           -dp Annotate the assembler output with a comment indicating which pattern and alternative was used.  The
               length of each instruction is also printed.


           -dP Dump the RTL in the assembler output as a comment before each instruction.  Also turns on -dp annota-
               tion.


           -dv For each of the other indicated dump files (either with -d or -fdump-rtl-pass), dump a representation
               of the control flow graph suitable for viewing with VCG to file.pass.vcg.


           -dx Just generate RTL for a function instead of compiling it.  Usually used with r (-fdump-rtl-expand).


           -dy Dump debugging information during parsing, to standard error.


       -fdump-unnumbered
           When doing debugging dumps (see -d option above), suppress instruction numbers and line number note output.
           This makes it more feasible to use diff on debugging dumps for compiler invocations with different options,
           in particular with and without -g.


       -fdump-translation-unit (C++ only)
       -fdump-translation-unit-options (C++ only)
           Dump a representation of the tree structure for the entire translation unit to a file.  The file name is
           made by appending .tu to the source file name.  If the -options form is used, options controls the details
           of the dump as described for the -fdump-tree options.


       -fdump-class-hierarchy (C++ only)
       -fdump-class-hierarchy-options (C++ only)
           Dump a representation of each class’s hierarchy and virtual function table layout to a file.  The file name
           is made by appending .class to the source file name.  If the -options form is used, options controls the
           details of the dump as described for the -fdump-tree options.


       -fdump-ipa-switch
           Control the dumping at various stages of inter-procedural analysis language tree to a file.  The file name
           is generated by appending a switch specific suffix to the source file name.  The following dumps are possi-
           ble:


           all Enables all inter-procedural analysis dumps; currently the only produced dump is the cgraph dump.


           cgraph
               Dumps information about call-graph optimization, unused function removal, and inlining decisions.


       -fdump-tree-switch
       -fdump-tree-switch-options
           Control the dumping at various stages of processing the intermediate language tree to a file.  The file
           name is generated by appending a switch specific suffix to the source file name.  If the -options form is
           used, options is a list of - separated options that control the details of the dump.  Not all options are
           applicable to all dumps, those which are not meaningful will be ignored.  The following options are avail-
           able


           address
               Print the address of each node.  Usually this is not meaningful as it changes according to the environ-
               ment and source file.  Its primary use is for tying up a dump file with a debug environment.


           slim
               Inhibit dumping of members of a scope or body of a function merely because that scope has been reached.
               Only dump such items when they are directly reachable by some other path.  When dumping pretty-printed
               trees, this option inhibits dumping the bodies of control structures.


           raw Print a raw representation of the tree.  By default, trees are pretty-printed into a C-like representa-
               tion.


           details
               Enable more detailed dumps (not honored by every dump option).


           stats
               Enable dumping various statistics about the pass (not honored by every dump option).


           blocks
               Enable showing basic block boundaries (disabled in raw dumps).


           vops
               Enable showing virtual operands for every statement.


           lineno
               Enable showing line numbers for statements.


           uid Enable showing the unique ID ("DECL_UID") for each variable.


           all Turn on all options, except raw, slim and lineno.


           The following tree dumps are possible:


           original
               Dump before any tree based optimization, to file.original.


           optimized
               Dump after all tree based optimization, to file.optimized.


           inlined
               Dump after function inlining, to file.inlined.


           gimple
               Dump each function before and after the gimplification pass to a file.  The file name is made by
               appending .gimple to the source file name.


           cfg Dump the control flow graph of each function to a file.  The file name is made by appending .cfg to the
               source file name.


           vcg Dump the control flow graph of each function to a file in VCG format.  The file name is made by append-
               ing .vcg to the source file name.  Note that if the file contains more than one function, the generated
               file cannot be used directly by VCG.  You will need to cut and paste each function’s graph into its own
               separate file first.


           ch  Dump each function after copying loop headers.  The file name is made by appending .ch to the source
               file name.


           ssa Dump SSA related information to a file.  The file name is made by appending .ssa to the source file
               name.


           salias
               Dump structure aliasing variable information to a file.  This file name is made by appending .salias to
               the source file name.


           alias
               Dump aliasing information for each function.  The file name is made by appending .alias to the source
               file name.


           ccp Dump each function after CCP.  The file name is made by appending .ccp to the source file name.


           storeccp
               Dump each function after STORE-CCP.  The file name is made by appending .storeccp to the source file
               name.


           pre Dump trees after partial redundancy elimination.  The file name is made by appending .pre to the source
               file name.


           fre Dump trees after full redundancy elimination.  The file name is made by appending .fre to the source
               file name.


           copyprop
               Dump trees after copy propagation.  The file name is made by appending .copyprop to the source file
               name.


           store_copyprop
               Dump trees after store copy-propagation.  The file name is made by appending .store_copyprop to the
               source file name.


           dce Dump each function after dead code elimination.  The file name is made by appending .dce to the source
               file name.


           mudflap
               Dump each function after adding mudflap instrumentation.  The file name is made by appending .mudflap
               to the source file name.


           sra Dump each function after performing scalar replacement of aggregates.  The file name is made by append-
               ing .sra to the source file name.


           sink
               Dump each function after performing code sinking.  The file name is made by appending .sink to the
               source file name.


           dom Dump each function after applying dominator tree optimizations.  The file name is made by appending
               .dom to the source file name.


           dse Dump each function after applying dead store elimination.  The file name is made by appending .dse to
               the source file name.


           phiopt
               Dump each function after optimizing PHI nodes into straightline code.  The file name is made by append-
               ing .phiopt to the source file name.


           forwprop
               Dump each function after forward propagating single use variables.  The file name is made by appending
               .forwprop to the source file name.


           copyrename
               Dump each function after applying the copy rename optimization.  The file name is made by appending
               .copyrename to the source file name.


           nrv Dump each function after applying the named return value optimization on generic trees.  The file name
               is made by appending .nrv to the source file name.


           vect
               Dump each function after applying vectorization of loops.  The file name is made by appending .vect to
               the source file name.


           vrp Dump each function after Value Range Propagation (VRP).  The file name is made by appending .vrp to the
               source file name.


           all Enable all the available tree dumps with the flags provided in this option.


       -ftree-vectorizer-verbose=n
           This option controls the amount of debugging output the vectorizer prints.  This information is written to
           standard error, unless -fdump-tree-all or -fdump-tree-vect is specified, in which case it is output to the
           usual dump listing file, .vect.


       -frandom-seed=string
           This option provides a seed that GCC uses when it would otherwise use random numbers.  It is used to gener-
           ate certain symbol names that have to be different in every compiled file.  It is also used to place unique
           stamps in coverage data files and the object files that produce them.  You can use the -frandom-seed option
           to produce reproducibly identical object files.


           The string should be different for every file you compile.


       -fsched-verbose=n
           On targets that use instruction scheduling, this option controls the amount of debugging output the sched-
           uler prints.  This information is written to standard error, unless -dS or -dR is specified, in which case
           it is output to the usual dump listing file, .sched or .sched2 respectively.  However for n greater than
           nine, the output is always printed to standard error.


           For n greater than zero, -fsched-verbose outputs the same information as -dRS.  For n greater than one, it
           also output basic block probabilities, detailed ready list information and unit/insn info.  For n greater
           than two, it includes RTL at abort point, control-flow and regions info.  And for n over four, -fsched-ver-
           bose also includes dependence info.


       -save-temps
           Store the usual "temporary" intermediate files permanently; place them in the current directory and name
           them based on the source file.  Thus, compiling foo.c with -c -save-temps would produce files foo.i and
           foo.s, as well as foo.o.  This creates a preprocessed foo.i output file even though the compiler now nor-
           mally uses an integrated preprocessor.


           When used in combination with the -x command line option, -save-temps is sensible enough to avoid over
           writing an input source file with the same extension as an intermediate file.  The corresponding intermedi-
           ate file may be obtained by renaming the source file before using -save-temps.


       -time
           Report the CPU time taken by each subprocess in the compilation sequence.  For C source files, this is the
           compiler proper and assembler (plus the linker if linking is done).  The output looks like this:


                   # cc1 0.12 0.01
                   # as 0.00 0.01


           The first number on each line is the "user time", that is time spent executing the program itself.  The
           second number is "system time", time spent executing operating system routines on behalf of the program.
           Both numbers are in seconds.


       -fvar-tracking
           Run variable tracking pass.  It computes where variables are stored at each position in code.  Better
           debugging information is then generated (if the debugging information format supports this information).


           It is enabled by default when compiling with optimization (-Os, -O, -O2, ...), debugging information (-g)
           and the debug info format supports it.


       -print-file-name=library
           Print the full absolute name of the library file library that would be used when linking---and don’t do
           anything else.  With this option, GCC does not compile or link anything; it just prints the file name.


       -print-multi-directory
           Print the directory name corresponding to the multilib selected by any other switches present in the com-
           mand line.  This directory is supposed to exist in GCC_EXEC_PREFIX.


       -print-multi-lib
           Print the mapping from multilib directory names to compiler switches that enable them.  The directory name
           is separated from the switches by ;, and each switch starts with an @} instead of the @samp{-, without
           spaces between multiple switches.  This is supposed to ease shell-processing.


       -print-multi-os-directory
           Print the path to OS libraries for the selected multilib, relative to some lib subdirectory.  If OS
           libraries are present in the lib subdirectory and no multilibs are used, this is usually just ., if OS
           libraries are present in libsuffix sibling directories this prints e.g. ../lib64, ../lib or ../lib32, or if
           OS libraries are present in lib/subdir subdirectories it prints e.g. amd64, sparcv9 or ev6.


       -print-prog-name=program
           Like -print-file-name, but searches for a program such as cpp.


       -print-libgcc-file-name
           Same as -print-file-name=libgcc.a.


           This is useful when you use -nostdlib or -nodefaultlibs but you do want to link with libgcc.a.  You can do


                   gcc -nostdlib <files>... ‘gcc -print-libgcc-file-name‘


       -print-search-dirs
           Print the name of the configured installation directory and a list of program and library directories gcc
           will search---and don’t do anything else.


           This is useful when gcc prints the error message installation problem, cannot exec cpp0: No such file or
           directory.  To resolve this you either need to put cpp0 and the other compiler components where gcc expects
           to find them, or you can set the environment variable GCC_EXEC_PREFIX to the directory where you installed
           them.  Don’t forget the trailing /.


       -dumpmachine
           Print the compiler’s target machine (for example, i686-pc-linux-gnu)---and don’t do anything else.


       -dumpversion
           Print the compiler version (for example, 3.0)---and don’t do anything else.


       -dumpspecs
           Print the compiler’s built-in specs---and don’t do anything else.  (This is used when GCC itself is being
           built.)


       -feliminate-unused-debug-types
           Normally, when producing DWARF2 output, GCC will emit debugging information for all types declared in a
           compilation unit, regardless of whether or not they are actually used in that compilation unit.  Sometimes
           this is useful, such as if, in the debugger, you want to cast a value to a type that is not actually used
           in your program (but is declared).  More often, however, this results in a significant amount of wasted
           space.  With this option, GCC will avoid producing debug symbol output for types that are nowhere used in
           the source file being compiled.


       Options That Control Optimization


       These options control various sorts of optimizations.


       Without any optimization option, the compiler’s goal is to reduce the cost of compilation and to make debugging
       produce the expected results.  Statements are independent: if you stop the program with a breakpoint between
       statements, you can then assign a new value to any variable or change the program counter to any other state-
       ment in the function and get exactly the results you would expect from the source code.


       Turning on optimization flags makes the compiler attempt to improve the performance and/or code size at the
       expense of compilation time and possibly the ability to debug the program.


       The compiler performs optimization based on the knowledge it has of the program.  Optimization levels -O and
       above, in particular, enable unit-at-a-time mode, which allows the compiler to consider information gained from
       later functions in the file when compiling a function.  Compiling multiple files at once to a single output
       file in unit-at-a-time mode allows the compiler to use information gained from all of the files when compiling
       each of them.


       Not all optimizations are controlled directly by a flag.  Only optimizations that have a flag are listed.


       -O
       -O1 Optimize.  Optimizing compilation takes somewhat more time, and a lot more memory for a large function.


           With -O, the compiler tries to reduce code size and execution time, without performing any optimizations
           that take a great deal of compilation time.


           -O turns on the following optimization flags: -fdefer-pop -fdelayed-branch -fguess-branch-probability
           -fcprop-registers -floop-optimize -fif-conversion -fif-conversion2 -ftree-ccp -ftree-dce -ftree-domina-
           tor-opts -ftree-dse -ftree-ter -ftree-lrs -ftree-sra -ftree-copyrename -ftree-fre -ftree-ch
           -funit-at-a-time -fmerge-constants


           -O also turns on -fomit-frame-pointer on machines where doing so does not interfere with debugging.


           -O doesn’t turn on -ftree-sra for the Ada compiler.  This option must be explicitly specified on the com-
           mand line to be enabled for the Ada compiler.


       -O2 Optimize even more.  GCC performs nearly all supported optimizations that do not involve a space-speed
           tradeoff.  The compiler does not perform loop unrolling or function inlining when you specify -O2.  As com-
           pared to -O, this option increases both compilation time and the performance of the generated code.


           -O2 turns on all optimization flags specified by -O.  It also turns on the following optimization flags:
           -fthread-jumps -fcrossjumping -foptimize-sibling-calls -fcse-follow-jumps  -fcse-skip-blocks -fgcse
           -fgcse-lm -fexpensive-optimizations -fstrength-reduce -frerun-cse-after-loop  -frerun-loop-opt
           -fcaller-saves -fpeephole2 -fschedule-insns  -fschedule-insns2 -fsched-interblock  -fsched-spec -fregmove
           -fstrict-aliasing -fdelete-null-pointer-checks -freorder-blocks  -freorder-functions -falign-functions
           -falign-jumps -falign-loops  -falign-labels -ftree-vrp -ftree-pre


           Please note the warning under -fgcse about invoking -O2 on programs that use computed gotos.


       -O3 Optimize yet more.  -O3 turns on all optimizations specified by -O2 and also turns on the -finline-func-
           tions, -funswitch-loops and -fgcse-after-reload options.


       -O0 Do not optimize.  This is the default.


       -Os Optimize for size.  -Os enables all -O2 optimizations that do not typically increase code size.  It also
           performs further optimizations designed to reduce code size.


           -Os disables the following optimization flags: -falign-functions  -falign-jumps  -falign-loops
           -falign-labels  -freorder-blocks  -freorder-blocks-and-partition -fprefetch-loop-arrays
           -ftree-vect-loop-version


           If you use multiple -O options, with or without level numbers, the last such option is the one that is
           effective.


       Options of the form -fflag specify machine-independent flags.  Most flags have both positive and negative
       forms; the negative form of -ffoo would be -fno-foo.  In the table below, only one of the forms is listed---the
       one you typically will use.  You can figure out the other form by either removing no- or adding it.


       The following options control specific optimizations.  They are either activated by -O options or are related
       to ones that are.  You can use the following flags in the rare cases when "fine-tuning" of optimizations to be
       performed is desired.


       -fno-default-inline
           Do not make member functions inline by default merely because they are defined inside the class scope (C++
           only).  Otherwise, when you specify -O, member functions defined inside class scope are compiled inline by
           default; i.e., you don’t need to add inline in front of the member function name.


       -fno-defer-pop
           Always pop the arguments to each function call as soon as that function returns.  For machines which must
           pop arguments after a function call, the compiler normally lets arguments accumulate on the stack for sev-
           eral function calls and pops them all at once.


           Disabled at levels -O, -O2, -O3, -Os.


       -fforce-mem
           Force memory operands to be copied into registers before doing arithmetic on them.  This produces better
           code by making all memory references potential common subexpressions.  When they are not common subexpres-
           sions, instruction combination should eliminate the separate register-load. This option is now a nop and
           will be removed in 4.2.


       -fforce-addr
           Force memory address constants to be copied into registers before doing arithmetic on them.


       -fomit-frame-pointer
           Don’t keep the frame pointer in a register for functions that don’t need one.  This avoids the instructions
           to save, set up and restore frame pointers; it also makes an extra register available in many functions.
           It also makes debugging impossible on some machines.


           On some machines, such as the VAX, this flag has no effect, because the standard calling sequence automati-
           cally handles the frame pointer and nothing is saved by pretending it doesn’t exist.  The machine-descrip-
           tion macro "FRAME_POINTER_REQUIRED" controls whether a target machine supports this flag.


           Enabled at levels -O, -O2, -O3, -Os.


       -foptimize-sibling-calls
           Optimize sibling and tail recursive calls.


           Enabled at levels -O2, -O3, -Os.


       -fno-inline
           Don’t pay attention to the "inline" keyword.  Normally this option is used to keep the compiler from
           expanding any functions inline.  Note that if you are not optimizing, no functions can be expanded inline.


       -finline-functions
           Integrate all simple functions into their callers.  The compiler heuristically decides which functions are
           simple enough to be worth integrating in this way.


           If all calls to a given function are integrated, and the function is declared "static", then the function
           is normally not output as assembler code in its own right.


           Enabled at level -O3.


       -finline-functions-called-once
           Consider all "static" functions called once for inlining into their caller even if they are not marked
           "inline".  If a call to a given function is integrated, then the function is not output as assembler code
           in its own right.


           Enabled if -funit-at-a-time is enabled.


       -fearly-inlining
           Inline functions marked by "always_inline" and functions whose body seems smaller than the function call
           overhead early before doing -fprofile-generate instrumentation and real inlining pass.  Doing so makes pro-
           filing significantly cheaper and usually inlining faster on programs having large chains of nested wrapper
           functions.


           Enabled by default.


       -finline-limit=n
           By default, GCC limits the size of functions that can be inlined.  This flag allows the control of this
           limit for functions that are explicitly marked as inline (i.e., marked with the inline keyword or defined
           within the class definition in c++).  n is the size of functions that can be inlined in number of pseudo
           instructions (not counting parameter handling).  The default value of n is 600.  Increasing this value can
           result in more inlined code at the cost of compilation time and memory consumption.  Decreasing usually
           makes the compilation faster and less code will be inlined (which presumably means slower programs).  This
           option is particularly useful for programs that use inlining heavily such as those based on recursive tem-
           plates with C++.


           Inlining is actually controlled by a number of parameters, which may be specified individually by using
           --param name=value.  The -finline-limit=n option sets some of these parameters as follows:


           max-inline-insns-single
                is set to I<n>/2.


           max-inline-insns-auto
                is set to I<n>/2.


           min-inline-insns
                is set to 130 or I<n>/4, whichever is smaller.


           max-inline-insns-rtl
                is set to I<n>.


           See below for a documentation of the individual parameters controlling inlining.


           Note: pseudo instruction represents, in this particular context, an abstract measurement of function’s
           size.  In no way does it represent a count of assembly instructions and as such its exact meaning might
           change from one release to an another.


       -fkeep-inline-functions
           In C, emit "static" functions that are declared "inline" into the object file, even if the function has
           been inlined into all of its callers.  This switch does not affect functions using the "extern inline"
           extension in GNU C.  In C++, emit any and all inline functions into the object file.


       -fkeep-static-consts
           Emit variables declared "static const" when optimization isn’t turned on, even if the variables aren’t ref-
           erenced.


           GCC enables this option by default.  If you want to force the compiler to check if the variable was refer-
           enced, regardless of whether or not optimization is turned on, use the -fno-keep-static-consts option.


       -fmerge-constants
           Attempt to merge identical constants (string constants and floating point constants) across compilation
           units.


           This option is the default for optimized compilation if the assembler and linker support it.  Use
           -fno-merge-constants to inhibit this behavior.


           Enabled at levels -O, -O2, -O3, -Os.


       -fmerge-all-constants
           Attempt to merge identical constants and identical variables.


           This option implies -fmerge-constants.  In addition to -fmerge-constants this considers e.g. even constant
           initialized arrays or initialized constant variables with integral or floating point types.  Languages like
           C or C++ require each non-automatic variable to have distinct location, so using this option will result in
           non-conforming behavior.


       -fmodulo-sched
           Perform swing modulo scheduling immediately before the first scheduling pass.  This pass looks at innermost
           loops and reorders their instructions by overlapping different iterations.


       -fno-branch-count-reg
           Do not use "decrement and branch" instructions on a count register, but instead generate a sequence of
           instructions that decrement a register, compare it against zero, then branch based upon the result.  This
           option is only meaningful on architectures that support such instructions, which include x86, PowerPC,
           IA-64 and S/390.


           The default is -fbranch-count-reg, enabled when -fstrength-reduce is enabled.


       -fno-function-cse
           Do not put function addresses in registers; make each instruction that calls a constant function contain
           the function’s address explicitly.


           This option results in less efficient code, but some strange hacks that alter the assembler output may be
           confused by the optimizations performed when this option is not used.


           The default is -ffunction-cse


       -fno-zero-initialized-in-bss
           If the target supports a BSS section, GCC by default puts variables that are initialized to zero into BSS.
           This can save space in the resulting code.


           This option turns off this behavior because some programs explicitly rely on variables going to the data
           section.  E.g., so that the resulting executable can find the beginning of that section and/or make assump-
           tions based on that.


           The default is -fzero-initialized-in-bss.


       -fmudflap -fmudflapth -fmudflapir
           For front-ends that support it (C and C++), instrument all risky pointer/array dereferencing operations,
           some standard library string/heap functions, and some other associated constructs with range/validity
           tests.  Modules so instrumented should be immune to buffer overflows, invalid heap use, and some other
           classes of C/C++ programming errors.  The instrumentation relies on a separate runtime library (libmud-
           flap), which will be linked into a program if -fmudflap is given at link time.  Run-time behavior of the
           instrumented program is controlled by the MUDFLAP_OPTIONS environment variable.  See "env MUD-
           FLAP_OPTIONS=-help a.out" for its options.


           Use -fmudflapth instead of -fmudflap to compile and to link if your program is multi-threaded.  Use -fmud-
           flapir, in addition to -fmudflap or -fmudflapth, if instrumentation should ignore pointer reads.  This pro-
           duces less instrumentation (and therefore faster execution) and still provides some protection against out-
           right memory corrupting writes, but allows erroneously read data to propagate within a program.


       -fstrength-reduce
           Perform the optimizations of loop strength reduction and elimination of iteration variables.


           Enabled at levels -O2, -O3, -Os.


       -fthread-jumps
           Perform optimizations where we check to see if a jump branches to a location where another comparison sub-
           sumed by the first is found.  If so, the first branch is redirected to either the destination of the second
           branch or a point immediately following it, depending on whether the condition is known to be true or
           false.


           Enabled at levels -O2, -O3, -Os.


       -fcse-follow-jumps
           In common subexpression elimination, scan through jump instructions when the target of the jump is not
           reached by any other path.  For example, when CSE encounters an "if" statement with an "else" clause, CSE
           will follow the jump when the condition tested is false.


           Enabled at levels -O2, -O3, -Os.


       -fcse-skip-blocks
           This is similar to -fcse-follow-jumps, but causes CSE to follow jumps which conditionally skip over blocks.
           When CSE encounters a simple "if" statement with no else clause, -fcse-skip-blocks causes CSE to follow the
           jump around the body of the "if".


           Enabled at levels -O2, -O3, -Os.


       -frerun-cse-after-loop
           Re-run common subexpression elimination after loop optimizations has been performed.


           Enabled at levels -O2, -O3, -Os.


       -frerun-loop-opt
           Run the loop optimizer twice.


           Enabled at levels -O2, -O3, -Os.


       -fgcse
           Perform a global common subexpression elimination pass.  This pass also performs global constant and copy
           propagation.


           Note: When compiling a program using computed gotos, a GCC extension, you may get better runtime perfor-
           mance if you disable the global common subexpression elimination pass by adding -fno-gcse to the command
           line.


           Enabled at levels -O2, -O3, -Os.


       -fgcse-lm
           When -fgcse-lm is enabled, global common subexpression elimination will attempt to move loads which are
           only killed by stores into themselves.  This allows a loop containing a load/store sequence to be changed
           to a load outside the loop, and a copy/store within the loop.


           Enabled by default when gcse is enabled.


       -fgcse-sm
           When -fgcse-sm is enabled, a store motion pass is run after global common subexpression elimination.  This
           pass will attempt to move stores out of loops.  When used in conjunction with -fgcse-lm, loops containing a
           load/store sequence can be changed to a load before the loop and a store after the loop.


           Not enabled at any optimization level.


       -fgcse-las
           When -fgcse-las is enabled, the global common subexpression elimination pass eliminates redundant loads
           that come after stores to the same memory location (both partial and full redundancies).


           Not enabled at any optimization level.


       -fgcse-after-reload
           When -fgcse-after-reload is enabled, a redundant load elimination pass is performed after reload.  The pur-
           pose of this pass is to cleanup redundant spilling.


       -floop-optimize
           Perform loop optimizations: move constant expressions out of loops, simplify exit test conditions and
           optionally do strength-reduction as well.


           Enabled at levels -O, -O2, -O3, -Os.


       -floop-optimize2
           Perform loop optimizations using the new loop optimizer.  The optimizations (loop unrolling, peeling and
           unswitching, loop invariant motion) are enabled by separate flags.


       -funsafe-loop-optimizations
           If given, the loop optimizer will assume that loop indices do not overflow, and that the loops with non-
           trivial exit condition are not infinite.  This enables a wider range of loop optimizations even if the loop
           optimizer itself cannot prove that these assumptions are valid.  Using -Wunsafe-loop-optimizations, the
           compiler will warn you if it finds this kind of loop.


       -fcrossjumping
           Perform cross-jumping transformation.  This transformation unifies equivalent code and save code size.  The
           resulting code may or may not perform better than without cross-jumping.


           Enabled at levels -O2, -O3, -Os.


       -fif-conversion
           Attempt to transform conditional jumps into branch-less equivalents.  This include use of conditional
           moves, min, max, set flags and abs instructions, and some tricks doable by standard arithmetics.  The use
           of conditional execution on chips where it is available is controlled by "if-conversion2".


           Enabled at levels -O, -O2, -O3, -Os.


       -fif-conversion2
           Use conditional execution (where available) to transform conditional jumps into branch-less equivalents.


           Enabled at levels -O, -O2, -O3, -Os.


       -fdelete-null-pointer-checks
           Use global dataflow analysis to identify and eliminate useless checks for null pointers.  The compiler
           assumes that dereferencing a null pointer would have halted the program.  If a pointer is checked after it
           has already been dereferenced, it cannot be null.


           In some environments, this assumption is not true, and programs can safely dereference null pointers.  Use
           -fno-delete-null-pointer-checks to disable this optimization for programs which depend on that behavior.


           Enabled at levels -O2, -O3, -Os.


       -fexpensive-optimizations
           Perform a number of minor optimizations that are relatively expensive.


           Enabled at levels -O2, -O3, -Os.


       -foptimize-register-move
       -fregmove
           Attempt to reassign register numbers in move instructions and as operands of other simple instructions in
           order to maximize the amount of register tying.  This is especially helpful on machines with two-operand
           instructions.


           Note -fregmove and -foptimize-register-move are the same optimization.


           Enabled at levels -O2, -O3, -Os.


       -fdelayed-branch
           If supported for the target machine, attempt to reorder instructions to exploit instruction slots available
           after delayed branch instructions.


           Enabled at levels -O, -O2, -O3, -Os.


       -fschedule-insns
           If supported for the target machine, attempt to reorder instructions to eliminate execution stalls due to
           required data being unavailable.  This helps machines that have slow floating point or memory load instruc-
           tions by allowing other instructions to be issued until the result of the load or floating point instruc-
           tion is required.


           Enabled at levels -O2, -O3, -Os.


       -fschedule-insns2
           Similar to -fschedule-insns, but requests an additional pass of instruction scheduling after register allo-
           cation has been done.  This is especially useful on machines with a relatively small number of registers
           and where memory load instructions take more than one cycle.


           Enabled at levels -O2, -O3, -Os.


       -fno-sched-interblock
           Don’t schedule instructions across basic blocks.  This is normally enabled by default when scheduling
           before register allocation, i.e.  with -fschedule-insns or at -O2 or higher.


       -fno-sched-spec
           Don’t allow speculative motion of non-load instructions.  This is normally enabled by default when schedul-
           ing before register allocation, i.e.  with -fschedule-insns or at -O2 or higher.


       -fsched-spec-load
           Allow speculative motion of some load instructions.  This only makes sense when scheduling before register
           allocation, i.e. with -fschedule-insns or at -O2 or higher.


       -fsched-spec-load-dangerous
           Allow speculative motion of more load instructions.  This only makes sense when scheduling before register
           allocation, i.e. with -fschedule-insns or at -O2 or higher.


       -fsched-stalled-insns
       -fsched-stalled-insns=n
           Define how many insns (if any) can be moved prematurely from the queue of stalled insns into the ready
           list, during the second scheduling pass.  -fno-fsched-stalled-insns and -fsched-stalled-insns=0 are equiva-
           lent and mean that no insns will be moved prematurely.  If n is unspecified then there is no limit on how
           many queued insns can be moved prematurely.


       -fsched-stalled-insns-dep
       -fsched-stalled-insns-dep=n
           Define how many insn groups (cycles) will be examined for a dependency on a stalled insn that is candidate
           for premature removal from the queue of stalled insns.  This has an effect only during the second schedul-
           ing pass, and only if -fsched-stalled-insns is used and its value is not zero.
           +-fno-sched-stalled-insns-dep is equivalent to +-fsched-stalled-insns-dep=0.  +-fsched-stalled-insns-dep
           without a value is equivalent to +-fsched-stalled-insns-dep=1.


       -fsched2-use-superblocks
           When scheduling after register allocation, do use superblock scheduling algorithm.  Superblock scheduling
           allows motion across basic block boundaries resulting on faster schedules.  This option is experimental, as
           not all machine descriptions used by GCC model the CPU closely enough to avoid unreliable results from the
           algorithm.


           This only makes sense when scheduling after register allocation, i.e. with -fschedule-insns2 or at -O2 or
           higher.


       -fsched2-use-traces
           Use -fsched2-use-superblocks algorithm when scheduling after register allocation and additionally perform
           code duplication in order to increase the size of superblocks using tracer pass.  See -ftracer for details
           on trace formation.


           This mode should produce faster but significantly longer programs.  Also without -fbranch-probabilities the
           traces constructed may not match the reality and hurt the performance.  This only makes sense when schedul-
           ing after register allocation, i.e. with -fschedule-insns2 or at -O2 or higher.


       -freschedule-modulo-scheduled-loops
           The modulo scheduling comes before the traditional scheduling, if a loop was modulo scheduled we may want
           to prevent the later scheduling passes from changing its schedule, we use this option to control that.


       -fcaller-saves
           Enable values to be allocated in registers that will be clobbered by function calls, by emitting extra
           instructions to save and restore the registers around such calls.  Such allocation is done only when it
           seems to result in better code than would otherwise be produced.


           This option is always enabled by default on certain machines, usually those which have no call-preserved
           registers to use instead.


           Enabled at levels -O2, -O3, -Os.


       -ftree-pre
           Perform Partial Redundancy Elimination (PRE) on trees.  This flag is enabled by default at -O2 and -O3.


       -ftree-fre
           Perform Full Redundancy Elimination (FRE) on trees.  The difference between FRE and PRE is that FRE only
           considers expressions that are computed on all paths leading to the redundant computation.  This analysis
           faster than PRE, though it exposes fewer redundancies.  This flag is enabled by default at -O and higher.


       -ftree-copy-prop
           Perform copy propagation on trees.  This pass eliminates unnecessary copy operations.  This flag is enabled
           by default at -O and higher.


       -ftree-store-copy-prop
           Perform copy propagation of memory loads and stores.  This pass eliminates unnecessary copy operations in
           memory references (structures, global variables, arrays, etc).  This flag is enabled by default at -O2 and
           higher.


       -ftree-salias
           Perform structural alias analysis on trees.  This flag is enabled by default at -O and higher.


       -ftree-sink
           Perform forward store motion  on trees.  This flag is enabled by default at -O and higher.


       -ftree-ccp
           Perform sparse conditional constant propagation (CCP) on trees.  This pass only operates on local scalar
           variables and is enabled by default at -O and higher.


       -ftree-store-ccp
           Perform sparse conditional constant propagation (CCP) on trees.  This pass operates on both local scalar
           variables and memory stores and loads (global variables, structures, arrays, etc).  This flag is enabled by
           default at -O2 and higher.


       -ftree-dce
           Perform dead code elimination (DCE) on trees.  This flag is enabled by default at -O and higher.


       -ftree-dominator-opts
           Perform a variety of simple scalar cleanups (constant/copy propagation, redundancy elimination, range prop-
           agation and expression simplification) based on a dominator tree traversal.  This also performs jump
           threading (to reduce jumps to jumps). This flag is enabled by default at -O and higher.


       -ftree-ch
           Perform loop header copying on trees.  This is beneficial since it increases effectiveness of code motion
           optimizations.  It also saves one jump.  This flag is enabled by default at -O and higher.  It is not
           enabled for -Os, since it usually increases code size.


       -ftree-loop-optimize
           Perform loop optimizations on trees.  This flag is enabled by default at -O and higher.


       -ftree-loop-linear
           Perform linear loop transformations on tree.  This flag can improve cache performance and allow further
           loop optimizations to take place.


       -ftree-loop-im
           Perform loop invariant motion on trees.  This pass moves only invariants that would be hard to handle at
           RTL level (function calls, operations that expand to nontrivial sequences of insns).  With -funswitch-loops
           it also moves operands of conditions that are invariant out of the loop, so that we can use just trivial
           invariantness analysis in loop unswitching.  The pass also includes store motion.


       -ftree-loop-ivcanon
           Create a canonical counter for number of iterations in the loop for that determining number of iterations
           requires complicated analysis.  Later optimizations then may determine the number easily.  Useful espe-
           cially in connection with unrolling.


       -fivopts
           Perform induction variable optimizations (strength reduction, induction variable merging and induction
           variable elimination) on trees.


       -ftree-sra
           Perform scalar replacement of aggregates.  This pass replaces structure references with scalars to prevent
           committing structures to memory too early.  This flag is enabled by default at -O and higher.


       -ftree-copyrename
           Perform copy renaming on trees.  This pass attempts to rename compiler temporaries to other variables at
           copy locations, usually resulting in variable names which more closely resemble the original variables.
           This flag is enabled by default at -O and higher.


