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 remain-
       der.  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 docu-
       mented 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 particular 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 matter when you use several options
       of the same kind; for example, if you specify -L more than once, the
       directories are searched in the order specified.
 
       Many options have long names starting with -f or with -W---for example,
       -fforce-mem, -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  --version
 
       C Language Options
           -ansi  -std=standard  -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  -funsigned-char
 
       C++ Language Options
           -fabi-version=n  -fno-access-control  -fcheck-new -fconserve-space
           -fno-const-strings -fno-elide-constructors -fno-enforce-eh-specs
           -ffor-scope  -fno-for-scope  -fno-gnu-keywords -fno-implicit-tem-
           plates -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  -fvisibil-
           ity-inlines-hidden -Wabi  -Wctor-dtor-privacy -Wnon-virtual-dtor
           -Wreorder -Weffc++  -Wno-deprecated  -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-exceptions -freplace-objc-classes
           -fzero-link -gen-decls -Wno-protocol  -Wselector -Wunde-
           clared-selector
 
       Language Independent Options
           -fmessage-length=n -fdiagnostics-show-location=[once|every-line]
 
       Warning Options
           -fsyntax-only  -pedantic  -pedantic-errors -w -Wextra  -Wall
           -Waggregate-return -Wcast-align  -Wcast-qual  -Wchar-subscripts
           -Wcomment -Wconversion  -Wno-deprecated-declarations -Wdis-
           abled-optimization  -Wno-div-by-zero  -Wno-endif-labels -Werror
           -Werror-implicit-function-declaration -Wfatal-errors  -Wfloat-equal
           -Wformat  -Wformat=2 -Wno-format-extra-args -Wformat-nonliteral
           -Wformat-security  -Wformat-y2k -Wimplicit  -Wimplicit-func-
           tion-declaration  -Wimplicit-int -Wimport  -Wno-import  -Winit-self
           -Winline -Wno-invalid-offsetof  -Winvalid-pch -Wlarger-than-len
           -Wlong-long -Wmain  -Wmissing-braces  -Wmissing-field-initializers
           -Wmissing-format-attribute  -Wmissing-include-dirs -Wmissing-nore-
           turn -Wno-multichar  -Wnonnull  -Wpacked  -Wpadded -Wparentheses
           -Wpointer-arith  -Wredundant-decls -Wreturn-type  -Wsequence-point
           -Wshadow -Wsign-compare  -Wstrict-aliasing -Wstrict-aliasing=2
           -Wswitch  -Wswitch-default  -Wswitch-enum -Wsystem-headers  -Wtri-
           graphs  -Wundef  -Wuninitialized -Wunknown-pragmas  -Wunreach-
           able-code -Wunused  -Wunused-function  -Wunused-label
           -Wunused-parameter -Wunused-value  -Wunused-variable
           -Wwrite-strings -Wvariadic-macros
 
       C-only Warning Options
           -Wbad-function-cast  -Wmissing-declarations -Wmissing-prototypes
           -Wnested-externs  -Wold-style-definition -Wstrict-prototypes
           -Wtraditional -Wdeclaration-after-statement -Wno-pointer-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-origi-
           nal[-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-mud-
           flap[-n] -fdump-tree-dom[-n] -fdump-tree-dse[-n]
           -fdump-tree-phiopt[-n] -fdump-tree-forwprop[-n] -fdump-tree-copyre-
           name[-n] -fdump-tree-nrv -fdump-tree-vect -fdump-tree-sra[-n]
           -fdump-tree-fre[-n] -ftree-vectorizer-verbose=n -felimi-
           nate-dwarf2-dups -feliminate-unused-debug-types -felimi-
           nate-unused-debug-symbols -fmem-report -fprofile-arcs
           -ftree-based-profiling -frandom-seed=string -fsched-verbose=n
           -ftest-coverage  -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-prog-name=program
           -print-search-dirs  -Q -save-temps  -time
 
       Optimization Options
           -falign-functions=n  -falign-jumps=n -falign-labels=n
           -falign-loops=n -fbounds-check -fmudflap -fmudflapth -fmudflapir
           -fbranch-probabilities -fprofile-values -fvpt -fbranch-tar-
           get-load-optimize -fbranch-target-load-optimize2 -fbtr-bb-exclusive
           -fcaller-saves  -fcprop-registers  -fcse-follow-jumps
           -fcse-skip-blocks  -fcx-limited-range  -fdata-sections -fde-
           layed-branch  -fdelete-null-pointer-checks -fexpensive-optimiza-
           tions  -ffast-math  -ffloat-store -fforce-addr  -fforce-mem
           -ffunction-sections -fgcse  -fgcse-lm  -fgcse-sm  -fgcse-las
           -fgcse-after-reload -floop-optimize -fcrossjumping  -fif-conversion
           -fif-conversion2 -finline-functions  -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-proba-
           bility -fno-inline  -fno-math-errno  -fno-peephole  -fno-peephole2
           -funsafe-math-optimizations  -ffinite-math-only -fno-trapping-math
           -fno-zero-initialized-in-bss -fomit-frame-pointer  -foptimize-reg-
           ister-move -foptimize-sibling-calls  -fprefetch-loop-arrays -fpro-
           file-generate -fprofile-use -fregmove  -frename-registers -fre-
           order-blocks  -freorder-blocks-and-partition -freorder-functions
           -frerun-cse-after-loop  -frerun-loop-opt -frounding-math -fsched-
           ule-insns  -fschedule-insns2 -fno-sched-interblock  -fno-sched-spec
           -fsched-spec-load -fsched-spec-load-dangerous
           -fsched-stalled-insns=n -sched-stalled-insns-dep=n
           -fsched2-use-superblocks -fsched2-use-traces -freschedule-mod-
           ulo-scheduled-loops -fsignaling-nans -fsingle-precision-constant
           -fspeculative-prefetching -fstrength-reduce  -fstrict-aliasing
           -ftracer  -fthread-jumps -funroll-all-loops  -funroll-loops
           -fpeel-loops -fsplit-ivs-in-unroller -funswitch-loops -fvari-
           able-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-ch -ftree-sra -ftree-ter -ftree-lrs
           -ftree-fre -ftree-vectorize -fweb --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 -M  -MM  -MF  -MG  -MP  -MQ  -MT  -nostdinc -P  -fwork-
           ing-directory  -remap -trigraphs  -undef  -Umacro  -Wp,option
           -Xpreprocessor option
 
