1@c Copyright (C) 1988-2014 Free Software Foundation, Inc. 2@c This is part of the GCC manual. 3@c For copying conditions, see the file gcc.texi. 4 5@ignore 6@c man begin INCLUDE 7@include gcc-vers.texi 8@c man end 9 10@c man begin COPYRIGHT 11Copyright @copyright{} 1988-2014 Free Software Foundation, Inc. 12 13Permission is granted to copy, distribute and/or modify this document 14under the terms of the GNU Free Documentation License, Version 1.3 or 15any later version published by the Free Software Foundation; with the 16Invariant Sections being ``GNU General Public License'' and ``Funding 17Free Software'', the Front-Cover texts being (a) (see below), and with 18the Back-Cover Texts being (b) (see below). A copy of the license is 19included in the gfdl(7) man page. 20 21(a) The FSF's Front-Cover Text is: 22 23 A GNU Manual 24 25(b) The FSF's Back-Cover Text is: 26 27 You have freedom to copy and modify this GNU Manual, like GNU 28 software. Copies published by the Free Software Foundation raise 29 funds for GNU development. 30@c man end 31@c Set file name and title for the man page. 32@setfilename gcc 33@settitle GNU project C and C++ compiler 34@c man begin SYNOPSIS 35gcc [@option{-c}|@option{-S}|@option{-E}] [@option{-std=}@var{standard}] 36 [@option{-g}] [@option{-pg}] [@option{-O}@var{level}] 37 [@option{-W}@var{warn}@dots{}] [@option{-Wpedantic}] 38 [@option{-I}@var{dir}@dots{}] [@option{-L}@var{dir}@dots{}] 39 [@option{-D}@var{macro}[=@var{defn}]@dots{}] [@option{-U}@var{macro}] 40 [@option{-f}@var{option}@dots{}] [@option{-m}@var{machine-option}@dots{}] 41 [@option{-o} @var{outfile}] [@@@var{file}] @var{infile}@dots{} 42 43Only the most useful options are listed here; see below for the 44remainder. @samp{g++} accepts mostly the same options as @samp{gcc}. 45@c man end 46@c man begin SEEALSO 47gpl(7), gfdl(7), fsf-funding(7), 48cpp(1), gcov(1), as(1), ld(1), gdb(1), adb(1), dbx(1), sdb(1) 49and the Info entries for @file{gcc}, @file{cpp}, @file{as}, 50@file{ld}, @file{binutils} and @file{gdb}. 51@c man end 52@c man begin BUGS 53For instructions on reporting bugs, see 54@w{@value{BUGURL}}. 55@c man end 56@c man begin AUTHOR 57See the Info entry for @command{gcc}, or 58@w{@uref{http://gcc.gnu.org/onlinedocs/gcc/Contributors.html}}, 59for contributors to GCC@. 60@c man end 61@end ignore 62 63@node Invoking GCC 64@chapter GCC Command Options 65@cindex GCC command options 66@cindex command options 67@cindex options, GCC command 68 69@c man begin DESCRIPTION 70When you invoke GCC, it normally does preprocessing, compilation, 71assembly and linking. The ``overall options'' allow you to stop this 72process at an intermediate stage. For example, the @option{-c} option 73says not to run the linker. Then the output consists of object files 74output by the assembler. 75 76Other options are passed on to one stage of processing. Some options 77control the preprocessor and others the compiler itself. Yet other 78options control the assembler and linker; most of these are not 79documented here, since you rarely need to use any of them. 80 81@cindex C compilation options 82Most of the command-line options that you can use with GCC are useful 83for C programs; when an option is only useful with another language 84(usually C++), the explanation says so explicitly. If the description 85for a particular option does not mention a source language, you can use 86that option with all supported languages. 87 88@cindex C++ compilation options 89@xref{Invoking G++,,Compiling C++ Programs}, for a summary of special 90options for compiling C++ programs. 91 92@cindex grouping options 93@cindex options, grouping 94The @command{gcc} program accepts options and file names as operands. Many 95options have multi-letter names; therefore multiple single-letter options 96may @emph{not} be grouped: @option{-dv} is very different from @w{@samp{-d 97-v}}. 98 99@cindex order of options 100@cindex options, order 101You can mix options and other arguments. For the most part, the order 102you use doesn't matter. Order does matter when you use several 103options of the same kind; for example, if you specify @option{-L} more 104than once, the directories are searched in the order specified. Also, 105the placement of the @option{-l} option is significant. 106 107Many options have long names starting with @samp{-f} or with 108@samp{-W}---for example, 109@option{-fmove-loop-invariants}, @option{-Wformat} and so on. Most of 110these have both positive and negative forms; the negative form of 111@option{-ffoo} is @option{-fno-foo}. This manual documents 112only one of these two forms, whichever one is not the default. 113 114@c man end 115 116@xref{Option Index}, for an index to GCC's options. 117 118@menu 119* Option Summary:: Brief list of all options, without explanations. 120* Overall Options:: Controlling the kind of output: 121 an executable, object files, assembler files, 122 or preprocessed source. 123* Invoking G++:: Compiling C++ programs. 124* C Dialect Options:: Controlling the variant of C language compiled. 125* C++ Dialect Options:: Variations on C++. 126* Objective-C and Objective-C++ Dialect Options:: Variations on Objective-C 127 and Objective-C++. 128* Language Independent Options:: Controlling how diagnostics should be 129 formatted. 130* Warning Options:: How picky should the compiler be? 131* Debugging Options:: Symbol tables, measurements, and debugging dumps. 132* Optimize Options:: How much optimization? 133* Preprocessor Options:: Controlling header files and macro definitions. 134 Also, getting dependency information for Make. 135* Assembler Options:: Passing options to the assembler. 136* Link Options:: Specifying libraries and so on. 137* Directory Options:: Where to find header files and libraries. 138 Where to find the compiler executable files. 139* Spec Files:: How to pass switches to sub-processes. 140* Target Options:: Running a cross-compiler, or an old version of GCC. 141* Submodel Options:: Specifying minor hardware or convention variations, 142 such as 68010 vs 68020. 143* Code Gen Options:: Specifying conventions for function calls, data layout 144 and register usage. 145* Environment Variables:: Env vars that affect GCC. 146* Precompiled Headers:: Compiling a header once, and using it many times. 147@end menu 148 149@c man begin OPTIONS 150 151@node Option Summary 152@section Option Summary 153 154Here is a summary of all the options, grouped by type. Explanations are 155in the following sections. 156 157@table @emph 158@item Overall Options 159@xref{Overall Options,,Options Controlling the Kind of Output}. 160@gccoptlist{-c -S -E -o @var{file} -no-canonical-prefixes @gol 161-pipe -pass-exit-codes @gol 162-x @var{language} -v -### --help@r{[}=@var{class}@r{[},@dots{}@r{]]} --target-help @gol 163--version -wrapper @@@var{file} -fplugin=@var{file} -fplugin-arg-@var{name}=@var{arg} @gol 164-fdump-ada-spec@r{[}-slim@r{]} -fada-spec-parent=@var{unit} -fdump-go-spec=@var{file}} 165 166@item C Language Options 167@xref{C Dialect Options,,Options Controlling C Dialect}. 168@gccoptlist{-ansi -std=@var{standard} -fgnu89-inline @gol 169-aux-info @var{filename} -fallow-parameterless-variadic-functions @gol 170-fno-asm -fno-builtin -fno-builtin-@var{function} @gol 171-fhosted -ffreestanding -fopenmp -fopenmp-simd -fms-extensions @gol 172-fplan9-extensions -trigraphs -traditional -traditional-cpp @gol 173-fallow-single-precision -fcond-mismatch -flax-vector-conversions @gol 174-fsigned-bitfields -fsigned-char @gol 175-funsigned-bitfields -funsigned-char} 176 177@item C++ Language Options 178@xref{C++ Dialect Options,,Options Controlling C++ Dialect}. 179@gccoptlist{-fabi-version=@var{n} -fno-access-control -fcheck-new @gol 180-fconstexpr-depth=@var{n} -ffriend-injection @gol 181-fno-elide-constructors @gol 182-fno-enforce-eh-specs @gol 183-ffor-scope -fno-for-scope -fno-gnu-keywords @gol 184-fno-implicit-templates @gol 185-fno-implicit-inline-templates @gol 186-fno-implement-inlines -fms-extensions @gol 187-fno-nonansi-builtins -fnothrow-opt -fno-operator-names @gol 188-fno-optional-diags -fpermissive @gol 189-fno-pretty-templates @gol 190-frepo -fno-rtti -fstats -ftemplate-backtrace-limit=@var{n} @gol 191-ftemplate-depth=@var{n} @gol 192-fno-threadsafe-statics -fuse-cxa-atexit -fno-weak -nostdinc++ @gol 193-fvisibility-inlines-hidden @gol 194-fvtable-verify=@var{std|preinit|none} @gol 195-fvtv-counts -fvtv-debug @gol 196-fvisibility-ms-compat @gol 197-fext-numeric-literals @gol 198-Wabi -Wconversion-null -Wctor-dtor-privacy @gol 199-Wdelete-non-virtual-dtor -Wliteral-suffix -Wnarrowing @gol 200-Wnoexcept -Wnon-virtual-dtor -Wreorder @gol 201-Weffc++ -Wstrict-null-sentinel @gol 202-Wno-non-template-friend -Wold-style-cast @gol 203-Woverloaded-virtual -Wno-pmf-conversions @gol 204-Wsign-promo} 205 206@item Objective-C and Objective-C++ Language Options 207@xref{Objective-C and Objective-C++ Dialect Options,,Options Controlling 208Objective-C and Objective-C++ Dialects}. 209@gccoptlist{-fconstant-string-class=@var{class-name} @gol 210-fgnu-runtime -fnext-runtime @gol 211-fno-nil-receivers @gol 212-fobjc-abi-version=@var{n} @gol 213-fobjc-call-cxx-cdtors @gol 214-fobjc-direct-dispatch @gol 215-fobjc-exceptions @gol 216-fobjc-gc @gol 217-fobjc-nilcheck @gol 218-fobjc-std=objc1 @gol 219-freplace-objc-classes @gol 220-fzero-link @gol 221-gen-decls @gol 222-Wassign-intercept @gol 223-Wno-protocol -Wselector @gol 224-Wstrict-selector-match @gol 225-Wundeclared-selector} 226 227@item Language Independent Options 228@xref{Language Independent Options,,Options to Control Diagnostic Messages Formatting}. 229@gccoptlist{-fmessage-length=@var{n} @gol 230-fdiagnostics-show-location=@r{[}once@r{|}every-line@r{]} @gol 231-fdiagnostics-color=@r{[}auto@r{|}never@r{|}always@r{]} @gol 232-fno-diagnostics-show-option -fno-diagnostics-show-caret} 233 234@item Warning Options 235@xref{Warning Options,,Options to Request or Suppress Warnings}. 236@gccoptlist{-fsyntax-only -fmax-errors=@var{n} -Wpedantic @gol 237-pedantic-errors @gol 238-w -Wextra -Wall -Waddress -Waggregate-return @gol 239-Waggressive-loop-optimizations -Warray-bounds @gol 240-Wno-attributes -Wno-builtin-macro-redefined @gol 241-Wc++-compat -Wc++11-compat -Wcast-align -Wcast-qual @gol 242-Wchar-subscripts -Wclobbered -Wcomment -Wconditionally-supported @gol 243-Wconversion -Wcoverage-mismatch -Wdate-time -Wdelete-incomplete -Wno-cpp @gol 244-Wno-deprecated -Wno-deprecated-declarations -Wdisabled-optimization @gol 245-Wno-div-by-zero -Wdouble-promotion -Wempty-body -Wenum-compare @gol 246-Wno-endif-labels -Werror -Werror=* @gol 247-Wfatal-errors -Wfloat-equal -Wformat -Wformat=2 @gol 248-Wno-format-contains-nul -Wno-format-extra-args -Wformat-nonliteral @gol 249-Wformat-security -Wformat-y2k @gol 250-Wframe-larger-than=@var{len} -Wno-free-nonheap-object -Wjump-misses-init @gol 251-Wignored-qualifiers @gol 252-Wimplicit -Wimplicit-function-declaration -Wimplicit-int @gol 253-Winit-self -Winline -Wmaybe-uninitialized @gol 254-Wno-int-to-pointer-cast -Wno-invalid-offsetof @gol 255-Winvalid-pch -Wlarger-than=@var{len} -Wunsafe-loop-optimizations @gol 256-Wlogical-op -Wlong-long @gol 257-Wmain -Wmaybe-uninitialized -Wmissing-braces -Wmissing-field-initializers @gol 258-Wmissing-include-dirs @gol 259-Wno-multichar -Wnonnull -Wno-overflow -Wopenmp-simd @gol 260-Woverlength-strings -Wpacked -Wpacked-bitfield-compat -Wpadded @gol 261-Wparentheses -Wpedantic-ms-format -Wno-pedantic-ms-format @gol 262-Wpointer-arith -Wno-pointer-to-int-cast @gol 263-Wredundant-decls -Wno-return-local-addr @gol 264-Wreturn-type -Wsequence-point -Wshadow @gol 265-Wsign-compare -Wsign-conversion -Wfloat-conversion @gol 266-Wsizeof-pointer-memaccess @gol 267-Wstack-protector -Wstack-usage=@var{len} -Wstrict-aliasing @gol 268-Wstrict-aliasing=n @gol -Wstrict-overflow -Wstrict-overflow=@var{n} @gol 269-Wsuggest-attribute=@r{[}pure@r{|}const@r{|}noreturn@r{|}format@r{]} @gol 270-Wmissing-format-attribute @gol 271-Wswitch -Wswitch-default -Wswitch-enum -Wsync-nand @gol 272-Wsystem-headers -Wtrampolines -Wtrigraphs -Wtype-limits -Wundef @gol 273-Wuninitialized -Wunknown-pragmas -Wno-pragmas @gol 274-Wunsuffixed-float-constants -Wunused -Wunused-function @gol 275-Wunused-label -Wunused-local-typedefs -Wunused-parameter @gol 276-Wno-unused-result -Wunused-value @gol -Wunused-variable @gol 277-Wunused-but-set-parameter -Wunused-but-set-variable @gol 278-Wuseless-cast -Wvariadic-macros -Wvector-operation-performance @gol 279-Wvla -Wvolatile-register-var -Wwrite-strings -Wzero-as-null-pointer-constant} 280 281@item C and Objective-C-only Warning Options 282@gccoptlist{-Wbad-function-cast -Wmissing-declarations @gol 283-Wmissing-parameter-type -Wmissing-prototypes -Wnested-externs @gol 284-Wold-style-declaration -Wold-style-definition @gol 285-Wstrict-prototypes -Wtraditional -Wtraditional-conversion @gol 286-Wdeclaration-after-statement -Wpointer-sign} 287 288@item Debugging Options 289@xref{Debugging Options,,Options for Debugging Your Program or GCC}. 290@gccoptlist{-d@var{letters} -dumpspecs -dumpmachine -dumpversion @gol 291-fsanitize=@var{style} @gol 292-fdbg-cnt-list -fdbg-cnt=@var{counter-value-list} @gol 293-fdisable-ipa-@var{pass_name} @gol 294-fdisable-rtl-@var{pass_name} @gol 295-fdisable-rtl-@var{pass-name}=@var{range-list} @gol 296-fdisable-tree-@var{pass_name} @gol 297-fdisable-tree-@var{pass-name}=@var{range-list} @gol 298-fdump-noaddr -fdump-unnumbered -fdump-unnumbered-links @gol 299-fdump-translation-unit@r{[}-@var{n}@r{]} @gol 300-fdump-class-hierarchy@r{[}-@var{n}@r{]} @gol 301-fdump-ipa-all -fdump-ipa-cgraph -fdump-ipa-inline @gol 302-fdump-passes @gol 303-fdump-statistics @gol 304-fdump-tree-all @gol 305-fdump-tree-original@r{[}-@var{n}@r{]} @gol 306-fdump-tree-optimized@r{[}-@var{n}@r{]} @gol 307-fdump-tree-cfg -fdump-tree-alias @gol 308-fdump-tree-ch @gol 309-fdump-tree-ssa@r{[}-@var{n}@r{]} -fdump-tree-pre@r{[}-@var{n}@r{]} @gol 310-fdump-tree-ccp@r{[}-@var{n}@r{]} -fdump-tree-dce@r{[}-@var{n}@r{]} @gol 311-fdump-tree-gimple@r{[}-raw@r{]} @gol 312-fdump-tree-dom@r{[}-@var{n}@r{]} @gol 313-fdump-tree-dse@r{[}-@var{n}@r{]} @gol 314-fdump-tree-phiprop@r{[}-@var{n}@r{]} @gol 315-fdump-tree-phiopt@r{[}-@var{n}@r{]} @gol 316-fdump-tree-forwprop@r{[}-@var{n}@r{]} @gol 317-fdump-tree-copyrename@r{[}-@var{n}@r{]} @gol 318-fdump-tree-nrv -fdump-tree-vect @gol 319-fdump-tree-sink @gol 320-fdump-tree-sra@r{[}-@var{n}@r{]} @gol 321-fdump-tree-forwprop@r{[}-@var{n}@r{]} @gol 322-fdump-tree-fre@r{[}-@var{n}@r{]} @gol 323-fdump-tree-vtable-verify @gol 324-fdump-tree-vrp@r{[}-@var{n}@r{]} @gol 325-fdump-tree-storeccp@r{[}-@var{n}@r{]} @gol 326-fdump-final-insns=@var{file} @gol 327-fcompare-debug@r{[}=@var{opts}@r{]} -fcompare-debug-second @gol 328-feliminate-dwarf2-dups -fno-eliminate-unused-debug-types @gol 329-feliminate-unused-debug-symbols -femit-class-debug-always @gol 330-fenable-@var{kind}-@var{pass} @gol 331-fenable-@var{kind}-@var{pass}=@var{range-list} @gol 332-fdebug-types-section -fmem-report-wpa @gol 333-fmem-report -fpre-ipa-mem-report -fpost-ipa-mem-report -fprofile-arcs @gol 334-fopt-info @gol 335-fopt-info-@var{options}@r{[}=@var{file}@r{]} @gol 336-frandom-seed=@var{string} -fsched-verbose=@var{n} @gol 337-fsel-sched-verbose -fsel-sched-dump-cfg -fsel-sched-pipelining-verbose @gol 338-fstack-usage -ftest-coverage -ftime-report -fvar-tracking @gol 339-fvar-tracking-assignments -fvar-tracking-assignments-toggle @gol 340-g -g@var{level} -gtoggle -gcoff -gdwarf-@var{version} @gol 341-ggdb -grecord-gcc-switches -gno-record-gcc-switches @gol 342-gstabs -gstabs+ -gstrict-dwarf -gno-strict-dwarf @gol 343-gvms -gxcoff -gxcoff+ @gol 344-fno-merge-debug-strings -fno-dwarf2-cfi-asm @gol 345-fdebug-prefix-map=@var{old}=@var{new} @gol 346-femit-struct-debug-baseonly -femit-struct-debug-reduced @gol 347-femit-struct-debug-detailed@r{[}=@var{spec-list}@r{]} @gol 348-p -pg -print-file-name=@var{library} -print-libgcc-file-name @gol 349-print-multi-directory -print-multi-lib -print-multi-os-directory @gol 350-print-prog-name=@var{program} -print-search-dirs -Q @gol 351-print-sysroot -print-sysroot-headers-suffix @gol 352-save-temps -save-temps=cwd -save-temps=obj -time@r{[}=@var{file}@r{]}} 353 354@item Optimization Options 355@xref{Optimize Options,,Options that Control Optimization}. 356@gccoptlist{-faggressive-loop-optimizations -falign-functions[=@var{n}] @gol 357-falign-jumps[=@var{n}] @gol 358-falign-labels[=@var{n}] -falign-loops[=@var{n}] @gol 359-fassociative-math -fauto-inc-dec -fbranch-probabilities @gol 360-fbranch-target-load-optimize -fbranch-target-load-optimize2 @gol 361-fbtr-bb-exclusive -fcaller-saves @gol 362-fcheck-data-deps -fcombine-stack-adjustments -fconserve-stack @gol 363-fcompare-elim -fcprop-registers -fcrossjumping @gol 364-fcse-follow-jumps -fcse-skip-blocks -fcx-fortran-rules @gol 365-fcx-limited-range @gol 366-fdata-sections -fdce -fdelayed-branch @gol 367-fdelete-null-pointer-checks -fdevirtualize -fdevirtualize-speculatively -fdse @gol 368-fearly-inlining -fipa-sra -fexpensive-optimizations -ffat-lto-objects @gol 369-ffast-math -ffinite-math-only -ffloat-store -fexcess-precision=@var{style} @gol 370-fforward-propagate -ffp-contract=@var{style} -ffunction-sections @gol 371-fgcse -fgcse-after-reload -fgcse-las -fgcse-lm -fgraphite-identity @gol 372-fgcse-sm -fhoist-adjacent-loads -fif-conversion @gol 373-fif-conversion2 -findirect-inlining @gol 374-finline-functions -finline-functions-called-once -finline-limit=@var{n} @gol 375-finline-small-functions -fipa-cp -fipa-cp-clone @gol 376-fipa-pta -fipa-profile -fipa-pure-const -fipa-reference @gol 377-fira-algorithm=@var{algorithm} @gol 378-fira-region=@var{region} -fira-hoist-pressure @gol 379-fira-loop-pressure -fno-ira-share-save-slots @gol 380-fno-ira-share-spill-slots -fira-verbose=@var{n} @gol 381-fisolate-erroneous-paths-dereference -fisolate-erroneous-paths-attribute 382-fivopts -fkeep-inline-functions -fkeep-static-consts -flive-range-shrinkage @gol 383-floop-block -floop-interchange -floop-strip-mine -floop-nest-optimize @gol 384-floop-parallelize-all -flto -flto-compression-level @gol 385-flto-partition=@var{alg} -flto-report -flto-report-wpa -fmerge-all-constants @gol 386-fmerge-constants -fmodulo-sched -fmodulo-sched-allow-regmoves @gol 387-fmove-loop-invariants -fno-branch-count-reg @gol 388-fno-defer-pop -fno-function-cse -fno-guess-branch-probability @gol 389-fno-inline -fno-math-errno -fno-peephole -fno-peephole2 @gol 390-fno-sched-interblock -fno-sched-spec -fno-signed-zeros @gol 391-fno-toplevel-reorder -fno-trapping-math -fno-zero-initialized-in-bss @gol 392-fomit-frame-pointer -foptimize-sibling-calls @gol 393-fpartial-inlining -fpeel-loops -fpredictive-commoning @gol 394-fprefetch-loop-arrays -fprofile-report @gol 395-fprofile-correction -fprofile-dir=@var{path} -fprofile-generate @gol 396-fprofile-generate=@var{path} @gol 397-fprofile-use -fprofile-use=@var{path} -fprofile-values -fprofile-reorder-functions @gol 398-freciprocal-math -free -frename-registers -freorder-blocks @gol 399-freorder-blocks-and-partition -freorder-functions @gol 400-frerun-cse-after-loop -freschedule-modulo-scheduled-loops @gol 401-frounding-math -fsched2-use-superblocks -fsched-pressure @gol 402-fsched-spec-load -fsched-spec-load-dangerous @gol 403-fsched-stalled-insns-dep[=@var{n}] -fsched-stalled-insns[=@var{n}] @gol 404-fsched-group-heuristic -fsched-critical-path-heuristic @gol 405-fsched-spec-insn-heuristic -fsched-rank-heuristic @gol 406-fsched-last-insn-heuristic -fsched-dep-count-heuristic @gol 407-fschedule-insns -fschedule-insns2 -fsection-anchors @gol 408-fselective-scheduling -fselective-scheduling2 @gol 409-fsel-sched-pipelining -fsel-sched-pipelining-outer-loops @gol 410-fshrink-wrap -fsignaling-nans -fsingle-precision-constant @gol 411-fsplit-ivs-in-unroller -fsplit-wide-types -fstack-protector @gol 412-fstack-protector-all -fstack-protector-strong -fstrict-aliasing @gol 413-fstrict-overflow -fthread-jumps -ftracer -ftree-bit-ccp @gol 414-ftree-builtin-call-dce -ftree-ccp -ftree-ch @gol 415-ftree-coalesce-inline-vars -ftree-coalesce-vars -ftree-copy-prop @gol 416-ftree-copyrename -ftree-dce -ftree-dominator-opts -ftree-dse @gol 417-ftree-forwprop -ftree-fre -ftree-loop-if-convert @gol 418-ftree-loop-if-convert-stores -ftree-loop-im @gol 419-ftree-phiprop -ftree-loop-distribution -ftree-loop-distribute-patterns @gol 420-ftree-loop-ivcanon -ftree-loop-linear -ftree-loop-optimize @gol 421-ftree-loop-vectorize @gol 422-ftree-parallelize-loops=@var{n} -ftree-pre -ftree-partial-pre -ftree-pta @gol 423-ftree-reassoc -ftree-sink -ftree-slsr -ftree-sra @gol 424-ftree-switch-conversion -ftree-tail-merge -ftree-ter @gol 425-ftree-vectorize -ftree-vrp @gol 426-funit-at-a-time -funroll-all-loops -funroll-loops @gol 427-funsafe-loop-optimizations -funsafe-math-optimizations -funswitch-loops @gol 428-fvariable-expansion-in-unroller -fvect-cost-model -fvpt -fweb @gol 429-fwhole-program -fwpa -fuse-ld=@var{linker} -fuse-linker-plugin @gol 430--param @var{name}=@var{value} 431-O -O0 -O1 -O2 -O3 -Os -Ofast -Og} 432 433@item Preprocessor Options 434@xref{Preprocessor Options,,Options Controlling the Preprocessor}. 435@gccoptlist{-A@var{question}=@var{answer} @gol 436-A-@var{question}@r{[}=@var{answer}@r{]} @gol 437-C -dD -dI -dM -dN @gol 438-D@var{macro}@r{[}=@var{defn}@r{]} -E -H @gol 439-idirafter @var{dir} @gol 440-include @var{file} -imacros @var{file} @gol 441-iprefix @var{file} -iwithprefix @var{dir} @gol 442-iwithprefixbefore @var{dir} -isystem @var{dir} @gol 443-imultilib @var{dir} -isysroot @var{dir} @gol 444-M -MM -MF -MG -MP -MQ -MT -nostdinc @gol 445-P -fdebug-cpp -ftrack-macro-expansion -fworking-directory @gol 446-remap -trigraphs -undef -U@var{macro} @gol 447-Wp,@var{option} -Xpreprocessor @var{option} -no-integrated-cpp} 448 449@item Assembler Option 450@xref{Assembler Options,,Passing Options to the Assembler}. 451@gccoptlist{-Wa,@var{option} -Xassembler @var{option}} 452 453@item Linker Options 454@xref{Link Options,,Options for Linking}. 455@gccoptlist{@var{object-file-name} -l@var{library} @gol 456-nostartfiles -nodefaultlibs -nostdlib -pie -rdynamic @gol 457-s -static -static-libgcc -static-libstdc++ @gol 458-static-libasan -static-libtsan -static-liblsan -static-libubsan @gol 459-shared -shared-libgcc -symbolic @gol 460-T @var{script} -Wl,@var{option} -Xlinker @var{option} @gol 461-u @var{symbol}} 462 463@item Directory Options 464@xref{Directory Options,,Options for Directory Search}. 465@gccoptlist{-B@var{prefix} -I@var{dir} -iplugindir=@var{dir} @gol 466-iquote@var{dir} -L@var{dir} -specs=@var{file} -I- @gol 467--sysroot=@var{dir} --no-sysroot-suffix} 468 469@item Machine Dependent Options 470@xref{Submodel Options,,Hardware Models and Configurations}. 471@c This list is ordered alphanumerically by subsection name. 472@c Try and put the significant identifier (CPU or system) first, 473@c so users have a clue at guessing where the ones they want will be. 474 475@emph{AArch64 Options} 476@gccoptlist{-mabi=@var{name} -mbig-endian -mlittle-endian @gol 477-mgeneral-regs-only @gol 478-mcmodel=tiny -mcmodel=small -mcmodel=large @gol 479-mstrict-align @gol 480-momit-leaf-frame-pointer -mno-omit-leaf-frame-pointer @gol 481-mtls-dialect=desc -mtls-dialect=traditional @gol 482-mfix-cortex-a53-835769 -mno-fix-cortex-a53-835769 @gol 483-march=@var{name} -mcpu=@var{name} -mtune=@var{name}} 484 485@emph{Adapteva Epiphany Options} 486@gccoptlist{-mhalf-reg-file -mprefer-short-insn-regs @gol 487-mbranch-cost=@var{num} -mcmove -mnops=@var{num} -msoft-cmpsf @gol 488-msplit-lohi -mpost-inc -mpost-modify -mstack-offset=@var{num} @gol 489-mround-nearest -mlong-calls -mshort-calls -msmall16 @gol 490-mfp-mode=@var{mode} -mvect-double -max-vect-align=@var{num} @gol 491-msplit-vecmove-early -m1reg-@var{reg}} 492 493@emph{ARC Options} 494@gccoptlist{-mbarrel-shifter @gol 495-mcpu=@var{cpu} -mA6 -mARC600 -mA7 -mARC700 @gol 496-mdpfp -mdpfp-compact -mdpfp-fast -mno-dpfp-lrsr @gol 497-mea -mno-mpy -mmul32x16 -mmul64 @gol 498-mnorm -mspfp -mspfp-compact -mspfp-fast -msimd -msoft-float -mswap @gol 499-mcrc -mdsp-packa -mdvbf -mlock -mmac-d16 -mmac-24 -mrtsc -mswape @gol 500-mtelephony -mxy -misize -mannotate-align -marclinux -marclinux_prof @gol 501-mepilogue-cfi -mlong-calls -mmedium-calls -msdata @gol 502-mucb-mcount -mvolatile-cache @gol 503-malign-call -mauto-modify-reg -mbbit-peephole -mno-brcc @gol 504-mcase-vector-pcrel -mcompact-casesi -mno-cond-exec -mearly-cbranchsi @gol 505-mexpand-adddi -mindexed-loads -mlra -mlra-priority-none @gol 506-mlra-priority-compact mlra-priority-noncompact -mno-millicode @gol 507-mmixed-code -mq-class -mRcq -mRcw -msize-level=@var{level} @gol 508-mtune=@var{cpu} -mmultcost=@var{num} -munalign-prob-threshold=@var{probability}} 509 510@emph{ARM Options} 511@gccoptlist{-mapcs-frame -mno-apcs-frame @gol 512-mabi=@var{name} @gol 513-mapcs-stack-check -mno-apcs-stack-check @gol 514-mapcs-float -mno-apcs-float @gol 515-mapcs-reentrant -mno-apcs-reentrant @gol 516-msched-prolog -mno-sched-prolog @gol 517-mlittle-endian -mbig-endian -mwords-little-endian @gol 518-mfloat-abi=@var{name} @gol 519-mfp16-format=@var{name} 520-mthumb-interwork -mno-thumb-interwork @gol 521-mcpu=@var{name} -march=@var{name} -mfpu=@var{name} @gol 522-mstructure-size-boundary=@var{n} @gol 523-mabort-on-noreturn @gol 524-mlong-calls -mno-long-calls @gol 525-msingle-pic-base -mno-single-pic-base @gol 526-mpic-register=@var{reg} @gol 527-mnop-fun-dllimport @gol 528-mpoke-function-name @gol 529-mthumb -marm @gol 530-mtpcs-frame -mtpcs-leaf-frame @gol 531-mcaller-super-interworking -mcallee-super-interworking @gol 532-mtp=@var{name} -mtls-dialect=@var{dialect} @gol 533-mword-relocations @gol 534-mfix-cortex-m3-ldrd @gol 535-munaligned-access @gol 536-mneon-for-64bits @gol 537-mslow-flash-data @gol 538-mrestrict-it} 539 540@emph{AVR Options} 541@gccoptlist{-mmcu=@var{mcu} -maccumulate-args -mbranch-cost=@var{cost} @gol 542-mcall-prologues -mint8 -mno-interrupts -mrelax @gol 543-mstrict-X -mtiny-stack -Waddr-space-convert} 544 545@emph{Blackfin Options} 546@gccoptlist{-mcpu=@var{cpu}@r{[}-@var{sirevision}@r{]} @gol 547-msim -momit-leaf-frame-pointer -mno-omit-leaf-frame-pointer @gol 548-mspecld-anomaly -mno-specld-anomaly -mcsync-anomaly -mno-csync-anomaly @gol 549-mlow-64k -mno-low64k -mstack-check-l1 -mid-shared-library @gol 550-mno-id-shared-library -mshared-library-id=@var{n} @gol 551-mleaf-id-shared-library -mno-leaf-id-shared-library @gol 552-msep-data -mno-sep-data -mlong-calls -mno-long-calls @gol 553-mfast-fp -minline-plt -mmulticore -mcorea -mcoreb -msdram @gol 554-micplb} 555 556@emph{C6X Options} 557@gccoptlist{-mbig-endian -mlittle-endian -march=@var{cpu} @gol 558-msim -msdata=@var{sdata-type}} 559 560@emph{CRIS Options} 561@gccoptlist{-mcpu=@var{cpu} -march=@var{cpu} -mtune=@var{cpu} @gol 562-mmax-stack-frame=@var{n} -melinux-stacksize=@var{n} @gol 563-metrax4 -metrax100 -mpdebug -mcc-init -mno-side-effects @gol 564-mstack-align -mdata-align -mconst-align @gol 565-m32-bit -m16-bit -m8-bit -mno-prologue-epilogue -mno-gotplt @gol 566-melf -maout -melinux -mlinux -sim -sim2 @gol 567-mmul-bug-workaround -mno-mul-bug-workaround} 568 569@emph{CR16 Options} 570@gccoptlist{-mmac @gol 571-mcr16cplus -mcr16c @gol 572-msim -mint32 -mbit-ops 573-mdata-model=@var{model}} 574 575@emph{Darwin Options} 576@gccoptlist{-all_load -allowable_client -arch -arch_errors_fatal @gol 577-arch_only -bind_at_load -bundle -bundle_loader @gol 578-client_name -compatibility_version -current_version @gol 579-dead_strip @gol 580-dependency-file -dylib_file -dylinker_install_name @gol 581-dynamic -dynamiclib -exported_symbols_list @gol 582-filelist -flat_namespace -force_cpusubtype_ALL @gol 583-force_flat_namespace -headerpad_max_install_names @gol 584-iframework @gol 585-image_base -init -install_name -keep_private_externs @gol 586-multi_module -multiply_defined -multiply_defined_unused @gol 587-noall_load -no_dead_strip_inits_and_terms @gol 588-nofixprebinding -nomultidefs -noprebind -noseglinkedit @gol 589-pagezero_size -prebind -prebind_all_twolevel_modules @gol 590-private_bundle -read_only_relocs -sectalign @gol 591-sectobjectsymbols -whyload -seg1addr @gol 592-sectcreate -sectobjectsymbols -sectorder @gol 593-segaddr -segs_read_only_addr -segs_read_write_addr @gol 594-seg_addr_table -seg_addr_table_filename -seglinkedit @gol 595-segprot -segs_read_only_addr -segs_read_write_addr @gol 596-single_module -static -sub_library -sub_umbrella @gol 597-twolevel_namespace -umbrella -undefined @gol 598-unexported_symbols_list -weak_reference_mismatches @gol 599-whatsloaded -F -gused -gfull -mmacosx-version-min=@var{version} @gol 600-mkernel -mone-byte-bool} 601 602@emph{DEC Alpha Options} 603@gccoptlist{-mno-fp-regs -msoft-float @gol 604-mieee -mieee-with-inexact -mieee-conformant @gol 605-mfp-trap-mode=@var{mode} -mfp-rounding-mode=@var{mode} @gol 606-mtrap-precision=@var{mode} -mbuild-constants @gol 607-mcpu=@var{cpu-type} -mtune=@var{cpu-type} @gol 608-mbwx -mmax -mfix -mcix @gol 609-mfloat-vax -mfloat-ieee @gol 610-mexplicit-relocs -msmall-data -mlarge-data @gol 611-msmall-text -mlarge-text @gol 612-mmemory-latency=@var{time}} 613 614@emph{FR30 Options} 615@gccoptlist{-msmall-model -mno-lsim} 616 617@emph{FRV Options} 618@gccoptlist{-mgpr-32 -mgpr-64 -mfpr-32 -mfpr-64 @gol 619-mhard-float -msoft-float @gol 620-malloc-cc -mfixed-cc -mdword -mno-dword @gol 621-mdouble -mno-double @gol 622-mmedia -mno-media -mmuladd -mno-muladd @gol 623-mfdpic -minline-plt -mgprel-ro -multilib-library-pic @gol 624-mlinked-fp -mlong-calls -malign-labels @gol 625-mlibrary-pic -macc-4 -macc-8 @gol 626-mpack -mno-pack -mno-eflags -mcond-move -mno-cond-move @gol 627-moptimize-membar -mno-optimize-membar @gol 628-mscc -mno-scc -mcond-exec -mno-cond-exec @gol 629-mvliw-branch -mno-vliw-branch @gol 630-mmulti-cond-exec -mno-multi-cond-exec -mnested-cond-exec @gol 631-mno-nested-cond-exec -mtomcat-stats @gol 632-mTLS -mtls @gol 633-mcpu=@var{cpu}} 634 635@emph{GNU/Linux Options} 636@gccoptlist{-mglibc -muclibc -mbionic -mandroid @gol 637-tno-android-cc -tno-android-ld} 638 639@emph{H8/300 Options} 640@gccoptlist{-mrelax -mh -ms -mn -mexr -mno-exr -mint32 -malign-300} 641 642@emph{HPPA Options} 643@gccoptlist{-march=@var{architecture-type} @gol 644-mdisable-fpregs -mdisable-indexing @gol 645-mfast-indirect-calls -mgas -mgnu-ld -mhp-ld @gol 646-mfixed-range=@var{register-range} @gol 647-mjump-in-delay -mlinker-opt -mlong-calls @gol 648-mlong-load-store -mno-disable-fpregs @gol 649-mno-disable-indexing -mno-fast-indirect-calls -mno-gas @gol 650-mno-jump-in-delay -mno-long-load-store @gol 651-mno-portable-runtime -mno-soft-float @gol 652-mno-space-regs -msoft-float -mpa-risc-1-0 @gol 653-mpa-risc-1-1 -mpa-risc-2-0 -mportable-runtime @gol 654-mschedule=@var{cpu-type} -mspace-regs -msio -mwsio @gol 655-munix=@var{unix-std} -nolibdld -static -threads} 656 657@emph{i386 and x86-64 Options} 658@gccoptlist{-mtune=@var{cpu-type} -march=@var{cpu-type} @gol 659-mtune-ctrl=@var{feature-list} -mdump-tune-features -mno-default @gol 660-mfpmath=@var{unit} @gol 661-masm=@var{dialect} -mno-fancy-math-387 @gol 662-mno-fp-ret-in-387 -msoft-float @gol 663-mno-wide-multiply -mrtd -malign-double @gol 664-mpreferred-stack-boundary=@var{num} @gol 665-mincoming-stack-boundary=@var{num} @gol 666-mcld -mcx16 -msahf -mmovbe -mcrc32 @gol 667-mrecip -mrecip=@var{opt} @gol 668-mvzeroupper -mprefer-avx128 @gol 669-mmmx -msse -msse2 -msse3 -mssse3 -msse4.1 -msse4.2 -msse4 -mavx @gol 670-mavx2 -mavx512f -mavx512pf -mavx512er -mavx512cd -msha @gol 671-maes -mpclmul -mfsgsbase -mrdrnd -mf16c -mfma -mprefetchwt1 @gol 672-msse4a -m3dnow -mpopcnt -mabm -mbmi -mtbm -mfma4 -mxop -mlzcnt @gol 673-mbmi2 -mfxsr -mxsave -mxsaveopt -mrtm -mlwp -mthreads @gol 674-mno-align-stringops -minline-all-stringops @gol 675-minline-stringops-dynamically -mstringop-strategy=@var{alg} @gol 676-mmemcpy-strategy=@var{strategy} -mmemset-strategy=@var{strategy} 677-mpush-args -maccumulate-outgoing-args -m128bit-long-double @gol 678-m96bit-long-double -mlong-double-64 -mlong-double-80 -mlong-double-128 @gol 679-mregparm=@var{num} -msseregparm @gol 680-mveclibabi=@var{type} -mvect8-ret-in-mem @gol 681-mpc32 -mpc64 -mpc80 -mstackrealign @gol 682-momit-leaf-frame-pointer -mno-red-zone -mno-tls-direct-seg-refs @gol 683-mcmodel=@var{code-model} -mabi=@var{name} -maddress-mode=@var{mode} @gol 684-m32 -m64 -mx32 -m16 -mlarge-data-threshold=@var{num} @gol 685-msse2avx -mfentry -m8bit-idiv @gol 686-mavx256-split-unaligned-load -mavx256-split-unaligned-store @gol 687-mstack-protector-guard=@var{guard}} 688 689@emph{i386 and x86-64 Windows Options} 690@gccoptlist{-mconsole -mcygwin -mno-cygwin -mdll @gol 691-mnop-fun-dllimport -mthread @gol 692-municode -mwin32 -mwindows -fno-set-stack-executable} 693 694@emph{IA-64 Options} 695@gccoptlist{-mbig-endian -mlittle-endian -mgnu-as -mgnu-ld -mno-pic @gol 696-mvolatile-asm-stop -mregister-names -msdata -mno-sdata @gol 697-mconstant-gp -mauto-pic -mfused-madd @gol 698-minline-float-divide-min-latency @gol 699-minline-float-divide-max-throughput @gol 700-mno-inline-float-divide @gol 701-minline-int-divide-min-latency @gol 702-minline-int-divide-max-throughput @gol 703-mno-inline-int-divide @gol 704-minline-sqrt-min-latency -minline-sqrt-max-throughput @gol 705-mno-inline-sqrt @gol 706-mdwarf2-asm -mearly-stop-bits @gol 707-mfixed-range=@var{register-range} -mtls-size=@var{tls-size} @gol 708-mtune=@var{cpu-type} -milp32 -mlp64 @gol 709-msched-br-data-spec -msched-ar-data-spec -msched-control-spec @gol 710-msched-br-in-data-spec -msched-ar-in-data-spec -msched-in-control-spec @gol 711-msched-spec-ldc -msched-spec-control-ldc @gol 712-msched-prefer-non-data-spec-insns -msched-prefer-non-control-spec-insns @gol 713-msched-stop-bits-after-every-cycle -msched-count-spec-in-critical-path @gol 714-msel-sched-dont-check-control-spec -msched-fp-mem-deps-zero-cost @gol 715-msched-max-memory-insns-hard-limit -msched-max-memory-insns=@var{max-insns}} 716 717@emph{LM32 Options} 718@gccoptlist{-mbarrel-shift-enabled -mdivide-enabled -mmultiply-enabled @gol 719-msign-extend-enabled -muser-enabled} 720 721@emph{M32R/D Options} 722@gccoptlist{-m32r2 -m32rx -m32r @gol 723-mdebug @gol 724-malign-loops -mno-align-loops @gol 725-missue-rate=@var{number} @gol 726-mbranch-cost=@var{number} @gol 727-mmodel=@var{code-size-model-type} @gol 728-msdata=@var{sdata-type} @gol 729-mno-flush-func -mflush-func=@var{name} @gol 730-mno-flush-trap -mflush-trap=@var{number} @gol 731-G @var{num}} 732 733@emph{M32C Options} 734@gccoptlist{-mcpu=@var{cpu} -msim -memregs=@var{number}} 735 736@emph{M680x0 Options} 737@gccoptlist{-march=@var{arch} -mcpu=@var{cpu} -mtune=@var{tune} 738-m68000 -m68020 -m68020-40 -m68020-60 -m68030 -m68040 @gol 739-m68060 -mcpu32 -m5200 -m5206e -m528x -m5307 -m5407 @gol 740-mcfv4e -mbitfield -mno-bitfield -mc68000 -mc68020 @gol 741-mnobitfield -mrtd -mno-rtd -mdiv -mno-div -mshort @gol 742-mno-short -mhard-float -m68881 -msoft-float -mpcrel @gol 743-malign-int -mstrict-align -msep-data -mno-sep-data @gol 744-mshared-library-id=n -mid-shared-library -mno-id-shared-library @gol 745-mxgot -mno-xgot} 746 747@emph{MCore Options} 748@gccoptlist{-mhardlit -mno-hardlit -mdiv -mno-div -mrelax-immediates @gol 749-mno-relax-immediates -mwide-bitfields -mno-wide-bitfields @gol 750-m4byte-functions -mno-4byte-functions -mcallgraph-data @gol 751-mno-callgraph-data -mslow-bytes -mno-slow-bytes -mno-lsim @gol 752-mlittle-endian -mbig-endian -m210 -m340 -mstack-increment} 753 754@emph{MeP Options} 755@gccoptlist{-mabsdiff -mall-opts -maverage -mbased=@var{n} -mbitops @gol 756-mc=@var{n} -mclip -mconfig=@var{name} -mcop -mcop32 -mcop64 -mivc2 @gol 757-mdc -mdiv -meb -mel -mio-volatile -ml -mleadz -mm -mminmax @gol 758-mmult -mno-opts -mrepeat -ms -msatur -msdram -msim -msimnovec -mtf @gol 759-mtiny=@var{n}} 760 761@emph{MicroBlaze Options} 762@gccoptlist{-msoft-float -mhard-float -msmall-divides -mcpu=@var{cpu} @gol 763-mmemcpy -mxl-soft-mul -mxl-soft-div -mxl-barrel-shift @gol 764-mxl-pattern-compare -mxl-stack-check -mxl-gp-opt -mno-clearbss @gol 765-mxl-multiply-high -mxl-float-convert -mxl-float-sqrt @gol 766-mbig-endian -mlittle-endian -mxl-reorder -mxl-mode-@var{app-model}} 767 768@emph{MIPS Options} 769@gccoptlist{-EL -EB -march=@var{arch} -mtune=@var{arch} @gol 770-mips1 -mips2 -mips3 -mips4 -mips32 -mips32r2 @gol 771-mips64 -mips64r2 @gol 772-mips16 -mno-mips16 -mflip-mips16 @gol 773-minterlink-compressed -mno-interlink-compressed @gol 774-minterlink-mips16 -mno-interlink-mips16 @gol 775-mabi=@var{abi} -mabicalls -mno-abicalls @gol 776-mshared -mno-shared -mplt -mno-plt -mxgot -mno-xgot @gol 777-mgp32 -mgp64 -mfp32 -mfp64 -mhard-float -msoft-float @gol 778-mno-float -msingle-float -mdouble-float @gol 779-mabs=@var{mode} -mnan=@var{encoding} @gol 780-mdsp -mno-dsp -mdspr2 -mno-dspr2 @gol 781-mmcu -mmno-mcu @gol 782-meva -mno-eva @gol 783-mvirt -mno-virt @gol 784-mmicromips -mno-micromips @gol 785-mfpu=@var{fpu-type} @gol 786-msmartmips -mno-smartmips @gol 787-mpaired-single -mno-paired-single -mdmx -mno-mdmx @gol 788-mips3d -mno-mips3d -mmt -mno-mt -mllsc -mno-llsc @gol 789-mlong64 -mlong32 -msym32 -mno-sym32 @gol 790-G@var{num} -mlocal-sdata -mno-local-sdata @gol 791-mextern-sdata -mno-extern-sdata -mgpopt -mno-gopt @gol 792-membedded-data -mno-embedded-data @gol 793-muninit-const-in-rodata -mno-uninit-const-in-rodata @gol 794-mcode-readable=@var{setting} @gol 795-msplit-addresses -mno-split-addresses @gol 796-mexplicit-relocs -mno-explicit-relocs @gol 797-mcheck-zero-division -mno-check-zero-division @gol 798-mdivide-traps -mdivide-breaks @gol 799-mmemcpy -mno-memcpy -mlong-calls -mno-long-calls @gol 800-mmad -mno-mad -mimadd -mno-imadd -mfused-madd -mno-fused-madd -nocpp @gol 801-mfix-24k -mno-fix-24k @gol 802-mfix-r4000 -mno-fix-r4000 -mfix-r4400 -mno-fix-r4400 @gol 803-mfix-r10000 -mno-fix-r10000 -mfix-rm7000 -mno-fix-rm7000 @gol 804-mfix-vr4120 -mno-fix-vr4120 @gol 805-mfix-vr4130 -mno-fix-vr4130 -mfix-sb1 -mno-fix-sb1 @gol 806-mflush-func=@var{func} -mno-flush-func @gol 807-mbranch-cost=@var{num} -mbranch-likely -mno-branch-likely @gol 808-mfp-exceptions -mno-fp-exceptions @gol 809-mvr4130-align -mno-vr4130-align -msynci -mno-synci @gol 810-mrelax-pic-calls -mno-relax-pic-calls -mmcount-ra-address} 811 812@emph{MMIX Options} 813@gccoptlist{-mlibfuncs -mno-libfuncs -mepsilon -mno-epsilon -mabi=gnu @gol 814-mabi=mmixware -mzero-extend -mknuthdiv -mtoplevel-symbols @gol 815-melf -mbranch-predict -mno-branch-predict -mbase-addresses @gol 816-mno-base-addresses -msingle-exit -mno-single-exit} 817 818@emph{MN10300 Options} 819@gccoptlist{-mmult-bug -mno-mult-bug @gol 820-mno-am33 -mam33 -mam33-2 -mam34 @gol 821-mtune=@var{cpu-type} @gol 822-mreturn-pointer-on-d0 @gol 823-mno-crt0 -mrelax -mliw -msetlb} 824 825@emph{Moxie Options} 826@gccoptlist{-meb -mel -mno-crt0} 827 828@emph{MSP430 Options} 829@gccoptlist{-msim -masm-hex -mmcu= -mcpu= -mlarge -msmall -mrelax @gol 830-mhwmult= -minrt} 831 832@emph{NDS32 Options} 833@gccoptlist{-mbig-endian -mlittle-endian @gol 834-mreduced-regs -mfull-regs @gol 835-mcmov -mno-cmov @gol 836-mperf-ext -mno-perf-ext @gol 837-mv3push -mno-v3push @gol 838-m16bit -mno-16bit @gol 839-mgp-direct -mno-gp-direct @gol 840-misr-vector-size=@var{num} @gol 841-mcache-block-size=@var{num} @gol 842-march=@var{arch} @gol 843-mforce-fp-as-gp -mforbid-fp-as-gp @gol 844-mex9 -mctor-dtor -mrelax} 845 846@emph{Nios II Options} 847@gccoptlist{-G @var{num} -mgpopt -mno-gpopt -mel -meb @gol 848-mno-bypass-cache -mbypass-cache @gol 849-mno-cache-volatile -mcache-volatile @gol 850-mno-fast-sw-div -mfast-sw-div @gol 851-mhw-mul -mno-hw-mul -mhw-mulx -mno-hw-mulx -mno-hw-div -mhw-div @gol 852-mcustom-@var{insn}=@var{N} -mno-custom-@var{insn} @gol 853-mcustom-fpu-cfg=@var{name} @gol 854-mhal -msmallc -msys-crt0=@var{name} -msys-lib=@var{name}} 855 856@emph{PDP-11 Options} 857@gccoptlist{-mfpu -msoft-float -mac0 -mno-ac0 -m40 -m45 -m10 @gol 858-mbcopy -mbcopy-builtin -mint32 -mno-int16 @gol 859-mint16 -mno-int32 -mfloat32 -mno-float64 @gol 860-mfloat64 -mno-float32 -mabshi -mno-abshi @gol 861-mbranch-expensive -mbranch-cheap @gol 862-munix-asm -mdec-asm} 863 864@emph{picoChip Options} 865@gccoptlist{-mae=@var{ae_type} -mvliw-lookahead=@var{N} @gol 866-msymbol-as-address -mno-inefficient-warnings} 867 868@emph{PowerPC Options} 869See RS/6000 and PowerPC Options. 870 871@emph{RL78 Options} 872@gccoptlist{-msim -mmul=none -mmul=g13 -mmul=rl78} 873 874@emph{RS/6000 and PowerPC Options} 875@gccoptlist{-mcpu=@var{cpu-type} @gol 876-mtune=@var{cpu-type} @gol 877-mcmodel=@var{code-model} @gol 878-mpowerpc64 @gol 879-maltivec -mno-altivec @gol 880-mpowerpc-gpopt -mno-powerpc-gpopt @gol 881-mpowerpc-gfxopt -mno-powerpc-gfxopt @gol 882-mmfcrf -mno-mfcrf -mpopcntb -mno-popcntb -mpopcntd -mno-popcntd @gol 883-mfprnd -mno-fprnd @gol 884-mcmpb -mno-cmpb -mmfpgpr -mno-mfpgpr -mhard-dfp -mno-hard-dfp @gol 885-mfull-toc -mminimal-toc -mno-fp-in-toc -mno-sum-in-toc @gol 886-m64 -m32 -mxl-compat -mno-xl-compat -mpe @gol 887-malign-power -malign-natural @gol 888-msoft-float -mhard-float -mmultiple -mno-multiple @gol 889-msingle-float -mdouble-float -msimple-fpu @gol 890-mstring -mno-string -mupdate -mno-update @gol 891-mavoid-indexed-addresses -mno-avoid-indexed-addresses @gol 892-mfused-madd -mno-fused-madd -mbit-align -mno-bit-align @gol 893-mstrict-align -mno-strict-align -mrelocatable @gol 894-mno-relocatable -mrelocatable-lib -mno-relocatable-lib @gol 895-mtoc -mno-toc -mlittle -mlittle-endian -mbig -mbig-endian @gol 896-mdynamic-no-pic -maltivec -mswdiv -msingle-pic-base @gol 897-mprioritize-restricted-insns=@var{priority} @gol 898-msched-costly-dep=@var{dependence_type} @gol 899-minsert-sched-nops=@var{scheme} @gol 900-mcall-sysv -mcall-netbsd @gol 901-maix-struct-return -msvr4-struct-return @gol 902-mabi=@var{abi-type} -msecure-plt -mbss-plt @gol 903-mblock-move-inline-limit=@var{num} @gol 904-misel -mno-isel @gol 905-misel=yes -misel=no @gol 906-mspe -mno-spe @gol 907-mspe=yes -mspe=no @gol 908-mpaired @gol 909-mgen-cell-microcode -mwarn-cell-microcode @gol 910-mvrsave -mno-vrsave @gol 911-mmulhw -mno-mulhw @gol 912-mdlmzb -mno-dlmzb @gol 913-mfloat-gprs=yes -mfloat-gprs=no -mfloat-gprs=single -mfloat-gprs=double @gol 914-mprototype -mno-prototype @gol 915-msim -mmvme -mads -myellowknife -memb -msdata @gol 916-msdata=@var{opt} -mvxworks -G @var{num} -pthread @gol 917-mrecip -mrecip=@var{opt} -mno-recip -mrecip-precision @gol 918-mno-recip-precision @gol 919-mveclibabi=@var{type} -mfriz -mno-friz @gol 920-mpointers-to-nested-functions -mno-pointers-to-nested-functions @gol 921-msave-toc-indirect -mno-save-toc-indirect @gol 922-mpower8-fusion -mno-mpower8-fusion -mpower8-vector -mno-power8-vector @gol 923-mcrypto -mno-crypto -mdirect-move -mno-direct-move @gol 924-mquad-memory -mno-quad-memory @gol 925-mquad-memory-atomic -mno-quad-memory-atomic @gol 926-mcompat-align-parm -mno-compat-align-parm} 927 928@emph{RX Options} 929@gccoptlist{-m64bit-doubles -m32bit-doubles -fpu -nofpu@gol 930-mcpu=@gol 931-mbig-endian-data -mlittle-endian-data @gol 932-msmall-data @gol 933-msim -mno-sim@gol 934-mas100-syntax -mno-as100-syntax@gol 935-mrelax@gol 936-mmax-constant-size=@gol 937-mint-register=@gol 938-mpid@gol 939-mno-warn-multiple-fast-interrupts@gol 940-msave-acc-in-interrupts} 941 942@emph{S/390 and zSeries Options} 943@gccoptlist{-mtune=@var{cpu-type} -march=@var{cpu-type} @gol 944-mhard-float -msoft-float -mhard-dfp -mno-hard-dfp @gol 945-mlong-double-64 -mlong-double-128 @gol 946-mbackchain -mno-backchain -mpacked-stack -mno-packed-stack @gol 947-msmall-exec -mno-small-exec -mmvcle -mno-mvcle @gol 948-m64 -m31 -mdebug -mno-debug -mesa -mzarch @gol 949-mtpf-trace -mno-tpf-trace -mfused-madd -mno-fused-madd @gol 950-mwarn-framesize -mwarn-dynamicstack -mstack-size -mstack-guard @gol 951-mhotpatch[=@var{halfwords}] -mno-hotpatch} 952 953@emph{Score Options} 954@gccoptlist{-meb -mel @gol 955-mnhwloop @gol 956-muls @gol 957-mmac @gol 958-mscore5 -mscore5u -mscore7 -mscore7d} 959 960@emph{SH Options} 961@gccoptlist{-m1 -m2 -m2e @gol 962-m2a-nofpu -m2a-single-only -m2a-single -m2a @gol 963-m3 -m3e @gol 964-m4-nofpu -m4-single-only -m4-single -m4 @gol 965-m4a-nofpu -m4a-single-only -m4a-single -m4a -m4al @gol 966-m5-64media -m5-64media-nofpu @gol 967-m5-32media -m5-32media-nofpu @gol 968-m5-compact -m5-compact-nofpu @gol 969-mb -ml -mdalign -mrelax @gol 970-mbigtable -mfmovd -mhitachi -mrenesas -mno-renesas -mnomacsave @gol 971-mieee -mno-ieee -mbitops -misize -minline-ic_invalidate -mpadstruct @gol 972-mspace -mprefergot -musermode -multcost=@var{number} -mdiv=@var{strategy} @gol 973-mdivsi3_libfunc=@var{name} -mfixed-range=@var{register-range} @gol 974-mindexed-addressing -mgettrcost=@var{number} -mpt-fixed @gol 975-maccumulate-outgoing-args -minvalid-symbols @gol 976-matomic-model=@var{atomic-model} @gol 977-mbranch-cost=@var{num} -mzdcbranch -mno-zdcbranch @gol 978-mfused-madd -mno-fused-madd -mfsca -mno-fsca -mfsrra -mno-fsrra @gol 979-mpretend-cmove -mtas} 980 981@emph{Solaris 2 Options} 982@gccoptlist{-mclear-hwcap -mno-clear-hwcap -mimpure-text -mno-impure-text @gol 983-pthreads -pthread} 984 985@emph{SPARC Options} 986@gccoptlist{-mcpu=@var{cpu-type} @gol 987-mtune=@var{cpu-type} @gol 988-mcmodel=@var{code-model} @gol 989-mmemory-model=@var{mem-model} @gol 990-m32 -m64 -mapp-regs -mno-app-regs @gol 991-mfaster-structs -mno-faster-structs -mflat -mno-flat @gol 992-mfpu -mno-fpu -mhard-float -msoft-float @gol 993-mhard-quad-float -msoft-quad-float @gol 994-mstack-bias -mno-stack-bias @gol 995-munaligned-doubles -mno-unaligned-doubles @gol 996-muser-mode -mno-user-mode @gol 997-mv8plus -mno-v8plus -mvis -mno-vis @gol 998-mvis2 -mno-vis2 -mvis3 -mno-vis3 @gol 999-mcbcond -mno-cbcond @gol 1000-mfmaf -mno-fmaf -mpopc -mno-popc @gol 1001-mfix-at697f -mfix-ut699} 1002 1003@emph{SPU Options} 1004@gccoptlist{-mwarn-reloc -merror-reloc @gol 1005-msafe-dma -munsafe-dma @gol 1006-mbranch-hints @gol 1007-msmall-mem -mlarge-mem -mstdmain @gol 1008-mfixed-range=@var{register-range} @gol 1009-mea32 -mea64 @gol 1010-maddress-space-conversion -mno-address-space-conversion @gol 1011-mcache-size=@var{cache-size} @gol 1012-matomic-updates -mno-atomic-updates} 1013 1014@emph{System V Options} 1015@gccoptlist{-Qy -Qn -YP,@var{paths} -Ym,@var{dir}} 1016 1017@emph{TILE-Gx Options} 1018@gccoptlist{-mcpu=CPU -m32 -m64 -mbig-endian -mlittle-endian @gol 1019-mcmodel=@var{code-model}} 1020 1021@emph{TILEPro Options} 1022@gccoptlist{-mcpu=@var{cpu} -m32} 1023 1024@emph{V850 Options} 1025@gccoptlist{-mlong-calls -mno-long-calls -mep -mno-ep @gol 1026-mprolog-function -mno-prolog-function -mspace @gol 1027-mtda=@var{n} -msda=@var{n} -mzda=@var{n} @gol 1028-mapp-regs -mno-app-regs @gol 1029-mdisable-callt -mno-disable-callt @gol 1030-mv850e2v3 -mv850e2 -mv850e1 -mv850es @gol 1031-mv850e -mv850 -mv850e3v5 @gol 1032-mloop @gol 1033-mrelax @gol 1034-mlong-jumps @gol 1035-msoft-float @gol 1036-mhard-float @gol 1037-mgcc-abi @gol 1038-mrh850-abi @gol 1039-mbig-switch} 1040 1041@emph{VAX Options} 1042@gccoptlist{-mg -mgnu -munix} 1043 1044@emph{VMS Options} 1045@gccoptlist{-mvms-return-codes -mdebug-main=@var{prefix} -mmalloc64 @gol 1046-mpointer-size=@var{size}} 1047 1048@emph{VxWorks Options} 1049@gccoptlist{-mrtp -non-static -Bstatic -Bdynamic @gol 1050-Xbind-lazy -Xbind-now} 1051 1052@emph{x86-64 Options} 1053See i386 and x86-64 Options. 1054 1055@emph{Xstormy16 Options} 1056@gccoptlist{-msim} 1057 1058@emph{Xtensa Options} 1059@gccoptlist{-mconst16 -mno-const16 @gol 1060-mfused-madd -mno-fused-madd @gol 1061-mforce-no-pic @gol 1062-mserialize-volatile -mno-serialize-volatile @gol 1063-mtext-section-literals -mno-text-section-literals @gol 1064-mtarget-align -mno-target-align @gol 1065-mlongcalls -mno-longcalls} 1066 1067@emph{zSeries Options} 1068See S/390 and zSeries Options. 1069 1070@item Code Generation Options 1071@xref{Code Gen Options,,Options for Code Generation Conventions}. 1072@gccoptlist{-fcall-saved-@var{reg} -fcall-used-@var{reg} @gol 1073-ffixed-@var{reg} -fexceptions @gol 1074-fnon-call-exceptions -fdelete-dead-exceptions -funwind-tables @gol 1075-fasynchronous-unwind-tables @gol 1076-fno-gnu-unique @gol 1077-finhibit-size-directive -finstrument-functions @gol 1078-finstrument-functions-exclude-function-list=@var{sym},@var{sym},@dots{} @gol 1079-finstrument-functions-exclude-file-list=@var{file},@var{file},@dots{} @gol 1080-fno-common -fno-ident @gol 1081-fpcc-struct-return -fpic -fPIC -fpie -fPIE @gol 1082-fno-jump-tables @gol 1083-frecord-gcc-switches @gol 1084-freg-struct-return -fshort-enums @gol 1085-fshort-double -fshort-wchar @gol 1086-fverbose-asm -fpack-struct[=@var{n}] -fstack-check @gol 1087-fstack-limit-register=@var{reg} -fstack-limit-symbol=@var{sym} @gol 1088-fno-stack-limit -fsplit-stack @gol 1089-fleading-underscore -ftls-model=@var{model} @gol 1090-fstack-reuse=@var{reuse_level} @gol 1091-ftrapv -fwrapv -fbounds-check @gol 1092-fvisibility -fstrict-volatile-bitfields -fsync-libcalls} 1093@end table 1094 1095 1096@node Overall Options 1097@section Options Controlling the Kind of Output 1098 1099Compilation can involve up to four stages: preprocessing, compilation 1100proper, assembly and linking, always in that order. GCC is capable of 1101preprocessing and compiling several files either into several 1102assembler input files, or into one assembler input file; then each 1103assembler input file produces an object file, and linking combines all 1104the object files (those newly compiled, and those specified as input) 1105into an executable file. 1106 1107@cindex file name suffix 1108For any given input file, the file name suffix determines what kind of 1109compilation is done: 1110 1111@table @gcctabopt 1112@item @var{file}.c 1113C source code that must be preprocessed. 1114 1115@item @var{file}.i 1116C source code that should not be preprocessed. 1117 1118@item @var{file}.ii 1119C++ source code that should not be preprocessed. 1120 1121@item @var{file}.m 1122Objective-C source code. Note that you must link with the @file{libobjc} 1123library to make an Objective-C program work. 1124 1125@item @var{file}.mi 1126Objective-C source code that should not be preprocessed. 1127 1128@item @var{file}.mm 1129@itemx @var{file}.M 1130Objective-C++ source code. Note that you must link with the @file{libobjc} 1131library to make an Objective-C++ program work. Note that @samp{.M} refers 1132to a literal capital M@. 1133 1134@item @var{file}.mii 1135Objective-C++ source code that should not be preprocessed. 1136 1137@item @var{file}.h 1138C, C++, Objective-C or Objective-C++ header file to be turned into a 1139precompiled header (default), or C, C++ header file to be turned into an 1140Ada spec (via the @option{-fdump-ada-spec} switch). 1141 1142@item @var{file}.cc 1143@itemx @var{file}.cp 1144@itemx @var{file}.cxx 1145@itemx @var{file}.cpp 1146@itemx @var{file}.CPP 1147@itemx @var{file}.c++ 1148@itemx @var{file}.C 1149C++ source code that must be preprocessed. Note that in @samp{.cxx}, 1150the last two letters must both be literally @samp{x}. Likewise, 1151@samp{.C} refers to a literal capital C@. 1152 1153@item @var{file}.mm 1154@itemx @var{file}.M 1155Objective-C++ source code that must be preprocessed. 1156 1157@item @var{file}.mii 1158Objective-C++ source code that should not be preprocessed. 1159 1160@item @var{file}.hh 1161@itemx @var{file}.H 1162@itemx @var{file}.hp 1163@itemx @var{file}.hxx 1164@itemx @var{file}.hpp 1165@itemx @var{file}.HPP 1166@itemx @var{file}.h++ 1167@itemx @var{file}.tcc 1168C++ header file to be turned into a precompiled header or Ada spec. 1169 1170@item @var{file}.f 1171@itemx @var{file}.for 1172@itemx @var{file}.ftn 1173Fixed form Fortran source code that should not be preprocessed. 1174 1175@item @var{file}.F 1176@itemx @var{file}.FOR 1177@itemx @var{file}.fpp 1178@itemx @var{file}.FPP 1179@itemx @var{file}.FTN 1180Fixed form Fortran source code that must be preprocessed (with the traditional 1181preprocessor). 1182 1183@item @var{file}.f90 1184@itemx @var{file}.f95 1185@itemx @var{file}.f03 1186@itemx @var{file}.f08 1187Free form Fortran source code that should not be preprocessed. 1188 1189@item @var{file}.F90 1190@itemx @var{file}.F95 1191@itemx @var{file}.F03 1192@itemx @var{file}.F08 1193Free form Fortran source code that must be preprocessed (with the 1194traditional preprocessor). 1195 1196@item @var{file}.go 1197Go source code. 1198 1199@c FIXME: Descriptions of Java file types. 1200@c @var{file}.java 1201@c @var{file}.class 1202@c @var{file}.zip 1203@c @var{file}.jar 1204 1205@item @var{file}.ads 1206Ada source code file that contains a library unit declaration (a 1207declaration of a package, subprogram, or generic, or a generic 1208instantiation), or a library unit renaming declaration (a package, 1209generic, or subprogram renaming declaration). Such files are also 1210called @dfn{specs}. 1211 1212@item @var{file}.adb 1213Ada source code file containing a library unit body (a subprogram or 1214package body). Such files are also called @dfn{bodies}. 1215 1216@c GCC also knows about some suffixes for languages not yet included: 1217@c Pascal: 1218@c @var{file}.p 1219@c @var{file}.pas 1220@c Ratfor: 1221@c @var{file}.r 1222 1223@item @var{file}.s 1224Assembler code. 1225 1226@item @var{file}.S 1227@itemx @var{file}.sx 1228Assembler code that must be preprocessed. 1229 1230@item @var{other} 1231An object file to be fed straight into linking. 1232Any file name with no recognized suffix is treated this way. 1233@end table 1234 1235@opindex x 1236You can specify the input language explicitly with the @option{-x} option: 1237 1238@table @gcctabopt 1239@item -x @var{language} 1240Specify explicitly the @var{language} for the following input files 1241(rather than letting the compiler choose a default based on the file 1242name suffix). This option applies to all following input files until 1243the next @option{-x} option. Possible values for @var{language} are: 1244@smallexample 1245c c-header cpp-output 1246c++ c++-header c++-cpp-output 1247objective-c objective-c-header objective-c-cpp-output 1248objective-c++ objective-c++-header objective-c++-cpp-output 1249assembler assembler-with-cpp 1250ada 1251f77 f77-cpp-input f95 f95-cpp-input 1252go 1253java 1254@end smallexample 1255 1256@item -x none 1257Turn off any specification of a language, so that subsequent files are 1258handled according to their file name suffixes (as they are if @option{-x} 1259has not been used at all). 1260 1261@item -pass-exit-codes 1262@opindex pass-exit-codes 1263Normally the @command{gcc} program exits with the code of 1 if any 1264phase of the compiler returns a non-success return code. If you specify 1265@option{-pass-exit-codes}, the @command{gcc} program instead returns with 1266the numerically highest error produced by any phase returning an error 1267indication. The C, C++, and Fortran front ends return 4 if an internal 1268compiler error is encountered. 1269@end table 1270 1271If you only want some of the stages of compilation, you can use 1272@option{-x} (or filename suffixes) to tell @command{gcc} where to start, and 1273one of the options @option{-c}, @option{-S}, or @option{-E} to say where 1274@command{gcc} is to stop. Note that some combinations (for example, 1275@samp{-x cpp-output -E}) instruct @command{gcc} to do nothing at all. 1276 1277@table @gcctabopt 1278@item -c 1279@opindex c 1280Compile or assemble the source files, but do not link. The linking 1281stage simply is not done. The ultimate output is in the form of an 1282object file for each source file. 1283 1284By default, the object file name for a source file is made by replacing 1285the suffix @samp{.c}, @samp{.i}, @samp{.s}, etc., with @samp{.o}. 1286 1287Unrecognized input files, not requiring compilation or assembly, are 1288ignored. 1289 1290@item -S 1291@opindex S 1292Stop after the stage of compilation proper; do not assemble. The output 1293is in the form of an assembler code file for each non-assembler input 1294file specified. 1295 1296By default, the assembler file name for a source file is made by 1297replacing the suffix @samp{.c}, @samp{.i}, etc., with @samp{.s}. 1298 1299Input files that don't require compilation are ignored. 1300 1301@item -E 1302@opindex E 1303Stop after the preprocessing stage; do not run the compiler proper. The 1304output is in the form of preprocessed source code, which is sent to the 1305standard output. 1306 1307Input files that don't require preprocessing are ignored. 1308 1309@cindex output file option 1310@item -o @var{file} 1311@opindex o 1312Place output in file @var{file}. This applies to whatever 1313sort of output is being produced, whether it be an executable file, 1314an object file, an assembler file or preprocessed C code. 1315 1316If @option{-o} is not specified, the default is to put an executable 1317file in @file{a.out}, the object file for 1318@file{@var{source}.@var{suffix}} in @file{@var{source}.o}, its 1319assembler file in @file{@var{source}.s}, a precompiled header file in 1320@file{@var{source}.@var{suffix}.gch}, and all preprocessed C source on 1321standard output. 1322 1323@item -v 1324@opindex v 1325Print (on standard error output) the commands executed to run the stages 1326of compilation. Also print the version number of the compiler driver 1327program and of the preprocessor and the compiler proper. 1328 1329@item -### 1330@opindex ### 1331Like @option{-v} except the commands are not executed and arguments 1332are quoted unless they contain only alphanumeric characters or @code{./-_}. 1333This is useful for shell scripts to capture the driver-generated command lines. 1334 1335@item -pipe 1336@opindex pipe 1337Use pipes rather than temporary files for communication between the 1338various stages of compilation. This fails to work on some systems where 1339the assembler is unable to read from a pipe; but the GNU assembler has 1340no trouble. 1341 1342@item --help 1343@opindex help 1344Print (on the standard output) a description of the command-line options 1345understood by @command{gcc}. If the @option{-v} option is also specified 1346then @option{--help} is also passed on to the various processes 1347invoked by @command{gcc}, so that they can display the command-line options 1348they accept. If the @option{-Wextra} option has also been specified 1349(prior to the @option{--help} option), then command-line options that 1350have no documentation associated with them are also displayed. 1351 1352@item --target-help 1353@opindex target-help 1354Print (on the standard output) a description of target-specific command-line 1355options for each tool. For some targets extra target-specific 1356information may also be printed. 1357 1358@item --help=@{@var{class}@r{|[}^@r{]}@var{qualifier}@}@r{[},@dots{}@r{]} 1359Print (on the standard output) a description of the command-line 1360options understood by the compiler that fit into all specified classes 1361and qualifiers. These are the supported classes: 1362 1363@table @asis 1364@item @samp{optimizers} 1365Display all of the optimization options supported by the 1366compiler. 1367 1368@item @samp{warnings} 1369Display all of the options controlling warning messages 1370produced by the compiler. 