       -ftree-ter
           Perform temporary expression replacement during the SSA->normal phase.  Single use/single def temporaries
           are replaced at their use location with their defining expression.  This results in non-GIMPLE code, but
           gives the expanders much more complex trees to work on resulting in better RTL generation.  This is enabled
           by default at -O and higher.


       -ftree-lrs
           Perform live range splitting during the SSA->normal phase.  Distinct live ranges of a variable are split
           into unique variables, allowing for better optimization later.  This is enabled by default at -O and
           higher.


       -ftree-vectorize
           Perform loop vectorization on trees.


       -ftree-vect-loop-version
           Perform loop versioning when doing loop vectorization on trees.  When a loop appears to be vectorizable
           except that data alignment or data dependence cannot be determined at compile time then vectorized and non-
           vectorized versions of the loop are generated along with runtime checks for alignment or dependence to con-
           trol which version is executed.  This option is enabled by default except at level -Os where it is dis-
           abled.


       -ftree-vrp
           Perform Value Range Propagation on trees.  This is similar to the constant propagation pass, but instead of
           values, ranges of values are propagated.  This allows the optimizers to remove unnecessary range checks
           like array bound checks and null pointer checks.  This is enabled by default at -O2 and higher.  Null
           pointer check elimination is only done if -fdelete-null-pointer-checks is enabled.


       -ftracer
           Perform tail duplication to enlarge superblock size.  This transformation simplifies the control flow of
           the function allowing other optimizations to do better job.


       -funroll-loops
           Unroll loops whose number of iterations can be determined at compile time or upon entry to the loop.  -fun-
           roll-loops implies both -fstrength-reduce and -frerun-cse-after-loop.  This option makes code larger, and
           may or may not make it run faster.


       -funroll-all-loops
           Unroll all loops, even if their number of iterations is uncertain when the loop is entered.  This usually
           makes programs run more slowly.  -funroll-all-loops implies the same options as -funroll-loops,


       -fsplit-ivs-in-unroller
           Enables expressing of values of induction variables in later iterations of the unrolled loop using the
           value in the first iteration.  This breaks long dependency chains, thus improving efficiency of the
           scheduling passes.


           Combination of -fweb and CSE is often sufficient to obtain the same effect.  However in cases the loop body
           is more complicated than a single basic block, this is not reliable.  It also does not work at all on some
           of the architectures due to restrictions in the CSE pass.


           This optimization is enabled by default.


       -fvariable-expansion-in-unroller
           With this option, the compiler will create multiple copies of some local variables when unrolling a loop
           which can result in superior code.


       -fprefetch-loop-arrays
           If supported by the target machine, generate instructions to prefetch memory to improve the performance of
           loops that access large arrays.


           These options may generate better or worse code; results are highly dependent on the structure of loops
           within the source code.


       -fno-peephole
       -fno-peephole2
           Disable any machine-specific peephole optimizations.  The difference between -fno-peephole and -fno-peep-
           hole2 is in how they are implemented in the compiler; some targets use one, some use the other, a few use
           both.


           -fpeephole is enabled by default.  -fpeephole2 enabled at levels -O2, -O3, -Os.


       -fno-guess-branch-probability
           Do not guess branch probabilities using heuristics.


           GCC will use heuristics to guess branch probabilities if they are not provided by profiling feedback
           (-fprofile-arcs).  These heuristics are based on the control flow graph.  If some branch probabilities are
           specified by __builtin_expect, then the heuristics will be used to guess branch probabilities for the rest
           of the control flow graph, taking the __builtin_expect info into account.  The interactions between the
           heuristics and __builtin_expect can be complex, and in some cases, it may be useful to disable the heuris-
           tics so that the effects of __builtin_expect are easier to understand.


           The default is -fguess-branch-probability at levels -O, -O2, -O3, -Os.


       -freorder-blocks
           Reorder basic blocks in the compiled function in order to reduce number of taken branches and improve code
           locality.


           Enabled at levels -O2, -O3.


       -freorder-blocks-and-partition
           In addition to reordering basic blocks in the compiled function, in order to reduce number of taken
           branches, partitions hot and cold basic blocks into separate sections of the assembly and .o files, to
           improve paging and cache locality performance.


           This optimization is automatically turned off in the presence of exception handling, for linkonce sections,
           for functions with a user-defined section attribute and on any architecture that does not support named
           sections.


       -freorder-functions
           Reorder functions in the object file in order to improve code locality.  This is implemented by using spe-
           cial subsections ".text.hot" for most frequently executed functions and ".text.unlikely" for unlikely exe-
           cuted functions.  Reordering is done by the linker so object file format must support named sections and
           linker must place them in a reasonable way.


           Also profile feedback must be available in to make this option effective.  See -fprofile-arcs for details.


           Enabled at levels -O2, -O3, -Os.


       -fstrict-aliasing
           Allows the compiler to assume the strictest aliasing rules applicable to the language being compiled.  For
           C (and C++), this activates optimizations based on the type of expressions.  In particular, an object of
           one type is assumed never to reside at the same address as an object of a different type, unless the types
           are almost the same.  For example, an "unsigned int" can alias an "int", but not a "void*" or a "double".
           A character type may alias any other type.


           Pay special attention to code like this:


                   union a_union {
                     int i;
                     double d;
                   };


                   int f() {
                     a_union t;
                     t.d = 3.0;
                     return t.i;
                   }


           The practice of reading from a different union member than the one most recently written to (called
           "type-punning") is common.  Even with -fstrict-aliasing, type-punning is allowed, provided the memory is
           accessed through the union type.  So, the code above will work as expected.  However, this code might not:


                   int f() {
                     a_union t;
                     int* ip;
                     t.d = 3.0;
                     ip = &t.i;
                     return *ip;
                   }


           Every language that wishes to perform language-specific alias analysis should define a function that com-
           putes, given an "tree" node, an alias set for the node.  Nodes in different alias sets are not allowed to
           alias.  For an example, see the C front-end function "c_get_alias_set".


           Enabled at levels -O2, -O3, -Os.


       -falign-functions
       -falign-functions=n
           Align the start of functions to the next power-of-two greater than n, skipping up to n bytes.  For
           instance, -falign-functions=32 aligns functions to the next 32-byte boundary, but -falign-functions=24
           would align to the next 32-byte boundary only if this can be done by skipping 23 bytes or less.


           -fno-align-functions and -falign-functions=1 are equivalent and mean that functions will not be aligned.


           Some assemblers only support this flag when n is a power of two; in that case, it is rounded up.


           If n is not specified or is zero, use a machine-dependent default.


           Enabled at levels -O2, -O3.


       -falign-labels
       -falign-labels=n
           Align all branch targets to a power-of-two boundary, skipping up to n bytes like -falign-functions.  This
           option can easily make code slower, because it must insert dummy operations for when the branch target is
           reached in the usual flow of the code.


           -fno-align-labels and -falign-labels=1 are equivalent and mean that labels will not be aligned.


           If -falign-loops or -falign-jumps are applicable and are greater than this value, then their values are
           used instead.


           If n is not specified or is zero, use a machine-dependent default which is very likely to be 1, meaning no
           alignment.


           Enabled at levels -O2, -O3.


       -falign-loops
       -falign-loops=n
           Align loops to a power-of-two boundary, skipping up to n bytes like -falign-functions.  The hope is that
           the loop will be executed many times, which will make up for any execution of the dummy operations.


           -fno-align-loops and -falign-loops=1 are equivalent and mean that loops will not be aligned.


           If n is not specified or is zero, use a machine-dependent default.


           Enabled at levels -O2, -O3.


       -falign-jumps
       -falign-jumps=n
           Align branch targets to a power-of-two boundary, for branch targets where the targets can only be reached
           by jumping, skipping up to n bytes like -falign-functions.  In this case, no dummy operations need be exe-
           cuted.


           -fno-align-jumps and -falign-jumps=1 are equivalent and mean that loops will not be aligned.


           If n is not specified or is zero, use a machine-dependent default.


           Enabled at levels -O2, -O3.


       -funit-at-a-time
           Parse the whole compilation unit before starting to produce code.  This allows some extra optimizations to
           take place but consumes more memory (in general).  There are some compatibility issues with unit-at-at-time
           mode:


           *   enabling unit-at-a-time mode may change the order in which functions, variables, and top-level "asm"
               statements are emitted, and will likely break code relying on some particular ordering.  The majority
               of such top-level "asm" statements, though, can be replaced by "section" attributes.


           *   unit-at-a-time mode removes unreferenced static variables and functions.  This may result in undefined
               references when an "asm" statement refers directly to variables or functions that are otherwise unused.
               In that case either the variable/function shall be listed as an operand of the "asm" statement operand
               or, in the case of top-level "asm" statements the attribute "used" shall be used on the declaration.


           *   Static functions now can use non-standard passing conventions that may break "asm" statements calling
               functions directly.  Again, attribute "used" will prevent this behavior.


           As a temporary workaround, -fno-unit-at-a-time can be used, but this scheme may not be supported by future
           releases of GCC.


           Enabled at levels -O, -O2, -O3, -Os.


       -fweb
           Constructs webs as commonly used for register allocation purposes and assign each web individual pseudo
           register.  This allows the register allocation pass to operate on pseudos directly, but also strengthens
           several other optimization passes, such as CSE, loop optimizer and trivial dead code remover.  It can, how-
           ever, make debugging impossible, since variables will no longer stay in a "home register".


           Enabled by default with -funroll-loops.


       -fwhole-program
           Assume that the current compilation unit represents whole program being compiled.  All public functions and
           variables with the exception of "main" and those merged by attribute "externally_visible" become static
           functions and in a affect gets more aggressively optimized by interprocedural optimizers.  While this
           option is equivalent to proper use of "static" keyword for programs consisting of single file, in combina-
           tion with option --combine this flag can be used to compile most of smaller scale C programs since the
           functions and variables become local for the whole combined compilation unit, not for the single source
           file itself.


       -fcprop-registers
           After register allocation and post-register allocation instruction splitting, we perform a copy-propagation
           pass to try to reduce scheduling dependencies and occasionally eliminate the copy.


           Enabled at levels -O, -O2, -O3, -Os.


       -fprofile-generate
           Enable options usually used for instrumenting application to produce profile useful for later recompilation
           with profile feedback based optimization.  You must use -fprofile-generate both when compiling and when
           linking your program.


           The following options are enabled: "-fprofile-arcs", "-fprofile-values", "-fvpt".


       -fprofile-use
           Enable profile feedback directed optimizations, and optimizations generally profitable only with profile
           feedback available.


           The following options are enabled: "-fbranch-probabilities", "-fvpt", "-funroll-loops", "-fpeel-loops",
           "-ftracer", "-fno-loop-optimize".


       The following options control compiler behavior regarding floating point arithmetic.  These options trade off
       between speed and correctness.  All must be specifically enabled.


       -ffloat-store
           Do not store floating point variables in registers, and inhibit other options that might change whether a
           floating point value is taken from a register or memory.


           This option prevents undesirable excess precision on machines such as the 68000 where the floating regis-
           ters (of the 68881) keep more precision than a "double" is supposed to have.  Similarly for the x86 archi-
           tecture.  For most programs, the excess precision does only good, but a few programs rely on the precise
           definition of IEEE floating point.  Use -ffloat-store for such programs, after modifying them to store all
           pertinent intermediate computations into variables.


       -ffast-math
           Sets -fno-math-errno, -funsafe-math-optimizations, -fno-trapping-math, -ffinite-math-only, -fno-round-
           ing-math, -fno-signaling-nans and fcx-limited-range.


           This option causes the preprocessor macro "__FAST_MATH__" to be defined.


           This option should never be turned on by any -O option since it can result in incorrect output for programs
           which depend on an exact implementation of IEEE or ISO rules/specifications for math functions.


       -fno-math-errno
           Do not set ERRNO after calling math functions that are executed with a single instruction, e.g., sqrt.  A
           program that relies on IEEE exceptions for math error handling may want to use this flag for speed while
           maintaining IEEE arithmetic compatibility.


           This option should never be turned on by any -O option since it can result in incorrect output for programs
           which depend on an exact implementation of IEEE or ISO rules/specifications for math functions.


           The default is -fmath-errno.


           On Darwin systems, the math library never sets "errno".  There is therefore no reason for the compiler to
           consider the possibility that it might, and -fno-math-errno is the default.


       -funsafe-math-optimizations
           Allow optimizations for floating-point arithmetic that (a) assume that arguments and results are valid and
           (b) may violate IEEE or ANSI standards.  When used at link-time, it may include libraries or startup files
           that change the default FPU control word or other similar optimizations.


           This option should never be turned on by any -O option since it can result in incorrect output for programs
           which depend on an exact implementation of IEEE or ISO rules/specifications for math functions.


           The default is -fno-unsafe-math-optimizations.


       -ffinite-math-only
           Allow optimizations for floating-point arithmetic that assume that arguments and results are not NaNs or
           +-Infs.


           This option should never be turned on by any -O option since it can result in incorrect output for programs
           which depend on an exact implementation of IEEE or ISO rules/specifications.


           The default is -fno-finite-math-only.


       -fno-trapping-math
           Compile code assuming that floating-point operations cannot generate user-visible traps.  These traps
           include division by zero, overflow, underflow, inexact result and invalid operation.  This option implies
           -fno-signaling-nans.  Setting this option may allow faster code if one relies on "non-stop" IEEE arith-
           metic, for example.


           This option should never be turned on by any -O option since it can result in incorrect output for programs
           which depend on an exact implementation of IEEE or ISO rules/specifications for math functions.


           The default is -ftrapping-math.


       -frounding-math
           Disable transformations and optimizations that assume default floating point rounding behavior.  This is
           round-to-zero for all floating point to integer conversions, and round-to-nearest for all other arithmetic
           truncations.  This option should be specified for programs that change the FP rounding mode dynamically, or
           that may be executed with a non-default rounding mode.  This option disables constant folding of floating
           point expressions at compile-time (which may be affected by rounding mode) and arithmetic transformations
           that are unsafe in the presence of sign-dependent rounding modes.


           The default is -fno-rounding-math.


           This option is experimental and does not currently guarantee to disable all GCC optimizations that are
           affected by rounding mode.  Future versions of GCC may provide finer control of this setting using C99’s
           "FENV_ACCESS" pragma.  This command line option will be used to specify the default state for
           "FENV_ACCESS".


       -fsignaling-nans
           Compile code assuming that IEEE signaling NaNs may generate user-visible traps during floating-point opera-
           tions.  Setting this option disables optimizations that may change the number of exceptions visible with
           signaling NaNs.  This option implies -ftrapping-math.


           This option causes the preprocessor macro "__SUPPORT_SNAN__" to be defined.


           The default is -fno-signaling-nans.


           This option is experimental and does not currently guarantee to disable all GCC optimizations that affect
           signaling NaN behavior.


       -fsingle-precision-constant
           Treat floating point constant as single precision constant instead of implicitly converting it to double
           precision constant.


       -fcx-limited-range
           When enabled, this option states that a range reduction step is not needed when performing complex divi-
           sion.  The default is -fno-cx-limited-range, but is enabled by -ffast-math.


           This option controls the default setting of the ISO C99 "CX_LIMITED_RANGE" pragma.  Nevertheless, the
           option applies to all languages.


       The following options control optimizations that may improve performance, but are not enabled by any -O
       options.  This section includes experimental options that may produce broken code.


       -fbranch-probabilities
           After running a program compiled with -fprofile-arcs, you can compile it a second time using -fbranch-prob-
           abilities, to improve optimizations based on the number of times each branch was taken.  When the program
           compiled with -fprofile-arcs exits it saves arc execution counts to a file called sourcename.gcda for each
           source file  The information in this data file is very dependent on the structure of the generated code, so
           you must use the same source code and the same optimization options for both compilations.


           With -fbranch-probabilities, GCC puts a REG_BR_PROB note on each JUMP_INSN and CALL_INSN.  These can be
           used to improve optimization.  Currently, they are only used in one place: in reorg.c, instead of guessing
           which path a branch is mostly to take, the REG_BR_PROB values are used to exactly determine which path is
           taken more often.


       -fprofile-values
           If combined with -fprofile-arcs, it adds code so that some data about values of expressions in the program
           is gathered.


           With -fbranch-probabilities, it reads back the data gathered from profiling values of expressions and adds
           REG_VALUE_PROFILE notes to instructions for their later usage in optimizations.


           Enabled with -fprofile-generate and -fprofile-use.


       -fvpt
           If combined with -fprofile-arcs, it instructs the compiler to add a code to gather information about values
           of expressions.


           With -fbranch-probabilities, it reads back the data gathered and actually performs the optimizations based
           on them.  Currently the optimizations include specialization of division operation using the knowledge
           about the value of the denominator.


       -frename-registers
           Attempt to avoid false dependencies in scheduled code by making use of registers left over after register
           allocation.  This optimization will most benefit processors with lots of registers.  Depending on the debug
           information format adopted by the target, however, it can make debugging impossible, since variables will
           no longer stay in a "home register".


           Enabled by default with -funroll-loops.


       -ftracer
           Perform tail duplication to enlarge superblock size.  This transformation simplifies the control flow of
           the function allowing other optimizations to do better job.


           Enabled with -fprofile-use.


       -funroll-loops
           Unroll loops whose number of iterations can be determined at compile time or upon entry to the loop.  -fun-
           roll-loops implies -frerun-cse-after-loop, -fweb and -frename-registers.  It also turns on complete loop
           peeling (i.e. complete removal of loops with small constant number of iterations).  This option makes code
           larger, and may or may not make it run faster.


           Enabled with -fprofile-use.


       -funroll-all-loops
           Unroll all loops, even if their number of iterations is uncertain when the loop is entered.  This usually
           makes programs run more slowly.  -funroll-all-loops implies the same options as -funroll-loops.


       -fpeel-loops
           Peels the loops for that there is enough information that they do not roll much (from profile feedback).
           It also turns on complete loop peeling (i.e. complete removal of loops with small constant number of itera-
           tions).


           Enabled with -fprofile-use.


       -fmove-loop-invariants
           Enables the loop invariant motion pass in the new loop optimizer.  Enabled at level -O1


       -funswitch-loops
           Move branches with loop invariant conditions out of the loop, with duplicates of the loop on both branches
           (modified according to result of the condition).


       -fprefetch-loop-arrays
           If supported by the target machine, generate instructions to prefetch memory to improve the performance of
           loops that access large arrays.


           Disabled at level -Os.


       -ffunction-sections
       -fdata-sections
           Place each function or data item into its own section in the output file if the target supports arbitrary
           sections.  The name of the function or the name of the data item determines the section’s name in the out-
           put file.


           Use these options on systems where the linker can perform optimizations to improve locality of reference in
           the instruction space.  Most systems using the ELF object format and SPARC processors running Solaris 2
           have linkers with such optimizations.  AIX may have these optimizations in the future.


           Only use these options when there are significant benefits from doing so.  When you specify these options,
           the assembler and linker will create larger object and executable files and will also be slower.  You will
           not be able to use "gprof" on all systems if you specify this option and you may have problems with debug-
           ging if you specify both this option and -g.


       -fbranch-target-load-optimize
           Perform branch target register load optimization before prologue / epilogue threading.  The use of target
           registers can typically be exposed only during reload, thus hoisting loads out of loops and doing inter-
           block scheduling needs a separate optimization pass.


       -fbranch-target-load-optimize2
           Perform branch target register load optimization after prologue / epilogue threading.


       -fbtr-bb-exclusive
           When performing branch target register load optimization, don’t reuse branch target registers in within any
           basic block.


       -fstack-protector
           Emit extra code to check for buffer overflows, such as stack smashing attacks.  This is done by adding a
           guard variable to functions with vulnerable objects.  This includes functions that call alloca, and func-
           tions with buffers larger than 8 bytes.  The guards are initialized when a function is entered and then
           checked when the function exits.  If a guard check fails, an error message is printed and the program
           exits.


       -fstack-protector-all
           Like -fstack-protector except that all functions are protected.


       --param name=value
           In some places, GCC uses various constants to control the amount of optimization that is done.  For exam-
           ple, GCC will not inline functions that contain more that a certain number of instructions.  You can con-
           trol some of these constants on the command-line using the --param option.


           The names of specific parameters, and the meaning of the values, are tied to the internals of the compiler,
           and are subject to change without notice in future releases.


           In each case, the value is an integer.  The allowable choices for name are given in the following table:


           salias-max-implicit-fields
               The maximum number of fields in a variable without direct structure accesses for which structure alias-
               ing will consider trying to track each field.  The default is 5


           sra-max-structure-size
               The maximum structure size, in bytes, at which the scalar replacement of aggregates (SRA) optimization
               will perform block copies.  The default value, 0, implies that GCC will select the most appropriate
               size itself.


           sra-field-structure-ratio
               The threshold ratio (as a percentage) between instantiated fields and the complete structure size.  We
               say that if the ratio of the number of bytes in instantiated fields to the number of bytes in the com-
               plete structure exceeds this parameter, then block copies are not used.  The default is 75.


           max-crossjump-edges
               The maximum number of incoming edges to consider for crossjumping.  The algorithm used by -fcrossjump-
               ing is O(N^2) in the number of edges incoming to each block.  Increasing values mean more aggressive
               optimization, making the compile time increase with probably small improvement in executable size.


           min-crossjump-insns
               The minimum number of instructions which must be matched at the end of two blocks before crossjumping
               will be performed on them.  This value is ignored in the case where all instructions in the block being
               crossjumped from are matched.  The default value is 5.


           max-grow-copy-bb-insns
               The maximum code size expansion factor when copying basic blocks instead of jumping.  The expansion is
               relative to a jump instruction.  The default value is 8.


           max-goto-duplication-insns
               The maximum number of instructions to duplicate to a block that jumps to a computed goto.  To avoid
               O(N^2) behavior in a number of passes, GCC factors computed gotos early in the compilation process, and
               unfactors them as late as possible.  Only computed jumps at the end of a basic blocks with no more than
               max-goto-duplication-insns are unfactored.  The default value is 8.


           max-delay-slot-insn-search
               The maximum number of instructions to consider when looking for an instruction to fill a delay slot.
               If more than this arbitrary number of instructions is searched, the time savings from filling the delay
               slot will be minimal so stop searching.  Increasing values mean more aggressive optimization, making
               the compile time increase with probably small improvement in executable run time.


           max-delay-slot-live-search
               When trying to fill delay slots, the maximum number of instructions to consider when searching for a
               block with valid live register information.  Increasing this arbitrarily chosen value means more
               aggressive optimization, increasing the compile time.  This parameter should be removed when the delay
               slot code is rewritten to maintain the control-flow graph.


           max-gcse-memory
               The approximate maximum amount of memory that will be allocated in order to perform the global common
               subexpression elimination optimization.  If more memory than specified is required, the optimization
               will not be done.


           max-gcse-passes
               The maximum number of passes of GCSE to run.  The default is 1.


           max-pending-list-length
               The maximum number of pending dependencies scheduling will allow before flushing the current state and
               starting over.  Large functions with few branches or calls can create excessively large lists which
               needlessly consume memory and resources.


           max-inline-insns-single
               Several parameters control the tree inliner used in gcc.  This number sets the maximum number of
               instructions (counted in GCC’s internal representation) in a single function that the tree inliner will
               consider for inlining.  This only affects functions declared inline and methods implemented in a class
               declaration (C++).  The default value is 450.


           max-inline-insns-auto
               When you use -finline-functions (included in -O3), a lot of functions that would otherwise not be con-
               sidered for inlining by the compiler will be investigated.  To those functions, a different (more
               restrictive) limit compared to functions declared inline can be applied.  The default value is 90.


           large-function-insns
               The limit specifying really large functions.  For functions larger than this limit after inlining
               inlining is constrained by --param large-function-growth.  This parameter is useful primarily to avoid
               extreme compilation time caused by non-linear algorithms used by the backend.  This parameter is
               ignored when -funit-at-a-time is not used.  The default value is 2700.


           large-function-growth
               Specifies maximal growth of large function caused by inlining in percents.  This parameter is ignored
               when -funit-at-a-time is not used.  The default value is 100 which limits large function growth to 2.0
               times the original size.


           large-unit-insns
               The limit specifying large translation unit.  Growth caused by inlining of units larger than this limit
               is limited by --param inline-unit-growth.  For small units this might be too tight (consider unit con-
               sisting of function A that is inline and B that just calls A three time.  If B is small relative to A,
               the growth of unit is 300\% and yet such inlining is very sane.  For very large units consisting of
               small inlininable functions however the overall unit growth limit is needed to avoid exponential explo-
               sion of code size.  Thus for smaller units, the size is increased to --param large-unit-insns before
               aplying --param inline-unit-growth.  The default is 10000


           inline-unit-growth
               Specifies maximal overall growth of the compilation unit caused by inlining.  This parameter is ignored
               when -funit-at-a-time is not used.  The default value is 50 which limits unit growth to 1.5 times the
               original size.


           max-inline-insns-recursive
           max-inline-insns-recursive-auto
               Specifies maximum number of instructions out-of-line copy of self recursive inline function can grow
               into by performing recursive inlining.


               For functions declared inline --param max-inline-insns-recursive is taken into account.  For function
               not declared inline, recursive inlining happens only when -finline-functions (included in -O3) is
               enabled and --param max-inline-insns-recursive-auto is used.  The default value is 450.


           max-inline-recursive-depth
           max-inline-recursive-depth-auto
               Specifies maximum recursion depth used by the recursive inlining.


               For functions declared inline --param max-inline-recursive-depth is taken into account.  For function
               not declared inline, recursive inlining happens only when -finline-functions (included in -O3) is
               enabled and --param max-inline-recursive-depth-auto is used.  The default value is 450.


           min-inline-recursive-probability
               Recursive inlining is profitable only for function having deep recursion in average and can hurt for
               function having little recursion depth by increasing the prologue size or complexity of function body
               to other optimizers.


               When profile feedback is available (see -fprofile-generate) the actual recursion depth can be guessed
               from probability that function will recurse via given call expression.  This parameter limits inlining
               only to call expression whose probability exceeds given threshold (in percents).  The default value is
               10.


           inline-call-cost
               Specify cost of call instruction relative to simple arithmetics operations (having cost of 1).
               Increasing this cost disqualifies inlining of non-leaf functions and at the same time increases size of
               leaf function that is believed to reduce function size by being inlined.  In effect it increases amount
               of inlining for code having large abstraction penalty (many functions that just pass the arguments to
               other functions) and decrease inlining for code with low abstraction penalty.  The default value is 16.


           max-unrolled-insns
               The maximum number of instructions that a loop should have if that loop is unrolled, and if the loop is
               unrolled, it determines how many times the loop code is unrolled.


           max-average-unrolled-insns
               The maximum number of instructions biased by probabilities of their execution that a loop should have
               if that loop is unrolled, and if the loop is unrolled, it determines how many times the loop code is
               unrolled.


           max-unroll-times
               The maximum number of unrollings of a single loop.


           max-peeled-insns
               The maximum number of instructions that a loop should have if that loop is peeled, and if the loop is
               peeled, it determines how many times the loop code is peeled.


           max-peel-times
               The maximum number of peelings of a single loop.


           max-completely-peeled-insns
               The maximum number of insns of a completely peeled loop.


           max-completely-peel-times
               The maximum number of iterations of a loop to be suitable for complete peeling.


           max-unswitch-insns
               The maximum number of insns of an unswitched loop.


           max-unswitch-level
               The maximum number of branches unswitched in a single loop.


           lim-expensive
               The minimum cost of an expensive expression in the loop invariant motion.


           iv-consider-all-candidates-bound
               Bound on number of candidates for induction variables below that all candidates are considered for each
               use in induction variable optimizations.  Only the most relevant candidates are considered if there are
               more candidates, to avoid quadratic time complexity.


           iv-max-considered-uses
               The induction variable optimizations give up on loops that contain more induction variable uses.


           iv-always-prune-cand-set-bound
               If number of candidates in the set is smaller than this value, we always try to remove unnecessary ivs
               from the set during its optimization when a new iv is added to the set.


           scev-max-expr-size
               Bound on size of expressions used in the scalar evolutions analyzer.  Large expressions slow the ana-
               lyzer.


           vect-max-version-checks
               The maximum number of runtime checks that can be performed when doing loop versioning in the vector-
               izer.  See option ftree-vect-loop-version for more information.


           max-iterations-to-track
               The maximum number of iterations of a loop the brute force algorithm for analysis of # of iterations of
               the loop tries to evaluate.


           hot-bb-count-fraction
               Select fraction of the maximal count of repetitions of basic block in program given basic block needs
               to have to be considered hot.


           hot-bb-frequency-fraction
               Select fraction of the maximal frequency of executions of basic block in function given basic block
               needs to have to be considered hot


           max-predicted-iterations
               The maximum number of loop iterations we predict statically.  This is useful in cases where function
               contain single loop with known bound and other loop with unknown.  We predict the known number of iter-
               ations correctly, while the unknown number of iterations average to roughly 10.  This means that the
               loop without bounds would appear artificially cold relative to the other one.


           tracer-dynamic-coverage
           tracer-dynamic-coverage-feedback
               This value is used to limit superblock formation once the given percentage of executed instructions is
               covered.  This limits unnecessary code size expansion.


               The tracer-dynamic-coverage-feedback is used only when profile feedback is available.  The real pro-
               files (as opposed to statically estimated ones) are much less balanced allowing the threshold to be
               larger value.


           tracer-max-code-growth
               Stop tail duplication once code growth has reached given percentage.  This is rather hokey argument, as
               most of the duplicates will be eliminated later in cross jumping, so it may be set to much higher val-
               ues than is the desired code growth.


           tracer-min-branch-ratio
               Stop reverse growth when the reverse probability of best edge is less than this threshold (in percent).


           tracer-min-branch-ratio
           tracer-min-branch-ratio-feedback
               Stop forward growth if the best edge do have probability lower than this threshold.


               Similarly to tracer-dynamic-coverage two values are present, one for compilation for profile feedback
               and one for compilation without.  The value for compilation with profile feedback needs to be more con-
               servative (higher) in order to make tracer effective.


           max-cse-path-length
               Maximum number of basic blocks on path that cse considers.  The default is 10.


           max-cse-insns
               The maximum instructions CSE process before flushing. The default is 1000.


           global-var-threshold
               Counts the number of function calls (n) and the number of call-clobbered variables (v).  If nxv is
               larger than this limit, a single artificial variable will be created to represent all the call-clob-
               bered variables at function call sites.  This artificial variable will then be made to alias every
               call-clobbered variable.  (done as "int * size_t" on the host machine; beware overflow).


           max-aliased-vops
               Maximum number of virtual operands allowed to represent aliases before triggering the alias grouping
               heuristic.  Alias grouping reduces compile times and memory consumption needed for aliasing at the
               expense of precision loss in alias information.


           ggc-min-expand
               GCC uses a garbage collector to manage its own memory allocation.  This parameter specifies the minimum
               percentage by which the garbage collector’s heap should be allowed to expand between collections.  Tun-
               ing this may improve compilation speed; it has no effect on code generation.


               The default is 30% + 70% * (RAM/1GB) with an upper bound of 100% when RAM >= 1GB.  If "getrlimit" is
               available, the notion of "RAM" is the smallest of actual RAM and "RLIMIT_DATA" or "RLIMIT_AS".  If GCC
               is not able to calculate RAM on a particular platform, the lower bound of 30% is used.  Setting this
               parameter and ggc-min-heapsize to zero causes a full collection to occur at every opportunity.  This is
               extremely slow, but can be useful for debugging.


           ggc-min-heapsize
               Minimum size of the garbage collector’s heap before it begins bothering to collect garbage.  The first
               collection occurs after the heap expands by ggc-min-expand% beyond ggc-min-heapsize.  Again, tuning
               this may improve compilation speed, and has no effect on code generation.


               The default is the smaller of RAM/8, RLIMIT_RSS, or a limit which tries to ensure that RLIMIT_DATA or
               RLIMIT_AS are not exceeded, but with a lower bound of 4096 (four megabytes) and an upper bound of
               131072 (128 megabytes).  If GCC is not able to calculate RAM on a particular platform, the lower bound
               is used.  Setting this parameter very large effectively disables garbage collection.  Setting this
               parameter and ggc-min-expand to zero causes a full collection to occur at every opportunity.


           max-reload-search-insns
               The maximum number of instruction reload should look backward for equivalent register.  Increasing val-
               ues mean more aggressive optimization, making the compile time increase with probably slightly better
               performance.  The default value is 100.


           max-cselib-memory-location
               The maximum number of memory locations cselib should take into account.  Increasing values mean more
               aggressive optimization, making the compile time increase with probably slightly better performance.
               The default value is 500.


           max-flow-memory-location
               Similar as max-cselib-memory-location but for dataflow liveness.  The default value is 100.


           reorder-blocks-duplicate
           reorder-blocks-duplicate-feedback
               Used by basic block reordering pass to decide whether to use unconditional branch or duplicate the code
               on its destination.  Code is duplicated when its estimated size is smaller than this value multiplied
               by the estimated size of unconditional jump in the hot spots of the program.


               The reorder-block-duplicate-feedback is used only when profile feedback is available and may be set to
               higher values than reorder-block-duplicate since information about the hot spots is more accurate.


           max-sched-ready-insns
               The maximum number of instructions ready to be issued the scheduler should consider at any given time
               during the first scheduling pass.  Increasing values mean more thorough searches, making the compila-
               tion time increase with probably little benefit.  The default value is 100.


           max-sched-region-blocks
               The maximum number of blocks in a region to be considered for interblock scheduling.  The default value
               is 10.


           max-sched-region-insns
               The maximum number of insns in a region to be considered for interblock scheduling.  The default value
               is 100.


           min-sched-prob
               The minimum probability of reaching a source block for interblock speculative scheduling.  The default
               value is 40.


           max-last-value-rtl
               The maximum size measured as number of RTLs that can be recorded in an expression in combiner for a
               pseudo register as last known value of that register.  The default is 10000.


           integer-share-limit
               Small integer constants can use a shared data structure, reducing the compiler’s memory usage and
               increasing its speed.  This sets the maximum value of a shared integer constant’s.  The default value
               is 256.


           min-virtual-mappings
               Specifies the minimum number of virtual mappings in the incremental SSA updater that should be regis-
               tered to trigger the virtual mappings heuristic defined by virtual-mappings-ratio.  The default value
               is 100.


           virtual-mappings-ratio
               If the number of virtual mappings is virtual-mappings-ratio bigger than the number of virtual symbols
               to be updated, then the incremental SSA updater switches to a full update for those symbols.  The
               default ratio is 3.


           ssp-buffer-size
               The minimum size of buffers (i.e. arrays) that will receive stack smashing protection when -fstack-pro-
               tection is used.


           max-jump-thread-duplication-stmts
               Maximum number of statements allowed in a block that needs to be duplicated when threading jumps.


           max-fields-for-field-sensitive
               Maximum number of fields in a structure we will treat in a field sensitive manner during pointer analy-
               sis.