       Assembler Option
           -Wa,option  -Xassembler option
 
       Linker Options
           object-file-name  -llibrary -nostartfiles  -nodefaultlibs  -nost-
           dlib -pie -s  -static  -static-libgcc  -shared  -shared-libgcc
           -symbolic -Wl,option  -Xlinker option -u symbol
 
       Directory Options
           -Bprefix  -Idir  -iquotedir  -Ldir  -specs=file  -I-
 
       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-pro-
           log  -mno-sched-prolog -mlittle-endian  -mbig-endian  -mwords-lit-
           tle-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-sin-
           gle-pic-base -mpic-register=reg -mnop-fun-dllimport -mcir-
           rus-fix-invalid-insns -mno-cirrus-fix-invalid-insns -mpoke-func-
           tion-name -mthumb  -marm -mtpcs-frame  -mtpcs-leaf-frame
           -mcaller-super-interworking  -mcallee-super-interworking
 
           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
 
           Darwin Options -all_load  -allowable_client  -arch
           -arch_errors_fatal -arch_only  -bind_at_load  -bundle  -bun-
           dle_loader -client_name  -compatibility_version  -current_version
           -dead_strip -dependency-file  -dylib_file  -dylinker_install_name
           -dynamic  -dynamiclib  -exported_symbols_list -filelist
           -flat_namespace  -force_cpusubtype_ALL -force_flat_namespace
           -headerpad_max_install_names -image_base  -init  -install_name
           -keep_private_externs -multi_module  -multiply_defined  -multi-
           ply_defined_unused -noall_load   -no_dead_strip_inits_and_terms
           -nofixprebinding -nomultidefs  -noprebind  -noseglinkedit
           -pagezero_size  -prebind  -prebind_all_twolevel_modules -pri-
           vate_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  -unde-
           fined -unexported_symbols_list  -weak_reference_mismatches -what-
           sloaded -F -gused -gfull -mone-byte-bool
 
           DEC Alpha Options -mno-fp-regs  -msoft-float  -malpha-as  -mgas
           -mieee  -mieee-with-inexact  -mieee-conformant -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 -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  -mdis-
           able-fpregs  -mdisable-indexing -mfast-indirect-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-indirect-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 -mschedule=cpu-type  -mspace-regs  -msio  -mwsio
           -munix=unix-std  -nolibdld  -static  -threads
 
           i386 and x86-64 Options -mtune=cpu-type  -march=cpu-type -mfp-
           math=unit -masm=dialect  -mno-fancy-math-387 -mno-fp-ret-in-387
           -msoft-float  -msvr3-shlib -mno-wide-multiply -mrtd  -malign-dou-
           ble -mpreferred-stack-boundary=num -mmmx  -msse  -msse2 -msse3
           -m3dnow -mthreads  -mno-align-stringops  -minline-all-stringops
           -mpush-args  -maccumulate-outgoing-args  -m128bit-long-double
           -m96bit-long-double  -mregparm=num  -momit-leaf-frame-pointer
           -mno-red-zone -mno-tls-direct-seg-refs -mcmodel=code-model -m32
           -m64
 
           IA-64 Options -mbig-endian  -mlittle-endian  -mgnu-as  -mgnu-ld
           -mno-pic -mvolatile-asm-stop  -mregister-names  -mno-sdata -mcon-
           stant-gp  -mauto-pic  -minline-float-divide-min-latency -min-
           line-float-divide-max-throughput -minline-int-divide-min-latency
           -minline-int-divide-max-throughput -minline-sqrt-min-latency -min-
           line-sqrt-max-throughput -mno-dwarf2-asm -mearly-stop-bits
           -mfixed-range=register-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
 
           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  -mre-
           lax-immediates -mno-relax-immediates  -mwide-bitfields
           -mno-wide-bitfields -m4byte-functions  -mno-4byte-functions
           -mcallgraph-data -mno-callgraph-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 -mpaired-single  -mips3d -mint64  -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-divi-
           sion -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-excep-
           tions -mvr4130-align -mno-vr4130-align
 
           MMIX Options -mlibfuncs  -mno-libfuncs  -mepsilon  -mno-epsilon
           -mabi=gnu -mabi=mmixware  -mzero-extend  -mknuthdiv  -mto-
           plevel-symbols -melf  -mbranch-predict  -mno-branch-predict
           -mbase-addresses -mno-base-addresses  -msingle-exit  -mno-sin-
           gle-exit
 
           MN10300 Options -mmult-bug  -mno-mult-bug -mam33  -mno-am33
           -mam33-2  -mno-am33-2 -mno-crt0  -mrelax
 
           NS32K Options -m32032  -m32332  -m32532  -m32081  -m32381
           -mmult-add  -mnomult-add  -msoft-float  -mrtd -mnortd -mregparam
           -mnoregparam  -msb  -mnosb -mbitfield  -mnobitfield  -mhimem  -mno-
           himem
 
           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  -mab-
           shi  -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-pow-
           erpc -maltivec  -mno-altivec -mpowerpc-gpopt  -mno-powerpc-gpopt
           -mpowerpc-gfxopt  -mno-powerpc-gfxopt -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  -mmulti-
           ple  -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
           -mrelocatable-lib  -mno-relocatable-lib -mtoc -mno-toc  -mlittle
           -mlittle-endian  -mbig  -mbig-endian -mdynamic-no-pic -mpriori-
           tize-restricted-insns=priority -msched-costly-dep=dependence_type
           -minsert-sched-nops=scheme -mcall-sysv  -mcall-netbsd
           -maix-struct-return  -msvr4-struct-return -mabi=altivec
           -mabi=no-altivec -mabi=spe  -mabi=no-spe -misel=yes  -misel=no
           -mspe=yes  -mspe=no -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  -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-single-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 -mnomacsave
           -mieee  -misize  -mpadstruct  -mspace -mprefergot  -musermode
 
           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
 
           System V Options -Qy  -Qn  -YP,paths  -Ym,dir
 
           TMS320C3x⁄C4x Options -mcpu=cpu  -mbig  -msmall  -mregparm  -mmem-
           parm -mfast-fix  -mmpyi  -mbk  -mti  -mdp-isr-reload -mrpts=count
           -mrptb  -mdb  -mloop-unsigned -mparallel-insns  -mparallel-mpy
           -mpreserve-float
 