1371 1372@item @samp{target} 1373Display target-specific options. Unlike the 1374@option{--target-help} option however, target-specific options of the 1375linker and assembler are not displayed. This is because those 1376tools do not currently support the extended @option{--help=} syntax. 1377 1378@item @samp{params} 1379Display the values recognized by the @option{--param} 1380option. 1381 1382@item @var{language} 1383Display the options supported for @var{language}, where 1384@var{language} is the name of one of the languages supported in this 1385version of GCC@. 1386 1387@item @samp{common} 1388Display the options that are common to all languages. 1389@end table 1390 1391These are the supported qualifiers: 1392 1393@table @asis 1394@item @samp{undocumented} 1395Display only those options that are undocumented. 1396 1397@item @samp{joined} 1398Display options taking an argument that appears after an equal 1399sign in the same continuous piece of text, such as: 1400@samp{--help=target}. 1401 1402@item @samp{separate} 1403Display options taking an argument that appears as a separate word 1404following the original option, such as: @samp{-o output-file}. 1405@end table 1406 1407Thus for example to display all the undocumented target-specific 1408switches supported by the compiler, use: 1409 1410@smallexample 1411--help=target,undocumented 1412@end smallexample 1413 1414The sense of a qualifier can be inverted by prefixing it with the 1415@samp{^} character, so for example to display all binary warning 1416options (i.e., ones that are either on or off and that do not take an 1417argument) that have a description, use: 1418 1419@smallexample 1420--help=warnings,^joined,^undocumented 1421@end smallexample 1422 1423The argument to @option{--help=} should not consist solely of inverted 1424qualifiers. 1425 1426Combining several classes is possible, although this usually 1427restricts the output so much that there is nothing to display. One 1428case where it does work, however, is when one of the classes is 1429@var{target}. For example, to display all the target-specific 1430optimization options, use: 1431 1432@smallexample 1433--help=target,optimizers 1434@end smallexample 1435 1436The @option{--help=} option can be repeated on the command line. Each 1437successive use displays its requested class of options, skipping 1438those that have already been displayed. 1439 1440If the @option{-Q} option appears on the command line before the 1441@option{--help=} option, then the descriptive text displayed by 1442@option{--help=} is changed. Instead of describing the displayed 1443options, an indication is given as to whether the option is enabled, 1444disabled or set to a specific value (assuming that the compiler 1445knows this at the point where the @option{--help=} option is used). 1446 1447Here is a truncated example from the ARM port of @command{gcc}: 1448 1449@smallexample 1450 % gcc -Q -mabi=2 --help=target -c 1451 The following options are target specific: 1452 -mabi= 2 1453 -mabort-on-noreturn [disabled] 1454 -mapcs [disabled] 1455@end smallexample 1456 1457The output is sensitive to the effects of previous command-line 1458options, so for example it is possible to find out which optimizations 1459are enabled at @option{-O2} by using: 1460 1461@smallexample 1462-Q -O2 --help=optimizers 1463@end smallexample 1464 1465Alternatively you can discover which binary optimizations are enabled 1466by @option{-O3} by using: 1467 1468@smallexample 1469gcc -c -Q -O3 --help=optimizers > /tmp/O3-opts 1470gcc -c -Q -O2 --help=optimizers > /tmp/O2-opts 1471diff /tmp/O2-opts /tmp/O3-opts | grep enabled 1472@end smallexample 1473 1474@item -no-canonical-prefixes 1475@opindex no-canonical-prefixes 1476Do not expand any symbolic links, resolve references to @samp{/../} 1477or @samp{/./}, or make the path absolute when generating a relative 1478prefix. 1479 1480@item --version 1481@opindex version 1482Display the version number and copyrights of the invoked GCC@. 1483 1484@item -wrapper 1485@opindex wrapper 1486Invoke all subcommands under a wrapper program. The name of the 1487wrapper program and its parameters are passed as a comma separated 1488list. 1489 1490@smallexample 1491gcc -c t.c -wrapper gdb,--args 1492@end smallexample 1493 1494@noindent 1495This invokes all subprograms of @command{gcc} under 1496@samp{gdb --args}, thus the invocation of @command{cc1} is 1497@samp{gdb --args cc1 @dots{}}. 1498 1499@item -fplugin=@var{name}.so 1500@opindex fplugin 1501Load the plugin code in file @var{name}.so, assumed to be a 1502shared object to be dlopen'd by the compiler. The base name of 1503the shared object file is used to identify the plugin for the 1504purposes of argument parsing (See 1505@option{-fplugin-arg-@var{name}-@var{key}=@var{value}} below). 1506Each plugin should define the callback functions specified in the 1507Plugins API. 1508 1509@item -fplugin-arg-@var{name}-@var{key}=@var{value} 1510@opindex fplugin-arg 1511Define an argument called @var{key} with a value of @var{value} 1512for the plugin called @var{name}. 1513 1514@item -fdump-ada-spec@r{[}-slim@r{]} 1515@opindex fdump-ada-spec 1516For C and C++ source and include files, generate corresponding Ada specs. 1517@xref{Generating Ada Bindings for C and C++ headers,,, gnat_ugn, 1518GNAT User's Guide}, which provides detailed documentation on this feature. 1519 1520@item -fada-spec-parent=@var{unit} 1521@opindex fada-spec-parent 1522In conjunction with @option{-fdump-ada-spec@r{[}-slim@r{]}} above, generate 1523Ada specs as child units of parent @var{unit}. 1524 1525@item -fdump-go-spec=@var{file} 1526@opindex fdump-go-spec 1527For input files in any language, generate corresponding Go 1528declarations in @var{file}. This generates Go @code{const}, 1529@code{type}, @code{var}, and @code{func} declarations which may be a 1530useful way to start writing a Go interface to code written in some 1531other language. 1532 1533@include @value{srcdir}/../libiberty/at-file.texi 1534@end table 1535 1536@node Invoking G++ 1537@section Compiling C++ Programs 1538 1539@cindex suffixes for C++ source 1540@cindex C++ source file suffixes 1541C++ source files conventionally use one of the suffixes @samp{.C}, 1542@samp{.cc}, @samp{.cpp}, @samp{.CPP}, @samp{.c++}, @samp{.cp}, or 1543@samp{.cxx}; C++ header files often use @samp{.hh}, @samp{.hpp}, 1544@samp{.H}, or (for shared template code) @samp{.tcc}; and 1545preprocessed C++ files use the suffix @samp{.ii}. GCC recognizes 1546files with these names and compiles them as C++ programs even if you 1547call the compiler the same way as for compiling C programs (usually 1548with the name @command{gcc}). 1549 1550@findex g++ 1551@findex c++ 1552However, the use of @command{gcc} does not add the C++ library. 1553@command{g++} is a program that calls GCC and automatically specifies linking 1554against the C++ library. It treats @samp{.c}, 1555@samp{.h} and @samp{.i} files as C++ source files instead of C source 1556files unless @option{-x} is used. This program is also useful when 1557precompiling a C header file with a @samp{.h} extension for use in C++ 1558compilations. On many systems, @command{g++} is also installed with 1559the name @command{c++}. 1560 1561@cindex invoking @command{g++} 1562When you compile C++ programs, you may specify many of the same 1563command-line options that you use for compiling programs in any 1564language; or command-line options meaningful for C and related 1565languages; or options that are meaningful only for C++ programs. 1566@xref{C Dialect Options,,Options Controlling C Dialect}, for 1567explanations of options for languages related to C@. 1568@xref{C++ Dialect Options,,Options Controlling C++ Dialect}, for 1569explanations of options that are meaningful only for C++ programs. 1570 1571@node C Dialect Options 1572@section Options Controlling C Dialect 1573@cindex dialect options 1574@cindex language dialect options 1575@cindex options, dialect 1576 1577The following options control the dialect of C (or languages derived 1578from C, such as C++, Objective-C and Objective-C++) that the compiler 1579accepts: 1580 1581@table @gcctabopt 1582@cindex ANSI support 1583@cindex ISO support 1584@item -ansi 1585@opindex ansi 1586In C mode, this is equivalent to @option{-std=c90}. In C++ mode, it is 1587equivalent to @option{-std=c++98}. 1588 1589This turns off certain features of GCC that are incompatible with ISO 1590C90 (when compiling C code), or of standard C++ (when compiling C++ code), 1591such as the @code{asm} and @code{typeof} keywords, and 1592predefined macros such as @code{unix} and @code{vax} that identify the 1593type of system you are using. It also enables the undesirable and 1594rarely used ISO trigraph feature. For the C compiler, 1595it disables recognition of C++ style @samp{//} comments as well as 1596the @code{inline} keyword. 1597 1598The alternate keywords @code{__asm__}, @code{__extension__}, 1599@code{__inline__} and @code{__typeof__} continue to work despite 1600@option{-ansi}. You would not want to use them in an ISO C program, of 1601course, but it is useful to put them in header files that might be included 1602in compilations done with @option{-ansi}. Alternate predefined macros 1603such as @code{__unix__} and @code{__vax__} are also available, with or 1604without @option{-ansi}. 1605 1606The @option{-ansi} option does not cause non-ISO programs to be 1607rejected gratuitously. For that, @option{-Wpedantic} is required in 1608addition to @option{-ansi}. @xref{Warning Options}. 1609 1610The macro @code{__STRICT_ANSI__} is predefined when the @option{-ansi} 1611option is used. Some header files may notice this macro and refrain 1612from declaring certain functions or defining certain macros that the 1613ISO standard doesn't call for; this is to avoid interfering with any 1614programs that might use these names for other things. 1615 1616Functions that are normally built in but do not have semantics 1617defined by ISO C (such as @code{alloca} and @code{ffs}) are not built-in 1618functions when @option{-ansi} is used. @xref{Other Builtins,,Other 1619built-in functions provided by GCC}, for details of the functions 1620affected. 1621 1622@item -std= 1623@opindex std 1624Determine the language standard. @xref{Standards,,Language Standards 1625Supported by GCC}, for details of these standard versions. This option 1626is currently only supported when compiling C or C++. 1627 1628The compiler can accept several base standards, such as @samp{c90} or 1629@samp{c++98}, and GNU dialects of those standards, such as 1630@samp{gnu90} or @samp{gnu++98}. When a base standard is specified, the 1631compiler accepts all programs following that standard plus those 1632using GNU extensions that do not contradict it. For example, 1633@option{-std=c90} turns off certain features of GCC that are 1634incompatible with ISO C90, such as the @code{asm} and @code{typeof} 1635keywords, but not other GNU extensions that do not have a meaning in 1636ISO C90, such as omitting the middle term of a @code{?:} 1637expression. On the other hand, when a GNU dialect of a standard is 1638specified, all features supported by the compiler are enabled, even when 1639those features change the meaning of the base standard. As a result, some 1640strict-conforming programs may be rejected. The particular standard 1641is used by @option{-Wpedantic} to identify which features are GNU 1642extensions given that version of the standard. For example 1643@option{-std=gnu90 -Wpedantic} warns about C++ style @samp{//} 1644comments, while @option{-std=gnu99 -Wpedantic} does not. 1645 1646A value for this option must be provided; possible values are 1647 1648@table @samp 1649@item c90 1650@itemx c89 1651@itemx iso9899:1990 1652Support all ISO C90 programs (certain GNU extensions that conflict 1653with ISO C90 are disabled). Same as @option{-ansi} for C code. 1654 1655@item iso9899:199409 1656ISO C90 as modified in amendment 1. 1657 1658@item c99 1659@itemx c9x 1660@itemx iso9899:1999 1661@itemx iso9899:199x 1662ISO C99. This standard is substantially completely supported, modulo 1663bugs, extended identifiers (supported except for corner cases when 1664@option{-fextended-identifiers} is used) and floating-point issues 1665(mainly but not entirely relating to optional C99 features from 1666Annexes F and G). See 1667@w{@uref{http://gcc.gnu.org/c99status.html}} for more information. The 1668names @samp{c9x} and @samp{iso9899:199x} are deprecated. 1669 1670@item c11 1671@itemx c1x 1672@itemx iso9899:2011 1673ISO C11, the 2011 revision of the ISO C standard. This standard is 1674substantially completely supported, modulo bugs, extended identifiers 1675(supported except for corner cases when 1676@option{-fextended-identifiers} is used), floating-point issues 1677(mainly but not entirely relating to optional C11 features from 1678Annexes F and G) and the optional Annexes K (Bounds-checking 1679interfaces) and L (Analyzability). The name @samp{c1x} is deprecated. 1680 1681@item gnu90 1682@itemx gnu89 1683GNU dialect of ISO C90 (including some C99 features). This 1684is the default for C code. 1685 1686@item gnu99 1687@itemx gnu9x 1688GNU dialect of ISO C99. The name @samp{gnu9x} is deprecated. 1689 1690@item gnu11 1691@itemx gnu1x 1692GNU dialect of ISO C11. This is intended to become the default in a 1693future release of GCC. The name @samp{gnu1x} is deprecated. 1694 1695@item c++98 1696@itemx c++03 1697The 1998 ISO C++ standard plus the 2003 technical corrigendum and some 1698additional defect reports. Same as @option{-ansi} for C++ code. 1699 1700@item gnu++98 1701@itemx gnu++03 1702GNU dialect of @option{-std=c++98}. This is the default for 1703C++ code. 1704 1705@item c++11 1706@itemx c++0x 1707The 2011 ISO C++ standard plus amendments. 1708The name @samp{c++0x} is deprecated. 1709 1710@item gnu++11 1711@itemx gnu++0x 1712GNU dialect of @option{-std=c++11}. 1713The name @samp{gnu++0x} is deprecated. 1714 1715@item c++1y 1716The next revision of the ISO C++ standard, tentatively planned for 17172014. Support is highly experimental, and will almost certainly 1718change in incompatible ways in future releases. 1719 1720@item gnu++1y 1721GNU dialect of @option{-std=c++1y}. Support is highly experimental, 1722and will almost certainly change in incompatible ways in future 1723releases. 1724@end table 1725 1726@item -fgnu89-inline 1727@opindex fgnu89-inline 1728The option @option{-fgnu89-inline} tells GCC to use the traditional 1729GNU semantics for @code{inline} functions when in C99 mode. 1730@xref{Inline,,An Inline Function is As Fast As a Macro}. This option 1731is accepted and ignored by GCC versions 4.1.3 up to but not including 17324.3. In GCC versions 4.3 and later it changes the behavior of GCC in 1733C99 mode. Using this option is roughly equivalent to adding the 1734@code{gnu_inline} function attribute to all inline functions 1735(@pxref{Function Attributes}). 1736 1737The option @option{-fno-gnu89-inline} explicitly tells GCC to use the 1738C99 semantics for @code{inline} when in C99 or gnu99 mode (i.e., it 1739specifies the default behavior). This option was first supported in 1740GCC 4.3. This option is not supported in @option{-std=c90} or 1741@option{-std=gnu90} mode. 1742 1743The preprocessor macros @code{__GNUC_GNU_INLINE__} and 1744@code{__GNUC_STDC_INLINE__} may be used to check which semantics are 1745in effect for @code{inline} functions. @xref{Common Predefined 1746Macros,,,cpp,The C Preprocessor}. 1747 1748@item -aux-info @var{filename} 1749@opindex aux-info 1750Output to the given filename prototyped declarations for all functions 1751declared and/or defined in a translation unit, including those in header 1752files. This option is silently ignored in any language other than C@. 1753 1754Besides declarations, the file indicates, in comments, the origin of 1755each declaration (source file and line), whether the declaration was 1756implicit, prototyped or unprototyped (@samp{I}, @samp{N} for new or 1757@samp{O} for old, respectively, in the first character after the line 1758number and the colon), and whether it came from a declaration or a 1759definition (@samp{C} or @samp{F}, respectively, in the following 1760character). In the case of function definitions, a K&R-style list of 1761arguments followed by their declarations is also provided, inside 1762comments, after the declaration. 1763 1764@item -fallow-parameterless-variadic-functions 1765@opindex fallow-parameterless-variadic-functions 1766Accept variadic functions without named parameters. 1767 1768Although it is possible to define such a function, this is not very 1769useful as it is not possible to read the arguments. This is only 1770supported for C as this construct is allowed by C++. 1771 1772@item -fno-asm 1773@opindex fno-asm 1774Do not recognize @code{asm}, @code{inline} or @code{typeof} as a 1775keyword, so that code can use these words as identifiers. You can use 1776the keywords @code{__asm__}, @code{__inline__} and @code{__typeof__} 1777instead. @option{-ansi} implies @option{-fno-asm}. 1778 1779In C++, this switch only affects the @code{typeof} keyword, since 1780@code{asm} and @code{inline} are standard keywords. You may want to 1781use the @option{-fno-gnu-keywords} flag instead, which has the same 1782effect. In C99 mode (@option{-std=c99} or @option{-std=gnu99}), this 1783switch only affects the @code{asm} and @code{typeof} keywords, since 1784@code{inline} is a standard keyword in ISO C99. 1785 1786@item -fno-builtin 1787@itemx -fno-builtin-@var{function} 1788@opindex fno-builtin 1789@cindex built-in functions 1790Don't recognize built-in functions that do not begin with 1791@samp{__builtin_} as prefix. @xref{Other Builtins,,Other built-in 1792functions provided by GCC}, for details of the functions affected, 1793including those which are not built-in functions when @option{-ansi} or 1794@option{-std} options for strict ISO C conformance are used because they 1795do not have an ISO standard meaning. 1796 1797GCC normally generates special code to handle certain built-in functions 1798more efficiently; for instance, calls to @code{alloca} may become single 1799instructions which adjust the stack directly, and calls to @code{memcpy} 1800may become inline copy loops. The resulting code is often both smaller 1801and faster, but since the function calls no longer appear as such, you 1802cannot set a breakpoint on those calls, nor can you change the behavior 1803of the functions by linking with a different library. In addition, 1804when a function is recognized as a built-in function, GCC may use 1805information about that function to warn about problems with calls to 1806that function, or to generate more efficient code, even if the 1807resulting code still contains calls to that function. For example, 1808warnings are given with @option{-Wformat} for bad calls to 1809@code{printf} when @code{printf} is built in and @code{strlen} is 1810known not to modify global memory. 1811 1812With the @option{-fno-builtin-@var{function}} option 1813only the built-in function @var{function} is 1814disabled. @var{function} must not begin with @samp{__builtin_}. If a 1815function is named that is not built-in in this version of GCC, this 1816option is ignored. There is no corresponding 1817@option{-fbuiltin-@var{function}} option; if you wish to enable 1818built-in functions selectively when using @option{-fno-builtin} or 1819@option{-ffreestanding}, you may define macros such as: 1820 1821@smallexample 1822#define abs(n) __builtin_abs ((n)) 1823#define strcpy(d, s) __builtin_strcpy ((d), (s)) 1824@end smallexample 1825 1826@item -fhosted 1827@opindex fhosted 1828@cindex hosted environment 1829 1830Assert that compilation targets a hosted environment. This implies 1831@option{-fbuiltin}. A hosted environment is one in which the 1832entire standard library is available, and in which @code{main} has a return 1833type of @code{int}. Examples are nearly everything except a kernel. 1834This is equivalent to @option{-fno-freestanding}. 1835 1836@item -ffreestanding 1837@opindex ffreestanding 1838@cindex hosted environment 1839 1840Assert that compilation targets a freestanding environment. This 1841implies @option{-fno-builtin}. A freestanding environment 1842is one in which the standard library may not exist, and program startup may 1843not necessarily be at @code{main}. The most obvious example is an OS kernel. 1844This is equivalent to @option{-fno-hosted}. 1845 1846@xref{Standards,,Language Standards Supported by GCC}, for details of 1847freestanding and hosted environments. 1848 1849@item -fopenmp 1850@opindex fopenmp 1851@cindex OpenMP parallel 1852Enable handling of OpenMP directives @code{#pragma omp} in C/C++ and 1853@code{!$omp} in Fortran. When @option{-fopenmp} is specified, the 1854compiler generates parallel code according to the OpenMP Application 1855Program Interface v4.0 @w{@uref{http://www.openmp.org/}}. This option 1856implies @option{-pthread}, and thus is only supported on targets that 1857have support for @option{-pthread}. @option{-fopenmp} implies 1858@option{-fopenmp-simd}. 1859 1860@item -fopenmp-simd 1861@opindex fopenmp-simd 1862@cindex OpenMP SIMD 1863@cindex SIMD 1864Enable handling of OpenMP's SIMD directives with @code{#pragma omp} 1865in C/C++ and @code{!$omp} in Fortran. Other OpenMP directives 1866are ignored. 1867 1868@item -fcilkplus 1869@opindex fcilkplus 1870@cindex Enable Cilk Plus 1871Enable the usage of Cilk Plus language extension features for C/C++. 1872When the option @option{-fcilkplus} is specified, enable the usage of 1873the Cilk Plus Language extension features for C/C++. The present 1874implementation follows ABI version 1.2. This is an experimental 1875feature that is only partially complete, and whose interface may 1876change in future versions of GCC as the official specification 1877changes. Currently, all features but @code{_Cilk_for} have been 1878implemented. 1879 1880@item -fgnu-tm 1881@opindex fgnu-tm 1882When the option @option{-fgnu-tm} is specified, the compiler 1883generates code for the Linux variant of Intel's current Transactional 1884Memory ABI specification document (Revision 1.1, May 6 2009). This is 1885an experimental feature whose interface may change in future versions 1886of GCC, as the official specification changes. Please note that not 1887all architectures are supported for this feature. 1888 1889For more information on GCC's support for transactional memory, 1890@xref{Enabling libitm,,The GNU Transactional Memory Library,libitm,GNU 1891Transactional Memory Library}. 1892 1893Note that the transactional memory feature is not supported with 1894non-call exceptions (@option{-fnon-call-exceptions}). 1895 1896@item -fms-extensions 1897@opindex fms-extensions 1898Accept some non-standard constructs used in Microsoft header files. 1899 1900In C++ code, this allows member names in structures to be similar 1901to previous types declarations. 1902 1903@smallexample 1904typedef int UOW; 1905struct ABC @{ 1906 UOW UOW; 1907@}; 1908@end smallexample 1909 1910Some cases of unnamed fields in structures and unions are only 1911accepted with this option. @xref{Unnamed Fields,,Unnamed struct/union 1912fields within structs/unions}, for details. 1913 1914Note that this option is off for all targets but i?86 and x86_64 1915targets using ms-abi. 1916@item -fplan9-extensions 1917Accept some non-standard constructs used in Plan 9 code. 1918 1919This enables @option{-fms-extensions}, permits passing pointers to 1920structures with anonymous fields to functions that expect pointers to 1921elements of the type of the field, and permits referring to anonymous 1922fields declared using a typedef. @xref{Unnamed Fields,,Unnamed 1923struct/union fields within structs/unions}, for details. This is only 1924supported for C, not C++. 1925 1926@item -trigraphs 1927@opindex trigraphs 1928Support ISO C trigraphs. The @option{-ansi} option (and @option{-std} 1929options for strict ISO C conformance) implies @option{-trigraphs}. 1930 1931@cindex traditional C language 1932@cindex C language, traditional 1933@item -traditional 1934@itemx -traditional-cpp 1935@opindex traditional-cpp 1936@opindex traditional 1937Formerly, these options caused GCC to attempt to emulate a pre-standard 1938C compiler. They are now only supported with the @option{-E} switch. 1939The preprocessor continues to support a pre-standard mode. See the GNU 1940CPP manual for details. 1941 1942@item -fcond-mismatch 1943@opindex fcond-mismatch 1944Allow conditional expressions with mismatched types in the second and 1945third arguments. The value of such an expression is void. This option 1946is not supported for C++. 1947 1948@item -flax-vector-conversions 1949@opindex flax-vector-conversions 1950Allow implicit conversions between vectors with differing numbers of 1951elements and/or incompatible element types. This option should not be 1952used for new code. 1953 1954@item -funsigned-char 1955@opindex funsigned-char 1956Let the type @code{char} be unsigned, like @code{unsigned char}. 1957 1958Each kind of machine has a default for what @code{char} should 1959be. It is either like @code{unsigned char} by default or like 1960@code{signed char} by default. 1961 1962Ideally, a portable program should always use @code{signed char} or 1963@code{unsigned char} when it depends on the signedness of an object. 1964But many programs have been written to use plain @code{char} and 1965expect it to be signed, or expect it to be unsigned, depending on the 1966machines they were written for. This option, and its inverse, let you 1967make such a program work with the opposite default. 1968 1969The type @code{char} is always a distinct type from each of 1970@code{signed char} or @code{unsigned char}, even though its behavior 1971is always just like one of those two. 1972 1973@item -fsigned-char 1974@opindex fsigned-char 1975Let the type @code{char} be signed, like @code{signed char}. 1976 1977Note that this is equivalent to @option{-fno-unsigned-char}, which is 1978the negative form of @option{-funsigned-char}. Likewise, the option 1979@option{-fno-signed-char} is equivalent to @option{-funsigned-char}. 1980 1981@item -fsigned-bitfields 1982@itemx -funsigned-bitfields 1983@itemx -fno-signed-bitfields 1984@itemx -fno-unsigned-bitfields 1985@opindex fsigned-bitfields 1986@opindex funsigned-bitfields 1987@opindex fno-signed-bitfields 1988@opindex fno-unsigned-bitfields 1989These options control whether a bit-field is signed or unsigned, when the 1990declaration does not use either @code{signed} or @code{unsigned}. By 1991default, such a bit-field is signed, because this is consistent: the 1992basic integer types such as @code{int} are signed types. 1993@end table 1994 1995@node C++ Dialect Options 1996@section Options Controlling C++ Dialect 1997 1998@cindex compiler options, C++ 1999@cindex C++ options, command-line 2000@cindex options, C++ 2001This section describes the command-line options that are only meaningful 2002for C++ programs. You can also use most of the GNU compiler options 2003regardless of what language your program is in. For example, you 2004might compile a file @code{firstClass.C} like this: 2005 2006@smallexample 2007g++ -g -frepo -O -c firstClass.C 2008@end smallexample 2009 2010@noindent 2011In this example, only @option{-frepo} is an option meant 2012only for C++ programs; you can use the other options with any 2013language supported by GCC@. 2014 2015Here is a list of options that are @emph{only} for compiling C++ programs: 2016 2017@table @gcctabopt 2018 2019@item -fabi-version=@var{n} 2020@opindex fabi-version 2021Use version @var{n} of the C++ ABI@. The default is version 2. 2022 2023Version 0 refers to the version conforming most closely to 2024the C++ ABI specification. Therefore, the ABI obtained using version 0 2025will change in different versions of G++ as ABI bugs are fixed. 2026 2027Version 1 is the version of the C++ ABI that first appeared in G++ 3.2. 2028 2029Version 2 is the version of the C++ ABI that first appeared in G++ 3.4. 2030 2031Version 3 corrects an error in mangling a constant address as a 2032template argument. 2033 2034Version 4, which first appeared in G++ 4.5, implements a standard 2035mangling for vector types. 2036 2037Version 5, which first appeared in G++ 4.6, corrects the mangling of 2038attribute const/volatile on function pointer types, decltype of a 2039plain decl, and use of a function parameter in the declaration of 2040another parameter. 2041 2042Version 6, which first appeared in G++ 4.7, corrects the promotion 2043behavior of C++11 scoped enums and the mangling of template argument 2044packs, const/static_cast, prefix ++ and --, and a class scope function 2045used as a template argument. 2046 2047See also @option{-Wabi}. 2048 2049@item -fno-access-control 2050@opindex fno-access-control 2051Turn off all access checking. This switch is mainly useful for working 2052around bugs in the access control code. 2053 2054@item -fcheck-new 2055@opindex fcheck-new 2056Check that the pointer returned by @code{operator new} is non-null 2057before attempting to modify the storage allocated. This check is 2058normally unnecessary because the C++ standard specifies that 2059@code{operator new} only returns @code{0} if it is declared 2060@samp{throw()}, in which case the compiler always checks the 2061return value even without this option. In all other cases, when 2062@code{operator new} has a non-empty exception specification, memory 2063exhaustion is signalled by throwing @code{std::bad_alloc}. See also 2064@samp{new (nothrow)}. 2065 2066@item -fconstexpr-depth=@var{n} 2067@opindex fconstexpr-depth 2068Set the maximum nested evaluation depth for C++11 constexpr functions 2069to @var{n}. A limit is needed to detect endless recursion during 2070constant expression evaluation. The minimum specified by the standard 2071is 512. 2072 2073@item -fdeduce-init-list 2074@opindex fdeduce-init-list 2075Enable deduction of a template type parameter as 2076@code{std::initializer_list} from a brace-enclosed initializer list, i.e.@: 2077 2078@smallexample 2079template <class T> auto forward(T t) -> decltype (realfn (t)) 2080@{ 2081 return realfn (t); 2082@} 2083 2084void f() 2085@{ 2086 forward(@{1,2@}); // call forward<std::initializer_list<int>> 2087@} 2088@end smallexample 2089 2090This deduction was implemented as a possible extension to the 2091originally proposed semantics for the C++11 standard, but was not part 2092of the final standard, so it is disabled by default. This option is 2093deprecated, and may be removed in a future version of G++. 2094 2095@item -ffriend-injection 2096@opindex ffriend-injection 2097Inject friend functions into the enclosing namespace, so that they are 2098visible outside the scope of the class in which they are declared. 2099Friend functions were documented to work this way in the old Annotated 2100C++ Reference Manual, and versions of G++ before 4.1 always worked 2101that way. However, in ISO C++ a friend function that is not declared 2102in an enclosing scope can only be found using argument dependent 2103lookup. This option causes friends to be injected as they were in 2104earlier releases. 2105 2106This option is for compatibility, and may be removed in a future 2107release of G++. 2108 2109@item -fno-elide-constructors 2110@opindex fno-elide-constructors 2111The C++ standard allows an implementation to omit creating a temporary 2112that is only used to initialize another object of the same type. 2113Specifying this option disables that optimization, and forces G++ to 2114call the copy constructor in all cases. 2115 2116@item -fno-enforce-eh-specs 2117@opindex fno-enforce-eh-specs 2118Don't generate code to check for violation of exception specifications 2119at run time. This option violates the C++ standard, but may be useful 2120for reducing code size in production builds, much like defining 2121@samp{NDEBUG}. This does not give user code permission to throw 2122exceptions in violation of the exception specifications; the compiler 2123still optimizes based on the specifications, so throwing an 2124unexpected exception results in undefined behavior at run time. 2125 2126@item -fextern-tls-init 2127@itemx -fno-extern-tls-init 2128@opindex fextern-tls-init 2129@opindex fno-extern-tls-init 2130The C++11 and OpenMP standards allow @samp{thread_local} and 2131@samp{threadprivate} variables to have dynamic (runtime) 2132initialization. To support this, any use of such a variable goes 2133through a wrapper function that performs any necessary initialization. 2134When the use and definition of the variable are in the same 2135translation unit, this overhead can be optimized away, but when the 2136use is in a different translation unit there is significant overhead 2137even if the variable doesn't actually need dynamic initialization. If 2138the programmer can be sure that no use of the variable in a 2139non-defining TU needs to trigger dynamic initialization (either 2140because the variable is statically initialized, or a use of the 2141variable in the defining TU will be executed before any uses in 2142another TU), they can avoid this overhead with the 2143@option{-fno-extern-tls-init} option. 2144 2145On targets that support symbol aliases, the default is 2146@option{-fextern-tls-init}. On targets that do not support symbol 2147aliases, the default is @option{-fno-extern-tls-init}. 2148 2149@item -ffor-scope 2150@itemx -fno-for-scope 2151@opindex ffor-scope 2152@opindex fno-for-scope 2153If @option{-ffor-scope} is specified, the scope of variables declared in 2154a @i{for-init-statement} is limited to the @samp{for} loop itself, 2155as specified by the C++ standard. 2156If @option{-fno-for-scope} is specified, the scope of variables declared in 2157a @i{for-init-statement} extends to the end of the enclosing scope, 2158as was the case in old versions of G++, and other (traditional) 2159implementations of C++. 2160 2161If neither flag is given, the default is to follow the standard, 2162but to allow and give a warning for old-style code that would 2163otherwise be invalid, or have different behavior. 2164 2165@item -fno-gnu-keywords 2166@opindex fno-gnu-keywords 2167Do not recognize @code{typeof} as a keyword, so that code can use this 2168word as an identifier. You can use the keyword @code{__typeof__} instead. 2169@option{-ansi} implies @option{-fno-gnu-keywords}. 2170 2171@item -fno-implicit-templates 2172@opindex fno-implicit-templates 2173Never emit code for non-inline templates that are instantiated 2174implicitly (i.e.@: by use); only emit code for explicit instantiations. 2175@xref{Template Instantiation}, for more information. 2176 2177@item -fno-implicit-inline-templates 2178@opindex fno-implicit-inline-templates 2179Don't emit code for implicit instantiations of inline templates, either. 2180The default is to handle inlines differently so that compiles with and 2181without optimization need the same set of explicit instantiations. 2182 2183@item -fno-implement-inlines 2184@opindex fno-implement-inlines 2185To save space, do not emit out-of-line copies of inline functions 2186controlled by @samp{#pragma implementation}. This causes linker 2187errors if these functions are not inlined everywhere they are called. 2188 2189@item -fms-extensions 2190@opindex fms-extensions 2191Disable Wpedantic warnings about constructs used in MFC, such as implicit 2192int and getting a pointer to member function via non-standard syntax. 2193 2194@item -fno-nonansi-builtins 2195@opindex fno-nonansi-builtins 2196Disable built-in declarations of functions that are not mandated by 2197ANSI/ISO C@. These include @code{ffs}, @code{alloca}, @code{_exit}, 2198@code{index}, @code{bzero}, @code{conjf}, and other related functions. 2199 2200@item -fnothrow-opt 2201@opindex fnothrow-opt 2202Treat a @code{throw()} exception specification as if it were a 2203@code{noexcept} specification to reduce or eliminate the text size 2204overhead relative to a function with no exception specification. If 2205the function has local variables of types with non-trivial 2206destructors, the exception specification actually makes the 2207function smaller because the EH cleanups for those variables can be 2208optimized away. The semantic effect is that an exception thrown out of 2209a function with such an exception specification results in a call 2210to @code{terminate} rather than @code{unexpected}. 2211 2212@item -fno-operator-names 2213@opindex fno-operator-names 2214Do not treat the operator name keywords @code{and}, @code{bitand}, 2215@code{bitor}, @code{compl}, @code{not}, @code{or} and @code{xor} as 2216synonyms as keywords. 2217 2218@item -fno-optional-diags 2219@opindex fno-optional-diags 2220Disable diagnostics that the standard says a compiler does not need to 2221issue. Currently, the only such diagnostic issued by G++ is the one for 2222a name having multiple meanings within a class. 2223 2224@item -fpermissive 2225@opindex fpermissive 2226Downgrade some diagnostics about nonconformant code from errors to 2227warnings. Thus, using @option{-fpermissive} allows some 2228nonconforming code to compile. 2229 2230@item -fno-pretty-templates 2231@opindex fno-pretty-templates 2232When an error message refers to a specialization of a function 2233template, the compiler normally prints the signature of the 2234template followed by the template arguments and any typedefs or 2235typenames in the signature (e.g. @code{void f(T) [with T = int]} 2236rather than @code{void f(int)}) so that it's clear which template is 2237involved. When an error message refers to a specialization of a class 2238template, the compiler omits any template arguments that match 2239the default template arguments for that template. If either of these 2240behaviors make it harder to understand the error message rather than 2241easier, you can use @option{-fno-pretty-templates} to disable them. 2242 2243@item -frepo 2244@opindex frepo 2245Enable automatic template instantiation at link time. This option also 2246implies @option{-fno-implicit-templates}. @xref{Template 2247Instantiation}, for more information. 2248 2249@item -fno-rtti 2250@opindex fno-rtti 2251Disable generation of information about every class with virtual 2252functions for use by the C++ run-time type identification features 2253(@samp{dynamic_cast} and @samp{typeid}). If you don't use those parts 2254of the language, you can save some space by using this flag. Note that 2255exception handling uses the same information, but G++ generates it as 2256needed. The @samp{dynamic_cast} operator can still be used for casts that 2257do not require run-time type information, i.e.@: casts to @code{void *} or to 2258unambiguous base classes. 2259 2260@item -fstats 2261@opindex fstats 2262Emit statistics about front-end processing at the end of the compilation. 2263This information is generally only useful to the G++ development team. 2264 2265@item -fstrict-enums 2266@opindex fstrict-enums 2267Allow the compiler to optimize using the assumption that a value of 2268enumerated type can only be one of the values of the enumeration (as 2269defined in the C++ standard; basically, a value that can be 2270represented in the minimum number of bits needed to represent all the 2271enumerators). This assumption may not be valid if the program uses a 2272cast to convert an arbitrary integer value to the enumerated type. 2273 2274@item -ftemplate-backtrace-limit=@var{n} 2275@opindex ftemplate-backtrace-limit 2276Set the maximum number of template instantiation notes for a single 2277warning or error to @var{n}. The default value is 10. 2278 2279@item -ftemplate-depth=@var{n} 2280@opindex ftemplate-depth 2281Set the maximum instantiation depth for template classes to @var{n}. 2282A limit on the template instantiation depth is needed to detect 2283endless recursions during template class instantiation. ANSI/ISO C++ 2284conforming programs must not rely on a maximum depth greater than 17 2285(changed to 1024 in C++11). The default value is 900, as the compiler 2286can run out of stack space before hitting 1024 in some situations. 2287 2288@item -fno-threadsafe-statics 2289@opindex fno-threadsafe-statics 2290Do not emit the extra code to use the routines specified in the C++ 2291ABI for thread-safe initialization of local statics. You can use this 2292option to reduce code size slightly in code that doesn't need to be 2293thread-safe. 2294 2295@item -fuse-cxa-atexit 2296@opindex fuse-cxa-atexit 2297Register destructors for objects with static storage duration with the 2298@code{__cxa_atexit} function rather than the @code{atexit} function. 2299This option is required for fully standards-compliant handling of static 2300destructors, but only works if your C library supports 2301@code{__cxa_atexit}. 2302 2303@item -fno-use-cxa-get-exception-ptr 2304@opindex fno-use-cxa-get-exception-ptr 2305Don't use the @code{__cxa_get_exception_ptr} runtime routine. This 2306causes @code{std::uncaught_exception} to be incorrect, but is necessary 2307if the runtime routine is not available. 2308 2309@item -fvisibility-inlines-hidden 2310@opindex fvisibility-inlines-hidden 2311This switch declares that the user does not attempt to compare 2312pointers to inline functions or methods where the addresses of the two functions 2313are taken in different shared objects. 2314 2315The effect of this is that GCC may, effectively, mark inline methods with 2316@code{__attribute__ ((visibility ("hidden")))} so that they do not 2317appear in the export table of a DSO and do not require a PLT indirection 2318when used within the DSO@. Enabling this option can have a dramatic effect 2319on load and link times of a DSO as it massively reduces the size of the 2320dynamic export table when the library makes heavy use of templates. 2321 2322The behavior of this switch is not quite the same as marking the 2323methods as hidden directly, because it does not affect static variables 2324local to the function or cause the compiler to deduce that 2325the function is defined in only one shared object. 2326 2327You may mark a method as having a visibility explicitly to negate the 2328effect of the switch for that method. For example, if you do want to 2329compare pointers to a particular inline method, you might mark it as 2330having default visibility. Marking the enclosing class with explicit 2331visibility has no effect. 2332 2333Explicitly instantiated inline methods are unaffected by this option 2334as their linkage might otherwise cross a shared library boundary. 2335@xref{Template Instantiation}. 2336 2337@item -fvisibility-ms-compat 2338@opindex fvisibility-ms-compat 2339This flag attempts to use visibility settings to make GCC's C++ 2340linkage model compatible with that of Microsoft Visual Studio. 2341 2342The flag makes these changes to GCC's linkage model: 2343 2344@enumerate 2345@item 2346It sets the default visibility to @code{hidden}, like 2347@option{-fvisibility=hidden}. 2348 2349@item 2350Types, but not their members, are not hidden by default. 2351 2352@item 2353The One Definition Rule is relaxed for types without explicit 2354visibility specifications that are defined in more than one 2355shared object: those declarations are permitted if they are 2356permitted when this option is not used. 2357@end enumerate 2358 2359In new code it is better to use @option{-fvisibility=hidden} and 2360export those classes that are intended to be externally visible. 2361Unfortunately it is possible for code to rely, perhaps accidentally, 2362on the Visual Studio behavior. 2363 2364Among the consequences of these changes are that static data members 2365of the same type with the same name but defined in different shared 2366objects are different, so changing one does not change the other; 2367and that pointers to function members defined in different shared 2368objects may not compare equal. When this flag is given, it is a 2369violation of the ODR to define types with the same name differently. 2370 2371@item -fvtable-verify=@var{std|preinit|none} 2372@opindex fvtable-verify 2373Turn on (or off, if using @option{-fvtable-verify=none}) the security 2374feature that verifies at runtime, for every virtual call that is made, that 2375the vtable pointer through which the call is made is valid for the type of 2376the object, and has not been corrupted or overwritten. If an invalid vtable 2377pointer is detected (at runtime), an error is reported and execution of the 2378program is immediately halted. 2379 2380This option causes runtime data structures to be built, at program start up, 2381for verifying the vtable pointers. The options @code{std} and @code{preinit} 2382control the timing of when these data structures are built. In both cases the 2383data structures are built before execution reaches 'main'. The 2384@option{-fvtable-verify=std} causes these data structure to be built after the 2385shared libraries have been loaded and initialized. 2386@option{-fvtable-verify=preinit} causes them to be built before the shared 2387libraries have been loaded and initialized. 2388 2389If this option appears multiple times in the compiler line, with different 2390values specified, 'none' will take highest priority over both 'std' and 2391'preinit'; 'preinit' will take priority over 'std'. 2392 2393@item -fvtv-debug 2394@opindex (fvtv-debug) 2395Causes debug versions of the runtime functions for the vtable verification 2396feature to be called. This assumes the @option{-fvtable-verify=std} or 2397@option{-fvtable-verify=preinit} has been used. This flag will also cause the 2398compiler to keep track of which vtable pointers it found for each class, and 2399record that information in the file ``vtv_set_ptr_data.log'', in the dump 2400file directory on the user's machine. 2401 2402Note: This feature APPENDS data to the log file. If you want a fresh log 2403file, be sure to delete any existing one. 2404 2405@item -fvtv-counts 2406@opindex fvtv-counts 2407This is a debugging flag. When used in conjunction with 2408@option{-fvtable-verify=std} or @option{-fvtable-verify=preinit}, this 2409causes the compiler to keep track of the total number of virtual calls 2410it encountered and the number of verifications it inserted. It also 2411counts the number of calls to certain runtime library functions 2412that it inserts. This information, for each compilation unit, is written 2413to a file named ``vtv_count_data.log'', in the dump_file directory on 2414the user's machine. It also counts the size of the vtable pointer sets 2415for each class, and writes this information to ``vtv_class_set_sizes.log'' 2416in the same directory. 2417 2418Note: This feature APPENDS data to the log files. To get a fresh log 2419files, be sure to delete any existing ones. 2420 2421@item -fno-weak 2422@opindex fno-weak 2423Do not use weak symbol support, even if it is provided by the linker. 2424By default, G++ uses weak symbols if they are available. This 2425option exists only for testing, and should not be used by end-users; 2426it results in inferior code and has no benefits. This option may 2427be removed in a future release of G++. 2428 2429@item -nostdinc++ 2430@opindex nostdinc++ 2431Do not search for header files in the standard directories specific to 2432C++, but do still search the other standard directories. (This option 2433is used when building the C++ library.) 2434@end table 2435 2436In addition, these optimization, warning, and code generation options 2437have meanings only for C++ programs: 2438 2439@table @gcctabopt 2440@item -Wabi @r{(C, Objective-C, C++ and Objective-C++ only)} 2441@opindex Wabi 2442@opindex Wno-abi 2443Warn when G++ generates code that is probably not compatible with the 2444vendor-neutral C++ ABI@. Although an effort has been made to warn about 2445all such cases, there are probably some cases that are not warned about, 2446even though G++ is generating incompatible code. There may also be 2447cases where warnings are emitted even though the code that is generated 2448is compatible. 2449 2450You should rewrite your code to avoid these warnings if you are 2451concerned about the fact that code generated by G++ may not be binary 2452compatible with code generated by other compilers. 2453 2454The known incompatibilities in @option{-fabi-version=2} (the default) include: 2455 2456@itemize @bullet 2457 2458@item 2459A template with a non-type template parameter of reference type is 2460mangled incorrectly: 2461@smallexample 2462extern int N; 2463template <int &> struct S @{@}; 2464void n (S<N>) @{2@} 2465@end smallexample 2466 2467This is fixed in @option{-fabi-version=3}. 2468 2469@item 2470SIMD vector types declared using @code{__attribute ((vector_size))} are 2471mangled in a non-standard way that does not allow for overloading of 2472functions taking vectors of different sizes. 2473 2474The mangling is changed in @option{-fabi-version=4}. 2475@end itemize 2476 2477The known incompatibilities in @option{-fabi-version=1} include: 2478 2479@itemize @bullet 2480 2481@item 2482Incorrect handling of tail-padding for bit-fields. G++ may attempt to 2483pack data into the same byte as a base class. For example: 2484 2485@smallexample 2486struct A @{ virtual void f(); int f1 : 1; @}; 2487struct B : public A @{ int f2 : 1; @}; 2488@end smallexample 2489 2490@noindent 2491In this case, G++ places @code{B::f2} into the same byte 2492as @code{A::f1}; other compilers do not. You can avoid this problem 2493by explicitly padding @code{A} so that its size is a multiple of the 2494byte size on your platform; that causes G++ and other compilers to 2495lay out @code{B} identically. 2496 2497@item 2498Incorrect handling of tail-padding for virtual bases. G++ does not use 2499tail padding when laying out virtual bases. For example: 2500 2501@smallexample 2502struct A @{ virtual void f(); char c1; @}; 2503struct B @{ B(); char c2; @}; 2504struct C : public A, public virtual B @{@}; 2505@end smallexample 2506 2507@noindent 2508In this case, G++ does not place @code{B} into the tail-padding for 2509@code{A}; other compilers do. You can avoid this problem by 2510explicitly padding @code{A} so that its size is a multiple of its 2511alignment (ignoring virtual base classes); that causes G++ and other 2512compilers to lay out @code{C} identically. 2513 2514@item 2515Incorrect handling of bit-fields with declared widths greater than that 2516of their underlying types, when the bit-fields appear in a union. For 2517example: 2518 2519@smallexample 2520union U @{ int i : 4096; @}; 2521@end smallexample 2522 2523@noindent 2524Assuming that an @code{int} does not have 4096 bits, G++ makes the 2525union too small by the number of bits in an @code{int}. 2526 2527@item 2528Empty classes can be placed at incorrect offsets. For example: 2529 2530@smallexample 2531struct A @{@}; 2532 2533struct B @{ 2534 A a; 2535 virtual void f (); 2536@}; 2537 2538struct C : public B, public A @{@}; 2539@end smallexample 2540 2541@noindent 2542G++ places the @code{A} base class of @code{C} at a nonzero offset; 2543it should be placed at offset zero. G++ mistakenly believes that the 2544@code{A} data member of @code{B} is already at offset zero. 2545 2546@item 2547Names of template functions whose types involve @code{typename} or 2548template template parameters can be mangled incorrectly. 2549 2550@smallexample 2551template <typename Q> 2552void f(typename Q::X) @{@} 2553 2554template <template <typename> class Q> 2555void f(typename Q<int>::X) @{@} 2556@end smallexample 2557 2558@noindent 2559Instantiations of these templates may be mangled incorrectly. 2560 2561@end itemize 2562 2563It also warns about psABI-related changes. The known psABI changes at this 2564point include: 2565 2566@itemize @bullet 2567 2568@item 2569For SysV/x86-64, unions with @code{long double} members are 2570passed in memory as specified in psABI. For example: 2571 2572@smallexample 2573union U @{ 2574 long double ld; 2575 int i; 2576@}; 2577@end smallexample 2578 2579@noindent 2580@code{union U} is always passed in memory. 2581 2582@end itemize 2583 2584@item -Wctor-dtor-privacy @r{(C++ and Objective-C++ only)} 2585@opindex Wctor-dtor-privacy 2586@opindex Wno-ctor-dtor-privacy 2587Warn when a class seems unusable because all the constructors or 2588destructors in that class are private, and it has neither friends nor 2589public static member functions. Also warn if there are no non-private 2590methods, and there's at least one private member function that isn't 2591a constructor or destructor. 2592 2593@item -Wdelete-non-virtual-dtor @r{(C++ and Objective-C++ only)} 2594@opindex Wdelete-non-virtual-dtor 2595@opindex Wno-delete-non-virtual-dtor 2596Warn when @samp{delete} is used to destroy an instance of a class that 2597has virtual functions and non-virtual destructor. It is unsafe to delete 2598an instance of a derived class through a pointer to a base class if the 2599base class does not have a virtual destructor. This warning is enabled 2600by @option{-Wall}. 2601 2602@item -Wliteral-suffix @r{(C++ and Objective-C++ only)} 2603@opindex Wliteral-suffix 2604@opindex Wno-literal-suffix 2605Warn when a string or character literal is followed by a ud-suffix which does 2606not begin with an underscore. As a conforming extension, GCC treats such 2607suffixes as separate preprocessing tokens in order to maintain backwards 2608compatibility with code that uses formatting macros from @code{<inttypes.h>}. 2609For example: 2610 2611@smallexample 2612#define __STDC_FORMAT_MACROS 2613#include <inttypes.h> 2614#include <stdio.h> 2615 2616int main() @{ 2617 int64_t i64 = 123; 2618 printf("My int64: %"PRId64"\n", i64); 2619@} 2620@end smallexample 2621 2622In this case, @code{PRId64} is treated as a separate preprocessing token. 2623 2624This warning is enabled by default. 2625 2626@item -Wnarrowing @r{(C++ and Objective-C++ only)} 2627@opindex Wnarrowing 2628@opindex Wno-narrowing 2629Warn when a narrowing conversion prohibited by C++11 occurs within 2630@samp{@{ @}}, e.g. 2631 2632@smallexample 2633int i = @{ 2.2 @}; // error: narrowing from double to int 2634@end smallexample 2635 2636This flag is included in @option{-Wall} and @option{-Wc++11-compat}. 2637 2638With @option{-std=c++11}, @option{-Wno-narrowing} suppresses the diagnostic 2639required by the standard. Note that this does not affect the meaning 2640of well-formed code; narrowing conversions are still considered 2641ill-formed in SFINAE context. 2642 2643@item -Wnoexcept @r{(C++ and Objective-C++ only)} 2644@opindex Wnoexcept 2645@opindex Wno-noexcept 2646Warn when a noexcept-expression evaluates to false because of a call 2647to a function that does not have a non-throwing exception 2648specification (i.e. @samp{throw()} or @samp{noexcept}) but is known by 2649the compiler to never throw an exception. 2650 2651@item -Wnon-virtual-dtor @r{(C++ and Objective-C++ only)} 2652@opindex Wnon-virtual-dtor 2653@opindex Wno-non-virtual-dtor 2654Warn when a class has virtual functions and an accessible non-virtual 2655destructor itself or in an accessible polymorphic base class, in which 2656case it is possible but unsafe to delete an instance of a derived 2657class through a pointer to the class itself or base class. This 2658warning is automatically enabled if @option{-Weffc++} is specified. 2659 2660@item -Wreorder @r{(C++ and Objective-C++ only)} 2661@opindex Wreorder 2662@opindex Wno-reorder 2663@cindex reordering, warning 2664@cindex warning for reordering of member initializers 2665Warn when the order of member initializers given in the code does not 2666match the order in which they must be executed. For instance: 2667 2668@smallexample 2669struct A @{ 2670 int i; 2671 int j; 2672 A(): j (0), i (1) @{ @} 2673@}; 2674@end smallexample 2675 2676@noindent 2677The compiler rearranges the member initializers for @samp{i} 2678and @samp{j} to match the declaration order of the members, emitting 2679a warning to that effect. This warning is enabled by @option{-Wall}. 2680 2681@item -fext-numeric-literals @r{(C++ and Objective-C++ only)} 2682@opindex fext-numeric-literals 2683@opindex fno-ext-numeric-literals 2684Accept imaginary, fixed-point, or machine-defined 2685literal number suffixes as GNU extensions. 2686When this option is turned off these suffixes are treated 2687as C++11 user-defined literal numeric suffixes. 2688This is on by default for all pre-C++11 dialects and all GNU dialects: 2689@option{-std=c++98}, @option{-std=gnu++98}, @option{-std=gnu++11}, 2690@option{-std=gnu++1y}. 2691This option is off by default 2692for ISO C++11 onwards (@option{-std=c++11}, ...). 2693@end table 2694 2695The following @option{-W@dots{}} options are not affected by @option{-Wall}. 2696 2697@table @gcctabopt 2698@item -Weffc++ @r{(C++ and Objective-C++ only)} 2699@opindex Weffc++ 2700@opindex Wno-effc++ 2701Warn about violations of the following style guidelines from Scott Meyers' 2702@cite{Effective C++} series of books: 2703 2704@itemize @bullet 2705@item 2706Define a copy constructor and an assignment operator for classes 2707with dynamically-allocated memory. 2708 2709@item 2710Prefer initialization to assignment in constructors. 2711 2712@item 2713Have @code{operator=} return a reference to @code{*this}. 2714 2715@item 2716Don't try to return a reference when you must return an object. 2717 2718@item 2719Distinguish between prefix and postfix forms of increment and 2720decrement operators. 2721 2722@item 2723Never overload @code{&&}, @code{||}, or @code{,}. 2724 2725@end itemize 2726 2727This option also enables @option{-Wnon-virtual-dtor}, which is also 2728one of the effective C++ recommendations. However, the check is 2729extended to warn about the lack of virtual destructor in accessible 2730non-polymorphic bases classes too. 2731 2732When selecting this option, be aware that the standard library 2733headers do not obey all of these guidelines; use @samp{grep -v} 2734to filter out those warnings. 2735 2736@item -Wstrict-null-sentinel @r{(C++ and Objective-C++ only)} 2737@opindex Wstrict-null-sentinel 2738@opindex Wno-strict-null-sentinel 2739Warn about the use of an uncasted @code{NULL} as sentinel. When 2740compiling only with GCC this is a valid sentinel, as @code{NULL} is defined 2741to @code{__null}. Although it is a null pointer constant rather than a 2742null pointer, it is guaranteed to be of the same size as a pointer. 2743But this use is not portable across different compilers. 2744 2745@item -Wno-non-template-friend @r{(C++ and Objective-C++ only)} 2746@opindex Wno-non-template-friend 2747@opindex Wnon-template-friend 2748Disable warnings when non-templatized friend functions are declared 2749within a template. Since the advent of explicit template specification 2750support in G++, if the name of the friend is an unqualified-id (i.e., 2751@samp{friend foo(int)}), the C++ language specification demands that the 2752friend declare or define an ordinary, nontemplate function. (Section 275314.5.3). Before G++ implemented explicit specification, unqualified-ids 2754could be interpreted as a particular specialization of a templatized 2755function. Because this non-conforming behavior is no longer the default 2756behavior for G++, @option{-Wnon-template-friend} allows the compiler to 2757check existing code for potential trouble spots and is on by default. 