       Options Controlling the Preprocessor


       These options control the C preprocessor, which is run on each C source file before actual compilation.


       If you use the -E option, nothing is done except preprocessing.  Some of these options make sense only together
       with -E because they cause the preprocessor output to be unsuitable for actual compilation.


           You can use -Wp,option to bypass the compiler driver and pass option directly through to the preprocessor.
           If option contains commas, it is split into multiple options at the commas.  However, many options are mod-
           ified, translated or interpreted by the compiler driver before being passed to the preprocessor, and -Wp
           forcibly bypasses this phase.  The preprocessor’s direct interface is undocumented and subject to change,
           so whenever possible you should avoid using -Wp and let the driver handle the options instead.


       -Xpreprocessor option
           Pass option as an option to the preprocessor.  You can use this to supply system-specific preprocessor
           options which GCC does not know how to recognize.


           If you want to pass an option that takes an argument, you must use -Xpreprocessor twice, once for the
           option and once for the argument.


       -D name
           Predefine name as a macro, with definition 1.


       -D name=definition
           The contents of definition are tokenized and processed as if they appeared during translation phase three
           in a #define directive.  In particular, the definition will be truncated by embedded newline characters.


           If you are invoking the preprocessor from a shell or shell-like program you may need to use the shell’s
           quoting syntax to protect characters such as spaces that have a meaning in the shell syntax.


           If you wish to define a function-like macro on the command line, write its argument list with surrounding
           parentheses before the equals sign (if any).  Parentheses are meaningful to most shells, so you will need
           to quote the option.  With sh and csh, -D’name(args...)=definition’ works.


           -D and -U options are processed in the order they are given on the command line.  All -imacros file and
           -include file options are processed after all -D and -U options.


       -U name
           Cancel any previous definition of name, either built in or provided with a -D option.


       -undef
           Do not predefine any system-specific or GCC-specific macros.  The standard predefined macros remain
           defined.


       -I dir
           Add the directory dir to the list of directories to be searched for header files.  Directories named by -I
           are searched before the standard system include directories.  If the directory dir is a standard system
           include directory, the option is ignored to ensure that the default search order for system directories and
           the special treatment of system headers are not defeated .


       -o file
           Write output to file.  This is the same as specifying file as the second non-option argument to cpp.  gcc
           has a different interpretation of a second non-option argument, so you must use -o to specify the output
           file.


       -Wall
           Turns on all optional warnings which are desirable for normal code.  At present this is -Wcomment, -Wtri-
           graphs, -Wmultichar and a warning about integer promotion causing a change of sign in "#if" expressions.
           Note that many of the preprocessor’s warnings are on by default and have no options to control them.


       -Wcomment
       -Wcomments
           Warn whenever a comment-start sequence /* appears in a /* comment, or whenever a backslash-newline appears
           in a // comment.  (Both forms have the same effect.)


       -Wtrigraphs
           Most trigraphs in comments cannot affect the meaning of the program.  However, a trigraph that would form
           an escaped newline (??/ at the end of a line) can, by changing where the comment begins or ends.  There-
           fore, only trigraphs that would form escaped newlines produce warnings inside a comment.


           This option is implied by -Wall.  If -Wall is not given, this option is still enabled unless trigraphs are
           enabled.  To get trigraph conversion without warnings, but get the other -Wall warnings, use -trigraphs
           -Wall -Wno-trigraphs.


       -Wtraditional
           Warn about certain constructs that behave differently in traditional and ISO C.  Also warn about ISO C con-
           structs that have no traditional C equivalent, and problematic constructs which should be avoided.


       -Wimport
           Warn the first time #import is used.


       -Wundef
           Warn whenever an identifier which is not a macro is encountered in an #if directive, outside of defined.
           Such identifiers are replaced with zero.


       -Wunused-macros
           Warn about macros defined in the main file that are unused.  A macro is used if it is expanded or tested
           for existence at least once.  The preprocessor will also warn if the macro has not been used at the time it
           is redefined or undefined.


           Built-in macros, macros defined on the command line, and macros defined in include files are not warned
           about.


           Note: If a macro is actually used, but only used in skipped conditional blocks, then CPP will report it as
           unused.  To avoid the warning in such a case, you might improve the scope of the macro’s definition by, for
           example, moving it into the first skipped block.  Alternatively, you could provide a dummy use with some-
           thing like:


                   #if defined the_macro_causing_the_warning
                   #endif


       -Wendif-labels
           Warn whenever an #else or an #endif are followed by text.  This usually happens in code of the form


                   #if FOO
                   ...
                   #else FOO
                   ...
                   #endif FOO


           The second and third "FOO" should be in comments, but often are not in older programs.  This warning is on
           by default.


       -Werror
           Make all warnings into hard errors.  Source code which triggers warnings will be rejected.


       -Wsystem-headers
           Issue warnings for code in system headers.  These are normally unhelpful in finding bugs in your own code,
           therefore suppressed.  If you are responsible for the system library, you may want to see them.


       -w  Suppress all warnings, including those which GNU CPP issues by default.


       -pedantic
           Issue all the mandatory diagnostics listed in the C standard.  Some of them are left out by default, since
           they trigger frequently on harmless code.


       -pedantic-errors
           Issue all the mandatory diagnostics, and make all mandatory diagnostics into errors.  This includes manda-
           tory diagnostics that GCC issues without -pedantic but treats as warnings.


       -M  Instead of outputting the result of preprocessing, output a rule suitable for make describing the dependen-
           cies of the main source file.  The preprocessor outputs one make rule containing the object file name for
           that source file, a colon, and the names of all the included files, including those coming from -include or
           -imacros command line options.


           Unless specified explicitly (with -MT or -MQ), the object file name consists of the basename of the source
           file with any suffix replaced with object file suffix.  If there are many included files then the rule is
           split into several lines using \-newline.  The rule has no commands.


           This option does not suppress the preprocessor’s debug output, such as -dM.  To avoid mixing such debug
           output with the dependency rules you should explicitly specify the dependency output file with -MF, or use
           an environment variable like DEPENDENCIES_OUTPUT.  Debug output will still be sent to the regular output
           stream as normal.


           Passing -M to the driver implies -E, and suppresses warnings with an implicit -w.


       -MM Like -M but do not mention header files that are found in system header directories, nor header files that
           are included, directly or indirectly, from such a header.


           This implies that the choice of angle brackets or double quotes in an #include directive does not in itself
           determine whether that header will appear in -MM dependency output.  This is a slight change in semantics
           from GCC versions 3.0 and earlier.


       -MF file
           When used with -M or -MM, specifies a file to write the dependencies to.  If no -MF switch is given the
           preprocessor sends the rules to the same place it would have sent preprocessed output.


           When used with the driver options -MD or -MMD, -MF overrides the default dependency output file.


       -MG In conjunction with an option such as -M requesting dependency generation, -MG assumes missing header files
           are generated files and adds them to the dependency list without raising an error.  The dependency filename
           is taken directly from the "#include" directive without prepending any path.  -MG also suppresses prepro-
           cessed output, as a missing header file renders this useless.


           This feature is used in automatic updating of makefiles.


       -MP This option instructs CPP to add a phony target for each dependency other than the main file, causing each
           to depend on nothing.  These dummy rules work around errors make gives if you remove header files without
           updating the Makefile to match.


           This is typical output:


                   test.o: test.c test.h


                   test.h:


       -MT target
           Change the target of the rule emitted by dependency generation.  By default CPP takes the name of the main
           input file, including any path, deletes any file suffix such as .c, and appends the platform’s usual object
           suffix.  The result is the target.


           An -MT option will set the target to be exactly the string you specify.  If you want multiple targets, you
           can specify them as a single argument to -MT, or use multiple -MT options.


           For example, -MT ’$(objpfx)foo.o’ might give


                   $(objpfx)foo.o: foo.c


       -MQ target
           Same as -MT, but it quotes any characters which are special to Make.  -MQ ’$(objpfx)foo.o’ gives


                   $$(objpfx)foo.o: foo.c


           The default target is automatically quoted, as if it were given with -MQ.


       -MD -MD is equivalent to -M -MF file, except that -E is not implied.  The driver determines file based on
           whether an -o option is given.  If it is, the driver uses its argument but with a suffix of .d, otherwise
           it take the basename of the input file and applies a .d suffix.


           If -MD is used in conjunction with -E, any -o switch is understood to specify the dependency output file
           (but @pxref{dashMF,,-MF}), but if used without -E, each -o is understood to specify a target object file.


           Since -E is not implied, -MD can be used to generate a dependency output file as a side-effect of the com-
           pilation process.


       -MMD
           Like -MD except mention only user header files, not system header files.


       -fpch-deps
           When using precompiled headers, this flag will cause the dependency-output flags to also list the files
           from the precompiled header’s dependencies.  If not specified only the precompiled header would be listed
           and not the files that were used to create it because those files are not consulted when a precompiled
           header is used.


       -fpch-preprocess
           This option allows use of a precompiled header together with -E.  It inserts a special "#pragma", "#pragma
           GCC pch_preprocess "<filename>"" in the output to mark the place where the precompiled header was found,
           and its filename.  When -fpreprocessed is in use, GCC recognizes this "#pragma" and loads the PCH.


           This option is off by default, because the resulting preprocessed output is only really suitable as input
           to GCC.  It is switched on by -save-temps.


           You should not write this "#pragma" in your own code, but it is safe to edit the filename if the PCH file
           is available in a different location.  The filename may be absolute or it may be relative to GCC’s current
           directory.


       -x c
       -x c++
       -x objective-c
       -x assembler-with-cpp
           Specify the source language: C, C++, Objective-C, or assembly.  This has nothing to do with standards con-
           formance or extensions; it merely selects which base syntax to expect.  If you give none of these options,
           cpp will deduce the language from the extension of the source file: .c, .cc, .m, or .S.  Some other common
           extensions for C++ and assembly are also recognized.  If cpp does not recognize the extension, it will
           treat the file as C; this is the most generic mode.


           Note: Previous versions of cpp accepted a -lang option which selected both the language and the standards
           conformance level.  This option has been removed, because it conflicts with the -l option.


       -std=standard
       -ansi
           Specify the standard to which the code should conform.  Currently CPP knows about C and C++ standards; oth-
           ers may be added in the future.


           standard may be one of:


           "iso9899:1990"
           "c89"
               The ISO C standard from 1990.  c89 is the customary shorthand for this version of the standard.


               The -ansi option is equivalent to -std=c89.


           "iso9899:199409"
               The 1990 C standard, as amended in 1994.


           "iso9899:1999"
           "c99"
           "iso9899:199x"
           "c9x"
               The revised ISO C standard, published in December 1999.  Before publication, this was known as C9X.


           "gnu89"
               The 1990 C standard plus GNU extensions.  This is the default.


           "gnu99"
           "gnu9x"
               The 1999 C standard plus GNU extensions.


           "c++98"
               The 1998 ISO C++ standard plus amendments.


           "gnu++98"
               The same as -std=c++98 plus GNU extensions.  This is the default for C++ code.


       -I- Split the include path.  Any directories specified with -I options before -I- are searched only for headers
           requested with "#include "file""; they are not searched for "#include <file>".  If additional directories
           are specified with -I options after the -I-, those directories are searched for all #include directives.


           In addition, -I- inhibits the use of the directory of the current file directory as the first search direc-
           tory for "#include "file"".  This option has been deprecated.


       -nostdinc
           Do not search the standard system directories for header files.  Only the directories you have specified
           with -I options (and the directory of the current file, if appropriate) are searched.


       -nostdinc++
           Do not search for header files in the C++-specific standard directories, but do still search the other
           standard directories.  (This option is used when building the C++ library.)


       -include file
           Process file as if "#include "file"" appeared as the first line of the primary source file.  However, the
           first directory searched for file is the preprocessor’s working directory instead of the directory contain-
           ing the main source file.  If not found there, it is searched for in the remainder of the "#include "...""
           search chain as normal.


           If multiple -include options are given, the files are included in the order they appear on the command
           line.


       -imacros file
           Exactly like -include, except that any output produced by scanning file is thrown away.  Macros it defines
           remain defined.  This allows you to acquire all the macros from a header without also processing its decla-
           rations.


           All files specified by -imacros are processed before all files specified by -include.


       -idirafter dir
           Search dir for header files, but do it after all directories specified with -I and the standard system
           directories have been exhausted.  dir is treated as a system include directory.


       -iprefix prefix
           Specify prefix as the prefix for subsequent -iwithprefix options.  If the prefix represents a directory,
           you should include the final /.


       -iwithprefix dir
       -iwithprefixbefore dir
           Append dir to the prefix specified previously with -iprefix, and add the resulting directory to the include
           search path.  -iwithprefixbefore puts it in the same place -I would; -iwithprefix puts it where -idirafter
           would.


       -isysroot dir
           This option is like the --sysroot option, but applies only to header files.  See the --sysroot option for
           more information.


       -isystem dir
           Search dir for header files, after all directories specified by -I but before the standard system directo-
           ries.  Mark it as a system directory, so that it gets the same special treatment as is applied to the stan-
           dard system directories.


       -iquote dir
           Search dir only for header files requested with "#include "file""; they are not searched for
           "#include <file>", before all directories specified by -I and before the standard system directories.


       -fdollars-in-identifiers
           Accept $ in identifiers.


       -fextended-identifiers
           Accept universal character names in identifiers.  This option is experimental; in a future version of GCC,
           it will be enabled by default for C99 and C++.


       -fpreprocessed
           Indicate to the preprocessor that the input file has already been preprocessed.  This suppresses things
           like macro expansion, trigraph conversion, escaped newline splicing, and processing of most directives.
           The preprocessor still recognizes and removes comments, so that you can pass a file preprocessed with -C to
           the compiler without problems.  In this mode the integrated preprocessor is little more than a tokenizer
           for the front ends.


           -fpreprocessed is implicit if the input file has one of the extensions .i, .ii or .mi.  These are the
           extensions that GCC uses for preprocessed files created by -save-temps.


       -ftabstop=width
           Set the distance between tab stops.  This helps the preprocessor report correct column numbers in warnings
           or errors, even if tabs appear on the line.  If the value is less than 1 or greater than 100, the option is
           ignored.  The default is 8.


       -fexec-charset=charset
           Set the execution character set, used for string and character constants.  The default is UTF-8.  charset
           can be any encoding supported by the system’s "iconv" library routine.


       -fwide-exec-charset=charset
           Set the wide execution character set, used for wide string and character constants.  The default is UTF-32
           or UTF-16, whichever corresponds to the width of "wchar_t".  As with -fexec-charset, charset can be any
           encoding supported by the system’s "iconv" library routine; however, you will have problems with encodings
           that do not fit exactly in "wchar_t".


       -finput-charset=charset
           Set the input character set, used for translation from the character set of the input file to the source
           character set used by GCC.  If the locale does not specify, or GCC cannot get this information from the
           locale, the default is UTF-8.  This can be overridden by either the locale or this command line option.
           Currently the command line option takes precedence if there’s a conflict.  charset can be any encoding sup-
           ported by the system’s "iconv" library routine.


       -fworking-directory
           Enable generation of linemarkers in the preprocessor output that will let the compiler know the current
           working directory at the time of preprocessing.  When this option is enabled, the preprocessor will emit,
           after the initial linemarker, a second linemarker with the current working directory followed by two
           slashes.  GCC will use this directory, when it’s present in the preprocessed input, as the directory emit-
           ted as the current working directory in some debugging information formats.  This option is implicitly
           enabled if debugging information is enabled, but this can be inhibited with the negated form -fno-work-
           ing-directory.  If the -P flag is present in the command line, this option has no effect, since no "#line"
           directives are emitted whatsoever.


       -fno-show-column
           Do not print column numbers in diagnostics.  This may be necessary if diagnostics are being scanned by a
           program that does not understand the column numbers, such as dejagnu.


       -A predicate=answer
           Make an assertion with the predicate predicate and answer answer.  This form is preferred to the older form
           -A predicate(answer), which is still supported, because it does not use shell special characters.


       -A -predicate=answer
           Cancel an assertion with the predicate predicate and answer answer.


       -dCHARS
           CHARS is a sequence of one or more of the following characters, and must not be preceded by a space.  Other
           characters are interpreted by the compiler proper, or reserved for future versions of GCC, and so are
           silently ignored.  If you specify characters whose behavior conflicts, the result is undefined.


           M   Instead of the normal output, generate a list of #define directives for all the macros defined during
               the execution of the preprocessor, including predefined macros.  This gives you a way of finding out
               what is predefined in your version of the preprocessor.  Assuming you have no file foo.h, the command


                       touch foo.h; cpp -dM foo.h


               will show all the predefined macros.


           D   Like M except in two respects: it does not include the predefined macros, and it outputs both the
               #define directives and the result of preprocessing.  Both kinds of output go to the standard output
               file.


           N   Like D, but emit only the macro names, not their expansions.


           I   Output #include directives in addition to the result of preprocessing.


       -P  Inhibit generation of linemarkers in the output from the preprocessor.  This might be useful when running
           the preprocessor on something that is not C code, and will be sent to a program which might be confused by
           the linemarkers.


       -C  Do not discard comments.  All comments are passed through to the output file, except for comments in pro-
           cessed directives, which are deleted along with the directive.


           You should be prepared for side effects when using -C; it causes the preprocessor to treat comments as
           tokens in their own right.  For example, comments appearing at the start of what would be a directive line
           have the effect of turning that line into an ordinary source line, since the first token on the line is no
           longer a #.


       -CC Do not discard comments, including during macro expansion.  This is like -C, except that comments contained
           within macros are also passed through to the output file where the macro is expanded.


           In addition to the side-effects of the -C option, the -CC option causes all C++-style comments inside a
           macro to be converted to C-style comments.  This is to prevent later use of that macro from inadvertently
           commenting out the remainder of the source line.


           The -CC option is generally used to support lint comments.


       -traditional-cpp
           Try to imitate the behavior of old-fashioned C preprocessors, as opposed to ISO C preprocessors.


       -trigraphs
           Process trigraph sequences.  These are three-character sequences, all starting with ??, that are defined by
           ISO C to stand for single characters.  For example, ??/ stands for \, so ’??/n’ is a character constant for
           a newline.  By default, GCC ignores trigraphs, but in standard-conforming modes it converts them.  See the
           -std and -ansi options.


           The nine trigraphs and their replacements are


                   Trigraph:       ??(  ??)  ??<  ??>  ??=  ??/  ??’  ??!  ??-
                   Replacement:      [    ]    {    }    #    \    ^    │    ~


       -remap
           Enable special code to work around file systems which only permit very short file names, such as MS-DOS.


       --help
       --target-help
           Print text describing all the command line options instead of preprocessing anything.


       -v  Verbose mode.  Print out GNU CPP’s version number at the beginning of execution, and report the final form
           of the include path.


       -H  Print the name of each header file used, in addition to other normal activities.  Each name is indented to
           show how deep in the #include stack it is.  Precompiled header files are also printed, even if they are
           found to be invalid; an invalid precompiled header file is printed with ...x and a valid one with ...! .


       -version
       --version
           Print out GNU CPP’s version number.  With one dash, proceed to preprocess as normal.  With two dashes, exit
           immediately.


       Passing Options to the Assembler


       You can pass options to the assembler.


       -Wa,option
           Pass option as an option to the assembler.  If option contains commas, it is split into multiple options at
           the commas.


       -Xassembler option
           Pass option as an option to the assembler.  You can use this to supply system-specific assembler options
           which GCC does not know how to recognize.


           If you want to pass an option that takes an argument, you must use -Xassembler twice, once for the option
           and once for the argument.


       Options for Linking


       These options come into play when the compiler links object files into an executable output file.  They are
       meaningless if the compiler is not doing a link step.


       object-file-name
           A file name that does not end in a special recognized suffix is considered to name an object file or
           library.  (Object files are distinguished from libraries by the linker according to the file contents.)  If
           linking is done, these object files are used as input to the linker.


       -c
       -S
       -E  If any of these options is used, then the linker is not run, and object file names should not be used as
           arguments.


       -llibrary
       -l library
           Search the library named library when linking.  (The second alternative with the library as a separate
           argument is only for POSIX compliance and is not recommended.)


           It makes a difference where in the command you write this option; the linker searches and processes
           libraries and object files in the order they are specified.  Thus, foo.o -lz bar.o searches library z after
           file foo.o but before bar.o.  If bar.o refers to functions in z, those functions may not be loaded.


           The linker searches a standard list of directories for the library, which is actually a file named libli-
           brary.a.  The linker then uses this file as if it had been specified precisely by name.


           The directories searched include several standard system directories plus any that you specify with -L.


           Normally the files found this way are library files---archive files whose members are object files.  The
           linker handles an archive file by scanning through it for members which define symbols that have so far
           been referenced but not defined.  But if the file that is found is an ordinary object file, it is linked in
           the usual fashion.  The only difference between using an -l option and specifying a file name is that -l
           surrounds library with lib and .a and searches several directories.


       -lobjc
           You need this special case of the -l option in order to link an Objective-C or Objective-C++ program.


       -nostartfiles
           Do not use the standard system startup files when linking.  The standard system libraries are used nor-
           mally, unless -nostdlib or -nodefaultlibs is used.


       -nodefaultlibs
           Do not use the standard system libraries when linking.  Only the libraries you specify will be passed to
           the linker.  The standard startup files are used normally, unless -nostartfiles is used.  The compiler may
           generate calls to "memcmp", "memset", "memcpy" and "memmove".  These entries are usually resolved by
           entries in libc.  These entry points should be supplied through some other mechanism when this option is
           specified.


       -nostdlib
           Do not use the standard system startup files or libraries when linking.  No startup files and only the
           libraries you specify will be passed to the linker.  The compiler may generate calls to "memcmp", "memset",
           "memcpy" and "memmove".  These entries are usually resolved by entries in libc.  These entry points should
           be supplied through some other mechanism when this option is specified.


           One of the standard libraries bypassed by -nostdlib and -nodefaultlibs is libgcc.a, a library of internal
           subroutines that GCC uses to overcome shortcomings of particular machines, or special needs for some lan-
           guages.


           In most cases, you need libgcc.a even when you want to avoid other standard libraries.  In other words,
           when you specify -nostdlib or -nodefaultlibs you should usually specify -lgcc as well.  This ensures that
           you have no unresolved references to internal GCC library subroutines.  (For example, __main, used to
           ensure C++ constructors will be called.)


       -pie
           Produce a position independent executable on targets which support it.  For predictable results, you must
           also specify the same set of options that were used to generate code (-fpie, -fPIE, or model suboptions)
           when you specify this option.


       -rdynamic
           Pass the flag -export-dynamic to the ELF linker, on targets that support it. This instructs the linker to
           add all symbols, not only used ones, to the dynamic symbol table. This option is needed for some uses of
           "dlopen" or to allow obtaining backtraces from within a program.


       -s  Remove all symbol table and relocation information from the executable.


       -static
           On systems that support dynamic linking, this prevents linking with the shared libraries.  On other sys-
           tems, this option has no effect.


       -shared
           Produce a shared object which can then be linked with other objects to form an executable.  Not all systems
           support this option.  For predictable results, you must also specify the same set of options that were used
           to generate code (-fpic, -fPIC, or model suboptions) when you specify this option.[1]


       -shared-libgcc
       -static-libgcc
           On systems that provide libgcc as a shared library, these options force the use of either the shared or
           static version respectively.  If no shared version of libgcc was built when the compiler was configured,
           these options have no effect.


           There are several situations in which an application should use the shared libgcc instead of the static
           version.  The most common of these is when the application wishes to throw and catch exceptions across dif-
           ferent shared libraries.  In that case, each of the libraries as well as the application itself should use
           the shared libgcc.


           Therefore, the G++ and GCJ drivers automatically add -shared-libgcc whenever you build a shared library or
           a main executable, because C++ and Java programs typically use exceptions, so this is the right thing to
           do.


           If, instead, you use the GCC driver to create shared libraries, you may find that they will not always be
           linked with the shared libgcc.  If GCC finds, at its configuration time, that you have a non-GNU linker or
           a GNU linker that does not support option --eh-frame-hdr, it will link the shared version of libgcc into
           shared libraries by default.  Otherwise, it will take advantage of the linker and optimize away the linking
           with the shared version of libgcc, linking with the static version of libgcc by default.  This allows
           exceptions to propagate through such shared libraries, without incurring relocation costs at library load
           time.


           However, if a library or main executable is supposed to throw or catch exceptions, you must link it using
           the G++ or GCJ driver, as appropriate for the languages used in the program, or using the option
           -shared-libgcc, such that it is linked with the shared libgcc.


       -symbolic
           Bind references to global symbols when building a shared object.  Warn about any unresolved references
           (unless overridden by the link editor option -Xlinker -z -Xlinker defs).  Only a few systems support this
           option.


       -Xlinker option
           Pass option as an option to the linker.  You can use this to supply system-specific linker options which
           GCC does not know how to recognize.


           If you want to pass an option that takes an argument, you must use -Xlinker twice, once for the option and
           once for the argument.  For example, to pass -assert definitions, you must write -Xlinker -assert -Xlinker
           definitions.  It does not work to write -Xlinker "-assert definitions", because this passes the entire
           string as a single argument, which is not what the linker expects.


       -Wl,option
           Pass option as an option to the linker.  If option contains commas, it is split into multiple options at
           the commas.


       -u symbol
           Pretend the symbol symbol is undefined, to force linking of library modules to define it.  You can use -u
           multiple times with different symbols to force loading of additional library modules.


       Options for Directory Search


       These options specify directories to search for header files, for libraries and for parts of the compiler:


       -Idir
           Add the directory dir to the head of the list of directories to be searched for header files.  This can be
           used to override a system header file, substituting your own version, since these directories are searched
           before the system header file directories.  However, you should not use this option to add directories that
           contain vendor-supplied system header files (use -isystem for that).  If you use more than one -I option,
           the directories are scanned in left-to-right order; the standard system directories come after.


           If a standard system include directory, or a directory specified with -isystem, is also specified with -I,
           the -I option will be ignored.  The directory will still be searched but as a system directory at its nor-
           mal position in the system include chain.  This is to ensure that GCC’s procedure to fix buggy system head-
           ers and the ordering for the include_next directive are not inadvertently changed.  If you really need to
           change the search order for system directories, use the -nostdinc and/or -isystem options.


       -iquotedir
           Add the directory dir to the head of the list of directories to be searched for header files only for the
           case of #include "file"; they are not searched for #include <file>, otherwise just like -I.


       -Ldir
           Add directory dir to the list of directories to be searched for -l.


       -Bprefix
           This option specifies where to find the executables, libraries, include files, and data files of the com-
           piler itself.


           The compiler driver program runs one or more of the subprograms cpp, cc1, as and ld.  It tries prefix as a
           prefix for each program it tries to run, both with and without machine/version/.


           For each subprogram to be run, the compiler driver first tries the -B prefix, if any.  If that name is not
           found, or if -B was not specified, the driver tries two standard prefixes, which are /usr/lib/gcc/ and
           /usr/local/lib/gcc/.  If neither of those results in a file name that is found, the unmodified program name
           is searched for using the directories specified in your PATH environment variable.


           The compiler will check to see if the path provided by the -B refers to a directory, and if necessary it
           will add a directory separator character at the end of the path.


           -B prefixes that effectively specify directory names also apply to libraries in the linker, because the
           compiler translates these options into -L options for the linker.  They also apply to includes files in the
           preprocessor, because the compiler translates these options into -isystem options for the preprocessor.  In
           this case, the compiler appends include to the prefix.


           The run-time support file libgcc.a can also be searched for using the -B prefix, if needed.  If it is not
           found there, the two standard prefixes above are tried, and that is all.  The file is left out of the link
           if it is not found by those means.


           Another way to specify a prefix much like the -B prefix is to use the environment variable GCC_EXEC_PREFIX.


           As a special kludge, if the path provided by -B is [dir/]stageN/, where N is a number in the range 0 to 9,
           then it will be replaced by [dir/]include.  This is to help with boot-strapping the compiler.


       -specs=file
           Process file after the compiler reads in the standard specs file, in order to override the defaults that
           the gcc driver program uses when determining what switches to pass to cc1, cc1plus, as, ld, etc.  More than
           one -specs=file can be specified on the command line, and they are processed in order, from left to right.


       --sysroot=dir
           Use dir as the logical root directory for headers and libraries.  For example, if the compiler would nor-
           mally search for headers in /usr/include and libraries in /usr/lib, it will instead search dir/usr/include
           and dir/usr/lib.


           If you use both this option and the -isysroot option, then the --sysroot option will apply to libraries,
           but the -isysroot option will apply to header files.


           The GNU linker (beginning with version 2.16) has the necessary support for this option.  If your linker
           does not support this option, the header file aspect of --sysroot will still work, but the library aspect
           will not.


       -I- This option has been deprecated.  Please use -iquote instead for -I directories before the -I- and remove
           the -I-.  Any directories you specify with -I options before the -I- option are searched only for the case
           of #include "file"; they are not searched for #include <file>.


           If additional directories are specified with -I options after the -I-, these directories are searched for
           all #include directives.  (Ordinarily all -I directories are used this way.)


           In addition, the -I- option inhibits the use of the current directory (where the current input file came
           from) as the first search directory for #include "file".  There is no way to override this effect of -I-.
           With -I. you can specify searching the directory which was current when the compiler was invoked.  That is
           not exactly the same as what the preprocessor does by default, but it is often satisfactory.


           -I- does not inhibit the use of the standard system directories for header files.  Thus, -I- and -nostdinc
           are independent.


       Specifying Target Machine and Compiler Version


       The usual way to run GCC is to run the executable called gcc, or <machine>-gcc when cross-compiling, or
       <machine>-gcc-<version> to run a version other than the one that was installed last.  Sometimes this is incon-
       venient, so GCC provides options that will switch to another cross-compiler or version.


       -b machine
           The argument machine specifies the target machine for compilation.


           The value to use for machine is the same as was specified as the machine type when configuring GCC as a
           cross-compiler.  For example, if a cross-compiler was configured with configure arm-elf, meaning to compile
           for an arm processor with elf binaries, then you would specify -b arm-elf to run that cross compiler.
           Because there are other options beginning with -b, the configuration must contain a hyphen.


       -V version
           The argument version specifies which version of GCC to run.  This is useful when multiple versions are
           installed.  For example, version might be 4.0, meaning to run GCC version 4.0.


       The -V and -b options work by running the <machine>-gcc-<version> executable, so there’s no real reason to use
       them if you can just run that directly.


       Hardware Models and Configurations


       Earlier we discussed the standard option -b which chooses among different installed compilers for completely
       different target machines, such as VAX vs. 68000 vs. 80386.


       In addition, each of these target machine types can have its own special options, starting with -m, to choose
       among various hardware models or configurations---for example, 68010 vs 68020, floating coprocessor or none.  A
       single installed version of the compiler can compile for any model or configuration, according to the options
       specified.


       Some configurations of the compiler also support additional special options, usually for compatibility with
       other compilers on the same platform.


       ARC Options


       These options are defined for ARC implementations:


       -EL Compile code for little endian mode.  This is the default.


       -EB Compile code for big endian mode.


       -mmangle-cpu
           Prepend the name of the cpu to all public symbol names.  In multiple-processor systems, there are many ARC
           variants with different instruction and register set characteristics.  This flag prevents code compiled for
           one cpu to be linked with code compiled for another.  No facility exists for handling variants that are
           "almost identical".  This is an all or nothing option.


       -mcpu=cpu
           Compile code for ARC variant cpu.  Which variants are supported depend on the configuration.  All variants
           support -mcpu=base, this is the default.


       -mtext=text-section
       -mdata=data-section
       -mrodata=readonly-data-section
           Put functions, data, and readonly data in text-section, data-section, and readonly-data-section respec-
           tively by default.  This can be overridden with the "section" attribute.


       ARM Options


       These -m options are defined for Advanced RISC Machines (ARM) architectures:


       -mabi=name
           Generate code for the specified ABI.  Permissible values are: apcs-gnu, atpcs, aapcs, aapcs-linux and iwm-
           mxt.


       -mapcs-frame
           Generate a stack frame that is compliant with the ARM Procedure Call Standard for all functions, even if
           this is not strictly necessary for correct execution of the code.  Specifying -fomit-frame-pointer with
           this option will cause the stack frames not to be generated for leaf functions.  The default is
           -mno-apcs-frame.


       -mapcs
           This is a synonym for -mapcs-frame.


       -mthumb-interwork
           Generate code which supports calling between the ARM and Thumb instruction sets.  Without this option the
           two instruction sets cannot be reliably used inside one program.  The default is -mno-thumb-interwork,
           since slightly larger code is generated when -mthumb-interwork is specified.


       -mno-sched-prolog
           Prevent the reordering of instructions in the function prolog, or the merging of those instruction with the
           instructions in the function’s body.  This means that all functions will start with a recognizable set of
           instructions (or in fact one of a choice from a small set of different function prologues), and this infor-
           mation can be used to locate the start if functions inside an executable piece of code.  The default is
           -msched-prolog.


       -mhard-float
           Generate output containing floating point instructions.  This is the default.


       -msoft-float
           Generate output containing library calls for floating point.  Warning: the requisite libraries are not
           available for all ARM targets.  Normally the facilities of the machine’s usual C compiler are used, but
           this cannot be done directly in cross-compilation.  You must make your own arrangements to provide suitable
           library functions for cross-compilation.


           -msoft-float changes the calling convention in the output file; therefore, it is only useful if you compile
           all of a program with this option.  In particular, you need to compile libgcc.a, the library that comes
           with GCC, with -msoft-float in order for this to work.


       -mfloat-abi=name
           Specifies which ABI to use for floating point values.  Permissible values are: soft, softfp and hard.


           soft and hard are equivalent to -msoft-float and -mhard-float respectively.  softfp allows the generation
           of floating point instructions, but still uses the soft-float calling conventions.