           V850 Options -mlong-calls  -mno-long-calls  -mep  -mno-ep -mpro-
           log-function  -mno-prolog-function  -mspace -mtda=n  -msda=n
           -mzda=n -mapp-regs  -mno-app-regs -mdisable-callt  -mno-dis-
           able-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-section-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 -fasynchronous-unwind-tables
           -finhibit-size-directive  -finstrument-functions -fno-common
           -fno-ident -fpcc-struct-return  -fpic  -fPIC -fpie -fPIE
           -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 -fargu-
           ment-alias  -fargument-noalias -fargument-noalias-global  -flead-
           ing-underscore -ftls-model=model -ftrapv  -fwrapv  -fbounds-check
           -fvisibility
 
       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 exe-
       cutable 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++ program 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.hh
       file.H
           C++ header file to be turned into a precompiled header.
 
       file.f
       file.for
       file.FOR
           Fortran source code which should not be preprocessed.
 
       file.F
       file.fpp
       file.FPP
           Fortran source code which must be preprocessed (with the tradi-
           tional preprocessor).
 
       file.r
           Fortran source code which must be preprocessed with a RATFOR pre-
           processor (not included with GCC).
 
       file.f90
       file.f95
           Fortran 90⁄95 source code which should not be preprocessed.
 
       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 subprogram 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
                   f77  f77-cpp-input  ratfor
                   f95
                   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 numeri-
           cally 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-assem-
           bler 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 preprocessed 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.suffix 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 com-
           piler driver program and of the preprocessor and the compiler
           proper.
 
       -###
           Like -v except the commands are not executed and all command argu-
           ments 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 intermodule analysis (IMA) to be performed by the compiler.
           Currently the only language for which this is supported 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 appropri-
           ate 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 con-
           junction 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 dis-
           played.
 
       --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 com-
       piler the same way as for compiling C programs (usually with the name
       gcc).
 
       However, C++ programs often require class libraries as well as a com-
       piler that understands the C++ language---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 language 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 com-
       mand-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 seman-
           tics 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-4.0⁄c99status.html> for more
               information.  The names c9x and iso9899:199x are deprecated.
 
           gnu89
               Default, ISO C90 plus GNU extensions (including some C99 fea-
               tures).
 
           gnu99
           gnu9x
               ISO C99 plus GNU extensions.  When ISO C99 is fully implemented
               in GCC, this will become the default.  The name gnu9x is depre-
               cated.
 
           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.
 
       -aux-info filename
           Output to the given filename prototyped declarations for all func-
           tions declared and⁄or defined in a translation unit, including
           those in header files.  This option is silently ignored in any lan-
           guage 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, respectively, 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 function.  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-func-
           tion 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 freestanding environment is one in
           which the standard library may not exist, and program startup may
           not necessarily 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 -trigraphs.
 
       -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 (inter-
           nal 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 supported 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.  Likewise, 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 meaning-
       ful 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 compile 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 version that conforms most closely to the C++ ABI specifica-
           tion.  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 allocated.  This check is
           normally unnecessary because the C++ standard specifies that "oper-
           ator 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, memory 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.
 
       -fno-const-strings
           Give string constants type "char *" instead of type "const char *".
           By default, G++ uses type "const char *" as required by the stan-
           dard.  Even if you use -fno-const-strings, you cannot actually mod-
           ify the value of a string constant.
 
           This option might be removed in a future release of G++.  For maxi-
           mum 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 tempo-
           rary 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 check for violation of exception specifications at runtime.
           This option violates the C++ standard, but may be useful for reduc-
           ing code size in production builds, much like defining NDEBUG.  The
           compiler will still optimize based on the exception specifications.
 
       -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 instantia-
           tions.
 
       -fno-implicit-inline-templates
           Don't emit code for implicit instantiations of inline templates,
           either.  The default is to handle inlines differently so that com-
           piles with and without optimization will need the same set of
           explicit instantiations.
 
       -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 member function via non-stan-
           dard 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 synonyms 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 compi-
           lation.  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 end-
           less recursions during template class instantiation.  ANSI⁄ISO C++
           conforming programs 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 destructors, but will only work if your C library supports
           "__cxa_atexit".
 
       -fvisibility-inlines-hidden
           Causes all inlined methods to be marked with "__attribute__ ((visi-
           bility ("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.  While
           it can cause bloating through duplication of code within each DSO
           where it is used, often the wastage is less than the considerable
           space occupied by a long symbol name in the export table which is
           typical when using templates and namespaces.  For even more sav-
           ings, combine with the -fvisibility=hidden switch.
 
       -fno-weak
           Do not use weak symbol support, even if it is provided by the
           linker.  By default, G++ will use weak symbols if they are avail-
           able.  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 func-
           tions; 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 generated 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 prob-
               lem 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 exam-
               ple:
 
                       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 mangled 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 warn-
           ing 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 members, 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 incre-
               ment 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 be 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 specifi-
           cation 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 func-
           tion.  (Section 14.5.3).  Before G++ implemented explicit specifi-
           cation, unqualified-ids could be interpreted as a particular spe-
           cialization of a templatized function.  Because this non-conforming
           behavior is no longer the default behavior for G++, -Wnon-tem-
           plate-friend allows the compiler to check existing code for poten-
           tial trouble spots and is on by default.  This new compiler behav-
           ior can be turned off with -Wno-non-template-friend which keeps the
           conformant compiler code but disables the helpful warning.
 
       -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 (static_cast, reinter-
           pret_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 pre-
           serve unsignedness, 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 meaning-
       ful for Objective-C and Objective-C++ programs, but you can also use
       most of the language-independent GNU compiler options.  For example,
       you might compile 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 lan-
       guage supported by GCC.
 
       Note that since Objective-C is an extension of the C language, Objec-
       tive-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 "NSConstantString" 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 run-
           time to be used.  Currently, this option is only available in con-
           junction with the NeXT runtime on Mac OS X 10.3 and later.
 
       -fobjc-exceptions
           Enable syntactic support for structured exception handling in
           Objective-C, similar to what is offered by C++ and Java.  Cur-
           rently, 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 synchroniza-
           tion 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 @syn-
           chronized blocks is allowed, and will cause the guarding object to
           be unlocked properly.
 