2758This new compiler behavior can be turned off with 2759@option{-Wno-non-template-friend}, which keeps the conformant compiler code 2760but disables the helpful warning. 2761 2762@item -Wold-style-cast @r{(C++ and Objective-C++ only)} 2763@opindex Wold-style-cast 2764@opindex Wno-old-style-cast 2765Warn if an old-style (C-style) cast to a non-void type is used within 2766a C++ program. The new-style casts (@samp{dynamic_cast}, 2767@samp{static_cast}, @samp{reinterpret_cast}, and @samp{const_cast}) are 2768less vulnerable to unintended effects and much easier to search for. 2769 2770@item -Woverloaded-virtual @r{(C++ and Objective-C++ only)} 2771@opindex Woverloaded-virtual 2772@opindex Wno-overloaded-virtual 2773@cindex overloaded virtual function, warning 2774@cindex warning for overloaded virtual function 2775Warn when a function declaration hides virtual functions from a 2776base class. For example, in: 2777 2778@smallexample 2779struct A @{ 2780 virtual void f(); 2781@}; 2782 2783struct B: public A @{ 2784 void f(int); 2785@}; 2786@end smallexample 2787 2788the @code{A} class version of @code{f} is hidden in @code{B}, and code 2789like: 2790 2791@smallexample 2792B* b; 2793b->f(); 2794@end smallexample 2795 2796@noindent 2797fails to compile. 2798 2799@item -Wno-pmf-conversions @r{(C++ and Objective-C++ only)} 2800@opindex Wno-pmf-conversions 2801@opindex Wpmf-conversions 2802Disable the diagnostic for converting a bound pointer to member function 2803to a plain pointer. 2804 2805@item -Wsign-promo @r{(C++ and Objective-C++ only)} 2806@opindex Wsign-promo 2807@opindex Wno-sign-promo 2808Warn when overload resolution chooses a promotion from unsigned or 2809enumerated type to a signed type, over a conversion to an unsigned type of 2810the same size. Previous versions of G++ tried to preserve 2811unsignedness, but the standard mandates the current behavior. 2812@end table 2813 2814@node Objective-C and Objective-C++ Dialect Options 2815@section Options Controlling Objective-C and Objective-C++ Dialects 2816 2817@cindex compiler options, Objective-C and Objective-C++ 2818@cindex Objective-C and Objective-C++ options, command-line 2819@cindex options, Objective-C and Objective-C++ 2820(NOTE: This manual does not describe the Objective-C and Objective-C++ 2821languages themselves. @xref{Standards,,Language Standards 2822Supported by GCC}, for references.) 2823 2824This section describes the command-line options that are only meaningful 2825for Objective-C and Objective-C++ programs. You can also use most of 2826the language-independent GNU compiler options. 2827For example, you might compile a file @code{some_class.m} like this: 2828 2829@smallexample 2830gcc -g -fgnu-runtime -O -c some_class.m 2831@end smallexample 2832 2833@noindent 2834In this example, @option{-fgnu-runtime} is an option meant only for 2835Objective-C and Objective-C++ programs; you can use the other options with 2836any language supported by GCC@. 2837 2838Note that since Objective-C is an extension of the C language, Objective-C 2839compilations may also use options specific to the C front-end (e.g., 2840@option{-Wtraditional}). Similarly, Objective-C++ compilations may use 2841C++-specific options (e.g., @option{-Wabi}). 2842 2843Here is a list of options that are @emph{only} for compiling Objective-C 2844and Objective-C++ programs: 2845 2846@table @gcctabopt 2847@item -fconstant-string-class=@var{class-name} 2848@opindex fconstant-string-class 2849Use @var{class-name} as the name of the class to instantiate for each 2850literal string specified with the syntax @code{@@"@dots{}"}. The default 2851class name is @code{NXConstantString} if the GNU runtime is being used, and 2852@code{NSConstantString} if the NeXT runtime is being used (see below). The 2853@option{-fconstant-cfstrings} option, if also present, overrides the 2854@option{-fconstant-string-class} setting and cause @code{@@"@dots{}"} literals 2855to be laid out as constant CoreFoundation strings. 2856 2857@item -fgnu-runtime 2858@opindex fgnu-runtime 2859Generate object code compatible with the standard GNU Objective-C 2860runtime. This is the default for most types of systems. 2861 2862@item -fnext-runtime 2863@opindex fnext-runtime 2864Generate output compatible with the NeXT runtime. This is the default 2865for NeXT-based systems, including Darwin and Mac OS X@. The macro 2866@code{__NEXT_RUNTIME__} is predefined if (and only if) this option is 2867used. 2868 2869@item -fno-nil-receivers 2870@opindex fno-nil-receivers 2871Assume that all Objective-C message dispatches (@code{[receiver 2872message:arg]}) in this translation unit ensure that the receiver is 2873not @code{nil}. This allows for more efficient entry points in the 2874runtime to be used. This option is only available in conjunction with 2875the NeXT runtime and ABI version 0 or 1. 2876 2877@item -fobjc-abi-version=@var{n} 2878@opindex fobjc-abi-version 2879Use version @var{n} of the Objective-C ABI for the selected runtime. 2880This option is currently supported only for the NeXT runtime. In that 2881case, Version 0 is the traditional (32-bit) ABI without support for 2882properties and other Objective-C 2.0 additions. Version 1 is the 2883traditional (32-bit) ABI with support for properties and other 2884Objective-C 2.0 additions. Version 2 is the modern (64-bit) ABI. If 2885nothing is specified, the default is Version 0 on 32-bit target 2886machines, and Version 2 on 64-bit target machines. 2887 2888@item -fobjc-call-cxx-cdtors 2889@opindex fobjc-call-cxx-cdtors 2890For each Objective-C class, check if any of its instance variables is a 2891C++ object with a non-trivial default constructor. If so, synthesize a 2892special @code{- (id) .cxx_construct} instance method which runs 2893non-trivial default constructors on any such instance variables, in order, 2894and then return @code{self}. Similarly, check if any instance variable 2895is a C++ object with a non-trivial destructor, and if so, synthesize a 2896special @code{- (void) .cxx_destruct} method which runs 2897all such default destructors, in reverse order. 2898 2899The @code{- (id) .cxx_construct} and @code{- (void) .cxx_destruct} 2900methods thusly generated only operate on instance variables 2901declared in the current Objective-C class, and not those inherited 2902from superclasses. It is the responsibility of the Objective-C 2903runtime to invoke all such methods in an object's inheritance 2904hierarchy. The @code{- (id) .cxx_construct} methods are invoked 2905by the runtime immediately after a new object instance is allocated; 2906the @code{- (void) .cxx_destruct} methods are invoked immediately 2907before the runtime deallocates an object instance. 2908 2909As of this writing, only the NeXT runtime on Mac OS X 10.4 and later has 2910support for invoking the @code{- (id) .cxx_construct} and 2911@code{- (void) .cxx_destruct} methods. 2912 2913@item -fobjc-direct-dispatch 2914@opindex fobjc-direct-dispatch 2915Allow fast jumps to the message dispatcher. On Darwin this is 2916accomplished via the comm page. 2917 2918@item -fobjc-exceptions 2919@opindex fobjc-exceptions 2920Enable syntactic support for structured exception handling in 2921Objective-C, similar to what is offered by C++ and Java. This option 2922is required to use the Objective-C keywords @code{@@try}, 2923@code{@@throw}, @code{@@catch}, @code{@@finally} and 2924@code{@@synchronized}. This option is available with both the GNU 2925runtime and the NeXT runtime (but not available in conjunction with 2926the NeXT runtime on Mac OS X 10.2 and earlier). 2927 2928@item -fobjc-gc 2929@opindex fobjc-gc 2930Enable garbage collection (GC) in Objective-C and Objective-C++ 2931programs. This option is only available with the NeXT runtime; the 2932GNU runtime has a different garbage collection implementation that 2933does not require special compiler flags. 2934 2935@item -fobjc-nilcheck 2936@opindex fobjc-nilcheck 2937For the NeXT runtime with version 2 of the ABI, check for a nil 2938receiver in method invocations before doing the actual method call. 2939This is the default and can be disabled using 2940@option{-fno-objc-nilcheck}. Class methods and super calls are never 2941checked for nil in this way no matter what this flag is set to. 2942Currently this flag does nothing when the GNU runtime, or an older 2943version of the NeXT runtime ABI, is used. 2944 2945@item -fobjc-std=objc1 2946@opindex fobjc-std 2947Conform to the language syntax of Objective-C 1.0, the language 2948recognized by GCC 4.0. This only affects the Objective-C additions to 2949the C/C++ language; it does not affect conformance to C/C++ standards, 2950which is controlled by the separate C/C++ dialect option flags. When 2951this option is used with the Objective-C or Objective-C++ compiler, 2952any Objective-C syntax that is not recognized by GCC 4.0 is rejected. 2953This is useful if you need to make sure that your Objective-C code can 2954be compiled with older versions of GCC@. 2955 2956@item -freplace-objc-classes 2957@opindex freplace-objc-classes 2958Emit a special marker instructing @command{ld(1)} not to statically link in 2959the resulting object file, and allow @command{dyld(1)} to load it in at 2960run time instead. This is used in conjunction with the Fix-and-Continue 2961debugging mode, where the object file in question may be recompiled and 2962dynamically reloaded in the course of program execution, without the need 2963to restart the program itself. Currently, Fix-and-Continue functionality 2964is only available in conjunction with the NeXT runtime on Mac OS X 10.3 2965and later. 2966 2967@item -fzero-link 2968@opindex fzero-link 2969When compiling for the NeXT runtime, the compiler ordinarily replaces calls 2970to @code{objc_getClass("@dots{}")} (when the name of the class is known at 2971compile time) with static class references that get initialized at load time, 2972which improves run-time performance. Specifying the @option{-fzero-link} flag 2973suppresses this behavior and causes calls to @code{objc_getClass("@dots{}")} 2974to be retained. This is useful in Zero-Link debugging mode, since it allows 2975for individual class implementations to be modified during program execution. 2976The GNU runtime currently always retains calls to @code{objc_get_class("@dots{}")} 2977regardless of command-line options. 2978 2979@item -gen-decls 2980@opindex gen-decls 2981Dump interface declarations for all classes seen in the source file to a 2982file named @file{@var{sourcename}.decl}. 2983 2984@item -Wassign-intercept @r{(Objective-C and Objective-C++ only)} 2985@opindex Wassign-intercept 2986@opindex Wno-assign-intercept 2987Warn whenever an Objective-C assignment is being intercepted by the 2988garbage collector. 2989 2990@item -Wno-protocol @r{(Objective-C and Objective-C++ only)} 2991@opindex Wno-protocol 2992@opindex Wprotocol 2993If a class is declared to implement a protocol, a warning is issued for 2994every method in the protocol that is not implemented by the class. The 2995default behavior is to issue a warning for every method not explicitly 2996implemented in the class, even if a method implementation is inherited 2997from the superclass. If you use the @option{-Wno-protocol} option, then 2998methods inherited from the superclass are considered to be implemented, 2999and no warning is issued for them. 3000 3001@item -Wselector @r{(Objective-C and Objective-C++ only)} 3002@opindex Wselector 3003@opindex Wno-selector 3004Warn if multiple methods of different types for the same selector are 3005found during compilation. The check is performed on the list of methods 3006in the final stage of compilation. Additionally, a check is performed 3007for each selector appearing in a @code{@@selector(@dots{})} 3008expression, and a corresponding method for that selector has been found 3009during compilation. Because these checks scan the method table only at 3010the end of compilation, these warnings are not produced if the final 3011stage of compilation is not reached, for example because an error is 3012found during compilation, or because the @option{-fsyntax-only} option is 3013being used. 3014 3015@item -Wstrict-selector-match @r{(Objective-C and Objective-C++ only)} 3016@opindex Wstrict-selector-match 3017@opindex Wno-strict-selector-match 3018Warn if multiple methods with differing argument and/or return types are 3019found for a given selector when attempting to send a message using this 3020selector to a receiver of type @code{id} or @code{Class}. When this flag 3021is off (which is the default behavior), the compiler omits such warnings 3022if any differences found are confined to types that share the same size 3023and alignment. 3024 3025@item -Wundeclared-selector @r{(Objective-C and Objective-C++ only)} 3026@opindex Wundeclared-selector 3027@opindex Wno-undeclared-selector 3028Warn if a @code{@@selector(@dots{})} expression referring to an 3029undeclared selector is found. A selector is considered undeclared if no 3030method with that name has been declared before the 3031@code{@@selector(@dots{})} expression, either explicitly in an 3032@code{@@interface} or @code{@@protocol} declaration, or implicitly in 3033an @code{@@implementation} section. This option always performs its 3034checks as soon as a @code{@@selector(@dots{})} expression is found, 3035while @option{-Wselector} only performs its checks in the final stage of 3036compilation. This also enforces the coding style convention 3037that methods and selectors must be declared before being used. 3038 3039@item -print-objc-runtime-info 3040@opindex print-objc-runtime-info 3041Generate C header describing the largest structure that is passed by 3042value, if any. 3043 3044@end table 3045 3046@node Language Independent Options 3047@section Options to Control Diagnostic Messages Formatting 3048@cindex options to control diagnostics formatting 3049@cindex diagnostic messages 3050@cindex message formatting 3051 3052Traditionally, diagnostic messages have been formatted irrespective of 3053the output device's aspect (e.g.@: its width, @dots{}). You can use the 3054options described below 3055to control the formatting algorithm for diagnostic messages, 3056e.g.@: how many characters per line, how often source location 3057information should be reported. Note that some language front ends may not 3058honor these options. 3059 3060@table @gcctabopt 3061@item -fmessage-length=@var{n} 3062@opindex fmessage-length 3063Try to format error messages so that they fit on lines of about @var{n} 3064characters. The default is 72 characters for @command{g++} and 0 for the rest of 3065the front ends supported by GCC@. If @var{n} is zero, then no 3066line-wrapping is done; each error message appears on a single 3067line. 3068 3069@item -fdiagnostics-show-location=once 3070@opindex fdiagnostics-show-location 3071Only meaningful in line-wrapping mode. Instructs the diagnostic messages 3072reporter to emit source location information @emph{once}; that is, in 3073case the message is too long to fit on a single physical line and has to 3074be wrapped, the source location won't be emitted (as prefix) again, 3075over and over, in subsequent continuation lines. This is the default 3076behavior. 3077 3078@item -fdiagnostics-show-location=every-line 3079Only meaningful in line-wrapping mode. Instructs the diagnostic 3080messages reporter to emit the same source location information (as 3081prefix) for physical lines that result from the process of breaking 3082a message which is too long to fit on a single line. 3083 3084@item -fdiagnostics-color[=@var{WHEN}] 3085@itemx -fno-diagnostics-color 3086@opindex fdiagnostics-color 3087@cindex highlight, color, colour 3088@vindex GCC_COLORS @r{environment variable} 3089Use color in diagnostics. @var{WHEN} is @samp{never}, @samp{always}, 3090or @samp{auto}. The default is @samp{never} if @env{GCC_COLORS} environment 3091variable isn't present in the environment, and @samp{auto} otherwise. 3092@samp{auto} means to use color only when the standard error is a terminal. 3093The forms @option{-fdiagnostics-color} and @option{-fno-diagnostics-color} are 3094aliases for @option{-fdiagnostics-color=always} and 3095@option{-fdiagnostics-color=never}, respectively. 3096 3097The colors are defined by the environment variable @env{GCC_COLORS}. 3098Its value is a colon-separated list of capabilities and Select Graphic 3099Rendition (SGR) substrings. SGR commands are interpreted by the 3100terminal or terminal emulator. (See the section in the documentation 3101of your text terminal for permitted values and their meanings as 3102character attributes.) These substring values are integers in decimal 3103representation and can be concatenated with semicolons. 3104Common values to concatenate include 3105@samp{1} for bold, 3106@samp{4} for underline, 3107@samp{5} for blink, 3108@samp{7} for inverse, 3109@samp{39} for default foreground color, 3110@samp{30} to @samp{37} for foreground colors, 3111@samp{90} to @samp{97} for 16-color mode foreground colors, 3112@samp{38;5;0} to @samp{38;5;255} 3113for 88-color and 256-color modes foreground colors, 3114@samp{49} for default background color, 3115@samp{40} to @samp{47} for background colors, 3116@samp{100} to @samp{107} for 16-color mode background colors, 3117and @samp{48;5;0} to @samp{48;5;255} 3118for 88-color and 256-color modes background colors. 3119 3120The default @env{GCC_COLORS} is 3121@samp{error=01;31:warning=01;35:note=01;36:caret=01;32:locus=01:quote=01} 3122where @samp{01;31} is bold red, @samp{01;35} is bold magenta, 3123@samp{01;36} is bold cyan, @samp{01;32} is bold green and 3124@samp{01} is bold. Setting @env{GCC_COLORS} to the empty 3125string disables colors. 3126Supported capabilities are as follows. 3127 3128@table @code 3129@item error= 3130@vindex error GCC_COLORS @r{capability} 3131SGR substring for error: markers. 3132 3133@item warning= 3134@vindex warning GCC_COLORS @r{capability} 3135SGR substring for warning: markers. 3136 3137@item note= 3138@vindex note GCC_COLORS @r{capability} 3139SGR substring for note: markers. 3140 3141@item caret= 3142@vindex caret GCC_COLORS @r{capability} 3143SGR substring for caret line. 3144 3145@item locus= 3146@vindex locus GCC_COLORS @r{capability} 3147SGR substring for location information, @samp{file:line} or 3148@samp{file:line:column} etc. 3149 3150@item quote= 3151@vindex quote GCC_COLORS @r{capability} 3152SGR substring for information printed within quotes. 3153@end table 3154 3155@item -fno-diagnostics-show-option 3156@opindex fno-diagnostics-show-option 3157@opindex fdiagnostics-show-option 3158By default, each diagnostic emitted includes text indicating the 3159command-line option that directly controls the diagnostic (if such an 3160option is known to the diagnostic machinery). Specifying the 3161@option{-fno-diagnostics-show-option} flag suppresses that behavior. 3162 3163@item -fno-diagnostics-show-caret 3164@opindex fno-diagnostics-show-caret 3165@opindex fdiagnostics-show-caret 3166By default, each diagnostic emitted includes the original source line 3167and a caret '^' indicating the column. This option suppresses this 3168information. 3169 3170@end table 3171 3172@node Warning Options 3173@section Options to Request or Suppress Warnings 3174@cindex options to control warnings 3175@cindex warning messages 3176@cindex messages, warning 3177@cindex suppressing warnings 3178 3179Warnings are diagnostic messages that report constructions that 3180are not inherently erroneous but that are risky or suggest there 3181may have been an error. 3182 3183The following language-independent options do not enable specific 3184warnings but control the kinds of diagnostics produced by GCC@. 3185 3186@table @gcctabopt 3187@cindex syntax checking 3188@item -fsyntax-only 3189@opindex fsyntax-only 3190Check the code for syntax errors, but don't do anything beyond that. 3191 3192@item -fmax-errors=@var{n} 3193@opindex fmax-errors 3194Limits the maximum number of error messages to @var{n}, at which point 3195GCC bails out rather than attempting to continue processing the source 3196code. If @var{n} is 0 (the default), there is no limit on the number 3197of error messages produced. If @option{-Wfatal-errors} is also 3198specified, then @option{-Wfatal-errors} takes precedence over this 3199option. 3200 3201@item -w 3202@opindex w 3203Inhibit all warning messages. 3204 3205@item -Werror 3206@opindex Werror 3207@opindex Wno-error 3208Make all warnings into errors. 3209 3210@item -Werror= 3211@opindex Werror= 3212@opindex Wno-error= 3213Make the specified warning into an error. The specifier for a warning 3214is appended; for example @option{-Werror=switch} turns the warnings 3215controlled by @option{-Wswitch} into errors. This switch takes a 3216negative form, to be used to negate @option{-Werror} for specific 3217warnings; for example @option{-Wno-error=switch} makes 3218@option{-Wswitch} warnings not be errors, even when @option{-Werror} 3219is in effect. 3220 3221The warning message for each controllable warning includes the 3222option that controls the warning. That option can then be used with 3223@option{-Werror=} and @option{-Wno-error=} as described above. 3224(Printing of the option in the warning message can be disabled using the 3225@option{-fno-diagnostics-show-option} flag.) 3226 3227Note that specifying @option{-Werror=}@var{foo} automatically implies 3228@option{-W}@var{foo}. However, @option{-Wno-error=}@var{foo} does not 3229imply anything. 3230 3231@item -Wfatal-errors 3232@opindex Wfatal-errors 3233@opindex Wno-fatal-errors 3234This option causes the compiler to abort compilation on the first error 3235occurred rather than trying to keep going and printing further error 3236messages. 3237 3238@end table 3239 3240You can request many specific warnings with options beginning with 3241@samp{-W}, for example @option{-Wimplicit} to request warnings on 3242implicit declarations. Each of these specific warning options also 3243has a negative form beginning @samp{-Wno-} to turn off warnings; for 3244example, @option{-Wno-implicit}. This manual lists only one of the 3245two forms, whichever is not the default. For further 3246language-specific options also refer to @ref{C++ Dialect Options} and 3247@ref{Objective-C and Objective-C++ Dialect Options}. 3248 3249When an unrecognized warning option is requested (e.g., 3250@option{-Wunknown-warning}), GCC emits a diagnostic stating 3251that the option is not recognized. However, if the @option{-Wno-} form 3252is used, the behavior is slightly different: no diagnostic is 3253produced for @option{-Wno-unknown-warning} unless other diagnostics 3254are being produced. This allows the use of new @option{-Wno-} options 3255with old compilers, but if something goes wrong, the compiler 3256warns that an unrecognized option is present. 3257 3258@table @gcctabopt 3259@item -Wpedantic 3260@itemx -pedantic 3261@opindex pedantic 3262@opindex Wpedantic 3263Issue all the warnings demanded by strict ISO C and ISO C++; 3264reject all programs that use forbidden extensions, and some other 3265programs that do not follow ISO C and ISO C++. For ISO C, follows the 3266version of the ISO C standard specified by any @option{-std} option used. 3267 3268Valid ISO C and ISO C++ programs should compile properly with or without 3269this option (though a rare few require @option{-ansi} or a 3270@option{-std} option specifying the required version of ISO C)@. However, 3271without this option, certain GNU extensions and traditional C and C++ 3272features are supported as well. With this option, they are rejected. 3273 3274@option{-Wpedantic} does not cause warning messages for use of the 3275alternate keywords whose names begin and end with @samp{__}. Pedantic 3276warnings are also disabled in the expression that follows 3277@code{__extension__}. However, only system header files should use 3278these escape routes; application programs should avoid them. 3279@xref{Alternate Keywords}. 3280 3281Some users try to use @option{-Wpedantic} to check programs for strict ISO 3282C conformance. They soon find that it does not do quite what they want: 3283it finds some non-ISO practices, but not all---only those for which 3284ISO C @emph{requires} a diagnostic, and some others for which 3285diagnostics have been added. 3286 3287A feature to report any failure to conform to ISO C might be useful in 3288some instances, but would require considerable additional work and would 3289be quite different from @option{-Wpedantic}. We don't have plans to 3290support such a feature in the near future. 3291 3292Where the standard specified with @option{-std} represents a GNU 3293extended dialect of C, such as @samp{gnu90} or @samp{gnu99}, there is a 3294corresponding @dfn{base standard}, the version of ISO C on which the GNU 3295extended dialect is based. Warnings from @option{-Wpedantic} are given 3296where they are required by the base standard. (It does not make sense 3297for such warnings to be given only for features not in the specified GNU 3298C dialect, since by definition the GNU dialects of C include all 3299features the compiler supports with the given option, and there would be 3300nothing to warn about.) 3301 3302@item -pedantic-errors 3303@opindex pedantic-errors 3304Like @option{-Wpedantic}, except that errors are produced rather than 3305warnings. 3306 3307@item -Wall 3308@opindex Wall 3309@opindex Wno-all 3310This enables all the warnings about constructions that some users 3311consider questionable, and that are easy to avoid (or modify to 3312prevent the warning), even in conjunction with macros. This also 3313enables some language-specific warnings described in @ref{C++ Dialect 3314Options} and @ref{Objective-C and Objective-C++ Dialect Options}. 3315 3316@option{-Wall} turns on the following warning flags: 3317 3318@gccoptlist{-Waddress @gol 3319-Warray-bounds @r{(only with} @option{-O2}@r{)} @gol 3320-Wc++11-compat @gol 3321-Wchar-subscripts @gol 3322-Wenum-compare @r{(in C/ObjC; this is on by default in C++)} @gol 3323-Wimplicit-int @r{(C and Objective-C only)} @gol 3324-Wimplicit-function-declaration @r{(C and Objective-C only)} @gol 3325-Wcomment @gol 3326-Wformat @gol 3327-Wmain @r{(only for C/ObjC and unless} @option{-ffreestanding}@r{)} @gol 3328-Wmaybe-uninitialized @gol 3329-Wmissing-braces @r{(only for C/ObjC)} @gol 3330-Wnonnull @gol 3331-Wopenmp-simd @gol 3332-Wparentheses @gol 3333-Wpointer-sign @gol 3334-Wreorder @gol 3335-Wreturn-type @gol 3336-Wsequence-point @gol 3337-Wsign-compare @r{(only in C++)} @gol 3338-Wstrict-aliasing @gol 3339-Wstrict-overflow=1 @gol 3340-Wswitch @gol 3341-Wtrigraphs @gol 3342-Wuninitialized @gol 3343-Wunknown-pragmas @gol 3344-Wunused-function @gol 3345-Wunused-label @gol 3346-Wunused-value @gol 3347-Wunused-variable @gol 3348-Wvolatile-register-var @gol 3349} 3350 3351Note that some warning flags are not implied by @option{-Wall}. Some of 3352them warn about constructions that users generally do not consider 3353questionable, but which occasionally you might wish to check for; 3354others warn about constructions that are necessary or hard to avoid in 3355some cases, and there is no simple way to modify the code to suppress 3356the warning. Some of them are enabled by @option{-Wextra} but many of 3357them must be enabled individually. 3358 3359@item -Wextra 3360@opindex W 3361@opindex Wextra 3362@opindex Wno-extra 3363This enables some extra warning flags that are not enabled by 3364@option{-Wall}. (This option used to be called @option{-W}. The older 3365name is still supported, but the newer name is more descriptive.) 3366 3367@gccoptlist{-Wclobbered @gol 3368-Wempty-body @gol 3369-Wignored-qualifiers @gol 3370-Wmissing-field-initializers @gol 3371-Wmissing-parameter-type @r{(C only)} @gol 3372-Wold-style-declaration @r{(C only)} @gol 3373-Woverride-init @gol 3374-Wsign-compare @gol 3375-Wtype-limits @gol 3376-Wuninitialized @gol 3377-Wunused-parameter @r{(only with} @option{-Wunused} @r{or} @option{-Wall}@r{)} @gol 3378-Wunused-but-set-parameter @r{(only with} @option{-Wunused} @r{or} @option{-Wall}@r{)} @gol 3379} 3380 3381The option @option{-Wextra} also prints warning messages for the 3382following cases: 3383 3384@itemize @bullet 3385 3386@item 3387A pointer is compared against integer zero with @samp{<}, @samp{<=}, 3388@samp{>}, or @samp{>=}. 3389 3390@item 3391(C++ only) An enumerator and a non-enumerator both appear in a 3392conditional expression. 3393 3394@item 3395(C++ only) Ambiguous virtual bases. 3396 3397@item 3398(C++ only) Subscripting an array that has been declared @samp{register}. 3399 3400@item 3401(C++ only) Taking the address of a variable that has been declared 3402@samp{register}. 3403 3404@item 3405(C++ only) A base class is not initialized in a derived class's copy 3406constructor. 3407 3408@end itemize 3409 3410@item -Wchar-subscripts 3411@opindex Wchar-subscripts 3412@opindex Wno-char-subscripts 3413Warn if an array subscript has type @code{char}. This is a common cause 3414of error, as programmers often forget that this type is signed on some 3415machines. 3416This warning is enabled by @option{-Wall}. 3417 3418@item -Wcomment 3419@opindex Wcomment 3420@opindex Wno-comment 3421Warn whenever a comment-start sequence @samp{/*} appears in a @samp{/*} 3422comment, or whenever a Backslash-Newline appears in a @samp{//} comment. 3423This warning is enabled by @option{-Wall}. 3424 3425@item -Wno-coverage-mismatch 3426@opindex Wno-coverage-mismatch 3427Warn if feedback profiles do not match when using the 3428@option{-fprofile-use} option. 3429If a source file is changed between compiling with @option{-fprofile-gen} and 3430with @option{-fprofile-use}, the files with the profile feedback can fail 3431to match the source file and GCC cannot use the profile feedback 3432information. By default, this warning is enabled and is treated as an 3433error. @option{-Wno-coverage-mismatch} can be used to disable the 3434warning or @option{-Wno-error=coverage-mismatch} can be used to 3435disable the error. Disabling the error for this warning can result in 3436poorly optimized code and is useful only in the 3437case of very minor changes such as bug fixes to an existing code-base. 3438Completely disabling the warning is not recommended. 3439 3440@item -Wno-cpp 3441@r{(C, Objective-C, C++, Objective-C++ and Fortran only)} 3442 3443Suppress warning messages emitted by @code{#warning} directives. 3444 3445@item -Wdouble-promotion @r{(C, C++, Objective-C and Objective-C++ only)} 3446@opindex Wdouble-promotion 3447@opindex Wno-double-promotion 3448Give a warning when a value of type @code{float} is implicitly 3449promoted to @code{double}. CPUs with a 32-bit ``single-precision'' 3450floating-point unit implement @code{float} in hardware, but emulate 3451@code{double} in software. On such a machine, doing computations 3452using @code{double} values is much more expensive because of the 3453overhead required for software emulation. 3454 3455It is easy to accidentally do computations with @code{double} because 3456floating-point literals are implicitly of type @code{double}. For 3457example, in: 3458@smallexample 3459@group 3460float area(float radius) 3461@{ 3462 return 3.14159 * radius * radius; 3463@} 3464@end group 3465@end smallexample 3466the compiler performs the entire computation with @code{double} 3467because the floating-point literal is a @code{double}. 3468 3469@item -Wformat 3470@itemx -Wformat=@var{n} 3471@opindex Wformat 3472@opindex Wno-format 3473@opindex ffreestanding 3474@opindex fno-builtin 3475@opindex Wformat= 3476Check calls to @code{printf} and @code{scanf}, etc., to make sure that 3477the arguments supplied have types appropriate to the format string 3478specified, and that the conversions specified in the format string make 3479sense. This includes standard functions, and others specified by format 3480attributes (@pxref{Function Attributes}), in the @code{printf}, 3481@code{scanf}, @code{strftime} and @code{strfmon} (an X/Open extension, 3482not in the C standard) families (or other target-specific families). 3483Which functions are checked without format attributes having been 3484specified depends on the standard version selected, and such checks of 3485functions without the attribute specified are disabled by 3486@option{-ffreestanding} or @option{-fno-builtin}. 3487 3488The formats are checked against the format features supported by GNU 3489libc version 2.2. These include all ISO C90 and C99 features, as well 3490as features from the Single Unix Specification and some BSD and GNU 3491extensions. Other library implementations may not support all these 3492features; GCC does not support warning about features that go beyond a 3493particular library's limitations. However, if @option{-Wpedantic} is used 3494with @option{-Wformat}, warnings are given about format features not 3495in the selected standard version (but not for @code{strfmon} formats, 3496since those are not in any version of the C standard). @xref{C Dialect 3497Options,,Options Controlling C Dialect}. 3498 3499@table @gcctabopt 3500@item -Wformat=1 3501@itemx -Wformat 3502@opindex Wformat 3503@opindex Wformat=1 3504Option @option{-Wformat} is equivalent to @option{-Wformat=1}, and 3505@option{-Wno-format} is equivalent to @option{-Wformat=0}. Since 3506@option{-Wformat} also checks for null format arguments for several 3507functions, @option{-Wformat} also implies @option{-Wnonnull}. Some 3508aspects of this level of format checking can be disabled by the 3509options: @option{-Wno-format-contains-nul}, 3510@option{-Wno-format-extra-args}, and @option{-Wno-format-zero-length}. 3511@option{-Wformat} is enabled by @option{-Wall}. 3512 3513@item -Wno-format-contains-nul 3514@opindex Wno-format-contains-nul 3515@opindex Wformat-contains-nul 3516If @option{-Wformat} is specified, do not warn about format strings that 3517contain NUL bytes. 3518 3519@item -Wno-format-extra-args 3520@opindex Wno-format-extra-args 3521@opindex Wformat-extra-args 3522If @option{-Wformat} is specified, do not warn about excess arguments to a 3523@code{printf} or @code{scanf} format function. The C standard specifies 3524that such arguments are ignored. 3525 3526Where the unused arguments lie between used arguments that are 3527specified with @samp{$} operand number specifications, normally 3528warnings are still given, since the implementation could not know what 3529type to pass to @code{va_arg} to skip the unused arguments. However, 3530in the case of @code{scanf} formats, this option suppresses the 3531warning if the unused arguments are all pointers, since the Single 3532Unix Specification says that such unused arguments are allowed. 3533 3534@item -Wno-format-zero-length 3535@opindex Wno-format-zero-length 3536@opindex Wformat-zero-length 3537If @option{-Wformat} is specified, do not warn about zero-length formats. 3538The C standard specifies that zero-length formats are allowed. 3539 3540 3541@item -Wformat=2 3542@opindex Wformat=2 3543Enable @option{-Wformat} plus additional format checks. Currently 3544equivalent to @option{-Wformat -Wformat-nonliteral -Wformat-security 3545-Wformat-y2k}. 3546 3547@item -Wformat-nonliteral 3548@opindex Wformat-nonliteral 3549@opindex Wno-format-nonliteral 3550If @option{-Wformat} is specified, also warn if the format string is not a 3551string literal and so cannot be checked, unless the format function 3552takes its format arguments as a @code{va_list}. 3553 3554@item -Wformat-security 3555@opindex Wformat-security 3556@opindex Wno-format-security 3557If @option{-Wformat} is specified, also warn about uses of format 3558functions that represent possible security problems. At present, this 3559warns about calls to @code{printf} and @code{scanf} functions where the 3560format string is not a string literal and there are no format arguments, 3561as in @code{printf (foo);}. This may be a security hole if the format 3562string came from untrusted input and contains @samp{%n}. (This is 3563currently a subset of what @option{-Wformat-nonliteral} warns about, but 3564in future warnings may be added to @option{-Wformat-security} that are not 3565included in @option{-Wformat-nonliteral}.) 3566 3567@item -Wformat-y2k 3568@opindex Wformat-y2k 3569@opindex Wno-format-y2k 3570If @option{-Wformat} is specified, also warn about @code{strftime} 3571formats that may yield only a two-digit year. 3572@end table 3573 3574@item -Wnonnull 3575@opindex Wnonnull 3576@opindex Wno-nonnull 3577Warn about passing a null pointer for arguments marked as 3578requiring a non-null value by the @code{nonnull} function attribute. 3579 3580@option{-Wnonnull} is included in @option{-Wall} and @option{-Wformat}. It 3581can be disabled with the @option{-Wno-nonnull} option. 3582 3583@item -Winit-self @r{(C, C++, Objective-C and Objective-C++ only)} 3584@opindex Winit-self 3585@opindex Wno-init-self 3586Warn about uninitialized variables that are initialized with themselves. 3587Note this option can only be used with the @option{-Wuninitialized} option. 3588 3589For example, GCC warns about @code{i} being uninitialized in the 3590following snippet only when @option{-Winit-self} has been specified: 3591@smallexample 3592@group 3593int f() 3594@{ 3595 int i = i; 3596 return i; 3597@} 3598@end group 3599@end smallexample 3600 3601This warning is enabled by @option{-Wall} in C++. 3602 3603@item -Wimplicit-int @r{(C and Objective-C only)} 3604@opindex Wimplicit-int 3605@opindex Wno-implicit-int 3606Warn when a declaration does not specify a type. 3607This warning is enabled by @option{-Wall}. 3608 3609@item -Wimplicit-function-declaration @r{(C and Objective-C only)} 3610@opindex Wimplicit-function-declaration 3611@opindex Wno-implicit-function-declaration 3612Give a warning whenever a function is used before being declared. In 3613C99 mode (@option{-std=c99} or @option{-std=gnu99}), this warning is 3614enabled by default and it is made into an error by 3615@option{-pedantic-errors}. This warning is also enabled by 3616@option{-Wall}. 3617 3618@item -Wimplicit @r{(C and Objective-C only)} 3619@opindex Wimplicit 3620@opindex Wno-implicit 3621Same as @option{-Wimplicit-int} and @option{-Wimplicit-function-declaration}. 3622This warning is enabled by @option{-Wall}. 3623 3624@item -Wignored-qualifiers @r{(C and C++ only)} 3625@opindex Wignored-qualifiers 3626@opindex Wno-ignored-qualifiers 3627Warn if the return type of a function has a type qualifier 3628such as @code{const}. For ISO C such a type qualifier has no effect, 3629since the value returned by a function is not an lvalue. 3630For C++, the warning is only emitted for scalar types or @code{void}. 3631ISO C prohibits qualified @code{void} return types on function 3632definitions, so such return types always receive a warning 3633even without this option. 3634 3635This warning is also enabled by @option{-Wextra}. 3636 3637@item -Wmain 3638@opindex Wmain 3639@opindex Wno-main 3640Warn if the type of @samp{main} is suspicious. @samp{main} should be 3641a function with external linkage, returning int, taking either zero 3642arguments, two, or three arguments of appropriate types. This warning 3643is enabled by default in C++ and is enabled by either @option{-Wall} 3644or @option{-Wpedantic}. 3645 3646@item -Wmissing-braces 3647@opindex Wmissing-braces 3648@opindex Wno-missing-braces 3649Warn if an aggregate or union initializer is not fully bracketed. In 3650the following example, the initializer for @samp{a} is not fully 3651bracketed, but that for @samp{b} is fully bracketed. This warning is 3652enabled by @option{-Wall} in C. 3653 3654@smallexample 3655int a[2][2] = @{ 0, 1, 2, 3 @}; 3656int b[2][2] = @{ @{ 0, 1 @}, @{ 2, 3 @} @}; 3657@end smallexample 3658 3659This warning is enabled by @option{-Wall}. 3660 3661@item -Wmissing-include-dirs @r{(C, C++, Objective-C and Objective-C++ only)} 3662@opindex Wmissing-include-dirs 3663@opindex Wno-missing-include-dirs 3664Warn if a user-supplied include directory does not exist. 3665 3666@item -Wparentheses 3667@opindex Wparentheses 3668@opindex Wno-parentheses 3669Warn if parentheses are omitted in certain contexts, such 3670as when there is an assignment in a context where a truth value 3671is expected, or when operators are nested whose precedence people 3672often get confused about. 3673 3674Also warn if a comparison like @samp{x<=y<=z} appears; this is 3675equivalent to @samp{(x<=y ? 1 : 0) <= z}, which is a different 3676interpretation from that of ordinary mathematical notation. 3677 3678Also warn about constructions where there may be confusion to which 3679@code{if} statement an @code{else} branch belongs. Here is an example of 3680such a case: 3681 3682@smallexample 3683@group 3684@{ 3685 if (a) 3686 if (b) 3687 foo (); 3688 else 3689 bar (); 3690@} 3691@end group 3692@end smallexample 3693 3694In C/C++, every @code{else} branch belongs to the innermost possible 3695@code{if} statement, which in this example is @code{if (b)}. This is 3696often not what the programmer expected, as illustrated in the above 3697example by indentation the programmer chose. When there is the 3698potential for this confusion, GCC issues a warning when this flag 3699is specified. To eliminate the warning, add explicit braces around 3700the innermost @code{if} statement so there is no way the @code{else} 3701can belong to the enclosing @code{if}. The resulting code 3702looks like this: 3703 3704@smallexample 3705@group 3706@{ 3707 if (a) 3708 @{ 3709 if (b) 3710 foo (); 3711 else 3712 bar (); 3713 @} 3714@} 3715@end group 3716@end smallexample 3717 3718Also warn for dangerous uses of the GNU extension to 3719@code{?:} with omitted middle operand. When the condition 3720in the @code{?}: operator is a boolean expression, the omitted value is 3721always 1. Often programmers expect it to be a value computed 3722inside the conditional expression instead. 3723 3724This warning is enabled by @option{-Wall}. 3725 3726@item -Wsequence-point 3727@opindex Wsequence-point 3728@opindex Wno-sequence-point 3729Warn about code that may have undefined semantics because of violations 3730of sequence point rules in the C and C++ standards. 3731 3732The C and C++ standards define the order in which expressions in a C/C++ 3733program are evaluated in terms of @dfn{sequence points}, which represent 3734a partial ordering between the execution of parts of the program: those 3735executed before the sequence point, and those executed after it. These 3736occur after the evaluation of a full expression (one which is not part 3737of a larger expression), after the evaluation of the first operand of a 3738@code{&&}, @code{||}, @code{? :} or @code{,} (comma) operator, before a 3739function is called (but after the evaluation of its arguments and the 3740expression denoting the called function), and in certain other places. 3741Other than as expressed by the sequence point rules, the order of 3742evaluation of subexpressions of an expression is not specified. All 3743these rules describe only a partial order rather than a total order, 3744since, for example, if two functions are called within one expression 3745with no sequence point between them, the order in which the functions 3746are called is not specified. However, the standards committee have 3747ruled that function calls do not overlap. 3748 3749It is not specified when between sequence points modifications to the 3750values of objects take effect. Programs whose behavior depends on this 3751have undefined behavior; the C and C++ standards specify that ``Between 3752the previous and next sequence point an object shall have its stored 3753value modified at most once by the evaluation of an expression. 3754Furthermore, the prior value shall be read only to determine the value 3755to be stored.''. If a program breaks these rules, the results on any 3756particular implementation are entirely unpredictable. 3757 3758Examples of code with undefined behavior are @code{a = a++;}, @code{a[n] 3759= b[n++]} and @code{a[i++] = i;}. Some more complicated cases are not 3760diagnosed by this option, and it may give an occasional false positive 3761result, but in general it has been found fairly effective at detecting 3762this sort of problem in programs. 3763 3764The standard is worded confusingly, therefore there is some debate 3765over the precise meaning of the sequence point rules in subtle cases. 3766Links to discussions of the problem, including proposed formal 3767definitions, may be found on the GCC readings page, at 3768@uref{http://gcc.gnu.org/@/readings.html}. 3769 3770This warning is enabled by @option{-Wall} for C and C++. 3771 3772@item -Wno-return-local-addr 3773@opindex Wno-return-local-addr 3774@opindex Wreturn-local-addr 3775Do not warn about returning a pointer (or in C++, a reference) to a 3776variable that goes out of scope after the function returns. 3777 3778@item -Wreturn-type 3779@opindex Wreturn-type 3780@opindex Wno-return-type 3781Warn whenever a function is defined with a return type that defaults 3782to @code{int}. Also warn about any @code{return} statement with no 3783return value in a function whose return type is not @code{void} 3784(falling off the end of the function body is considered returning 3785without a value), and about a @code{return} statement with an 3786expression in a function whose return type is @code{void}. 3787 3788For C++, a function without return type always produces a diagnostic 3789message, even when @option{-Wno-return-type} is specified. The only 3790exceptions are @samp{main} and functions defined in system headers. 3791 3792This warning is enabled by @option{-Wall}. 3793 3794@item -Wswitch 3795@opindex Wswitch 3796@opindex Wno-switch 3797Warn whenever a @code{switch} statement has an index of enumerated type 3798and lacks a @code{case} for one or more of the named codes of that 3799enumeration. (The presence of a @code{default} label prevents this 3800warning.) @code{case} labels outside the enumeration range also 3801provoke warnings when this option is used (even if there is a 3802@code{default} label). 3803This warning is enabled by @option{-Wall}. 3804 3805@item -Wswitch-default 3806@opindex Wswitch-default 3807@opindex Wno-switch-default 3808Warn whenever a @code{switch} statement does not have a @code{default} 3809case. 3810 3811@item -Wswitch-enum 3812@opindex Wswitch-enum 3813@opindex Wno-switch-enum 3814Warn whenever a @code{switch} statement has an index of enumerated type 3815and lacks a @code{case} for one or more of the named codes of that 3816enumeration. @code{case} labels outside the enumeration range also 3817provoke warnings when this option is used. The only difference 3818between @option{-Wswitch} and this option is that this option gives a 3819warning about an omitted enumeration code even if there is a 3820@code{default} label. 3821 3822@item -Wsync-nand @r{(C and C++ only)} 3823@opindex Wsync-nand 3824@opindex Wno-sync-nand 3825Warn when @code{__sync_fetch_and_nand} and @code{__sync_nand_and_fetch} 3826built-in functions are used. These functions changed semantics in GCC 4.4. 3827 3828@item -Wtrigraphs 3829@opindex Wtrigraphs 3830@opindex Wno-trigraphs 3831Warn if any trigraphs are encountered that might change the meaning of 3832the program (trigraphs within comments are not warned about). 3833This warning is enabled by @option{-Wall}. 3834 3835@item -Wunused-but-set-parameter 3836@opindex Wunused-but-set-parameter 3837@opindex Wno-unused-but-set-parameter 3838Warn whenever a function parameter is assigned to, but otherwise unused 3839(aside from its declaration). 3840 3841To suppress this warning use the @samp{unused} attribute 3842(@pxref{Variable Attributes}). 3843 3844This warning is also enabled by @option{-Wunused} together with 3845@option{-Wextra}. 3846 3847@item -Wunused-but-set-variable 3848@opindex Wunused-but-set-variable 3849@opindex Wno-unused-but-set-variable 3850Warn whenever a local variable is assigned to, but otherwise unused 3851(aside from its declaration). 3852This warning is enabled by @option{-Wall}. 3853 3854To suppress this warning use the @samp{unused} attribute 3855(@pxref{Variable Attributes}). 3856 3857This warning is also enabled by @option{-Wunused}, which is enabled 3858by @option{-Wall}. 3859 3860@item -Wunused-function 3861@opindex Wunused-function 3862@opindex Wno-unused-function 3863Warn whenever a static function is declared but not defined or a 3864non-inline static function is unused. 3865This warning is enabled by @option{-Wall}. 3866 3867@item -Wunused-label 3868@opindex Wunused-label 3869@opindex Wno-unused-label 3870Warn whenever a label is declared but not used. 3871This warning is enabled by @option{-Wall}. 3872 3873To suppress this warning use the @samp{unused} attribute 3874(@pxref{Variable Attributes}). 3875 3876@item -Wunused-local-typedefs @r{(C, Objective-C, C++ and Objective-C++ only)} 3877@opindex Wunused-local-typedefs 3878Warn when a typedef locally defined in a function is not used. 3879This warning is enabled by @option{-Wall}. 3880 3881@item -Wunused-parameter 3882@opindex Wunused-parameter 3883@opindex Wno-unused-parameter 3884Warn whenever a function parameter is unused aside from its declaration. 3885 3886To suppress this warning use the @samp{unused} attribute 3887(@pxref{Variable Attributes}). 3888 3889@item -Wno-unused-result 3890@opindex Wunused-result 3891@opindex Wno-unused-result 3892Do not warn if a caller of a function marked with attribute 3893@code{warn_unused_result} (@pxref{Function Attributes}) does not use 3894its return value. The default is @option{-Wunused-result}. 3895 3896@item -Wunused-variable 3897@opindex Wunused-variable 3898@opindex Wno-unused-variable 3899Warn whenever a local variable or non-constant static variable is unused 3900aside from its declaration. 3901This warning is enabled by @option{-Wall}. 3902 3903To suppress this warning use the @samp{unused} attribute 3904(@pxref{Variable Attributes}). 3905 3906@item -Wunused-value 3907@opindex Wunused-value 3908@opindex Wno-unused-value 3909Warn whenever a statement computes a result that is explicitly not 3910used. To suppress this warning cast the unused expression to 3911@samp{void}. This includes an expression-statement or the left-hand 3912side of a comma expression that contains no side effects. For example, 3913an expression such as @samp{x[i,j]} causes a warning, while 3914@samp{x[(void)i,j]} does not. 3915 3916This warning is enabled by @option{-Wall}. 3917 3918@item -Wunused 3919@opindex Wunused 3920@opindex Wno-unused 3921All the above @option{-Wunused} options combined. 3922 3923In order to get a warning about an unused function parameter, you must 3924either specify @option{-Wextra -Wunused} (note that @option{-Wall} implies 3925@option{-Wunused}), or separately specify @option{-Wunused-parameter}. 3926 3927@item -Wuninitialized 3928@opindex Wuninitialized 3929@opindex Wno-uninitialized 3930Warn if an automatic variable is used without first being initialized 3931or if a variable may be clobbered by a @code{setjmp} call. In C++, 3932warn if a non-static reference or non-static @samp{const} member 3933appears in a class without constructors. 3934 3935If you want to warn about code that uses the uninitialized value of the 3936variable in its own initializer, use the @option{-Winit-self} option. 3937 3938These warnings occur for individual uninitialized or clobbered 3939elements of structure, union or array variables as well as for 3940variables that are uninitialized or clobbered as a whole. They do 3941not occur for variables or elements declared @code{volatile}. Because 3942these warnings depend on optimization, the exact variables or elements 3943for which there are warnings depends on the precise optimization 3944options and version of GCC used. 3945 3946Note that there may be no warning about a variable that is used only 3947to compute a value that itself is never used, because such 3948computations may be deleted by data flow analysis before the warnings 3949are printed. 3950 3951@item -Wmaybe-uninitialized 3952@opindex Wmaybe-uninitialized 3953@opindex Wno-maybe-uninitialized 3954For an automatic variable, if there exists a path from the function 3955entry to a use of the variable that is initialized, but there exist 3956some other paths for which the variable is not initialized, the compiler 3957emits a warning if it cannot prove the uninitialized paths are not 3958executed at run time. These warnings are made optional because GCC is 3959not smart enough to see all the reasons why the code might be correct 3960in spite of appearing to have an error. Here is one example of how 3961this can happen: 3962 3963@smallexample 3964@group 3965@{ 3966 int x; 3967 switch (y) 3968 @{ 3969 case 1: x = 1; 3970 break; 3971 case 2: x = 4; 3972 break; 3973 case 3: x = 5; 3974 @} 3975 foo (x); 3976@} 3977@end group 3978@end smallexample 3979 3980@noindent 3981If the value of @code{y} is always 1, 2 or 3, then @code{x} is 3982always initialized, but GCC doesn't know this. To suppress the 3983warning, you need to provide a default case with assert(0) or 3984similar code. 3985 3986@cindex @code{longjmp} warnings 3987This option also warns when a non-volatile automatic variable might be 3988changed by a call to @code{longjmp}. These warnings as well are possible 3989only in optimizing compilation. 3990 3991The compiler sees only the calls to @code{setjmp}. It cannot know 3992where @code{longjmp} will be called; in fact, a signal handler could 3993call it at any point in the code. As a result, you may get a warning 3994even when there is in fact no problem because @code{longjmp} cannot 3995in fact be called at the place that would cause a problem. 3996 3997Some spurious warnings can be avoided if you declare all the functions 3998you use that never return as @code{noreturn}. @xref{Function 3999Attributes}. 4000 4001This warning is enabled by @option{-Wall} or @option{-Wextra}. 4002 4003@item -Wunknown-pragmas 4004@opindex Wunknown-pragmas 4005@opindex Wno-unknown-pragmas 4006@cindex warning for unknown pragmas 4007@cindex unknown pragmas, warning 4008@cindex pragmas, warning of unknown 4009Warn when a @code{#pragma} directive is encountered that is not understood by 4010GCC@. If this command-line option is used, warnings are even issued 4011for unknown pragmas in system header files. This is not the case if 4012the warnings are only enabled by the @option{-Wall} command-line option. 4013 4014@item -Wno-pragmas 4015@opindex Wno-pragmas 4016@opindex Wpragmas 4017Do not warn about misuses of pragmas, such as incorrect parameters, 4018invalid syntax, or conflicts between pragmas. See also 4019@option{-Wunknown-pragmas}. 4020 4021@item -Wstrict-aliasing 4022@opindex Wstrict-aliasing 4023@opindex Wno-strict-aliasing 4024This option is only active when @option{-fstrict-aliasing} is active. 4025It warns about code that might break the strict aliasing rules that the 4026compiler is using for optimization. The warning does not catch all 4027cases, but does attempt to catch the more common pitfalls. It is 4028included in @option{-Wall}. 4029It is equivalent to @option{-Wstrict-aliasing=3} 4030 4031@item -Wstrict-aliasing=n 4032@opindex Wstrict-aliasing=n 4033This option is only active when @option{-fstrict-aliasing} is active. 4034It warns about code that might break the strict aliasing rules that the 4035compiler is using for optimization. 4036Higher levels correspond to higher accuracy (fewer false positives). 4037Higher levels also correspond to more effort, similar to the way @option{-O} 4038works. 4039@option{-Wstrict-aliasing} is equivalent to @option{-Wstrict-aliasing=3}. 4040 4041Level 1: Most aggressive, quick, least accurate. 4042Possibly useful when higher levels 4043do not warn but @option{-fstrict-aliasing} still breaks the code, as it has very few 4044false negatives. However, it has many false positives. 4045Warns for all pointer conversions between possibly incompatible types, 4046even if never dereferenced. Runs in the front end only. 4047 4048Level 2: Aggressive, quick, not too precise. 4049May still have many false positives (not as many as level 1 though), 4050and few false negatives (but possibly more than level 1). 4051Unlike level 1, it only warns when an address is taken. Warns about 4052incomplete types. Runs in the front end only. 4053 4054Level 3 (default for @option{-Wstrict-aliasing}): 4055Should have very few false positives and few false 4056negatives. Slightly slower than levels 1 or 2 when optimization is enabled. 4057Takes care of the common pun+dereference pattern in the front end: 4058@code{*(int*)&some_float}. 4059If optimization is enabled, it also runs in the back end, where it deals 4060with multiple statement cases using flow-sensitive points-to information. 4061Only warns when the converted pointer is dereferenced. 4062Does not warn about incomplete types. 4063 4064@item -Wstrict-overflow 4065@itemx -Wstrict-overflow=@var{n} 4066@opindex Wstrict-overflow 4067@opindex Wno-strict-overflow 4068This option is only active when @option{-fstrict-overflow} is active. 4069It warns about cases where the compiler optimizes based on the 4070assumption that signed overflow does not occur. Note that it does not 4071warn about all cases where the code might overflow: it only warns 4072about cases where the compiler implements some optimization. Thus 4073this warning depends on the optimization level. 4074 4075An optimization that assumes that signed overflow does not occur is 4076perfectly safe if the values of the variables involved are such that 4077overflow never does, in fact, occur. Therefore this warning can 4078easily give a false positive: a warning about code that is not 4079actually a problem. To help focus on important issues, several 4080warning levels are defined. No warnings are issued for the use of 4081undefined signed overflow when estimating how many iterations a loop 4082requires, in particular when determining whether a loop will be 4083executed at all. 4084 4085@table @gcctabopt 4086@item -Wstrict-overflow=1 4087Warn about cases that are both questionable and easy to avoid. For 4088example, with @option{-fstrict-overflow}, the compiler simplifies 4089@code{x + 1 > x} to @code{1}. This level of 4090@option{-Wstrict-overflow} is enabled by @option{-Wall}; higher levels 4091are not, and must be explicitly requested. 4092 4093@item -Wstrict-overflow=2 4094Also warn about other cases where a comparison is simplified to a 4095constant. For example: @code{abs (x) >= 0}. This can only be 4096simplified when @option{-fstrict-overflow} is in effect, because 4097@code{abs (INT_MIN)} overflows to @code{INT_MIN}, which is less than 4098zero. @option{-Wstrict-overflow} (with no level) is the same as 4099@option{-Wstrict-overflow=2}. 4100 4101@item -Wstrict-overflow=3 4102Also warn about other cases where a comparison is simplified. For 4103example: @code{x + 1 > 1} is simplified to @code{x > 0}. 4104 4105@item -Wstrict-overflow=4 4106Also warn about other simplifications not covered by the above cases. 4107For example: @code{(x * 10) / 5} is simplified to @code{x * 2}. 