       -mlittle-endian
           Generate code for a processor running in little-endian mode.  This is the default for all standard configu-
           rations.


       -mbig-endian
           Generate code for a processor running in big-endian mode; the default is to compile code for a little-
           endian processor.


       -mwords-little-endian
           This option only applies when generating code for big-endian processors.  Generate code for a little-endian
           word order but a big-endian byte order.  That is, a byte order of the form 32107654.  Note: this option
           should only be used if you require compatibility with code for big-endian ARM processors generated by ver-
           sions of the compiler prior to 2.8.


       -mcpu=name
           This specifies the name of the target ARM processor.  GCC uses this name to determine what kind of instruc-
           tions it can emit when generating assembly code.  Permissible names are: arm2, arm250, arm3, arm6, arm60,
           arm600, arm610, arm620, arm7, arm7m, arm7d, arm7dm, arm7di, arm7dmi, arm70, arm700, arm700i, arm710,
           arm710c, arm7100, arm7500, arm7500fe, arm7tdmi, arm7tdmi-s, arm8, strongarm, strongarm110, strongarm1100,
           arm8, arm810, arm9, arm9e, arm920, arm920t, arm922t, arm946e-s, arm966e-s, arm968e-s, arm926ej-s, arm940t,
           arm9tdmi, arm10tdmi, arm1020t, arm1026ej-s, arm10e, arm1020e, arm1022e, arm1136j-s, arm1136jf-s, mpcore,
           mpcorenovfp, arm1176jz-s, arm1176jzf-s, xscale, iwmmxt, ep9312.


       -mtune=name
           This option is very similar to the -mcpu= option, except that instead of specifying the actual target pro-
           cessor type, and hence restricting which instructions can be used, it specifies that GCC should tune the
           performance of the code as if the target were of the type specified in this option, but still choosing the
           instructions that it will generate based on the cpu specified by a -mcpu= option.  For some ARM implementa-
           tions better performance can be obtained by using this option.


       -march=name
           This specifies the name of the target ARM architecture.  GCC uses this name to determine what kind of
           instructions it can emit when generating assembly code.  This option can be used in conjunction with or
           instead of the -mcpu= option.  Permissible names are: armv2, armv2a, armv3, armv3m, armv4, armv4t, armv5,
           armv5t, armv5te, armv6, armv6j, iwmmxt, ep9312.


       -mfpu=name
       -mfpe=number
       -mfp=number
           This specifies what floating point hardware (or hardware emulation) is available on the target.  Permissi-
           ble names are: fpa, fpe2, fpe3, maverick, vfp.  -mfp and -mfpe are synonyms for -mfpu=fpenumber, for com-
           patibility with older versions of GCC.


           If -msoft-float is specified this specifies the format of floating point values.


       -mstructure-size-boundary=n
           The size of all structures and unions will be rounded up to a multiple of the number of bits set by this
           option.  Permissible values are 8, 32 and 64.  The default value varies for different toolchains.  For the
           COFF targeted toolchain the default value is 8.  A value of 64 is only allowed if the underlying ABI sup-
           ports it.


           Specifying the larger number can produce faster, more efficient code, but can also increase the size of the
           program.  Different values are potentially incompatible.  Code compiled with one value cannot necessarily
           expect to work with code or libraries compiled with another value, if they exchange information using
           structures or unions.


       -mabort-on-noreturn
           Generate a call to the function "abort" at the end of a "noreturn" function.  It will be executed if the
           function tries to return.


       -mlong-calls
       -mno-long-calls
           Tells the compiler to perform function calls by first loading the address of the function into a register
           and then performing a subroutine call on this register.  This switch is needed if the target function will
           lie outside of the 64 megabyte addressing range of the offset based version of subroutine call instruction.


           Even if this switch is enabled, not all function calls will be turned into long calls.  The heuristic is
           that static functions, functions which have the short-call attribute, functions that are inside the scope
           of a #pragma no_long_calls directive and functions whose definitions have already been compiled within the
           current compilation unit, will not be turned into long calls.  The exception to this rule is that weak
           function definitions, functions with the long-call attribute or the section attribute, and functions that
           are within the scope of a #pragma long_calls directive, will always be turned into long calls.


           This feature is not enabled by default.  Specifying -mno-long-calls will restore the default behavior, as
           will placing the function calls within the scope of a #pragma long_calls_off directive.  Note these
           switches have no effect on how the compiler generates code to handle function calls via function pointers.


       -mnop-fun-dllimport
           Disable support for the "dllimport" attribute.


       -msingle-pic-base
           Treat the register used for PIC addressing as read-only, rather than loading it in the prologue for each
           function.  The run-time system is responsible for initializing this register with an appropriate value
           before execution begins.


       -mpic-register=reg
           Specify the register to be used for PIC addressing.  The default is R10 unless stack-checking is enabled,
           when R9 is used.


       -mcirrus-fix-invalid-insns
           Insert NOPs into the instruction stream to in order to work around problems with invalid Maverick instruc-
           tion combinations.  This option is only valid if the -mcpu=ep9312 option has been used to enable generation
           of instructions for the Cirrus Maverick floating point co-processor.  This option is not enabled by
           default, since the problem is only present in older Maverick implementations.  The default can be re-
           enabled by use of the -mno-cirrus-fix-invalid-insns switch.


       -mpoke-function-name
           Write the name of each function into the text section, directly preceding the function prologue.  The gen-
           erated code is similar to this:


                        t0
                            .ascii "arm_poke_function_name", 0
                            .align
                        t1
                            .word 0xff000000 + (t1 - t0)
                        arm_poke_function_name
                            mov     ip, sp
                            stmfd   sp!, {fp, ip, lr, pc}
                            sub     fp, ip, #4


           When performing a stack backtrace, code can inspect the value of "pc" stored at "fp + 0".  If the trace
           function then looks at location "pc - 12" and the top 8 bits are set, then we know that there is a function
           name embedded immediately preceding this location and has length "((pc[-3]) & 0xff000000)".


       -mthumb
           Generate code for the 16-bit Thumb instruction set.  The default is to use the 32-bit ARM instruction set.


       -mtpcs-frame
           Generate a stack frame that is compliant with the Thumb Procedure Call Standard for all non-leaf functions.
           (A leaf function is one that does not call any other functions.)  The default is -mno-tpcs-frame.


       -mtpcs-leaf-frame
           Generate a stack frame that is compliant with the Thumb Procedure Call Standard for all leaf functions.  (A
           leaf function is one that does not call any other functions.)  The default is -mno-apcs-leaf-frame.


       -mcallee-super-interworking
           Gives all externally visible functions in the file being compiled an ARM instruction set header which
           switches to Thumb mode before executing the rest of the function.  This allows these functions to be called
           from non-interworking code.


       -mcaller-super-interworking
           Allows calls via function pointers (including virtual functions) to execute correctly regardless of whether
           the target code has been compiled for interworking or not.  There is a small overhead in the cost of exe-
           cuting a function pointer if this option is enabled.


       -mtp=name
           Specify the access model for the thread local storage pointer.  The valid models are soft, which generates
           calls to "__aeabi_read_tp", cp15, which fetches the thread pointer from "cp15" directly (supported in the
           arm6k architecture), and auto, which uses the best available method for the selected processor.  The
           default setting is auto.


       AVR Options


       These options are defined for AVR implementations:


       -mmcu=mcu
           Specify ATMEL AVR instruction set or MCU type.


           Instruction set avr1 is for the minimal AVR core, not supported by the C compiler, only for assembler pro-
           grams (MCU types: at90s1200, attiny10, attiny11, attiny12, attiny15, attiny28).


           Instruction set avr2 (default) is for the classic AVR core with up to 8K program memory space (MCU types:
           at90s2313, at90s2323, attiny22, at90s2333, at90s2343, at90s4414, at90s4433, at90s4434, at90s8515,
           at90c8534, at90s8535).


           Instruction set avr3 is for the classic AVR core with up to 128K program memory space (MCU types:
           atmega103, atmega603, at43usb320, at76c711).


           Instruction set avr4 is for the enhanced AVR core with up to 8K program memory space (MCU types: atmega8,
           atmega83, atmega85).


           Instruction set avr5 is for the enhanced AVR core with up to 128K program memory space (MCU types:
           atmega16, atmega161, atmega163, atmega32, atmega323, atmega64, atmega128, at43usb355, at94k).


       -msize
           Output instruction sizes to the asm file.


       -minit-stack=N
           Specify the initial stack address, which may be a symbol or numeric value, __stack is the default.


       -mno-interrupts
           Generated code is not compatible with hardware interrupts.  Code size will be smaller.


       -mcall-prologues
           Functions prologues/epilogues expanded as call to appropriate subroutines.  Code size will be smaller.


       -mno-tablejump
           Do not generate tablejump insns which sometimes increase code size.


       -mtiny-stack
           Change only the low 8 bits of the stack pointer.


       -mint8
           Assume int to be 8 bit integer.  This affects the sizes of all types: A char will be 1 byte, an int will be
           1 byte, an long will be 2 bytes and long long will be 4 bytes.  Please note that this option does not com-
           ply to the C standards, but it will provide you with smaller code size.


       Blackfin Options


       -momit-leaf-frame-pointer
           Don’t keep the frame pointer in a register for leaf functions.  This avoids the instructions to save, set
           up and restore frame pointers and makes an extra register available in leaf functions.  The option
           -fomit-frame-pointer removes the frame pointer for all functions which might make debugging harder.


       -mspecld-anomaly
           When enabled, the compiler will ensure that the generated code does not contain speculative loads after
           jump instructions.  This option is enabled by default.


       -mno-specld-anomaly
           Don’t generate extra code to prevent speculative loads from occurring.


       -mcsync-anomaly
           When enabled, the compiler will ensure that the generated code does not contain CSYNC or SSYNC instructions
           too soon after conditional branches.  This option is enabled by default.


       -mno-csync-anomaly
           Don’t generate extra code to prevent CSYNC or SSYNC instructions from occurring too soon after a condi-
           tional branch.


       -mlow-64k
           When enabled, the compiler is free to take advantage of the knowledge that the entire program fits into the
           low 64k of memory.


       -mno-low-64k
           Assume that the program is arbitrarily large.  This is the default.


       -mid-shared-library
           Generate code that supports shared libraries via the library ID method.  This allows for execute in place
           and shared libraries in an environment without virtual memory management.  This option implies -fPIC.


       -mno-id-shared-library
           Generate code that doesn’t assume ID based shared libraries are being used.  This is the default.


       -mshared-library-id=n
           Specified the identification number of the ID based shared library being compiled.  Specifying a value of 0
           will generate more compact code, specifying other values will force the allocation of that number to the
           current library but is no more space or time efficient than omitting this option.


       -mlong-calls
       -mno-long-calls
           Tells the compiler to perform function calls by first loading the address of the function into a register
           and then performing a subroutine call on this register.  This switch is needed if the target function will
           lie outside of the 24 bit addressing range of the offset based version of subroutine call instruction.


           This feature is not enabled by default.  Specifying -mno-long-calls will restore the default behavior.
           Note these switches have no effect on how the compiler generates code to handle function calls via function
           pointers.


       CRIS Options


       These options are defined specifically for the CRIS ports.


       -march=architecture-type
       -mcpu=architecture-type
           Generate code for the specified architecture.  The choices for architecture-type are v3, v8 and v10 for
           respectively ETRAX 4, ETRAX 100, and ETRAX 100 LX.  Default is v0 except for cris-axis-linux-gnu, where the
           default is v10.


       -mtune=architecture-type
           Tune to architecture-type everything applicable about the generated code, except for the ABI and the set of
           available instructions.  The choices for architecture-type are the same as for -march=architecture-type.


       -mmax-stack-frame=n
           Warn when the stack frame of a function exceeds n bytes.


       -melinux-stacksize=n
           Only available with the cris-axis-aout target.  Arranges for indications in the program to the kernel
           loader that the stack of the program should be set to n bytes.


       -metrax4
       -metrax100
           The options -metrax4 and -metrax100 are synonyms for -march=v3 and -march=v8 respectively.


       -mmul-bug-workaround
       -mno-mul-bug-workaround
           Work around a bug in the "muls" and "mulu" instructions for CPU models where it applies.  This option is
           active by default.


       -mpdebug
           Enable CRIS-specific verbose debug-related information in the assembly code.  This option also has the
           effect to turn off the #NO_APP formatted-code indicator to the assembler at the beginning of the assembly
           file.


       -mcc-init
           Do not use condition-code results from previous instruction; always emit compare and test instructions
           before use of condition codes.


       -mno-side-effects
           Do not emit instructions with side-effects in addressing modes other than post-increment.


       -mstack-align
       -mno-stack-align
       -mdata-align
       -mno-data-align
       -mconst-align
       -mno-const-align
           These options (no-options) arranges (eliminate arrangements) for the stack-frame, individual data and con-
           stants to be aligned for the maximum single data access size for the chosen CPU model.  The default is to
           arrange for 32-bit alignment.  ABI details such as structure layout are not affected by these options.


       -m32-bit
       -m16-bit
       -m8-bit
           Similar to the stack- data- and const-align options above, these options arrange for stack-frame, writable
           data and constants to all be 32-bit, 16-bit or 8-bit aligned.  The default is 32-bit alignment.


       -mno-prologue-epilogue
       -mprologue-epilogue
           With -mno-prologue-epilogue, the normal function prologue and epilogue that sets up the stack-frame are
           omitted and no return instructions or return sequences are generated in the code.  Use this option only
           together with visual inspection of the compiled code: no warnings or errors are generated when call-saved
           registers must be saved, or storage for local variable needs to be allocated.


       -mno-gotplt
       -mgotplt
           With -fpic and -fPIC, don’t generate (do generate) instruction sequences that load addresses for functions
           from the PLT part of the GOT rather than (traditional on other architectures) calls to the PLT.  The
           default is -mgotplt.


       -maout
           Legacy no-op option only recognized with the cris-axis-aout target.


       -melf
           Legacy no-op option only recognized with the cris-axis-elf and cris-axis-linux-gnu targets.


       -melinux
           Only recognized with the cris-axis-aout target, where it selects a GNU/linux-like multilib, include files
           and instruction set for -march=v8.


       -mlinux
           Legacy no-op option only recognized with the cris-axis-linux-gnu target.


       -sim
           This option, recognized for the cris-axis-aout and cris-axis-elf arranges to link with input-output func-
           tions from a simulator library.  Code, initialized data and zero-initialized data are allocated consecu-
           tively.


       -sim2
           Like -sim, but pass linker options to locate initialized data at 0x40000000 and zero-initialized data at
           0x80000000.


       CRX Options


       These options are defined specifically for the CRX ports.


       -mmac
           Enable the use of multiply-accumulate instructions. Disabled by default.


       -mpush-args
           Push instructions will be used to pass outgoing arguments when functions are called. Enabled by default.


       Darwin Options


       These options are defined for all architectures running the Darwin operating system.


       FSF GCC on Darwin does not create "fat" object files; it will create an object file for the single architecture
       that it was built to target.  Apple’s GCC on Darwin does create "fat" files if multiple -arch options are used;
       it does so by running the compiler or linker multiple times and joining the results together with lipo.


       The subtype of the file created (like ppc7400 or ppc970 or i686) is determined by the flags that specify the
       ISA that GCC is targetting, like -mcpu or -march.  The -force_cpusubtype_ALL option can be used to override
       this.


       The Darwin tools vary in their behavior when presented with an ISA mismatch.  The assembler, as, will only per-
       mit instructions to be used that are valid for the subtype of the file it is generating, so you cannot put
       64-bit instructions in an ppc750 object file.  The linker for shared libraries, /usr/bin/libtool, will fail and
       print an error if asked to create a shared library with a less restrictive subtype than its input files (for
       instance, trying to put a ppc970 object file in a ppc7400 library).  The linker for executables, ld, will qui-
       etly give the executable the most restrictive subtype of any of its input files.


       -Fdir
           Add the framework directory dir to the head of the list of directories to be searched for header files.
           These directories are interleaved with those specified by -I options and are scanned in a left-to-right
           order.


           A framework directory is a directory with frameworks in it.  A framework is a directory with a "Headers"
           and/or "PrivateHeaders" directory contained directly in it that ends in ".framework".  The name of a frame-
           work is the name of this directory excluding the ".framework".  Headers associated with the framework are
           found in one of those two directories, with "Headers" being searched first.  A subframework is a framework
           directory that is in a framework’s "Frameworks" directory.  Includes of subframework headers can only
           appear in a header of a framework that contains the subframework, or in a sibling subframework header.  Two
           subframeworks are siblings if they occur in the same framework.  A subframework should not have the same
           name as a framework, a warning will be issued if this is violated.  Currently a subframework cannot have
           subframeworks, in the future, the mechanism may be extended to support this.  The standard frameworks can
           be found in "/System/Library/Frameworks" and "/Library/Frameworks".  An example include looks like
           "#include <Framework/header.h>", where Framework denotes the name of the framework and header.h is found in
           the "PrivateHeaders" or "Headers" directory.


       -gused
           Emit debugging information for symbols that are used.  For STABS debugging format, this enables -felimi-
           nate-unused-debug-symbols.  This is by default ON.


       -gfull
           Emit debugging information for all symbols and types.


       -mmacosx-version-min=version
           The earliest version of MacOS X that this executable will run on is version.  Typical values of version
           include 10.1, 10.2, and 10.3.9.


           The default for this option is to make choices that seem to be most useful.


       -mone-byte-bool
           Override the defaults for bool so that sizeof(bool)==1.  By default sizeof(bool) is 4 when compiling for
           Darwin/PowerPC and 1 when compiling for Darwin/x86, so this option has no effect on x86.


           Warning: The -mone-byte-bool switch causes GCC to generate code that is not binary compatible with code
           generated without that switch.  Using this switch may require recompiling all other modules in a program,
           including system libraries.  Use this switch to conform to a non-default data model.


       -mfix-and-continue
       -ffix-and-continue
       -findirect-data
           Generate code suitable for fast turn around development.  Needed to enable gdb to dynamically load ".o"
           files into already running programs.  -findirect-data and -ffix-and-continue are provided for backwards
           compatibility.


       -all_load
           Loads all members of static archive libraries.  See man ld(1) for more information.


       -arch_errors_fatal
           Cause the errors having to do with files that have the wrong architecture to be fatal.


       -bind_at_load
           Causes the output file to be marked such that the dynamic linker will bind all undefined references when
           the file is loaded or launched.


       -bundle
           Produce a Mach-o bundle format file.  See man ld(1) for more information.


       -bundle_loader executable
           This option specifies the executable that will be loading the build output file being linked.  See man
           ld(1) for more information.


       -dynamiclib
           When passed this option, GCC will produce a dynamic library instead of an executable when linking, using
           the Darwin libtool command.


       -force_cpusubtype_ALL
           This causes GCC’s output file to have the ALL subtype, instead of one controlled by the -mcpu or -march
           option.


       -allowable_client  client_name
       -client_name
       -compatibility_version
       -current_version
       -dead_strip
       -dependency-file
       -dylib_file
       -dylinker_install_name
       -dynamic
       -exported_symbols_list
       -filelist
       -flat_namespace
       -force_flat_namespace
       -headerpad_max_install_names
       -image_base
       -init
       -install_name
       -keep_private_externs
       -multi_module
       -multiply_defined
       -multiply_defined_unused
       -noall_load
       -no_dead_strip_inits_and_terms
       -nofixprebinding
       -nomultidefs
       -noprebind
       -noseglinkedit
       -pagezero_size
       -prebind
       -prebind_all_twolevel_modules
       -private_bundle
       -read_only_relocs
       -sectalign
       -sectobjectsymbols
       -whyload
       -seg1addr
       -sectcreate
       -sectobjectsymbols
       -sectorder
       -segaddr
       -segs_read_only_addr
       -segs_read_write_addr
       -seg_addr_table
       -seg_addr_table_filename
       -seglinkedit
       -segprot
       -segs_read_only_addr
       -segs_read_write_addr
       -single_module
       -static
       -sub_library
       -sub_umbrella
       -twolevel_namespace
       -umbrella
       -undefined
       -unexported_symbols_list
       -weak_reference_mismatches
       -whatsloaded
           These options are passed to the Darwin linker.  The Darwin linker man page describes them in detail.


       DEC Alpha Options


       These -m options are defined for the DEC Alpha implementations:


       -mno-soft-float
       -msoft-float
           Use (do not use) the hardware floating-point instructions for floating-point operations.  When -msoft-float
           is specified, functions in libgcc.a will be used to perform floating-point operations.  Unless they are
           replaced by routines that emulate the floating-point operations, or compiled in such a way as to call such
           emulations routines, these routines will issue floating-point operations.   If you are compiling for an
           Alpha without floating-point operations, you must ensure that the library is built so as not to call them.


           Note that Alpha implementations without floating-point operations are required to have floating-point reg-
           isters.


       -mfp-reg
       -mno-fp-regs
           Generate code that uses (does not use) the floating-point register set.  -mno-fp-regs implies -msoft-float.
           If the floating-point register set is not used, floating point operands are passed in integer registers as
           if they were integers and floating-point results are passed in $0 instead of $f0.  This is a non-standard
           calling sequence, so any function with a floating-point argument or return value called by code compiled
           with -mno-fp-regs must also be compiled with that option.


           A typical use of this option is building a kernel that does not use, and hence need not save and restore,
           any floating-point registers.


       -mieee
           The Alpha architecture implements floating-point hardware optimized for maximum performance.  It is mostly
           compliant with the IEEE floating point standard.  However, for full compliance, software assistance is
           required.  This option generates code fully IEEE compliant code except that the inexact-flag is not main-
           tained (see below).  If this option is turned on, the preprocessor macro "_IEEE_FP" is defined during com-
           pilation.  The resulting code is less efficient but is able to correctly support denormalized numbers and
           exceptional IEEE values such as not-a-number and plus/minus infinity.  Other Alpha compilers call this
           option -ieee_with_no_inexact.


       -mieee-with-inexact
           This is like -mieee except the generated code also maintains the IEEE inexact-flag.  Turning on this option
           causes the generated code to implement fully-compliant IEEE math.  In addition to "_IEEE_FP",
           "_IEEE_FP_EXACT" is defined as a preprocessor macro.  On some Alpha implementations the resulting code may
           execute significantly slower than the code generated by default.  Since there is very little code that
           depends on the inexact-flag, you should normally not specify this option.  Other Alpha compilers call this
           option -ieee_with_inexact.


       -mfp-trap-mode=trap-mode
           This option controls what floating-point related traps are enabled.  Other Alpha compilers call this option
           -fptm trap-mode.  The trap mode can be set to one of four values:


           n   This is the default (normal) setting.  The only traps that are enabled are the ones that cannot be dis-
               abled in software (e.g., division by zero trap).


           u   In addition to the traps enabled by n, underflow traps are enabled as well.


           su  Like su, but the instructions are marked to be safe for software completion (see Alpha architecture
               manual for details).


           sui Like su, but inexact traps are enabled as well.


       -mfp-rounding-mode=rounding-mode
           Selects the IEEE rounding mode.  Other Alpha compilers call this option -fprm rounding-mode.  The rounding-
           mode can be one of:


           n   Normal IEEE rounding mode.  Floating point numbers are rounded towards the nearest machine number or
               towards the even machine number in case of a tie.


           m   Round towards minus infinity.


           c   Chopped rounding mode.  Floating point numbers are rounded towards zero.


           d   Dynamic rounding mode.  A field in the floating point control register (fpcr, see Alpha architecture
               reference manual) controls the rounding mode in effect.  The C library initializes this register for
               rounding towards plus infinity.  Thus, unless your program modifies the fpcr, d corresponds to round
               towards plus infinity.


       -mtrap-precision=trap-precision
           In the Alpha architecture, floating point traps are imprecise.  This means without software assistance it
           is impossible to recover from a floating trap and program execution normally needs to be terminated.  GCC
           can generate code that can assist operating system trap handlers in determining the exact location that
           caused a floating point trap.  Depending on the requirements of an application, different levels of preci-
           sions can be selected:


           p   Program precision.  This option is the default and means a trap handler can only identify which program
               caused a floating point exception.


           f   Function precision.  The trap handler can determine the function that caused a floating point excep-
               tion.


           i   Instruction precision.  The trap handler can determine the exact instruction that caused a floating
               point exception.


           Other Alpha compilers provide the equivalent options called -scope_safe and -resumption_safe.


       -mieee-conformant
           This option marks the generated code as IEEE conformant.  You must not use this option unless you also
           specify -mtrap-precision=i and either -mfp-trap-mode=su or -mfp-trap-mode=sui.  Its only effect is to emit
           the line .eflag 48 in the function prologue of the generated assembly file.  Under DEC Unix, this has the
           effect that IEEE-conformant math library routines will be linked in.


       -mbuild-constants
           Normally GCC examines a 32- or 64-bit integer constant to see if it can construct it from smaller constants
           in two or three instructions.  If it cannot, it will output the constant as a literal and generate code to
           load it from the data segment at runtime.


           Use this option to require GCC to construct all integer constants using code, even if it takes more
           instructions (the maximum is six).


           You would typically use this option to build a shared library dynamic loader.  Itself a shared library, it
           must relocate itself in memory before it can find the variables and constants in its own data segment.


       -malpha-as
       -mgas
           Select whether to generate code to be assembled by the vendor-supplied assembler (-malpha-as) or by the GNU
           assembler -mgas.


       -mbwx
       -mno-bwx
       -mcix
       -mno-cix
       -mfix
       -mno-fix
       -mmax
       -mno-max
           Indicate whether GCC should generate code to use the optional BWX, CIX, FIX and MAX instruction sets.  The
           default is to use the instruction sets supported by the CPU type specified via -mcpu= option or that of the
           CPU on which GCC was built if none was specified.


       -mfloat-vax
       -mfloat-ieee
           Generate code that uses (does not use) VAX F and G floating point arithmetic instead of IEEE single and
           double precision.


       -mexplicit-relocs
       -mno-explicit-relocs
           Older Alpha assemblers provided no way to generate symbol relocations except via assembler macros.  Use of
           these macros does not allow optimal instruction scheduling.  GNU binutils as of version 2.12 supports a new
           syntax that allows the compiler to explicitly mark which relocations should apply to which instructions.
           This option is mostly useful for debugging, as GCC detects the capabilities of the assembler when it is
           built and sets the default accordingly.


       -msmall-data
       -mlarge-data
           When -mexplicit-relocs is in effect, static data is accessed via gp-relative relocations.  When
           -msmall-data is used, objects 8 bytes long or smaller are placed in a small data area (the ".sdata" and
           ".sbss" sections) and are accessed via 16-bit relocations off of the $gp register.  This limits the size of
           the small data area to 64KB, but allows the variables to be directly accessed via a single instruction.


           The default is -mlarge-data.  With this option the data area is limited to just below 2GB.  Programs that
           require more than 2GB of data must use "malloc" or "mmap" to allocate the data in the heap instead of in
           the program’s data segment.


           When generating code for shared libraries, -fpic implies -msmall-data and -fPIC implies -mlarge-data.


       -msmall-text
       -mlarge-text
           When -msmall-text is used, the compiler assumes that the code of the entire program (or shared library)
           fits in 4MB, and is thus reachable with a branch instruction.  When -msmall-data is used, the compiler can
           assume that all local symbols share the same $gp value, and thus reduce the number of instructions required
           for a function call from 4 to 1.


           The default is -mlarge-text.


       -mcpu=cpu_type
           Set the instruction set and instruction scheduling parameters for machine type cpu_type.  You can specify
           either the EV style name or the corresponding chip number.  GCC supports scheduling parameters for the EV4,
           EV5 and EV6 family of processors and will choose the default values for the instruction set from the pro-
           cessor you specify.  If you do not specify a processor type, GCC will default to the processor on which the
           compiler was built.


           Supported values for cpu_type are


           ev4
           ev45
           21064
               Schedules as an EV4 and has no instruction set extensions.


           ev5
           21164
               Schedules as an EV5 and has no instruction set extensions.


           ev56
           21164a
               Schedules as an EV5 and supports the BWX extension.


           pca56
           21164pc
           21164PC
               Schedules as an EV5 and supports the BWX and MAX extensions.


           ev6
           21264
               Schedules as an EV6 and supports the BWX, FIX, and MAX extensions.


           ev67
           21264a
               Schedules as an EV6 and supports the BWX, CIX, FIX, and MAX extensions.


       -mtune=cpu_type
           Set only the instruction scheduling parameters for machine type cpu_type.  The instruction set is not
           changed.


       -mmemory-latency=time
           Sets the latency the scheduler should assume for typical memory references as seen by the application.
           This number is highly dependent on the memory access patterns used by the application and the size of the
           external cache on the machine.


           Valid options for time are


           number
               A decimal number representing clock cycles.


           L1
           L2
           L3
           main
               The compiler contains estimates of the number of clock cycles for "typical" EV4 & EV5 hardware for the
               Level 1, 2 & 3 caches (also called Dcache, Scache, and Bcache), as well as to main memory.  Note that
               L3 is only valid for EV5.


       DEC Alpha/VMS Options


       These -m options are defined for the DEC Alpha/VMS implementations:


       -mvms-return-codes
           Return VMS condition codes from main.  The default is to return POSIX style condition (e.g. error) codes.


       FRV Options


       -mgpr-32
           Only use the first 32 general purpose registers.


       -mgpr-64
           Use all 64 general purpose registers.


       -mfpr-32
           Use only the first 32 floating point registers.


       -mfpr-64
           Use all 64 floating point registers


       -mhard-float
           Use hardware instructions for floating point operations.


       -msoft-float
           Use library routines for floating point operations.


       -malloc-cc
           Dynamically allocate condition code registers.


       -mfixed-cc
           Do not try to dynamically allocate condition code registers, only use "icc0" and "fcc0".


       -mdword
           Change ABI to use double word insns.


       -mno-dword
           Do not use double word instructions.


       -mdouble
           Use floating point double instructions.


       -mno-double
           Do not use floating point double instructions.


       -mmedia
           Use media instructions.


       -mno-media
           Do not use media instructions.


       -mmuladd
           Use multiply and add/subtract instructions.


       -mno-muladd
           Do not use multiply and add/subtract instructions.


       -mfdpic
           Select the FDPIC ABI, that uses function descriptors to represent pointers to functions.  Without any
           PIC/PIE-related options, it implies -fPIE.  With -fpic or -fpie, it assumes GOT entries and small data are
           within a 12-bit range from the GOT base address; with -fPIC or -fPIE, GOT offsets are computed with 32
           bits.


       -minline-plt
           Enable inlining of PLT entries in function calls to functions that are not known to bind locally.  It has
           no effect without -mfdpic.  It’s enabled by default if optimizing for speed and compiling for shared
           libraries (i.e., -fPIC or -fpic), or when an optimization option such as -O3 or above is present in the
           command line.


       -mTLS
           Assume a large TLS segment when generating thread-local code.


       -mtls
           Do not assume a large TLS segment when generating thread-local code.


       -mgprel-ro
           Enable the use of "GPREL" relocations in the FDPIC ABI for data that is known to be in read-only sections.
           It’s enabled by default, except for -fpic or -fpie: even though it may help make the global offset table
           smaller, it trades 1 instruction for 4.  With -fPIC or -fPIE, it trades 3 instructions for 4, one of which
           may be shared by multiple symbols, and it avoids the need for a GOT entry for the referenced symbol, so
           it’s more likely to be a win.  If it is not, -mno-gprel-ro can be used to disable it.


       -multilib-library-pic
           Link with the (library, not FD) pic libraries.  It’s implied by -mlibrary-pic, as well as by -fPIC and
           -fpic without -mfdpic.  You should never have to use it explicitly.


       -mlinked-fp
           Follow the EABI requirement of always creating a frame pointer whenever a stack frame is allocated.  This
           option is enabled by default and can be disabled with -mno-linked-fp.


       -mlong-calls
           Use indirect addressing to call functions outside the current compilation unit.  This allows the functions
           to be placed anywhere within the 32-bit address space.


       -malign-labels
           Try to align labels to an 8-byte boundary by inserting nops into the previous packet.  This option only has
           an effect when VLIW packing is enabled.  It doesn’t create new packets; it merely adds nops to existing
           ones.


       -mlibrary-pic
           Generate position-independent EABI code.


       -macc-4
           Use only the first four media accumulator registers.


       -macc-8
           Use all eight media accumulator registers.


       -mpack
           Pack VLIW instructions.


       -mno-pack
           Do not pack VLIW instructions.


       -mno-eflags
           Do not mark ABI switches in e_flags.


       -mcond-move
           Enable the use of conditional-move instructions (default).


           This switch is mainly for debugging the compiler and will likely be removed in a future version.


       -mno-cond-move
           Disable the use of conditional-move instructions.


           This switch is mainly for debugging the compiler and will likely be removed in a future version.


       -mscc
           Enable the use of conditional set instructions (default).


           This switch is mainly for debugging the compiler and will likely be removed in a future version.


       -mno-scc
           Disable the use of conditional set instructions.


           This switch is mainly for debugging the compiler and will likely be removed in a future version.


       -mcond-exec
           Enable the use of conditional execution (default).


           This switch is mainly for debugging the compiler and will likely be removed in a future version.


       -mno-cond-exec
           Disable the use of conditional execution.


           This switch is mainly for debugging the compiler and will likely be removed in a future version.


       -mvliw-branch
           Run a pass to pack branches into VLIW instructions (default).


           This switch is mainly for debugging the compiler and will likely be removed in a future version.


       -mno-vliw-branch
           Do not run a pass to pack branches into VLIW instructions.


           This switch is mainly for debugging the compiler and will likely be removed in a future version.


       -mmulti-cond-exec
           Enable optimization of "&&" and "││" in conditional execution (default).


           This switch is mainly for debugging the compiler and will likely be removed in a future version.


       -mno-multi-cond-exec
           Disable optimization of "&&" and "││" in conditional execution.


           This switch is mainly for debugging the compiler and will likely be removed in a future version.


       -mnested-cond-exec
           Enable nested conditional execution optimizations (default).


           This switch is mainly for debugging the compiler and will likely be removed in a future version.


       -mno-nested-cond-exec
           Disable nested conditional execution optimizations.


           This switch is mainly for debugging the compiler and will likely be removed in a future version.