       -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-Con-
           tinue debugging mode, where the object file in question may be
           recompiled and dynamically reloaded in the course of program execu-
           tion, 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.
 
       -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 explicitly implemented in the class, even if a method implemen-
           tation is inherited from the superclass.  If you use the -Wno-pro-
           tocol option, then methods inherited from the superclass are con-
           sidered 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 "@selec-
           tor(...)"  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 compilation, 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.
 
       -Wundeclared-selector
           Warn if a "@selector(...)" expression referring to an undeclared
           selector is found.  A selector is considered undeclared if no
           method with that name has been declared before the "@selector(...)"
           expression, either explicitly in an @interface or @protocol decla-
           ration, 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 format-
       ting 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 cor-
       rectly.
 
       -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 continuation 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.
 
       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 extensions, and some other programs
           that do not follow ISO C and ISO C++.  For ISO C, follows the ver-
           sion 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++ fea-
           tures 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__".  How-
           ever, 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 diag-
           nostics 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 warn-
           ings.
 
       -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 fur-
           ther 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", "strftime" and
           "strfmon" (an X⁄Open extension, not in the C standard) families (or
           other target-specific families).  Which functions are checked with-
           out format attributes having been specified depends on the standard
           version selected, and such checks of functions without the
           attribute specified are disabled by -ffreestanding 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 for-
           mat 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 -Wnonnull.
 
           -Wformat is included in -Wall.  For more control over some aspects
           of format checking, the options -Wformat-y2k, -Wno-for-
           mat-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 specifications, 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 Speci-
           fication 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 problems.  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 cur-
           rently a subset of what -Wformat-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.  Cur-
           rently equivalent to -Wformat -Wformat-nonliteral -Wformat-security
           -Wformat-y2k.
 
       -Wnonnull
           Warn about passing a null pointer for arguments marked as requiring
           a non-null value by the "nonnull" function 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 them-
           selves.  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 initializer for a is not fully brack-
           eted, 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 different 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 viola-
           tions of sequence point rules in the C standard.
 
           The C standard defines the order in which expressions in a C pro-
           gram 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 expression (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 argu-
           ments and the expression denoting the called function), and in cer-
           tain 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.  How-
           ever, 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.  Programs 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 evaluation 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 complicated cases are not
           diagnosed by this option, and it may give an occasional false posi-
           tive 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 pro-
           grams.  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 sub-
           tle cases.  Links to discussions of the problem, including proposed
           formal definitions, 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 quali-
           fier such as "const".  Such a type qualifier 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 diagnos-
           tic 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 enu-
           meration.  (The presence of a "default" label prevents this warn-
           ing.)  "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 enu-
           meration.  "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 comments 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 decla-
           ration.
 
           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 initial-
           ized 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 opti-
           mizing.  If you don't specify -O, you simply won't get these warn-
           ings.
 
           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 ele-
           ments of structure, union or array variables as well as for vari-
           ables 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 variables or elements
           for which there are warnings will depend on the precise optimiza-
           tion 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 warn-
           ings 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 ini-
           tialized, 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 pos-
           sible 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 func-
           tions you use that never return as "noreturn".
 
           This warning is enabled by -Wall.
 
       -Wunknown-pragmas
           Warn when a #pragma directive is encountered which is not under-
           stood 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.
 
       -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 warn-
           ings 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 Objective-C and
           Objective-C++ Dialect Options.
 
       The following -W... options are not implied by -Wall.  Some of them
       warn about constructions that users generally do not consider question-
       able, but which occasionally you might wish to check for; others warn
       about constructions 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 sup-
           ported, but the newer name is more descriptive.)  Print extra warn-
           ing 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 with-
               out 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 suppress 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 argu-
               ments.
 
           *   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 speci-
               fied.)
 
           *   An aggregate has an initializer which does not initialize all
               members.  This warning can be independently 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 regis-
               ter.
 
           *<(C++ only)>
               A base class is not initialized in a derived class' copy con-
               structor.
 
       -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 legiti-
           mate way of obtaining infinities and NaNs.
 
       -Wsystem-headers
           Print warning messages for constructs found in system header files.
           Warnings from system headers are normally 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 compute (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 probably mistaken.
 
       -Wtraditional (C only)
           Warn about certain constructs that behave differently in tradi-
           tional and ISO C.  Also warn about ISO C constructs 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 replacement 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 under-
               stands but would ignore because the # does not appear as the
               first character on the line.  It also suggests you hide direc-
               tives 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 system 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 sepa-
               rate 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 missing 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 inten-
               tionally is not issued for prototype declarations or variadic
               functions because these ISO C features will appear in your code
               when using libiberty's traditional C compatibility macros,
               "PARAMS" and "VPARAMS".  This warning is also bypassed for
               nested functions because that feature is already a GCC exten-
               sion 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 introduced 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.
 
       -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 conve-
           nience in calculations with "void *" pointers and pointers to func-
           tions.
 
       -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 *".
 
       -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 conversion 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 care-
           ful about using "const" in declarations 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 proto-
           type.  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 assign-
           ment "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.)
 
       -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 prototype.
 
       -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 declara-
           tion.  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 "nore-
           turn".  Note these are only possible candidates, not absolute ones.
           Care should be taken to manually verify functions actually do not
           ever return before adding the "noreturn" attribute, otherwise sub-
           tle code generation bugs could be introduced. You will not get a
           warning for "main" in hosted C environments.
 
       -Wmissing-format-attribute
           If -Wformat is enabled, also warn about functions which might be
           candidates for "format" attributes.  Note these are only possible
           candidates, not absolute ones.  GCC will guess that "format"
           attributes might be appropriate for any function that calls a func-
           tion like "vprintf" or "vscanf", but this might not always be the
           case, and some functions for which "format" attributes are appro-
           priate may not be detected.  This option has no effect unless
           -Wformat is enabled (possibly by -Wall).
 
       -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 implementa-
           tion-defined values, and should not be used in portable code.
 
       -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 ele-
           ment 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 proce-
           dure 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 ver-
           sion of a program there is often substantial code which checks cor-
           rect functioning of the program and is, hopefully, unreachable
           because the program does work.  Another common use of unreachable
           code is to provide behavior which is selectable at compile-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 program
           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++ standard, applying offsetof to
           a non-POD type is undefined.  In existing C++ implementations, how-
           ever, 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 non-
           portable code and who have deliberately chosen to ignore the warn-
           ing about it.
 