4108 4109@item -Wstrict-overflow=5 4110Also warn about cases where the compiler reduces the magnitude of a 4111constant involved in a comparison. For example: @code{x + 2 > y} is 4112simplified to @code{x + 1 >= y}. This is reported only at the 4113highest warning level because this simplification applies to many 4114comparisons, so this warning level gives a very large number of 4115false positives. 4116@end table 4117 4118@item -Wsuggest-attribute=@r{[}pure@r{|}const@r{|}noreturn@r{|}format@r{]} 4119@opindex Wsuggest-attribute= 4120@opindex Wno-suggest-attribute= 4121Warn for cases where adding an attribute may be beneficial. The 4122attributes currently supported are listed below. 4123 4124@table @gcctabopt 4125@item -Wsuggest-attribute=pure 4126@itemx -Wsuggest-attribute=const 4127@itemx -Wsuggest-attribute=noreturn 4128@opindex Wsuggest-attribute=pure 4129@opindex Wno-suggest-attribute=pure 4130@opindex Wsuggest-attribute=const 4131@opindex Wno-suggest-attribute=const 4132@opindex Wsuggest-attribute=noreturn 4133@opindex Wno-suggest-attribute=noreturn 4134 4135Warn about functions that might be candidates for attributes 4136@code{pure}, @code{const} or @code{noreturn}. The compiler only warns for 4137functions visible in other compilation units or (in the case of @code{pure} and 4138@code{const}) if it cannot prove that the function returns normally. A function 4139returns normally if it doesn't contain an infinite loop or return abnormally 4140by throwing, calling @code{abort()} or trapping. This analysis requires option 4141@option{-fipa-pure-const}, which is enabled by default at @option{-O} and 4142higher. Higher optimization levels improve the accuracy of the analysis. 4143 4144@item -Wsuggest-attribute=format 4145@itemx -Wmissing-format-attribute 4146@opindex Wsuggest-attribute=format 4147@opindex Wmissing-format-attribute 4148@opindex Wno-suggest-attribute=format 4149@opindex Wno-missing-format-attribute 4150@opindex Wformat 4151@opindex Wno-format 4152 4153Warn about function pointers that might be candidates for @code{format} 4154attributes. Note these are only possible candidates, not absolute ones. 4155GCC guesses that function pointers with @code{format} attributes that 4156are used in assignment, initialization, parameter passing or return 4157statements should have a corresponding @code{format} attribute in the 4158resulting type. I.e.@: the left-hand side of the assignment or 4159initialization, the type of the parameter variable, or the return type 4160of the containing function respectively should also have a @code{format} 4161attribute to avoid the warning. 4162 4163GCC also warns about function definitions that might be 4164candidates for @code{format} attributes. Again, these are only 4165possible candidates. GCC guesses that @code{format} attributes 4166might be appropriate for any function that calls a function like 4167@code{vprintf} or @code{vscanf}, but this might not always be the 4168case, and some functions for which @code{format} attributes are 4169appropriate may not be detected. 4170@end table 4171 4172@item -Warray-bounds 4173@opindex Wno-array-bounds 4174@opindex Warray-bounds 4175This option is only active when @option{-ftree-vrp} is active 4176(default for @option{-O2} and above). It warns about subscripts to arrays 4177that are always out of bounds. This warning is enabled by @option{-Wall}. 4178 4179@item -Wno-div-by-zero 4180@opindex Wno-div-by-zero 4181@opindex Wdiv-by-zero 4182Do not warn about compile-time integer division by zero. Floating-point 4183division by zero is not warned about, as it can be a legitimate way of 4184obtaining infinities and NaNs. 4185 4186@item -Wsystem-headers 4187@opindex Wsystem-headers 4188@opindex Wno-system-headers 4189@cindex warnings from system headers 4190@cindex system headers, warnings from 4191Print warning messages for constructs found in system header files. 4192Warnings from system headers are normally suppressed, on the assumption 4193that they usually do not indicate real problems and would only make the 4194compiler output harder to read. Using this command-line option tells 4195GCC to emit warnings from system headers as if they occurred in user 4196code. However, note that using @option{-Wall} in conjunction with this 4197option does @emph{not} warn about unknown pragmas in system 4198headers---for that, @option{-Wunknown-pragmas} must also be used. 4199 4200@item -Wtrampolines 4201@opindex Wtrampolines 4202@opindex Wno-trampolines 4203 Warn about trampolines generated for pointers to nested functions. 4204 4205 A trampoline is a small piece of data or code that is created at run 4206 time on the stack when the address of a nested function is taken, and 4207 is used to call the nested function indirectly. For some targets, it 4208 is made up of data only and thus requires no special treatment. But, 4209 for most targets, it is made up of code and thus requires the stack 4210 to be made executable in order for the program to work properly. 4211 4212@item -Wfloat-equal 4213@opindex Wfloat-equal 4214@opindex Wno-float-equal 4215Warn if floating-point values are used in equality comparisons. 4216 4217The idea behind this is that sometimes it is convenient (for the 4218programmer) to consider floating-point values as approximations to 4219infinitely precise real numbers. If you are doing this, then you need 4220to compute (by analyzing the code, or in some other way) the maximum or 4221likely maximum error that the computation introduces, and allow for it 4222when performing comparisons (and when producing output, but that's a 4223different problem). In particular, instead of testing for equality, you 4224should check to see whether the two values have ranges that overlap; and 4225this is done with the relational operators, so equality comparisons are 4226probably mistaken. 4227 4228@item -Wtraditional @r{(C and Objective-C only)} 4229@opindex Wtraditional 4230@opindex Wno-traditional 4231Warn about certain constructs that behave differently in traditional and 4232ISO C@. Also warn about ISO C constructs that have no traditional C 4233equivalent, and/or problematic constructs that should be avoided. 4234 4235@itemize @bullet 4236@item 4237Macro parameters that appear within string literals in the macro body. 4238In traditional C macro replacement takes place within string literals, 4239but in ISO C it does not. 4240 4241@item 4242In traditional C, some preprocessor directives did not exist. 4243Traditional preprocessors only considered a line to be a directive 4244if the @samp{#} appeared in column 1 on the line. Therefore 4245@option{-Wtraditional} warns about directives that traditional C 4246understands but ignores because the @samp{#} does not appear as the 4247first character on the line. It also suggests you hide directives like 4248@samp{#pragma} not understood by traditional C by indenting them. Some 4249traditional implementations do not recognize @samp{#elif}, so this option 4250suggests avoiding it altogether. 4251 4252@item 4253A function-like macro that appears without arguments. 4254 4255@item 4256The unary plus operator. 4257 4258@item 4259The @samp{U} integer constant suffix, or the @samp{F} or @samp{L} floating-point 4260constant suffixes. (Traditional C does support the @samp{L} suffix on integer 4261constants.) Note, these suffixes appear in macros defined in the system 4262headers of most modern systems, e.g.@: the @samp{_MIN}/@samp{_MAX} macros in @code{<limits.h>}. 4263Use of these macros in user code might normally lead to spurious 4264warnings, however GCC's integrated preprocessor has enough context to 4265avoid warning in these cases. 4266 4267@item 4268A function declared external in one block and then used after the end of 4269the block. 4270 4271@item 4272A @code{switch} statement has an operand of type @code{long}. 4273 4274@item 4275A non-@code{static} function declaration follows a @code{static} one. 4276This construct is not accepted by some traditional C compilers. 4277 4278@item 4279The ISO type of an integer constant has a different width or 4280signedness from its traditional type. This warning is only issued if 4281the base of the constant is ten. I.e.@: hexadecimal or octal values, which 4282typically represent bit patterns, are not warned about. 4283 4284@item 4285Usage of ISO string concatenation is detected. 4286 4287@item 4288Initialization of automatic aggregates. 4289 4290@item 4291Identifier conflicts with labels. Traditional C lacks a separate 4292namespace for labels. 4293 4294@item 4295Initialization of unions. If the initializer is zero, the warning is 4296omitted. This is done under the assumption that the zero initializer in 4297user code appears conditioned on e.g.@: @code{__STDC__} to avoid missing 4298initializer warnings and relies on default initialization to zero in the 4299traditional C case. 4300 4301@item 4302Conversions by prototypes between fixed/floating-point values and vice 4303versa. The absence of these prototypes when compiling with traditional 4304C causes serious problems. This is a subset of the possible 4305conversion warnings; for the full set use @option{-Wtraditional-conversion}. 4306 4307@item 4308Use of ISO C style function definitions. This warning intentionally is 4309@emph{not} issued for prototype declarations or variadic functions 4310because these ISO C features appear in your code when using 4311libiberty's traditional C compatibility macros, @code{PARAMS} and 4312@code{VPARAMS}. This warning is also bypassed for nested functions 4313because that feature is already a GCC extension and thus not relevant to 4314traditional C compatibility. 4315@end itemize 4316 4317@item -Wtraditional-conversion @r{(C and Objective-C only)} 4318@opindex Wtraditional-conversion 4319@opindex Wno-traditional-conversion 4320Warn if a prototype causes a type conversion that is different from what 4321would happen to the same argument in the absence of a prototype. This 4322includes conversions of fixed point to floating and vice versa, and 4323conversions changing the width or signedness of a fixed-point argument 4324except when the same as the default promotion. 4325 4326@item -Wdeclaration-after-statement @r{(C and Objective-C only)} 4327@opindex Wdeclaration-after-statement 4328@opindex Wno-declaration-after-statement 4329Warn when a declaration is found after a statement in a block. This 4330construct, known from C++, was introduced with ISO C99 and is by default 4331allowed in GCC@. It is not supported by ISO C90 and was not supported by 4332GCC versions before GCC 3.0. @xref{Mixed Declarations}. 4333 4334@item -Wundef 4335@opindex Wundef 4336@opindex Wno-undef 4337Warn if an undefined identifier is evaluated in an @samp{#if} directive. 4338 4339@item -Wno-endif-labels 4340@opindex Wno-endif-labels 4341@opindex Wendif-labels 4342Do not warn whenever an @samp{#else} or an @samp{#endif} are followed by text. 4343 4344@item -Wshadow 4345@opindex Wshadow 4346@opindex Wno-shadow 4347Warn whenever a local variable or type declaration shadows another variable, 4348parameter, type, or class member (in C++), or whenever a built-in function 4349is shadowed. Note that in C++, the compiler warns if a local variable 4350shadows an explicit typedef, but not if it shadows a struct/class/enum. 4351 4352@item -Wlarger-than=@var{len} 4353@opindex Wlarger-than=@var{len} 4354@opindex Wlarger-than-@var{len} 4355Warn whenever an object of larger than @var{len} bytes is defined. 4356 4357@item -Wframe-larger-than=@var{len} 4358@opindex Wframe-larger-than 4359Warn if the size of a function frame is larger than @var{len} bytes. 4360The computation done to determine the stack frame size is approximate 4361and not conservative. 4362The actual requirements may be somewhat greater than @var{len} 4363even if you do not get a warning. In addition, any space allocated 4364via @code{alloca}, variable-length arrays, or related constructs 4365is not included by the compiler when determining 4366whether or not to issue a warning. 4367 4368@item -Wno-free-nonheap-object 4369@opindex Wno-free-nonheap-object 4370@opindex Wfree-nonheap-object 4371Do not warn when attempting to free an object that was not allocated 4372on the heap. 4373 4374@item -Wstack-usage=@var{len} 4375@opindex Wstack-usage 4376Warn if the stack usage of a function might be larger than @var{len} bytes. 4377The computation done to determine the stack usage is conservative. 4378Any space allocated via @code{alloca}, variable-length arrays, or related 4379constructs is included by the compiler when determining whether or not to 4380issue a warning. 4381 4382The message is in keeping with the output of @option{-fstack-usage}. 4383 4384@itemize 4385@item 4386If the stack usage is fully static but exceeds the specified amount, it's: 4387 4388@smallexample 4389 warning: stack usage is 1120 bytes 4390@end smallexample 4391@item 4392If the stack usage is (partly) dynamic but bounded, it's: 4393 4394@smallexample 4395 warning: stack usage might be 1648 bytes 4396@end smallexample 4397@item 4398If the stack usage is (partly) dynamic and not bounded, it's: 4399 4400@smallexample 4401 warning: stack usage might be unbounded 4402@end smallexample 4403@end itemize 4404 4405@item -Wunsafe-loop-optimizations 4406@opindex Wunsafe-loop-optimizations 4407@opindex Wno-unsafe-loop-optimizations 4408Warn if the loop cannot be optimized because the compiler cannot 4409assume anything on the bounds of the loop indices. With 4410@option{-funsafe-loop-optimizations} warn if the compiler makes 4411such assumptions. 4412 4413@item -Wno-pedantic-ms-format @r{(MinGW targets only)} 4414@opindex Wno-pedantic-ms-format 4415@opindex Wpedantic-ms-format 4416When used in combination with @option{-Wformat} 4417and @option{-pedantic} without GNU extensions, this option 4418disables the warnings about non-ISO @code{printf} / @code{scanf} format 4419width specifiers @code{I32}, @code{I64}, and @code{I} used on Windows targets, 4420which depend on the MS runtime. 4421 4422@item -Wpointer-arith 4423@opindex Wpointer-arith 4424@opindex Wno-pointer-arith 4425Warn about anything that depends on the ``size of'' a function type or 4426of @code{void}. GNU C assigns these types a size of 1, for 4427convenience in calculations with @code{void *} pointers and pointers 4428to functions. In C++, warn also when an arithmetic operation involves 4429@code{NULL}. This warning is also enabled by @option{-Wpedantic}. 4430 4431@item -Wtype-limits 4432@opindex Wtype-limits 4433@opindex Wno-type-limits 4434Warn if a comparison is always true or always false due to the limited 4435range of the data type, but do not warn for constant expressions. For 4436example, warn if an unsigned variable is compared against zero with 4437@samp{<} or @samp{>=}. This warning is also enabled by 4438@option{-Wextra}. 4439 4440@item -Wbad-function-cast @r{(C and Objective-C only)} 4441@opindex Wbad-function-cast 4442@opindex Wno-bad-function-cast 4443Warn whenever a function call is cast to a non-matching type. 4444For example, warn if @code{int malloc()} is cast to @code{anything *}. 4445 4446@item -Wc++-compat @r{(C and Objective-C only)} 4447Warn about ISO C constructs that are outside of the common subset of 4448ISO C and ISO C++, e.g.@: request for implicit conversion from 4449@code{void *} to a pointer to non-@code{void} type. 4450 4451@item -Wc++11-compat @r{(C++ and Objective-C++ only)} 4452Warn about C++ constructs whose meaning differs between ISO C++ 1998 4453and ISO C++ 2011, e.g., identifiers in ISO C++ 1998 that are keywords 4454in ISO C++ 2011. This warning turns on @option{-Wnarrowing} and is 4455enabled by @option{-Wall}. 4456 4457@item -Wcast-qual 4458@opindex Wcast-qual 4459@opindex Wno-cast-qual 4460Warn whenever a pointer is cast so as to remove a type qualifier from 4461the target type. For example, warn if a @code{const char *} is cast 4462to an ordinary @code{char *}. 4463 4464Also warn when making a cast that introduces a type qualifier in an 4465unsafe way. For example, casting @code{char **} to @code{const char **} 4466is unsafe, as in this example: 4467 4468@smallexample 4469 /* p is char ** value. */ 4470 const char **q = (const char **) p; 4471 /* Assignment of readonly string to const char * is OK. */ 4472 *q = "string"; 4473 /* Now char** pointer points to read-only memory. */ 4474 **p = 'b'; 4475@end smallexample 4476 4477@item -Wcast-align 4478@opindex Wcast-align 4479@opindex Wno-cast-align 4480Warn whenever a pointer is cast such that the required alignment of the 4481target is increased. For example, warn if a @code{char *} is cast to 4482an @code{int *} on machines where integers can only be accessed at 4483two- or four-byte boundaries. 4484 4485@item -Wwrite-strings 4486@opindex Wwrite-strings 4487@opindex Wno-write-strings 4488When compiling C, give string constants the type @code{const 4489char[@var{length}]} so that copying the address of one into a 4490non-@code{const} @code{char *} pointer produces a warning. These 4491warnings help you find at compile time code that can try to write 4492into a string constant, but only if you have been very careful about 4493using @code{const} in declarations and prototypes. Otherwise, it is 4494just a nuisance. This is why we did not make @option{-Wall} request 4495these warnings. 4496 4497When compiling C++, warn about the deprecated conversion from string 4498literals to @code{char *}. This warning is enabled by default for C++ 4499programs. 4500 4501@item -Wclobbered 4502@opindex Wclobbered 4503@opindex Wno-clobbered 4504Warn for variables that might be changed by @samp{longjmp} or 4505@samp{vfork}. This warning is also enabled by @option{-Wextra}. 4506 4507@item -Wconditionally-supported @r{(C++ and Objective-C++ only)} 4508@opindex Wconditionally-supported 4509@opindex Wno-conditionally-supported 4510Warn for conditionally-supported (C++11 [intro.defs]) constructs. 4511 4512@item -Wconversion 4513@opindex Wconversion 4514@opindex Wno-conversion 4515Warn for implicit conversions that may alter a value. This includes 4516conversions between real and integer, like @code{abs (x)} when 4517@code{x} is @code{double}; conversions between signed and unsigned, 4518like @code{unsigned ui = -1}; and conversions to smaller types, like 4519@code{sqrtf (M_PI)}. Do not warn for explicit casts like @code{abs 4520((int) x)} and @code{ui = (unsigned) -1}, or if the value is not 4521changed by the conversion like in @code{abs (2.0)}. Warnings about 4522conversions between signed and unsigned integers can be disabled by 4523using @option{-Wno-sign-conversion}. 4524 4525For C++, also warn for confusing overload resolution for user-defined 4526conversions; and conversions that never use a type conversion 4527operator: conversions to @code{void}, the same type, a base class or a 4528reference to them. Warnings about conversions between signed and 4529unsigned integers are disabled by default in C++ unless 4530@option{-Wsign-conversion} is explicitly enabled. 4531 4532@item -Wno-conversion-null @r{(C++ and Objective-C++ only)} 4533@opindex Wconversion-null 4534@opindex Wno-conversion-null 4535Do not warn for conversions between @code{NULL} and non-pointer 4536types. @option{-Wconversion-null} is enabled by default. 4537 4538@item -Wzero-as-null-pointer-constant @r{(C++ and Objective-C++ only)} 4539@opindex Wzero-as-null-pointer-constant 4540@opindex Wno-zero-as-null-pointer-constant 4541Warn when a literal '0' is used as null pointer constant. This can 4542be useful to facilitate the conversion to @code{nullptr} in C++11. 4543 4544@item -Wdate-time 4545@opindex Wdate-time 4546@opindex Wno-date-time 4547Warn when macros @code{__TIME__}, @code{__DATE__} or @code{__TIMESTAMP__} 4548are encountered as they might prevent bit-wise-identical reproducible 4549compilations. 4550 4551@item -Wdelete-incomplete @r{(C++ and Objective-C++ only)} 4552@opindex Wdelete-incomplete 4553@opindex Wno-delete-incomplete 4554Warn when deleting a pointer to incomplete type, which may cause 4555undefined behavior at runtime. This warning is enabled by default. 4556 4557@item -Wuseless-cast @r{(C++ and Objective-C++ only)} 4558@opindex Wuseless-cast 4559@opindex Wno-useless-cast 4560Warn when an expression is casted to its own type. 4561 4562@item -Wempty-body 4563@opindex Wempty-body 4564@opindex Wno-empty-body 4565Warn if an empty body occurs in an @samp{if}, @samp{else} or @samp{do 4566while} statement. This warning is also enabled by @option{-Wextra}. 4567 4568@item -Wenum-compare 4569@opindex Wenum-compare 4570@opindex Wno-enum-compare 4571Warn about a comparison between values of different enumerated types. 4572In C++ enumeral mismatches in conditional expressions are also 4573diagnosed and the warning is enabled by default. In C this warning is 4574enabled by @option{-Wall}. 4575 4576@item -Wjump-misses-init @r{(C, Objective-C only)} 4577@opindex Wjump-misses-init 4578@opindex Wno-jump-misses-init 4579Warn if a @code{goto} statement or a @code{switch} statement jumps 4580forward across the initialization of a variable, or jumps backward to a 4581label after the variable has been initialized. This only warns about 4582variables that are initialized when they are declared. This warning is 4583only supported for C and Objective-C; in C++ this sort of branch is an 4584error in any case. 4585 4586@option{-Wjump-misses-init} is included in @option{-Wc++-compat}. It 4587can be disabled with the @option{-Wno-jump-misses-init} option. 4588 4589@item -Wsign-compare 4590@opindex Wsign-compare 4591@opindex Wno-sign-compare 4592@cindex warning for comparison of signed and unsigned values 4593@cindex comparison of signed and unsigned values, warning 4594@cindex signed and unsigned values, comparison warning 4595Warn when a comparison between signed and unsigned values could produce 4596an incorrect result when the signed value is converted to unsigned. 4597This warning is also enabled by @option{-Wextra}; to get the other warnings 4598of @option{-Wextra} without this warning, use @option{-Wextra -Wno-sign-compare}. 4599 4600@item -Wsign-conversion 4601@opindex Wsign-conversion 4602@opindex Wno-sign-conversion 4603Warn for implicit conversions that may change the sign of an integer 4604value, like assigning a signed integer expression to an unsigned 4605integer variable. An explicit cast silences the warning. In C, this 4606option is enabled also by @option{-Wconversion}. 4607 4608@item -Wfloat-conversion 4609@opindex Wfloat-conversion 4610@opindex Wno-float-conversion 4611Warn for implicit conversions that reduce the precision of a real value. 4612This includes conversions from real to integer, and from higher precision 4613real to lower precision real values. This option is also enabled by 4614@option{-Wconversion}. 4615 4616@item -Wsizeof-pointer-memaccess 4617@opindex Wsizeof-pointer-memaccess 4618@opindex Wno-sizeof-pointer-memaccess 4619Warn for suspicious length parameters to certain string and memory built-in 4620functions if the argument uses @code{sizeof}. This warning warns e.g.@: 4621about @code{memset (ptr, 0, sizeof (ptr));} if @code{ptr} is not an array, 4622but a pointer, and suggests a possible fix, or about 4623@code{memcpy (&foo, ptr, sizeof (&foo));}. This warning is enabled by 4624@option{-Wall}. 4625 4626@item -Waddress 4627@opindex Waddress 4628@opindex Wno-address 4629Warn about suspicious uses of memory addresses. These include using 4630the address of a function in a conditional expression, such as 4631@code{void func(void); if (func)}, and comparisons against the memory 4632address of a string literal, such as @code{if (x == "abc")}. Such 4633uses typically indicate a programmer error: the address of a function 4634always evaluates to true, so their use in a conditional usually 4635indicate that the programmer forgot the parentheses in a function 4636call; and comparisons against string literals result in unspecified 4637behavior and are not portable in C, so they usually indicate that the 4638programmer intended to use @code{strcmp}. This warning is enabled by 4639@option{-Wall}. 4640 4641@item -Wlogical-op 4642@opindex Wlogical-op 4643@opindex Wno-logical-op 4644Warn about suspicious uses of logical operators in expressions. 4645This includes using logical operators in contexts where a 4646bit-wise operator is likely to be expected. 4647 4648@item -Waggregate-return 4649@opindex Waggregate-return 4650@opindex Wno-aggregate-return 4651Warn if any functions that return structures or unions are defined or 4652called. (In languages where you can return an array, this also elicits 4653a warning.) 4654 4655@item -Wno-aggressive-loop-optimizations 4656@opindex Wno-aggressive-loop-optimizations 4657@opindex Waggressive-loop-optimizations 4658Warn if in a loop with constant number of iterations the compiler detects 4659undefined behavior in some statement during one or more of the iterations. 4660 4661@item -Wno-attributes 4662@opindex Wno-attributes 4663@opindex Wattributes 4664Do not warn if an unexpected @code{__attribute__} is used, such as 4665unrecognized attributes, function attributes applied to variables, 4666etc. This does not stop errors for incorrect use of supported 4667attributes. 4668 4669@item -Wno-builtin-macro-redefined 4670@opindex Wno-builtin-macro-redefined 4671@opindex Wbuiltin-macro-redefined 4672Do not warn if certain built-in macros are redefined. This suppresses 4673warnings for redefinition of @code{__TIMESTAMP__}, @code{__TIME__}, 4674@code{__DATE__}, @code{__FILE__}, and @code{__BASE_FILE__}. 4675 4676@item -Wstrict-prototypes @r{(C and Objective-C only)} 4677@opindex Wstrict-prototypes 4678@opindex Wno-strict-prototypes 4679Warn if a function is declared or defined without specifying the 4680argument types. (An old-style function definition is permitted without 4681a warning if preceded by a declaration that specifies the argument 4682types.) 4683 4684@item -Wold-style-declaration @r{(C and Objective-C only)} 4685@opindex Wold-style-declaration 4686@opindex Wno-old-style-declaration 4687Warn for obsolescent usages, according to the C Standard, in a 4688declaration. For example, warn if storage-class specifiers like 4689@code{static} are not the first things in a declaration. This warning 4690is also enabled by @option{-Wextra}. 4691 4692@item -Wold-style-definition @r{(C and Objective-C only)} 4693@opindex Wold-style-definition 4694@opindex Wno-old-style-definition 4695Warn if an old-style function definition is used. A warning is given 4696even if there is a previous prototype. 4697 4698@item -Wmissing-parameter-type @r{(C and Objective-C only)} 4699@opindex Wmissing-parameter-type 4700@opindex Wno-missing-parameter-type 4701A function parameter is declared without a type specifier in K&R-style 4702functions: 4703 4704@smallexample 4705void foo(bar) @{ @} 4706@end smallexample 4707 4708This warning is also enabled by @option{-Wextra}. 4709 4710@item -Wmissing-prototypes @r{(C and Objective-C only)} 4711@opindex Wmissing-prototypes 4712@opindex Wno-missing-prototypes 4713Warn if a global function is defined without a previous prototype 4714declaration. This warning is issued even if the definition itself 4715provides a prototype. Use this option to detect global functions 4716that do not have a matching prototype declaration in a header file. 4717This option is not valid for C++ because all function declarations 4718provide prototypes and a non-matching declaration will declare an 4719overload rather than conflict with an earlier declaration. 4720Use @option{-Wmissing-declarations} to detect missing declarations in C++. 4721 4722@item -Wmissing-declarations 4723@opindex Wmissing-declarations 4724@opindex Wno-missing-declarations 4725Warn if a global function is defined without a previous declaration. 4726Do so even if the definition itself provides a prototype. 4727Use this option to detect global functions that are not declared in 4728header files. In C, no warnings are issued for functions with previous 4729non-prototype declarations; use @option{-Wmissing-prototype} to detect 4730missing prototypes. In C++, no warnings are issued for function templates, 4731or for inline functions, or for functions in anonymous namespaces. 4732 4733@item -Wmissing-field-initializers 4734@opindex Wmissing-field-initializers 4735@opindex Wno-missing-field-initializers 4736@opindex W 4737@opindex Wextra 4738@opindex Wno-extra 4739Warn if a structure's initializer has some fields missing. For 4740example, the following code causes such a warning, because 4741@code{x.h} is implicitly zero: 4742 4743@smallexample 4744struct s @{ int f, g, h; @}; 4745struct s x = @{ 3, 4 @}; 4746@end smallexample 4747 4748This option does not warn about designated initializers, so the following 4749modification does not trigger a warning: 4750 4751@smallexample 4752struct s @{ int f, g, h; @}; 4753struct s x = @{ .f = 3, .g = 4 @}; 4754@end smallexample 4755 4756This warning is included in @option{-Wextra}. To get other @option{-Wextra} 4757warnings without this one, use @option{-Wextra -Wno-missing-field-initializers}. 4758 4759@item -Wno-multichar 4760@opindex Wno-multichar 4761@opindex Wmultichar 4762Do not warn if a multicharacter constant (@samp{'FOOF'}) is used. 4763Usually they indicate a typo in the user's code, as they have 4764implementation-defined values, and should not be used in portable code. 4765 4766@item -Wnormalized=<none|id|nfc|nfkc> 4767@opindex Wnormalized= 4768@cindex NFC 4769@cindex NFKC 4770@cindex character set, input normalization 4771In ISO C and ISO C++, two identifiers are different if they are 4772different sequences of characters. However, sometimes when characters 4773outside the basic ASCII character set are used, you can have two 4774different character sequences that look the same. To avoid confusion, 4775the ISO 10646 standard sets out some @dfn{normalization rules} which 4776when applied ensure that two sequences that look the same are turned into 4777the same sequence. GCC can warn you if you are using identifiers that 4778have not been normalized; this option controls that warning. 4779 4780There are four levels of warning supported by GCC@. The default is 4781@option{-Wnormalized=nfc}, which warns about any identifier that is 4782not in the ISO 10646 ``C'' normalized form, @dfn{NFC}. NFC is the 4783recommended form for most uses. 4784 4785Unfortunately, there are some characters allowed in identifiers by 4786ISO C and ISO C++ that, when turned into NFC, are not allowed in 4787identifiers. That is, there's no way to use these symbols in portable 4788ISO C or C++ and have all your identifiers in NFC@. 4789@option{-Wnormalized=id} suppresses the warning for these characters. 4790It is hoped that future versions of the standards involved will correct 4791this, which is why this option is not the default. 4792 4793You can switch the warning off for all characters by writing 4794@option{-Wnormalized=none}. You should only do this if you 4795are using some other normalization scheme (like ``D''), because 4796otherwise you can easily create bugs that are literally impossible to see. 4797 4798Some characters in ISO 10646 have distinct meanings but look identical 4799in some fonts or display methodologies, especially once formatting has 4800been applied. For instance @code{\u207F}, ``SUPERSCRIPT LATIN SMALL 4801LETTER N'', displays just like a regular @code{n} that has been 4802placed in a superscript. ISO 10646 defines the @dfn{NFKC} 4803normalization scheme to convert all these into a standard form as 4804well, and GCC warns if your code is not in NFKC if you use 4805@option{-Wnormalized=nfkc}. This warning is comparable to warning 4806about every identifier that contains the letter O because it might be 4807confused with the digit 0, and so is not the default, but may be 4808useful as a local coding convention if the programming environment 4809cannot be fixed to display these characters distinctly. 4810 4811@item -Wno-deprecated 4812@opindex Wno-deprecated 4813@opindex Wdeprecated 4814Do not warn about usage of deprecated features. @xref{Deprecated Features}. 4815 4816@item -Wno-deprecated-declarations 4817@opindex Wno-deprecated-declarations 4818@opindex Wdeprecated-declarations 4819Do not warn about uses of functions (@pxref{Function Attributes}), 4820variables (@pxref{Variable Attributes}), and types (@pxref{Type 4821Attributes}) marked as deprecated by using the @code{deprecated} 4822attribute. 4823 4824@item -Wno-overflow 4825@opindex Wno-overflow 4826@opindex Woverflow 4827Do not warn about compile-time overflow in constant expressions. 4828 4829@item -Wopenmp-simd 4830@opindex Wopenm-simd 4831Warn if the vectorizer cost model overrides the OpenMP or the Cilk Plus 4832simd directive set by user. The @option{-fsimd-cost-model=unlimited} can 4833be used to relax the cost model. 4834 4835@item -Woverride-init @r{(C and Objective-C only)} 4836@opindex Woverride-init 4837@opindex Wno-override-init 4838@opindex W 4839@opindex Wextra 4840@opindex Wno-extra 4841Warn if an initialized field without side effects is overridden when 4842using designated initializers (@pxref{Designated Inits, , Designated 4843Initializers}). 4844 4845This warning is included in @option{-Wextra}. To get other 4846@option{-Wextra} warnings without this one, use @option{-Wextra 4847-Wno-override-init}. 4848 4849@item -Wpacked 4850@opindex Wpacked 4851@opindex Wno-packed 4852Warn if a structure is given the packed attribute, but the packed 4853attribute has no effect on the layout or size of the structure. 4854Such structures may be mis-aligned for little benefit. For 4855instance, in this code, the variable @code{f.x} in @code{struct bar} 4856is misaligned even though @code{struct bar} does not itself 4857have the packed attribute: 4858 4859@smallexample 4860@group 4861struct foo @{ 4862 int x; 4863 char a, b, c, d; 4864@} __attribute__((packed)); 4865struct bar @{ 4866 char z; 4867 struct foo f; 4868@}; 4869@end group 4870@end smallexample 4871 4872@item -Wpacked-bitfield-compat 4873@opindex Wpacked-bitfield-compat 4874@opindex Wno-packed-bitfield-compat 4875The 4.1, 4.2 and 4.3 series of GCC ignore the @code{packed} attribute 4876on bit-fields of type @code{char}. This has been fixed in GCC 4.4 but 4877the change can lead to differences in the structure layout. GCC 4878informs you when the offset of such a field has changed in GCC 4.4. 4879For example there is no longer a 4-bit padding between field @code{a} 4880and @code{b} in this structure: 4881 4882@smallexample 4883struct foo 4884@{ 4885 char a:4; 4886 char b:8; 4887@} __attribute__ ((packed)); 4888@end smallexample 4889 4890This warning is enabled by default. Use 4891@option{-Wno-packed-bitfield-compat} to disable this warning. 4892 4893@item -Wpadded 4894@opindex Wpadded 4895@opindex Wno-padded 4896Warn if padding is included in a structure, either to align an element 4897of the structure or to align the whole structure. Sometimes when this 4898happens it is possible to rearrange the fields of the structure to 4899reduce the padding and so make the structure smaller. 4900 4901@item -Wredundant-decls 4902@opindex Wredundant-decls 4903@opindex Wno-redundant-decls 4904Warn if anything is declared more than once in the same scope, even in 4905cases where multiple declaration is valid and changes nothing. 4906 4907@item -Wnested-externs @r{(C and Objective-C only)} 4908@opindex Wnested-externs 4909@opindex Wno-nested-externs 4910Warn if an @code{extern} declaration is encountered within a function. 4911 4912@item -Wno-inherited-variadic-ctor 4913@opindex Winherited-variadic-ctor 4914@opindex Wno-inherited-variadic-ctor 4915Suppress warnings about use of C++11 inheriting constructors when the 4916base class inherited from has a C variadic constructor; the warning is 4917on by default because the ellipsis is not inherited. 4918 4919@item -Winline 4920@opindex Winline 4921@opindex Wno-inline 4922Warn if a function that is declared as inline cannot be inlined. 4923Even with this option, the compiler does not warn about failures to 4924inline functions declared in system headers. 4925 4926The compiler uses a variety of heuristics to determine whether or not 4927to inline a function. For example, the compiler takes into account 4928the size of the function being inlined and the amount of inlining 4929that has already been done in the current function. Therefore, 4930seemingly insignificant changes in the source program can cause the 4931warnings produced by @option{-Winline} to appear or disappear. 4932 4933@item -Wno-invalid-offsetof @r{(C++ and Objective-C++ only)} 4934@opindex Wno-invalid-offsetof 4935@opindex Winvalid-offsetof 4936Suppress warnings from applying the @samp{offsetof} macro to a non-POD 4937type. According to the 1998 ISO C++ standard, applying @samp{offsetof} 4938to a non-POD type is undefined. In existing C++ implementations, 4939however, @samp{offsetof} typically gives meaningful results even when 4940applied to certain kinds of non-POD types (such as a simple 4941@samp{struct} that fails to be a POD type only by virtue of having a 4942constructor). This flag is for users who are aware that they are 4943writing nonportable code and who have deliberately chosen to ignore the 4944warning about it. 4945 4946The restrictions on @samp{offsetof} may be relaxed in a future version 4947of the C++ standard. 4948 4949@item -Wno-int-to-pointer-cast 4950@opindex Wno-int-to-pointer-cast 4951@opindex Wint-to-pointer-cast 4952Suppress warnings from casts to pointer type of an integer of a 4953different size. In C++, casting to a pointer type of smaller size is 4954an error. @option{Wint-to-pointer-cast} is enabled by default. 4955 4956 4957@item -Wno-pointer-to-int-cast @r{(C and Objective-C only)} 4958@opindex Wno-pointer-to-int-cast 4959@opindex Wpointer-to-int-cast 4960Suppress warnings from casts from a pointer to an integer type of a 4961different size. 4962 4963@item -Winvalid-pch 4964@opindex Winvalid-pch 4965@opindex Wno-invalid-pch 4966Warn if a precompiled header (@pxref{Precompiled Headers}) is found in 4967the search path but can't be used. 4968 4969@item -Wlong-long 4970@opindex Wlong-long 4971@opindex Wno-long-long 4972Warn if @samp{long long} type is used. This is enabled by either 4973@option{-Wpedantic} or @option{-Wtraditional} in ISO C90 and C++98 4974modes. To inhibit the warning messages, use @option{-Wno-long-long}. 4975 4976@item -Wvariadic-macros 4977@opindex Wvariadic-macros 4978@opindex Wno-variadic-macros 4979Warn if variadic macros are used in pedantic ISO C90 mode, or the GNU 4980alternate syntax when in pedantic ISO C99 mode. This is default. 4981To inhibit the warning messages, use @option{-Wno-variadic-macros}. 4982 4983@item -Wvarargs 4984@opindex Wvarargs 4985@opindex Wno-varargs 4986Warn upon questionable usage of the macros used to handle variable 4987arguments like @samp{va_start}. This is default. To inhibit the 4988warning messages, use @option{-Wno-varargs}. 4989 4990@item -Wvector-operation-performance 4991@opindex Wvector-operation-performance 4992@opindex Wno-vector-operation-performance 4993Warn if vector operation is not implemented via SIMD capabilities of the 4994architecture. Mainly useful for the performance tuning. 4995Vector operation can be implemented @code{piecewise}, which means that the 4996scalar operation is performed on every vector element; 4997@code{in parallel}, which means that the vector operation is implemented 4998using scalars of wider type, which normally is more performance efficient; 4999and @code{as a single scalar}, which means that vector fits into a 5000scalar type. 5001 5002@item -Wno-virtual-move-assign 5003@opindex Wvirtual-move-assign 5004@opindex Wno-virtual-move-assign 5005Suppress warnings about inheriting from a virtual base with a 5006non-trivial C++11 move assignment operator. This is dangerous because 5007if the virtual base is reachable along more than one path, it will be 5008moved multiple times, which can mean both objects end up in the 5009moved-from state. If the move assignment operator is written to avoid 5010moving from a moved-from object, this warning can be disabled. 5011 5012@item -Wvla 5013@opindex Wvla 5014@opindex Wno-vla 5015Warn if variable length array is used in the code. 5016@option{-Wno-vla} prevents the @option{-Wpedantic} warning of 5017the variable length array. 5018 5019@item -Wvolatile-register-var 5020@opindex Wvolatile-register-var 5021@opindex Wno-volatile-register-var 5022Warn if a register variable is declared volatile. The volatile 5023modifier does not inhibit all optimizations that may eliminate reads 5024and/or writes to register variables. This warning is enabled by 5025@option{-Wall}. 5026 5027@item -Wdisabled-optimization 5028@opindex Wdisabled-optimization 5029@opindex Wno-disabled-optimization 5030Warn if a requested optimization pass is disabled. This warning does 5031not generally indicate that there is anything wrong with your code; it 5032merely indicates that GCC's optimizers are unable to handle the code 5033effectively. Often, the problem is that your code is too big or too 5034complex; GCC refuses to optimize programs when the optimization 5035itself is likely to take inordinate amounts of time. 5036 5037@item -Wpointer-sign @r{(C and Objective-C only)} 5038@opindex Wpointer-sign 5039@opindex Wno-pointer-sign 5040Warn for pointer argument passing or assignment with different signedness. 5041This option is only supported for C and Objective-C@. It is implied by 5042@option{-Wall} and by @option{-Wpedantic}, which can be disabled with 5043@option{-Wno-pointer-sign}. 5044 5045@item -Wstack-protector 5046@opindex Wstack-protector 5047@opindex Wno-stack-protector 5048This option is only active when @option{-fstack-protector} is active. It 5049warns about functions that are not protected against stack smashing. 5050 5051@item -Woverlength-strings 5052@opindex Woverlength-strings 5053@opindex Wno-overlength-strings 5054Warn about string constants that are longer than the ``minimum 5055maximum'' length specified in the C standard. Modern compilers 5056generally allow string constants that are much longer than the 5057standard's minimum limit, but very portable programs should avoid 5058using longer strings. 5059 5060The limit applies @emph{after} string constant concatenation, and does 5061not count the trailing NUL@. In C90, the limit was 509 characters; in 5062C99, it was raised to 4095. C++98 does not specify a normative 5063minimum maximum, so we do not diagnose overlength strings in C++@. 5064 5065This option is implied by @option{-Wpedantic}, and can be disabled with 5066@option{-Wno-overlength-strings}. 5067 5068@item -Wunsuffixed-float-constants @r{(C and Objective-C only)} 5069@opindex Wunsuffixed-float-constants 5070 5071Issue a warning for any floating constant that does not have 5072a suffix. When used together with @option{-Wsystem-headers} it 5073warns about such constants in system header files. This can be useful 5074when preparing code to use with the @code{FLOAT_CONST_DECIMAL64} pragma 5075from the decimal floating-point extension to C99. 5076@end table 5077 5078@node Debugging Options 5079@section Options for Debugging Your Program or GCC 5080@cindex options, debugging 5081@cindex debugging information options 5082 5083GCC has various special options that are used for debugging 5084either your program or GCC: 5085 5086@table @gcctabopt 5087@item -g 5088@opindex g 5089Produce debugging information in the operating system's native format 5090(stabs, COFF, XCOFF, or DWARF 2)@. GDB can work with this debugging 5091information. 5092 5093On most systems that use stabs format, @option{-g} enables use of extra 5094debugging information that only GDB can use; this extra information 5095makes debugging work better in GDB but probably makes other debuggers 5096crash or 5097refuse to read the program. If you want to control for certain whether 5098to generate the extra information, use @option{-gstabs+}, @option{-gstabs}, 5099@option{-gxcoff+}, @option{-gxcoff}, or @option{-gvms} (see below). 5100 5101GCC allows you to use @option{-g} with 5102@option{-O}. The shortcuts taken by optimized code may occasionally 5103produce surprising results: some variables you declared may not exist 5104at all; flow of control may briefly move where you did not expect it; 5105some statements may not be executed because they compute constant 5106results or their values are already at hand; some statements may 5107execute in different places because they have been moved out of loops. 5108 5109Nevertheless it proves possible to debug optimized output. This makes 5110it reasonable to use the optimizer for programs that might have bugs. 5111 5112The following options are useful when GCC is generated with the 5113capability for more than one debugging format. 5114 5115@item -gsplit-dwarf 5116@opindex gsplit-dwarf 5117Separate as much dwarf debugging information as possible into a 5118separate output file with the extension .dwo. This option allows 5119the build system to avoid linking files with debug information. To 5120be useful, this option requires a debugger capable of reading .dwo 5121files. 5122 5123@item -ggdb 5124@opindex ggdb 5125Produce debugging information for use by GDB@. This means to use the 5126most expressive format available (DWARF 2, stabs, or the native format 5127if neither of those are supported), including GDB extensions if at all 5128possible. 5129 5130@item -gpubnames 5131@opindex gpubnames 5132Generate dwarf .debug_pubnames and .debug_pubtypes sections. 5133 5134@item -ggnu-pubnames 5135@opindex ggnu-pubnames 5136Generate .debug_pubnames and .debug_pubtypes sections in a format 5137suitable for conversion into a GDB@ index. This option is only useful 5138with a linker that can produce GDB@ index version 7. 5139 5140@item -gstabs 5141@opindex gstabs 5142Produce debugging information in stabs format (if that is supported), 5143without GDB extensions. This is the format used by DBX on most BSD 5144systems. On MIPS, Alpha and System V Release 4 systems this option 5145produces stabs debugging output that is not understood by DBX or SDB@. 5146On System V Release 4 systems this option requires the GNU assembler. 5147 5148@item -feliminate-unused-debug-symbols 5149@opindex feliminate-unused-debug-symbols 5150Produce debugging information in stabs format (if that is supported), 5151for only symbols that are actually used. 5152 5153@item -femit-class-debug-always 5154Instead of emitting debugging information for a C++ class in only one 5155object file, emit it in all object files using the class. This option 5156should be used only with debuggers that are unable to handle the way GCC 5157normally emits debugging information for classes because using this 5158option increases the size of debugging information by as much as a 5159factor of two. 5160 5161@item -fdebug-types-section 5162@opindex fdebug-types-section 5163@opindex fno-debug-types-section 5164When using DWARF Version 4 or higher, type DIEs can be put into 5165their own @code{.debug_types} section instead of making them part of the 5166@code{.debug_info} section. It is more efficient to put them in a separate 5167comdat sections since the linker can then remove duplicates. 5168But not all DWARF consumers support @code{.debug_types} sections yet 5169and on some objects @code{.debug_types} produces larger instead of smaller 5170debugging information. 5171 5172@item -gstabs+ 5173@opindex gstabs+ 5174Produce debugging information in stabs format (if that is supported), 5175using GNU extensions understood only by the GNU debugger (GDB)@. The 5176use of these extensions is likely to make other debuggers crash or 5177refuse to read the program. 5178 5179@item -gcoff 5180@opindex gcoff 5181Produce debugging information in COFF format (if that is supported). 5182This is the format used by SDB on most System V systems prior to 5183System V Release 4. 5184 5185@item -gxcoff 5186@opindex gxcoff 5187Produce debugging information in XCOFF format (if that is supported). 5188This is the format used by the DBX debugger on IBM RS/6000 systems. 5189 5190@item -gxcoff+ 5191@opindex gxcoff+ 5192Produce debugging information in XCOFF format (if that is supported), 5193using GNU extensions understood only by the GNU debugger (GDB)@. The 5194use of these extensions is likely to make other debuggers crash or 5195refuse to read the program, and may cause assemblers other than the GNU 5196assembler (GAS) to fail with an error. 5197 5198@item -gdwarf-@var{version} 5199@opindex gdwarf-@var{version} 5200Produce debugging information in DWARF format (if that is supported). 5201The value of @var{version} may be either 2, 3 or 4; the default version 5202for most targets is 4. 5203 5204Note that with DWARF Version 2, some ports require and always 5205use some non-conflicting DWARF 3 extensions in the unwind tables. 5206 5207Version 4 may require GDB 7.0 and @option{-fvar-tracking-assignments} 5208for maximum benefit. 5209 5210@item -grecord-gcc-switches 5211@opindex grecord-gcc-switches 5212This switch causes the command-line options used to invoke the 5213compiler that may affect code generation to be appended to the 5214DW_AT_producer attribute in DWARF debugging information. The options 5215are concatenated with spaces separating them from each other and from 5216the compiler version. See also @option{-frecord-gcc-switches} for another 5217way of storing compiler options into the object file. This is the default. 5218 5219@item -gno-record-gcc-switches 5220@opindex gno-record-gcc-switches 5221Disallow appending command-line options to the DW_AT_producer attribute 5222in DWARF debugging information. 5223 5224@item -gstrict-dwarf 5225@opindex gstrict-dwarf 5226Disallow using extensions of later DWARF standard version than selected 5227with @option{-gdwarf-@var{version}}. On most targets using non-conflicting 5228DWARF extensions from later standard versions is allowed. 5229 5230@item -gno-strict-dwarf 5231@opindex gno-strict-dwarf 5232Allow using extensions of later DWARF standard version than selected with 5233@option{-gdwarf-@var{version}}. 5234 5235@item -gvms 5236@opindex gvms 5237Produce debugging information in Alpha/VMS debug format (if that is 5238supported). This is the format used by DEBUG on Alpha/VMS systems. 5239 5240@item -g@var{level} 5241@itemx -ggdb@var{level} 5242@itemx -gstabs@var{level} 5243@itemx -gcoff@var{level} 5244@itemx -gxcoff@var{level} 5245@itemx -gvms@var{level} 5246Request debugging information and also use @var{level} to specify how 5247much information. The default level is 2. 5248 5249Level 0 produces no debug information at all. Thus, @option{-g0} negates 5250@option{-g}. 5251 5252Level 1 produces minimal information, enough for making backtraces in 5253parts of the program that you don't plan to debug. This includes 5254descriptions of functions and external variables, and line number 5255tables, but no information about local variables. 5256 5257Level 3 includes extra information, such as all the macro definitions 5258present in the program. Some debuggers support macro expansion when 5259you use @option{-g3}. 5260 5261@option{-gdwarf-2} does not accept a concatenated debug level, because 5262GCC used to support an option @option{-gdwarf} that meant to generate 5263debug information in version 1 of the DWARF format (which is very 5264different from version 2), and it would have been too confusing. That 5265debug format is long obsolete, but the option cannot be changed now. 5266Instead use an additional @option{-g@var{level}} option to change the 5267debug level for DWARF. 5268 5269@item -gtoggle 5270@opindex gtoggle 5271Turn off generation of debug info, if leaving out this option 5272generates it, or turn it on at level 2 otherwise. The position of this 5273argument in the command line does not matter; it takes effect after all 5274other options are processed, and it does so only once, no matter how 5275many times it is given. This is mainly intended to be used with 5276@option{-fcompare-debug}. 5277 5278@item -fsanitize=address 5279@opindex fsanitize=address 5280Enable AddressSanitizer, a fast memory error detector. 5281Memory access instructions will be instrumented to detect 5282out-of-bounds and use-after-free bugs. 5283See @uref{http://code.google.com/p/address-sanitizer/} for 5284more details. The run-time behavior can be influenced using the 5285@env{ASAN_OPTIONS} environment variable; see 5286@url{https://code.google.com/p/address-sanitizer/wiki/Flags#Run-time_flags} for 5287a list of supported options. 5288 5289@item -fsanitize=kernel-address 5290@opindex fsanitize=kernel-address 5291Enable AddressSanitizer for Linux kernel. 5292See @uref{http://code.google.com/p/address-sanitizer/wiki/AddressSanitizerForKernel} for more details. 5293 5294@item -fsanitize=thread 5295@opindex fsanitize=thread 5296Enable ThreadSanitizer, a fast data race detector. 5297Memory access instructions will be instrumented to detect 5298data race bugs. See @uref{http://code.google.com/p/thread-sanitizer/} for more 5299details. The run-time behavior can be influenced using the @env{TSAN_OPTIONS} 5300environment variable; see 5301@url{https://code.google.com/p/thread-sanitizer/wiki/Flags} for a list of 5302supported options. 5303 5304@item -fsanitize=leak 5305@opindex fsanitize=leak 5306Enable LeakSanitizer, a memory leak detector. 5307This option only matters for linking of executables and if neither 5308@option{-fsanitize=address} nor @option{-fsanitize=thread} is used. In that 5309case it will link the executable against a library that overrides @code{malloc} 5310and other allocator functions. See 5311@uref{https://code.google.com/p/address-sanitizer/wiki/LeakSanitizer} for more 5312details. The run-time behavior can be influenced using the 5313@env{LSAN_OPTIONS} environment variable. 5314 5315@item -fsanitize=undefined 5316@opindex fsanitize=undefined 5317Enable UndefinedBehaviorSanitizer, a fast undefined behavior detector. 5318Various computations will be instrumented to detect undefined behavior 5319at runtime. Current suboptions are: 5320 5321@table @gcctabopt 5322 5323@item -fsanitize=shift 5324@opindex fsanitize=shift 5325 5326This option enables checking that the result of a shift operation is 5327not undefined. Note that what exactly is considered undefined differs 5328slightly between C and C++, as well as between ISO C90 and C99, etc. 5329 5330@item -fsanitize=integer-divide-by-zero 5331@opindex fsanitize=integer-divide-by-zero 5332 5333Detect integer division by zero as well as @code{INT_MIN / -1} division. 5334 5335@item -fsanitize=unreachable 5336@opindex fsanitize=unreachable 5337 5338With this option, the compiler will turn the @code{__builtin_unreachable} 5339call into a diagnostics message call instead. When reaching the 5340@code{__builtin_unreachable} call, the behavior is undefined. 5341 5342@item -fsanitize=vla-bound 5343@opindex fsanitize=vla-bound 5344 5345This option instructs the compiler to check that the size of a variable 5346length array is positive. This option does not have any effect in 5347@option{-std=c++1y} mode, as the standard requires the exception be thrown 5348instead. 5349 5350@item -fsanitize=null 5351@opindex fsanitize=null 5352 5353This option enables pointer checking. Particularly, the application 5354built with this option turned on will issue an error message when it 5355tries to dereference a NULL pointer, or if a reference (possibly an 5356rvalue reference) is bound to a NULL pointer. 5357 5358@item -fsanitize=return 5359@opindex fsanitize=return 5360 5361This option enables return statement checking. Programs 5362built with this option turned on will issue an error message 5363when the end of a non-void function is reached without actually 5364returning a value. This option works in C++ only. 5365 5366@item -fsanitize=signed-integer-overflow 5367@opindex fsanitize=signed-integer-overflow 5368 5369This option enables signed integer overflow checking. We check that 5370the result of @code{+}, @code{*}, and both unary and binary @code{-} 5371does not overflow in the signed arithmetics. Note, integer promotion 5372rules must be taken into account. That is, the following is not an 5373overflow: 5374@smallexample 5375signed char a = SCHAR_MAX; 5376a++; 5377@end smallexample 5378 5379@end table 5380 5381While @option{-ftrapv} causes traps for signed overflows to be emitted, 5382@option{-fsanitize=undefined} gives a diagnostic message. 5383This currently works only for the C family of languages. 5384 5385@item -fdump-final-insns@r{[}=@var{file}@r{]} 5386@opindex fdump-final-insns 5387Dump the final internal representation (RTL) to @var{file}. If the 5388optional argument is omitted (or if @var{file} is @code{.}), the name 5389of the dump file is determined by appending @code{.gkd} to the 5390compilation output file name. 5391 5392@item -fcompare-debug@r{[}=@var{opts}@r{]} 5393@opindex fcompare-debug 5394@opindex fno-compare-debug 5395If no error occurs during compilation, run the compiler a second time, 5396adding @var{opts} and @option{-fcompare-debug-second} to the arguments 5397passed to the second compilation. Dump the final internal 5398representation in both compilations, and print an error if they differ. 5399 5400If the equal sign is omitted, the default @option{-gtoggle} is used. 5401 5402The environment variable @env{GCC_COMPARE_DEBUG}, if defined, non-empty 5403and nonzero, implicitly enables @option{-fcompare-debug}. If 5404@env{GCC_COMPARE_DEBUG} is defined to a string starting with a dash, 5405then it is used for @var{opts}, otherwise the default @option{-gtoggle} 5406is used. 5407 5408@option{-fcompare-debug=}, with the equal sign but without @var{opts}, 5409is equivalent to @option{-fno-compare-debug}, which disables the dumping 5410of the final representation and the second compilation, preventing even 5411@env{GCC_COMPARE_DEBUG} from taking effect. 5412 5413To verify full coverage during @option{-fcompare-debug} testing, set 5414@env{GCC_COMPARE_DEBUG} to say @samp{-fcompare-debug-not-overridden}, 5415which GCC rejects as an invalid option in any actual compilation 5416(rather than preprocessing, assembly or linking). To get just a 5417warning, setting @env{GCC_COMPARE_DEBUG} to @samp{-w%n-fcompare-debug 5418not overridden} will do. 5419 5420@item -fcompare-debug-second 5421@opindex fcompare-debug-second 5422This option is implicitly passed to the compiler for the second 5423compilation requested by @option{-fcompare-debug}, along with options to 5424silence warnings, and omitting other options that would cause 5425side-effect compiler outputs to files or to the standard output. Dump 5426files and preserved temporary files are renamed so as to contain the 5427@code{.gk} additional extension during the second compilation, to avoid 5428overwriting those generated by the first. 5429 5430When this option is passed to the compiler driver, it causes the 5431@emph{first} compilation to be skipped, which makes it useful for little 5432other than debugging the compiler proper. 5433 5434@item -feliminate-dwarf2-dups 5435@opindex feliminate-dwarf2-dups 5436Compress DWARF 2 debugging information by eliminating duplicated 5437information about each symbol. This option only makes sense when 5438generating DWARF 2 debugging information with @option{-gdwarf-2}. 