       -moptimize-membar
           This switch removes redundant "membar" instructions from the compiler generated code.  It is enabled by
           default.


       -mno-optimize-membar
           This switch disables the automatic removal of redundant "membar" instructions from the generated code.


       -mtomcat-stats
           Cause gas to print out tomcat statistics.


       -mcpu=cpu
           Select the processor type for which to generate code.  Possible values are frv, fr550, tomcat, fr500,
           fr450, fr405, fr400, fr300 and simple.


       H8/300 Options


       These -m options are defined for the H8/300 implementations:


       -mrelax
           Shorten some address references at link time, when possible; uses the linker option -relax.


       -mh Generate code for the H8/300H.


       -ms Generate code for the H8S.


       -mn Generate code for the H8S and H8/300H in the normal mode.  This switch must be used either with -mh or -ms.


       -ms2600
           Generate code for the H8S/2600.  This switch must be used with -ms.


       -mint32
           Make "int" data 32 bits by default.


       -malign-300
           On the H8/300H and H8S, use the same alignment rules as for the H8/300.  The default for the H8/300H and
           H8S is to align longs and floats on 4 byte boundaries.  -malign-300 causes them to be aligned on 2 byte
           boundaries.  This option has no effect on the H8/300.


       HPPA Options


       These -m options are defined for the HPPA family of computers:


       -march=architecture-type
           Generate code for the specified architecture.  The choices for architecture-type are 1.0 for PA 1.0, 1.1
           for PA 1.1, and 2.0 for PA 2.0 processors.  Refer to /usr/lib/sched.models on an HP-UX system to determine
           the proper architecture option for your machine.  Code compiled for lower numbered architectures will run
           on higher numbered architectures, but not the other way around.


       -mpa-risc-1-0
       -mpa-risc-1-1
       -mpa-risc-2-0
           Synonyms for -march=1.0, -march=1.1, and -march=2.0 respectively.


       -mbig-switch
           Generate code suitable for big switch tables.  Use this option only if the assembler/linker complain about
           out of range branches within a switch table.


       -mjump-in-delay
           Fill delay slots of function calls with unconditional jump instructions by modifying the return pointer for
           the function call to be the target of the conditional jump.


       -mdisable-fpregs
           Prevent floating point registers from being used in any manner.  This is necessary for compiling kernels
           which perform lazy context switching of floating point registers.  If you use this option and attempt to
           perform floating point operations, the compiler will abort.


       -mdisable-indexing
           Prevent the compiler from using indexing address modes.  This avoids some rather obscure problems when com-
           piling MIG generated code under MACH.


       -mno-space-regs
           Generate code that assumes the target has no space registers.  This allows GCC to generate faster indirect
           calls and use unscaled index address modes.


           Such code is suitable for level 0 PA systems and kernels.


       -mfast-indirect-calls
           Generate code that assumes calls never cross space boundaries.  This allows GCC to emit code which performs
           faster indirect calls.


           This option will not work in the presence of shared libraries or nested functions.


       -mfixed-range=register-range
           Generate code treating the given register range as fixed registers.  A fixed register is one that the reg-
           ister allocator can not use.  This is useful when compiling kernel code.  A register range is specified as
           two registers separated by a dash.  Multiple register ranges can be specified separated by a comma.


       -mlong-load-store
           Generate 3-instruction load and store sequences as sometimes required by the HP-UX 10 linker.  This is
           equivalent to the +k option to the HP compilers.


       -mportable-runtime
           Use the portable calling conventions proposed by HP for ELF systems.


       -mgas
           Enable the use of assembler directives only GAS understands.


       -mschedule=cpu-type
           Schedule code according to the constraints for the machine type cpu-type.  The choices for cpu-type are 700
           7100, 7100LC, 7200, 7300 and 8000.  Refer to /usr/lib/sched.models on an HP-UX system to determine the
           proper scheduling option for your machine.  The default scheduling is 8000.


       -mlinker-opt
           Enable the optimization pass in the HP-UX linker.  Note this makes symbolic debugging impossible.  It also
           triggers a bug in the HP-UX 8 and HP-UX 9 linkers in which they give bogus error messages when linking some
           programs.


       -msoft-float
           Generate output containing library calls for floating point.  Warning: the requisite libraries are not
           available for all HPPA targets.  Normally the facilities of the machine’s usual C compiler are used, but
           this cannot be done directly in cross-compilation.  You must make your own arrangements to provide suitable
           library functions for cross-compilation.  The embedded target hppa1.1-*-pro does provide software floating
           point support.


           -msoft-float changes the calling convention in the output file; therefore, it is only useful if you compile
           all of a program with this option.  In particular, you need to compile libgcc.a, the library that comes
           with GCC, with -msoft-float in order for this to work.


       -msio
           Generate the predefine, "_SIO", for server IO.  The default is -mwsio.  This generates the predefines,
           "__hp9000s700", "__hp9000s700__" and "_WSIO", for workstation IO.  These options are available under HP-UX
           and HI-UX.


       -mgnu-ld
           Use GNU ld specific options.  This passes -shared to ld when building a shared library.  It is the default
           when GCC is configured, explicitly or implicitly, with the GNU linker.  This option does not have any
           affect on which ld is called, it only changes what parameters are passed to that ld.  The ld that is called
           is determined by the --with-ld configure option, GCC’s program search path, and finally by the user’s PATH.
           The linker used by GCC can be printed using which ‘gcc -print-prog-name=ld‘.  This option is only available
           on the 64 bit HP-UX GCC, i.e. configured with hppa*64*-*-hpux*.


       -mhp-ld
           Use HP ld specific options.  This passes -b to ld when building a shared library and passes +Accept
           TypeMismatch to ld on all links.  It is the default when GCC is configured, explicitly or implicitly, with
           the HP linker.  This option does not have any affect on which ld is called, it only changes what parameters
           are passed to that ld.  The ld that is called is determined by the --with-ld configure option, GCC’s pro-
           gram search path, and finally by the user’s PATH.  The linker used by GCC can be printed using which ‘gcc
           -print-prog-name=ld‘.  This option is only available on the 64 bit HP-UX GCC, i.e. configured with
           hppa*64*-*-hpux*.


       -mlong-calls
           Generate code that uses long call sequences.  This ensures that a call is always able to reach linker gen-
           erated stubs.  The default is to generate long calls only when the distance from the call site to the
           beginning of the function or translation unit, as the case may be, exceeds a predefined limit set by the
           branch type being used.  The limits for normal calls are 7,600,000 and 240,000 bytes, respectively for the
           PA 2.0 and PA 1.X architectures.  Sibcalls are always limited at 240,000 bytes.


           Distances are measured from the beginning of functions when using the -ffunction-sections option, or when
           using the -mgas and -mno-portable-runtime options together under HP-UX with the SOM linker.


           It is normally not desirable to use this option as it will degrade performance.  However, it may be useful
           in large applications, particularly when partial linking is used to build the application.


           The types of long calls used depends on the capabilities of the assembler and linker, and the type of code
           being generated.  The impact on systems that support long absolute calls, and long pic symbol-difference or
           pc-relative calls should be relatively small.  However, an indirect call is used on 32-bit ELF systems in
           pic code and it is quite long.


       -munix=unix-std
           Generate compiler predefines and select a startfile for the specified UNIX standard.  The choices for unix-
           std are 93, 95 and 98.  93 is supported on all HP-UX versions.  95 is available on HP-UX 10.10 and later.
           98 is available on HP-UX 11.11 and later.  The default values are 93 for HP-UX 10.00, 95 for HP-UX 10.10
           though to 11.00, and 98 for HP-UX 11.11 and later.


           -munix=93 provides the same predefines as GCC 3.3 and 3.4.  -munix=95 provides additional predefines for
           "XOPEN_UNIX" and "_XOPEN_SOURCE_EXTENDED", and the startfile unix95.o.  -munix=98 provides additional pre-
           defines for "_XOPEN_UNIX", "_XOPEN_SOURCE_EXTENDED", "_INCLUDE__STDC_A1_SOURCE" and
           "_INCLUDE_XOPEN_SOURCE_500", and the startfile unix98.o.


           It is important to note that this option changes the interfaces for various library routines.  It also
           affects the operational behavior of the C library.  Thus, extreme care is needed in using this option.


           Library code that is intended to operate with more than one UNIX standard must test, set and restore the
           variable __xpg4_extended_mask as appropriate.  Most GNU software doesn’t provide this capability.


       -nolibdld
           Suppress the generation of link options to search libdld.sl when the -static option is specified on HP-UX
           10 and later.


       -static
           The HP-UX implementation of setlocale in libc has a dependency on libdld.sl.  There isn’t an archive ver-
           sion of libdld.sl.  Thus, when the -static option is specified, special link options are needed to resolve
           this dependency.


           On HP-UX 10 and later, the GCC driver adds the necessary options to link with libdld.sl when the -static
           option is specified.  This causes the resulting binary to be dynamic.  On the 64-bit port, the linkers gen-
           erate dynamic binaries by default in any case.  The -nolibdld option can be used to prevent the GCC driver
           from adding these link options.


       -threads
           Add support for multithreading with the dce thread library under HP-UX.  This option sets flags for both
           the preprocessor and linker.


       Intel 386 and AMD x86-64 Options


       These -m options are defined for the i386 and x86-64 family of computers:


       -mtune=cpu-type
           Tune to cpu-type everything applicable about the generated code, except for the ABI and the set of avail-
           able instructions.  The choices for cpu-type are:


           generic
               Produce code optimized for the most common IA32/AMD64/EM64T processors.  If you know the CPU on which
               your code will run, then you should use the corresponding -mtune option instead of -mtune=generic.
               But, if you do not know exactly what CPU users of your application will have, then you should use this
               option.


               As new processors are deployed in the marketplace, the behavior of this option will change.  Therefore,
               if you upgrade to a newer version of GCC, the code generated option will change to reflect the proces-
               sors that were most common when that version of GCC was released.


               There is no -march=generic option because -march indicates the instruction set the compiler can use,
               and there is no generic instruction set applicable to all processors.  In contrast, -mtune indicates
               the processor (or, in this case, collection of processors) for which the code is optimized.


           i386
               Original Intel’s i386 CPU.


           i486
               Intel’s i486 CPU.  (No scheduling is implemented for this chip.)


           i586, pentium
               Intel Pentium CPU with no MMX support.


           pentium-mmx
               Intel PentiumMMX CPU based on Pentium core with MMX instruction set support.


           pentiumpro
               Intel PentiumPro CPU.


           i686
               Same as "generic", but when used as "march" option, PentiumPro instruction set will be used, so the
               code will run on all i686 familly chips.


           pentium2
               Intel Pentium2 CPU based on PentiumPro core with MMX instruction set support.


           pentium3, pentium3m
               Intel Pentium3 CPU based on PentiumPro core with MMX and SSE instruction set support.


           pentium-m
               Low power version of Intel Pentium3 CPU with MMX, SSE and SSE2 instruction set support.  Used by Cen-
               trino notebooks.


           pentium4, pentium4m
               Intel Pentium4 CPU with MMX, SSE and SSE2 instruction set support.


           prescott
               Improved version of Intel Pentium4 CPU with MMX, SSE, SSE2 and SSE3 instruction set support.


           nocona
               Improved version of Intel Pentium4 CPU with 64-bit extensions, MMX, SSE, SSE2 and SSE3 instruction set
               support.


           core2
               Intel Core2 CPU with 64-bit extensions, MMX, SSE, SSE2, SSE3 and SSSE3 instruction set support.


           k6  AMD K6 CPU with MMX instruction set support.


           k6-2, k6-3
               Improved versions of AMD K6 CPU with MMX and 3dNOW! instruction set support.


           athlon, athlon-tbird
               AMD Athlon CPU with MMX, 3dNOW!, enhanced 3dNOW! and SSE prefetch instructions support.


           athlon-4, athlon-xp, athlon-mp
               Improved AMD Athlon CPU with MMX, 3dNOW!, enhanced 3dNOW! and full SSE instruction set support.


           k8, opteron, athlon64, athlon-fx
               AMD K8 core based CPUs with x86-64 instruction set support.  (This supersets MMX, SSE, SSE2, 3dNOW!,
               enhanced 3dNOW! and 64-bit instruction set extensions.)


           amdfam10
               AMD Family 10 core based CPUs with x86-64 instruction set support.  (This supersets MMX, SSE, SSE2,
               SSE3, SSE4A, SSE5, 3dNOW!, enhanced 3dNOW!, ABM and 64-bit instruction set extensions.)


           winchip-c6
               IDT Winchip C6 CPU, dealt in same way as i486 with additional MMX instruction set support.


           winchip2
               IDT Winchip2 CPU, dealt in same way as i486 with additional MMX and 3dNOW!  instruction set support.


           c3  Via C3 CPU with MMX and 3dNOW! instruction set support.  (No scheduling is implemented for this chip.)


           c3-2
               Via C3-2 CPU with MMX and SSE instruction set support.  (No scheduling is implemented for this chip.)


           While picking a specific cpu-type will schedule things appropriately for that particular chip, the compiler
           will not generate any code that does not run on the i386 without the -march=cpu-type option being used.


       -march=cpu-type
           Generate instructions for the machine type cpu-type.  The choices for cpu-type are the same as for -mtune.
           Moreover, specifying -march=cpu-type implies -mtune=cpu-type.


       -mcpu=cpu-type
           A deprecated synonym for -mtune.


       -m386
       -m486
       -mpentium
       -mpentiumpro
           These options are synonyms for -mtune=i386, -mtune=i486, -mtune=pentium, and -mtune=pentiumpro respec-
           tively.  These synonyms are deprecated.


       -mfpmath=unit
           Generate floating point arithmetics for selected unit unit.  The choices for unit are:


           387 Use the standard 387 floating point coprocessor present majority of chips and emulated otherwise.  Code
               compiled with this option will run almost everywhere.  The temporary results are computed in 80bit pre-
               cision instead of precision specified by the type resulting in slightly different results compared to
               most of other chips.  See -ffloat-store for more detailed description.


               This is the default choice for i386 compiler.


           sse Use scalar floating point instructions present in the SSE instruction set.  This instruction set is
               supported by Pentium3 and newer chips, in the AMD line by Athlon-4, Athlon-xp and Athlon-mp chips.  The
               earlier version of SSE instruction set supports only single precision arithmetics, thus the double and
               extended precision arithmetics is still done using 387.  Later version, present only in Pentium4 and
               the future AMD x86-64 chips supports double precision arithmetics too.


               For the i386 compiler, you need to use -march=cpu-type, -msse or -msse2 switches to enable SSE exten-
               sions and make this option effective.  For the x86-64 compiler, these extensions are enabled by
               default.


               The resulting code should be considerably faster in the majority of cases and avoid the numerical
               instability problems of 387 code, but may break some existing code that expects temporaries to be
               80bit.


               This is the default choice for the x86-64 compiler.


           sse,387
               Attempt to utilize both instruction sets at once.  This effectively double the amount of available reg-
               isters and on chips with separate execution units for 387 and SSE the execution resources too.  Use
               this option with care, as it is still experimental, because the GCC register allocator does not model
               separate functional units well resulting in instable performance.


       -masm=dialect
           Output asm instructions using selected dialect.  Supported choices are intel or att (the default one).
           Darwin does not support intel.


       -mieee-fp
       -mno-ieee-fp
           Control whether or not the compiler uses IEEE floating point comparisons.  These handle correctly the case
           where the result of a comparison is unordered.


       -msoft-float
           Generate output containing library calls for floating point.  Warning: the requisite libraries are not part
           of GCC.  Normally the facilities of the machine’s usual C compiler are used, but this can’t be done
           directly in cross-compilation.  You must make your own arrangements to provide suitable library functions
           for cross-compilation.


           On machines where a function returns floating point results in the 80387 register stack, some floating
           point opcodes may be emitted even if -msoft-float is used.


       -mno-fp-ret-in-387
           Do not use the FPU registers for return values of functions.


           The usual calling convention has functions return values of types "float" and "double" in an FPU register,
           even if there is no FPU.  The idea is that the operating system should emulate an FPU.


           The option -mno-fp-ret-in-387 causes such values to be returned in ordinary CPU registers instead.


       -mno-fancy-math-387
           Some 387 emulators do not support the "sin", "cos" and "sqrt" instructions for the 387.  Specify this
           option to avoid generating those instructions.  This option is the default on FreeBSD, OpenBSD and NetBSD.
           This option is overridden when -march indicates that the target cpu will always have an FPU and so the
           instruction will not need emulation.  As of revision 2.6.1, these instructions are not generated unless you
           also use the -funsafe-math-optimizations switch.


       -malign-double
       -mno-align-double
           Control whether GCC aligns "double", "long double", and "long long" variables on a two word boundary or a
           one word boundary.  Aligning "double" variables on a two word boundary will produce code that runs somewhat
           faster on a Pentium at the expense of more memory.


           On x86-64, -malign-double is enabled by default.


           Warning: if you use the -malign-double switch, structures containing the above types will be aligned dif-
           ferently than the published application binary interface specifications for the 386 and will not be binary
           compatible with structures in code compiled without that switch.


       -m96bit-long-double
       -m128bit-long-double
           These switches control the size of "long double" type.  The i386 application binary interface specifies the
           size to be 96 bits, so -m96bit-long-double is the default in 32 bit mode.


           Modern architectures (Pentium and newer) would prefer "long double" to be aligned to an 8 or 16 byte bound-
           ary.  In arrays or structures conforming to the ABI, this would not be possible.  So specifying a
           -m128bit-long-double will align "long double" to a 16 byte boundary by padding the "long double" with an
           additional 32 bit zero.


           In the x86-64 compiler, -m128bit-long-double is the default choice as its ABI specifies that "long double"
           is to be aligned on 16 byte boundary.


           Notice that neither of these options enable any extra precision over the x87 standard of 80 bits for a
           "long double".


           Warning: if you override the default value for your target ABI, the structures and arrays containing "long
           double" variables will change their size as well as function calling convention for function taking "long
           double" will be modified.  Hence they will not be binary compatible with arrays or structures in code com-
           piled without that switch.


       -mmlarge-data-threshold=number
           When -mcmodel=medium is specified, the data greater than threshold are placed in large data section.  This
           value must be the same across all object linked into the binary and defaults to 65535.


       -msvr3-shlib
       -mno-svr3-shlib
           Control whether GCC places uninitialized local variables into the "bss" or "data" segments.  -msvr3-shlib
           places them into "bss".  These options are meaningful only on System V Release 3.


       -mrtd
           Use a different function-calling convention, in which functions that take a fixed number of arguments
           return with the "ret" num instruction, which pops their arguments while returning.  This saves one instruc-
           tion in the caller since there is no need to pop the arguments there.


           You can specify that an individual function is called with this calling sequence with the function
           attribute stdcall.  You can also override the -mrtd option by using the function attribute cdecl.


           Warning: this calling convention is incompatible with the one normally used on Unix, so you cannot use it
           if you need to call libraries compiled with the Unix compiler.


           Also, you must provide function prototypes for all functions that take variable numbers of arguments
           (including "printf"); otherwise incorrect code will be generated for calls to those functions.


           In addition, seriously incorrect code will result if you call a function with too many arguments.  (Nor-
           mally, extra arguments are harmlessly ignored.)


       -mregparm=num
           Control how many registers are used to pass integer arguments.  By default, no registers are used to pass
           arguments, and at most 3 registers can be used.  You can control this behavior for a specific function by
           using the function attribute regparm.


           Warning: if you use this switch, and num is nonzero, then you must build all modules with the same value,
           including any libraries.  This includes the system libraries and startup modules.


       -msseregparm
           Use SSE register passing conventions for float and double arguments and return values.  You can control
           this behavior for a specific function by using the function attribute sseregparm.


           Warning: if you use this switch then you must build all modules with the same value, including any
           libraries.  This includes the system libraries and startup modules.


       -mpreferred-stack-boundary=num
           Attempt to keep the stack boundary aligned to a 2 raised to num byte boundary.  If -mpreferred-stack-bound-
           ary is not specified, the default is 4 (16 bytes or 128 bits).


           On Pentium and PentiumPro, "double" and "long double" values should be aligned to an 8 byte boundary (see
           -malign-double) or suffer significant run time performance penalties.  On Pentium III, the Streaming SIMD
           Extension (SSE) data type "__m128" may not work properly if it is not 16 byte aligned.


           To ensure proper alignment of this values on the stack, the stack boundary must be as aligned as that
           required by any value stored on the stack.  Further, every function must be generated such that it keeps
           the stack aligned.  Thus calling a function compiled with a higher preferred stack boundary from a function
           compiled with a lower preferred stack boundary will most likely misalign the stack.  It is recommended that
           libraries that use callbacks always use the default setting.


           This extra alignment does consume extra stack space, and generally increases code size.  Code that is sen-
           sitive to stack space usage, such as embedded systems and operating system kernels, may want to reduce the
           preferred alignment to -mpreferred-stack-boundary=2.


       -mmmx
       -mno-mmx
       -msse
       -mno-sse
       -msse2
       -mno-sse2
       -msse3
       -mno-sse3
       -mssse3
       -mno-ssse3
       -msse4a
       -mno-sse4a
       -msse5
       -mno-sse5
       -m3dnow
       -mno-3dnow
       -mpopcnt
       -mno-popcnt
       -mabm
       -mno-abm
           These switches enable or disable the use of instructions in the MMX, SSE, SSE2 or 3DNow! extended instruc-
           tion sets.  These extensions are also available as built-in functions: see X86 Built-in Functions, for
           details of the functions enabled and disabled by these switches.


           To have SSE/SSE2 instructions generated automatically from floating-point code (as opposed to 387 instruc-
           tions), see -mfpmath=sse.


           These options will enable GCC to use these extended instructions in generated code, even without -mfp-
           math=sse.  Applications which perform runtime CPU detection must compile separate files for each supported
           architecture, using the appropriate flags.  In particular, the file containing the CPU detection code
           should be compiled without these options.


       -mpush-args
       -mno-push-args
           Use PUSH operations to store outgoing parameters.  This method is shorter and usually equally fast as
           method using SUB/MOV operations and is enabled by default.  In some cases disabling it may improve perfor-
           mance because of improved scheduling and reduced dependencies.


       -maccumulate-outgoing-args
           If enabled, the maximum amount of space required for outgoing arguments will be computed in the function
           prologue.  This is faster on most modern CPUs because of reduced dependencies, improved scheduling and
           reduced stack usage when preferred stack boundary is not equal to 2.  The drawback is a notable increase in
           code size.  This switch implies -mno-push-args.


       -mthreads
           Support thread-safe exception handling on Mingw32.  Code that relies on thread-safe exception handling must
           compile and link all code with the -mthreads option.  When compiling, -mthreads defines -D_MT; when link-
           ing, it links in a special thread helper library -lmingwthrd which cleans up per thread exception handling
           data.


       -mno-align-stringops
           Do not align destination of inlined string operations.  This switch reduces code size and improves perfor-
           mance in case the destination is already aligned, but GCC doesn’t know about it.


       -minline-all-stringops
           By default GCC inlines string operations only when destination is known to be aligned at least to 4 byte
           boundary.  This enables more inlining, increase code size, but may improve performance of code that depends
           on fast memcpy, strlen and memset for short lengths.


       -momit-leaf-frame-pointer
           Don’t keep the frame pointer in a register for leaf functions.  This avoids the instructions to save, set
           up and restore frame pointers and makes an extra register available in leaf functions.  The option
           -fomit-frame-pointer removes the frame pointer for all functions which might make debugging harder.


       -mtls-direct-seg-refs
       -mno-tls-direct-seg-refs
           Controls whether TLS variables may be accessed with offsets from the TLS segment register (%gs for 32-bit,
           %fs for 64-bit), or whether the thread base pointer must be added.  Whether or not this is legal depends on
           the operating system, and whether it maps the segment to cover the entire TLS area.


           For systems that use GNU libc, the default is on.


       These -m switches are supported in addition to the above on AMD x86-64 processors in 64-bit environments.


       -m32
       -m64
           Generate code for a 32-bit or 64-bit environment.  The 32-bit environment sets int, long and pointer to 32
           bits and generates code that runs on any i386 system.  The 64-bit environment sets int to 32 bits and long
           and pointer to 64 bits and generates code for AMD’s x86-64 architecture.


       -mno-red-zone
           Do not use a so called red zone for x86-64 code.  The red zone is mandated by the x86-64 ABI, it is a
           128-byte area beyond the location of the stack pointer that will not be modified by signal or interrupt
           handlers and therefore can be used for temporary data without adjusting the stack pointer.  The flag
           -mno-red-zone disables this red zone.


       -mcmodel=small
           Generate code for the small code model: the program and its symbols must be linked in the lower 2 GB of the
           address space.  Pointers are 64 bits.  Programs can be statically or dynamically linked.  This is the
           default code model.


       -mcmodel=kernel
           Generate code for the kernel code model.  The kernel runs in the negative 2 GB of the address space.  This
           model has to be used for Linux kernel code.


       -mcmodel=medium
           Generate code for the medium model: The program is linked in the lower 2 GB of the address space but sym-
           bols can be located anywhere in the address space.  Programs can be statically or dynamically linked, but
           building of shared libraries are not supported with the medium model.


       -mcmodel=large
           Generate code for the large model: This model makes no assumptions about addresses and sizes of sections.
           Currently GCC does not implement this model.


       -mfused-madd
       -mno-fused-madd
           Enable automatic generation of fused floating point multiply-add instructions if the ISA supports such
           instructions.  The -mfused-madd option is on by default.


       IA-64 Options


       These are the -m options defined for the Intel IA-64 architecture.


       -mbig-endian
           Generate code for a big endian target.  This is the default for HP-UX.


       -mlittle-endian
           Generate code for a little endian target.  This is the default for AIX5 and GNU/Linux.


       -mgnu-as
       -mno-gnu-as
           Generate (or don’t) code for the GNU assembler.  This is the default.


       -mgnu-ld
       -mno-gnu-ld
           Generate (or don’t) code for the GNU linker.  This is the default.


       -mno-pic
           Generate code that does not use a global pointer register.  The result is not position independent code,
           and violates the IA-64 ABI.


       -mvolatile-asm-stop
       -mno-volatile-asm-stop
           Generate (or don’t) a stop bit immediately before and after volatile asm statements.


       -mregister-names
       -mno-register-names
           Generate (or don’t) in, loc, and out register names for the stacked registers.  This may make assembler
           output more readable.


       -mno-sdata
       -msdata
           Disable (or enable) optimizations that use the small data section.  This may be useful for working around
           optimizer bugs.


       -mconstant-gp
           Generate code that uses a single constant global pointer value.  This is useful when compiling kernel code.


       -mauto-pic
           Generate code that is self-relocatable.  This implies -mconstant-gp.  This is useful when compiling
           firmware code.


       -minline-float-divide-min-latency
           Generate code for inline divides of floating point values using the minimum latency algorithm.


       -minline-float-divide-max-throughput
           Generate code for inline divides of floating point values using the maximum throughput algorithm.


       -minline-int-divide-min-latency
           Generate code for inline divides of integer values using the minimum latency algorithm.


       -minline-int-divide-max-throughput
           Generate code for inline divides of integer values using the maximum throughput algorithm.


       -minline-sqrt-min-latency
           Generate code for inline square roots using the minimum latency algorithm.


       -minline-sqrt-max-throughput
           Generate code for inline square roots using the maximum throughput algorithm.


       -mno-dwarf2-asm
       -mdwarf2-asm
           Don’t (or do) generate assembler code for the DWARF2 line number debugging info.  This may be useful when
           not using the GNU assembler.


       -mearly-stop-bits
       -mno-early-stop-bits
           Allow stop bits to be placed earlier than immediately preceding the instruction that triggered the stop
           bit.  This can improve instruction scheduling, but does not always do so.


       -mfixed-range=register-range
           Generate code treating the given register range as fixed registers.  A fixed register is one that the reg-
           ister allocator can not use.  This is useful when compiling kernel code.  A register range is specified as
           two registers separated by a dash.  Multiple register ranges can be specified separated by a comma.


       -mtls-size=tls-size
           Specify bit size of immediate TLS offsets.  Valid values are 14, 22, and 64.


       -mtune=cpu-type
           Tune the instruction scheduling for a particular CPU, Valid values are itanium, itanium1, merced, itanium2,
           and mckinley.


       -mt
       -pthread
           Add support for multithreading using the POSIX threads library.  This option sets flags for both the pre-
           processor and linker.  It does not affect the thread safety of object code produced by the compiler or that
           of libraries supplied with it.  These are HP-UX specific flags.


       -milp32
       -mlp64
           Generate code for a 32-bit or 64-bit environment.  The 32-bit environment sets int, long and pointer to 32
           bits.  The 64-bit environment sets int to 32 bits and long and pointer to 64 bits.  These are HP-UX spe-
           cific flags.


       M32C Options


       -mcpu=name
           Select the CPU for which code is generated.  name may be one of r8c for the R8C/Tiny series, m16c for the
           M16C (up to /60) series, m32cm for the M16C/80 series, or m32c for the M32C/80 series.


       -msim
           Specifies that the program will be run on the simulator.  This causes an alternate runtime library to be
           linked in which supports, for example, file I/O.  You must not use this option when generating programs
           that will run on real hardware; you must provide your own runtime library for whatever I/O functions are
           needed.


       -memregs=number
           Specifies the number of memory-based pseudo-registers GCC will use during code generation.  These pseudo-
           registers will be used like real registers, so there is a tradeoff between GCC’s ability to fit the code
           into available registers, and the performance penalty of using memory instead of registers.  Note that all
           modules in a program must be compiled with the same value for this option.  Because of that, you must not
           use this option with the default runtime libraries gcc builds.


       M32R/D Options


       These -m options are defined for Renesas M32R/D architectures:


       -m32r2
           Generate code for the M32R/2.


       -m32rx
           Generate code for the M32R/X.


       -m32r
           Generate code for the M32R.  This is the default.


       -mmodel=small
           Assume all objects live in the lower 16MB of memory (so that their addresses can be loaded with the "ld24"
           instruction), and assume all subroutines are reachable with the "bl" instruction.  This is the default.


           The addressability of a particular object can be set with the "model" attribute.


       -mmodel=medium
           Assume objects may be anywhere in the 32-bit address space (the compiler will generate "seth/add3" instruc-
           tions to load their addresses), and assume all subroutines are reachable with the "bl" instruction.


       -mmodel=large
           Assume objects may be anywhere in the 32-bit address space (the compiler will generate "seth/add3" instruc-
           tions to load their addresses), and assume subroutines may not be reachable with the "bl" instruction (the
           compiler will generate the much slower "seth/add3/jl" instruction sequence).


       -msdata=none
           Disable use of the small data area.  Variables will be put into one of .data, bss, or .rodata (unless the
           "section" attribute has been specified).  This is the default.


           The small data area consists of sections .sdata and .sbss.  Objects may be explicitly put in the small data
           area with the "section" attribute using one of these sections.


       -msdata=sdata
           Put small global and static data in the small data area, but do not generate special code to reference
           them.


       -msdata=use
           Put small global and static data in the small data area, and generate special instructions to reference
           them.


       -G num
           Put global and static objects less than or equal to num bytes into the small data or bss sections instead
           of the normal data or bss sections.  The default value of num is 8.  The -msdata option must be set to one
           of sdata or use for this option to have any effect.


           All modules should be compiled with the same -G num value.  Compiling with different values of num may or
           may not work; if it doesn’t the linker will give an error message---incorrect code will not be generated.


       -mdebug
           Makes the M32R specific code in the compiler display some statistics that might help in debugging programs.


       -malign-loops
           Align all loops to a 32-byte boundary.


       -mno-align-loops
           Do not enforce a 32-byte alignment for loops.  This is the default.


       -missue-rate=number
           Issue number instructions per cycle.  number can only be 1 or 2.


       -mbranch-cost=number
           number can only be 1 or 2.  If it is 1 then branches will be preferred over conditional code, if it is 2,
           then the opposite will apply.


       -mflush-trap=number
           Specifies the trap number to use to flush the cache.  The default is 12.  Valid numbers are between 0 and
           15 inclusive.


       -mno-flush-trap
           Specifies that the cache cannot be flushed by using a trap.


       -mflush-func=name
           Specifies the name of the operating system function to call to flush the cache.  The default is
           _flush_cache, but a function call will only be used if a trap is not available.


       -mno-flush-func
           Indicates that there is no OS function for flushing the cache.


       M680x0 Options


       These are the -m options defined for the 68000 series.  The default values for these options depends on which
       style of 68000 was selected when the compiler was configured; the defaults for the most common choices are
       given below.


       -m68000
       -mc68000
           Generate output for a 68000.  This is the default when the compiler is configured for 68000-based systems.


           Use this option for microcontrollers with a 68000 or EC000 core, including the 68008, 68302, 68306, 68307,
           68322, 68328 and 68356.


       -m68020
       -mc68020
           Generate output for a 68020.  This is the default when the compiler is configured for 68020-based systems.


       -m68881
           Generate output containing 68881 instructions for floating point.  This is the default for most 68020 sys-
           tems unless --nfp was specified when the compiler was configured.


       -m68030
           Generate output for a 68030.  This is the default when the compiler is configured for 68030-based systems.


       -m68040
           Generate output for a 68040.  This is the default when the compiler is configured for 68040-based systems.


           This option inhibits the use of 68881/68882 instructions that have to be emulated by software on the 68040.
           Use this option if your 68040 does not have code to emulate those instructions.


       -m68060
           Generate output for a 68060.  This is the default when the compiler is configured for 68060-based systems.


           This option inhibits the use of 68020 and 68881/68882 instructions that have to be emulated by software on
           the 68060.  Use this option if your 68060 does not have code to emulate those instructions.


       -mcpu32
           Generate output for a CPU32.  This is the default when the compiler is configured for CPU32-based systems.


           Use this option for microcontrollers with a CPU32 or CPU32+ core, including the 68330, 68331, 68332, 68333,
           68334, 68336, 68340, 68341, 68349 and 68360.


       -m5200
           Generate output for a 520X "coldfire" family cpu.  This is the default when the compiler is configured for
           520X-based systems.


           Use this option for microcontroller with a 5200 core, including the MCF5202, MCF5203, MCF5204 and MCF5202.


       -m68020-40
           Generate output for a 68040, without using any of the new instructions.  This results in code which can run
           relatively efficiently on either a 68020/68881 or a 68030 or a 68040.  The generated code does use the
           68881 instructions that are emulated on the 68040.