           The restrictions on offsetof may be relaxed in a future version of
           the C++ standard.
 
       -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-vari-
           adic-macros.
 
       -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 han-
           dle 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.
 
       -Wno-pointer-sign
           Don't warn for pointer argument passing or assignment with differ-
           ent signedness.  Only useful in the negative form since this warn-
           ing is enabled by default.  This option is only supported for C and
           Objective-C.
 
       -Werror
           Make all warnings into errors.
 
       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 for-
           mat (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 con-
           trol 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 surprising 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 values 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 format.
 
       -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 sup-
           ported), 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 sup-
           ported), for only symbols that are actually used.
 
       -gstabs+
           Produce debugging information in stabs format (if that is sup-
           ported), using GNU extensions understood only by the GNU debugger
           (GDB).  The use of these extensions is likely to make other debug-
           gers crash or refuse to read the program.
 
       -gcoff
           Produce debugging information in COFF format (if that is sup-
           ported).  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 sup-
           ported).  This is the format used by the DBX debugger on IBM
           RS⁄6000 systems.
 
       -gxcoff+
           Produce debugging information in XCOFF format (if that is sup-
           ported), using GNU extensions understood only by the GNU debugger
           (GDB).  The use of these extensions is likely to make other debug-
           gers 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 sup-
           ported).  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 defini-
           tions present in the program.  Some debuggers 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 differ-
           ent 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 execu-
           tion the program records how many times each branch and call is
           executed and how many times it is taken or returns.  When the com-
           piled program exits it saves this data to a file called aux-
           name.gcda for each source file.  The data may be used for profile-
           directed optimizations (-fbranch-probabilities), or for test cover-
           age analysis (-ftest-coverage).  Each object file's auxname is gen-
           erated 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).
 
           @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 pro-
               file every source file in a program.
 
           @cvmmfu
               Link your object files with -lgcov or -fprofile-arcs (the lat-
               ter 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 spanning tree for the graph.
           Only arcs that are not on the spanning tree have to be instru-
           mented: 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.
 
       -ftree-based-profiling
           This option is used in addition to -fprofile-arcs or -fbranch-prob-
           abilities to control whether those optimizations are performed on a
           tree-based or rtl-based internal representation.  If you use this
           option when compiling with -fprofile-arcs, you must also use it
           when compiling later with -fbranch-probabilities.  Currently the
           tree-based optimization is in an early stage of development, and
           this option is recommended only for those people working on improv-
           ing it.
 
       -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 aux-
           name.gcno.  Refer to the -fprofile-arcs option above for a descrip-
           tion 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 debugging 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 explicitly 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.com-
               bine.
 
           -dC
           -fdump-rtl-ce1
           -fdump-rtl-ce2
               -dC and -fdump-rtl-ce1 enable dumping after the first if con-
               version, 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 instruc-
               tion is also printed.
 
           -dP Dump the RTL in the assembler output as a comment before each
               instruction.  Also turns on -dp annotation.
 
           -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.  Usu-
               ally 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 instruc-
           tion numbers and line number note output.  This makes it more fea-
           sible to use diff on debugging dumps for compiler invocations with
           different options, in particular with and without -g.
 
       -fdump-translation-unit (C and C++ only)
       -fdump-translation-unit-options (C and C++ only)
           Dump a representation of the tree structure for the entire transla-
           tion 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 func-
           tion 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 possible:
 
           all Enables all inter-procedural analysis dumps; currently the only
               produced dump is the cgraph dump.
 
           cgraph
               Dumps information about call-graph optimization, unused func-
               tion removal, and inlining decisions.
 
       -fdump-tree-switch (C and C++ only)
       -fdump-tree-switch-options (C and C++ only)
           Control the dumping at various stages of processing the intermedi-
           ate language tree to a file.  The file name is generated by append-
           ing 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 applica-
           ble to all dumps, those which are not meaningful will be ignored.
           The following options are available
 
           address
               Print the address of each node.  Usually this is not meaningful
               as it changes according to the environment and source file.
               Its primary use is for tying up a dump file with a debug envi-
               ronment.
 
           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 representation.
 
           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 appending .vcg to the source
               file name.  Note that if the file contains more than one func-
               tion, 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.
 
           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 append-
               ing .ccp 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.
 
           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 appending .sra 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 appending .phiopt to the source
               file name.
 
           forwprop
               Dump each function after forward propagating single use vari-
               ables.  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 opti-
               mization 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.
 
           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 generate 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 scheduler prints.  This informa-
           tion 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 out-
           put is always printed to standard error.
 
           For n greater than zero, -fsched-verbose outputs the same informa-
           tion 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-verbose 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 prepro-
           cessed foo.i output file even though the compiler now normally 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 intermediate 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 infor-
           mation).
 
           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 command 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-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 prob-
           lem, 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 any-
           thing 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, regard-
           less 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 statement 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 -O2 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 opti-
       mizations 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 com-
           pilation time.
 
           -O turns on the following optimization flags: -fdefer-pop -fde-
           layed-branch -fguess-branch-probability -fcprop-registers
           -floop-optimize -fif-conversion -fif-conversion2 -ftree-ccp
           -ftree-dce -ftree-dominator-opts -ftree-dse -ftree-ter -ftree-lrs
           -ftree-sra -ftree-copyrename -ftree-fre -ftree-ch -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 command line to be enabled for
           the Ada compiler.
 
       -O2 Optimize even more.  GCC performs nearly all supported optimiza-
           tions that do not involve a space-speed tradeoff.  The compiler
           does not perform loop unrolling or function inlining when you spec-
           ify -O2.  As compared 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 -fre-
           run-cse-after-loop  -frerun-loop-opt -fcaller-saves -fforce-mem
           -fpeephole2 -fschedule-insns  -fschedule-insns2 -fsched-interblock
           -fsched-spec -fregmove -fstrict-aliasing
           -fdelete-null-pointer-checks -freorder-blocks -freorder-functions
           -funit-at-a-time -falign-functions  -falign-jumps -falign-loops
           -falign-labels -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-functions, -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 optimiza-
           tions 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
 
           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 optimiza-
       tions 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 com-
           piled 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 several 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 com-
           mon subexpressions, instruction combination should eliminate the
           separate register-load.
 
           Enabled at levels -O2, -O3, -Os.
 