5439 5440@item -femit-struct-debug-baseonly 5441@opindex femit-struct-debug-baseonly 5442Emit debug information for struct-like types 5443only when the base name of the compilation source file 5444matches the base name of file in which the struct is defined. 5445 5446This option substantially reduces the size of debugging information, 5447but at significant potential loss in type information to the debugger. 5448See @option{-femit-struct-debug-reduced} for a less aggressive option. 5449See @option{-femit-struct-debug-detailed} for more detailed control. 5450 5451This option works only with DWARF 2. 5452 5453@item -femit-struct-debug-reduced 5454@opindex femit-struct-debug-reduced 5455Emit debug information for struct-like types 5456only when the base name of the compilation source file 5457matches the base name of file in which the type is defined, 5458unless the struct is a template or defined in a system header. 5459 5460This option significantly reduces the size of debugging information, 5461with some potential loss in type information to the debugger. 5462See @option{-femit-struct-debug-baseonly} for a more aggressive option. 5463See @option{-femit-struct-debug-detailed} for more detailed control. 5464 5465This option works only with DWARF 2. 5466 5467@item -femit-struct-debug-detailed@r{[}=@var{spec-list}@r{]} 5468Specify the struct-like types 5469for which the compiler generates debug information. 5470The intent is to reduce duplicate struct debug information 5471between different object files within the same program. 5472 5473This option is a detailed version of 5474@option{-femit-struct-debug-reduced} and @option{-femit-struct-debug-baseonly}, 5475which serves for most needs. 5476 5477A specification has the syntax@* 5478[@samp{dir:}|@samp{ind:}][@samp{ord:}|@samp{gen:}](@samp{any}|@samp{sys}|@samp{base}|@samp{none}) 5479 5480The optional first word limits the specification to 5481structs that are used directly (@samp{dir:}) or used indirectly (@samp{ind:}). 5482A struct type is used directly when it is the type of a variable, member. 5483Indirect uses arise through pointers to structs. 5484That is, when use of an incomplete struct is valid, the use is indirect. 5485An example is 5486@samp{struct one direct; struct two * indirect;}. 5487 5488The optional second word limits the specification to 5489ordinary structs (@samp{ord:}) or generic structs (@samp{gen:}). 5490Generic structs are a bit complicated to explain. 5491For C++, these are non-explicit specializations of template classes, 5492or non-template classes within the above. 5493Other programming languages have generics, 5494but @option{-femit-struct-debug-detailed} does not yet implement them. 5495 5496The third word specifies the source files for those 5497structs for which the compiler should emit debug information. 5498The values @samp{none} and @samp{any} have the normal meaning. 5499The value @samp{base} means that 5500the base of name of the file in which the type declaration appears 5501must match the base of the name of the main compilation file. 5502In practice, this means that when compiling @file{foo.c}, debug information 5503is generated for types declared in that file and @file{foo.h}, 5504but not other header files. 5505The value @samp{sys} means those types satisfying @samp{base} 5506or declared in system or compiler headers. 5507 5508You may need to experiment to determine the best settings for your application. 5509 5510The default is @option{-femit-struct-debug-detailed=all}. 5511 5512This option works only with DWARF 2. 5513 5514@item -fno-merge-debug-strings 5515@opindex fmerge-debug-strings 5516@opindex fno-merge-debug-strings 5517Direct the linker to not merge together strings in the debugging 5518information that are identical in different object files. Merging is 5519not supported by all assemblers or linkers. Merging decreases the size 5520of the debug information in the output file at the cost of increasing 5521link processing time. Merging is enabled by default. 5522 5523@item -fdebug-prefix-map=@var{old}=@var{new} 5524@opindex fdebug-prefix-map 5525When compiling files in directory @file{@var{old}}, record debugging 5526information describing them as in @file{@var{new}} instead. 5527 5528@item -fno-dwarf2-cfi-asm 5529@opindex fdwarf2-cfi-asm 5530@opindex fno-dwarf2-cfi-asm 5531Emit DWARF 2 unwind info as compiler generated @code{.eh_frame} section 5532instead of using GAS @code{.cfi_*} directives. 5533 5534@cindex @command{prof} 5535@item -p 5536@opindex p 5537Generate extra code to write profile information suitable for the 5538analysis program @command{prof}. You must use this option when compiling 5539the source files you want data about, and you must also use it when 5540linking. 5541 5542@cindex @command{gprof} 5543@item -pg 5544@opindex pg 5545Generate extra code to write profile information suitable for the 5546analysis program @command{gprof}. You must use this option when compiling 5547the source files you want data about, and you must also use it when 5548linking. 5549 5550@item -Q 5551@opindex Q 5552Makes the compiler print out each function name as it is compiled, and 5553print some statistics about each pass when it finishes. 5554 5555@item -ftime-report 5556@opindex ftime-report 5557Makes the compiler print some statistics about the time consumed by each 5558pass when it finishes. 5559 5560@item -fmem-report 5561@opindex fmem-report 5562Makes the compiler print some statistics about permanent memory 5563allocation when it finishes. 5564 5565@item -fmem-report-wpa 5566@opindex fmem-report-wpa 5567Makes the compiler print some statistics about permanent memory 5568allocation for the WPA phase only. 5569 5570@item -fpre-ipa-mem-report 5571@opindex fpre-ipa-mem-report 5572@item -fpost-ipa-mem-report 5573@opindex fpost-ipa-mem-report 5574Makes the compiler print some statistics about permanent memory 5575allocation before or after interprocedural optimization. 5576 5577@item -fprofile-report 5578@opindex fprofile-report 5579Makes the compiler print some statistics about consistency of the 5580(estimated) profile and effect of individual passes. 5581 5582@item -fstack-usage 5583@opindex fstack-usage 5584Makes the compiler output stack usage information for the program, on a 5585per-function basis. The filename for the dump is made by appending 5586@file{.su} to the @var{auxname}. @var{auxname} is generated from the name of 5587the output file, if explicitly specified and it is not an executable, 5588otherwise it is the basename of the source file. An entry is made up 5589of three fields: 5590 5591@itemize 5592@item 5593The name of the function. 5594@item 5595A number of bytes. 5596@item 5597One or more qualifiers: @code{static}, @code{dynamic}, @code{bounded}. 5598@end itemize 5599 5600The qualifier @code{static} means that the function manipulates the stack 5601statically: a fixed number of bytes are allocated for the frame on function 5602entry and released on function exit; no stack adjustments are otherwise made 5603in the function. The second field is this fixed number of bytes. 5604 5605The qualifier @code{dynamic} means that the function manipulates the stack 5606dynamically: in addition to the static allocation described above, stack 5607adjustments are made in the body of the function, for example to push/pop 5608arguments around function calls. If the qualifier @code{bounded} is also 5609present, the amount of these adjustments is bounded at compile time and 5610the second field is an upper bound of the total amount of stack used by 5611the function. If it is not present, the amount of these adjustments is 5612not bounded at compile time and the second field only represents the 5613bounded part. 5614 5615@item -fprofile-arcs 5616@opindex fprofile-arcs 5617Add code so that program flow @dfn{arcs} are instrumented. During 5618execution the program records how many times each branch and call is 5619executed and how many times it is taken or returns. When the compiled 5620program exits it saves this data to a file called 5621@file{@var{auxname}.gcda} for each source file. The data may be used for 5622profile-directed optimizations (@option{-fbranch-probabilities}), or for 5623test coverage analysis (@option{-ftest-coverage}). Each object file's 5624@var{auxname} is generated from the name of the output file, if 5625explicitly specified and it is not the final executable, otherwise it is 5626the basename of the source file. In both cases any suffix is removed 5627(e.g.@: @file{foo.gcda} for input file @file{dir/foo.c}, or 5628@file{dir/foo.gcda} for output file specified as @option{-o dir/foo.o}). 5629@xref{Cross-profiling}. 5630 5631@cindex @command{gcov} 5632@item --coverage 5633@opindex coverage 5634 5635This option is used to compile and link code instrumented for coverage 5636analysis. The option is a synonym for @option{-fprofile-arcs} 5637@option{-ftest-coverage} (when compiling) and @option{-lgcov} (when 5638linking). See the documentation for those options for more details. 5639 5640@itemize 5641 5642@item 5643Compile the source files with @option{-fprofile-arcs} plus optimization 5644and code generation options. For test coverage analysis, use the 5645additional @option{-ftest-coverage} option. You do not need to profile 5646every source file in a program. 5647 5648@item 5649Link your object files with @option{-lgcov} or @option{-fprofile-arcs} 5650(the latter implies the former). 5651 5652@item 5653Run the program on a representative workload to generate the arc profile 5654information. This may be repeated any number of times. You can run 5655concurrent instances of your program, and provided that the file system 5656supports locking, the data files will be correctly updated. Also 5657@code{fork} calls are detected and correctly handled (double counting 5658will not happen). 5659 5660@item 5661For profile-directed optimizations, compile the source files again with 5662the same optimization and code generation options plus 5663@option{-fbranch-probabilities} (@pxref{Optimize Options,,Options that 5664Control Optimization}). 5665 5666@item 5667For test coverage analysis, use @command{gcov} to produce human readable 5668information from the @file{.gcno} and @file{.gcda} files. Refer to the 5669@command{gcov} documentation for further information. 5670 5671@end itemize 5672 5673With @option{-fprofile-arcs}, for each function of your program GCC 5674creates a program flow graph, then finds a spanning tree for the graph. 5675Only arcs that are not on the spanning tree have to be instrumented: the 5676compiler adds code to count the number of times that these arcs are 5677executed. When an arc is the only exit or only entrance to a block, the 5678instrumentation code can be added to the block; otherwise, a new basic 5679block must be created to hold the instrumentation code. 5680 5681@need 2000 5682@item -ftest-coverage 5683@opindex ftest-coverage 5684Produce a notes file that the @command{gcov} code-coverage utility 5685(@pxref{Gcov,, @command{gcov}---a Test Coverage Program}) can use to 5686show program coverage. Each source file's note file is called 5687@file{@var{auxname}.gcno}. Refer to the @option{-fprofile-arcs} option 5688above for a description of @var{auxname} and instructions on how to 5689generate test coverage data. Coverage data matches the source files 5690more closely if you do not optimize. 5691 5692@item -fdbg-cnt-list 5693@opindex fdbg-cnt-list 5694Print the name and the counter upper bound for all debug counters. 5695 5696 5697@item -fdbg-cnt=@var{counter-value-list} 5698@opindex fdbg-cnt 5699Set the internal debug counter upper bound. @var{counter-value-list} 5700is a comma-separated list of @var{name}:@var{value} pairs 5701which sets the upper bound of each debug counter @var{name} to @var{value}. 5702All debug counters have the initial upper bound of @code{UINT_MAX}; 5703thus @code{dbg_cnt()} returns true always unless the upper bound 5704is set by this option. 5705For example, with @option{-fdbg-cnt=dce:10,tail_call:0}, 5706@code{dbg_cnt(dce)} returns true only for first 10 invocations. 5707 5708@item -fenable-@var{kind}-@var{pass} 5709@itemx -fdisable-@var{kind}-@var{pass}=@var{range-list} 5710@opindex fdisable- 5711@opindex fenable- 5712 5713This is a set of options that are used to explicitly disable/enable 5714optimization passes. These options are intended for use for debugging GCC. 5715Compiler users should use regular options for enabling/disabling 5716passes instead. 5717 5718@table @gcctabopt 5719 5720@item -fdisable-ipa-@var{pass} 5721Disable IPA pass @var{pass}. @var{pass} is the pass name. If the same pass is 5722statically invoked in the compiler multiple times, the pass name should be 5723appended with a sequential number starting from 1. 5724 5725@item -fdisable-rtl-@var{pass} 5726@itemx -fdisable-rtl-@var{pass}=@var{range-list} 5727Disable RTL pass @var{pass}. @var{pass} is the pass name. If the same pass is 5728statically invoked in the compiler multiple times, the pass name should be 5729appended with a sequential number starting from 1. @var{range-list} is a 5730comma-separated list of function ranges or assembler names. Each range is a number 5731pair separated by a colon. The range is inclusive in both ends. If the range 5732is trivial, the number pair can be simplified as a single number. If the 5733function's call graph node's @var{uid} falls within one of the specified ranges, 5734the @var{pass} is disabled for that function. The @var{uid} is shown in the 5735function header of a dump file, and the pass names can be dumped by using 5736option @option{-fdump-passes}. 5737 5738@item -fdisable-tree-@var{pass} 5739@itemx -fdisable-tree-@var{pass}=@var{range-list} 5740Disable tree pass @var{pass}. See @option{-fdisable-rtl} for the description of 5741option arguments. 5742 5743@item -fenable-ipa-@var{pass} 5744Enable IPA pass @var{pass}. @var{pass} is the pass name. If the same pass is 5745statically invoked in the compiler multiple times, the pass name should be 5746appended with a sequential number starting from 1. 5747 5748@item -fenable-rtl-@var{pass} 5749@itemx -fenable-rtl-@var{pass}=@var{range-list} 5750Enable RTL pass @var{pass}. See @option{-fdisable-rtl} for option argument 5751description and examples. 5752 5753@item -fenable-tree-@var{pass} 5754@itemx -fenable-tree-@var{pass}=@var{range-list} 5755Enable tree pass @var{pass}. See @option{-fdisable-rtl} for the description 5756of option arguments. 5757 5758@end table 5759 5760Here are some examples showing uses of these options. 5761 5762@smallexample 5763 5764# disable ccp1 for all functions 5765 -fdisable-tree-ccp1 5766# disable complete unroll for function whose cgraph node uid is 1 5767 -fenable-tree-cunroll=1 5768# disable gcse2 for functions at the following ranges [1,1], 5769# [300,400], and [400,1000] 5770# disable gcse2 for functions foo and foo2 5771 -fdisable-rtl-gcse2=foo,foo2 5772# disable early inlining 5773 -fdisable-tree-einline 5774# disable ipa inlining 5775 -fdisable-ipa-inline 5776# enable tree full unroll 5777 -fenable-tree-unroll 5778 5779@end smallexample 5780 5781@item -d@var{letters} 5782@itemx -fdump-rtl-@var{pass} 5783@itemx -fdump-rtl-@var{pass}=@var{filename} 5784@opindex d 5785@opindex fdump-rtl-@var{pass} 5786Says to make debugging dumps during compilation at times specified by 5787@var{letters}. This is used for debugging the RTL-based passes of the 5788compiler. The file names for most of the dumps are made by appending 5789a pass number and a word to the @var{dumpname}, and the files are 5790created in the directory of the output file. In case of 5791@option{=@var{filename}} option, the dump is output on the given file 5792instead of the pass numbered dump files. Note that the pass number is 5793computed statically as passes get registered into the pass manager. 5794Thus the numbering is not related to the dynamic order of execution of 5795passes. In particular, a pass installed by a plugin could have a 5796number over 200 even if it executed quite early. @var{dumpname} is 5797generated from the name of the output file, if explicitly specified 5798and it is not an executable, otherwise it is the basename of the 5799source file. These switches may have different effects when 5800@option{-E} is used for preprocessing. 5801 5802Debug dumps can be enabled with a @option{-fdump-rtl} switch or some 5803@option{-d} option @var{letters}. Here are the possible 5804letters for use in @var{pass} and @var{letters}, and their meanings: 5805 5806@table @gcctabopt 5807 5808@item -fdump-rtl-alignments 5809@opindex fdump-rtl-alignments 5810Dump after branch alignments have been computed. 5811 5812@item -fdump-rtl-asmcons 5813@opindex fdump-rtl-asmcons 5814Dump after fixing rtl statements that have unsatisfied in/out constraints. 5815 5816@item -fdump-rtl-auto_inc_dec 5817@opindex fdump-rtl-auto_inc_dec 5818Dump after auto-inc-dec discovery. This pass is only run on 5819architectures that have auto inc or auto dec instructions. 5820 5821@item -fdump-rtl-barriers 5822@opindex fdump-rtl-barriers 5823Dump after cleaning up the barrier instructions. 5824 5825@item -fdump-rtl-bbpart 5826@opindex fdump-rtl-bbpart 5827Dump after partitioning hot and cold basic blocks. 5828 5829@item -fdump-rtl-bbro 5830@opindex fdump-rtl-bbro 5831Dump after block reordering. 5832 5833@item -fdump-rtl-btl1 5834@itemx -fdump-rtl-btl2 5835@opindex fdump-rtl-btl2 5836@opindex fdump-rtl-btl2 5837@option{-fdump-rtl-btl1} and @option{-fdump-rtl-btl2} enable dumping 5838after the two branch 5839target load optimization passes. 5840 5841@item -fdump-rtl-bypass 5842@opindex fdump-rtl-bypass 5843Dump after jump bypassing and control flow optimizations. 5844 5845@item -fdump-rtl-combine 5846@opindex fdump-rtl-combine 5847Dump after the RTL instruction combination pass. 5848 5849@item -fdump-rtl-compgotos 5850@opindex fdump-rtl-compgotos 5851Dump after duplicating the computed gotos. 5852 5853@item -fdump-rtl-ce1 5854@itemx -fdump-rtl-ce2 5855@itemx -fdump-rtl-ce3 5856@opindex fdump-rtl-ce1 5857@opindex fdump-rtl-ce2 5858@opindex fdump-rtl-ce3 5859@option{-fdump-rtl-ce1}, @option{-fdump-rtl-ce2}, and 5860@option{-fdump-rtl-ce3} enable dumping after the three 5861if conversion passes. 5862 5863@item -fdump-rtl-cprop_hardreg 5864@opindex fdump-rtl-cprop_hardreg 5865Dump after hard register copy propagation. 5866 5867@item -fdump-rtl-csa 5868@opindex fdump-rtl-csa 5869Dump after combining stack adjustments. 5870 5871@item -fdump-rtl-cse1 5872@itemx -fdump-rtl-cse2 5873@opindex fdump-rtl-cse1 5874@opindex fdump-rtl-cse2 5875@option{-fdump-rtl-cse1} and @option{-fdump-rtl-cse2} enable dumping after 5876the two common subexpression elimination passes. 5877 5878@item -fdump-rtl-dce 5879@opindex fdump-rtl-dce 5880Dump after the standalone dead code elimination passes. 5881 5882@item -fdump-rtl-dbr 5883@opindex fdump-rtl-dbr 5884Dump after delayed branch scheduling. 5885 5886@item -fdump-rtl-dce1 5887@itemx -fdump-rtl-dce2 5888@opindex fdump-rtl-dce1 5889@opindex fdump-rtl-dce2 5890@option{-fdump-rtl-dce1} and @option{-fdump-rtl-dce2} enable dumping after 5891the two dead store elimination passes. 5892 5893@item -fdump-rtl-eh 5894@opindex fdump-rtl-eh 5895Dump after finalization of EH handling code. 5896 5897@item -fdump-rtl-eh_ranges 5898@opindex fdump-rtl-eh_ranges 5899Dump after conversion of EH handling range regions. 5900 5901@item -fdump-rtl-expand 5902@opindex fdump-rtl-expand 5903Dump after RTL generation. 5904 5905@item -fdump-rtl-fwprop1 5906@itemx -fdump-rtl-fwprop2 5907@opindex fdump-rtl-fwprop1 5908@opindex fdump-rtl-fwprop2 5909@option{-fdump-rtl-fwprop1} and @option{-fdump-rtl-fwprop2} enable 5910dumping after the two forward propagation passes. 5911 5912@item -fdump-rtl-gcse1 5913@itemx -fdump-rtl-gcse2 5914@opindex fdump-rtl-gcse1 5915@opindex fdump-rtl-gcse2 5916@option{-fdump-rtl-gcse1} and @option{-fdump-rtl-gcse2} enable dumping 5917after global common subexpression elimination. 5918 5919@item -fdump-rtl-init-regs 5920@opindex fdump-rtl-init-regs 5921Dump after the initialization of the registers. 5922 5923@item -fdump-rtl-initvals 5924@opindex fdump-rtl-initvals 5925Dump after the computation of the initial value sets. 5926 5927@item -fdump-rtl-into_cfglayout 5928@opindex fdump-rtl-into_cfglayout 5929Dump after converting to cfglayout mode. 5930 5931@item -fdump-rtl-ira 5932@opindex fdump-rtl-ira 5933Dump after iterated register allocation. 5934 5935@item -fdump-rtl-jump 5936@opindex fdump-rtl-jump 5937Dump after the second jump optimization. 5938 5939@item -fdump-rtl-loop2 5940@opindex fdump-rtl-loop2 5941@option{-fdump-rtl-loop2} enables dumping after the rtl 5942loop optimization passes. 5943 5944@item -fdump-rtl-mach 5945@opindex fdump-rtl-mach 5946Dump after performing the machine dependent reorganization pass, if that 5947pass exists. 5948 5949@item -fdump-rtl-mode_sw 5950@opindex fdump-rtl-mode_sw 5951Dump after removing redundant mode switches. 5952 5953@item -fdump-rtl-rnreg 5954@opindex fdump-rtl-rnreg 5955Dump after register renumbering. 5956 5957@item -fdump-rtl-outof_cfglayout 5958@opindex fdump-rtl-outof_cfglayout 5959Dump after converting from cfglayout mode. 5960 5961@item -fdump-rtl-peephole2 5962@opindex fdump-rtl-peephole2 5963Dump after the peephole pass. 5964 5965@item -fdump-rtl-postreload 5966@opindex fdump-rtl-postreload 5967Dump after post-reload optimizations. 5968 5969@item -fdump-rtl-pro_and_epilogue 5970@opindex fdump-rtl-pro_and_epilogue 5971Dump after generating the function prologues and epilogues. 5972 5973@item -fdump-rtl-sched1 5974@itemx -fdump-rtl-sched2 5975@opindex fdump-rtl-sched1 5976@opindex fdump-rtl-sched2 5977@option{-fdump-rtl-sched1} and @option{-fdump-rtl-sched2} enable dumping 5978after the basic block scheduling passes. 5979 5980@item -fdump-rtl-ree 5981@opindex fdump-rtl-ree 5982Dump after sign/zero extension elimination. 5983 5984@item -fdump-rtl-seqabstr 5985@opindex fdump-rtl-seqabstr 5986Dump after common sequence discovery. 5987 5988@item -fdump-rtl-shorten 5989@opindex fdump-rtl-shorten 5990Dump after shortening branches. 5991 5992@item -fdump-rtl-sibling 5993@opindex fdump-rtl-sibling 5994Dump after sibling call optimizations. 5995 5996@item -fdump-rtl-split1 5997@itemx -fdump-rtl-split2 5998@itemx -fdump-rtl-split3 5999@itemx -fdump-rtl-split4 6000@itemx -fdump-rtl-split5 6001@opindex fdump-rtl-split1 6002@opindex fdump-rtl-split2 6003@opindex fdump-rtl-split3 6004@opindex fdump-rtl-split4 6005@opindex fdump-rtl-split5 6006@option{-fdump-rtl-split1}, @option{-fdump-rtl-split2}, 6007@option{-fdump-rtl-split3}, @option{-fdump-rtl-split4} and 6008@option{-fdump-rtl-split5} enable dumping after five rounds of 6009instruction splitting. 6010 6011@item -fdump-rtl-sms 6012@opindex fdump-rtl-sms 6013Dump after modulo scheduling. This pass is only run on some 6014architectures. 6015 6016@item -fdump-rtl-stack 6017@opindex fdump-rtl-stack 6018Dump after conversion from GCC's ``flat register file'' registers to the 6019x87's stack-like registers. This pass is only run on x86 variants. 6020 6021@item -fdump-rtl-subreg1 6022@itemx -fdump-rtl-subreg2 6023@opindex fdump-rtl-subreg1 6024@opindex fdump-rtl-subreg2 6025@option{-fdump-rtl-subreg1} and @option{-fdump-rtl-subreg2} enable dumping after 6026the two subreg expansion passes. 6027 6028@item -fdump-rtl-unshare 6029@opindex fdump-rtl-unshare 6030Dump after all rtl has been unshared. 6031 6032@item -fdump-rtl-vartrack 6033@opindex fdump-rtl-vartrack 6034Dump after variable tracking. 6035 6036@item -fdump-rtl-vregs 6037@opindex fdump-rtl-vregs 6038Dump after converting virtual registers to hard registers. 6039 6040@item -fdump-rtl-web 6041@opindex fdump-rtl-web 6042Dump after live range splitting. 6043 6044@item -fdump-rtl-regclass 6045@itemx -fdump-rtl-subregs_of_mode_init 6046@itemx -fdump-rtl-subregs_of_mode_finish 6047@itemx -fdump-rtl-dfinit 6048@itemx -fdump-rtl-dfinish 6049@opindex fdump-rtl-regclass 6050@opindex fdump-rtl-subregs_of_mode_init 6051@opindex fdump-rtl-subregs_of_mode_finish 6052@opindex fdump-rtl-dfinit 6053@opindex fdump-rtl-dfinish 6054These dumps are defined but always produce empty files. 6055 6056@item -da 6057@itemx -fdump-rtl-all 6058@opindex da 6059@opindex fdump-rtl-all 6060Produce all the dumps listed above. 6061 6062@item -dA 6063@opindex dA 6064Annotate the assembler output with miscellaneous debugging information. 6065 6066@item -dD 6067@opindex dD 6068Dump all macro definitions, at the end of preprocessing, in addition to 6069normal output. 6070 6071@item -dH 6072@opindex dH 6073Produce a core dump whenever an error occurs. 6074 6075@item -dp 6076@opindex dp 6077Annotate the assembler output with a comment indicating which 6078pattern and alternative is used. The length of each instruction is 6079also printed. 6080 6081@item -dP 6082@opindex dP 6083Dump the RTL in the assembler output as a comment before each instruction. 6084Also turns on @option{-dp} annotation. 6085 6086@item -dx 6087@opindex dx 6088Just generate RTL for a function instead of compiling it. Usually used 6089with @option{-fdump-rtl-expand}. 6090@end table 6091 6092@item -fdump-noaddr 6093@opindex fdump-noaddr 6094When doing debugging dumps, suppress address output. This makes it more 6095feasible to use diff on debugging dumps for compiler invocations with 6096different compiler binaries and/or different 6097text / bss / data / heap / stack / dso start locations. 6098 6099@item -fdump-unnumbered 6100@opindex fdump-unnumbered 6101When doing debugging dumps, suppress instruction numbers and address output. 6102This makes it more feasible to use diff on debugging dumps for compiler 6103invocations with different options, in particular with and without 6104@option{-g}. 6105 6106@item -fdump-unnumbered-links 6107@opindex fdump-unnumbered-links 6108When doing debugging dumps (see @option{-d} option above), suppress 6109instruction numbers for the links to the previous and next instructions 6110in a sequence. 6111 6112@item -fdump-translation-unit @r{(C++ only)} 6113@itemx -fdump-translation-unit-@var{options} @r{(C++ only)} 6114@opindex fdump-translation-unit 6115Dump a representation of the tree structure for the entire translation 6116unit to a file. The file name is made by appending @file{.tu} to the 6117source file name, and the file is created in the same directory as the 6118output file. If the @samp{-@var{options}} form is used, @var{options} 6119controls the details of the dump as described for the 6120@option{-fdump-tree} options. 6121 6122@item -fdump-class-hierarchy @r{(C++ only)} 6123@itemx -fdump-class-hierarchy-@var{options} @r{(C++ only)} 6124@opindex fdump-class-hierarchy 6125Dump a representation of each class's hierarchy and virtual function 6126table layout to a file. The file name is made by appending 6127@file{.class} to the source file name, and the file is created in the 6128same directory as the output file. If the @samp{-@var{options}} form 6129is used, @var{options} controls the details of the dump as described 6130for the @option{-fdump-tree} options. 6131 6132@item -fdump-ipa-@var{switch} 6133@opindex fdump-ipa 6134Control the dumping at various stages of inter-procedural analysis 6135language tree to a file. The file name is generated by appending a 6136switch specific suffix to the source file name, and the file is created 6137in the same directory as the output file. The following dumps are 6138possible: 6139 6140@table @samp 6141@item all 6142Enables all inter-procedural analysis dumps. 6143 6144@item cgraph 6145Dumps information about call-graph optimization, unused function removal, 6146and inlining decisions. 6147 6148@item inline 6149Dump after function inlining. 6150 6151@end table 6152 6153@item -fdump-passes 6154@opindex fdump-passes 6155Dump the list of optimization passes that are turned on and off by 6156the current command-line options. 6157 6158@item -fdump-statistics-@var{option} 6159@opindex fdump-statistics 6160Enable and control dumping of pass statistics in a separate file. The 6161file name is generated by appending a suffix ending in 6162@samp{.statistics} to the source file name, and the file is created in 6163the same directory as the output file. If the @samp{-@var{option}} 6164form is used, @samp{-stats} causes counters to be summed over the 6165whole compilation unit while @samp{-details} dumps every event as 6166the passes generate them. The default with no option is to sum 6167counters for each function compiled. 6168 6169@item -fdump-tree-@var{switch} 6170@itemx -fdump-tree-@var{switch}-@var{options} 6171@itemx -fdump-tree-@var{switch}-@var{options}=@var{filename} 6172@opindex fdump-tree 6173Control the dumping at various stages of processing the intermediate 6174language tree to a file. The file name is generated by appending a 6175switch-specific suffix to the source file name, and the file is 6176created in the same directory as the output file. In case of 6177@option{=@var{filename}} option, the dump is output on the given file 6178instead of the auto named dump files. If the @samp{-@var{options}} 6179form is used, @var{options} is a list of @samp{-} separated options 6180which control the details of the dump. Not all options are applicable 6181to all dumps; those that are not meaningful are ignored. The 6182following options are available 6183 6184@table @samp 6185@item address 6186Print the address of each node. Usually this is not meaningful as it 6187changes according to the environment and source file. Its primary use 6188is for tying up a dump file with a debug environment. 6189@item asmname 6190If @code{DECL_ASSEMBLER_NAME} has been set for a given decl, use that 6191in the dump instead of @code{DECL_NAME}. Its primary use is ease of 6192use working backward from mangled names in the assembly file. 6193@item slim 6194When dumping front-end intermediate representations, inhibit dumping 6195of members of a scope or body of a function merely because that scope 6196has been reached. Only dump such items when they are directly reachable 6197by some other path. 6198 6199When dumping pretty-printed trees, this option inhibits dumping the 6200bodies of control structures. 6201 6202When dumping RTL, print the RTL in slim (condensed) form instead of 6203the default LISP-like representation. 6204@item raw 6205Print a raw representation of the tree. By default, trees are 6206pretty-printed into a C-like representation. 6207@item details 6208Enable more detailed dumps (not honored by every dump option). Also 6209include information from the optimization passes. 6210@item stats 6211Enable dumping various statistics about the pass (not honored by every dump 6212option). 6213@item blocks 6214Enable showing basic block boundaries (disabled in raw dumps). 6215@item graph 6216For each of the other indicated dump files (@option{-fdump-rtl-@var{pass}}), 6217dump a representation of the control flow graph suitable for viewing with 6218GraphViz to @file{@var{file}.@var{passid}.@var{pass}.dot}. Each function in 6219the file is pretty-printed as a subgraph, so that GraphViz can render them 6220all in a single plot. 6221 6222This option currently only works for RTL dumps, and the RTL is always 6223dumped in slim form. 6224@item vops 6225Enable showing virtual operands for every statement. 6226@item lineno 6227Enable showing line numbers for statements. 6228@item uid 6229Enable showing the unique ID (@code{DECL_UID}) for each variable. 6230@item verbose 6231Enable showing the tree dump for each statement. 6232@item eh 6233Enable showing the EH region number holding each statement. 6234@item scev 6235Enable showing scalar evolution analysis details. 6236@item optimized 6237Enable showing optimization information (only available in certain 6238passes). 6239@item missed 6240Enable showing missed optimization information (only available in certain 6241passes). 6242@item notes 6243Enable other detailed optimization information (only available in 6244certain passes). 6245@item =@var{filename} 6246Instead of an auto named dump file, output into the given file 6247name. The file names @file{stdout} and @file{stderr} are treated 6248specially and are considered already open standard streams. For 6249example, 6250 6251@smallexample 6252gcc -O2 -ftree-vectorize -fdump-tree-vect-blocks=foo.dump 6253 -fdump-tree-pre=stderr file.c 6254@end smallexample 6255 6256outputs vectorizer dump into @file{foo.dump}, while the PRE dump is 6257output on to @file{stderr}. If two conflicting dump filenames are 6258given for the same pass, then the latter option overrides the earlier 6259one. 6260 6261@item all 6262Turn on all options, except @option{raw}, @option{slim}, @option{verbose} 6263and @option{lineno}. 6264 6265@item optall 6266Turn on all optimization options, i.e., @option{optimized}, 6267@option{missed}, and @option{note}. 6268@end table 6269 6270The following tree dumps are possible: 6271@table @samp 6272 6273@item original 6274@opindex fdump-tree-original 6275Dump before any tree based optimization, to @file{@var{file}.original}. 6276 6277@item optimized 6278@opindex fdump-tree-optimized 6279Dump after all tree based optimization, to @file{@var{file}.optimized}. 6280 6281@item gimple 6282@opindex fdump-tree-gimple 6283Dump each function before and after the gimplification pass to a file. The 6284file name is made by appending @file{.gimple} to the source file name. 6285 6286@item cfg 6287@opindex fdump-tree-cfg 6288Dump the control flow graph of each function to a file. The file name is 6289made by appending @file{.cfg} to the source file name. 6290 6291@item ch 6292@opindex fdump-tree-ch 6293Dump each function after copying loop headers. The file name is made by 6294appending @file{.ch} to the source file name. 6295 6296@item ssa 6297@opindex fdump-tree-ssa 6298Dump SSA related information to a file. The file name is made by appending 6299@file{.ssa} to the source file name. 6300 6301@item alias 6302@opindex fdump-tree-alias 6303Dump aliasing information for each function. The file name is made by 6304appending @file{.alias} to the source file name. 6305 6306@item ccp 6307@opindex fdump-tree-ccp 6308Dump each function after CCP@. The file name is made by appending 6309@file{.ccp} to the source file name. 6310 6311@item storeccp 6312@opindex fdump-tree-storeccp 6313Dump each function after STORE-CCP@. The file name is made by appending 6314@file{.storeccp} to the source file name. 6315 6316@item pre 6317@opindex fdump-tree-pre 6318Dump trees after partial redundancy elimination. The file name is made 6319by appending @file{.pre} to the source file name. 6320 6321@item fre 6322@opindex fdump-tree-fre 6323Dump trees after full redundancy elimination. The file name is made 6324by appending @file{.fre} to the source file name. 6325 6326@item copyprop 6327@opindex fdump-tree-copyprop 6328Dump trees after copy propagation. The file name is made 6329by appending @file{.copyprop} to the source file name. 6330 6331@item store_copyprop 6332@opindex fdump-tree-store_copyprop 6333Dump trees after store copy-propagation. The file name is made 6334by appending @file{.store_copyprop} to the source file name. 6335 6336@item dce 6337@opindex fdump-tree-dce 6338Dump each function after dead code elimination. The file name is made by 6339appending @file{.dce} to the source file name. 6340 6341@item sra 6342@opindex fdump-tree-sra 6343Dump each function after performing scalar replacement of aggregates. The 6344file name is made by appending @file{.sra} to the source file name. 6345 6346@item sink 6347@opindex fdump-tree-sink 6348Dump each function after performing code sinking. The file name is made 6349by appending @file{.sink} to the source file name. 6350 6351@item dom 6352@opindex fdump-tree-dom 6353Dump each function after applying dominator tree optimizations. The file 6354name is made by appending @file{.dom} to the source file name. 6355 6356@item dse 6357@opindex fdump-tree-dse 6358Dump each function after applying dead store elimination. The file 6359name is made by appending @file{.dse} to the source file name. 6360 6361@item phiopt 6362@opindex fdump-tree-phiopt 6363Dump each function after optimizing PHI nodes into straightline code. The file 6364name is made by appending @file{.phiopt} to the source file name. 6365 6366@item forwprop 6367@opindex fdump-tree-forwprop 6368Dump each function after forward propagating single use variables. The file 6369name is made by appending @file{.forwprop} to the source file name. 6370 6371@item copyrename 6372@opindex fdump-tree-copyrename 6373Dump each function after applying the copy rename optimization. The file 6374name is made by appending @file{.copyrename} to the source file name. 6375 6376@item nrv 6377@opindex fdump-tree-nrv 6378Dump each function after applying the named return value optimization on 6379generic trees. The file name is made by appending @file{.nrv} to the source 6380file name. 6381 6382@item vect 6383@opindex fdump-tree-vect 6384Dump each function after applying vectorization of loops. The file name is 6385made by appending @file{.vect} to the source file name. 6386 6387@item slp 6388@opindex fdump-tree-slp 6389Dump each function after applying vectorization of basic blocks. The file name 6390is made by appending @file{.slp} to the source file name. 6391 6392@item vrp 6393@opindex fdump-tree-vrp 6394Dump each function after Value Range Propagation (VRP). The file name 6395is made by appending @file{.vrp} to the source file name. 6396 6397@item all 6398@opindex fdump-tree-all 6399Enable all the available tree dumps with the flags provided in this option. 6400@end table 6401 6402@item -fopt-info 6403@itemx -fopt-info-@var{options} 6404@itemx -fopt-info-@var{options}=@var{filename} 6405@opindex fopt-info 6406Controls optimization dumps from various optimization passes. If the 6407@samp{-@var{options}} form is used, @var{options} is a list of 6408@samp{-} separated options to select the dump details and 6409optimizations. If @var{options} is not specified, it defaults to 6410@option{optimized} for details and @option{optall} for optimization 6411groups. If the @var{filename} is not specified, it defaults to 6412@file{stderr}. Note that the output @var{filename} will be overwritten 6413in case of multiple translation units. If a combined output from 6414multiple translation units is desired, @file{stderr} should be used 6415instead. 6416 6417The options can be divided into two groups, 1) options describing the 6418verbosity of the dump, and 2) options describing which optimizations 6419should be included. The options from both the groups can be freely 6420mixed as they are non-overlapping. However, in case of any conflicts, 6421the latter options override the earlier options on the command 6422line. Though multiple -fopt-info options are accepted, only one of 6423them can have @option{=filename}. If other filenames are provided then 6424all but the first one are ignored. 6425 6426The dump verbosity has the following options 6427 6428@table @samp 6429@item optimized 6430Print information when an optimization is successfully applied. It is 6431up to a pass to decide which information is relevant. For example, the 6432vectorizer passes print the source location of loops which got 6433successfully vectorized. 6434@item missed 6435Print information about missed optimizations. Individual passes 6436control which information to include in the output. For example, 6437 6438@smallexample 6439gcc -O2 -ftree-vectorize -fopt-info-vec-missed 6440@end smallexample 6441 6442will print information about missed optimization opportunities from 6443vectorization passes on stderr. 6444@item note 6445Print verbose information about optimizations, such as certain 6446transformations, more detailed messages about decisions etc. 6447@item all 6448Print detailed optimization information. This includes 6449@var{optimized}, @var{missed}, and @var{note}. 6450@end table 6451 6452The second set of options describes a group of optimizations and may 6453include one or more of the following. 6454 6455@table @samp 6456@item ipa 6457Enable dumps from all interprocedural optimizations. 6458@item loop 6459Enable dumps from all loop optimizations. 6460@item inline 6461Enable dumps from all inlining optimizations. 6462@item vec 6463Enable dumps from all vectorization optimizations. 6464@item optall 6465Enable dumps from all optimizations. This is a superset of 6466the optimization groups listed above. 6467@end table 6468 6469For example, 6470@smallexample 6471gcc -O3 -fopt-info-missed=missed.all 6472@end smallexample 6473 6474outputs missed optimization report from all the passes into 6475@file{missed.all}. 6476 6477As another example, 6478@smallexample 6479gcc -O3 -fopt-info-inline-optimized-missed=inline.txt 6480@end smallexample 6481 6482will output information about missed optimizations as well as 6483optimized locations from all the inlining passes into 6484@file{inline.txt}. 6485 6486If the @var{filename} is provided, then the dumps from all the 6487applicable optimizations are concatenated into the @file{filename}. 6488Otherwise the dump is output onto @file{stderr}. If @var{options} is 6489omitted, it defaults to @option{all-optall}, which means dump all 6490available optimization info from all the passes. In the following 6491example, all optimization info is output on to @file{stderr}. 6492 6493@smallexample 6494gcc -O3 -fopt-info 6495@end smallexample 6496 6497Note that @option{-fopt-info-vec-missed} behaves the same as 6498@option{-fopt-info-missed-vec}. 6499 6500As another example, consider 6501 6502@smallexample 6503gcc -fopt-info-vec-missed=vec.miss -fopt-info-loop-optimized=loop.opt 6504@end smallexample 6505 6506Here the two output filenames @file{vec.miss} and @file{loop.opt} are 6507in conflict since only one output file is allowed. In this case, only 6508the first option takes effect and the subsequent options are 6509ignored. Thus only the @file{vec.miss} is produced which contains 6510dumps from the vectorizer about missed opportunities. 6511 6512@item -frandom-seed=@var{string} 6513@opindex frandom-seed 6514This option provides a seed that GCC uses in place of 6515random numbers in generating certain symbol names 6516that have to be different in every compiled file. It is also used to 6517place unique stamps in coverage data files and the object files that 6518produce them. You can use the @option{-frandom-seed} option to produce 6519reproducibly identical object files. 6520 6521The @var{string} should be different for every file you compile. 6522 6523@item -fsched-verbose=@var{n} 6524@opindex fsched-verbose 6525On targets that use instruction scheduling, this option controls the 6526amount of debugging output the scheduler prints. This information is 6527written to standard error, unless @option{-fdump-rtl-sched1} or 6528@option{-fdump-rtl-sched2} is specified, in which case it is output 6529to the usual dump listing file, @file{.sched1} or @file{.sched2} 6530respectively. However for @var{n} greater than nine, the output is 6531always printed to standard error. 6532 6533For @var{n} greater than zero, @option{-fsched-verbose} outputs the 6534same information as @option{-fdump-rtl-sched1} and @option{-fdump-rtl-sched2}. 6535For @var{n} greater than one, it also output basic block probabilities, 6536detailed ready list information and unit/insn info. For @var{n} greater 6537than two, it includes RTL at abort point, control-flow and regions info. 6538And for @var{n} over four, @option{-fsched-verbose} also includes 6539dependence info. 6540 6541@item -save-temps 6542@itemx -save-temps=cwd 6543@opindex save-temps 6544Store the usual ``temporary'' intermediate files permanently; place them 6545in the current directory and name them based on the source file. Thus, 6546compiling @file{foo.c} with @option{-c -save-temps} produces files 6547@file{foo.i} and @file{foo.s}, as well as @file{foo.o}. This creates a 6548preprocessed @file{foo.i} output file even though the compiler now 6549normally uses an integrated preprocessor. 6550 6551When used in combination with the @option{-x} command-line option, 6552@option{-save-temps} is sensible enough to avoid over writing an 6553input source file with the same extension as an intermediate file. 6554The corresponding intermediate file may be obtained by renaming the 6555source file before using @option{-save-temps}. 6556 6557If you invoke GCC in parallel, compiling several different source 6558files that share a common base name in different subdirectories or the 6559same source file compiled for multiple output destinations, it is 6560likely that the different parallel compilers will interfere with each 6561other, and overwrite the temporary files. For instance: 6562 6563@smallexample 6564gcc -save-temps -o outdir1/foo.o indir1/foo.c& 6565gcc -save-temps -o outdir2/foo.o indir2/foo.c& 6566@end smallexample 6567 6568may result in @file{foo.i} and @file{foo.o} being written to 6569simultaneously by both compilers. 6570 6571@item -save-temps=obj 6572@opindex save-temps=obj 6573Store the usual ``temporary'' intermediate files permanently. If the 6574@option{-o} option is used, the temporary files are based on the 6575object file. If the @option{-o} option is not used, the 6576@option{-save-temps=obj} switch behaves like @option{-save-temps}. 6577 6578For example: 6579 6580@smallexample 6581gcc -save-temps=obj -c foo.c 6582gcc -save-temps=obj -c bar.c -o dir/xbar.o 6583gcc -save-temps=obj foobar.c -o dir2/yfoobar 6584@end smallexample 6585 6586@noindent 6587creates @file{foo.i}, @file{foo.s}, @file{dir/xbar.i}, 6588@file{dir/xbar.s}, @file{dir2/yfoobar.i}, @file{dir2/yfoobar.s}, and 6589@file{dir2/yfoobar.o}. 6590 6591@item -time@r{[}=@var{file}@r{]} 6592@opindex time 6593Report the CPU time taken by each subprocess in the compilation 6594sequence. For C source files, this is the compiler proper and assembler 6595(plus the linker if linking is done). 6596 6597Without the specification of an output file, the output looks like this: 6598 6599@smallexample 6600# cc1 0.12 0.01 6601# as 0.00 0.01 6602@end smallexample 6603 6604The first number on each line is the ``user time'', that is time spent 6605executing the program itself. The second number is ``system time'', 6606time spent executing operating system routines on behalf of the program. 6607Both numbers are in seconds. 6608 6609With the specification of an output file, the output is appended to the 6610named file, and it looks like this: 6611 6612@smallexample 66130.12 0.01 cc1 @var{options} 66140.00 0.01 as @var{options} 6615@end smallexample 6616 6617The ``user time'' and the ``system time'' are moved before the program 6618name, and the options passed to the program are displayed, so that one 6619can later tell what file was being compiled, and with which options. 6620 6621@item -fvar-tracking 6622@opindex fvar-tracking 6623Run variable tracking pass. It computes where variables are stored at each 6624position in code. Better debugging information is then generated 6625(if the debugging information format supports this information). 6626 6627It is enabled by default when compiling with optimization (@option{-Os}, 6628@option{-O}, @option{-O2}, @dots{}), debugging information (@option{-g}) and 6629the debug info format supports it. 6630 6631@item -fvar-tracking-assignments 6632@opindex fvar-tracking-assignments 6633@opindex fno-var-tracking-assignments 6634Annotate assignments to user variables early in the compilation and 6635attempt to carry the annotations over throughout the compilation all the 6636way to the end, in an attempt to improve debug information while 6637optimizing. Use of @option{-gdwarf-4} is recommended along with it. 6638 6639It can be enabled even if var-tracking is disabled, in which case 6640annotations are created and maintained, but discarded at the end. 6641 6642@item -fvar-tracking-assignments-toggle 6643@opindex fvar-tracking-assignments-toggle 6644@opindex fno-var-tracking-assignments-toggle 6645Toggle @option{-fvar-tracking-assignments}, in the same way that 6646@option{-gtoggle} toggles @option{-g}. 6647 6648@item -print-file-name=@var{library} 6649@opindex print-file-name 6650Print the full absolute name of the library file @var{library} that 6651would be used when linking---and don't do anything else. With this 6652option, GCC does not compile or link anything; it just prints the 6653file name. 6654 6655@item -print-multi-directory 6656@opindex print-multi-directory 6657Print the directory name corresponding to the multilib selected by any 6658other switches present in the command line. This directory is supposed 6659to exist in @env{GCC_EXEC_PREFIX}. 6660 6661@item -print-multi-lib 6662@opindex print-multi-lib 6663Print the mapping from multilib directory names to compiler switches 6664that enable them. The directory name is separated from the switches by 6665@samp{;}, and each switch starts with an @samp{@@} instead of the 6666@samp{-}, without spaces between multiple switches. This is supposed to 6667ease shell processing. 6668 6669@item -print-multi-os-directory 6670@opindex print-multi-os-directory 6671Print the path to OS libraries for the selected 6672multilib, relative to some @file{lib} subdirectory. If OS libraries are 6673present in the @file{lib} subdirectory and no multilibs are used, this is 6674usually just @file{.}, if OS libraries are present in @file{lib@var{suffix}} 6675sibling directories this prints e.g.@: @file{../lib64}, @file{../lib} or 6676@file{../lib32}, or if OS libraries are present in @file{lib/@var{subdir}} 6677subdirectories it prints e.g.@: @file{amd64}, @file{sparcv9} or @file{ev6}. 6678 6679@item -print-multiarch 6680@opindex print-multiarch 6681Print the path to OS libraries for the selected multiarch, 6682relative to some @file{lib} subdirectory. 6683 6684@item -print-prog-name=@var{program} 6685@opindex print-prog-name 6686Like @option{-print-file-name}, but searches for a program such as @samp{cpp}. 6687 6688@item -print-libgcc-file-name 6689@opindex print-libgcc-file-name 6690Same as @option{-print-file-name=libgcc.a}. 6691 6692This is useful when you use @option{-nostdlib} or @option{-nodefaultlibs} 6693but you do want to link with @file{libgcc.a}. You can do: 6694 6695@smallexample 6696gcc -nostdlib @var{files}@dots{} `gcc -print-libgcc-file-name` 6697@end smallexample 6698 6699@item -print-search-dirs 6700@opindex print-search-dirs 6701Print the name of the configured installation directory and a list of 6702program and library directories @command{gcc} searches---and don't do anything else. 6703 6704This is useful when @command{gcc} prints the error message 6705@samp{installation problem, cannot exec cpp0: No such file or directory}. 6706To resolve this you either need to put @file{cpp0} and the other compiler 6707components where @command{gcc} expects to find them, or you can set the environment 6708variable @env{GCC_EXEC_PREFIX} to the directory where you installed them. 6709Don't forget the trailing @samp{/}. 6710@xref{Environment Variables}. 6711 6712@item -print-sysroot 6713@opindex print-sysroot 6714Print the target sysroot directory that is used during 6715compilation. This is the target sysroot specified either at configure 6716time or using the @option{--sysroot} option, possibly with an extra 6717suffix that depends on compilation options. If no target sysroot is 6718specified, the option prints nothing. 6719 6720@item -print-sysroot-headers-suffix 6721@opindex print-sysroot-headers-suffix 6722Print the suffix added to the target sysroot when searching for 6723headers, or give an error if the compiler is not configured with such 6724a suffix---and don't do anything else. 6725 6726@item -dumpmachine 6727@opindex dumpmachine 6728Print the compiler's target machine (for example, 6729@samp{i686-pc-linux-gnu})---and don't do anything else. 6730 6731@item -dumpversion 6732@opindex dumpversion 6733Print the compiler version (for example, @samp{3.0})---and don't do 6734anything else. 6735 6736@item -dumpspecs 6737@opindex dumpspecs 6738Print the compiler's built-in specs---and don't do anything else. (This 6739is used when GCC itself is being built.) @xref{Spec Files}. 6740 6741@item -fno-eliminate-unused-debug-types 6742@opindex feliminate-unused-debug-types 6743@opindex fno-eliminate-unused-debug-types 6744Normally, when producing DWARF 2 output, GCC avoids producing debug symbol 6745output for types that are nowhere used in the source file being compiled. 6746Sometimes it is useful to have GCC emit debugging 6747information for all types declared in a compilation 6748unit, regardless of whether or not they are actually used 6749in that compilation unit, for example 6750if, in the debugger, you want to cast a value to a type that is 6751not actually used in your program (but is declared). More often, 6752however, this results in a significant amount of wasted space. 6753@end table 6754 6755@node Optimize Options 6756@section Options That Control Optimization 6757@cindex optimize options 6758@cindex options, optimization 6759 6760These options control various sorts of optimizations. 6761 6762Without any optimization option, the compiler's goal is to reduce the 6763cost of compilation and to make debugging produce the expected 6764results. Statements are independent: if you stop the program with a 6765breakpoint between statements, you can then assign a new value to any 6766variable or change the program counter to any other statement in the 6767function and get exactly the results you expect from the source 6768code. 6769 6770Turning on optimization flags makes the compiler attempt to improve 6771the performance and/or code size at the expense of compilation time 6772and possibly the ability to debug the program. 6773 6774The compiler performs optimization based on the knowledge it has of the 6775program. Compiling multiple files at once to a single output file mode allows 6776the compiler to use information gained from all of the files when compiling 6777each of them. 6778 6779Not all optimizations are controlled directly by a flag. Only 6780optimizations that have a flag are listed in this section. 6781 6782Most optimizations are only enabled if an @option{-O} level is set on 6783the command line. Otherwise they are disabled, even if individual 6784optimization flags are specified. 6785 6786Depending on the target and how GCC was configured, a slightly different 6787set of optimizations may be enabled at each @option{-O} level than 6788those listed here. You can invoke GCC with @option{-Q --help=optimizers} 6789to find out the exact set of optimizations that are enabled at each level. 6790@xref{Overall Options}, for examples. 6791 6792@table @gcctabopt 6793@item -O 6794@itemx -O1 6795@opindex O 6796@opindex O1 6797Optimize. Optimizing compilation takes somewhat more time, and a lot 6798more memory for a large function. 6799 6800With @option{-O}, the compiler tries to reduce code size and execution 6801time, without performing any optimizations that take a great deal of 6802compilation time. 6803 6804@option{-O} turns on the following optimization flags: 6805@gccoptlist{ 6806-fauto-inc-dec @gol 6807-fcompare-elim @gol 6808-fcprop-registers @gol 6809-fdce @gol 6810-fdefer-pop @gol 6811-fdelayed-branch @gol 6812-fdse @gol 6813-fguess-branch-probability @gol 6814-fif-conversion2 @gol 6815-fif-conversion @gol 6816-fipa-pure-const @gol 6817-fipa-profile @gol 6818-fipa-reference @gol 6819-fmerge-constants 6820-fsplit-wide-types @gol 6821-ftree-bit-ccp @gol 6822-ftree-builtin-call-dce @gol 6823-ftree-ccp @gol 6824-ftree-ch @gol 6825-ftree-copyrename @gol 6826-ftree-dce @gol 6827-ftree-dominator-opts @gol 6828-ftree-dse @gol 6829-ftree-forwprop @gol 6830-ftree-fre @gol 6831-ftree-phiprop @gol 6832-ftree-slsr @gol 6833-ftree-sra @gol 6834-ftree-pta @gol 6835-ftree-ter @gol 6836-funit-at-a-time} 6837 6838@option{-O} also turns on @option{-fomit-frame-pointer} on machines 6839where doing so does not interfere with debugging. 6840 6841@item -O2 6842@opindex O2 6843Optimize even more. GCC performs nearly all supported optimizations 6844that do not involve a space-speed tradeoff. 6845As compared to @option{-O}, this option increases both compilation time 6846and the performance of the generated code. 6847 6848@option{-O2} turns on all optimization flags specified by @option{-O}. It 6849also turns on the following optimization flags: 6850@gccoptlist{-fthread-jumps @gol 6851-falign-functions -falign-jumps @gol 6852-falign-loops -falign-labels @gol 6853-fcaller-saves @gol 6854-fcrossjumping @gol 6855-fcse-follow-jumps -fcse-skip-blocks @gol 6856-fdelete-null-pointer-checks @gol 6857-fdevirtualize -fdevirtualize-speculatively @gol 6858-fexpensive-optimizations @gol 6859-fgcse -fgcse-lm @gol 6860-fhoist-adjacent-loads @gol 6861-finline-small-functions @gol 6862-findirect-inlining @gol 6863-fipa-sra @gol 6864-fisolate-erroneous-paths-dereference @gol 6865-foptimize-sibling-calls @gol 6866-fpartial-inlining @gol 6867-fpeephole2 @gol 6868-freorder-blocks -freorder-functions @gol 6869-frerun-cse-after-loop @gol 6870-fsched-interblock -fsched-spec @gol 6871-fschedule-insns -fschedule-insns2 @gol 6872-fstrict-aliasing -fstrict-overflow @gol 6873-ftree-switch-conversion -ftree-tail-merge @gol 6874-ftree-pre @gol 6875-ftree-vrp} 6876 6877Please note the warning under @option{-fgcse} about 6878invoking @option{-O2} on programs that use computed gotos. 6879 6880@item -O3 6881@opindex O3 6882Optimize yet more. @option{-O3} turns on all optimizations specified 6883by @option{-O2} and also turns on the @option{-finline-functions}, 6884@option{-funswitch-loops}, @option{-fpredictive-commoning}, 6885@option{-fgcse-after-reload}, @option{-ftree-loop-vectorize}, 6886@option{-ftree-slp-vectorize}, @option{-fvect-cost-model}, 6887@option{-ftree-partial-pre} and @option{-fipa-cp-clone} options. 6888 6889@item -O0 6890@opindex O0 6891Reduce compilation time and make debugging produce the expected 6892results. This is the default. 6893 6894@item -Os 6895@opindex Os 6896Optimize for size. @option{-Os} enables all @option{-O2} optimizations that 6897do not typically increase code size. It also performs further 6898optimizations designed to reduce code size. 6899 6900@option{-Os} disables the following optimization flags: 6901@gccoptlist{-falign-functions -falign-jumps -falign-loops @gol 6902-falign-labels -freorder-blocks -freorder-blocks-and-partition @gol 6903-fprefetch-loop-arrays} 6904 6905@item -Ofast 6906@opindex Ofast 6907Disregard strict standards compliance. @option{-Ofast} enables all 6908@option{-O3} optimizations. It also enables optimizations that are not 6909valid for all standard-compliant programs. 6910It turns on @option{-ffast-math} and the Fortran-specific 6911@option{-fno-protect-parens} and @option{-fstack-arrays}. 6912 6913@item -Og 6914@opindex Og 6915Optimize debugging experience. @option{-Og} enables optimizations 6916that do not interfere with debugging. It should be the optimization 6917level of choice for the standard edit-compile-debug cycle, offering 6918a reasonable level of optimization while maintaining fast compilation 6919and a good debugging experience. 6920 6921If you use multiple @option{-O} options, with or without level numbers, 6922the last such option is the one that is effective. 