       -m68020-60
           Generate output for a 68060, without using any of the new instructions.  This results in code which can run
           relatively efficiently on either a 68020/68881 or a 68030 or a 68040.  The generated code does use the
           68881 instructions that are emulated on the 68060.


       -msoft-float
           Generate output containing library calls for floating point.  Warning: the requisite libraries are not
           available for all m68k targets.  Normally the facilities of the machine’s usual C compiler are used, but
           this can’t be done directly in cross-compilation.  You must make your own arrangements to provide suitable
           library functions for cross-compilation.  The embedded targets m68k-*-aout and m68k-*-coff do provide soft-
           ware floating point support.


       -mshort
           Consider type "int" to be 16 bits wide, like "short int".  Additionally, parameters passed on the stack are
           also aligned to a 16-bit boundary even on targets whose API mandates promotion to 32-bit.


       -mnobitfield
           Do not use the bit-field instructions.  The -m68000, -mcpu32 and -m5200 options imply -mnobitfield.


       -mbitfield
           Do use the bit-field instructions.  The -m68020 option implies -mbitfield.  This is the default if you use
           a configuration designed for a 68020.


       -mrtd
           Use a different function-calling convention, in which functions that take a fixed number of arguments
           return with the "rtd" instruction, which pops their arguments while returning.  This saves one instruction
           in the caller since there is no need to pop the arguments there.


           This calling convention is incompatible with the one normally used on Unix, so you cannot use it if you
           need to call libraries compiled with the Unix compiler.


           Also, you must provide function prototypes for all functions that take variable numbers of arguments
           (including "printf"); otherwise incorrect code will be generated for calls to those functions.


           In addition, seriously incorrect code will result if you call a function with too many arguments.  (Nor-
           mally, extra arguments are harmlessly ignored.)


           The "rtd" instruction is supported by the 68010, 68020, 68030, 68040, 68060 and CPU32 processors, but not
           by the 68000 or 5200.


       -malign-int
       -mno-align-int
           Control whether GCC aligns "int", "long", "long long", "float", "double", and "long double" variables on a
           32-bit boundary (-malign-int) or a 16-bit boundary (-mno-align-int).  Aligning variables on 32-bit
           boundaries produces code that runs somewhat faster on processors with 32-bit busses at the expense of more
           memory.


           Warning: if you use the -malign-int switch, GCC will align structures containing the above types  differ-
           ently than most published application binary interface specifications for the m68k.


       -mpcrel
           Use the pc-relative addressing mode of the 68000 directly, instead of using a global offset table.  At
           present, this option implies -fpic, allowing at most a 16-bit offset for pc-relative addressing.  -fPIC is
           not presently supported with -mpcrel, though this could be supported for 68020 and higher processors.


       -mno-strict-align
       -mstrict-align
           Do not (do) assume that unaligned memory references will be handled by the system.


       -msep-data
           Generate code that allows the data segment to be located in a different area of memory from the text seg-
           ment.  This allows for execute in place in an environment without virtual memory management.  This option
           implies -fPIC.


       -mno-sep-data
           Generate code that assumes that the data segment follows the text segment.  This is the default.


       -mid-shared-library
           Generate code that supports shared libraries via the library ID method.  This allows for execute in place
           and shared libraries in an environment without virtual memory management.  This option implies -fPIC.


       -mno-id-shared-library
           Generate code that doesn’t assume ID based shared libraries are being used.  This is the default.


       -mshared-library-id=n
           Specified the identification number of the ID based shared library being compiled.  Specifying a value of 0
           will generate more compact code, specifying other values will force the allocation of that number to the
           current library but is no more space or time efficient than omitting this option.


       M68hc1x Options


       These are the -m options defined for the 68hc11 and 68hc12 microcontrollers.  The default values for these
       options depends on which style of microcontroller was selected when the compiler was configured; the defaults
       for the most common choices are given below.


       -m6811
       -m68hc11
           Generate output for a 68HC11.  This is the default when the compiler is configured for 68HC11-based sys-
           tems.


       -m6812
       -m68hc12
           Generate output for a 68HC12.  This is the default when the compiler is configured for 68HC12-based sys-
           tems.


       -m68S12
       -m68hcs12
           Generate output for a 68HCS12.


       -mauto-incdec
           Enable the use of 68HC12 pre and post auto-increment and auto-decrement addressing modes.


       -minmax
       -nominmax
           Enable the use of 68HC12 min and max instructions.


       -mlong-calls
       -mno-long-calls
           Treat all calls as being far away (near).  If calls are assumed to be far away, the compiler will use the
           "call" instruction to call a function and the "rtc" instruction for returning.


       -mshort
           Consider type "int" to be 16 bits wide, like "short int".


       -msoft-reg-count=count
           Specify the number of pseudo-soft registers which are used for the code generation.  The maximum number is
           32.  Using more pseudo-soft register may or may not result in better code depending on the program.  The
           default is 4 for 68HC11 and 2 for 68HC12.


       MCore Options


       These are the -m options defined for the Motorola M*Core processors.


       -mhardlit
       -mno-hardlit
           Inline constants into the code stream if it can be done in two instructions or less.


       -mdiv
       -mno-div
           Use the divide instruction.  (Enabled by default).


       -mrelax-immediate
       -mno-relax-immediate
           Allow arbitrary sized immediates in bit operations.


       -mwide-bitfields
       -mno-wide-bitfields
           Always treat bit-fields as int-sized.


       -m4byte-functions
       -mno-4byte-functions
           Force all functions to be aligned to a four byte boundary.


       -mcallgraph-data
       -mno-callgraph-data
           Emit callgraph information.


       -mslow-bytes
       -mno-slow-bytes
           Prefer word access when reading byte quantities.


       -mlittle-endian
       -mbig-endian
           Generate code for a little endian target.


       -m210
       -m340
           Generate code for the 210 processor.


       MIPS Options


       -EB Generate big-endian code.


       -EL Generate little-endian code.  This is the default for mips*el-*-* configurations.


       -march=arch
           Generate code that will run on arch, which can be the name of a generic MIPS ISA, or the name of a particu-
           lar processor.  The ISA names are: mips1, mips2, mips3, mips4, mips32, mips32r2, and mips64.  The processor
           names are: 4kc, 4km, 4kp, 5kc, 5kf, 20kc, 24k, 24kc, 24kf, 24kx, m4k, orion, r2000, r3000, r3900, r4000,
           r4400, r4600, r4650, r6000, r8000, rm7000, rm9000, sb1, sr71000, vr4100, vr4111, vr4120, vr4130, vr4300,
           vr5000, vr5400 and vr5500.  The special value from-abi selects the most compatible architecture for the
           selected ABI (that is, mips1 for 32-bit ABIs and mips3 for 64-bit ABIs).


           In processor names, a final 000 can be abbreviated as k (for example, -march=r2k).  Prefixes are optional,
           and vr may be written r.


           GCC defines two macros based on the value of this option.  The first is _MIPS_ARCH, which gives the name of
           target architecture, as a string.  The second has the form _MIPS_ARCH_foo, where foo is the capitalized
           value of _MIPS_ARCH.  For example, -march=r2000 will set _MIPS_ARCH to "r2000" and define the macro
           _MIPS_ARCH_R2000.


           Note that the _MIPS_ARCH macro uses the processor names given above.  In other words, it will have the full
           prefix and will not abbreviate 000 as k.  In the case of from-abi, the macro names the resolved architec-
           ture (either "mips1" or "mips3").  It names the default architecture when no -march option is given.


       -mtune=arch
           Optimize for arch.  Among other things, this option controls the way instructions are scheduled, and the
           perceived cost of arithmetic operations.  The list of arch values is the same as for -march.


           When this option is not used, GCC will optimize for the processor specified by -march.  By using -march and
           -mtune together, it is possible to generate code that will run on a family of processors, but optimize the
           code for one particular member of that family.


           -mtune defines the macros _MIPS_TUNE and _MIPS_TUNE_foo, which work in the same way as the -march ones
           described above.


       -mips1
           Equivalent to -march=mips1.


       -mips2
           Equivalent to -march=mips2.


       -mips3
           Equivalent to -march=mips3.


       -mips4
           Equivalent to -march=mips4.


       -mips32
           Equivalent to -march=mips32.


       -mips32r2
           Equivalent to -march=mips32r2.


       -mips64
           Equivalent to -march=mips64.


       -mips16
       -mno-mips16
           Generate (do not generate) MIPS16 code.  If GCC is targetting a MIPS32 or MIPS64 architecture, it will make
           use of the MIPS16e ASE.


       -mabi=32
       -mabi=o64
       -mabi=n32
       -mabi=64
       -mabi=eabi
           Generate code for the given ABI.


           Note that the EABI has a 32-bit and a 64-bit variant.  GCC normally generates 64-bit code when you select a
           64-bit architecture, but you can use -mgp32 to get 32-bit code instead.


           For information about the O64 ABI, see <http://gcc.gnu.org/projects/mipso64-abi.html>.


       -mabicalls
       -mno-abicalls
           Generate (do not generate) SVR4-style position-independent code.  -mabicalls is the default for SVR4-based
           systems.


       -mxgot
       -mno-xgot
           Lift (do not lift) the usual restrictions on the size of the global offset table.


           GCC normally uses a single instruction to load values from the GOT.  While this is relatively efficient, it
           will only work if the GOT is smaller than about 64k.  Anything larger will cause the linker to report an
           error such as:


                   relocation truncated to fit: R_MIPS_GOT16 foobar


           If this happens, you should recompile your code with -mxgot.  It should then work with very large GOTs,
           although it will also be less efficient, since it will take three instructions to fetch the value of a
           global symbol.


           Note that some linkers can create multiple GOTs.  If you have such a linker, you should only need to use
           -mxgot when a single object file accesses more than 64k’s worth of GOT entries.  Very few do.


           These options have no effect unless GCC is generating position independent code.


       -mgp32
           Assume that general-purpose registers are 32 bits wide.


       -mgp64
           Assume that general-purpose registers are 64 bits wide.


       -mfp32
           Assume that floating-point registers are 32 bits wide.


       -mfp64
           Assume that floating-point registers are 64 bits wide.


       -mhard-float
           Use floating-point coprocessor instructions.


       -msoft-float
           Do not use floating-point coprocessor instructions.  Implement floating-point calculations using library
           calls instead.


       -msingle-float
           Assume that the floating-point coprocessor only supports single-precision operations.


       -mdouble-float
           Assume that the floating-point coprocessor supports double-precision operations.  This is the default.


       -mdsp
       -mno-dsp
           Use (do not use) the MIPS DSP ASE.


       -mpaired-single
       -mno-paired-single
           Use (do not use) paired-single floating-point instructions.
             This option can only be used when generating 64-bit code and requires hardware floating-point support to
           be enabled.


       -mips3d
       -mno-mips3d
           Use (do not use) the MIPS-3D ASE.  The option -mips3d implies -mpaired-single.


       -mlong64
           Force "long" types to be 64 bits wide.  See -mlong32 for an explanation of the default and the way that the
           pointer size is determined.


       -mlong32
           Force "long", "int", and pointer types to be 32 bits wide.


           The default size of "int"s, "long"s and pointers depends on the ABI.  All the supported ABIs use 32-bit
           "int"s.  The n64 ABI uses 64-bit "long"s, as does the 64-bit EABI; the others use 32-bit "long"s.  Pointers
           are the same size as "long"s, or the same size as integer registers, whichever is smaller.


       -msym32
       -mno-sym32
           Assume (do not assume) that all symbols have 32-bit values, regardless of the selected ABI.  This option is
           useful in combination with -mabi=64 and -mno-abicalls because it allows GCC to generate shorter and faster
           references to symbolic addresses.


       -G num
           Put global and static items less than or equal to num bytes into the small data or bss section instead of
           the normal data or bss section.  This allows the data to be accessed using a single instruction.


           All modules should be compiled with the same -G num value.


       -membedded-data
       -mno-embedded-data
           Allocate variables to the read-only data section first if possible, then next in the small data section if
           possible, otherwise in data.  This gives slightly slower code than the default, but reduces the amount of
           RAM required when executing, and thus may be preferred for some embedded systems.


       -muninit-const-in-rodata
       -mno-uninit-const-in-rodata
           Put uninitialized "const" variables in the read-only data section.  This option is only meaningful in con-
           junction with -membedded-data.


       -msplit-addresses
       -mno-split-addresses
           Enable (disable) use of the "%hi()" and "%lo()" assembler relocation operators.  This option has been
           superseded by -mexplicit-relocs but is retained for backwards compatibility.


       -mexplicit-relocs
       -mno-explicit-relocs
           Use (do not use) assembler relocation operators when dealing with symbolic addresses.  The alternative,
           selected by -mno-explicit-relocs, is to use assembler macros instead.


           -mexplicit-relocs is the default if GCC was configured to use an assembler that supports relocation opera-
           tors.


       -mcheck-zero-division
       -mno-check-zero-division
           Trap (do not trap) on integer division by zero.  The default is -mcheck-zero-division.


       -mdivide-traps
       -mdivide-breaks
           MIPS systems check for division by zero by generating either a conditional trap or a break instruction.
           Using traps results in smaller code, but is only supported on MIPS II and later.  Also, some versions of
           the Linux kernel have a bug that prevents trap from generating the proper signal ("SIGFPE").  Use -mdi-
           vide-traps to allow conditional traps on architectures that support them and -mdivide-breaks to force the
           use of breaks.


           The default is usually -mdivide-traps, but this can be overridden at configure time using
           --with-divide=breaks.  Divide-by-zero checks can be completely disabled using -mno-check-zero-division.


       -mmemcpy
       -mno-memcpy
           Force (do not force) the use of "memcpy()" for non-trivial block moves.  The default is -mno-memcpy, which
           allows GCC to inline most constant-sized copies.


       -mlong-calls
       -mno-long-calls
           Disable (do not disable) use of the "jal" instruction.  Calling functions using "jal" is more efficient but
           requires the caller and callee to be in the same 256 megabyte segment.


           This option has no effect on abicalls code.  The default is -mno-long-calls.


       -mmad
       -mno-mad
           Enable (disable) use of the "mad", "madu" and "mul" instructions, as provided by the R4650 ISA.


       -mfused-madd
       -mno-fused-madd
           Enable (disable) use of the floating point multiply-accumulate instructions, when they are available.  The
           default is -mfused-madd.


           When multiply-accumulate instructions are used, the intermediate product is calculated to infinite preci-
           sion and is not subject to the FCSR Flush to Zero bit.  This may be undesirable in some circumstances.


       -nocpp
           Tell the MIPS assembler to not run its preprocessor over user assembler files (with a .s suffix) when
           assembling them.


       -mfix-r4000
       -mno-fix-r4000
           Work around certain R4000 CPU errata:


           -   A double-word or a variable shift may give an incorrect result if executed immediately after starting
               an integer division.


           -   A double-word or a variable shift may give an incorrect result if executed while an integer multiplica-
               tion is in progress.


           -   An integer division may give an incorrect result if started in a delay slot of a taken branch or a
               jump.


       -mfix-r4400
       -mno-fix-r4400
           Work around certain R4400 CPU errata:


           -   A double-word or a variable shift may give an incorrect result if executed immediately after starting
               an integer division.


       -mfix-vr4120
       -mno-fix-vr4120
           Work around certain VR4120 errata:


           -   "dmultu" does not always produce the correct result.


           -   "div" and "ddiv" do not always produce the correct result if one of the operands is negative.


           The workarounds for the division errata rely on special functions in libgcc.a.  At present, these functions
           are only provided by the "mips64vr*-elf" configurations.


           Other VR4120 errata require a nop to be inserted between certain pairs of instructions.  These errata are
           handled by the assembler, not by GCC itself.


       -mfix-vr4130
           Work around the VR4130 "mflo"/"mfhi" errata.  The workarounds are implemented by the assembler rather than
           by GCC, although GCC will avoid using "mflo" and "mfhi" if the VR4130 "macc", "macchi", "dmacc" and "dmac-
           chi" instructions are available instead.


       -mfix-sb1
       -mno-fix-sb1
           Work around certain SB-1 CPU core errata.  (This flag currently works around the SB-1 revision 2 "F1" and
           "F2" floating point errata.)


       -mflush-func=func
       -mno-flush-func
           Specifies the function to call to flush the I and D caches, or to not call any such function.  If called,
           the function must take the same arguments as the common "_flush_func()", that is, the address of the memory
           range for which the cache is being flushed, the size of the memory range, and the number 3 (to flush both
           caches).  The default depends on the target GCC was configured for, but commonly is either _flush_func or
           __cpu_flush.


       -mbranch-likely
       -mno-branch-likely
           Enable or disable use of Branch Likely instructions, regardless of the default for the selected architec-
           ture.  By default, Branch Likely instructions may be generated if they are supported by the selected archi-
           tecture.  An exception is for the MIPS32 and MIPS64 architectures and processors which implement those
           architectures; for those, Branch Likely instructions will not be generated by default because the MIPS32
           and MIPS64 architectures specifically deprecate their use.


       -mfp-exceptions
       -mno-fp-exceptions
           Specifies whether FP exceptions are enabled.  This affects how we schedule FP instructions for some proces-
           sors.  The default is that FP exceptions are enabled.


           For instance, on the SB-1, if FP exceptions are disabled, and we are emitting 64-bit code, then we can use
           both FP pipes.  Otherwise, we can only use one FP pipe.


       -mvr4130-align
       -mno-vr4130-align
           The VR4130 pipeline is two-way superscalar, but can only issue two instructions together if the first one
           is 8-byte aligned.  When this option is enabled, GCC will align pairs of instructions that it thinks should
           execute in parallel.


           This option only has an effect when optimizing for the VR4130.  It normally makes code faster, but at the
           expense of making it bigger.  It is enabled by default at optimization level -O3.


       MMIX Options


       These options are defined for the MMIX:


       -mlibfuncs
       -mno-libfuncs
           Specify that intrinsic library functions are being compiled, passing all values in registers, no matter the
           size.


       -mepsilon
       -mno-epsilon
           Generate floating-point comparison instructions that compare with respect to the "rE" epsilon register.


       -mabi=mmixware
       -mabi=gnu
           Generate code that passes function parameters and return values that (in the called function) are seen as
           registers $0 and up, as opposed to the GNU ABI which uses global registers $231 and up.


       -mzero-extend
       -mno-zero-extend
           When reading data from memory in sizes shorter than 64 bits, use (do not use) zero-extending load instruc-
           tions by default, rather than sign-extending ones.


       -mknuthdiv
       -mno-knuthdiv
           Make the result of a division yielding a remainder have the same sign as the divisor.  With the default,
           -mno-knuthdiv, the sign of the remainder follows the sign of the dividend.  Both methods are arithmetically
           valid, the latter being almost exclusively used.


       -mtoplevel-symbols
       -mno-toplevel-symbols
           Prepend (do not prepend) a : to all global symbols, so the assembly code can be used with the "PREFIX"
           assembly directive.


       -melf
           Generate an executable in the ELF format, rather than the default mmo format used by the mmix simulator.


       -mbranch-predict
       -mno-branch-predict
           Use (do not use) the probable-branch instructions, when static branch prediction indicates a probable
           branch.


       -mbase-addresses
       -mno-base-addresses
           Generate (do not generate) code that uses base addresses.  Using a base address automatically generates a
           request (handled by the assembler and the linker) for a constant to be set up in a global register.  The
           register is used for one or more base address requests within the range 0 to 255 from the value held in the
           register.  The generally leads to short and fast code, but the number of different data items that can be
           addressed is limited.  This means that a program that uses lots of static data may require
           -mno-base-addresses.


       -msingle-exit
       -mno-single-exit
           Force (do not force) generated code to have a single exit point in each function.


       MN10300 Options


       These -m options are defined for Matsushita MN10300 architectures:


       -mmult-bug
           Generate code to avoid bugs in the multiply instructions for the MN10300 processors.  This is the default.


       -mno-mult-bug
           Do not generate code to avoid bugs in the multiply instructions for the MN10300 processors.


       -mam33
           Generate code which uses features specific to the AM33 processor.


       -mno-am33
           Do not generate code which uses features specific to the AM33 processor.  This is the default.


       -mreturn-pointer-on-d0
           When generating a function which returns a pointer, return the pointer in both "a0" and "d0".  Otherwise,
           the pointer is returned only in a0, and attempts to call such functions without a prototype would result in
           errors.  Note that this option is on by default; use -mno-return-pointer-on-d0 to disable it.


       -mno-crt0
           Do not link in the C run-time initialization object file.


       -mrelax
           Indicate to the linker that it should perform a relaxation optimization pass to shorten branches, calls and
           absolute memory addresses.  This option only has an effect when used on the command line for the final link
           step.


           This option makes symbolic debugging impossible.


       MT Options


       These -m options are defined for Morpho MT architectures:


       -march=cpu-type
           Generate code that will run on cpu-type, which is the name of a system representing a certain processor
           type.  Possible values for cpu-type are ms1-64-001, ms1-16-002, ms1-16-003 and ms2.


           When this option is not used, the default is -march=ms1-16-002.


       -mbacc
           Use byte loads and stores when generating code.


       -mno-bacc
           Do not use byte loads and stores when generating code.


       -msim
           Use simulator runtime


       -mno-crt0
           Do not link in the C run-time initialization object file crti.o.  Other run-time initialization and termi-
           nation files such as startup.o and exit.o are still included on the linker command line.


       PDP-11 Options


       These options are defined for the PDP-11:


       -mfpu
           Use hardware FPP floating point.  This is the default.  (FIS floating point on the PDP-11/40 is not sup-
           ported.)


       -msoft-float
           Do not use hardware floating point.


       -mac0
           Return floating-point results in ac0 (fr0 in Unix assembler syntax).


       -mno-ac0
           Return floating-point results in memory.  This is the default.


       -m40
           Generate code for a PDP-11/40.


       -m45
           Generate code for a PDP-11/45.  This is the default.


       -m10
           Generate code for a PDP-11/10.


       -mbcopy-builtin
           Use inline "movmemhi" patterns for copying memory.  This is the default.


       -mbcopy
           Do not use inline "movmemhi" patterns for copying memory.


       -mint16
       -mno-int32
           Use 16-bit "int".  This is the default.


       -mint32
       -mno-int16
           Use 32-bit "int".


       -mfloat64
       -mno-float32
           Use 64-bit "float".  This is the default.


       -mfloat32
       -mno-float64
           Use 32-bit "float".


       -mabshi
           Use "abshi2" pattern.  This is the default.


       -mno-abshi
           Do not use "abshi2" pattern.


       -mbranch-expensive
           Pretend that branches are expensive.  This is for experimenting with code generation only.


       -mbranch-cheap
           Do not pretend that branches are expensive.  This is the default.


       -msplit
           Generate code for a system with split I&D.


       -mno-split
           Generate code for a system without split I&D.  This is the default.


       -munix-asm
           Use Unix assembler syntax.  This is the default when configured for pdp11-*-bsd.


       -mdec-asm
           Use DEC assembler syntax.  This is the default when configured for any PDP-11 target other than
           pdp11-*-bsd.


       PowerPC Options


       These are listed under


       IBM RS/6000 and PowerPC Options


       These -m options are defined for the IBM RS/6000 and PowerPC:


       -mpower
       -mno-power
       -mpower2
       -mno-power2
       -mpowerpc
       -mno-powerpc
       -mpowerpc-gpopt
       -mno-powerpc-gpopt
       -mpowerpc-gfxopt
       -mno-powerpc-gfxopt
       -mpowerpc64
       -mno-powerpc64
       -mmfcrf
       -mno-mfcrf
       -mpopcntb
       -mno-popcntb
       -mfprnd
       -mno-fprnd
       -mmfpgpr
       -mno-mfpgpr
           GCC supports two related instruction set architectures for the RS/6000 and PowerPC.  The POWER instruction
           set are those instructions supported by the rios chip set used in the original RS/6000 systems and the Pow-
           erPC instruction set is the architecture of the Freescale MPC5xx, MPC6xx, MPC8xx microprocessors, and the
           IBM 4xx, 6xx, and follow-on microprocessors.


           Neither architecture is a subset of the other.  However there is a large common subset of instructions sup-
           ported by both.  An MQ register is included in processors supporting the POWER architecture.


           You use these options to specify which instructions are available on the processor you are using.  The
           default value of these options is determined when configuring GCC.  Specifying the -mcpu=cpu_type overrides
           the specification of these options.  We recommend you use the -mcpu=cpu_type option rather than the options
           listed above.


           The -mpower option allows GCC to generate instructions that are found only in the POWER architecture and to
           use the MQ register.  Specifying -mpower2 implies -power and also allows GCC to generate instructions that
           are present in the POWER2 architecture but not the original POWER architecture.


           The -mpowerpc option allows GCC to generate instructions that are found only in the 32-bit subset of the
           PowerPC architecture.  Specifying -mpowerpc-gpopt implies -mpowerpc and also allows GCC to use the optional
           PowerPC architecture instructions in the General Purpose group, including floating-point square root.
           Specifying -mpowerpc-gfxopt implies -mpowerpc and also allows GCC to use the optional PowerPC architecture
           instructions in the Graphics group, including floating-point select.


           The -mmfcrf option allows GCC to generate the move from condition register field instruction implemented on
           the POWER4 processor and other processors that support the PowerPC V2.01 architecture.  The -mpopcntb
           option allows GCC to generate the popcount and double precision FP reciprocal estimate instruction imple-
           mented on the POWER5 processor and other processors that support the PowerPC V2.02 architecture.  The
           -mfprnd option allows GCC to generate the FP round to integer instructions implemented on the POWER5+ pro-
           cessor and other processors that support the PowerPC V2.03 architecture.  The -mmfpgpr option allows GCC to
           generate the FP move to/from general purpose register instructions implemented on the POWER6X processor and
           other processors that support the extended PowerPC V2.05 architecture.


           The -mpowerpc64 option allows GCC to generate the additional 64-bit instructions that are found in the full
           PowerPC64 architecture and to treat GPRs as 64-bit, doubleword quantities.  GCC defaults to -mno-powerpc64.


           If you specify both -mno-power and -mno-powerpc, GCC will use only the instructions in the common subset of
           both architectures plus some special AIX common-mode calls, and will not use the MQ register.  Specifying
           both -mpower and -mpowerpc permits GCC to use any instruction from either architecture and to allow use of
           the MQ register; specify this for the Motorola MPC601.


       -mnew-mnemonics
       -mold-mnemonics
           Select which mnemonics to use in the generated assembler code.  With -mnew-mnemonics, GCC uses the assem-
           bler mnemonics defined for the PowerPC architecture.  With -mold-mnemonics it uses the assembler mnemonics
           defined for the POWER architecture.  Instructions defined in only one architecture have only one mnemonic;
           GCC uses that mnemonic irrespective of which of these options is specified.


           GCC defaults to the mnemonics appropriate for the architecture in use.  Specifying -mcpu=cpu_type sometimes
           overrides the value of these option.  Unless you are building a cross-compiler, you should normally not
           specify either -mnew-mnemonics or -mold-mnemonics, but should instead accept the default.


       -mcpu=cpu_type
           Set architecture type, register usage, choice of mnemonics, and instruction scheduling parameters for
           machine type cpu_type.  Supported values for cpu_type are 401, 403, 405, 405fp, 440, 440fp, 505, 601, 602,
           603, 603e, 604, 604e, 620, 630, 740, 7400, 7450, 750, 801, 821, 823, 860, 970, 8540, ec603e, G3, G4, G5,
           power, power2, power3, power4, power5, power5+, power6, power6x, common, powerpc, powerpc64, rios, rios1,
           rios2, rsc, and rs64.


           -mcpu=common selects a completely generic processor.  Code generated under this option will run on any
           POWER or PowerPC processor.  GCC will use only the instructions in the common subset of both architectures,
           and will not use the MQ register.  GCC assumes a generic processor model for scheduling purposes.


           -mcpu=power, -mcpu=power2, -mcpu=powerpc, and -mcpu=powerpc64 specify generic POWER, POWER2, pure 32-bit
           PowerPC (i.e., not MPC601), and 64-bit PowerPC architecture machine types, with an appropriate, generic
           processor model assumed for scheduling purposes.


           The other options specify a specific processor.  Code generated under those options will run best on that
           processor, and may not run at all on others.


           The -mcpu options automatically enable or disable the following options: -maltivec, -mfprnd, -mhard-float,
           -mmfcrf, -mmultiple, -mnew-mnemonics, -mpopcntb, -mpower, -mpower2, -mpowerpc64, -mpowerpc-gpopt, -mpow-
           erpc-gfxopt, -mstring, -mmfpgpr.  The particular options set for any particular CPU will vary between com-
           piler versions, depending on what setting seems to produce optimal code for that CPU; it doesn’t necessar-
           ily reflect the actual hardware’s capabilities.  If you wish to set an individual option to a particular
           value, you may specify it after the -mcpu option, like -mcpu=970 -mno-altivec.


           On AIX, the -maltivec and -mpowerpc64 options are not enabled or disabled by the -mcpu option at present
           because AIX does not have full support for these options.  You may still enable or disable them individu-
           ally if you’re sure it’ll work in your environment.


       -mtune=cpu_type
           Set the instruction scheduling parameters for machine type cpu_type, but do not set the architecture type,
           register usage, or choice of mnemonics, as -mcpu=cpu_type would.  The same values for cpu_type are used for
           -mtune as for -mcpu.  If both are specified, the code generated will use the architecture, registers, and
           mnemonics set by -mcpu, but the scheduling parameters set by -mtune.


       -mswdiv
       -mno-swdiv
           Generate code to compute division as reciprocal estimate and iterative refinement, creating opportunities
           for increased throughput.  This feature requires: optional PowerPC Graphics instruction set for single pre-
           cision and FRE instruction for double precision, assuming divides cannot generate user-visible traps, and
           the domain values not include Infinities, denormals or zero denominator.


       -maltivec
       -mno-altivec
           Generate code that uses (does not use) AltiVec instructions, and also enable the use of built-in functions
           that allow more direct access to the AltiVec instruction set.  You may also need to set -mabi=altivec to
           adjust the current ABI with AltiVec ABI enhancements.


       -mvrsave
       -mno-vrsave
           Generate VRSAVE instructions when generating AltiVec code.


       -msecure-plt
           Generate code that allows ld and ld.so to build executables and shared libraries with non-exec .plt and
           .got sections.  This is a PowerPC 32-bit SYSV ABI option.


       -mbss-plt
           Generate code that uses a BSS .plt section that ld.so fills in, and requires .plt and .got sections that
           are both writable and executable.  This is a PowerPC 32-bit SYSV ABI option.


       -misel
       -mno-isel
           This switch enables or disables the generation of ISEL instructions.


       -misel=yes/no
           This switch has been deprecated.  Use -misel and -mno-isel instead.


       -mspe
       -mno-isel
           This switch enables or disables the generation of SPE simd instructions.


       -mspe=yes/no
           This option has been deprecated.  Use -mspe and -mno-spe instead.


       -mfloat-gprs=yes/single/double/no
       -mfloat-gprs
           This switch enables or disables the generation of floating point operations on the general purpose regis-
           ters for architectures that support it.


           The argument yes or single enables the use of single-precision floating point operations.


           The argument double enables the use of single and double-precision floating point operations.


           The argument no disables floating point operations on the general purpose registers.


           This option is currently only available on the MPC854x.


       -m32
       -m64
           Generate code for 32-bit or 64-bit environments of Darwin and SVR4 targets (including GNU/Linux).  The
           32-bit environment sets int, long and pointer to 32 bits and generates code that runs on any PowerPC vari-
           ant.  The 64-bit environment sets int to 32 bits and long and pointer to 64 bits, and generates code for
           PowerPC64, as for -mpowerpc64.


       -mfull-toc
       -mno-fp-in-toc
       -mno-sum-in-toc
       -mminimal-toc
           Modify generation of the TOC (Table Of Contents), which is created for every executable file.  The
           -mfull-toc option is selected by default.  In that case, GCC will allocate at least one TOC entry for each
           unique non-automatic variable reference in your program.  GCC will also place floating-point constants in
           the TOC.  However, only 16,384 entries are available in the TOC.


           If you receive a linker error message that saying you have overflowed the available TOC space, you can
           reduce the amount of TOC space used with the -mno-fp-in-toc and -mno-sum-in-toc options.  -mno-fp-in-toc
           prevents GCC from putting floating-point constants in the TOC and -mno-sum-in-toc forces GCC to generate
           code to calculate the sum of an address and a constant at run-time instead of putting that sum into the
           TOC.  You may specify one or both of these options.  Each causes GCC to produce very slightly slower and
           larger code at the expense of conserving TOC space.


           If you still run out of space in the TOC even when you specify both of these options, specify -mminimal-toc
           instead.  This option causes GCC to make only one TOC entry for every file.  When you specify this option,
           GCC will produce code that is slower and larger but which uses extremely little TOC space.  You may wish to
           use this option only on files that contain less frequently executed code.


       -maix64
       -maix32
           Enable 64-bit AIX ABI and calling convention: 64-bit pointers, 64-bit "long" type, and the infrastructure
           needed to support them.  Specifying -maix64 implies -mpowerpc64 and -mpowerpc, while -maix32 disables the
           64-bit ABI and implies -mno-powerpc64.  GCC defaults to -maix32.


       -mxl-compat
       -mno-xl-compat
           Produce code that conforms more closely to IBM XL compiler semantics when using AIX-compatible ABI.  Pass
           floating-point arguments to prototyped functions beyond the register save area (RSA) on the stack in addi-
           tion to argument FPRs.  Do not assume that most significant double in 128-bit long double value is properly
           rounded when comparing values and converting to double.  Use XL symbol names for long double support rou-
           tines.


           The AIX calling convention was extended but not initially documented to handle an obscure K&R C case of
           calling a function that takes the address of its arguments with fewer arguments than declared.  IBM XL com-
           pilers access floating point arguments which do not fit in the RSA from the stack when a subroutine is com-
           piled without optimization.  Because always storing floating-point arguments on the stack is inefficient
           and rarely needed, this option is not enabled by default and only is necessary when calling subroutines
           compiled by IBM XL compilers without optimization.