       -fforce-addr
           Force memory address constants to be copied into registers before
           doing arithmetic on them.  This may produce better code just as
           -fforce-mem may.
 
       -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 automatically handles the frame
           pointer and nothing is saved by pretending it doesn't exist.  The
           machine-description 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-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 templates 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:
 
            @item max-inline-insns-single
             is set to I<n>⁄2.
            @item max-inline-insns-auto
             is set to I<n>⁄2.
            @item min-inline-insns
             is set to 130 or I<n>⁄4, whichever is smaller.
            @item max-inline-insns-rtl
             is set to I<n>.
 
           See below for a documentation of the individual parameters control-
           ling inlining.
 
           Note: pseudo instruction represents, in this particular context, an
           abstract measurement of function's size.  In no way, it represents
           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 func-
           tions into the object file.
 
       -fkeep-static-consts
           Emit variables declared "static const" when optimization isn't
           turned on, even if the variables aren't referenced.
 
           GCC enables this option by default.  If you want to force the com-
           piler to check if the variable was referenced, 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 assem-
           bler 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-con-
           stants this considers e.g. even constant initialized arrays or ini-
           tialized 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-conform-
           ing 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 regis-
           ter, 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 sup-
           port 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 optimiza-
           tions 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 explic-
           itly 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 assumptions based on that.
 
           The default is -fzero-initialized-in-bss.
 
       -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 respec-
           tively.
 
       -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 run-
           time library (libmudflap), which will be linked into a program if
           -fmudflap is given at link time.  Run-time behavior of the instru-
           mented program is controlled by the MUDFLAP_OPTIONS environment
           variable.  See "env MUDFLAP_OPTIONS=-help a.out" for its options.
 
           Use -fmudflapth instead of -fmudflap to compile and to link if your
           program is multi-threaded.  Use -fmudflapir, in addition to -fmud-
           flap or -fmudflapth, if instrumentation should ignore pointer
           reads.  This produces less instrumentation (and therefore faster
           execution) and still provides some protection against outright mem-
           ory corrupting writes, but allows erroneously read data to propa-
           gate within a program.
 
       -fstrength-reduce
           Perform the optimizations of loop strength reduction and elimina-
           tion 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 subsumed 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 exten-
           sion, you may get better runtime performance 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 elimi-
           nation 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 purpose 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 opti-
           mizations (loop unrolling, peeling and unswitching, loop invariant
           motion) are enabled by separate flags.
 
       -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 equiva-
           lents.  This include use of conditional moves, min, max, set flags
           and abs instructions, and some tricks doable by standard arith-
           metics.  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 condi-
           tional 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 expen-
           sive.
 
           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 optimiza-
           tion.
 
           Enabled at levels -O2, -O3, -Os.
 
       -fdelayed-branch
           If supported for the target machine, attempt to reorder instruc-
           tions 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 instruc-
           tions to eliminate execution stalls due to required data being
           unavailable.  This helps machines that have slow floating point or
           memory load instructions by allowing other instructions to be
           issued until the result of the load or floating point instruction
           is required.
 
           Enabled at levels -O2, -O3, -Os.
 
       -fschedule-insns2
           Similar to -fschedule-insns, but requests an additional pass of
           instruction scheduling after register allocation has been done.
           This is especially useful on machines with a relatively small num-
           ber 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 scheduling before register alloca-
           tion, 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=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.
 
       -fsched-stalled-insns-dep=n
           Define how many insn groups (cycles) will be examined for a depen-
           dency on a stalled insn that is candidate for premature removal
           from the queue of stalled insns.  Has an effect only during the
           second scheduling pass, and only if -fsched-stalled-insns is used
           and its value is not zero.
 
       -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 algo-
           rithm.
 
           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 regis-
           ter 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 scheduling after register allocation, i.e. with -fsched-
           ule-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, usu-
           ally 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-ccp
           Perform sparse conditional constant propagation (CCP) on trees.
           This flag is enabled by default at -O 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 propaga-
           tion, redundancy elimination, range propagation and expression sim-
           plification) based on a dominator tree traversal.  This also per-
           forms 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 invari-
           antness 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 analy-
           sis.  Later optimizations then may determine the number easily.
           Useful especially 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 com-
           piler temporaries to other variables at copy locations, usually
           resulting in variable names which more closely resemble the origi-
           nal 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-GIM-
           PLE 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.  Dis-
           tinct 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.
 
       -ftracer
           Perform tail duplication to enlarge superblock size.  This trans-
           formation 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 com-
           pile time or upon entry to the loop.  -funroll-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 -fun-
           roll-loops,
 
       -fsplit-ivs-in-unroller
           Enables expressing of values of induction variables in later itera-
           tions of the unrolled loop using the value in the first iteration.
           This breaks long dependency chains, thus improving efficiency of
           the scheduling passes (for best results, -fweb should be used as
           well).
 
           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 differ-
           ence between -fno-peephole and -fno-peephole2 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 heuris-
           tics are based on the control flow graph.  If some branch probabil-
           ities 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 under-
           stand.
 
           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 special subsections
           ".text.hot" for most frequently executed functions and
           ".text.unlikely" for unlikely executed functions.  Reordering is
           done by the linker so object file format must support named sec-
           tions 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 applica-
           ble to the language being compiled.  For C (and C++), this acti-
           vates optimizations based on the type of expressions.  In particu-
           lar, 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 mem-
           ory 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 anal-
           ysis should define a function that computes, 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-func-
           tions=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 opera-
           tions.
 
           -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 executed.
 
           -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 emit-
               ted, 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 are removed.  This may result in undefined references
               when an "asm" statement refers directly to variables or func-
               tions that are otherwise unused.  In that case either the vari-
               able⁄function shall be listed as an operand of the "asm" state-
               ment 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 -O2, -O3.
 
       -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, however, make
           debugging impossible, since variables will no longer stay in a
           ``home register''.
 
           Enabled at levels -O2, -O3, -Os, on targets where the default for-
           mat for debugging information supports variable tracking.
 
       -fno-cprop-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.
 
           Disabled at levels -O, -O2, -O3, -Os.
 
       -fprofile-generate
           Enable options usually used for instrumenting application to pro-
           duce profile useful for later recompilation with profile feedback
           based optimization.  You must use -fprofile-generate both when com-
           piling and when linking your program.
 
           The following options are enabled: "-fprofile-arcs", "-fpro-
           file-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".
 