6923@end table 6924 6925Options of the form @option{-f@var{flag}} specify machine-independent 6926flags. Most flags have both positive and negative forms; the negative 6927form of @option{-ffoo} is @option{-fno-foo}. In the table 6928below, only one of the forms is listed---the one you typically 6929use. You can figure out the other form by either removing @samp{no-} 6930or adding it. 6931 6932The following options control specific optimizations. They are either 6933activated by @option{-O} options or are related to ones that are. You 6934can use the following flags in the rare cases when ``fine-tuning'' of 6935optimizations to be performed is desired. 6936 6937@table @gcctabopt 6938@item -fno-defer-pop 6939@opindex fno-defer-pop 6940Always pop the arguments to each function call as soon as that function 6941returns. For machines that must pop arguments after a function call, 6942the compiler normally lets arguments accumulate on the stack for several 6943function calls and pops them all at once. 6944 6945Disabled at levels @option{-O}, @option{-O2}, @option{-O3}, @option{-Os}. 6946 6947@item -fforward-propagate 6948@opindex fforward-propagate 6949Perform a forward propagation pass on RTL@. The pass tries to combine two 6950instructions and checks if the result can be simplified. If loop unrolling 6951is active, two passes are performed and the second is scheduled after 6952loop unrolling. 6953 6954This option is enabled by default at optimization levels @option{-O}, 6955@option{-O2}, @option{-O3}, @option{-Os}. 6956 6957@item -ffp-contract=@var{style} 6958@opindex ffp-contract 6959@option{-ffp-contract=off} disables floating-point expression contraction. 6960@option{-ffp-contract=fast} enables floating-point expression contraction 6961such as forming of fused multiply-add operations if the target has 6962native support for them. 6963@option{-ffp-contract=on} enables floating-point expression contraction 6964if allowed by the language standard. This is currently not implemented 6965and treated equal to @option{-ffp-contract=off}. 6966 6967The default is @option{-ffp-contract=fast}. 6968 6969@item -fomit-frame-pointer 6970@opindex fomit-frame-pointer 6971Don't keep the frame pointer in a register for functions that 6972don't need one. This avoids the instructions to save, set up and 6973restore frame pointers; it also makes an extra register available 6974in many functions. @strong{It also makes debugging impossible on 6975some machines.} 6976 6977On some machines, such as the VAX, this flag has no effect, because 6978the standard calling sequence automatically handles the frame pointer 6979and nothing is saved by pretending it doesn't exist. The 6980machine-description macro @code{FRAME_POINTER_REQUIRED} controls 6981whether a target machine supports this flag. @xref{Registers,,Register 6982Usage, gccint, GNU Compiler Collection (GCC) Internals}. 6983 6984Starting with GCC version 4.6, the default setting (when not optimizing for 6985size) for 32-bit GNU/Linux x86 and 32-bit Darwin x86 targets has been changed to 6986@option{-fomit-frame-pointer}. The default can be reverted to 6987@option{-fno-omit-frame-pointer} by configuring GCC with the 6988@option{--enable-frame-pointer} configure option. 6989 6990Enabled at levels @option{-O}, @option{-O2}, @option{-O3}, @option{-Os}. 6991 6992@item -foptimize-sibling-calls 6993@opindex foptimize-sibling-calls 6994Optimize sibling and tail recursive calls. 6995 6996Enabled at levels @option{-O2}, @option{-O3}, @option{-Os}. 6997 6998@item -fno-inline 6999@opindex fno-inline 7000Do not expand any functions inline apart from those marked with 7001the @code{always_inline} attribute. This is the default when not 7002optimizing. 7003 7004Single functions can be exempted from inlining by marking them 7005with the @code{noinline} attribute. 7006 7007@item -finline-small-functions 7008@opindex finline-small-functions 7009Integrate functions into their callers when their body is smaller than expected 7010function call code (so overall size of program gets smaller). The compiler 7011heuristically decides which functions are simple enough to be worth integrating 7012in this way. This inlining applies to all functions, even those not declared 7013inline. 7014 7015Enabled at level @option{-O2}. 7016 7017@item -findirect-inlining 7018@opindex findirect-inlining 7019Inline also indirect calls that are discovered to be known at compile 7020time thanks to previous inlining. This option has any effect only 7021when inlining itself is turned on by the @option{-finline-functions} 7022or @option{-finline-small-functions} options. 7023 7024Enabled at level @option{-O2}. 7025 7026@item -finline-functions 7027@opindex finline-functions 7028Consider all functions for inlining, even if they are not declared inline. 7029The compiler heuristically decides which functions are worth integrating 7030in this way. 7031 7032If all calls to a given function are integrated, and the function is 7033declared @code{static}, then the function is normally not output as 7034assembler code in its own right. 7035 7036Enabled at level @option{-O3}. 7037 7038@item -finline-functions-called-once 7039@opindex finline-functions-called-once 7040Consider all @code{static} functions called once for inlining into their 7041caller even if they are not marked @code{inline}. If a call to a given 7042function is integrated, then the function is not output as assembler code 7043in its own right. 7044 7045Enabled at levels @option{-O1}, @option{-O2}, @option{-O3} and @option{-Os}. 7046 7047@item -fearly-inlining 7048@opindex fearly-inlining 7049Inline functions marked by @code{always_inline} and functions whose body seems 7050smaller than the function call overhead early before doing 7051@option{-fprofile-generate} instrumentation and real inlining pass. Doing so 7052makes profiling significantly cheaper and usually inlining faster on programs 7053having large chains of nested wrapper functions. 7054 7055Enabled by default. 7056 7057@item -fipa-sra 7058@opindex fipa-sra 7059Perform interprocedural scalar replacement of aggregates, removal of 7060unused parameters and replacement of parameters passed by reference 7061by parameters passed by value. 7062 7063Enabled at levels @option{-O2}, @option{-O3} and @option{-Os}. 7064 7065@item -finline-limit=@var{n} 7066@opindex finline-limit 7067By default, GCC limits the size of functions that can be inlined. This flag 7068allows coarse control of this limit. @var{n} is the size of functions that 7069can be inlined in number of pseudo instructions. 7070 7071Inlining is actually controlled by a number of parameters, which may be 7072specified individually by using @option{--param @var{name}=@var{value}}. 7073The @option{-finline-limit=@var{n}} option sets some of these parameters 7074as follows: 7075 7076@table @gcctabopt 7077@item max-inline-insns-single 7078is set to @var{n}/2. 7079@item max-inline-insns-auto 7080is set to @var{n}/2. 7081@end table 7082 7083See below for a documentation of the individual 7084parameters controlling inlining and for the defaults of these parameters. 7085 7086@emph{Note:} there may be no value to @option{-finline-limit} that results 7087in default behavior. 7088 7089@emph{Note:} pseudo instruction represents, in this particular context, an 7090abstract measurement of function's size. In no way does it represent a count 7091of assembly instructions and as such its exact meaning might change from one 7092release to an another. 7093 7094@item -fno-keep-inline-dllexport 7095@opindex -fno-keep-inline-dllexport 7096This is a more fine-grained version of @option{-fkeep-inline-functions}, 7097which applies only to functions that are declared using the @code{dllexport} 7098attribute or declspec (@xref{Function Attributes,,Declaring Attributes of 7099Functions}.) 7100 7101@item -fkeep-inline-functions 7102@opindex fkeep-inline-functions 7103In C, emit @code{static} functions that are declared @code{inline} 7104into the object file, even if the function has been inlined into all 7105of its callers. This switch does not affect functions using the 7106@code{extern inline} extension in GNU C90@. In C++, emit any and all 7107inline functions into the object file. 7108 7109@item -fkeep-static-consts 7110@opindex fkeep-static-consts 7111Emit variables declared @code{static const} when optimization isn't turned 7112on, even if the variables aren't referenced. 7113 7114GCC enables this option by default. If you want to force the compiler to 7115check if a variable is referenced, regardless of whether or not 7116optimization is turned on, use the @option{-fno-keep-static-consts} option. 7117 7118@item -fmerge-constants 7119@opindex fmerge-constants 7120Attempt to merge identical constants (string constants and floating-point 7121constants) across compilation units. 7122 7123This option is the default for optimized compilation if the assembler and 7124linker support it. Use @option{-fno-merge-constants} to inhibit this 7125behavior. 7126 7127Enabled at levels @option{-O}, @option{-O2}, @option{-O3}, @option{-Os}. 7128 7129@item -fmerge-all-constants 7130@opindex fmerge-all-constants 7131Attempt to merge identical constants and identical variables. 7132 7133This option implies @option{-fmerge-constants}. In addition to 7134@option{-fmerge-constants} this considers e.g.@: even constant initialized 7135arrays or initialized constant variables with integral or floating-point 7136types. Languages like C or C++ require each variable, including multiple 7137instances of the same variable in recursive calls, to have distinct locations, 7138so using this option results in non-conforming 7139behavior. 7140 7141@item -fmodulo-sched 7142@opindex fmodulo-sched 7143Perform swing modulo scheduling immediately before the first scheduling 7144pass. This pass looks at innermost loops and reorders their 7145instructions by overlapping different iterations. 7146 7147@item -fmodulo-sched-allow-regmoves 7148@opindex fmodulo-sched-allow-regmoves 7149Perform more aggressive SMS-based modulo scheduling with register moves 7150allowed. By setting this flag certain anti-dependences edges are 7151deleted, which triggers the generation of reg-moves based on the 7152life-range analysis. This option is effective only with 7153@option{-fmodulo-sched} enabled. 7154 7155@item -fno-branch-count-reg 7156@opindex fno-branch-count-reg 7157Do not use ``decrement and branch'' instructions on a count register, 7158but instead generate a sequence of instructions that decrement a 7159register, compare it against zero, then branch based upon the result. 7160This option is only meaningful on architectures that support such 7161instructions, which include x86, PowerPC, IA-64 and S/390. 7162 7163The default is @option{-fbranch-count-reg}. 7164 7165@item -fno-function-cse 7166@opindex fno-function-cse 7167Do not put function addresses in registers; make each instruction that 7168calls a constant function contain the function's address explicitly. 7169 7170This option results in less efficient code, but some strange hacks 7171that alter the assembler output may be confused by the optimizations 7172performed when this option is not used. 7173 7174The default is @option{-ffunction-cse} 7175 7176@item -fno-zero-initialized-in-bss 7177@opindex fno-zero-initialized-in-bss 7178If the target supports a BSS section, GCC by default puts variables that 7179are initialized to zero into BSS@. This can save space in the resulting 7180code. 7181 7182This option turns off this behavior because some programs explicitly 7183rely on variables going to the data section---e.g., so that the 7184resulting executable can find the beginning of that section and/or make 7185assumptions based on that. 7186 7187The default is @option{-fzero-initialized-in-bss}. 7188 7189@item -fthread-jumps 7190@opindex fthread-jumps 7191Perform optimizations that check to see if a jump branches to a 7192location where another comparison subsumed by the first is found. If 7193so, the first branch is redirected to either the destination of the 7194second branch or a point immediately following it, depending on whether 7195the condition is known to be true or false. 7196 7197Enabled at levels @option{-O2}, @option{-O3}, @option{-Os}. 7198 7199@item -fsplit-wide-types 7200@opindex fsplit-wide-types 7201When using a type that occupies multiple registers, such as @code{long 7202long} on a 32-bit system, split the registers apart and allocate them 7203independently. This normally generates better code for those types, 7204but may make debugging more difficult. 7205 7206Enabled at levels @option{-O}, @option{-O2}, @option{-O3}, 7207@option{-Os}. 7208 7209@item -fcse-follow-jumps 7210@opindex fcse-follow-jumps 7211In common subexpression elimination (CSE), scan through jump instructions 7212when the target of the jump is not reached by any other path. For 7213example, when CSE encounters an @code{if} statement with an 7214@code{else} clause, CSE follows the jump when the condition 7215tested is false. 7216 7217Enabled at levels @option{-O2}, @option{-O3}, @option{-Os}. 7218 7219@item -fcse-skip-blocks 7220@opindex fcse-skip-blocks 7221This is similar to @option{-fcse-follow-jumps}, but causes CSE to 7222follow jumps that conditionally skip over blocks. When CSE 7223encounters a simple @code{if} statement with no else clause, 7224@option{-fcse-skip-blocks} causes CSE to follow the jump around the 7225body of the @code{if}. 7226 7227Enabled at levels @option{-O2}, @option{-O3}, @option{-Os}. 7228 7229@item -frerun-cse-after-loop 7230@opindex frerun-cse-after-loop 7231Re-run common subexpression elimination after loop optimizations are 7232performed. 7233 7234Enabled at levels @option{-O2}, @option{-O3}, @option{-Os}. 7235 7236@item -fgcse 7237@opindex fgcse 7238Perform a global common subexpression elimination pass. 7239This pass also performs global constant and copy propagation. 7240 7241@emph{Note:} When compiling a program using computed gotos, a GCC 7242extension, you may get better run-time performance if you disable 7243the global common subexpression elimination pass by adding 7244@option{-fno-gcse} to the command line. 7245 7246Enabled at levels @option{-O2}, @option{-O3}, @option{-Os}. 7247 7248@item -fgcse-lm 7249@opindex fgcse-lm 7250When @option{-fgcse-lm} is enabled, global common subexpression elimination 7251attempts to move loads that are only killed by stores into themselves. This 7252allows a loop containing a load/store sequence to be changed to a load outside 7253the loop, and a copy/store within the loop. 7254 7255Enabled by default when @option{-fgcse} is enabled. 7256 7257@item -fgcse-sm 7258@opindex fgcse-sm 7259When @option{-fgcse-sm} is enabled, a store motion pass is run after 7260global common subexpression elimination. This pass attempts to move 7261stores out of loops. When used in conjunction with @option{-fgcse-lm}, 7262loops containing a load/store sequence can be changed to a load before 7263the loop and a store after the loop. 7264 7265Not enabled at any optimization level. 7266 7267@item -fgcse-las 7268@opindex fgcse-las 7269When @option{-fgcse-las} is enabled, the global common subexpression 7270elimination pass eliminates redundant loads that come after stores to the 7271same memory location (both partial and full redundancies). 7272 7273Not enabled at any optimization level. 7274 7275@item -fgcse-after-reload 7276@opindex fgcse-after-reload 7277When @option{-fgcse-after-reload} is enabled, a redundant load elimination 7278pass is performed after reload. The purpose of this pass is to clean up 7279redundant spilling. 7280 7281@item -faggressive-loop-optimizations 7282@opindex faggressive-loop-optimizations 7283This option tells the loop optimizer to use language constraints to 7284derive bounds for the number of iterations of a loop. This assumes that 7285loop code does not invoke undefined behavior by for example causing signed 7286integer overflows or out-of-bound array accesses. The bounds for the 7287number of iterations of a loop are used to guide loop unrolling and peeling 7288and loop exit test optimizations. 7289This option is enabled by default. 7290 7291@item -funsafe-loop-optimizations 7292@opindex funsafe-loop-optimizations 7293This option tells the loop optimizer to assume that loop indices do not 7294overflow, and that loops with nontrivial exit condition are not 7295infinite. This enables a wider range of loop optimizations even if 7296the loop optimizer itself cannot prove that these assumptions are valid. 7297If you use @option{-Wunsafe-loop-optimizations}, the compiler warns you 7298if it finds this kind of loop. 7299 7300@item -fcrossjumping 7301@opindex fcrossjumping 7302Perform cross-jumping transformation. 7303This transformation unifies equivalent code and saves code size. The 7304resulting code may or may not perform better than without cross-jumping. 7305 7306Enabled at levels @option{-O2}, @option{-O3}, @option{-Os}. 7307 7308@item -fauto-inc-dec 7309@opindex fauto-inc-dec 7310Combine increments or decrements of addresses with memory accesses. 7311This pass is always skipped on architectures that do not have 7312instructions to support this. Enabled by default at @option{-O} and 7313higher on architectures that support this. 7314 7315@item -fdce 7316@opindex fdce 7317Perform dead code elimination (DCE) on RTL@. 7318Enabled by default at @option{-O} and higher. 7319 7320@item -fdse 7321@opindex fdse 7322Perform dead store elimination (DSE) on RTL@. 7323Enabled by default at @option{-O} and higher. 7324 7325@item -fif-conversion 7326@opindex fif-conversion 7327Attempt to transform conditional jumps into branch-less equivalents. This 7328includes use of conditional moves, min, max, set flags and abs instructions, and 7329some tricks doable by standard arithmetics. The use of conditional execution 7330on chips where it is available is controlled by @code{if-conversion2}. 7331 7332Enabled at levels @option{-O}, @option{-O2}, @option{-O3}, @option{-Os}. 7333 7334@item -fif-conversion2 7335@opindex fif-conversion2 7336Use conditional execution (where available) to transform conditional jumps into 7337branch-less equivalents. 7338 7339Enabled at levels @option{-O}, @option{-O2}, @option{-O3}, @option{-Os}. 7340 7341@item -fdeclone-ctor-dtor 7342@opindex fdeclone-ctor-dtor 7343The C++ ABI requires multiple entry points for constructors and 7344destructors: one for a base subobject, one for a complete object, and 7345one for a virtual destructor that calls operator delete afterwards. 7346For a hierarchy with virtual bases, the base and complete variants are 7347clones, which means two copies of the function. With this option, the 7348base and complete variants are changed to be thunks that call a common 7349implementation. 7350 7351Enabled by @option{-Os}. 7352 7353@item -fdelete-null-pointer-checks 7354@opindex fdelete-null-pointer-checks 7355Assume that programs cannot safely dereference null pointers, and that 7356no code or data element resides there. This enables simple constant 7357folding optimizations at all optimization levels. In addition, other 7358optimization passes in GCC use this flag to control global dataflow 7359analyses that eliminate useless checks for null pointers; these assume 7360that if a pointer is checked after it has already been dereferenced, 7361it cannot be null. 7362 7363Note however that in some environments this assumption is not true. 7364Use @option{-fno-delete-null-pointer-checks} to disable this optimization 7365for programs that depend on that behavior. 7366 7367Some targets, especially embedded ones, disable this option at all levels. 7368Otherwise it is enabled at all levels: @option{-O0}, @option{-O1}, 7369@option{-O2}, @option{-O3}, @option{-Os}. Passes that use the information 7370are enabled independently at different optimization levels. 7371 7372@item -fdevirtualize 7373@opindex fdevirtualize 7374Attempt to convert calls to virtual functions to direct calls. This 7375is done both within a procedure and interprocedurally as part of 7376indirect inlining (@code{-findirect-inlining}) and interprocedural constant 7377propagation (@option{-fipa-cp}). 7378Enabled at levels @option{-O2}, @option{-O3}, @option{-Os}. 7379 7380@item -fdevirtualize-speculatively 7381@opindex fdevirtualize-speculatively 7382Attempt to convert calls to virtual functions to speculative direct calls. 7383Based on the analysis of the type inheritance graph, determine for a given call 7384the set of likely targets. If the set is small, preferably of size 1, change 7385the call into an conditional deciding on direct and indirect call. The 7386speculative calls enable more optimizations, such as inlining. When they seem 7387useless after further optimization, they are converted back into original form. 7388 7389@item -fexpensive-optimizations 7390@opindex fexpensive-optimizations 7391Perform a number of minor optimizations that are relatively expensive. 7392 7393Enabled at levels @option{-O2}, @option{-O3}, @option{-Os}. 7394 7395@item -free 7396@opindex free 7397Attempt to remove redundant extension instructions. This is especially 7398helpful for the x86-64 architecture, which implicitly zero-extends in 64-bit 7399registers after writing to their lower 32-bit half. 7400 7401Enabled for Alpha, AArch64 and x86 at levels @option{-O2}, 7402@option{-O3}, @option{-Os}. 7403 7404@item -flive-range-shrinkage 7405@opindex flive-range-shrinkage 7406Attempt to decrease register pressure through register live range 7407shrinkage. This is helpful for fast processors with small or moderate 7408size register sets. 7409 7410@item -fira-algorithm=@var{algorithm} 7411Use the specified coloring algorithm for the integrated register 7412allocator. The @var{algorithm} argument can be @samp{priority}, which 7413specifies Chow's priority coloring, or @samp{CB}, which specifies 7414Chaitin-Briggs coloring. Chaitin-Briggs coloring is not implemented 7415for all architectures, but for those targets that do support it, it is 7416the default because it generates better code. 7417 7418@item -fira-region=@var{region} 7419Use specified regions for the integrated register allocator. The 7420@var{region} argument should be one of the following: 7421 7422@table @samp 7423 7424@item all 7425Use all loops as register allocation regions. 7426This can give the best results for machines with a small and/or 7427irregular register set. 7428 7429@item mixed 7430Use all loops except for loops with small register pressure 7431as the regions. This value usually gives 7432the best results in most cases and for most architectures, 7433and is enabled by default when compiling with optimization for speed 7434(@option{-O}, @option{-O2}, @dots{}). 7435 7436@item one 7437Use all functions as a single region. 7438This typically results in the smallest code size, and is enabled by default for 7439@option{-Os} or @option{-O0}. 7440 7441@end table 7442 7443@item -fira-hoist-pressure 7444@opindex fira-hoist-pressure 7445Use IRA to evaluate register pressure in the code hoisting pass for 7446decisions to hoist expressions. This option usually results in smaller 7447code, but it can slow the compiler down. 7448 7449This option is enabled at level @option{-Os} for all targets. 7450 7451@item -fira-loop-pressure 7452@opindex fira-loop-pressure 7453Use IRA to evaluate register pressure in loops for decisions to move 7454loop invariants. This option usually results in generation 7455of faster and smaller code on machines with large register files (>= 32 7456registers), but it can slow the compiler down. 7457 7458This option is enabled at level @option{-O3} for some targets. 7459 7460@item -fno-ira-share-save-slots 7461@opindex fno-ira-share-save-slots 7462Disable sharing of stack slots used for saving call-used hard 7463registers living through a call. Each hard register gets a 7464separate stack slot, and as a result function stack frames are 7465larger. 7466 7467@item -fno-ira-share-spill-slots 7468@opindex fno-ira-share-spill-slots 7469Disable sharing of stack slots allocated for pseudo-registers. Each 7470pseudo-register that does not get a hard register gets a separate 7471stack slot, and as a result function stack frames are larger. 7472 7473@item -fira-verbose=@var{n} 7474@opindex fira-verbose 7475Control the verbosity of the dump file for the integrated register allocator. 7476The default value is 5. If the value @var{n} is greater or equal to 10, 7477the dump output is sent to stderr using the same format as @var{n} minus 10. 7478 7479@item -fdelayed-branch 7480@opindex fdelayed-branch 7481If supported for the target machine, attempt to reorder instructions 7482to exploit instruction slots available after delayed branch 7483instructions. 7484 7485Enabled at levels @option{-O}, @option{-O2}, @option{-O3}, @option{-Os}. 7486 7487@item -fschedule-insns 7488@opindex fschedule-insns 7489If supported for the target machine, attempt to reorder instructions to 7490eliminate execution stalls due to required data being unavailable. This 7491helps machines that have slow floating point or memory load instructions 7492by allowing other instructions to be issued until the result of the load 7493or floating-point instruction is required. 7494 7495Enabled at levels @option{-O2}, @option{-O3}. 7496 7497@item -fschedule-insns2 7498@opindex fschedule-insns2 7499Similar to @option{-fschedule-insns}, but requests an additional pass of 7500instruction scheduling after register allocation has been done. This is 7501especially useful on machines with a relatively small number of 7502registers and where memory load instructions take more than one cycle. 7503 7504Enabled at levels @option{-O2}, @option{-O3}, @option{-Os}. 7505 7506@item -fno-sched-interblock 7507@opindex fno-sched-interblock 7508Don't schedule instructions across basic blocks. This is normally 7509enabled by default when scheduling before register allocation, i.e.@: 7510with @option{-fschedule-insns} or at @option{-O2} or higher. 7511 7512@item -fno-sched-spec 7513@opindex fno-sched-spec 7514Don't allow speculative motion of non-load instructions. This is normally 7515enabled by default when scheduling before register allocation, i.e.@: 7516with @option{-fschedule-insns} or at @option{-O2} or higher. 7517 7518@item -fsched-pressure 7519@opindex fsched-pressure 7520Enable register pressure sensitive insn scheduling before register 7521allocation. This only makes sense when scheduling before register 7522allocation is enabled, i.e.@: with @option{-fschedule-insns} or at 7523@option{-O2} or higher. Usage of this option can improve the 7524generated code and decrease its size by preventing register pressure 7525increase above the number of available hard registers and subsequent 7526spills in register allocation. 7527 7528@item -fsched-spec-load 7529@opindex fsched-spec-load 7530Allow speculative motion of some load instructions. This only makes 7531sense when scheduling before register allocation, i.e.@: with 7532@option{-fschedule-insns} or at @option{-O2} or higher. 7533 7534@item -fsched-spec-load-dangerous 7535@opindex fsched-spec-load-dangerous 7536Allow speculative motion of more load instructions. This only makes 7537sense when scheduling before register allocation, i.e.@: with 7538@option{-fschedule-insns} or at @option{-O2} or higher. 7539 7540@item -fsched-stalled-insns 7541@itemx -fsched-stalled-insns=@var{n} 7542@opindex fsched-stalled-insns 7543Define how many insns (if any) can be moved prematurely from the queue 7544of stalled insns into the ready list during the second scheduling pass. 7545@option{-fno-sched-stalled-insns} means that no insns are moved 7546prematurely, @option{-fsched-stalled-insns=0} means there is no limit 7547on how many queued insns can be moved prematurely. 7548@option{-fsched-stalled-insns} without a value is equivalent to 7549@option{-fsched-stalled-insns=1}. 7550 7551@item -fsched-stalled-insns-dep 7552@itemx -fsched-stalled-insns-dep=@var{n} 7553@opindex fsched-stalled-insns-dep 7554Define how many insn groups (cycles) are examined for a dependency 7555on a stalled insn that is a candidate for premature removal from the queue 7556of stalled insns. This has an effect only during the second scheduling pass, 7557and only if @option{-fsched-stalled-insns} is used. 7558@option{-fno-sched-stalled-insns-dep} is equivalent to 7559@option{-fsched-stalled-insns-dep=0}. 7560@option{-fsched-stalled-insns-dep} without a value is equivalent to 7561@option{-fsched-stalled-insns-dep=1}. 7562 7563@item -fsched2-use-superblocks 7564@opindex fsched2-use-superblocks 7565When scheduling after register allocation, use superblock scheduling. 7566This allows motion across basic block boundaries, 7567resulting in faster schedules. This option is experimental, as not all machine 7568descriptions used by GCC model the CPU closely enough to avoid unreliable 7569results from the algorithm. 7570 7571This only makes sense when scheduling after register allocation, i.e.@: with 7572@option{-fschedule-insns2} or at @option{-O2} or higher. 7573 7574@item -fsched-group-heuristic 7575@opindex fsched-group-heuristic 7576Enable the group heuristic in the scheduler. This heuristic favors 7577the instruction that belongs to a schedule group. This is enabled 7578by default when scheduling is enabled, i.e.@: with @option{-fschedule-insns} 7579or @option{-fschedule-insns2} or at @option{-O2} or higher. 7580 7581@item -fsched-critical-path-heuristic 7582@opindex fsched-critical-path-heuristic 7583Enable the critical-path heuristic in the scheduler. This heuristic favors 7584instructions on the critical path. This is enabled by default when 7585scheduling is enabled, i.e.@: with @option{-fschedule-insns} 7586or @option{-fschedule-insns2} or at @option{-O2} or higher. 7587 7588@item -fsched-spec-insn-heuristic 7589@opindex fsched-spec-insn-heuristic 7590Enable the speculative instruction heuristic in the scheduler. This 7591heuristic favors speculative instructions with greater dependency weakness. 7592This is enabled by default when scheduling is enabled, i.e.@: 7593with @option{-fschedule-insns} or @option{-fschedule-insns2} 7594or at @option{-O2} or higher. 7595 7596@item -fsched-rank-heuristic 7597@opindex fsched-rank-heuristic 7598Enable the rank heuristic in the scheduler. This heuristic favors 7599the instruction belonging to a basic block with greater size or frequency. 7600This is enabled by default when scheduling is enabled, i.e.@: 7601with @option{-fschedule-insns} or @option{-fschedule-insns2} or 7602at @option{-O2} or higher. 7603 7604@item -fsched-last-insn-heuristic 7605@opindex fsched-last-insn-heuristic 7606Enable the last-instruction heuristic in the scheduler. This heuristic 7607favors the instruction that is less dependent on the last instruction 7608scheduled. This is enabled by default when scheduling is enabled, 7609i.e.@: with @option{-fschedule-insns} or @option{-fschedule-insns2} or 7610at @option{-O2} or higher. 7611 7612@item -fsched-dep-count-heuristic 7613@opindex fsched-dep-count-heuristic 7614Enable the dependent-count heuristic in the scheduler. This heuristic 7615favors the instruction that has more instructions depending on it. 7616This is enabled by default when scheduling is enabled, i.e.@: 7617with @option{-fschedule-insns} or @option{-fschedule-insns2} or 7618at @option{-O2} or higher. 7619 7620@item -freschedule-modulo-scheduled-loops 7621@opindex freschedule-modulo-scheduled-loops 7622Modulo scheduling is performed before traditional scheduling. If a loop 7623is modulo scheduled, later scheduling passes may change its schedule. 7624Use this option to control that behavior. 7625 7626@item -fselective-scheduling 7627@opindex fselective-scheduling 7628Schedule instructions using selective scheduling algorithm. Selective 7629scheduling runs instead of the first scheduler pass. 7630 7631@item -fselective-scheduling2 7632@opindex fselective-scheduling2 7633Schedule instructions using selective scheduling algorithm. Selective 7634scheduling runs instead of the second scheduler pass. 7635 7636@item -fsel-sched-pipelining 7637@opindex fsel-sched-pipelining 7638Enable software pipelining of innermost loops during selective scheduling. 7639This option has no effect unless one of @option{-fselective-scheduling} or 7640@option{-fselective-scheduling2} is turned on. 7641 7642@item -fsel-sched-pipelining-outer-loops 7643@opindex fsel-sched-pipelining-outer-loops 7644When pipelining loops during selective scheduling, also pipeline outer loops. 7645This option has no effect unless @option{-fsel-sched-pipelining} is turned on. 7646 7647@item -fshrink-wrap 7648@opindex fshrink-wrap 7649Emit function prologues only before parts of the function that need it, 7650rather than at the top of the function. This flag is enabled by default at 7651@option{-O} and higher. 7652 7653@item -fcaller-saves 7654@opindex fcaller-saves 7655Enable allocation of values to registers that are clobbered by 7656function calls, by emitting extra instructions to save and restore the 7657registers around such calls. Such allocation is done only when it 7658seems to result in better code. 7659 7660This option is always enabled by default on certain machines, usually 7661those which have no call-preserved registers to use instead. 7662 7663Enabled at levels @option{-O2}, @option{-O3}, @option{-Os}. 7664 7665@item -fcombine-stack-adjustments 7666@opindex fcombine-stack-adjustments 7667Tracks stack adjustments (pushes and pops) and stack memory references 7668and then tries to find ways to combine them. 7669 7670Enabled by default at @option{-O1} and higher. 7671 7672@item -fconserve-stack 7673@opindex fconserve-stack 7674Attempt to minimize stack usage. The compiler attempts to use less 7675stack space, even if that makes the program slower. This option 7676implies setting the @option{large-stack-frame} parameter to 100 7677and the @option{large-stack-frame-growth} parameter to 400. 7678 7679@item -ftree-reassoc 7680@opindex ftree-reassoc 7681Perform reassociation on trees. This flag is enabled by default 7682at @option{-O} and higher. 7683 7684@item -ftree-pre 7685@opindex ftree-pre 7686Perform partial redundancy elimination (PRE) on trees. This flag is 7687enabled by default at @option{-O2} and @option{-O3}. 7688 7689@item -ftree-partial-pre 7690@opindex ftree-partial-pre 7691Make partial redundancy elimination (PRE) more aggressive. This flag is 7692enabled by default at @option{-O3}. 7693 7694@item -ftree-forwprop 7695@opindex ftree-forwprop 7696Perform forward propagation on trees. This flag is enabled by default 7697at @option{-O} and higher. 7698 7699@item -ftree-fre 7700@opindex ftree-fre 7701Perform full redundancy elimination (FRE) on trees. The difference 7702between FRE and PRE is that FRE only considers expressions 7703that are computed on all paths leading to the redundant computation. 7704This analysis is faster than PRE, though it exposes fewer redundancies. 7705This flag is enabled by default at @option{-O} and higher. 7706 7707@item -ftree-phiprop 7708@opindex ftree-phiprop 7709Perform hoisting of loads from conditional pointers on trees. This 7710pass is enabled by default at @option{-O} and higher. 7711 7712@item -fhoist-adjacent-loads 7713@opindex hoist-adjacent-loads 7714Speculatively hoist loads from both branches of an if-then-else if the 7715loads are from adjacent locations in the same structure and the target 7716architecture has a conditional move instruction. This flag is enabled 7717by default at @option{-O2} and higher. 7718 7719@item -ftree-copy-prop 7720@opindex ftree-copy-prop 7721Perform copy propagation on trees. This pass eliminates unnecessary 7722copy operations. This flag is enabled by default at @option{-O} and 7723higher. 7724 7725@item -fipa-pure-const 7726@opindex fipa-pure-const 7727Discover which functions are pure or constant. 7728Enabled by default at @option{-O} and higher. 7729 7730@item -fipa-reference 7731@opindex fipa-reference 7732Discover which static variables do not escape the 7733compilation unit. 7734Enabled by default at @option{-O} and higher. 7735 7736@item -fipa-pta 7737@opindex fipa-pta 7738Perform interprocedural pointer analysis and interprocedural modification 7739and reference analysis. This option can cause excessive memory and 7740compile-time usage on large compilation units. It is not enabled by 7741default at any optimization level. 7742 7743@item -fipa-profile 7744@opindex fipa-profile 7745Perform interprocedural profile propagation. The functions called only from 7746cold functions are marked as cold. Also functions executed once (such as 7747@code{cold}, @code{noreturn}, static constructors or destructors) are identified. Cold 7748functions and loop less parts of functions executed once are then optimized for 7749size. 7750Enabled by default at @option{-O} and higher. 7751 7752@item -fipa-cp 7753@opindex fipa-cp 7754Perform interprocedural constant propagation. 7755This optimization analyzes the program to determine when values passed 7756to functions are constants and then optimizes accordingly. 7757This optimization can substantially increase performance 7758if the application has constants passed to functions. 7759This flag is enabled by default at @option{-O2}, @option{-Os} and @option{-O3}. 7760 7761@item -fipa-cp-clone 7762@opindex fipa-cp-clone 7763Perform function cloning to make interprocedural constant propagation stronger. 7764When enabled, interprocedural constant propagation performs function cloning 7765when externally visible function can be called with constant arguments. 7766Because this optimization can create multiple copies of functions, 7767it may significantly increase code size 7768(see @option{--param ipcp-unit-growth=@var{value}}). 7769This flag is enabled by default at @option{-O3}. 7770 7771@item -fisolate-erroneous-paths-dereference 7772Detect paths which trigger erroneous or undefined behaviour due to 7773dereferencing a NULL pointer. Isolate those paths from the main control 7774flow and turn the statement with erroneous or undefined behaviour into a trap. 7775 7776@item -fisolate-erroneous-paths-attribute 7777Detect paths which trigger erroneous or undefined behaviour due a NULL value 7778being used in a way which is forbidden by a @code{returns_nonnull} or @code{nonnull} 7779attribute. Isolate those paths from the main control flow and turn the 7780statement with erroneous or undefined behaviour into a trap. This is not 7781currently enabled, but may be enabled by @code{-O2} in the future. 7782 7783@item -ftree-sink 7784@opindex ftree-sink 7785Perform forward store motion on trees. This flag is 7786enabled by default at @option{-O} and higher. 7787 7788@item -ftree-bit-ccp 7789@opindex ftree-bit-ccp 7790Perform sparse conditional bit constant propagation on trees and propagate 7791pointer alignment information. 7792This pass only operates on local scalar variables and is enabled by default 7793at @option{-O} and higher. It requires that @option{-ftree-ccp} is enabled. 7794 7795@item -ftree-ccp 7796@opindex ftree-ccp 7797Perform sparse conditional constant propagation (CCP) on trees. This 7798pass only operates on local scalar variables and is enabled by default 7799at @option{-O} and higher. 7800 7801@item -ftree-switch-conversion 7802Perform conversion of simple initializations in a switch to 7803initializations from a scalar array. This flag is enabled by default 7804at @option{-O2} and higher. 7805 7806@item -ftree-tail-merge 7807Look for identical code sequences. When found, replace one with a jump to the 7808other. This optimization is known as tail merging or cross jumping. This flag 7809is enabled by default at @option{-O2} and higher. The compilation time 7810in this pass can 7811be limited using @option{max-tail-merge-comparisons} parameter and 7812@option{max-tail-merge-iterations} parameter. 7813 7814@item -ftree-dce 7815@opindex ftree-dce 7816Perform dead code elimination (DCE) on trees. This flag is enabled by 7817default at @option{-O} and higher. 7818 7819@item -ftree-builtin-call-dce 7820@opindex ftree-builtin-call-dce 7821Perform conditional dead code elimination (DCE) for calls to built-in functions 7822that may set @code{errno} but are otherwise side-effect free. This flag is 7823enabled by default at @option{-O2} and higher if @option{-Os} is not also 7824specified. 7825 7826@item -ftree-dominator-opts 7827@opindex ftree-dominator-opts 7828Perform a variety of simple scalar cleanups (constant/copy 7829propagation, redundancy elimination, range propagation and expression 7830simplification) based on a dominator tree traversal. This also 7831performs jump threading (to reduce jumps to jumps). This flag is 7832enabled by default at @option{-O} and higher. 7833 7834@item -ftree-dse 7835@opindex ftree-dse 7836Perform dead store elimination (DSE) on trees. A dead store is a store into 7837a memory location that is later overwritten by another store without 7838any intervening loads. In this case the earlier store can be deleted. This 7839flag is enabled by default at @option{-O} and higher. 7840 7841@item -ftree-ch 7842@opindex ftree-ch 7843Perform loop header copying on trees. This is beneficial since it increases 7844effectiveness of code motion optimizations. It also saves one jump. This flag 7845is enabled by default at @option{-O} and higher. It is not enabled 7846for @option{-Os}, since it usually increases code size. 7847 7848@item -ftree-loop-optimize 7849@opindex ftree-loop-optimize 7850Perform loop optimizations on trees. This flag is enabled by default 7851at @option{-O} and higher. 7852 7853@item -ftree-loop-linear 7854@opindex ftree-loop-linear 7855Perform loop interchange transformations on tree. Same as 7856@option{-floop-interchange}. To use this code transformation, GCC has 7857to be configured with @option{--with-ppl} and @option{--with-cloog} to 7858enable the Graphite loop transformation infrastructure. 7859 7860@item -floop-interchange 7861@opindex floop-interchange 7862Perform loop interchange transformations on loops. Interchanging two 7863nested loops switches the inner and outer loops. For example, given a 7864loop like: 7865@smallexample 7866DO J = 1, M 7867 DO I = 1, N 7868 A(J, I) = A(J, I) * C 7869 ENDDO 7870ENDDO 7871@end smallexample 7872loop interchange transforms the loop as if it were written: 7873@smallexample 7874DO I = 1, N 7875 DO J = 1, M 7876 A(J, I) = A(J, I) * C 7877 ENDDO 7878ENDDO 7879@end smallexample 7880which can be beneficial when @code{N} is larger than the caches, 7881because in Fortran, the elements of an array are stored in memory 7882contiguously by column, and the original loop iterates over rows, 7883potentially creating at each access a cache miss. This optimization 7884applies to all the languages supported by GCC and is not limited to 7885Fortran. To use this code transformation, GCC has to be configured 7886with @option{--with-ppl} and @option{--with-cloog} to enable the 7887Graphite loop transformation infrastructure. 7888 7889@item -floop-strip-mine 7890@opindex floop-strip-mine 7891Perform loop strip mining transformations on loops. Strip mining 7892splits a loop into two nested loops. The outer loop has strides 7893equal to the strip size and the inner loop has strides of the 7894original loop within a strip. The strip length can be changed 7895using the @option{loop-block-tile-size} parameter. For example, 7896given a loop like: 7897@smallexample 7898DO I = 1, N 7899 A(I) = A(I) + C 7900ENDDO 7901@end smallexample 7902loop strip mining transforms the loop as if it were written: 7903@smallexample 7904DO II = 1, N, 51 7905 DO I = II, min (II + 50, N) 7906 A(I) = A(I) + C 7907 ENDDO 7908ENDDO 7909@end smallexample 7910This optimization applies to all the languages supported by GCC and is 7911not limited to Fortran. To use this code transformation, GCC has to 7912be configured with @option{--with-ppl} and @option{--with-cloog} to 7913enable the Graphite loop transformation infrastructure. 7914 7915@item -floop-block 7916@opindex floop-block 7917Perform loop blocking transformations on loops. Blocking strip mines 7918each loop in the loop nest such that the memory accesses of the 7919element loops fit inside caches. The strip length can be changed 7920using the @option{loop-block-tile-size} parameter. For example, given 7921a loop like: 7922@smallexample 7923DO I = 1, N 7924 DO J = 1, M 7925 A(J, I) = B(I) + C(J) 7926 ENDDO 7927ENDDO 7928@end smallexample 7929loop blocking transforms the loop as if it were written: 7930@smallexample 7931DO II = 1, N, 51 7932 DO JJ = 1, M, 51 7933 DO I = II, min (II + 50, N) 7934 DO J = JJ, min (JJ + 50, M) 7935 A(J, I) = B(I) + C(J) 7936 ENDDO 7937 ENDDO 7938 ENDDO 7939ENDDO 7940@end smallexample 7941which can be beneficial when @code{M} is larger than the caches, 7942because the innermost loop iterates over a smaller amount of data 7943which can be kept in the caches. This optimization applies to all the 7944languages supported by GCC and is not limited to Fortran. To use this 7945code transformation, GCC has to be configured with @option{--with-ppl} 7946and @option{--with-cloog} to enable the Graphite loop transformation 7947infrastructure. 7948 7949@item -fgraphite-identity 7950@opindex fgraphite-identity 7951Enable the identity transformation for graphite. For every SCoP we generate 7952the polyhedral representation and transform it back to gimple. Using 7953@option{-fgraphite-identity} we can check the costs or benefits of the 7954GIMPLE -> GRAPHITE -> GIMPLE transformation. Some minimal optimizations 7955are also performed by the code generator CLooG, like index splitting and 7956dead code elimination in loops. 7957 7958@item -floop-nest-optimize 7959@opindex floop-nest-optimize 7960Enable the ISL based loop nest optimizer. This is a generic loop nest 7961optimizer based on the Pluto optimization algorithms. It calculates a loop 7962structure optimized for data-locality and parallelism. This option 7963is experimental. 7964 7965@item -floop-parallelize-all 7966@opindex floop-parallelize-all 7967Use the Graphite data dependence analysis to identify loops that can 7968be parallelized. Parallelize all the loops that can be analyzed to 7969not contain loop carried dependences without checking that it is 7970profitable to parallelize the loops. 7971 7972@item -fcheck-data-deps 7973@opindex fcheck-data-deps 7974Compare the results of several data dependence analyzers. This option 7975is used for debugging the data dependence analyzers. 7976 7977@item -ftree-loop-if-convert 7978Attempt to transform conditional jumps in the innermost loops to 7979branch-less equivalents. The intent is to remove control-flow from 7980the innermost loops in order to improve the ability of the 7981vectorization pass to handle these loops. This is enabled by default 7982if vectorization is enabled. 7983 7984@item -ftree-loop-if-convert-stores 7985Attempt to also if-convert conditional jumps containing memory writes. 7986This transformation can be unsafe for multi-threaded programs as it 7987transforms conditional memory writes into unconditional memory writes. 7988For example, 7989@smallexample 7990for (i = 0; i < N; i++) 7991 if (cond) 7992 A[i] = expr; 7993@end smallexample 7994is transformed to 7995@smallexample 7996for (i = 0; i < N; i++) 7997 A[i] = cond ? expr : A[i]; 7998@end smallexample 7999potentially producing data races. 8000 8001@item -ftree-loop-distribution 8002Perform loop distribution. This flag can improve cache performance on 8003big loop bodies and allow further loop optimizations, like 8004parallelization or vectorization, to take place. For example, the loop 8005@smallexample 8006DO I = 1, N 8007 A(I) = B(I) + C 8008 D(I) = E(I) * F 8009ENDDO 8010@end smallexample 8011is transformed to 8012@smallexample 8013DO I = 1, N 8014 A(I) = B(I) + C 8015ENDDO 8016DO I = 1, N 8017 D(I) = E(I) * F 8018ENDDO 8019@end smallexample 8020 8021@item -ftree-loop-distribute-patterns 8022Perform loop distribution of patterns that can be code generated with 8023calls to a library. This flag is enabled by default at @option{-O3}. 8024 8025This pass distributes the initialization loops and generates a call to 8026memset zero. For example, the loop 8027@smallexample 8028DO I = 1, N 8029 A(I) = 0 8030 B(I) = A(I) + I 8031ENDDO 8032@end smallexample 8033is transformed to 8034@smallexample 8035DO I = 1, N 8036 A(I) = 0 8037ENDDO 8038DO I = 1, N 8039 B(I) = A(I) + I 8040ENDDO 8041@end smallexample 8042and the initialization loop is transformed into a call to memset zero. 8043 8044@item -ftree-loop-im 8045@opindex ftree-loop-im 8046Perform loop invariant motion on trees. This pass moves only invariants that 8047are hard to handle at RTL level (function calls, operations that expand to 8048nontrivial sequences of insns). With @option{-funswitch-loops} it also moves 8049operands of conditions that are invariant out of the loop, so that we can use 8050just trivial invariantness analysis in loop unswitching. The pass also includes 8051store motion. 8052 8053@item -ftree-loop-ivcanon 8054@opindex ftree-loop-ivcanon 8055Create a canonical counter for number of iterations in loops for which 8056determining number of iterations requires complicated analysis. Later 8057optimizations then may determine the number easily. Useful especially 8058in connection with unrolling. 8059 8060@item -fivopts 8061@opindex fivopts 8062Perform induction variable optimizations (strength reduction, induction 8063variable merging and induction variable elimination) on trees. 8064 8065@item -ftree-parallelize-loops=n 8066@opindex ftree-parallelize-loops 8067Parallelize loops, i.e., split their iteration space to run in n threads. 8068This is only possible for loops whose iterations are independent 8069and can be arbitrarily reordered. The optimization is only 8070profitable on multiprocessor machines, for loops that are CPU-intensive, 8071rather than constrained e.g.@: by memory bandwidth. This option 8072implies @option{-pthread}, and thus is only supported on targets 8073that have support for @option{-pthread}. 8074 8075@item -ftree-pta 8076@opindex ftree-pta 8077Perform function-local points-to analysis on trees. This flag is 8078enabled by default at @option{-O} and higher. 8079 8080@item -ftree-sra 8081@opindex ftree-sra 8082Perform scalar replacement of aggregates. This pass replaces structure 8083references with scalars to prevent committing structures to memory too 8084early. This flag is enabled by default at @option{-O} and higher. 8085 8086@item -ftree-copyrename 8087@opindex ftree-copyrename 8088Perform copy renaming on trees. This pass attempts to rename compiler 8089temporaries to other variables at copy locations, usually resulting in 8090variable names which more closely resemble the original variables. This flag 8091is enabled by default at @option{-O} and higher. 8092 8093@item -ftree-coalesce-inlined-vars 8094@opindex ftree-coalesce-inlined-vars 8095Tell the copyrename pass (see @option{-ftree-copyrename}) to attempt to 8096combine small user-defined variables too, but only if they were inlined 8097from other functions. It is a more limited form of 8098@option{-ftree-coalesce-vars}. This may harm debug information of such 8099inlined variables, but it will keep variables of the inlined-into 8100function apart from each other, such that they are more likely to 8101contain the expected values in a debugging session. This was the 8102default in GCC versions older than 4.7. 8103 8104@item -ftree-coalesce-vars 8105@opindex ftree-coalesce-vars 8106Tell the copyrename pass (see @option{-ftree-copyrename}) to attempt to 8107combine small user-defined variables too, instead of just compiler 8108temporaries. This may severely limit the ability to debug an optimized 8109program compiled with @option{-fno-var-tracking-assignments}. In the 8110negated form, this flag prevents SSA coalescing of user variables, 8111including inlined ones. This option is enabled by default. 8112 8113@item -ftree-ter 8114@opindex ftree-ter 8115Perform temporary expression replacement during the SSA->normal phase. Single 8116use/single def temporaries are replaced at their use location with their 8117defining expression. This results in non-GIMPLE code, but gives the expanders 8118much more complex trees to work on resulting in better RTL generation. This is 8119enabled by default at @option{-O} and higher. 8120 8121@item -ftree-slsr 8122@opindex ftree-slsr 8123Perform straight-line strength reduction on trees. This recognizes related 8124expressions involving multiplications and replaces them by less expensive 8125calculations when possible. This is enabled by default at @option{-O} and 8126higher. 8127 8128@item -ftree-vectorize 8129@opindex ftree-vectorize 8130Perform vectorization on trees. This flag enables @option{-ftree-loop-vectorize} 8131and @option{-ftree-slp-vectorize} if not explicitly specified. 8132 8133@item -ftree-loop-vectorize 8134@opindex ftree-loop-vectorize 8135Perform loop vectorization on trees. This flag is enabled by default at 8136@option{-O3} and when @option{-ftree-vectorize} is enabled. 8137 8138@item -ftree-slp-vectorize 8139@opindex ftree-slp-vectorize 8140Perform basic block vectorization on trees. This flag is enabled by default at 8141@option{-O3} and when @option{-ftree-vectorize} is enabled. 8142 8143@item -fvect-cost-model=@var{model} 8144@opindex fvect-cost-model 8145Alter the cost model used for vectorization. The @var{model} argument 8146should be one of @code{unlimited}, @code{dynamic} or @code{cheap}. 8147With the @code{unlimited} model the vectorized code-path is assumed 8148to be profitable while with the @code{dynamic} model a runtime check 8149will guard the vectorized code-path to enable it only for iteration 8150counts that will likely execute faster than when executing the original 8151scalar loop. The @code{cheap} model will disable vectorization of 8152loops where doing so would be cost prohibitive for example due to 8153required runtime checks for data dependence or alignment but otherwise 8154is equal to the @code{dynamic} model. 8155The default cost model depends on other optimization flags and is 8156either @code{dynamic} or @code{cheap}. 8157 8158@item -fsimd-cost-model=@var{model} 8159@opindex fsimd-cost-model 8160Alter the cost model used for vectorization of loops marked with the OpenMP 8161or Cilk Plus simd directive. The @var{model} argument should be one of 8162@code{unlimited}, @code{dynamic}, @code{cheap}. All values of @var{model} 8163have the same meaning as described in @option{-fvect-cost-model} and by 8164default a cost model defined with @option{-fvect-cost-model} is used. 8165 8166@item -ftree-vrp 8167@opindex ftree-vrp 8168Perform Value Range Propagation on trees. This is similar to the 8169constant propagation pass, but instead of values, ranges of values are 8170propagated. This allows the optimizers to remove unnecessary range 8171checks like array bound checks and null pointer checks. This is 8172enabled by default at @option{-O2} and higher. Null pointer check 8173elimination is only done if @option{-fdelete-null-pointer-checks} is 8174enabled. 8175 8176@item -ftracer 8177@opindex ftracer 8178Perform tail duplication to enlarge superblock size. This transformation 8179simplifies the control flow of the function allowing other optimizations to do 8180a better job. 8181 8182@item -funroll-loops 8183@opindex funroll-loops 8184Unroll loops whose number of iterations can be determined at compile 8185time or upon entry to the loop. @option{-funroll-loops} implies 8186@option{-frerun-cse-after-loop}. This option makes code larger, 8187and may or may not make it run faster. 8188 8189@item -funroll-all-loops 8190@opindex funroll-all-loops 8191Unroll all loops, even if their number of iterations is uncertain when 8192the loop is entered. This usually makes programs run more slowly. 8193@option{-funroll-all-loops} implies the same options as 8194@option{-funroll-loops}, 8195 8196@item -fsplit-ivs-in-unroller 8197@opindex fsplit-ivs-in-unroller 8198Enables expression of values of induction variables in later iterations 8199of the unrolled loop using the value in the first iteration. This breaks 8200long dependency chains, thus improving efficiency of the scheduling passes. 8201 8202A combination of @option{-fweb} and CSE is often sufficient to obtain the 8203same effect. However, that is not reliable in cases where the loop body 8204is more complicated than a single basic block. It also does not work at all 8205on some architectures due to restrictions in the CSE pass. 8206 8207This optimization is enabled by default. 8208 8209@item -fvariable-expansion-in-unroller 8210@opindex fvariable-expansion-in-unroller 8211With this option, the compiler creates multiple copies of some 8212local variables when unrolling a loop, which can result in superior code. 8213 8214@item -fpartial-inlining 8215@opindex fpartial-inlining 8216Inline parts of functions. This option has any effect only 8217when inlining itself is turned on by the @option{-finline-functions} 8218or @option{-finline-small-functions} options. 8219 8220Enabled at level @option{-O2}. 8221 8222@item -fpredictive-commoning 8223@opindex fpredictive-commoning 8224Perform predictive commoning optimization, i.e., reusing computations 8225(especially memory loads and stores) performed in previous 8226iterations of loops. 8227 8228This option is enabled at level @option{-O3}. 8229 8230@item -fprefetch-loop-arrays 8231@opindex fprefetch-loop-arrays 8232If supported by the target machine, generate instructions to prefetch 8233memory to improve the performance of loops that access large arrays. 8234 8235This option may generate better or worse code; results are highly 8236dependent on the structure of loops within the source code. 8237 8238Disabled at level @option{-Os}. 8239 8240@item -fno-peephole 8241@itemx -fno-peephole2 8242@opindex fno-peephole 8243@opindex fno-peephole2 8244Disable any machine-specific peephole optimizations. The difference 8245between @option{-fno-peephole} and @option{-fno-peephole2} is in how they 8246are implemented in the compiler; some targets use one, some use the 8247other, a few use both. 8248 8249@option{-fpeephole} is enabled by default. 8250@option{-fpeephole2} enabled at levels @option{-O2}, @option{-O3}, @option{-Os}. 8251 8252@item -fno-guess-branch-probability 8253@opindex fno-guess-branch-probability 8254Do not guess branch probabilities using heuristics. 8255 8256GCC uses heuristics to guess branch probabilities if they are 8257not provided by profiling feedback (@option{-fprofile-arcs}). These 8258heuristics are based on the control flow graph. If some branch probabilities 8259are specified by @samp{__builtin_expect}, then the heuristics are 8260used to guess branch probabilities for the rest of the control flow graph, 8261taking the @samp{__builtin_expect} info into account. The interactions 8262between the heuristics and @samp{__builtin_expect} can be complex, and in 8263some cases, it may be useful to disable the heuristics so that the effects 8264of @samp{__builtin_expect} are easier to understand. 8265 8266The default is @option{-fguess-branch-probability} at levels 8267@option{-O}, @option{-O2}, @option{-O3}, @option{-Os}. 