       -mpe
           Support IBM RS/6000 SP Parallel Environment (PE).  Link an application written to use message passing with
           special startup code to enable the application to run.  The system must have PE installed in the standard
           location (/usr/lpp/ppe.poe/), or the specs file must be overridden with the -specs= option to specify the
           appropriate directory location.  The Parallel Environment does not support threads, so the -mpe option and
           the -pthread option are incompatible.


       -malign-natural
       -malign-power
           On AIX, 32-bit Darwin, and 64-bit PowerPC GNU/Linux, the option -malign-natural overrides the ABI-defined
           alignment of larger types, such as floating-point doubles, on their natural size-based boundary.  The
           option -malign-power instructs GCC to follow the ABI-specified alignment rules.  GCC defaults to the stan-
           dard alignment defined in the ABI.


           On 64-bit Darwin, natural alignment is the default, and -malign-power is not supported.


       -msoft-float
       -mhard-float
           Generate code that does not use (uses) the floating-point register set.  Software floating point emulation
           is provided if you use the -msoft-float option, and pass the option to GCC when linking.


       -mmultiple
       -mno-multiple
           Generate code that uses (does not use) the load multiple word instructions and the store multiple word
           instructions.  These instructions are generated by default on POWER systems, and not generated on PowerPC
           systems.  Do not use -mmultiple on little endian PowerPC systems, since those instructions do not work when
           the processor is in little endian mode.  The exceptions are PPC740 and PPC750 which permit the instructions
           usage in little endian mode.


       -mstring
       -mno-string
           Generate code that uses (does not use) the load string instructions and the store string word instructions
           to save multiple registers and do small block moves.  These instructions are generated by default on POWER
           systems, and not generated on PowerPC systems.  Do not use -mstring on little endian PowerPC systems, since
           those instructions do not work when the processor is in little endian mode.  The exceptions are PPC740 and
           PPC750 which permit the instructions usage in little endian mode.


       -mupdate
       -mno-update
           Generate code that uses (does not use) the load or store instructions that update the base register to the
           address of the calculated memory location.  These instructions are generated by default.  If you use
           -mno-update, there is a small window between the time that the stack pointer is updated and the address of
           the previous frame is stored, which means code that walks the stack frame across interrupts or signals may
           get corrupted data.


       -mfused-madd
       -mno-fused-madd
           Generate code that uses (does not use) the floating point multiply and accumulate instructions.  These
           instructions are generated by default if hardware floating is used.


       -mno-bit-align
       -mbit-align
           On System V.4 and embedded PowerPC systems do not (do) force structures and unions that contain bit-fields
           to be aligned to the base type of the bit-field.


           For example, by default a structure containing nothing but 8 "unsigned" bit-fields of length 1 would be
           aligned to a 4 byte boundary and have a size of 4 bytes.  By using -mno-bit-align, the structure would be
           aligned to a 1 byte boundary and be one byte in size.


       -mno-strict-align
       -mstrict-align
           On System V.4 and embedded PowerPC systems do not (do) assume that unaligned memory references will be han-
           dled by the system.


       -mrelocatable
       -mno-relocatable
           On embedded PowerPC systems generate code that allows (does not allow) the program to be relocated to a
           different address at runtime.  If you use -mrelocatable on any module, all objects linked together must be
           compiled with -mrelocatable or -mrelocatable-lib.


       -mrelocatable-lib
       -mno-relocatable-lib
           On embedded PowerPC systems generate code that allows (does not allow) the program to be relocated to a
           different address at runtime.  Modules compiled with -mrelocatable-lib can be linked with either modules
           compiled without -mrelocatable and -mrelocatable-lib or with modules compiled with the -mrelocatable
           options.


       -mno-toc
       -mtoc
           On System V.4 and embedded PowerPC systems do not (do) assume that register 2 contains a pointer to a
           global area pointing to the addresses used in the program.


       -mlittle
       -mlittle-endian
           On System V.4 and embedded PowerPC systems compile code for the processor in little endian mode.  The
           -mlittle-endian option is the same as -mlittle.


       -mbig
       -mbig-endian
           On System V.4 and embedded PowerPC systems compile code for the processor in big endian mode.  The
           -mbig-endian option is the same as -mbig.


       -mdynamic-no-pic
           On Darwin and Mac OS X systems, compile code so that it is not relocatable, but that its external refer-
           ences are relocatable.  The resulting code is suitable for applications, but not shared libraries.


       -mprioritize-restricted-insns=priority
           This option controls the priority that is assigned to dispatch-slot restricted instructions during the sec-
           ond scheduling pass.  The argument priority takes the value 0/1/2 to assign no/highest/second-highest pri-
           ority to dispatch slot restricted instructions.


       -msched-costly-dep=dependence_type
           This option controls which dependences are considered costly by the target during instruction scheduling.
           The argument dependence_type takes one of the following values: no: no dependence is costly, all: all
           dependences are costly, true_store_to_load: a true dependence from store to load is costly, store_to_load:
           any dependence from store to load is costly, number: any dependence which latency >= number is costly.


       -minsert-sched-nops=scheme
           This option controls which nop insertion scheme will be used during the second scheduling pass.  The argu-
           ment scheme takes one of the following values: no: Don’t insert nops.  pad: Pad with nops any dispatch
           group which has vacant issue slots, according to the scheduler’s grouping.  regroup_exact: Insert nops to
           force costly dependent insns into separate groups.  Insert exactly as many nops as needed to force an insn
           to a new group, according to the estimated processor grouping.  number: Insert nops to force costly depen-
           dent insns into separate groups.  Insert number nops to force an insn to a new group.


       -mcall-sysv
           On System V.4 and embedded PowerPC systems compile code using calling conventions that adheres to the March
           1995 draft of the System V Application Binary Interface, PowerPC processor supplement.  This is the default
           unless you configured GCC using powerpc-*-eabiaix.


       -mcall-sysv-eabi
           Specify both -mcall-sysv and -meabi options.


       -mcall-sysv-noeabi
           Specify both -mcall-sysv and -mno-eabi options.


       -mcall-solaris
           On System V.4 and embedded PowerPC systems compile code for the Solaris operating system.


       -mcall-linux
           On System V.4 and embedded PowerPC systems compile code for the Linux-based GNU system.


       -mcall-gnu
           On System V.4 and embedded PowerPC systems compile code for the Hurd-based GNU system.


       -mcall-netbsd
           On System V.4 and embedded PowerPC systems compile code for the NetBSD operating system.


       -maix-struct-return
           Return all structures in memory (as specified by the AIX ABI).


       -msvr4-struct-return
           Return structures smaller than 8 bytes in registers (as specified by the SVR4 ABI).


       -mabi=abi-type
           Extend the current ABI with a particular extension, or remove such extension.  Valid values are altivec,
           no-altivec, spe, no-spe, ibmlongdouble, ieeelongdouble.


       -mabi=spe
           Extend the current ABI with SPE ABI extensions.  This does not change the default ABI, instead it adds the
           SPE ABI extensions to the current ABI.


       -mabi=no-spe
           Disable Booke SPE ABI extensions for the current ABI.


       -mabi=ibmlongdouble
           Change the current ABI to use IBM extended precision long double.  This is a PowerPC 32-bit SYSV ABI
           option.


       -mabi=ieeelongdouble
           Change the current ABI to use IEEE extended precision long double.  This is a PowerPC 32-bit Linux ABI
           option.


       -mprototype
       -mno-prototype
           On System V.4 and embedded PowerPC systems assume that all calls to variable argument functions are prop-
           erly prototyped.  Otherwise, the compiler must insert an instruction before every non prototyped call to
           set or clear bit 6 of the condition code register (CR) to indicate whether floating point values were
           passed in the floating point registers in case the function takes a variable arguments.  With -mprototype,
           only calls to prototyped variable argument functions will set or clear the bit.


       -msim
           On embedded PowerPC systems, assume that the startup module is called sim-crt0.o and that the standard C
           libraries are libsim.a and libc.a.  This is the default for powerpc-*-eabisim.  configurations.


       -mmvme
           On embedded PowerPC systems, assume that the startup module is called crt0.o and the standard C libraries
           are libmvme.a and libc.a.


       -mads
           On embedded PowerPC systems, assume that the startup module is called crt0.o and the standard C libraries
           are libads.a and libc.a.


       -myellowknife
           On embedded PowerPC systems, assume that the startup module is called crt0.o and the standard C libraries
           are libyk.a and libc.a.


       -mvxworks
           On System V.4 and embedded PowerPC systems, specify that you are compiling for a VxWorks system.


       -mwindiss
           Specify that you are compiling for the WindISS simulation environment.


       -memb
           On embedded PowerPC systems, set the PPC_EMB bit in the ELF flags header to indicate that eabi extended
           relocations are used.


       -meabi
       -mno-eabi
           On System V.4 and embedded PowerPC systems do (do not) adhere to the Embedded Applications Binary Interface
           (eabi) which is a set of modifications to the System V.4 specifications.  Selecting -meabi means that the
           stack is aligned to an 8 byte boundary, a function "__eabi" is called to from "main" to set up the eabi
           environment, and the -msdata option can use both "r2" and "r13" to point to two separate small data areas.
           Selecting -mno-eabi means that the stack is aligned to a 16 byte boundary, do not call an initialization
           function from "main", and the -msdata option will only use "r13" to point to a single small data area.  The
           -meabi option is on by default if you configured GCC using one of the powerpc*-*-eabi* options.


       -msdata=eabi
           On System V.4 and embedded PowerPC systems, put small initialized "const" global and static data in the
           .sdata2 section, which is pointed to by register "r2".  Put small initialized non-"const" global and static
           data in the .sdata section, which is pointed to by register "r13".  Put small uninitialized global and
           static data in the .sbss section, which is adjacent to the .sdata section.  The -msdata=eabi option is
           incompatible with the -mrelocatable option.  The -msdata=eabi option also sets the -memb option.


       -msdata=sysv
           On System V.4 and embedded PowerPC systems, put small global and static data in the .sdata section, which
           is pointed to by register "r13".  Put small uninitialized global and static data in the .sbss section,
           which is adjacent to the .sdata section.  The -msdata=sysv option is incompatible with the -mrelocatable
           option.


       -msdata=default
       -msdata
           On System V.4 and embedded PowerPC systems, if -meabi is used, compile code the same as -msdata=eabi, oth-
           erwise compile code the same as -msdata=sysv.


       -msdata-data
           On System V.4 and embedded PowerPC systems, put small global and static data in the .sdata section.  Put
           small uninitialized global and static data in the .sbss section.  Do not use register "r13" to address
           small data however.  This is the default behavior unless other -msdata options are used.


       -msdata=none
       -mno-sdata
           On embedded PowerPC systems, put all initialized global and static data in the .data section, and all
           uninitialized data in the .bss section.


       -G num
           On embedded PowerPC systems, put global and static items less than or equal to num bytes into the small
           data or bss sections instead of the normal data or bss section.  By default, num is 8.  The -G num switch
           is also passed to the linker.  All modules should be compiled with the same -G num value.


       -mregnames
       -mno-regnames
           On System V.4 and embedded PowerPC systems do (do not) emit register names in the assembly language output
           using symbolic forms.


       -mlongcall
       -mno-longcall
           Default to making all function calls indirectly, using a register, so that functions which reside further
           than 32 megabytes (33,554,432 bytes) from the current location can be called.  This setting can be overrid-
           den by the "shortcall" function attribute, or by "#pragma longcall(0)".


           Some linkers are capable of detecting out-of-range calls and generating glue code on the fly.  On these
           systems, long calls are unnecessary and generate slower code.  As of this writing, the AIX linker can do
           this, as can the GNU linker for PowerPC/64.  It is planned to add this feature to the GNU linker for 32-bit
           PowerPC systems as well.


           On Darwin/PPC systems, "#pragma longcall" will generate "jbsr callee, L42", plus a "branch island" (glue
           code).  The two target addresses represent the callee and the "branch island".  The Darwin/PPC linker will
           prefer the first address and generate a "bl callee" if the PPC "bl" instruction will reach the callee
           directly; otherwise, the linker will generate "bl L42" to call the "branch island".  The "branch island" is
           appended to the body of the calling function; it computes the full 32-bit address of the callee and jumps
           to it.


           On Mach-O (Darwin) systems, this option directs the compiler emit to the glue for every direct call, and
           the Darwin linker decides whether to use or discard it.


           In the future, we may cause GCC to ignore all longcall specifications when the linker is known to generate
           glue.


       -pthread
           Adds support for multithreading with the pthreads library.  This option sets flags for both the preproces-
           sor and linker.


       S/390 and zSeries Options


       These are the -m options defined for the S/390 and zSeries architecture.


       -mhard-float
       -msoft-float
           Use (do not use) the hardware floating-point instructions and registers for floating-point operations.
           When -msoft-float is specified, functions in libgcc.a will be used to perform floating-point operations.
           When -mhard-float is specified, the compiler generates IEEE floating-point instructions.  This is the
           default.


       -mlong-double-64
       -mlong-double-128
           These switches control the size of "long double" type. A size of 64bit makes the "long double" type equiva-
           lent to the "double" type. This is the default.


       -mbackchain
       -mno-backchain
           Store (do not store) the address of the caller’s frame as backchain pointer into the callee’s stack frame.
           A backchain may be needed to allow debugging using tools that do not understand DWARF-2 call frame informa-
           tion.  When -mno-packed-stack is in effect, the backchain pointer is stored at the bottom of the stack
           frame; when -mpacked-stack is in effect, the backchain is placed into the topmost word of the 96/160 byte
           register save area.


           In general, code compiled with -mbackchain is call-compatible with code compiled with -mmo-backchain; how-
           ever, use of the backchain for debugging purposes usually requires that the whole binary is built with
           -mbackchain.  Note that the combination of -mbackchain, -mpacked-stack and -mhard-float is not supported.
           In order to build a linux kernel use -msoft-float.


           The default is to not maintain the backchain.


       -mpacked-stack
       -mno-packed-stack
           Use (do not use) the packed stack layout.  When -mno-packed-stack is specified, the compiler uses the all
           fields of the 96/160 byte register save area only for their default purpose; unused fields still take up
           stack space.  When -mpacked-stack is specified, register save slots are densely packed at the top of the
           register save area; unused space is reused for other purposes, allowing for more efficient use of the
           available stack space.  However, when -mbackchain is also in effect, the topmost word of the save area is
           always used to store the backchain, and the return address register is always saved two words below the
           backchain.


           As long as the stack frame backchain is not used, code generated with -mpacked-stack is call-compatible
           with code generated with -mno-packed-stack.  Note that some non-FSF releases of GCC 2.95 for S/390 or
           zSeries generated code that uses the stack frame backchain at run time, not just for debugging purposes.
           Such code is not call-compatible with code compiled with -mpacked-stack.  Also, note that the combination
           of -mbackchain, -mpacked-stack and -mhard-float is not supported.  In order to build a linux kernel use
           -msoft-float.


           The default is to not use the packed stack layout.


       -msmall-exec
       -mno-small-exec
           Generate (or do not generate) code using the "bras" instruction to do subroutine calls.  This only works
           reliably if the total executable size does not exceed 64k.  The default is to use the "basr" instruction
           instead, which does not have this limitation.


       -m64
       -m31
           When -m31 is specified, generate code compliant to the GNU/Linux for S/390 ABI.  When -m64 is specified,
           generate code compliant to the GNU/Linux for zSeries ABI.  This allows GCC in particular to generate 64-bit
           instructions.  For the s390 targets, the default is -m31, while the s390x targets default to -m64.


       -mzarch
       -mesa
           When -mzarch is specified, generate code using the instructions available on z/Architecture.  When -mesa is
           specified, generate code using the instructions available on ESA/390.  Note that -mesa is not possible with
           -m64.  When generating code compliant to the GNU/Linux for S/390 ABI, the default is -mesa.  When generat-
           ing code compliant to the GNU/Linux for zSeries ABI, the default is -mzarch.


       -mmvcle
       -mno-mvcle
           Generate (or do not generate) code using the "mvcle" instruction to perform block moves.  When -mno-mvcle
           is specified, use a "mvc" loop instead.  This is the default unless optimizing for size.


       -mdebug
       -mno-debug
           Print (or do not print) additional debug information when compiling.  The default is to not print debug
           information.


       -march=cpu-type
           Generate code that will run on cpu-type, which is the name of a system representing a certain processor
           type.  Possible values for cpu-type are g5, g6, z900, and z990.  When generating code using the instruc-
           tions available on z/Architecture, the default is -march=z900.  Otherwise, the default is -march=g5.


       -mtune=cpu-type
           Tune to cpu-type everything applicable about the generated code, except for the ABI and the set of avail-
           able instructions.  The list of cpu-type values is the same as for -march.  The default is the value used
           for -march.


       -mtpf-trace
       -mno-tpf-trace
           Generate code that adds (does not add) in TPF OS specific branches to trace routines in the operating sys-
           tem.  This option is off by default, even when compiling for the TPF OS.


       -mfused-madd
       -mno-fused-madd
           Generate code that uses (does not use) the floating point multiply and accumulate instructions.  These
           instructions are generated by default if hardware floating point is used.


       -mwarn-framesize=framesize
           Emit a warning if the current function exceeds the given frame size.  Because this is a compile time check
           it doesn’t need to be a real problem when the program runs.  It is intended to identify functions which
           most probably cause a stack overflow.  It is useful to be used in an environment with limited stack size
           e.g. the linux kernel.


       -mwarn-dynamicstack
           Emit a warning if the function calls alloca or uses dynamically sized arrays.  This is generally a bad idea
           with a limited stack size.


       -mstack-guard=stack-guard
       -mstack-size=stack-size
           These arguments always have to be used in conjunction.  If they are present the s390 back end emits addi-
           tional instructions in the function prologue which trigger a trap if the stack size is stack-guard bytes
           above the stack-size (remember that the stack on s390 grows downward).  These options are intended to be
           used to help debugging stack overflow problems.  The additionally emitted code causes only little overhead
           and hence can also be used in production like systems without greater performance degradation.  The given
           values have to be exact powers of 2 and stack-size has to be greater than stack-guard without exceeding
           64k.  In order to be efficient the extra code makes the assumption that the stack starts at an address
           aligned to the value given by stack-size.


       SH Options


       These -m options are defined for the SH implementations:


       -m1 Generate code for the SH1.


       -m2 Generate code for the SH2.


       -m2e
           Generate code for the SH2e.


       -m3 Generate code for the SH3.


       -m3e
           Generate code for the SH3e.


       -m4-nofpu
           Generate code for the SH4 without a floating-point unit.


       -m4-single-only
           Generate code for the SH4 with a floating-point unit that only supports single-precision arithmetic.


       -m4-single
           Generate code for the SH4 assuming the floating-point unit is in single-precision mode by default.


       -m4 Generate code for the SH4.


       -m4a-nofpu
           Generate code for the SH4al-dsp, or for a SH4a in such a way that the floating-point unit is not used.


       -m4a-single-only
           Generate code for the SH4a, in such a way that no double-precision floating point operations are used.


       -m4a-single
           Generate code for the SH4a assuming the floating-point unit is in single-precision mode by default.


       -m4a
           Generate code for the SH4a.


       -m4al
           Same as -m4a-nofpu, except that it implicitly passes -dsp to the assembler.  GCC doesn’t generate any DSP
           instructions at the moment.


       -mb Compile code for the processor in big endian mode.


       -ml Compile code for the processor in little endian mode.


       -mdalign
           Align doubles at 64-bit boundaries.  Note that this changes the calling conventions, and thus some func-
           tions from the standard C library will not work unless you recompile it first with -mdalign.


       -mrelax
           Shorten some address references at link time, when possible; uses the linker option -relax.


       -mbigtable
           Use 32-bit offsets in "switch" tables.  The default is to use 16-bit offsets.


       -mfmovd
           Enable the use of the instruction "fmovd".


       -mhitachi
           Comply with the calling conventions defined by Renesas.


       -mrenesas
           Comply with the calling conventions defined by Renesas.


       -mno-renesas
           Comply with the calling conventions defined for GCC before the Renesas conventions were available.  This
           option is the default for all targets of the SH toolchain except for sh-symbianelf.


       -mnomacsave
           Mark the "MAC" register as call-clobbered, even if -mhitachi is given.


       -mieee
           Increase IEEE-compliance of floating-point code.  At the moment, this is equivalent to
           -fno-finite-math-only.  When generating 16 bit SH opcodes, getting IEEE-conforming results for comparisons
           of NANs / infinities incurs extra overhead in every floating point comparison, therefore the default is set
           to -ffinite-math-only.


       -misize
           Dump instruction size and location in the assembly code.


       -mpadstruct
           This option is deprecated.  It pads structures to multiple of 4 bytes, which is incompatible with the SH
           ABI.


       -mspace
           Optimize for space instead of speed.  Implied by -Os.


       -mprefergot
           When generating position-independent code, emit function calls using the Global Offset Table instead of the
           Procedure Linkage Table.


       -musermode
           Generate a library function call to invalidate instruction cache entries, after fixing up a trampoline.
           This library function call doesn’t assume it can write to the whole memory address space.  This is the
           default when the target is "sh-*-linux*".


       -multcost=number
           Set the cost to assume for a multiply insn.


       -mdiv=strategy
           Set the division strategy to use for SHmedia code.  strategy must be one of: call, call2, fp, inv, inv:min-
           lat, inv20u, inv20l, inv:call, inv:call2, inv:fp .  "fp" performs the operation in floating point.  This
           has a very high latency, but needs only a few instructions, so it might be a good choice if your code has
           enough easily exploitable ILP to allow the compiler to schedule the floating point instructions together
           with other instructions.  Division by zero causes a floating point exception.  "inv" uses integer opera-
           tions to calculate the inverse of the divisor, and then multiplies the dividend with the inverse.  This
           strategy allows cse and hoisting of the inverse calculation.  Division by zero calculates an unspecified
           result, but does not trap.  "inv:minlat" is a variant of "inv" where if no cse / hoisting opportunities
           have been found, or if the entire operation has been hoisted to the same place, the last stages of the
           inverse calculation are intertwined with the final multiply to reduce the overall latency, at the expense
           of using a few more instructions, and thus offering fewer scheduling opportunities with other code.  "call"
           calls a library function that usually implements the inv:minlat strategy.  This gives high code density for
           m5-*media-nofpu compilations.  "call2" uses a different entry point of the same library function, where it
           assumes that a pointer to a lookup table has already been set up, which exposes the pointer load to cse /
           code hoisting optimizations.  "inv:call", "inv:call2" and "inv:fp" all use the "inv" algorithm for initial
           code generation, but if the code stays unoptimized, revert to the "call", "call2", or "fp" strategies,
           respectively.  Note that the potentially-trapping side effect of division by zero is carried by a separate
           instruction, so it is possible that all the integer instructions are hoisted out, but the marker for the
           side effect stays where it is.  A recombination to fp operations or a call is not possible in that case.
           "inv20u" and "inv20l" are variants of the "inv:minlat" strategy.  In the case that the inverse calculation
           was nor separated from the multiply, they speed up division where the dividend fits into 20 bits (plus sign
           where applicable), by inserting a test to skip a number of operations in this case; this test slows down
           the case of larger dividends.  inv20u assumes the case of a such a small dividend to be unlikely, and
           inv20l assumes it to be likely.


       -mdivsi3_libfunc=name
           Set the name of the library function used for 32 bit signed division to name.  This only affect the name
           used in the call and inv:call division strategies, and the compiler will still expect the same sets of
           input/output/clobbered registers as if this option was not present.


       -madjust-unroll
           Throttle unrolling to avoid thrashing target registers.  This option only has an effect if the gcc code
           base supports the TARGET_ADJUST_UNROLL_MAX target hook.


       -mindexed-addressing
           Enable the use of the indexed addressing mode for SHmedia32/SHcompact.  This is only safe if the hardware
           and/or OS implement 32 bit wrap-around semantics for the indexed addressing mode.  The architecture allows
           the implementation of processors with 64 bit MMU, which the OS could use to get 32 bit addressing, but
           since no current hardware implementation supports this or any other way to make the indexed addressing mode
           safe to use in the 32 bit ABI, the default is -mno-indexed-addressing.


       -mgettrcost=number
           Set the cost assumed for the gettr instruction to number.  The default is 2 if -mpt-fixed is in effect, 100
           otherwise.


       -mpt-fixed
           Assume pt* instructions won’t trap.  This will generally generate better scheduled code, but is unsafe on
           current hardware.  The current architecture definition says that ptabs and ptrel trap when the target anded
           with 3 is 3.  This has the unintentional effect of making it unsafe to schedule ptabs / ptrel before a
           branch, or hoist it out of a loop.  For example, __do_global_ctors, a part of libgcc that runs constructors
           at program startup, calls functions in a list which is delimited by -1.  With the -mpt-fixed option, the
           ptabs will be done before testing against -1.  That means that all the constructors will be run a bit
           quicker, but when the loop comes to the end of the list, the program crashes because ptabs loads -1 into a
           target register.  Since this option is unsafe for any hardware implementing the current architecture speci-
           fication, the default is -mno-pt-fixed.  Unless the user specifies a specific cost with -mgettrcost,
           -mno-pt-fixed also implies -mgettrcost=100; this deters register allocation using target registers for
           storing ordinary integers.


       -minvalid-symbols
           Assume symbols might be invalid.  Ordinary function symbols generated by the compiler will always be valid
           to load with movi/shori/ptabs or movi/shori/ptrel, but with assembler and/or linker tricks it is possible
           to generate symbols that will cause ptabs / ptrel to trap.  This option is only meaningful when
           -mno-pt-fixed is in effect.  It will then prevent cross-basic-block cse, hoisting and most scheduling of
           symbol loads.  The default is -mno-invalid-symbols.


       SPARC Options


       These -m options are supported on the SPARC:


       -mno-app-regs
       -mapp-regs
           Specify -mapp-regs to generate output using the global registers 2 through 4, which the SPARC SVR4 ABI
           reserves for applications.  This is the default.


           To be fully SVR4 ABI compliant at the cost of some performance loss, specify -mno-app-regs.  You should
           compile libraries and system software with this option.


       -mfpu
       -mhard-float
           Generate output containing floating point instructions.  This is the default.


       -mno-fpu
       -msoft-float
           Generate output containing library calls for floating point.  Warning: the requisite libraries are not
           available for all SPARC targets.  Normally the facilities of the machine’s usual C compiler are used, but
           this cannot be done directly in cross-compilation.  You must make your own arrangements to provide suitable
           library functions for cross-compilation.  The embedded targets sparc-*-aout and sparclite-*-* do provide
           software floating point support.


           -msoft-float changes the calling convention in the output file; therefore, it is only useful if you compile
           all of a program with this option.  In particular, you need to compile libgcc.a, the library that comes
           with GCC, with -msoft-float in order for this to work.


       -mhard-quad-float
           Generate output containing quad-word (long double) floating point instructions.


       -msoft-quad-float
           Generate output containing library calls for quad-word (long double) floating point instructions.  The
           functions called are those specified in the SPARC ABI.  This is the default.


           As of this writing, there are no SPARC implementations that have hardware support for the quad-word float-
           ing point instructions.  They all invoke a trap handler for one of these instructions, and then the trap
           handler emulates the effect of the instruction.  Because of the trap handler overhead, this is much slower
           than calling the ABI library routines.  Thus the -msoft-quad-float option is the default.


       -mno-unaligned-doubles
       -munaligned-doubles
           Assume that doubles have 8 byte alignment.  This is the default.


           With -munaligned-doubles, GCC assumes that doubles have 8 byte alignment only if they are contained in
           another type, or if they have an absolute address.  Otherwise, it assumes they have 4 byte alignment.
           Specifying this option avoids some rare compatibility problems with code generated by other compilers.  It
           is not the default because it results in a performance loss, especially for floating point code.


       -mno-faster-structs
       -mfaster-structs
           With -mfaster-structs, the compiler assumes that structures should have 8 byte alignment.  This enables the
           use of pairs of "ldd" and "std" instructions for copies in structure assignment, in place of twice as many
           "ld" and "st" pairs.  However, the use of this changed alignment directly violates the SPARC ABI.  Thus,
           it’s intended only for use on targets where the developer acknowledges that their resulting code will not
           be directly in line with the rules of the ABI.


       -mimpure-text
           -mimpure-text, used in addition to -shared, tells the compiler to not pass -z text to the linker when link-
           ing a shared object.  Using this option, you can link position-dependent code into a shared object.


           -mimpure-text suppresses the "relocations remain against allocatable but non-writable sections" linker
           error message.  However, the necessary relocations will trigger copy-on-write, and the shared object is not
           actually shared across processes.  Instead of using -mimpure-text, you should compile all source code with
           -fpic or -fPIC.


           This option is only available on SunOS and Solaris.


       -mcpu=cpu_type
           Set the instruction set, register set, and instruction scheduling parameters for machine type cpu_type.
           Supported values for cpu_type are v7, cypress, v8, supersparc, sparclite, f930, f934, hypersparc, spar-
           clite86x, sparclet, tsc701, v9, ultrasparc, ultrasparc3, and niagara.


           Default instruction scheduling parameters are used for values that select an architecture and not an imple-
           mentation.  These are v7, v8, sparclite, sparclet, v9.


           Here is a list of each supported architecture and their supported implementations.


                       v7:             cypress
                       v8:             supersparc, hypersparc
                       sparclite:      f930, f934, sparclite86x
                       sparclet:       tsc701
                       v9:             ultrasparc, ultrasparc3, niagara


           By default (unless configured otherwise), GCC generates code for the V7 variant of the SPARC architecture.
           With -mcpu=cypress, the compiler additionally optimizes it for the Cypress CY7C602 chip, as used in the
           SPARCStation/SPARCServer 3xx series.  This is also appropriate for the older SPARCStation 1, 2, IPX etc.


           With -mcpu=v8, GCC generates code for the V8 variant of the SPARC architecture.  The only difference from
           V7 code is that the compiler emits the integer multiply and integer divide instructions which exist in
           SPARC-V8 but not in SPARC-V7.  With -mcpu=supersparc, the compiler additionally optimizes it for the Super-
           SPARC chip, as used in the SPARCStation 10, 1000 and 2000 series.


           With -mcpu=sparclite, GCC generates code for the SPARClite variant of the SPARC architecture.  This adds
           the integer multiply, integer divide step and scan ("ffs") instructions which exist in SPARClite but not in
           SPARC-V7.  With -mcpu=f930, the compiler additionally optimizes it for the Fujitsu MB86930 chip, which is
           the original SPARClite, with no FPU.  With -mcpu=f934, the compiler additionally optimizes it for the
           Fujitsu MB86934 chip, which is the more recent SPARClite with FPU.


           With -mcpu=sparclet, GCC generates code for the SPARClet variant of the SPARC architecture.  This adds the
           integer multiply, multiply/accumulate, integer divide step and scan ("ffs") instructions which exist in
           SPARClet but not in SPARC-V7.  With -mcpu=tsc701, the compiler additionally optimizes it for the TEMIC
           SPARClet chip.


           With -mcpu=v9, GCC generates code for the V9 variant of the SPARC architecture.  This adds 64-bit integer
           and floating-point move instructions, 3 additional floating-point condition code registers and conditional
           move instructions.  With -mcpu=ultrasparc, the compiler additionally optimizes it for the Sun UltraSPARC
           I/II/IIi chips.  With -mcpu=ultrasparc3, the compiler additionally optimizes it for the Sun UltraSPARC
           III/III+/IIIi/IIIi+/IV/IV+ chips.  With -mcpu=niagara, the compiler additionally optimizes it for Sun
           UltraSPARC T1 chips.


       -mtune=cpu_type
           Set the instruction scheduling parameters for machine type cpu_type, but do not set the instruction set or
           register set that the option -mcpu=cpu_type would.


           The same values for -mcpu=cpu_type can be used for -mtune=cpu_type, but the only useful values are those
           that select a particular cpu implementation.  Those are cypress, supersparc, hypersparc, f930, f934, spar-
           clite86x, tsc701, ultrasparc, ultrasparc3, and niagara.


       -mv8plus
       -mno-v8plus
           With -mv8plus, GCC generates code for the SPARC-V8+ ABI.  The difference from the V8 ABI is that the global
           and out registers are considered 64-bit wide.  This is enabled by default on Solaris in 32-bit mode for all
           SPARC-V9 processors.


       -mvis
       -mno-vis
           With -mvis, GCC generates code that takes advantage of the UltraSPARC Visual Instruction Set extensions.
           The default is -mno-vis.


       These -m options are supported in addition to the above on SPARC-V9 processors in 64-bit environments:


       -mlittle-endian
           Generate code for a processor running in little-endian mode.  It is only available for a few configurations
           and most notably not on Solaris and Linux.


       -m32
       -m64
           Generate code for a 32-bit or 64-bit environment.  The 32-bit environment sets int, long and pointer to 32
           bits.  The 64-bit environment sets int to 32 bits and long and pointer to 64 bits.


       -mcmodel=medlow
           Generate code for the Medium/Low code model: 64-bit addresses, programs must be linked in the low 32 bits
           of memory.  Programs can be statically or dynamically linked.


       -mcmodel=medmid
           Generate code for the Medium/Middle code model: 64-bit addresses, programs must be linked in the low 44
           bits of memory, the text and data segments must be less than 2GB in size and the data segment must be
           located within 2GB of the text segment.


       -mcmodel=medany
           Generate code for the Medium/Anywhere code model: 64-bit addresses, programs may be linked anywhere in mem-
           ory, the text and data segments must be less than 2GB in size and the data segment must be located within
           2GB of the text segment.


       -mcmodel=embmedany
           Generate code for the Medium/Anywhere code model for embedded systems: 64-bit addresses, the text and data
           segments must be less than 2GB in size, both starting anywhere in memory (determined at link time).  The
           global register %g4 points to the base of the data segment.  Programs are statically linked and PIC is not
           supported.


       -mstack-bias
       -mno-stack-bias
           With -mstack-bias, GCC assumes that the stack pointer, and frame pointer if present, are offset by -2047
           which must be added back when making stack frame references.  This is the default in 64-bit mode.  Other-
           wise, assume no such offset is present.


       These switches are supported in addition to the above on Solaris:


       -threads
           Add support for multithreading using the Solaris threads library.  This option sets flags for both the pre-
           processor and linker.  This option does not affect the thread safety of object code produced by the com-
           piler or that of libraries supplied with it.


       -pthreads
           Add support for multithreading using the POSIX threads library.  This option sets flags for both the pre-
           processor and linker.  This option does not affect the thread safety of object code produced  by the com-
           piler or that of libraries supplied with it.