       The following options control compiler behavior regarding floating
       point arithmetic.  These options trade off between speed and correct-
       ness.  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 registers (of the 68881) keep more
           precision than a "double" is supposed to have.  Similarly for the
           x86 architecture.  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-trap-
           ping-math, -ffinite-math-only, -fno-rounding-math, -fno-signal-
           ing-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 func-
           tions.
 
       -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 func-
           tions.
 
           The default is -fmath-errno.
 
       -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 func-
           tions.
 
           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 gener-
           ate 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 arithmetic, 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 func-
           tions.
 
           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 transforma-
           tions 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 operations.  Setting this
           option disables optimizations that may change the number of excep-
           tions visible with signaling NaNs.  This option implies -ftrap-
           ping-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
       -fno-cx-limited-range
           When enabled, this option states that a range reduction step is not
           needed when performing complex division.  The default is
           -fno-cx-limited-range, but is enabled by -ffast-math.
 
           This option controls the default setting of the ISO C99 "CX_LIM-
           ITED_RANGE" pragma.  Nevertheless, the option applies to all lan-
           guages.
 
       The following options control optimizations that may improve perfor-
       mance, 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 com-
           pile it a second time using -fbranch-probabilities, 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 optimiza-
           tion.  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.
 
       -fspeculative-prefetching
           If combined with -fprofile-arcs, it instructs the compiler to add a
           code to gather information about addresses of memory references in
           the program.
 
           With -fbranch-probabilities, it reads back the data gathered and
           issues prefetch instructions according to them.  In addition to the
           opportunities noticed by -fprefetch-loop-arrays, it also notices
           more complicated memory access patterns---for example accesses to
           the data stored in linked list whose elements are usually allocated
           sequentially.
 
           In order to prevent issuing double prefetches, usage of -fspecula-
           tive-prefetching implies -fno-prefetch-loop-arrays.
 
           Enabled with -fprofile-generate and -fprofile-use.
 
       -frename-registers
           Attempt to avoid false dependencies in scheduled code by making use
           of registers left over after register allocation.  This optimiza-
           tion will most benefit processors with lots of registers.  Depend-
           ing 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''.
 
           Not enabled by default at any level because it has known bugs.
 
       -ftracer
           Perform tail duplication to enlarge superblock size.  This trans-
           formation 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 com-
           pile time or upon entry to the loop.  -funroll-loops implies -fre-
           run-cse-after-loop.  It also turns on complete loop peeling (i.e.
           complete removal of loops with small constant number of itera-
           tions).  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 iterations).
 
           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 output file.
 
           Use these options on systems where the linker can perform optimiza-
           tions to improve locality of reference in the instruction space.
           Most systems using the ELF object format and SPARC processors run-
           ning 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 debugging 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.
 
       --param name=value
           In some places, GCC uses various constants to control the amount of
           optimization that is done.  For example, GCC will not inline func-
           tions that contain more that a certain number of instructions.  You
           can control 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:
 
           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 num-
               ber of bytes in the complete 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 crossjump-
               ing.  The algorithm used by -fcrossjumping 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-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 arbi-
               trary 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 exe-
               cutable run time.
 
           max-delay-slot-live-search
               When trying to fill delay slots, the maximum number of instruc-
               tions 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 exces-
               sively 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 considered 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-lin-
               ear 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 func-
               tion growth to 2.0 times the original size.
 
           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-recur-
               sive is taken into acount.  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 inlin-
               ing.
 
               For functions declared inline --param max-inline-recursive-
               depth is taken into acount.  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.
 
           inline-call-cost
               Specify cost of call instruction relative to simple arithmetics
               operations (having cost of 1).  Increasing this cost disqualify
               inlinining of non-leaf functions and at same time increase size
               of leaf function that is believed to reduce function size by
               being inlined.  In effect it increase amount of inlining for
               code having large abstraction penalty (many functions that just
               pass the argumetns to other functions) and decrease inlining
               for code with low abstraction penalty.  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 deter-
               mines 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 invari-
               ant 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 con-
               tain 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 ana-
               lyzer.  Large expressions slow the analyzer.
 
           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 consid-
               ered 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 consid-
               ered hot
 
           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 profiles (as opposed to stati-
               cally 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 per-
               centage.  This is rather hokey argument, as most of the dupli-
               cates will be eliminated later in cross jumping, so it may be
               set to much higher values 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 compila-
               tion without.  The value for compilation with profile feedback
               needs to be more conservative (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.
 
           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-clobbered variables at function call sites.  This artifi-
               cial 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 alias-
               ing at the expense of precision loss in alias information.
 
           ggc-min-expand
               GCC uses a garbage collector to manage its own memory alloca-
               tion.  This parameter specifies the minimum percentage by which
               the garbage collector's heap should be allowed to expand
               between collections.  Tuning 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 particu-
               lar 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-heap-
               size.  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 values mean more aggres-
               sive optimization, making the compile time increase with proba-
               bly slightly better performance.  The default value is 100.
 
           max-cselib-memory-location
               The maximum number of memory locations cselib should take into
               acount.  Increasing values mean more aggressive optimization,
               making the compile time increase with probably slightly better
               performance.  The default value is 500.
 
           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-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.
 
           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, reduc-
               ing 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.
 
       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 modified, translated or interpreted by the com-
           piler 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 argu-
           ment.
 
       -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 pro-
           cessed 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 spe-
           cial 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 interpreta-
           tion 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, -Wtrigraphs, -Wmultichar and a warn-
           ing 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
           @anchor{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.  Therefore, 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 tri-
           graph conversion without warnings, but get the other -Wall warn-
           ings, use -trigraphs -Wall -Wno-trigraphs.
 
       -Wtraditional
           Warn about certain constructs that behave differently in tradi-
           tional and ISO C.  Also warn about ISO C constructs 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 condi-
           tional 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 something 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 diag-
           nostics into errors.  This includes mandatory 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 dependencies 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.
 
           @anchor{dashMF}
 
       -MF file
           When used with -M or -MM, specifies a file to write the dependen-
           cies 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 gen-
           eration, -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 preprocessed out-
           put, 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 plat-
           form'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 compilation 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 depen-
           dency-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 "<file-
           name>"" 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 differ-
           ent 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 conformance 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 recog-
           nize 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; others 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 directory 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 direc-
           tories, 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 direc-
           tory containing 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 declarations.
 