8268 8269@item -freorder-blocks 8270@opindex freorder-blocks 8271Reorder basic blocks in the compiled function in order to reduce number of 8272taken branches and improve code locality. 8273 8274Enabled at levels @option{-O2}, @option{-O3}. 8275 8276@item -freorder-blocks-and-partition 8277@opindex freorder-blocks-and-partition 8278In addition to reordering basic blocks in the compiled function, in order 8279to reduce number of taken branches, partitions hot and cold basic blocks 8280into separate sections of the assembly and .o files, to improve 8281paging and cache locality performance. 8282 8283This optimization is automatically turned off in the presence of 8284exception handling, for linkonce sections, for functions with a user-defined 8285section attribute and on any architecture that does not support named 8286sections. 8287 8288Enabled for x86 at levels @option{-O2}, @option{-O3}. 8289 8290@item -freorder-functions 8291@opindex freorder-functions 8292Reorder functions in the object file in order to 8293improve code locality. This is implemented by using special 8294subsections @code{.text.hot} for most frequently executed functions and 8295@code{.text.unlikely} for unlikely executed functions. Reordering is done by 8296the linker so object file format must support named sections and linker must 8297place them in a reasonable way. 8298 8299Also profile feedback must be available to make this option effective. See 8300@option{-fprofile-arcs} for details. 8301 8302Enabled at levels @option{-O2}, @option{-O3}, @option{-Os}. 8303 8304@item -fstrict-aliasing 8305@opindex fstrict-aliasing 8306Allow the compiler to assume the strictest aliasing rules applicable to 8307the language being compiled. For C (and C++), this activates 8308optimizations based on the type of expressions. In particular, an 8309object of one type is assumed never to reside at the same address as an 8310object of a different type, unless the types are almost the same. For 8311example, an @code{unsigned int} can alias an @code{int}, but not a 8312@code{void*} or a @code{double}. A character type may alias any other 8313type. 8314 8315@anchor{Type-punning}Pay special attention to code like this: 8316@smallexample 8317union a_union @{ 8318 int i; 8319 double d; 8320@}; 8321 8322int f() @{ 8323 union a_union t; 8324 t.d = 3.0; 8325 return t.i; 8326@} 8327@end smallexample 8328The practice of reading from a different union member than the one most 8329recently written to (called ``type-punning'') is common. Even with 8330@option{-fstrict-aliasing}, type-punning is allowed, provided the memory 8331is accessed through the union type. So, the code above works as 8332expected. @xref{Structures unions enumerations and bit-fields 8333implementation}. However, this code might not: 8334@smallexample 8335int f() @{ 8336 union a_union t; 8337 int* ip; 8338 t.d = 3.0; 8339 ip = &t.i; 8340 return *ip; 8341@} 8342@end smallexample 8343 8344Similarly, access by taking the address, casting the resulting pointer 8345and dereferencing the result has undefined behavior, even if the cast 8346uses a union type, e.g.: 8347@smallexample 8348int f() @{ 8349 double d = 3.0; 8350 return ((union a_union *) &d)->i; 8351@} 8352@end smallexample 8353 8354The @option{-fstrict-aliasing} option is enabled at levels 8355@option{-O2}, @option{-O3}, @option{-Os}. 8356 8357@item -fstrict-overflow 8358@opindex fstrict-overflow 8359Allow the compiler to assume strict signed overflow rules, depending 8360on the language being compiled. For C (and C++) this means that 8361overflow when doing arithmetic with signed numbers is undefined, which 8362means that the compiler may assume that it does not happen. This 8363permits various optimizations. For example, the compiler assumes 8364that an expression like @code{i + 10 > i} is always true for 8365signed @code{i}. This assumption is only valid if signed overflow is 8366undefined, as the expression is false if @code{i + 10} overflows when 8367using twos complement arithmetic. When this option is in effect any 8368attempt to determine whether an operation on signed numbers 8369overflows must be written carefully to not actually involve overflow. 8370 8371This option also allows the compiler to assume strict pointer 8372semantics: given a pointer to an object, if adding an offset to that 8373pointer does not produce a pointer to the same object, the addition is 8374undefined. This permits the compiler to conclude that @code{p + u > 8375p} is always true for a pointer @code{p} and unsigned integer 8376@code{u}. This assumption is only valid because pointer wraparound is 8377undefined, as the expression is false if @code{p + u} overflows using 8378twos complement arithmetic. 8379 8380See also the @option{-fwrapv} option. Using @option{-fwrapv} means 8381that integer signed overflow is fully defined: it wraps. When 8382@option{-fwrapv} is used, there is no difference between 8383@option{-fstrict-overflow} and @option{-fno-strict-overflow} for 8384integers. With @option{-fwrapv} certain types of overflow are 8385permitted. For example, if the compiler gets an overflow when doing 8386arithmetic on constants, the overflowed value can still be used with 8387@option{-fwrapv}, but not otherwise. 8388 8389The @option{-fstrict-overflow} option is enabled at levels 8390@option{-O2}, @option{-O3}, @option{-Os}. 8391 8392@item -falign-functions 8393@itemx -falign-functions=@var{n} 8394@opindex falign-functions 8395Align the start of functions to the next power-of-two greater than 8396@var{n}, skipping up to @var{n} bytes. For instance, 8397@option{-falign-functions=32} aligns functions to the next 32-byte 8398boundary, but @option{-falign-functions=24} aligns to the next 839932-byte boundary only if this can be done by skipping 23 bytes or less. 8400 8401@option{-fno-align-functions} and @option{-falign-functions=1} are 8402equivalent and mean that functions are not aligned. 8403 8404Some assemblers only support this flag when @var{n} is a power of two; 8405in that case, it is rounded up. 8406 8407If @var{n} is not specified or is zero, use a machine-dependent default. 8408 8409Enabled at levels @option{-O2}, @option{-O3}. 8410 8411@item -falign-labels 8412@itemx -falign-labels=@var{n} 8413@opindex falign-labels 8414Align all branch targets to a power-of-two boundary, skipping up to 8415@var{n} bytes like @option{-falign-functions}. This option can easily 8416make code slower, because it must insert dummy operations for when the 8417branch target is reached in the usual flow of the code. 8418 8419@option{-fno-align-labels} and @option{-falign-labels=1} are 8420equivalent and mean that labels are not aligned. 8421 8422If @option{-falign-loops} or @option{-falign-jumps} are applicable and 8423are greater than this value, then their values are used instead. 8424 8425If @var{n} is not specified or is zero, use a machine-dependent default 8426which is very likely to be @samp{1}, meaning no alignment. 8427 8428Enabled at levels @option{-O2}, @option{-O3}. 8429 8430@item -falign-loops 8431@itemx -falign-loops=@var{n} 8432@opindex falign-loops 8433Align loops to a power-of-two boundary, skipping up to @var{n} bytes 8434like @option{-falign-functions}. If the loops are 8435executed many times, this makes up for any execution of the dummy 8436operations. 8437 8438@option{-fno-align-loops} and @option{-falign-loops=1} are 8439equivalent and mean that loops are not aligned. 8440 8441If @var{n} is not specified or is zero, use a machine-dependent default. 8442 8443Enabled at levels @option{-O2}, @option{-O3}. 8444 8445@item -falign-jumps 8446@itemx -falign-jumps=@var{n} 8447@opindex falign-jumps 8448Align branch targets to a power-of-two boundary, for branch targets 8449where the targets can only be reached by jumping, skipping up to @var{n} 8450bytes like @option{-falign-functions}. In this case, no dummy operations 8451need be executed. 8452 8453@option{-fno-align-jumps} and @option{-falign-jumps=1} are 8454equivalent and mean that loops are not aligned. 8455 8456If @var{n} is not specified or is zero, use a machine-dependent default. 8457 8458Enabled at levels @option{-O2}, @option{-O3}. 8459 8460@item -funit-at-a-time 8461@opindex funit-at-a-time 8462This option is left for compatibility reasons. @option{-funit-at-a-time} 8463has no effect, while @option{-fno-unit-at-a-time} implies 8464@option{-fno-toplevel-reorder} and @option{-fno-section-anchors}. 8465 8466Enabled by default. 8467 8468@item -fno-toplevel-reorder 8469@opindex fno-toplevel-reorder 8470Do not reorder top-level functions, variables, and @code{asm} 8471statements. Output them in the same order that they appear in the 8472input file. When this option is used, unreferenced static variables 8473are not removed. This option is intended to support existing code 8474that relies on a particular ordering. For new code, it is better to 8475use attributes when possible. 8476 8477Enabled at level @option{-O0}. When disabled explicitly, it also implies 8478@option{-fno-section-anchors}, which is otherwise enabled at @option{-O0} on some 8479targets. 8480 8481@item -fweb 8482@opindex fweb 8483Constructs webs as commonly used for register allocation purposes and assign 8484each web individual pseudo register. This allows the register allocation pass 8485to operate on pseudos directly, but also strengthens several other optimization 8486passes, such as CSE, loop optimizer and trivial dead code remover. It can, 8487however, make debugging impossible, since variables no longer stay in a 8488``home register''. 8489 8490Enabled by default with @option{-funroll-loops}. 8491 8492@item -fwhole-program 8493@opindex fwhole-program 8494Assume that the current compilation unit represents the whole program being 8495compiled. All public functions and variables with the exception of @code{main} 8496and those merged by attribute @code{externally_visible} become static functions 8497and in effect are optimized more aggressively by interprocedural optimizers. 8498 8499This option should not be used in combination with @code{-flto}. 8500Instead relying on a linker plugin should provide safer and more precise 8501information. 8502 8503@item -flto[=@var{n}] 8504@opindex flto 8505This option runs the standard link-time optimizer. When invoked 8506with source code, it generates GIMPLE (one of GCC's internal 8507representations) and writes it to special ELF sections in the object 8508file. When the object files are linked together, all the function 8509bodies are read from these ELF sections and instantiated as if they 8510had been part of the same translation unit. 8511 8512To use the link-time optimizer, @option{-flto} and optimization 8513options should be specified at compile time and during the final link. 8514For example: 8515 8516@smallexample 8517gcc -c -O2 -flto foo.c 8518gcc -c -O2 -flto bar.c 8519gcc -o myprog -flto -O2 foo.o bar.o 8520@end smallexample 8521 8522The first two invocations to GCC save a bytecode representation 8523of GIMPLE into special ELF sections inside @file{foo.o} and 8524@file{bar.o}. The final invocation reads the GIMPLE bytecode from 8525@file{foo.o} and @file{bar.o}, merges the two files into a single 8526internal image, and compiles the result as usual. Since both 8527@file{foo.o} and @file{bar.o} are merged into a single image, this 8528causes all the interprocedural analyses and optimizations in GCC to 8529work across the two files as if they were a single one. This means, 8530for example, that the inliner is able to inline functions in 8531@file{bar.o} into functions in @file{foo.o} and vice-versa. 8532 8533Another (simpler) way to enable link-time optimization is: 8534 8535@smallexample 8536gcc -o myprog -flto -O2 foo.c bar.c 8537@end smallexample 8538 8539The above generates bytecode for @file{foo.c} and @file{bar.c}, 8540merges them together into a single GIMPLE representation and optimizes 8541them as usual to produce @file{myprog}. 8542 8543The only important thing to keep in mind is that to enable link-time 8544optimizations you need to use the GCC driver to perform the link-step. 8545GCC then automatically performs link-time optimization if any of the 8546objects involved were compiled with the @option{-flto}. You generally 8547should specify the optimization options to be used for link-time 8548optimization though GCC will try to be clever at guessing an 8549optimization level to use from the options used at compile-time 8550if you fail to specify one at link-time. You can always override 8551the automatic decision to do link-time optimization at link-time 8552by passing @option{-fno-lto} to the link command. 8553 8554To make whole program optimization effective, it is necessary to make 8555certain whole program assumptions. The compiler needs to know 8556what functions and variables can be accessed by libraries and runtime 8557outside of the link-time optimized unit. When supported by the linker, 8558the linker plugin (see @option{-fuse-linker-plugin}) passes information 8559to the compiler about used and externally visible symbols. When 8560the linker plugin is not available, @option{-fwhole-program} should be 8561used to allow the compiler to make these assumptions, which leads 8562to more aggressive optimization decisions. 8563 8564When @option{-fuse-linker-plugin} is not enabled then, when a file is 8565compiled with @option{-flto}, the generated object file is larger than 8566a regular object file because it contains GIMPLE bytecodes and the usual 8567final code (see @option{-ffat-lto-objects}. This means that 8568object files with LTO information can be linked as normal object 8569files; if @option{-fno-lto} is passed to the linker, no 8570interprocedural optimizations are applied. Note that when 8571@option{-fno-fat-lto-objects} is enabled the compile-stage is faster 8572but you cannot perform a regular, non-LTO link on them. 8573 8574Additionally, the optimization flags used to compile individual files 8575are not necessarily related to those used at link time. For instance, 8576 8577@smallexample 8578gcc -c -O0 -ffat-lto-objects -flto foo.c 8579gcc -c -O0 -ffat-lto-objects -flto bar.c 8580gcc -o myprog -O3 foo.o bar.o 8581@end smallexample 8582 8583This produces individual object files with unoptimized assembler 8584code, but the resulting binary @file{myprog} is optimized at 8585@option{-O3}. If, instead, the final binary is generated with 8586@option{-fno-lto}, then @file{myprog} is not optimized. 8587 8588When producing the final binary, GCC only 8589applies link-time optimizations to those files that contain bytecode. 8590Therefore, you can mix and match object files and libraries with 8591GIMPLE bytecodes and final object code. GCC automatically selects 8592which files to optimize in LTO mode and which files to link without 8593further processing. 8594 8595There are some code generation flags preserved by GCC when 8596generating bytecodes, as they need to be used during the final link 8597stage. Generally options specified at link-time override those 8598specified at compile-time. 8599 8600If you do not specify an optimization level option @option{-O} at 8601link-time then GCC will compute one based on the optimization levels 8602used when compiling the object files. The highest optimization 8603level will win here. 8604 8605Currently, the following options and their setting are take from 8606the first object file that explicitely specified it: 8607@option{-fPIC}, @option{-fpic}, @option{-fpie}, @option{-fcommon}, 8608@option{-fexceptions}, @option{-fnon-call-exceptions}, @option{-fgnu-tm} 8609and all the @option{-m} target flags. 8610 8611Certain ABI changing flags are required to match in all compilation-units 8612and trying to override this at link-time with a conflicting value 8613is ignored. This includes options such as @option{-freg-struct-return} 8614and @option{-fpcc-struct-return}. 8615 8616Other options such as @option{-ffp-contract}, @option{-fno-strict-overflow}, 8617@option{-fwrapv}, @option{-fno-trapv} or @option{-fno-strict-aliasing} 8618are passed through to the link stage and merged conservatively for 8619conflicting translation units. Specifically 8620@option{-fno-strict-overflow}, @option{-fwrapv} and @option{-fno-trapv} take 8621precedence and for example @option{-ffp-contract=off} takes precedence 8622over @option{-ffp-contract=fast}. You can override them at linke-time. 8623 8624It is recommended that you compile all the files participating in the 8625same link with the same options and also specify those options at 8626link time. 8627 8628If LTO encounters objects with C linkage declared with incompatible 8629types in separate translation units to be linked together (undefined 8630behavior according to ISO C99 6.2.7), a non-fatal diagnostic may be 8631issued. The behavior is still undefined at run time. Similar 8632diagnostics may be raised for other languages. 8633 8634Another feature of LTO is that it is possible to apply interprocedural 8635optimizations on files written in different languages: 8636 8637@smallexample 8638gcc -c -flto foo.c 8639g++ -c -flto bar.cc 8640gfortran -c -flto baz.f90 8641g++ -o myprog -flto -O3 foo.o bar.o baz.o -lgfortran 8642@end smallexample 8643 8644Notice that the final link is done with @command{g++} to get the C++ 8645runtime libraries and @option{-lgfortran} is added to get the Fortran 8646runtime libraries. In general, when mixing languages in LTO mode, you 8647should use the same link command options as when mixing languages in a 8648regular (non-LTO) compilation. 8649 8650If object files containing GIMPLE bytecode are stored in a library archive, say 8651@file{libfoo.a}, it is possible to extract and use them in an LTO link if you 8652are using a linker with plugin support. To create static libraries suitable 8653for LTO, use @command{gcc-ar} and @command{gcc-ranlib} instead of @command{ar} 8654and @code{ranlib}; to show the symbols of object files with GIMPLE bytecode, use 8655@command{gcc-nm}. Those commands require that @command{ar}, @command{ranlib} 8656and @command{nm} have been compiled with plugin support. At link time, use the the 8657flag @option{-fuse-linker-plugin} to ensure that the library participates in 8658the LTO optimization process: 8659 8660@smallexample 8661gcc -o myprog -O2 -flto -fuse-linker-plugin a.o b.o -lfoo 8662@end smallexample 8663 8664With the linker plugin enabled, the linker extracts the needed 8665GIMPLE files from @file{libfoo.a} and passes them on to the running GCC 8666to make them part of the aggregated GIMPLE image to be optimized. 8667 8668If you are not using a linker with plugin support and/or do not 8669enable the linker plugin, then the objects inside @file{libfoo.a} 8670are extracted and linked as usual, but they do not participate 8671in the LTO optimization process. In order to make a static library suitable 8672for both LTO optimization and usual linkage, compile its object files with 8673@option{-flto} @code{-ffat-lto-objects}. 8674 8675Link-time optimizations do not require the presence of the whole program to 8676operate. If the program does not require any symbols to be exported, it is 8677possible to combine @option{-flto} and @option{-fwhole-program} to allow 8678the interprocedural optimizers to use more aggressive assumptions which may 8679lead to improved optimization opportunities. 8680Use of @option{-fwhole-program} is not needed when linker plugin is 8681active (see @option{-fuse-linker-plugin}). 8682 8683The current implementation of LTO makes no 8684attempt to generate bytecode that is portable between different 8685types of hosts. The bytecode files are versioned and there is a 8686strict version check, so bytecode files generated in one version of 8687GCC will not work with an older or newer version of GCC. 8688 8689Link-time optimization does not work well with generation of debugging 8690information. Combining @option{-flto} with 8691@option{-g} is currently experimental and expected to produce unexpected 8692results. 8693 8694If you specify the optional @var{n}, the optimization and code 8695generation done at link time is executed in parallel using @var{n} 8696parallel jobs by utilizing an installed @command{make} program. The 8697environment variable @env{MAKE} may be used to override the program 8698used. The default value for @var{n} is 1. 8699 8700You can also specify @option{-flto=jobserver} to use GNU make's 8701job server mode to determine the number of parallel jobs. This 8702is useful when the Makefile calling GCC is already executing in parallel. 8703You must prepend a @samp{+} to the command recipe in the parent Makefile 8704for this to work. This option likely only works if @env{MAKE} is 8705GNU make. 8706 8707@item -flto-partition=@var{alg} 8708@opindex flto-partition 8709Specify the partitioning algorithm used by the link-time optimizer. 8710The value is either @code{1to1} to specify a partitioning mirroring 8711the original source files or @code{balanced} to specify partitioning 8712into equally sized chunks (whenever possible) or @code{max} to create 8713new partition for every symbol where possible. Specifying @code{none} 8714as an algorithm disables partitioning and streaming completely. 8715The default value is @code{balanced}. While @code{1to1} can be used 8716as an workaround for various code ordering issues, the @code{max} 8717partitioning is intended for internal testing only. 8718 8719@item -flto-compression-level=@var{n} 8720This option specifies the level of compression used for intermediate 8721language written to LTO object files, and is only meaningful in 8722conjunction with LTO mode (@option{-flto}). Valid 8723values are 0 (no compression) to 9 (maximum compression). Values 8724outside this range are clamped to either 0 or 9. If the option is not 8725given, a default balanced compression setting is used. 8726 8727@item -flto-report 8728Prints a report with internal details on the workings of the link-time 8729optimizer. The contents of this report vary from version to version. 8730It is meant to be useful to GCC developers when processing object 8731files in LTO mode (via @option{-flto}). 8732 8733Disabled by default. 8734 8735@item -flto-report-wpa 8736Like @option{-flto-report}, but only print for the WPA phase of Link 8737Time Optimization. 8738 8739@item -fuse-linker-plugin 8740Enables the use of a linker plugin during link-time optimization. This 8741option relies on plugin support in the linker, which is available in gold 8742or in GNU ld 2.21 or newer. 8743 8744This option enables the extraction of object files with GIMPLE bytecode out 8745of library archives. This improves the quality of optimization by exposing 8746more code to the link-time optimizer. This information specifies what 8747symbols can be accessed externally (by non-LTO object or during dynamic 8748linking). Resulting code quality improvements on binaries (and shared 8749libraries that use hidden visibility) are similar to @code{-fwhole-program}. 8750See @option{-flto} for a description of the effect of this flag and how to 8751use it. 8752 8753This option is enabled by default when LTO support in GCC is enabled 8754and GCC was configured for use with 8755a linker supporting plugins (GNU ld 2.21 or newer or gold). 8756 8757@item -ffat-lto-objects 8758@opindex ffat-lto-objects 8759Fat LTO objects are object files that contain both the intermediate language 8760and the object code. This makes them usable for both LTO linking and normal 8761linking. This option is effective only when compiling with @option{-flto} 8762and is ignored at link time. 8763 8764@option{-fno-fat-lto-objects} improves compilation time over plain LTO, but 8765requires the complete toolchain to be aware of LTO. It requires a linker with 8766linker plugin support for basic functionality. Additionally, 8767@command{nm}, @command{ar} and @command{ranlib} 8768need to support linker plugins to allow a full-featured build environment 8769(capable of building static libraries etc). GCC provides the @command{gcc-ar}, 8770@command{gcc-nm}, @command{gcc-ranlib} wrappers to pass the right options 8771to these tools. With non fat LTO makefiles need to be modified to use them. 8772 8773The default is @option{-fno-fat-lto-objects} on targets with linker plugin 8774support. 8775 8776@item -fcompare-elim 8777@opindex fcompare-elim 8778After register allocation and post-register allocation instruction splitting, 8779identify arithmetic instructions that compute processor flags similar to a 8780comparison operation based on that arithmetic. If possible, eliminate the 8781explicit comparison operation. 8782 8783This pass only applies to certain targets that cannot explicitly represent 8784the comparison operation before register allocation is complete. 8785 8786Enabled at levels @option{-O}, @option{-O2}, @option{-O3}, @option{-Os}. 8787 8788@item -fuse-ld=bfd 8789@opindex fuse-ld=bfd 8790Use the @command{bfd} linker instead of the default linker. 8791 8792@item -fuse-ld=gold 8793@opindex fuse-ld=gold 8794Use the @command{gold} linker instead of the default linker. 8795 8796@item -fcprop-registers 8797@opindex fcprop-registers 8798After register allocation and post-register allocation instruction splitting, 8799perform a copy-propagation pass to try to reduce scheduling dependencies 8800and occasionally eliminate the copy. 8801 8802Enabled at levels @option{-O}, @option{-O2}, @option{-O3}, @option{-Os}. 8803 8804@item -fprofile-correction 8805@opindex fprofile-correction 8806Profiles collected using an instrumented binary for multi-threaded programs may 8807be inconsistent due to missed counter updates. When this option is specified, 8808GCC uses heuristics to correct or smooth out such inconsistencies. By 8809default, GCC emits an error message when an inconsistent profile is detected. 8810 8811@item -fprofile-dir=@var{path} 8812@opindex fprofile-dir 8813 8814Set the directory to search for the profile data files in to @var{path}. 8815This option affects only the profile data generated by 8816@option{-fprofile-generate}, @option{-ftest-coverage}, @option{-fprofile-arcs} 8817and used by @option{-fprofile-use} and @option{-fbranch-probabilities} 8818and its related options. Both absolute and relative paths can be used. 8819By default, GCC uses the current directory as @var{path}, thus the 8820profile data file appears in the same directory as the object file. 8821 8822@item -fprofile-generate 8823@itemx -fprofile-generate=@var{path} 8824@opindex fprofile-generate 8825 8826Enable options usually used for instrumenting application to produce 8827profile useful for later recompilation with profile feedback based 8828optimization. You must use @option{-fprofile-generate} both when 8829compiling and when linking your program. 8830 8831The following options are enabled: @code{-fprofile-arcs}, @code{-fprofile-values}, @code{-fvpt}. 8832 8833If @var{path} is specified, GCC looks at the @var{path} to find 8834the profile feedback data files. See @option{-fprofile-dir}. 8835 8836@item -fprofile-use 8837@itemx -fprofile-use=@var{path} 8838@opindex fprofile-use 8839Enable profile feedback directed optimizations, and optimizations 8840generally profitable only with profile feedback available. 8841 8842The following options are enabled: @code{-fbranch-probabilities}, @code{-fvpt}, 8843@code{-funroll-loops}, @code{-fpeel-loops}, @code{-ftracer}, @code{-ftree-vectorize}, 8844@code{ftree-loop-distribute-patterns} 8845 8846By default, GCC emits an error message if the feedback profiles do not 8847match the source code. This error can be turned into a warning by using 8848@option{-Wcoverage-mismatch}. Note this may result in poorly optimized 8849code. 8850 8851If @var{path} is specified, GCC looks at the @var{path} to find 8852the profile feedback data files. See @option{-fprofile-dir}. 8853@end table 8854 8855The following options control compiler behavior regarding floating-point 8856arithmetic. These options trade off between speed and 8857correctness. All must be specifically enabled. 8858 8859@table @gcctabopt 8860@item -ffloat-store 8861@opindex ffloat-store 8862Do not store floating-point variables in registers, and inhibit other 8863options that might change whether a floating-point value is taken from a 8864register or memory. 8865 8866@cindex floating-point precision 8867This option prevents undesirable excess precision on machines such as 8868the 68000 where the floating registers (of the 68881) keep more 8869precision than a @code{double} is supposed to have. Similarly for the 8870x86 architecture. For most programs, the excess precision does only 8871good, but a few programs rely on the precise definition of IEEE floating 8872point. Use @option{-ffloat-store} for such programs, after modifying 8873them to store all pertinent intermediate computations into variables. 8874 8875@item -fexcess-precision=@var{style} 8876@opindex fexcess-precision 8877This option allows further control over excess precision on machines 8878where floating-point registers have more precision than the IEEE 8879@code{float} and @code{double} types and the processor does not 8880support operations rounding to those types. By default, 8881@option{-fexcess-precision=fast} is in effect; this means that 8882operations are carried out in the precision of the registers and that 8883it is unpredictable when rounding to the types specified in the source 8884code takes place. When compiling C, if 8885@option{-fexcess-precision=standard} is specified then excess 8886precision follows the rules specified in ISO C99; in particular, 8887both casts and assignments cause values to be rounded to their 8888semantic types (whereas @option{-ffloat-store} only affects 8889assignments). This option is enabled by default for C if a strict 8890conformance option such as @option{-std=c99} is used. 8891 8892@opindex mfpmath 8893@option{-fexcess-precision=standard} is not implemented for languages 8894other than C, and has no effect if 8895@option{-funsafe-math-optimizations} or @option{-ffast-math} is 8896specified. On the x86, it also has no effect if @option{-mfpmath=sse} 8897or @option{-mfpmath=sse+387} is specified; in the former case, IEEE 8898semantics apply without excess precision, and in the latter, rounding 8899is unpredictable. 8900 8901@item -ffast-math 8902@opindex ffast-math 8903Sets @option{-fno-math-errno}, @option{-funsafe-math-optimizations}, 8904@option{-ffinite-math-only}, @option{-fno-rounding-math}, 8905@option{-fno-signaling-nans} and @option{-fcx-limited-range}. 8906 8907This option causes the preprocessor macro @code{__FAST_MATH__} to be defined. 8908 8909This option is not turned on by any @option{-O} option besides 8910@option{-Ofast} since it can result in incorrect output for programs 8911that depend on an exact implementation of IEEE or ISO rules/specifications 8912for math functions. It may, however, yield faster code for programs 8913that do not require the guarantees of these specifications. 8914 8915@item -fno-math-errno 8916@opindex fno-math-errno 8917Do not set @code{errno} after calling math functions that are executed 8918with a single instruction, e.g., @code{sqrt}. A program that relies on 8919IEEE exceptions for math error handling may want to use this flag 8920for speed while maintaining IEEE arithmetic compatibility. 8921 8922This option is not turned on by any @option{-O} option since 8923it can result in incorrect output for programs that depend on 8924an exact implementation of IEEE or ISO rules/specifications for 8925math functions. It may, however, yield faster code for programs 8926that do not require the guarantees of these specifications. 8927 8928The default is @option{-fmath-errno}. 8929 8930On Darwin systems, the math library never sets @code{errno}. There is 8931therefore no reason for the compiler to consider the possibility that 8932it might, and @option{-fno-math-errno} is the default. 8933 8934@item -funsafe-math-optimizations 8935@opindex funsafe-math-optimizations 8936 8937Allow optimizations for floating-point arithmetic that (a) assume 8938that arguments and results are valid and (b) may violate IEEE or 8939ANSI standards. When used at link-time, it may include libraries 8940or startup files that change the default FPU control word or other 8941similar optimizations. 8942 8943This option is not turned on by any @option{-O} option since 8944it can result in incorrect output for programs that depend on 8945an exact implementation of IEEE or ISO rules/specifications for 8946math functions. It may, however, yield faster code for programs 8947that do not require the guarantees of these specifications. 8948Enables @option{-fno-signed-zeros}, @option{-fno-trapping-math}, 8949@option{-fassociative-math} and @option{-freciprocal-math}. 8950 8951The default is @option{-fno-unsafe-math-optimizations}. 8952 8953@item -fassociative-math 8954@opindex fassociative-math 8955 8956Allow re-association of operands in series of floating-point operations. 8957This violates the ISO C and C++ language standard by possibly changing 8958computation result. NOTE: re-ordering may change the sign of zero as 8959well as ignore NaNs and inhibit or create underflow or overflow (and 8960thus cannot be used on code that relies on rounding behavior like 8961@code{(x + 2**52) - 2**52}. May also reorder floating-point comparisons 8962and thus may not be used when ordered comparisons are required. 8963This option requires that both @option{-fno-signed-zeros} and 8964@option{-fno-trapping-math} be in effect. Moreover, it doesn't make 8965much sense with @option{-frounding-math}. For Fortran the option 8966is automatically enabled when both @option{-fno-signed-zeros} and 8967@option{-fno-trapping-math} are in effect. 8968 8969The default is @option{-fno-associative-math}. 8970 8971@item -freciprocal-math 8972@opindex freciprocal-math 8973 8974Allow the reciprocal of a value to be used instead of dividing by 8975the value if this enables optimizations. For example @code{x / y} 8976can be replaced with @code{x * (1/y)}, which is useful if @code{(1/y)} 8977is subject to common subexpression elimination. Note that this loses 8978precision and increases the number of flops operating on the value. 8979 8980The default is @option{-fno-reciprocal-math}. 8981 8982@item -ffinite-math-only 8983@opindex ffinite-math-only 8984Allow optimizations for floating-point arithmetic that assume 8985that arguments and results are not NaNs or +-Infs. 8986 8987This option is not turned on by any @option{-O} option since 8988it can result in incorrect output for programs that depend on 8989an exact implementation of IEEE or ISO rules/specifications for 8990math functions. It may, however, yield faster code for programs 8991that do not require the guarantees of these specifications. 8992 8993The default is @option{-fno-finite-math-only}. 8994 8995@item -fno-signed-zeros 8996@opindex fno-signed-zeros 8997Allow optimizations for floating-point arithmetic that ignore the 8998signedness of zero. IEEE arithmetic specifies the behavior of 8999distinct +0.0 and @minus{}0.0 values, which then prohibits simplification 9000of expressions such as x+0.0 or 0.0*x (even with @option{-ffinite-math-only}). 9001This option implies that the sign of a zero result isn't significant. 9002 9003The default is @option{-fsigned-zeros}. 9004 9005@item -fno-trapping-math 9006@opindex fno-trapping-math 9007Compile code assuming that floating-point operations cannot generate 9008user-visible traps. These traps include division by zero, overflow, 9009underflow, inexact result and invalid operation. This option requires 9010that @option{-fno-signaling-nans} be in effect. Setting this option may 9011allow faster code if one relies on ``non-stop'' IEEE arithmetic, for example. 9012 9013This option should never be turned on by any @option{-O} option since 9014it can result in incorrect output for programs that depend on 9015an exact implementation of IEEE or ISO rules/specifications for 9016math functions. 9017 9018The default is @option{-ftrapping-math}. 9019 9020@item -frounding-math 9021@opindex frounding-math 9022Disable transformations and optimizations that assume default floating-point 9023rounding behavior. This is round-to-zero for all floating point 9024to integer conversions, and round-to-nearest for all other arithmetic 9025truncations. This option should be specified for programs that change 9026the FP rounding mode dynamically, or that may be executed with a 9027non-default rounding mode. This option disables constant folding of 9028floating-point expressions at compile time (which may be affected by 9029rounding mode) and arithmetic transformations that are unsafe in the 9030presence of sign-dependent rounding modes. 9031 9032The default is @option{-fno-rounding-math}. 9033 9034This option is experimental and does not currently guarantee to 9035disable all GCC optimizations that are affected by rounding mode. 9036Future versions of GCC may provide finer control of this setting 9037using C99's @code{FENV_ACCESS} pragma. This command-line option 9038will be used to specify the default state for @code{FENV_ACCESS}. 9039 9040@item -fsignaling-nans 9041@opindex fsignaling-nans 9042Compile code assuming that IEEE signaling NaNs may generate user-visible 9043traps during floating-point operations. Setting this option disables 9044optimizations that may change the number of exceptions visible with 9045signaling NaNs. This option implies @option{-ftrapping-math}. 9046 9047This option causes the preprocessor macro @code{__SUPPORT_SNAN__} to 9048be defined. 9049 9050The default is @option{-fno-signaling-nans}. 9051 9052This option is experimental and does not currently guarantee to 9053disable all GCC optimizations that affect signaling NaN behavior. 9054 9055@item -fsingle-precision-constant 9056@opindex fsingle-precision-constant 9057Treat floating-point constants as single precision instead of 9058implicitly converting them to double-precision constants. 9059 9060@item -fcx-limited-range 9061@opindex fcx-limited-range 9062When enabled, this option states that a range reduction step is not 9063needed when performing complex division. Also, there is no checking 9064whether the result of a complex multiplication or division is @code{NaN 9065+ I*NaN}, with an attempt to rescue the situation in that case. The 9066default is @option{-fno-cx-limited-range}, but is enabled by 9067@option{-ffast-math}. 9068 9069This option controls the default setting of the ISO C99 9070@code{CX_LIMITED_RANGE} pragma. Nevertheless, the option applies to 9071all languages. 9072 9073@item -fcx-fortran-rules 9074@opindex fcx-fortran-rules 9075Complex multiplication and division follow Fortran rules. Range 9076reduction is done as part of complex division, but there is no checking 9077whether the result of a complex multiplication or division is @code{NaN 9078+ I*NaN}, with an attempt to rescue the situation in that case. 9079 9080The default is @option{-fno-cx-fortran-rules}. 9081 9082@end table 9083 9084The following options control optimizations that may improve 9085performance, but are not enabled by any @option{-O} options. This 9086section includes experimental options that may produce broken code. 9087 9088@table @gcctabopt 9089@item -fbranch-probabilities 9090@opindex fbranch-probabilities 9091After running a program compiled with @option{-fprofile-arcs} 9092(@pxref{Debugging Options,, Options for Debugging Your Program or 9093@command{gcc}}), you can compile it a second time using 9094@option{-fbranch-probabilities}, to improve optimizations based on 9095the number of times each branch was taken. When a program 9096compiled with @option{-fprofile-arcs} exits, it saves arc execution 9097counts to a file called @file{@var{sourcename}.gcda} for each source 9098file. The information in this data file is very dependent on the 9099structure of the generated code, so you must use the same source code 9100and the same optimization options for both compilations. 9101 9102With @option{-fbranch-probabilities}, GCC puts a 9103@samp{REG_BR_PROB} note on each @samp{JUMP_INSN} and @samp{CALL_INSN}. 9104These can be used to improve optimization. Currently, they are only 9105used in one place: in @file{reorg.c}, instead of guessing which path a 9106branch is most likely to take, the @samp{REG_BR_PROB} values are used to 9107exactly determine which path is taken more often. 9108 9109@item -fprofile-values 9110@opindex fprofile-values 9111If combined with @option{-fprofile-arcs}, it adds code so that some 9112data about values of expressions in the program is gathered. 9113 9114With @option{-fbranch-probabilities}, it reads back the data gathered 9115from profiling values of expressions for usage in optimizations. 9116 9117Enabled with @option{-fprofile-generate} and @option{-fprofile-use}. 9118 9119@item -fprofile-reorder-functions 9120@opindex fprofile-reorder-functions 9121Function reordering based on profile instrumentation collects 9122first time of execution of a function and orders these functions 9123in ascending order. 9124 9125Enabled with @option{-fprofile-use}. 9126 9127@item -fvpt 9128@opindex fvpt 9129If combined with @option{-fprofile-arcs}, this option instructs the compiler 9130to add code to gather information about values of expressions. 9131 9132With @option{-fbranch-probabilities}, it reads back the data gathered 9133and actually performs the optimizations based on them. 9134Currently the optimizations include specialization of division operations 9135using the knowledge about the value of the denominator. 9136 9137@item -frename-registers 9138@opindex frename-registers 9139Attempt to avoid false dependencies in scheduled code by making use 9140of registers left over after register allocation. This optimization 9141most benefits processors with lots of registers. Depending on the 9142debug information format adopted by the target, however, it can 9143make debugging impossible, since variables no longer stay in 9144a ``home register''. 9145 9146Enabled by default with @option{-funroll-loops} and @option{-fpeel-loops}. 9147 9148@item -ftracer 9149@opindex ftracer 9150Perform tail duplication to enlarge superblock size. This transformation 9151simplifies the control flow of the function allowing other optimizations to do 9152a better job. 9153 9154Enabled with @option{-fprofile-use}. 9155 9156@item -funroll-loops 9157@opindex funroll-loops 9158Unroll loops whose number of iterations can be determined at compile time or 9159upon entry to the loop. @option{-funroll-loops} implies 9160@option{-frerun-cse-after-loop}, @option{-fweb} and @option{-frename-registers}. 9161It also turns on complete loop peeling (i.e.@: complete removal of loops with 9162a small constant number of iterations). This option makes code larger, and may 9163or may not make it run faster. 9164 9165Enabled with @option{-fprofile-use}. 9166 9167@item -funroll-all-loops 9168@opindex funroll-all-loops 9169Unroll all loops, even if their number of iterations is uncertain when 9170the loop is entered. This usually makes programs run more slowly. 9171@option{-funroll-all-loops} implies the same options as 9172@option{-funroll-loops}. 9173 9174@item -fpeel-loops 9175@opindex fpeel-loops 9176Peels loops for which there is enough information that they do not 9177roll much (from profile feedback). It also turns on complete loop peeling 9178(i.e.@: complete removal of loops with small constant number of iterations). 9179 9180Enabled with @option{-fprofile-use}. 9181 9182@item -fmove-loop-invariants 9183@opindex fmove-loop-invariants 9184Enables the loop invariant motion pass in the RTL loop optimizer. Enabled 9185at level @option{-O1} 9186 9187@item -funswitch-loops 9188@opindex funswitch-loops 9189Move branches with loop invariant conditions out of the loop, with duplicates 9190of the loop on both branches (modified according to result of the condition). 9191 9192@item -ffunction-sections 9193@itemx -fdata-sections 9194@opindex ffunction-sections 9195@opindex fdata-sections 9196Place each function or data item into its own section in the output 9197file if the target supports arbitrary sections. The name of the 9198function or the name of the data item determines the section's name 9199in the output file. 9200 9201Use these options on systems where the linker can perform optimizations 9202to improve locality of reference in the instruction space. Most systems 9203using the ELF object format and SPARC processors running Solaris 2 have 9204linkers with such optimizations. AIX may have these optimizations in 9205the future. 9206 9207Only use these options when there are significant benefits from doing 9208so. When you specify these options, the assembler and linker 9209create larger object and executable files and are also slower. 9210You cannot use @code{gprof} on all systems if you 9211specify this option, and you may have problems with debugging if 9212you specify both this option and @option{-g}. 9213 9214@item -fbranch-target-load-optimize 9215@opindex fbranch-target-load-optimize 9216Perform branch target register load optimization before prologue / epilogue 9217threading. 9218The use of target registers can typically be exposed only during reload, 9219thus hoisting loads out of loops and doing inter-block scheduling needs 9220a separate optimization pass. 9221 9222@item -fbranch-target-load-optimize2 9223@opindex fbranch-target-load-optimize2 9224Perform branch target register load optimization after prologue / epilogue 9225threading. 9226 9227@item -fbtr-bb-exclusive 9228@opindex fbtr-bb-exclusive 9229When performing branch target register load optimization, don't reuse 9230branch target registers within any basic block. 9231 9232@item -fstack-protector 9233@opindex fstack-protector 9234Emit extra code to check for buffer overflows, such as stack smashing 9235attacks. This is done by adding a guard variable to functions with 9236vulnerable objects. This includes functions that call @code{alloca}, and 9237functions with buffers larger than 8 bytes. The guards are initialized 9238when a function is entered and then checked when the function exits. 9239If a guard check fails, an error message is printed and the program exits. 9240 9241@item -fstack-protector-all 9242@opindex fstack-protector-all 9243Like @option{-fstack-protector} except that all functions are protected. 9244 9245@item -fstack-protector-strong 9246@opindex fstack-protector-strong 9247Like @option{-fstack-protector} but includes additional functions to 9248be protected --- those that have local array definitions, or have 9249references to local frame addresses. 9250 9251@item -fsection-anchors 9252@opindex fsection-anchors 9253Try to reduce the number of symbolic address calculations by using 9254shared ``anchor'' symbols to address nearby objects. This transformation 9255can help to reduce the number of GOT entries and GOT accesses on some 9256targets. 9257 9258For example, the implementation of the following function @code{foo}: 9259 9260@smallexample 9261static int a, b, c; 9262int foo (void) @{ return a + b + c; @} 9263@end smallexample 9264 9265@noindent 9266usually calculates the addresses of all three variables, but if you 9267compile it with @option{-fsection-anchors}, it accesses the variables 9268from a common anchor point instead. The effect is similar to the 9269following pseudocode (which isn't valid C): 9270 9271@smallexample 9272int foo (void) 9273@{ 9274 register int *xr = &x; 9275 return xr[&a - &x] + xr[&b - &x] + xr[&c - &x]; 9276@} 9277@end smallexample 9278 9279Not all targets support this option. 9280 9281@item --param @var{name}=@var{value} 9282@opindex param 9283In some places, GCC uses various constants to control the amount of 9284optimization that is done. For example, GCC does not inline functions 9285that contain more than a certain number of instructions. You can 9286control some of these constants on the command line using the 9287@option{--param} option. 9288 9289The names of specific parameters, and the meaning of the values, are 9290tied to the internals of the compiler, and are subject to change 9291without notice in future releases. 9292 9293In each case, the @var{value} is an integer. The allowable choices for 9294@var{name} are: 9295 9296@table @gcctabopt 9297@item predictable-branch-outcome 9298When branch is predicted to be taken with probability lower than this threshold 9299(in percent), then it is considered well predictable. The default is 10. 9300 9301@item max-crossjump-edges 9302The maximum number of incoming edges to consider for cross-jumping. 9303The algorithm used by @option{-fcrossjumping} is @math{O(N^2)} in 9304the number of edges incoming to each block. Increasing values mean 9305more aggressive optimization, making the compilation time increase with 9306probably small improvement in executable size. 9307 9308@item min-crossjump-insns 9309The minimum number of instructions that must be matched at the end 9310of two blocks before cross-jumping is performed on them. This 9311value is ignored in the case where all instructions in the block being 9312cross-jumped from are matched. The default value is 5. 9313 9314@item max-grow-copy-bb-insns 9315The maximum code size expansion factor when copying basic blocks 9316instead of jumping. The expansion is relative to a jump instruction. 9317The default value is 8. 9318 9319@item max-goto-duplication-insns 9320The maximum number of instructions to duplicate to a block that jumps 9321to a computed goto. To avoid @math{O(N^2)} behavior in a number of 9322passes, GCC factors computed gotos early in the compilation process, 9323and unfactors them as late as possible. Only computed jumps at the 9324end of a basic blocks with no more than max-goto-duplication-insns are 9325unfactored. The default value is 8. 9326 9327@item max-delay-slot-insn-search 9328The maximum number of instructions to consider when looking for an 9329instruction to fill a delay slot. If more than this arbitrary number of 9330instructions are searched, the time savings from filling the delay slot 9331are minimal, so stop searching. Increasing values mean more 9332aggressive optimization, making the compilation time increase with probably 9333small improvement in execution time. 9334 9335@item max-delay-slot-live-search 9336When trying to fill delay slots, the maximum number of instructions to 9337consider when searching for a block with valid live register 9338information. Increasing this arbitrarily chosen value means more 9339aggressive optimization, increasing the compilation time. This parameter 9340should be removed when the delay slot code is rewritten to maintain the 9341control-flow graph. 9342 9343@item max-gcse-memory 9344The approximate maximum amount of memory that can be allocated in 9345order to perform the global common subexpression elimination 9346optimization. If more memory than specified is required, the 9347optimization is not done. 9348 9349@item max-gcse-insertion-ratio 9350If the ratio of expression insertions to deletions is larger than this value 9351for any expression, then RTL PRE inserts or removes the expression and thus 9352leaves partially redundant computations in the instruction stream. The default value is 20. 9353 9354@item max-pending-list-length 9355The maximum number of pending dependencies scheduling allows 9356before flushing the current state and starting over. Large functions 9357with few branches or calls can create excessively large lists which 9358needlessly consume memory and resources. 9359 9360@item max-modulo-backtrack-attempts 9361The maximum number of backtrack attempts the scheduler should make 9362when modulo scheduling a loop. Larger values can exponentially increase 9363compilation time. 9364 9365@item max-inline-insns-single 9366Several parameters control the tree inliner used in GCC@. 9367This number sets the maximum number of instructions (counted in GCC's 9368internal representation) in a single function that the tree inliner 9369considers for inlining. This only affects functions declared 9370inline and methods implemented in a class declaration (C++). 9371The default value is 400. 9372 9373@item max-inline-insns-auto 9374When you use @option{-finline-functions} (included in @option{-O3}), 9375a lot of functions that would otherwise not be considered for inlining 9376by the compiler are investigated. To those functions, a different 9377(more restrictive) limit compared to functions declared inline can 9378be applied. 9379The default value is 40. 9380 9381@item inline-min-speedup 9382When estimated performance improvement of caller + callee runtime exceeds this 9383threshold (in precent), the function can be inlined regardless the limit on 9384@option{--param max-inline-insns-single} and @option{--param 9385max-inline-insns-auto}. 9386 9387@item large-function-insns 9388The limit specifying really large functions. For functions larger than this 9389limit after inlining, inlining is constrained by 9390@option{--param large-function-growth}. This parameter is useful primarily 9391to avoid extreme compilation time caused by non-linear algorithms used by the 9392back end. 9393The default value is 2700. 9394 9395@item large-function-growth 9396Specifies maximal growth of large function caused by inlining in percents. 9397The default value is 100 which limits large function growth to 2.0 times 9398the original size. 9399 9400@item large-unit-insns 9401The limit specifying large translation unit. Growth caused by inlining of 9402units larger than this limit is limited by @option{--param inline-unit-growth}. 9403For small units this might be too tight. 9404For example, consider a unit consisting of function A 9405that is inline and B that just calls A three times. If B is small relative to 9406A, the growth of unit is 300\% and yet such inlining is very sane. For very 9407large units consisting of small inlineable functions, however, the overall unit 9408growth limit is needed to avoid exponential explosion of code size. Thus for 9409smaller units, the size is increased to @option{--param large-unit-insns} 9410before applying @option{--param inline-unit-growth}. The default is 10000. 9411 9412@item inline-unit-growth 9413Specifies maximal overall growth of the compilation unit caused by inlining. 9414The default value is 30 which limits unit growth to 1.3 times the original 9415size. 9416 9417@item ipcp-unit-growth 9418Specifies maximal overall growth of the compilation unit caused by 9419interprocedural constant propagation. The default value is 10 which limits 9420unit growth to 1.1 times the original size. 9421 9422@item large-stack-frame 9423The limit specifying large stack frames. While inlining the algorithm is trying 9424to not grow past this limit too much. The default value is 256 bytes. 9425 9426@item large-stack-frame-growth 9427Specifies maximal growth of large stack frames caused by inlining in percents. 9428The default value is 1000 which limits large stack frame growth to 11 times 9429the original size. 9430 9431@item max-inline-insns-recursive 9432@itemx max-inline-insns-recursive-auto 9433Specifies the maximum number of instructions an out-of-line copy of a 9434self-recursive inline 9435function can grow into by performing recursive inlining. 9436 9437For functions declared inline, @option{--param max-inline-insns-recursive} is 9438taken into account. For functions not declared inline, recursive inlining 9439happens only when @option{-finline-functions} (included in @option{-O3}) is 9440enabled and @option{--param max-inline-insns-recursive-auto} is used. The 9441default value is 450. 9442 9443@item max-inline-recursive-depth 9444@itemx max-inline-recursive-depth-auto 9445Specifies the maximum recursion depth used for recursive inlining. 9446 9447For functions declared inline, @option{--param max-inline-recursive-depth} is 9448taken into account. For functions not declared inline, recursive inlining 9449happens only when @option{-finline-functions} (included in @option{-O3}) is 9450enabled and @option{--param max-inline-recursive-depth-auto} is used. The 9451default value is 8. 9452 9453@item min-inline-recursive-probability 9454Recursive inlining is profitable only for function having deep recursion 9455in average and can hurt for function having little recursion depth by 9456increasing the prologue size or complexity of function body to other 9457optimizers. 9458 9459When profile feedback is available (see @option{-fprofile-generate}) the actual 9460recursion depth can be guessed from probability that function recurses via a 9461given call expression. This parameter limits inlining only to call expressions 9462whose probability exceeds the given threshold (in percents). 9463The default value is 10. 9464 9465@item early-inlining-insns 9466Specify growth that the early inliner can make. In effect it increases 9467the amount of inlining for code having a large abstraction penalty. 9468The default value is 10. 9469 9470@item max-early-inliner-iterations 9471@itemx max-early-inliner-iterations 9472Limit of iterations of the early inliner. This basically bounds 9473the number of nested indirect calls the early inliner can resolve. 9474Deeper chains are still handled by late inlining. 9475 9476@item comdat-sharing-probability 9477@itemx comdat-sharing-probability 9478Probability (in percent) that C++ inline function with comdat visibility 9479are shared across multiple compilation units. The default value is 20. 9480 9481@item min-vect-loop-bound 9482The minimum number of iterations under which loops are not vectorized 9483when @option{-ftree-vectorize} is used. The number of iterations after 9484vectorization needs to be greater than the value specified by this option 9485to allow vectorization. The default value is 0. 9486 9487@item gcse-cost-distance-ratio 9488Scaling factor in calculation of maximum distance an expression 9489can be moved by GCSE optimizations. This is currently supported only in the 9490code hoisting pass. The bigger the ratio, the more aggressive code hoisting 9491is with simple expressions, i.e., the expressions that have cost 9492less than @option{gcse-unrestricted-cost}. Specifying 0 disables 9493hoisting of simple expressions. The default value is 10. 9494 9495@item gcse-unrestricted-cost 9496Cost, roughly measured as the cost of a single typical machine 9497instruction, at which GCSE optimizations do not constrain 9498the distance an expression can travel. This is currently 9499supported only in the code hoisting pass. The lesser the cost, 9500the more aggressive code hoisting is. Specifying 0 9501allows all expressions to travel unrestricted distances. 9502The default value is 3. 9503 9504@item max-hoist-depth 9505The depth of search in the dominator tree for expressions to hoist. 9506This is used to avoid quadratic behavior in hoisting algorithm. 9507The value of 0 does not limit on the search, but may slow down compilation 9508of huge functions. The default value is 30. 9509 9510@item max-tail-merge-comparisons 9511The maximum amount of similar bbs to compare a bb with. This is used to 9512avoid quadratic behavior in tree tail merging. The default value is 10. 9513 9514@item max-tail-merge-iterations 9515The maximum amount of iterations of the pass over the function. This is used to 9516limit compilation time in tree tail merging. The default value is 2. 