       -pthread
           This is a synonym for -pthreads.


       Options for System V


       These additional options are available on System V Release 4 for compatibility with other compilers on those
       systems:


       -G  Create a shared object.  It is recommended that -symbolic or -shared be used instead.


       -Qy Identify the versions of each tool used by the compiler, in a ".ident" assembler directive in the output.


       -Qn Refrain from adding ".ident" directives to the output file (this is the default).


       -YP,dirs
           Search the directories dirs, and no others, for libraries specified with -l.


       -Ym,dir
           Look in the directory dir to find the M4 preprocessor.  The assembler uses this option.


       TMS320C3x/C4x Options


       These -m options are defined for TMS320C3x/C4x implementations:


       -mcpu=cpu_type
           Set the instruction set, register set, and instruction scheduling parameters for machine type cpu_type.
           Supported values for cpu_type are c30, c31, c32, c40, and c44.  The default is c40 to generate code for the
           TMS320C40.


       -mbig-memory
       -mbig
       -msmall-memory
       -msmall
           Generates code for the big or small memory model.  The small memory model assumed that all data fits into
           one 64K word page.  At run-time the data page (DP) register must be set to point to the 64K page containing
           the .bss and .data program sections.  The big memory model is the default and requires reloading of the DP
           register for every direct memory access.


       -mbk
       -mno-bk
           Allow (disallow) allocation of general integer operands into the block count register BK.


       -mdb
       -mno-db
           Enable (disable) generation of code using decrement and branch, DBcond(D), instructions.  This is enabled
           by default for the C4x.  To be on the safe side, this is disabled for the C3x, since the maximum iteration
           count on the C3x is 2^{23 + 1} (but who iterates loops more than 2^{23} times on the C3x?).  Note that GCC
           will try to reverse a loop so that it can utilize the decrement and branch instruction, but will give up if
           there is more than one memory reference in the loop.  Thus a loop where the loop counter is decremented can
           generate slightly more efficient code, in cases where the RPTB instruction cannot be utilized.


       -mdp-isr-reload
       -mparanoid
           Force the DP register to be saved on entry to an interrupt service routine (ISR), reloaded to point to the
           data section, and restored on exit from the ISR.  This should not be required unless someone has violated
           the small memory model by modifying the DP register, say within an object library.


       -mmpyi
       -mno-mpyi
           For the C3x use the 24-bit MPYI instruction for integer multiplies instead of a library call to guarantee
           32-bit results.  Note that if one of the operands is a constant, then the multiplication will be performed
           using shifts and adds.  If the -mmpyi option is not specified for the C3x, then squaring operations are
           performed inline instead of a library call.


       -mfast-fix
       -mno-fast-fix
           The C3x/C4x FIX instruction to convert a floating point value to an integer value chooses the nearest inte-
           ger less than or equal to the floating point value rather than to the nearest integer.  Thus if the float-
           ing point number is negative, the result will be incorrectly truncated an additional code is necessary to
           detect and correct this case.  This option can be used to disable generation of the additional code
           required to correct the result.


       -mrptb
       -mno-rptb
           Enable (disable) generation of repeat block sequences using the RPTB instruction for zero overhead looping.
           The RPTB construct is only used for innermost loops that do not call functions or jump across the loop
           boundaries.  There is no advantage having nested RPTB loops due to the overhead required to save and
           restore the RC, RS, and RE registers.  This is enabled by default with -O2.


       -mrpts=count
       -mno-rpts
           Enable (disable) the use of the single instruction repeat instruction RPTS.  If a repeat block contains a
           single instruction, and the loop count can be guaranteed to be less than the value count, GCC will emit a
           RPTS instruction instead of a RPTB.  If no value is specified, then a RPTS will be emitted even if the loop
           count cannot be determined at compile time.  Note that the repeated instruction following RPTS does not
           have to be reloaded from memory each iteration, thus freeing up the CPU buses for operands.  However, since
           interrupts are blocked by this instruction, it is disabled by default.


       -mloop-unsigned
       -mno-loop-unsigned
           The maximum iteration count when using RPTS and RPTB (and DB on the C40) is 2^{31 + 1} since these instruc-
           tions test if the iteration count is negative to terminate the loop.  If the iteration count is unsigned
           there is a possibility than the 2^{31 + 1} maximum iteration count may be exceeded.  This switch allows an
           unsigned iteration count.


       -mti
           Try to emit an assembler syntax that the TI assembler (asm30) is happy with.  This also enforces compati-
           bility with the API employed by the TI C3x C compiler.  For example, long doubles are passed as structures
           rather than in floating point registers.


       -mregparm
       -mmemparm
           Generate code that uses registers (stack) for passing arguments to functions.  By default, arguments are
           passed in registers where possible rather than by pushing arguments on to the stack.


       -mparallel-insns
       -mno-parallel-insns
           Allow the generation of parallel instructions.  This is enabled by default with -O2.


       -mparallel-mpy
       -mno-parallel-mpy
           Allow the generation of MPY││ADD and MPY││SUB parallel instructions, provided -mparallel-insns is also
           specified.  These instructions have tight register constraints which can pessimize the code generation of
           large functions.


       V850 Options


       These -m options are defined for V850 implementations:


       -mlong-calls
       -mno-long-calls
           Treat all calls as being far away (near).  If calls are assumed to be far away, the compiler will always
           load the functions address up into a register, and call indirect through the pointer.


       -mno-ep
       -mep
           Do not optimize (do optimize) basic blocks that use the same index pointer 4 or more times to copy pointer
           into the "ep" register, and use the shorter "sld" and "sst" instructions.  The -mep option is on by default
           if you optimize.


       -mno-prolog-function
       -mprolog-function
           Do not use (do use) external functions to save and restore registers at the prologue and epilogue of a
           function.  The external functions are slower, but use less code space if more than one function saves the
           same number of registers.  The -mprolog-function option is on by default if you optimize.


       -mspace
           Try to make the code as small as possible.  At present, this just turns on the -mep and -mprolog-function
           options.


       -mtda=n
           Put static or global variables whose size is n bytes or less into the tiny data area that register "ep"
           points to.  The tiny data area can hold up to 256 bytes in total (128 bytes for byte references).


       -msda=n
           Put static or global variables whose size is n bytes or less into the small data area that register "gp"
           points to.  The small data area can hold up to 64 kilobytes.


       -mzda=n
           Put static or global variables whose size is n bytes or less into the first 32 kilobytes of memory.


       -mv850
           Specify that the target processor is the V850.


       -mbig-switch
           Generate code suitable for big switch tables.  Use this option only if the assembler/linker complain about
           out of range branches within a switch table.


       -mapp-regs
           This option will cause r2 and r5 to be used in the code generated by the compiler.  This setting is the
           default.


       -mno-app-regs
           This option will cause r2 and r5 to be treated as fixed registers.


       -mv850e1
           Specify that the target processor is the V850E1.  The preprocessor constants __v850e1__ and __v850e__ will
           be defined if this option is used.


       -mv850e
           Specify that the target processor is the V850E.  The preprocessor constant __v850e__ will be defined if
           this option is used.


           If neither -mv850 nor -mv850e nor -mv850e1 are defined then a default target processor will be chosen and
           the relevant __v850*__ preprocessor constant will be defined.


           The preprocessor constants __v850 and __v851__ are always defined, regardless of which processor variant is
           the target.


       -mdisable-callt
           This option will suppress generation of the CALLT instruction for the v850e and v850e1 flavors of the v850
           architecture.  The default is -mno-disable-callt which allows the CALLT instruction to be used.


       VAX Options


       These -m options are defined for the VAX:


       -munix
           Do not output certain jump instructions ("aobleq" and so on) that the Unix assembler for the VAX cannot
           handle across long ranges.


       -mgnu
           Do output those jump instructions, on the assumption that you will assemble with the GNU assembler.


       -mg Output code for g-format floating point numbers instead of d-format.


       x86-64 Options


       These are listed under


       Xstormy16 Options


       These options are defined for Xstormy16:


       -msim
           Choose startup files and linker script suitable for the simulator.


       Xtensa Options


       These options are supported for Xtensa targets:


       -mconst16
       -mno-const16
           Enable or disable use of "CONST16" instructions for loading constant values.  The "CONST16" instruction is
           currently not a standard option from Tensilica.  When enabled, "CONST16" instructions are always used in
           place of the standard "L32R" instructions.  The use of "CONST16" is enabled by default only if the "L32R"
           instruction is not available.


       -mfused-madd
       -mno-fused-madd
           Enable or disable use of fused multiply/add and multiply/subtract instructions in the floating-point
           option.  This has no effect if the floating-point option is not also enabled.  Disabling fused multiply/add
           and multiply/subtract instructions forces the compiler to use separate instructions for the multiply and
           add/subtract operations.  This may be desirable in some cases where strict IEEE 754-compliant results are
           required: the fused multiply add/subtract instructions do not round the intermediate result, thereby pro-
           ducing results with more bits of precision than specified by the IEEE standard.  Disabling fused multiply
           add/subtract instructions also ensures that the program output is not sensitive to the compiler’s ability
           to combine multiply and add/subtract operations.


       -mtext-section-literals
       -mno-text-section-literals
           Control the treatment of literal pools.  The default is -mno-text-section-literals, which places literals
           in a separate section in the output file.  This allows the literal pool to be placed in a data RAM/ROM, and
           it also allows the linker to combine literal pools from separate object files to remove redundant literals
           and improve code size.  With -mtext-section-literals, the literals are interspersed in the text section in
           order to keep them as close as possible to their references.  This may be necessary for large assembly
           files.


       -mtarget-align
       -mno-target-align
           When this option is enabled, GCC instructs the assembler to automatically align instructions to reduce
           branch penalties at the expense of some code density.  The assembler attempts to widen density instructions
           to align branch targets and the instructions following call instructions.  If there are not enough preced-
           ing safe density instructions to align a target, no widening will be performed.  The default is -mtar-
           get-align.  These options do not affect the treatment of auto-aligned instructions like "LOOP", which the
           assembler will always align, either by widening density instructions or by inserting no-op instructions.


       -mlongcalls
       -mno-longcalls
           When this option is enabled, GCC instructs the assembler to translate direct calls to indirect calls unless
           it can determine that the target of a direct call is in the range allowed by the call instruction.  This
           translation typically occurs for calls to functions in other source files.  Specifically, the assembler
           translates a direct "CALL" instruction into an "L32R" followed by a "CALLX" instruction.  The default is
           -mno-longcalls.  This option should be used in programs where the call target can potentially be out of
           range.  This option is implemented in the assembler, not the compiler, so the assembly code generated by
           GCC will still show direct call instructions---look at the disassembled object code to see the actual
           instructions.  Note that the assembler will use an indirect call for every cross-file call, not just those
           that really will be out of range.


       zSeries Options


       These are listed under


       Options for Code Generation Conventions


       These machine-independent options control the interface conventions used in code generation.


       Most of them have both positive and negative forms; the negative form of -ffoo would be -fno-foo.  In the table
       below, only one of the forms is listed---the one which is not the default.  You can figure out the other form
       by either removing no- or adding it.


       -fbounds-check
           For front-ends that support it, generate additional code to check that indices used to access arrays are
           within the declared range.  This is currently only supported by the Java and Fortran front-ends, where this
           option defaults to true and false respectively.


       -ftrapv
           This option generates traps for signed overflow on addition, subtraction, multiplication operations.


       -fwrapv
           This option instructs the compiler to assume that signed arithmetic overflow of addition, subtraction and
           multiplication wraps around using twos-complement representation.  This flag enables some optimizations and
           disables others.  This option is enabled by default for the Java front-end, as required by the Java lan-
           guage specification.


       -fexceptions
           Enable exception handling.  Generates extra code needed to propagate exceptions.  For some targets, this
           implies GCC will generate frame unwind information for all functions, which can produce significant data
           size overhead, although it does not affect execution.  If you do not specify this option, GCC will enable
           it by default for languages like C++ which normally require exception handling, and disable it for lan-
           guages like C that do not normally require it.  However, you may need to enable this option when compiling
           C code that needs to interoperate properly with exception handlers written in C++.  You may also wish to
           disable this option if you are compiling older C++ programs that don’t use exception handling.


       -fnon-call-exceptions
           Generate code that allows trapping instructions to throw exceptions.  Note that this requires platform-spe-
           cific runtime support that does not exist everywhere.  Moreover, it only allows trapping instructions to
           throw exceptions, i.e. memory references or floating point instructions.  It does not allow exceptions to
           be thrown from arbitrary signal handlers such as "SIGALRM".


       -funwind-tables
           Similar to -fexceptions, except that it will just generate any needed static data, but will not affect the
           generated code in any other way.  You will normally not enable this option; instead, a language processor
           that needs this handling would enable it on your behalf.


       -fasynchronous-unwind-tables
           Generate unwind table in dwarf2 format, if supported by target machine.  The table is exact at each
           instruction boundary, so it can be used for stack unwinding from asynchronous events (such as debugger or
           garbage collector).


       -fpcc-struct-return
           Return "short" "struct" and "union" values in memory like longer ones, rather than in registers.  This con-
           vention is less efficient, but it has the advantage of allowing intercallability between GCC-compiled files
           and files compiled with other compilers, particularly the Portable C Compiler (pcc).


           The precise convention for returning structures in memory depends on the target configuration macros.


           Short structures and unions are those whose size and alignment match that of some integer type.


           Warning: code compiled with the -fpcc-struct-return switch is not binary compatible with code compiled with
           the -freg-struct-return switch.  Use it to conform to a non-default application binary interface.


       -freg-struct-return
           Return "struct" and "union" values in registers when possible.  This is more efficient for small structures
           than -fpcc-struct-return.


           If you specify neither -fpcc-struct-return nor -freg-struct-return, GCC defaults to whichever convention is
           standard for the target.  If there is no standard convention, GCC defaults to -fpcc-struct-return, except
           on targets where GCC is the principal compiler.  In those cases, we can choose the standard, and we chose
           the more efficient register return alternative.


           Warning: code compiled with the -freg-struct-return switch is not binary compatible with code compiled with
           the -fpcc-struct-return switch.  Use it to conform to a non-default application binary interface.


       -fshort-enums
           Allocate to an "enum" type only as many bytes as it needs for the declared range of possible values.
           Specifically, the "enum" type will be equivalent to the smallest integer type which has enough room.


           Warning: the -fshort-enums switch causes GCC to generate code that is not binary compatible with code gen-
           erated without that switch.  Use it to conform to a non-default application binary interface.


       -fshort-double
           Use the same size for "double" as for "float".


           Warning: the -fshort-double switch causes GCC to generate code that is not binary compatible with code gen-
           erated without that switch.  Use it to conform to a non-default application binary interface.


       -fshort-wchar
           Override the underlying type for wchar_t to be short unsigned int instead of the default for the target.
           This option is useful for building programs to run under WINE.


           Warning: the -fshort-wchar switch causes GCC to generate code that is not binary compatible with code gen-
           erated without that switch.  Use it to conform to a non-default application binary interface.


       -fshared-data
           Requests that the data and non-"const" variables of this compilation be shared data rather than private
           data.  The distinction makes sense only on certain operating systems, where shared data is shared between
           processes running the same program, while private data exists in one copy per process.


       -fno-common
           In C, allocate even uninitialized global variables in the data section of the object file, rather than gen-
           erating them as common blocks.  This has the effect that if the same variable is declared (without
           "extern") in two different compilations, you will get an error when you link them.  The only reason this
           might be useful is if you wish to verify that the program will work on other systems which always work this
           way.


       -fno-ident
           Ignore the #ident directive.


       -finhibit-size-directive
           Don’t output a ".size" assembler directive, or anything else that would cause trouble if the function is
           split in the middle, and the two halves are placed at locations far apart in memory.  This option is used
           when compiling crtstuff.c; you should not need to use it for anything else.


       -fverbose-asm
           Put extra commentary information in the generated assembly code to make it more readable.  This option is
           generally only of use to those who actually need to read the generated assembly code (perhaps while debug-
           ging the compiler itself).


           -fno-verbose-asm, the default, causes the extra information to be omitted and is useful when comparing two
           assembler files.


       -fpic
           Generate position-independent code (PIC) suitable for use in a shared library, if supported for the target
           machine.  Such code accesses all constant addresses through a global offset table (GOT).  The dynamic
           loader resolves the GOT entries when the program starts (the dynamic loader is not part of GCC; it is part
           of the operating system).  If the GOT size for the linked executable exceeds a machine-specific maximum
           size, you get an error message from the linker indicating that -fpic does not work; in that case, recompile
           with -fPIC instead.  (These maximums are 8k on the SPARC and 32k on the m68k and RS/6000.  The 386 has no
           such limit.)


           Position-independent code requires special support, and therefore works only on certain machines.  For the
           386, GCC supports PIC for System V but not for the Sun 386i.  Code generated for the IBM RS/6000 is always
           position-independent.


       -fPIC
           If supported for the target machine, emit position-independent code, suitable for dynamic linking and
           avoiding any limit on the size of the global offset table.  This option makes a difference on the m68k,
           PowerPC and SPARC.


           Position-independent code requires special support, and therefore works only on certain machines.


       -fpie
       -fPIE
           These options are similar to -fpic and -fPIC, but generated position independent code can be only linked
           into executables.  Usually these options are used when -pie GCC option will be used during linking.


       -fno-jump-tables
           Do not use jump tables for switch statements even where it would be more efficient than other code genera-
           tion strategies.  This option is of use in conjunction with -fpic or -fPIC for building code which forms
           part of a dynamic linker and cannot reference the address of a jump table.  On some targets, jump tables do
           not require a GOT and this option is not needed.


       -ffixed-reg
           Treat the register named reg as a fixed register; generated code should never refer to it (except perhaps
           as a stack pointer, frame pointer or in some other fixed role).


           reg must be the name of a register.  The register names accepted are machine-specific and are defined in
           the "REGISTER_NAMES" macro in the machine description macro file.


           This flag does not have a negative form, because it specifies a three-way choice.


       -fcall-used-reg
           Treat the register named reg as an allocable register that is clobbered by function calls.  It may be allo-
           cated for temporaries or variables that do not live across a call.  Functions compiled this way will not
           save and restore the register reg.


           It is an error to used this flag with the frame pointer or stack pointer.  Use of this flag for other reg-
           isters that have fixed pervasive roles in the machine’s execution model will produce disastrous results.


           This flag does not have a negative form, because it specifies a three-way choice.


       -fcall-saved-reg
           Treat the register named reg as an allocable register saved by functions.  It may be allocated even for
           temporaries or variables that live across a call.  Functions compiled this way will save and restore the
           register reg if they use it.


           It is an error to used this flag with the frame pointer or stack pointer.  Use of this flag for other reg-
           isters that have fixed pervasive roles in the machine’s execution model will produce disastrous results.


           A different sort of disaster will result from the use of this flag for a register in which function values
           may be returned.


           This flag does not have a negative form, because it specifies a three-way choice.


       -fpack-struct[=n]
           Without a value specified, pack all structure members together without holes.  When a value is specified
           (which must be a small power of two), pack structure members according to this value, representing the max-
           imum alignment (that is, objects with default alignment requirements larger than this will be output poten-
           tially unaligned at the next fitting location.


           Warning: the -fpack-struct switch causes GCC to generate code that is not binary compatible with code gen-
           erated without that switch.  Additionally, it makes the code suboptimal.  Use it to conform to a non-
           default application binary interface.


       -finstrument-functions
           Generate instrumentation calls for entry and exit to functions.  Just after function entry and just before
           function exit, the following profiling functions will be called with the address of the current function
           and its call site.  (On some platforms, "__builtin_return_address" does not work beyond the current func-
           tion, so the call site information may not be available to the profiling functions otherwise.)


                   void __cyg_profile_func_enter (void *this_fn,
                                                  void *call_site);
                   void __cyg_profile_func_exit  (void *this_fn,
                                                  void *call_site);


           The first argument is the address of the start of the current function, which may be looked up exactly in
           the symbol table.


           This instrumentation is also done for functions expanded inline in other functions.  The profiling calls
           will indicate where, conceptually, the inline function is entered and exited.  This means that addressable
           versions of such functions must be available.  If all your uses of a function are expanded inline, this may
           mean an additional expansion of code size.  If you use extern inline in your C code, an addressable version
           of such functions must be provided.  (This is normally the case anyways, but if you get lucky and the opti-
           mizer always expands the functions inline, you might have gotten away without providing static copies.)


           A function may be given the attribute "no_instrument_function", in which case this instrumentation will not
           be done.  This can be used, for example, for the profiling functions listed above, high-priority interrupt
           routines, and any functions from which the profiling functions cannot safely be called (perhaps signal han-
           dlers, if the profiling routines generate output or allocate memory).


       -fstack-check
           Generate code to verify that you do not go beyond the boundary of the stack.  You should specify this flag
           if you are running in an environment with multiple threads, but only rarely need to specify it in a single-
           threaded environment since stack overflow is automatically detected on nearly all systems if there is only
           one stack.


           Note that this switch does not actually cause checking to be done; the operating system must do that.  The
           switch causes generation of code to ensure that the operating system sees the stack being extended.


       -fstack-limit-register=reg
       -fstack-limit-symbol=sym
       -fno-stack-limit
           Generate code to ensure that the stack does not grow beyond a certain value, either the value of a register
           or the address of a symbol.  If the stack would grow beyond the value, a signal is raised.  For most tar-
           gets, the signal is raised before the stack overruns the boundary, so it is possible to catch the signal
           without taking special precautions.


           For instance, if the stack starts at absolute address 0x80000000 and grows downwards, you can use the flags
           -fstack-limit-symbol=__stack_limit and -Wl,--defsym,__stack_limit=0x7ffe0000 to enforce a stack limit of
           128KB.  Note that this may only work with the GNU linker.


       -fargument-alias
       -fargument-noalias
       -fargument-noalias-global
           Specify the possible relationships among parameters and between parameters and global data.


           -fargument-alias specifies that arguments (parameters) may alias each other and may alias global stor-
           age.-fargument-noalias specifies that arguments do not alias each other, but may alias global storage.-far-
           gument-noalias-global specifies that arguments do not alias each other and do not alias global storage.


           Each language will automatically use whatever option is required by the language standard.  You should not
           need to use these options yourself.


       -fleading-underscore
           This option and its counterpart, -fno-leading-underscore, forcibly change the way C symbols are represented
           in the object file.  One use is to help link with legacy assembly code.


           Warning: the -fleading-underscore switch causes GCC to generate code that is not binary compatible with
           code generated without that switch.  Use it to conform to a non-default application binary interface.  Not
           all targets provide complete support for this switch.


       -ftls-model=model
           Alter the thread-local storage model to be used.  The model argument should be one of "global-dynamic",
           "local-dynamic", "initial-exec" or "local-exec".


           The default without -fpic is "initial-exec"; with -fpic the default is "global-dynamic".


       -fvisibility=default│internal│hidden│protected
           Set the default ELF image symbol visibility to the specified option---all symbols will be marked with this
           unless overridden within the code.  Using this feature can very substantially improve linking and load
           times of shared object libraries, produce more optimized code, provide near-perfect API export and prevent
           symbol clashes.  It is strongly recommended that you use this in any shared objects you distribute.


           Despite the nomenclature, "default" always means public ie; available to be linked against from outside the
           shared object.  "protected" and "internal" are pretty useless in real-world usage so the only other com-
           monly used option will be "hidden".  The default if -fvisibility isn’t specified is "default", i.e., make
           every symbol public---this causes the same behavior as previous versions of GCC.


           A good explanation of the benefits offered by ensuring ELF symbols have the correct visibility is given by
           "How To Write Shared Libraries" by Ulrich Drepper (which can be found at <http://people.redhat.com/~drep-
           per/>)---however a superior solution made possible by this option to marking things hidden when the default
           is public is to make the default hidden and mark things public.  This is the norm with DLL’s on Windows and
           with -fvisibility=hidden and "__attribute__ ((visibility("default")))" instead of "__declspec(dllexport)"
           you get almost identical semantics with identical syntax.  This is a great boon to those working with
           cross-platform projects.


           For those adding visibility support to existing code, you may find #pragma GCC visibility of use.  This
           works by you enclosing the declarations you wish to set visibility for with (for example) #pragma GCC visi-
           bility push(hidden) and #pragma GCC visibility pop.  Bear in mind that symbol visibility should be viewed
           as part of the API interface contract and thus all new code should always specify visibility when it is not
           the default ie; declarations only for use within the local DSO should always be marked explicitly as hidden
           as so to avoid PLT indirection overheads---making this abundantly clear also aids readability and self-doc-
           umentation of the code.  Note that due to ISO C++ specification requirements, operator new and operator
           delete must always be of default visibility.


           extern declarations are not affected by -fvisibility, so a lot of code can be recompiled with -fvisibil-
           ity=hidden with no modifications.  However, this means that calls to extern functions with no explicit vis-
           ibility will use the PLT, so it is more effective to use __attribute ((visibility)) and/or #pragma GCC vis-
           ibility to tell the compiler which extern declarations should be treated as hidden.


           Note that -fvisibility does affect C++ vague linkage entities. This means that, for instance, an exception
           class that will be thrown between DSOs must be explicitly marked with default visibility so that the
           type_info nodes will be unified between the DSOs.


           An overview of these techniques, their benefits and how to use them is at <http://gcc.gnu.org/wiki/Visibil-
           ity>.


       -fopenmp
           Enable handling of OpenMP directives "#pragma omp" in C/C++ and "!$omp" in Fortran.  When -fopenmp is spec-
           ified, the compiler generates parallel code according to the OpenMP Application Program Interface v2.5
           <http://www.openmp.org/>.


ENVIRONMENT
       This section describes several environment variables that affect how GCC operates.  Some of them work by speci-
       fying directories or prefixes to use when searching for various kinds of files.  Some are used to specify other
       aspects of the compilation environment.


       Note that you can also specify places to search using options such as -B, -I and -L.  These take precedence
       over places specified using environment variables, which in turn take precedence over those specified by the
       configuration of GCC.


       LANG
       LC_CTYPE
       LC_MESSAGES
       LC_ALL
           These environment variables control the way that GCC uses localization information that allow GCC to work
           with different national conventions.  GCC inspects the locale categories LC_CTYPE and LC_MESSAGES if it has
           been configured to do so.  These locale categories can be set to any value supported by your installation.
           A typical value is en_GB.UTF-8 for English in the United Kingdom encoded in UTF-8.


           The LC_CTYPE environment variable specifies character classification.  GCC uses it to determine the charac-
           ter boundaries in a string; this is needed for some multibyte encodings that contain quote and escape char-
           acters that would otherwise be interpreted as a string end or escape.


           The LC_MESSAGES environment variable specifies the language to use in diagnostic messages.


           If the LC_ALL environment variable is set, it overrides the value of LC_CTYPE and LC_MESSAGES; otherwise,
           LC_CTYPE and LC_MESSAGES default to the value of the LANG environment variable.  If none of these variables
           are set, GCC defaults to traditional C English behavior.


       TMPDIR
           If TMPDIR is set, it specifies the directory to use for temporary files.  GCC uses temporary files to hold
           the output of one stage of compilation which is to be used as input to the next stage: for example, the
           output of the preprocessor, which is the input to the compiler proper.


       GCC_EXEC_PREFIX
           If GCC_EXEC_PREFIX is set, it specifies a prefix to use in the names of the subprograms executed by the
           compiler.  No slash is added when this prefix is combined with the name of a subprogram, but you can spec-
           ify a prefix that ends with a slash if you wish.


           If GCC_EXEC_PREFIX is not set, GCC will attempt to figure out an appropriate prefix to use based on the
           pathname it was invoked with.


           If GCC cannot find the subprogram using the specified prefix, it tries looking in the usual places for the
           subprogram.


           The default value of GCC_EXEC_PREFIX is prefix/lib/gcc/ where prefix is the value of "prefix" when you ran
           the configure script.


           Other prefixes specified with -B take precedence over this prefix.


           This prefix is also used for finding files such as crt0.o that are used for linking.


           In addition, the prefix is used in an unusual way in finding the directories to search for header files.
           For each of the standard directories whose name normally begins with /usr/local/lib/gcc (more precisely,
           with the value of GCC_INCLUDE_DIR), GCC tries replacing that beginning with the specified prefix to produce
           an alternate directory name.  Thus, with -Bfoo/, GCC will search foo/bar where it would normally search
           /usr/local/lib/bar.  These alternate directories are searched first; the standard directories come next.


       COMPILER_PATH
           The value of COMPILER_PATH is a colon-separated list of directories, much like PATH.  GCC tries the direc-
           tories thus specified when searching for subprograms, if it can’t find the subprograms using GCC_EXEC_PRE-
           FIX.


       LIBRARY_PATH
           The value of LIBRARY_PATH is a colon-separated list of directories, much like PATH.  When configured as a
           native compiler, GCC tries the directories thus specified when searching for special linker files, if it
           can’t find them using GCC_EXEC_PREFIX.  Linking using GCC also uses these directories when searching for
           ordinary libraries for the -l option (but directories specified with -L come first).


       LANG
           This variable is used to pass locale information to the compiler.  One way in which this information is
           used is to determine the character set to be used when character literals, string literals and comments are
           parsed in C and C++.  When the compiler is configured to allow multibyte characters, the following values
           for LANG are recognized:


           C-JIS
               Recognize JIS characters.


           C-SJIS
               Recognize SJIS characters.


           C-EUCJP
               Recognize EUCJP characters.


           If LANG is not defined, or if it has some other value, then the compiler will use mblen and mbtowc as
           defined by the default locale to recognize and translate multibyte characters.


       Some additional environments variables affect the behavior of the preprocessor.


       CPATH
       C_INCLUDE_PATH
       CPLUS_INCLUDE_PATH
       OBJC_INCLUDE_PATH
           Each variable’s value is a list of directories separated by a special character, much like PATH, in which
           to look for header files.  The special character, "PATH_SEPARATOR", is target-dependent and determined at
           GCC build time.  For Microsoft Windows-based targets it is a semicolon, and for almost all other targets it
           is a colon.


           CPATH specifies a list of directories to be searched as if specified with -I, but after any paths given
           with -I options on the command line.  This environment variable is used regardless of which language is
           being preprocessed.


           The remaining environment variables apply only when preprocessing the particular language indicated.  Each
           specifies a list of directories to be searched as if specified with -isystem, but after any paths given
           with -isystem options on the command line.


           In all these variables, an empty element instructs the compiler to search its current working directory.
           Empty elements can appear at the beginning or end of a path.  For instance, if the value of CPATH is
           ":/special/include", that has the same effect as -I. -I/special/include.


       DEPENDENCIES_OUTPUT
           If this variable is set, its value specifies how to output dependencies for Make based on the non-system
           header files processed by the compiler.  System header files are ignored in the dependency output.


           The value of DEPENDENCIES_OUTPUT can be just a file name, in which case the Make rules are written to that
           file, guessing the target name from the source file name.  Or the value can have the form file target, in
           which case the rules are written to file file using target as the target name.


           In other words, this environment variable is equivalent to combining the options -MM and -MF, with an
           optional -MT switch too.


       SUNPRO_DEPENDENCIES
           This variable is the same as DEPENDENCIES_OUTPUT (see above), except that system header files are not
           ignored, so it implies -M rather than -MM.  However, the dependence on the main input file is omitted.


BUGS
       For instructions on reporting bugs, see <http://gcc.gnu.org/bugs.html>.


FOOTNOTES
       1.  On some systems, gcc -shared needs to build supplementary stub code for constructors to work.  On multi-
           libbed systems, gcc -shared must select the correct support libraries to link against.  Failing to supply
           the correct flags may lead to subtle defects.  Supplying them in cases where they are not necessary is
           innocuous.


SEE ALSO
       gpl(7), gfdl(7), fsf-funding(7), cpp(1), gcov(1), as(1), ld(1), gdb(1), adb(1), dbx(1), sdb(1) and the Info
       entries for gcc, cpp, as, ld, binutils and gdb.


AUTHOR
       See the Info entry for gcc, or <http://gcc.gnu.org/onlinedocs/gcc/Contributors.html>, for contributors to GCC.


COPYRIGHT
       Copyright (c) 1988, 1989, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001, 2002, 2003, 2004, 2005
       Free Software Foundation, Inc.


       Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documen-
       tation License, Version 1.2 or any later version published by the Free Software Foundation; with the Invariant
       Sections being "GNU General Public License" and "Funding Free Software", the Front-Cover texts being (a) (see
       below), and with the Back-Cover Texts being (b) (see below).  A copy of the license is included in the gfdl(7)
       man page.


       (a) The FSF’s Front-Cover Text is:


            A GNU Manual


       (b) The FSF’s Back-Cover Text is:


            You have freedom to copy and modify this GNU Manual, like GNU
            software.  Copies published by the Free Software Foundation raise
            funds for GNU development.


gcc-4.1.2                         2010-10-06                            GCC(1)
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1 Programming Languages Supported by GCC . . . . . . . . . . . . . . . 3 2 Language Standards Supported by GCC . . . . . . . . . . . . . . . . . . 5 3 GCC Command Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 4 C Implementation-defined behavior. . . . . . . . . . . . . . . . . . . . . 299 5 C++ Implementation-defined behavior . . . . . . . . . . . . . . . . . . 307 6 Extensions to the C Language Family . . . . . . . . . . . . . . . . . . . 309 7 Extensions to the C++ Language . . . . . . . . . . . . . . . . . . . . . . 623 8 GNU Objective-C features . . . . . . . . . . . . . . . . . . . . . . . . . . . . 635 9 Binary Compatibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 651 10 gcov—a Test Coverage Program . . . . . . . . . . . . . . . . . . . . . . . 655 11 Known Causes of Trouble with GCC. . . . . . . . . . . . . . . . . . . . 663 12 Reporting Bugs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 679 13 How To Get Help with GCC . . . . . . . . . . . . . . . . . . . . . . . . . . 681 14 Contributing to GCC Development . . . . . . . . . . . . . . . . . . . . . 683 Funding Free Software . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 685 The GNU Project and GNU/Linux. . . . . . . . . . . . . . . . . . . . . . . . . 687 GNU General Public License . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 689 GNU Free Documentation License . . . . . . . . . . . . . . . . . . . . . . . . . 701 Contributors to GCC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 709 Option Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 725
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值