           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 speci-
           fied 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.  -iwithpre-
           fixbefore puts it in the same place -I would; -iwithprefix puts it
           where -idirafter would.
 
       -isystem dir
           Search dir for header files, after all directories specified by -I
           but before the standard system directories.  Mark it as a system
           directory, so that it gets the same special treatment as is applied
           to the standard 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
           @anchor{fdollars-in-identifiers} Accept $ in identifiers.
 
       -fpreprocessed
           Indicate to the preprocessor that the input file has already been
           preprocessed.  This suppresses things like macro expansion, tri-
           graph conversion, escaped newline splicing, and processing of most
           directives.  The preprocessor still recognizes and removes com-
           ments, so that you can pass a file preprocessed with -C to the com-
           piler 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 exten-
           sions .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 con-
           stants.  The default is UTF-8.  charset can be any encoding sup-
           ported 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 charac-
           ter 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 com-
           mand line option takes precedence if there's a conflict.  charset
           can be any encoding supported by the system's "iconv" library rou-
           tine.
 
       -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 preproces-
           sor 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 emitted as the current working directory in
           some debugging information formats.  This option is implicitly
           enabled if debugging information is enabled, but this can be inhib-
           ited with the negated form -fno-working-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 under-
           stand 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 direc-
               tives 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 prede-
               fined 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 prepro-
               cessing.
 
       -P  Inhibit generation of linemarkers in the output from the preproces-
           sor.  This might be useful when running the preprocessor on some-
           thing 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 processed 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 ordi-
           nary 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 charac-
           ter 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 pre-
           processing 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 nor-
           mal 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 pre-
           process 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 com-
           mas, 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 alter-
           native 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 liblibrary.a.  The linker then uses
           this file as if it had been specified precisely by name.
 
           The directories searched include several standard system directo-
           ries 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 fash-
           ion.  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 normally, 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 "mem-
           cmp", "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 -nodefault-
           libs is libgcc.a, a library of internal subroutines that GCC uses
           to overcome shortcomings of particular machines, or special needs
           for some languages.
 
           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++ con-
           structors 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.
 
       -s  Remove all symbol table and relocation information from the exe-
           cutable.
 
       -static
           On systems that support dynamic linking, this prevents linking with
           the shared libraries.  On other systems, 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 con-
           figured, 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 different 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, with-
           out 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 sup-
           port 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 rec-
           ognize.
 
           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 ven-
           dor-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 normal position in the system include chain.  This
           is to ensure that GCC's procedure to fix buggy system headers 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 compiler 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 environ-
           ment 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 stan-
           dard 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 com-
           piler.
 
       -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.
 
       -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 direc-
           tory (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
       inconvenient, 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 exam-
           ple, if a cross-compiler was configured with configure i386v, mean-
           ing to compile for an 80386 running System V, then you would spec-
           ify -b i386v to run that cross compiler.
 
       -V version
           The argument version specifies which version of GCC to run.  This
           is useful when multiple versions are installed.  For example, ver-
           sion might be 2.0, meaning to run GCC version 2.0.
 
       The -V and -b options work by running the <machine>-gcc-<version> exe-
       cutable, 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 differ-
       ent 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 spe-
       cial 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.
 
       These options are defined by the macro "TARGET_SWITCHES" in the machine
       description.  The default for the options is also defined by that
       macro, which enables you to change the defaults.
 
       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 multi-
           ple-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-sec-
           tion, and readonly-data-section respectively by default.  This can
           be overridden with the "section" attribute.
 
       ARM Options
 
       These -m options are defined for Advanced RISC Machines (ARM) architec-
       tures:
 
       -mabi=name
           Generate code for the specified ABI.  Permissible values are: apcs-
           gnu, atpcs, aapcs and iwmmxt.
 
       -mapcs-frame
           Generate a stack frame that is compliant with the ARM Procedure
           Call Standard for all functions, even if this is not strictly nec-
           essary 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 func-
           tion's body.  This means that all functions will start with a rec-
           ognizable set of instructions (or in fact one of a choice from a
           small set of different function prologues), and this information
           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.  Warn-
           ing: 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 configurations.
 
       -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 pro-
           cessors.  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 versions 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 instructions it can emit when gener-
           ating 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, stron-
           garm110, 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 processor 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 instruc-
           tions that it will generate based on the cpu specified by a -mcpu=
           option.  For some ARM implementations 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.  Permissible names are: fpa, fpe2,
           fpe3, maverick, vfp.  -mfp and -mfpe are synonyms for -mfpu=fpenum-
           ber, for compatibility 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 mul-
           tiple 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 supports
           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 sub-
           routine 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 excep-
           tion 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 appro-
           priate 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 instruction 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-cir-
           rus-fix-invalid-insns switch.
 
       -mpoke-function-name
           Write the name of each function into the text section, directly
           preceding the function prologue.  The generated 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 loca-
           tion "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 executing a function pointer if this option is enabled.
 
       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 programs (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 sub-
           routines.  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 comply 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 occur-
           ring.
 
       -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 conditional branch.
 
       -mlow-64k
           When enabled, the compiler is free to take advantage of the knowl-
           edge 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 effi-
           cient 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 sub-
           routine 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=archi-
           tecture-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 indi-
           cations 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 constants 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 align-
           ment.
 
       -mno-prologue-epilogue
       -mprologue-epilogue
           With -mno-prologue-epilogue, the normal function prologue and epi-
           logue 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 functions from a simulator
           library.  Code, initialized data and zero-initialized data are
           allocated consecutively.
 
       -sim2
           Like -sim, but pass linker options to locate initialized data at
           0x40000000 and zero-initialized data at 0x80000000.
 
       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 mis-
       match.  The assembler, as, will only permit 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 quietly 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 directo-
           ries to be searched for header files.  These directories are inter-
           leaved 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 framework 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 sib-
           lings 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 subframe-
           works, in the future, the mechanism may be extended to support
           this.  The standard frameworks can be found in "⁄Sys-
           tem⁄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 -feliminate-unused-debug-symbols.
           This is by default ON.
 
       -gfull
           Emit debugging information for all symbols and types.
 
       -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 com-
           piling 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 mod-
           ules 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 pro-
           grams.  -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 archi-
           tecture to be fatal.
 
       -bind_at_load
           Causes the output file to be marked such that the dynamic linker
     &