9517 9518@item max-unrolled-insns 9519The maximum number of instructions that a loop may have to be unrolled. 9520If a loop is unrolled, this parameter also determines how many times 9521the loop code is unrolled. 9522 9523@item max-average-unrolled-insns 9524The maximum number of instructions biased by probabilities of their execution 9525that a loop may have to be unrolled. If a loop is unrolled, 9526this parameter also determines how many times the loop code is unrolled. 9527 9528@item max-unroll-times 9529The maximum number of unrollings of a single loop. 9530 9531@item max-peeled-insns 9532The maximum number of instructions that a loop may have to be peeled. 9533If a loop is peeled, this parameter also determines how many times 9534the loop code is peeled. 9535 9536@item max-peel-times 9537The maximum number of peelings of a single loop. 9538 9539@item max-peel-branches 9540The maximum number of branches on the hot path through the peeled sequence. 9541 9542@item max-completely-peeled-insns 9543The maximum number of insns of a completely peeled loop. 9544 9545@item max-completely-peel-times 9546The maximum number of iterations of a loop to be suitable for complete peeling. 9547 9548@item max-completely-peel-loop-nest-depth 9549The maximum depth of a loop nest suitable for complete peeling. 9550 9551@item max-unswitch-insns 9552The maximum number of insns of an unswitched loop. 9553 9554@item max-unswitch-level 9555The maximum number of branches unswitched in a single loop. 9556 9557@item lim-expensive 9558The minimum cost of an expensive expression in the loop invariant motion. 9559 9560@item iv-consider-all-candidates-bound 9561Bound on number of candidates for induction variables, below which 9562all candidates are considered for each use in induction variable 9563optimizations. If there are more candidates than this, 9564only the most relevant ones are considered to avoid quadratic time complexity. 9565 9566@item iv-max-considered-uses 9567The induction variable optimizations give up on loops that contain more 9568induction variable uses. 9569 9570@item iv-always-prune-cand-set-bound 9571If the number of candidates in the set is smaller than this value, 9572always try to remove unnecessary ivs from the set 9573when adding a new one. 9574 9575@item scev-max-expr-size 9576Bound on size of expressions used in the scalar evolutions analyzer. 9577Large expressions slow the analyzer. 9578 9579@item scev-max-expr-complexity 9580Bound on the complexity of the expressions in the scalar evolutions analyzer. 9581Complex expressions slow the analyzer. 9582 9583@item omega-max-vars 9584The maximum number of variables in an Omega constraint system. 9585The default value is 128. 9586 9587@item omega-max-geqs 9588The maximum number of inequalities in an Omega constraint system. 9589The default value is 256. 9590 9591@item omega-max-eqs 9592The maximum number of equalities in an Omega constraint system. 9593The default value is 128. 9594 9595@item omega-max-wild-cards 9596The maximum number of wildcard variables that the Omega solver is 9597able to insert. The default value is 18. 9598 9599@item omega-hash-table-size 9600The size of the hash table in the Omega solver. The default value is 9601550. 9602 9603@item omega-max-keys 9604The maximal number of keys used by the Omega solver. The default 9605value is 500. 9606 9607@item omega-eliminate-redundant-constraints 9608When set to 1, use expensive methods to eliminate all redundant 9609constraints. The default value is 0. 9610 9611@item vect-max-version-for-alignment-checks 9612The maximum number of run-time checks that can be performed when 9613doing loop versioning for alignment in the vectorizer. 9614 9615@item vect-max-version-for-alias-checks 9616The maximum number of run-time checks that can be performed when 9617doing loop versioning for alias in the vectorizer. 9618 9619@item vect-max-peeling-for-alignment 9620The maximum number of loop peels to enhance access alignment 9621for vectorizer. Value -1 means 'no limit'. 9622 9623@item max-iterations-to-track 9624The maximum number of iterations of a loop the brute-force algorithm 9625for analysis of the number of iterations of the loop tries to evaluate. 9626 9627@item hot-bb-count-ws-permille 9628A basic block profile count is considered hot if it contributes to 9629the given permillage (i.e. 0...1000) of the entire profiled execution. 9630 9631@item hot-bb-frequency-fraction 9632Select fraction of the entry block frequency of executions of basic block in 9633function given basic block needs to have to be considered hot. 9634 9635@item max-predicted-iterations 9636The maximum number of loop iterations we predict statically. This is useful 9637in cases where a function contains a single loop with known bound and 9638another loop with unknown bound. 9639The known number of iterations is predicted correctly, while 9640the unknown number of iterations average to roughly 10. This means that the 9641loop without bounds appears artificially cold relative to the other one. 9642 9643@item builtin-expect-probability 9644Control the probability of the expression having the specified value. This 9645parameter takes a percentage (i.e. 0 ... 100) as input. 9646The default probability of 90 is obtained empirically. 9647 9648@item align-threshold 9649 9650Select fraction of the maximal frequency of executions of a basic block in 9651a function to align the basic block. 9652 9653@item align-loop-iterations 9654 9655A loop expected to iterate at least the selected number of iterations is 9656aligned. 9657 9658@item tracer-dynamic-coverage 9659@itemx tracer-dynamic-coverage-feedback 9660 9661This value is used to limit superblock formation once the given percentage of 9662executed instructions is covered. This limits unnecessary code size 9663expansion. 9664 9665The @option{tracer-dynamic-coverage-feedback} is used only when profile 9666feedback is available. The real profiles (as opposed to statically estimated 9667ones) are much less balanced allowing the threshold to be larger value. 9668 9669@item tracer-max-code-growth 9670Stop tail duplication once code growth has reached given percentage. This is 9671a rather artificial limit, as most of the duplicates are eliminated later in 9672cross jumping, so it may be set to much higher values than is the desired code 9673growth. 9674 9675@item tracer-min-branch-ratio 9676 9677Stop reverse growth when the reverse probability of best edge is less than this 9678threshold (in percent). 9679 9680@item tracer-min-branch-ratio 9681@itemx tracer-min-branch-ratio-feedback 9682 9683Stop forward growth if the best edge has probability lower than this 9684threshold. 9685 9686Similarly to @option{tracer-dynamic-coverage} two values are present, one for 9687compilation for profile feedback and one for compilation without. The value 9688for compilation with profile feedback needs to be more conservative (higher) in 9689order to make tracer effective. 9690 9691@item max-cse-path-length 9692 9693The maximum number of basic blocks on path that CSE considers. 9694The default is 10. 9695 9696@item max-cse-insns 9697The maximum number of instructions CSE processes before flushing. 9698The default is 1000. 9699 9700@item ggc-min-expand 9701 9702GCC uses a garbage collector to manage its own memory allocation. This 9703parameter specifies the minimum percentage by which the garbage 9704collector's heap should be allowed to expand between collections. 9705Tuning this may improve compilation speed; it has no effect on code 9706generation. 9707 9708The default is 30% + 70% * (RAM/1GB) with an upper bound of 100% when 9709RAM >= 1GB@. If @code{getrlimit} is available, the notion of ``RAM'' is 9710the smallest of actual RAM and @code{RLIMIT_DATA} or @code{RLIMIT_AS}. If 9711GCC is not able to calculate RAM on a particular platform, the lower 9712bound of 30% is used. Setting this parameter and 9713@option{ggc-min-heapsize} to zero causes a full collection to occur at 9714every opportunity. This is extremely slow, but can be useful for 9715debugging. 9716 9717@item ggc-min-heapsize 9718 9719Minimum size of the garbage collector's heap before it begins bothering 9720to collect garbage. The first collection occurs after the heap expands 9721by @option{ggc-min-expand}% beyond @option{ggc-min-heapsize}. Again, 9722tuning this may improve compilation speed, and has no effect on code 9723generation. 9724 9725The default is the smaller of RAM/8, RLIMIT_RSS, or a limit that 9726tries to ensure that RLIMIT_DATA or RLIMIT_AS are not exceeded, but 9727with a lower bound of 4096 (four megabytes) and an upper bound of 9728131072 (128 megabytes). If GCC is not able to calculate RAM on a 9729particular platform, the lower bound is used. Setting this parameter 9730very large effectively disables garbage collection. Setting this 9731parameter and @option{ggc-min-expand} to zero causes a full collection 9732to occur at every opportunity. 9733 9734@item max-reload-search-insns 9735The maximum number of instruction reload should look backward for equivalent 9736register. Increasing values mean more aggressive optimization, making the 9737compilation time increase with probably slightly better performance. 9738The default value is 100. 9739 9740@item max-cselib-memory-locations 9741The maximum number of memory locations cselib should take into account. 9742Increasing values mean more aggressive optimization, making the compilation time 9743increase with probably slightly better performance. The default value is 500. 9744 9745@item reorder-blocks-duplicate 9746@itemx reorder-blocks-duplicate-feedback 9747 9748Used by the basic block reordering pass to decide whether to use unconditional 9749branch or duplicate the code on its destination. Code is duplicated when its 9750estimated size is smaller than this value multiplied by the estimated size of 9751unconditional jump in the hot spots of the program. 9752 9753The @option{reorder-block-duplicate-feedback} is used only when profile 9754feedback is available. It may be set to higher values than 9755@option{reorder-block-duplicate} since information about the hot spots is more 9756accurate. 9757 9758@item max-sched-ready-insns 9759The maximum number of instructions ready to be issued the scheduler should 9760consider at any given time during the first scheduling pass. Increasing 9761values mean more thorough searches, making the compilation time increase 9762with probably little benefit. The default value is 100. 9763 9764@item max-sched-region-blocks 9765The maximum number of blocks in a region to be considered for 9766interblock scheduling. The default value is 10. 9767 9768@item max-pipeline-region-blocks 9769The maximum number of blocks in a region to be considered for 9770pipelining in the selective scheduler. The default value is 15. 9771 9772@item max-sched-region-insns 9773The maximum number of insns in a region to be considered for 9774interblock scheduling. The default value is 100. 9775 9776@item max-pipeline-region-insns 9777The maximum number of insns in a region to be considered for 9778pipelining in the selective scheduler. The default value is 200. 9779 9780@item min-spec-prob 9781The minimum probability (in percents) of reaching a source block 9782for interblock speculative scheduling. The default value is 40. 9783 9784@item max-sched-extend-regions-iters 9785The maximum number of iterations through CFG to extend regions. 9786A value of 0 (the default) disables region extensions. 9787 9788@item max-sched-insn-conflict-delay 9789The maximum conflict delay for an insn to be considered for speculative motion. 9790The default value is 3. 9791 9792@item sched-spec-prob-cutoff 9793The minimal probability of speculation success (in percents), so that 9794speculative insns are scheduled. 9795The default value is 40. 9796 9797@item sched-spec-state-edge-prob-cutoff 9798The minimum probability an edge must have for the scheduler to save its 9799state across it. 9800The default value is 10. 9801 9802@item sched-mem-true-dep-cost 9803Minimal distance (in CPU cycles) between store and load targeting same 9804memory locations. The default value is 1. 9805 9806@item selsched-max-lookahead 9807The maximum size of the lookahead window of selective scheduling. It is a 9808depth of search for available instructions. 9809The default value is 50. 9810 9811@item selsched-max-sched-times 9812The maximum number of times that an instruction is scheduled during 9813selective scheduling. This is the limit on the number of iterations 9814through which the instruction may be pipelined. The default value is 2. 9815 9816@item selsched-max-insns-to-rename 9817The maximum number of best instructions in the ready list that are considered 9818for renaming in the selective scheduler. The default value is 2. 9819 9820@item sms-min-sc 9821The minimum value of stage count that swing modulo scheduler 9822generates. The default value is 2. 9823 9824@item max-last-value-rtl 9825The maximum size measured as number of RTLs that can be recorded in an expression 9826in combiner for a pseudo register as last known value of that register. The default 9827is 10000. 9828 9829@item integer-share-limit 9830Small integer constants can use a shared data structure, reducing the 9831compiler's memory usage and increasing its speed. This sets the maximum 9832value of a shared integer constant. The default value is 256. 9833 9834@item ssp-buffer-size 9835The minimum size of buffers (i.e.@: arrays) that receive stack smashing 9836protection when @option{-fstack-protection} is used. 9837 9838@item min-size-for-stack-sharing 9839The minimum size of variables taking part in stack slot sharing when not 9840optimizing. The default value is 32. 9841 9842@item max-jump-thread-duplication-stmts 9843Maximum number of statements allowed in a block that needs to be 9844duplicated when threading jumps. 9845 9846@item max-fields-for-field-sensitive 9847Maximum number of fields in a structure treated in 9848a field sensitive manner during pointer analysis. The default is zero 9849for @option{-O0} and @option{-O1}, 9850and 100 for @option{-Os}, @option{-O2}, and @option{-O3}. 9851 9852@item prefetch-latency 9853Estimate on average number of instructions that are executed before 9854prefetch finishes. The distance prefetched ahead is proportional 9855to this constant. Increasing this number may also lead to less 9856streams being prefetched (see @option{simultaneous-prefetches}). 9857 9858@item simultaneous-prefetches 9859Maximum number of prefetches that can run at the same time. 9860 9861@item l1-cache-line-size 9862The size of cache line in L1 cache, in bytes. 9863 9864@item l1-cache-size 9865The size of L1 cache, in kilobytes. 9866 9867@item l2-cache-size 9868The size of L2 cache, in kilobytes. 9869 9870@item min-insn-to-prefetch-ratio 9871The minimum ratio between the number of instructions and the 9872number of prefetches to enable prefetching in a loop. 9873 9874@item prefetch-min-insn-to-mem-ratio 9875The minimum ratio between the number of instructions and the 9876number of memory references to enable prefetching in a loop. 9877 9878@item use-canonical-types 9879Whether the compiler should use the ``canonical'' type system. By 9880default, this should always be 1, which uses a more efficient internal 9881mechanism for comparing types in C++ and Objective-C++. However, if 9882bugs in the canonical type system are causing compilation failures, 9883set this value to 0 to disable canonical types. 9884 9885@item switch-conversion-max-branch-ratio 9886Switch initialization conversion refuses to create arrays that are 9887bigger than @option{switch-conversion-max-branch-ratio} times the number of 9888branches in the switch. 9889 9890@item max-partial-antic-length 9891Maximum length of the partial antic set computed during the tree 9892partial redundancy elimination optimization (@option{-ftree-pre}) when 9893optimizing at @option{-O3} and above. For some sorts of source code 9894the enhanced partial redundancy elimination optimization can run away, 9895consuming all of the memory available on the host machine. This 9896parameter sets a limit on the length of the sets that are computed, 9897which prevents the runaway behavior. Setting a value of 0 for 9898this parameter allows an unlimited set length. 9899 9900@item sccvn-max-scc-size 9901Maximum size of a strongly connected component (SCC) during SCCVN 9902processing. If this limit is hit, SCCVN processing for the whole 9903function is not done and optimizations depending on it are 9904disabled. The default maximum SCC size is 10000. 9905 9906@item sccvn-max-alias-queries-per-access 9907Maximum number of alias-oracle queries we perform when looking for 9908redundancies for loads and stores. If this limit is hit the search 9909is aborted and the load or store is not considered redundant. The 9910number of queries is algorithmically limited to the number of 9911stores on all paths from the load to the function entry. 9912The default maxmimum number of queries is 1000. 9913 9914@item ira-max-loops-num 9915IRA uses regional register allocation by default. If a function 9916contains more loops than the number given by this parameter, only at most 9917the given number of the most frequently-executed loops form regions 9918for regional register allocation. The default value of the 9919parameter is 100. 9920 9921@item ira-max-conflict-table-size 9922Although IRA uses a sophisticated algorithm to compress the conflict 9923table, the table can still require excessive amounts of memory for 9924huge functions. If the conflict table for a function could be more 9925than the size in MB given by this parameter, the register allocator 9926instead uses a faster, simpler, and lower-quality 9927algorithm that does not require building a pseudo-register conflict table. 9928The default value of the parameter is 2000. 9929 9930@item ira-loop-reserved-regs 9931IRA can be used to evaluate more accurate register pressure in loops 9932for decisions to move loop invariants (see @option{-O3}). The number 9933of available registers reserved for some other purposes is given 9934by this parameter. The default value of the parameter is 2, which is 9935the minimal number of registers needed by typical instructions. 9936This value is the best found from numerous experiments. 9937 9938@item loop-invariant-max-bbs-in-loop 9939Loop invariant motion can be very expensive, both in compilation time and 9940in amount of needed compile-time memory, with very large loops. Loops 9941with more basic blocks than this parameter won't have loop invariant 9942motion optimization performed on them. The default value of the 9943parameter is 1000 for @option{-O1} and 10000 for @option{-O2} and above. 9944 9945@item loop-max-datarefs-for-datadeps 9946Building data dapendencies is expensive for very large loops. This 9947parameter limits the number of data references in loops that are 9948considered for data dependence analysis. These large loops are no 9949handled by the optimizations using loop data dependencies. 9950The default value is 1000. 9951 9952@item max-vartrack-size 9953Sets a maximum number of hash table slots to use during variable 9954tracking dataflow analysis of any function. If this limit is exceeded 9955with variable tracking at assignments enabled, analysis for that 9956function is retried without it, after removing all debug insns from 9957the function. If the limit is exceeded even without debug insns, var 9958tracking analysis is completely disabled for the function. Setting 9959the parameter to zero makes it unlimited. 9960 9961@item max-vartrack-expr-depth 9962Sets a maximum number of recursion levels when attempting to map 9963variable names or debug temporaries to value expressions. This trades 9964compilation time for more complete debug information. If this is set too 9965low, value expressions that are available and could be represented in 9966debug information may end up not being used; setting this higher may 9967enable the compiler to find more complex debug expressions, but compile 9968time and memory use may grow. The default is 12. 9969 9970@item min-nondebug-insn-uid 9971Use uids starting at this parameter for nondebug insns. The range below 9972the parameter is reserved exclusively for debug insns created by 9973@option{-fvar-tracking-assignments}, but debug insns may get 9974(non-overlapping) uids above it if the reserved range is exhausted. 9975 9976@item ipa-sra-ptr-growth-factor 9977IPA-SRA replaces a pointer to an aggregate with one or more new 9978parameters only when their cumulative size is less or equal to 9979@option{ipa-sra-ptr-growth-factor} times the size of the original 9980pointer parameter. 9981 9982@item tm-max-aggregate-size 9983When making copies of thread-local variables in a transaction, this 9984parameter specifies the size in bytes after which variables are 9985saved with the logging functions as opposed to save/restore code 9986sequence pairs. This option only applies when using 9987@option{-fgnu-tm}. 9988 9989@item graphite-max-nb-scop-params 9990To avoid exponential effects in the Graphite loop transforms, the 9991number of parameters in a Static Control Part (SCoP) is bounded. The 9992default value is 10 parameters. A variable whose value is unknown at 9993compilation time and defined outside a SCoP is a parameter of the SCoP. 9994 9995@item graphite-max-bbs-per-function 9996To avoid exponential effects in the detection of SCoPs, the size of 9997the functions analyzed by Graphite is bounded. The default value is 9998100 basic blocks. 9999 10000@item loop-block-tile-size 10001Loop blocking or strip mining transforms, enabled with 10002@option{-floop-block} or @option{-floop-strip-mine}, strip mine each 10003loop in the loop nest by a given number of iterations. The strip 10004length can be changed using the @option{loop-block-tile-size} 10005parameter. The default value is 51 iterations. 10006 10007@item ipa-cp-value-list-size 10008IPA-CP attempts to track all possible values and types passed to a function's 10009parameter in order to propagate them and perform devirtualization. 10010@option{ipa-cp-value-list-size} is the maximum number of values and types it 10011stores per one formal parameter of a function. 10012 10013@item ipa-cp-eval-threshold 10014IPA-CP calculates its own score of cloning profitability heuristics 10015and performs those cloning opportunities with scores that exceed 10016@option{ipa-cp-eval-threshold}. 10017 10018@item ipa-max-agg-items 10019IPA-CP is also capable to propagate a number of scalar values passed 10020in an aggregate. @option{ipa-max-agg-items} controls the maximum 10021number of such values per one parameter. 10022 10023@item ipa-cp-loop-hint-bonus 10024When IPA-CP determines that a cloning candidate would make the number 10025of iterations of a loop known, it adds a bonus of 10026@option{ipa-cp-loop-hint-bonus} bonus to the profitability score of 10027the candidate. 10028 10029@item ipa-cp-array-index-hint-bonus 10030When IPA-CP determines that a cloning candidate would make the index of 10031an array access known, it adds a bonus of 10032@option{ipa-cp-array-index-hint-bonus} bonus to the profitability 10033score of the candidate. 10034 10035@item lto-partitions 10036Specify desired number of partitions produced during WHOPR compilation. 10037The number of partitions should exceed the number of CPUs used for compilation. 10038The default value is 32. 10039 10040@item lto-minpartition 10041Size of minimal partition for WHOPR (in estimated instructions). 10042This prevents expenses of splitting very small programs into too many 10043partitions. 10044 10045@item cxx-max-namespaces-for-diagnostic-help 10046The maximum number of namespaces to consult for suggestions when C++ 10047name lookup fails for an identifier. The default is 1000. 10048 10049@item sink-frequency-threshold 10050The maximum relative execution frequency (in percents) of the target block 10051relative to a statement's original block to allow statement sinking of a 10052statement. Larger numbers result in more aggressive statement sinking. 10053The default value is 75. A small positive adjustment is applied for 10054statements with memory operands as those are even more profitable so sink. 10055 10056@item max-stores-to-sink 10057The maximum number of conditional stores paires that can be sunk. Set to 0 10058if either vectorization (@option{-ftree-vectorize}) or if-conversion 10059(@option{-ftree-loop-if-convert}) is disabled. The default is 2. 10060 10061@item allow-load-data-races 10062Allow optimizers to introduce new data races on loads. 10063Set to 1 to allow, otherwise to 0. This option is enabled by default 10064unless implicitly set by the @option{-fmemory-model=} option. 10065 10066@item allow-store-data-races 10067Allow optimizers to introduce new data races on stores. 10068Set to 1 to allow, otherwise to 0. This option is enabled by default 10069unless implicitly set by the @option{-fmemory-model=} option. 10070 10071@item allow-packed-load-data-races 10072Allow optimizers to introduce new data races on packed data loads. 10073Set to 1 to allow, otherwise to 0. This option is enabled by default 10074unless implicitly set by the @option{-fmemory-model=} option. 10075 10076@item allow-packed-store-data-races 10077Allow optimizers to introduce new data races on packed data stores. 10078Set to 1 to allow, otherwise to 0. This option is enabled by default 10079unless implicitly set by the @option{-fmemory-model=} option. 10080 10081@item case-values-threshold 10082The smallest number of different values for which it is best to use a 10083jump-table instead of a tree of conditional branches. If the value is 100840, use the default for the machine. The default is 0. 10085 10086@item tree-reassoc-width 10087Set the maximum number of instructions executed in parallel in 10088reassociated tree. This parameter overrides target dependent 10089heuristics used by default if has non zero value. 10090 10091@item sched-pressure-algorithm 10092Choose between the two available implementations of 10093@option{-fsched-pressure}. Algorithm 1 is the original implementation 10094and is the more likely to prevent instructions from being reordered. 10095Algorithm 2 was designed to be a compromise between the relatively 10096conservative approach taken by algorithm 1 and the rather aggressive 10097approach taken by the default scheduler. It relies more heavily on 10098having a regular register file and accurate register pressure classes. 10099See @file{haifa-sched.c} in the GCC sources for more details. 10100 10101The default choice depends on the target. 10102 10103@item max-slsr-cand-scan 10104Set the maximum number of existing candidates that will be considered when 10105seeking a basis for a new straight-line strength reduction candidate. 10106 10107@item asan-globals 10108Enable buffer overflow detection for global objects. This kind 10109of protection is enabled by default if you are using 10110@option{-fsanitize=address} option. 10111To disable global objects protection use @option{--param asan-globals=0}. 10112 10113@item asan-stack 10114Enable buffer overflow detection for stack objects. This kind of 10115protection is enabled by default when using@option{-fsanitize=address}. 10116To disable stack protection use @option{--param asan-stack=0} option. 10117 10118@item asan-instrument-reads 10119Enable buffer overflow detection for memory reads. This kind of 10120protection is enabled by default when using @option{-fsanitize=address}. 10121To disable memory reads protection use 10122@option{--param asan-instrument-reads=0}. 10123 10124@item asan-instrument-writes 10125Enable buffer overflow detection for memory writes. This kind of 10126protection is enabled by default when using @option{-fsanitize=address}. 10127To disable memory writes protection use 10128@option{--param asan-instrument-writes=0} option. 10129 10130@item asan-memintrin 10131Enable detection for built-in functions. This kind of protection 10132is enabled by default when using @option{-fsanitize=address}. 10133To disable built-in functions protection use 10134@option{--param asan-memintrin=0}. 10135 10136@item asan-use-after-return 10137Enable detection of use-after-return. This kind of protection 10138is enabled by default when using @option{-fsanitize=address} option. 10139To disable use-after-return detection use 10140@option{--param asan-use-after-return=0}. 10141 10142@item asan-instrumentation-with-call-threshold 10143If number of memory accesses in function being instrumented 10144is greater or equal to this number, use callbacks instead of inline checks. 10145E.g. to disable inline code use 10146@option{--param asan-instrumentation-with-call-threshold=0}. 10147 10148@end table 10149@end table 10150 10151@node Preprocessor Options 10152@section Options Controlling the Preprocessor 10153@cindex preprocessor options 10154@cindex options, preprocessor 10155 10156These options control the C preprocessor, which is run on each C source 10157file before actual compilation. 10158 10159If you use the @option{-E} option, nothing is done except preprocessing. 10160Some of these options make sense only together with @option{-E} because 10161they cause the preprocessor output to be unsuitable for actual 10162compilation. 10163 10164@table @gcctabopt 10165@item -Wp,@var{option} 10166@opindex Wp 10167You can use @option{-Wp,@var{option}} to bypass the compiler driver 10168and pass @var{option} directly through to the preprocessor. If 10169@var{option} contains commas, it is split into multiple options at the 10170commas. However, many options are modified, translated or interpreted 10171by the compiler driver before being passed to the preprocessor, and 10172@option{-Wp} forcibly bypasses this phase. The preprocessor's direct 10173interface is undocumented and subject to change, so whenever possible 10174you should avoid using @option{-Wp} and let the driver handle the 10175options instead. 10176 10177@item -Xpreprocessor @var{option} 10178@opindex Xpreprocessor 10179Pass @var{option} as an option to the preprocessor. You can use this to 10180supply system-specific preprocessor options that GCC does not 10181recognize. 10182 10183If you want to pass an option that takes an argument, you must use 10184@option{-Xpreprocessor} twice, once for the option and once for the argument. 10185 10186@item -no-integrated-cpp 10187@opindex no-integrated-cpp 10188Perform preprocessing as a separate pass before compilation. 10189By default, GCC performs preprocessing as an integrated part of 10190input tokenization and parsing. 10191If this option is provided, the appropriate language front end 10192(@command{cc1}, @command{cc1plus}, or @command{cc1obj} for C, C++, 10193and Objective-C, respectively) is instead invoked twice, 10194once for preprocessing only and once for actual compilation 10195of the preprocessed input. 10196This option may be useful in conjunction with the @option{-B} or 10197@option{-wrapper} options to specify an alternate preprocessor or 10198perform additional processing of the program source between 10199normal preprocessing and compilation. 10200@end table 10201 10202@include cppopts.texi 10203 10204@node Assembler Options 10205@section Passing Options to the Assembler 10206 10207@c prevent bad page break with this line 10208You can pass options to the assembler. 10209 10210@table @gcctabopt 10211@item -Wa,@var{option} 10212@opindex Wa 10213Pass @var{option} as an option to the assembler. If @var{option} 10214contains commas, it is split into multiple options at the commas. 10215 10216@item -Xassembler @var{option} 10217@opindex Xassembler 10218Pass @var{option} as an option to the assembler. You can use this to 10219supply system-specific assembler options that GCC does not 10220recognize. 10221 10222If you want to pass an option that takes an argument, you must use 10223@option{-Xassembler} twice, once for the option and once for the argument. 10224 10225@end table 10226 10227@node Link Options 10228@section Options for Linking 10229@cindex link options 10230@cindex options, linking 10231 10232These options come into play when the compiler links object files into 10233an executable output file. They are meaningless if the compiler is 10234not doing a link step. 10235 10236@table @gcctabopt 10237@cindex file names 10238@item @var{object-file-name} 10239A file name that does not end in a special recognized suffix is 10240considered to name an object file or library. (Object files are 10241distinguished from libraries by the linker according to the file 10242contents.) If linking is done, these object files are used as input 10243to the linker. 10244 10245@item -c 10246@itemx -S 10247@itemx -E 10248@opindex c 10249@opindex S 10250@opindex E 10251If any of these options is used, then the linker is not run, and 10252object file names should not be used as arguments. @xref{Overall 10253Options}. 10254 10255@cindex Libraries 10256@item -l@var{library} 10257@itemx -l @var{library} 10258@opindex l 10259Search the library named @var{library} when linking. (The second 10260alternative with the library as a separate argument is only for 10261POSIX compliance and is not recommended.) 10262 10263It makes a difference where in the command you write this option; the 10264linker searches and processes libraries and object files in the order they 10265are specified. Thus, @samp{foo.o -lz bar.o} searches library @samp{z} 10266after file @file{foo.o} but before @file{bar.o}. If @file{bar.o} refers 10267to functions in @samp{z}, those functions may not be loaded. 10268 10269The linker searches a standard list of directories for the library, 10270which is actually a file named @file{lib@var{library}.a}. The linker 10271then uses this file as if it had been specified precisely by name. 10272 10273The directories searched include several standard system directories 10274plus any that you specify with @option{-L}. 10275 10276Normally the files found this way are library files---archive files 10277whose members are object files. The linker handles an archive file by 10278scanning through it for members which define symbols that have so far 10279been referenced but not defined. But if the file that is found is an 10280ordinary object file, it is linked in the usual fashion. The only 10281difference between using an @option{-l} option and specifying a file name 10282is that @option{-l} surrounds @var{library} with @samp{lib} and @samp{.a} 10283and searches several directories. 10284 10285@item -lobjc 10286@opindex lobjc 10287You need this special case of the @option{-l} option in order to 10288link an Objective-C or Objective-C++ program. 10289 10290@item -nostartfiles 10291@opindex nostartfiles 10292Do not use the standard system startup files when linking. 10293The standard system libraries are used normally, unless @option{-nostdlib} 10294or @option{-nodefaultlibs} is used. 10295 10296@item -nodefaultlibs 10297@opindex nodefaultlibs 10298Do not use the standard system libraries when linking. 10299Only the libraries you specify are passed to the linker, and options 10300specifying linkage of the system libraries, such as @code{-static-libgcc} 10301or @code{-shared-libgcc}, are ignored. 10302The standard startup files are used normally, unless @option{-nostartfiles} 10303is used. 10304 10305The compiler may generate calls to @code{memcmp}, 10306@code{memset}, @code{memcpy} and @code{memmove}. 10307These entries are usually resolved by entries in 10308libc. These entry points should be supplied through some other 10309mechanism when this option is specified. 10310 10311@item -nostdlib 10312@opindex nostdlib 10313Do not use the standard system startup files or libraries when linking. 10314No startup files and only the libraries you specify are passed to 10315the linker, and options specifying linkage of the system libraries, such as 10316@code{-static-libgcc} or @code{-shared-libgcc}, are ignored. 10317 10318The compiler may generate calls to @code{memcmp}, @code{memset}, 10319@code{memcpy} and @code{memmove}. 10320These entries are usually resolved by entries in 10321libc. These entry points should be supplied through some other 10322mechanism when this option is specified. 10323 10324@cindex @option{-lgcc}, use with @option{-nostdlib} 10325@cindex @option{-nostdlib} and unresolved references 10326@cindex unresolved references and @option{-nostdlib} 10327@cindex @option{-lgcc}, use with @option{-nodefaultlibs} 10328@cindex @option{-nodefaultlibs} and unresolved references 10329@cindex unresolved references and @option{-nodefaultlibs} 10330One of the standard libraries bypassed by @option{-nostdlib} and 10331@option{-nodefaultlibs} is @file{libgcc.a}, a library of internal subroutines 10332which GCC uses to overcome shortcomings of particular machines, or special 10333needs for some languages. 10334(@xref{Interface,,Interfacing to GCC Output,gccint,GNU Compiler 10335Collection (GCC) Internals}, 10336for more discussion of @file{libgcc.a}.) 10337In most cases, you need @file{libgcc.a} even when you want to avoid 10338other standard libraries. In other words, when you specify @option{-nostdlib} 10339or @option{-nodefaultlibs} you should usually specify @option{-lgcc} as well. 10340This ensures that you have no unresolved references to internal GCC 10341library subroutines. 10342(An example of such an internal subroutine is @samp{__main}, used to ensure C++ 10343constructors are called; @pxref{Collect2,,@code{collect2}, gccint, 10344GNU Compiler Collection (GCC) Internals}.) 10345 10346@item -pie 10347@opindex pie 10348Produce a position independent executable on targets that support it. 10349For predictable results, you must also specify the same set of options 10350used for compilation (@option{-fpie}, @option{-fPIE}, 10351or model suboptions) when you specify this linker option. 10352 10353@item -rdynamic 10354@opindex rdynamic 10355Pass the flag @option{-export-dynamic} to the ELF linker, on targets 10356that support it. This instructs the linker to add all symbols, not 10357only used ones, to the dynamic symbol table. This option is needed 10358for some uses of @code{dlopen} or to allow obtaining backtraces 10359from within a program. 10360 10361@item -s 10362@opindex s 10363Remove all symbol table and relocation information from the executable. 10364 10365@item -static 10366@opindex static 10367On systems that support dynamic linking, this prevents linking with the shared 10368libraries. On other systems, this option has no effect. 10369 10370@item -shared 10371@opindex shared 10372Produce a shared object which can then be linked with other objects to 10373form an executable. Not all systems support this option. For predictable 10374results, you must also specify the same set of options used for compilation 10375(@option{-fpic}, @option{-fPIC}, or model suboptions) when 10376you specify this linker option.@footnote{On some systems, @samp{gcc -shared} 10377needs to build supplementary stub code for constructors to work. On 10378multi-libbed systems, @samp{gcc -shared} must select the correct support 10379libraries to link against. Failing to supply the correct flags may lead 10380to subtle defects. Supplying them in cases where they are not necessary 10381is innocuous.} 10382 10383@item -shared-libgcc 10384@itemx -static-libgcc 10385@opindex shared-libgcc 10386@opindex static-libgcc 10387On systems that provide @file{libgcc} as a shared library, these options 10388force the use of either the shared or static version, respectively. 10389If no shared version of @file{libgcc} was built when the compiler was 10390configured, these options have no effect. 10391 10392There are several situations in which an application should use the 10393shared @file{libgcc} instead of the static version. The most common 10394of these is when the application wishes to throw and catch exceptions 10395across different shared libraries. In that case, each of the libraries 10396as well as the application itself should use the shared @file{libgcc}. 10397 10398Therefore, the G++ and GCJ drivers automatically add 10399@option{-shared-libgcc} whenever you build a shared library or a main 10400executable, because C++ and Java programs typically use exceptions, so 10401this is the right thing to do. 10402 10403If, instead, you use the GCC driver to create shared libraries, you may 10404find that they are not always linked with the shared @file{libgcc}. 10405If GCC finds, at its configuration time, that you have a non-GNU linker 10406or a GNU linker that does not support option @option{--eh-frame-hdr}, 10407it links the shared version of @file{libgcc} into shared libraries 10408by default. Otherwise, it takes advantage of the linker and optimizes 10409away the linking with the shared version of @file{libgcc}, linking with 10410the static version of libgcc by default. This allows exceptions to 10411propagate through such shared libraries, without incurring relocation 10412costs at library load time. 10413 10414However, if a library or main executable is supposed to throw or catch 10415exceptions, you must link it using the G++ or GCJ driver, as appropriate 10416for the languages used in the program, or using the option 10417@option{-shared-libgcc}, such that it is linked with the shared 10418@file{libgcc}. 10419 10420@item -static-libasan 10421@opindex static-libasan 10422When the @option{-fsanitize=address} option is used to link a program, 10423the GCC driver automatically links against @option{libasan}. If 10424@file{libasan} is available as a shared library, and the @option{-static} 10425option is not used, then this links against the shared version of 10426@file{libasan}. The @option{-static-libasan} option directs the GCC 10427driver to link @file{libasan} statically, without necessarily linking 10428other libraries statically. 10429 10430@item -static-libtsan 10431@opindex static-libtsan 10432When the @option{-fsanitize=thread} option is used to link a program, 10433the GCC driver automatically links against @option{libtsan}. If 10434@file{libtsan} is available as a shared library, and the @option{-static} 10435option is not used, then this links against the shared version of 10436@file{libtsan}. The @option{-static-libtsan} option directs the GCC 10437driver to link @file{libtsan} statically, without necessarily linking 10438other libraries statically. 10439 10440@item -static-liblsan 10441@opindex static-liblsan 10442When the @option{-fsanitize=leak} option is used to link a program, 10443the GCC driver automatically links against @option{liblsan}. If 10444@file{liblsan} is available as a shared library, and the @option{-static} 10445option is not used, then this links against the shared version of 10446@file{liblsan}. The @option{-static-liblsan} option directs the GCC 10447driver to link @file{liblsan} statically, without necessarily linking 10448other libraries statically. 10449 10450@item -static-libubsan 10451@opindex static-libubsan 10452When the @option{-fsanitize=undefined} option is used to link a program, 10453the GCC driver automatically links against @option{libubsan}. If 10454@file{libubsan} is available as a shared library, and the @option{-static} 10455option is not used, then this links against the shared version of 10456@file{libubsan}. The @option{-static-libubsan} option directs the GCC 10457driver to link @file{libubsan} statically, without necessarily linking 10458other libraries statically. 10459 10460@item -static-libstdc++ 10461@opindex static-libstdc++ 10462When the @command{g++} program is used to link a C++ program, it 10463normally automatically links against @option{libstdc++}. If 10464@file{libstdc++} is available as a shared library, and the 10465@option{-static} option is not used, then this links against the 10466shared version of @file{libstdc++}. That is normally fine. However, it 10467is sometimes useful to freeze the version of @file{libstdc++} used by 10468the program without going all the way to a fully static link. The 10469@option{-static-libstdc++} option directs the @command{g++} driver to 10470link @file{libstdc++} statically, without necessarily linking other 10471libraries statically. 10472 10473@item -symbolic 10474@opindex symbolic 10475Bind references to global symbols when building a shared object. Warn 10476about any unresolved references (unless overridden by the link editor 10477option @option{-Xlinker -z -Xlinker defs}). Only a few systems support 10478this option. 10479 10480@item -T @var{script} 10481@opindex T 10482@cindex linker script 10483Use @var{script} as the linker script. This option is supported by most 10484systems using the GNU linker. On some targets, such as bare-board 10485targets without an operating system, the @option{-T} option may be required 10486when linking to avoid references to undefined symbols. 10487 10488@item -Xlinker @var{option} 10489@opindex Xlinker 10490Pass @var{option} as an option to the linker. You can use this to 10491supply system-specific linker options that GCC does not recognize. 10492 10493If you want to pass an option that takes a separate argument, you must use 10494@option{-Xlinker} twice, once for the option and once for the argument. 10495For example, to pass @option{-assert definitions}, you must write 10496@option{-Xlinker -assert -Xlinker definitions}. It does not work to write 10497@option{-Xlinker "-assert definitions"}, because this passes the entire 10498string as a single argument, which is not what the linker expects. 10499 10500When using the GNU linker, it is usually more convenient to pass 10501arguments to linker options using the @option{@var{option}=@var{value}} 10502syntax than as separate arguments. For example, you can specify 10503@option{-Xlinker -Map=output.map} rather than 10504@option{-Xlinker -Map -Xlinker output.map}. Other linkers may not support 10505this syntax for command-line options. 10506 10507@item -Wl,@var{option} 10508@opindex Wl 10509Pass @var{option} as an option to the linker. If @var{option} contains 10510commas, it is split into multiple options at the commas. You can use this 10511syntax to pass an argument to the option. 10512For example, @option{-Wl,-Map,output.map} passes @option{-Map output.map} to the 10513linker. When using the GNU linker, you can also get the same effect with 10514@option{-Wl,-Map=output.map}. 10515 10516@item -u @var{symbol} 10517@opindex u 10518Pretend the symbol @var{symbol} is undefined, to force linking of 10519library modules to define it. You can use @option{-u} multiple times with 10520different symbols to force loading of additional library modules. 10521@end table 10522 10523@node Directory Options 10524@section Options for Directory Search 10525@cindex directory options 10526@cindex options, directory search 10527@cindex search path 10528 10529These options specify directories to search for header files, for 10530libraries and for parts of the compiler: 10531 10532@table @gcctabopt 10533@item -I@var{dir} 10534@opindex I 10535Add the directory @var{dir} to the head of the list of directories to be 10536searched for header files. This can be used to override a system header 10537file, substituting your own version, since these directories are 10538searched before the system header file directories. However, you should 10539not use this option to add directories that contain vendor-supplied 10540system header files (use @option{-isystem} for that). If you use more than 10541one @option{-I} option, the directories are scanned in left-to-right 10542order; the standard system directories come after. 10543 10544If a standard system include directory, or a directory specified with 10545@option{-isystem}, is also specified with @option{-I}, the @option{-I} 10546option is ignored. The directory is still searched but as a 10547system directory at its normal position in the system include chain. 10548This is to ensure that GCC's procedure to fix buggy system headers and 10549the ordering for the @code{include_next} directive are not inadvertently changed. 10550If you really need to change the search order for system directories, 10551use the @option{-nostdinc} and/or @option{-isystem} options. 10552 10553@item -iplugindir=@var{dir} 10554@opindex iplugindir= 10555Set the directory to search for plugins that are passed 10556by @option{-fplugin=@var{name}} instead of 10557@option{-fplugin=@var{path}/@var{name}.so}. This option is not meant 10558to be used by the user, but only passed by the driver. 10559 10560@item -iquote@var{dir} 10561@opindex iquote 10562Add the directory @var{dir} to the head of the list of directories to 10563be searched for header files only for the case of @samp{#include 10564"@var{file}"}; they are not searched for @samp{#include <@var{file}>}, 10565otherwise just like @option{-I}. 10566 10567@item -L@var{dir} 10568@opindex L 10569Add directory @var{dir} to the list of directories to be searched 10570for @option{-l}. 10571 10572@item -B@var{prefix} 10573@opindex B 10574This option specifies where to find the executables, libraries, 10575include files, and data files of the compiler itself. 10576 10577The compiler driver program runs one or more of the subprograms 10578@command{cpp}, @command{cc1}, @command{as} and @command{ld}. It tries 10579@var{prefix} as a prefix for each program it tries to run, both with and 10580without @samp{@var{machine}/@var{version}/} (@pxref{Target Options}). 10581 10582For each subprogram to be run, the compiler driver first tries the 10583@option{-B} prefix, if any. If that name is not found, or if @option{-B} 10584is not specified, the driver tries two standard prefixes, 10585@file{/usr/lib/gcc/} and @file{/usr/local/lib/gcc/}. If neither of 10586those results in a file name that is found, the unmodified program 10587name is searched for using the directories specified in your 10588@env{PATH} environment variable. 10589 10590The compiler checks to see if the path provided by the @option{-B} 10591refers to a directory, and if necessary it adds a directory 10592separator character at the end of the path. 10593 10594@option{-B} prefixes that effectively specify directory names also apply 10595to libraries in the linker, because the compiler translates these 10596options into @option{-L} options for the linker. They also apply to 10597include files in the preprocessor, because the compiler translates these 10598options into @option{-isystem} options for the preprocessor. In this case, 10599the compiler appends @samp{include} to the prefix. 10600 10601The runtime support file @file{libgcc.a} can also be searched for using 10602the @option{-B} prefix, if needed. If it is not found there, the two 10603standard prefixes above are tried, and that is all. The file is left 10604out of the link if it is not found by those means. 10605 10606Another way to specify a prefix much like the @option{-B} prefix is to use 10607the environment variable @env{GCC_EXEC_PREFIX}. @xref{Environment 10608Variables}. 10609 10610As a special kludge, if the path provided by @option{-B} is 10611@file{[dir/]stage@var{N}/}, where @var{N} is a number in the range 0 to 106129, then it is replaced by @file{[dir/]include}. This is to help 10613with boot-strapping the compiler. 10614 10615@item -specs=@var{file} 10616@opindex specs 10617Process @var{file} after the compiler reads in the standard @file{specs} 10618file, in order to override the defaults which the @command{gcc} driver 10619program uses when determining what switches to pass to @command{cc1}, 10620@command{cc1plus}, @command{as}, @command{ld}, etc. More than one 10621@option{-specs=@var{file}} can be specified on the command line, and they 10622are processed in order, from left to right. 10623 10624@item --sysroot=@var{dir} 10625@opindex sysroot 10626Use @var{dir} as the logical root directory for headers and libraries. 10627For example, if the compiler normally searches for headers in 10628@file{/usr/include} and libraries in @file{/usr/lib}, it instead 10629searches @file{@var{dir}/usr/include} and @file{@var{dir}/usr/lib}. 10630 10631If you use both this option and the @option{-isysroot} option, then 10632the @option{--sysroot} option applies to libraries, but the 10633@option{-isysroot} option applies to header files. 10634 10635The GNU linker (beginning with version 2.16) has the necessary support 10636for this option. If your linker does not support this option, the 10637header file aspect of @option{--sysroot} still works, but the 10638library aspect does not. 10639 10640@item --no-sysroot-suffix 10641@opindex no-sysroot-suffix 10642For some targets, a suffix is added to the root directory specified 10643with @option{--sysroot}, depending on the other options used, so that 10644headers may for example be found in 10645@file{@var{dir}/@var{suffix}/usr/include} instead of 10646@file{@var{dir}/usr/include}. This option disables the addition of 10647such a suffix. 10648 10649@item -I- 10650@opindex I- 10651This option has been deprecated. Please use @option{-iquote} instead for 10652@option{-I} directories before the @option{-I-} and remove the @option{-I-}. 10653Any directories you specify with @option{-I} options before the @option{-I-} 10654option are searched only for the case of @samp{#include "@var{file}"}; 10655they are not searched for @samp{#include <@var{file}>}. 10656 10657If additional directories are specified with @option{-I} options after 10658the @option{-I-}, these directories are searched for all @samp{#include} 10659directives. (Ordinarily @emph{all} @option{-I} directories are used 10660this way.) 10661 10662In addition, the @option{-I-} option inhibits the use of the current 10663directory (where the current input file came from) as the first search 10664directory for @samp{#include "@var{file}"}. There is no way to 10665override this effect of @option{-I-}. With @option{-I.} you can specify 10666searching the directory that is current when the compiler is 10667invoked. That is not exactly the same as what the preprocessor does 10668by default, but it is often satisfactory. 10669 10670@option{-I-} does not inhibit the use of the standard system directories 10671for header files. Thus, @option{-I-} and @option{-nostdinc} are 10672independent. 10673@end table 10674 10675@c man end 10676 10677@node Spec Files 10678@section Specifying subprocesses and the switches to pass to them 10679@cindex Spec Files 10680 10681@command{gcc} is a driver program. It performs its job by invoking a 10682sequence of other programs to do the work of compiling, assembling and 10683linking. GCC interprets its command-line parameters and uses these to 10684deduce which programs it should invoke, and which command-line options 10685it ought to place on their command lines. This behavior is controlled 10686by @dfn{spec strings}. In most cases there is one spec string for each 10687program that GCC can invoke, but a few programs have multiple spec 10688strings to control their behavior. The spec strings built into GCC can 10689be overridden by using the @option{-specs=} command-line switch to specify 10690a spec file. 10691 10692@dfn{Spec files} are plaintext files that are used to construct spec 10693strings. They consist of a sequence of directives separated by blank 10694lines. The type of directive is determined by the first non-whitespace 10695character on the line, which can be one of the following: 10696 10697@table @code 10698@item %@var{command} 10699Issues a @var{command} to the spec file processor. The commands that can 10700appear here are: 10701 10702@table @code 10703@item %include <@var{file}> 10704@cindex @code{%include} 10705Search for @var{file} and insert its text at the current point in the 10706specs file. 10707 10708@item %include_noerr <@var{file}> 10709@cindex @code{%include_noerr} 10710Just like @samp{%include}, but do not generate an error message if the include 10711file cannot be found. 10712 10713@item %rename @var{old_name} @var{new_name} 10714@cindex @code{%rename} 10715Rename the spec string @var{old_name} to @var{new_name}. 10716 10717@end table 10718 10719@item *[@var{spec_name}]: 10720This tells the compiler to create, override or delete the named spec 10721string. All lines after this directive up to the next directive or 10722blank line are considered to be the text for the spec string. If this 10723results in an empty string then the spec is deleted. (Or, if the 10724spec did not exist, then nothing happens.) Otherwise, if the spec 10725does not currently exist a new spec is created. If the spec does 10726exist then its contents are overridden by the text of this 10727directive, unless the first character of that text is the @samp{+} 10728character, in which case the text is appended to the spec. 10729 10730@item [@var{suffix}]: 10731Creates a new @samp{[@var{suffix}] spec} pair. All lines after this directive 10732and up to the next directive or blank line are considered to make up the 10733spec string for the indicated suffix. When the compiler encounters an 10734input file with the named suffix, it processes the spec string in 10735order to work out how to compile that file. For example: 10736 10737@smallexample 10738.ZZ: 10739z-compile -input %i 10740@end smallexample 10741 10742This says that any input file whose name ends in @samp{.ZZ} should be 10743passed to the program @samp{z-compile}, which should be invoked with the 10744command-line switch @option{-input} and with the result of performing the 10745@samp{%i} substitution. (See below.) 10746 10747As an alternative to providing a spec string, the text following a 10748suffix directive can be one of the following: 10749 10750@table @code 10751@item @@@var{language} 10752This says that the suffix is an alias for a known @var{language}. This is 10753similar to using the @option{-x} command-line switch to GCC to specify a 10754language explicitly. For example: 10755 10756@smallexample 10757.ZZ: 10758@@c++ 10759@end smallexample 10760 10761Says that .ZZ files are, in fact, C++ source files. 10762 10763@item #@var{name} 10764This causes an error messages saying: 10765 10766@smallexample 10767@var{name} compiler not installed on this system. 10768@end smallexample 10769@end table 10770 10771GCC already has an extensive list of suffixes built into it. 10772This directive adds an entry to the end of the list of suffixes, but 10773since the list is searched from the end backwards, it is effectively 10774possible to override earlier entries using this technique. 10775 10776@end table 10777 10778GCC has the following spec strings built into it. Spec files can 10779override these strings or create their own. Note that individual 10780targets can also add their own spec strings to this list. 10781 10782@smallexample 10783asm Options to pass to the assembler 10784asm_final Options to pass to the assembler post-processor 10785cpp Options to pass to the C preprocessor 10786cc1 Options to pass to the C compiler 10787cc1plus Options to pass to the C++ compiler 10788endfile Object files to include at the end of the link 10789link Options to pass to the linker 10790lib Libraries to include on the command line to the linker 10791libgcc Decides which GCC support library to pass to the linker 10792linker Sets the name of the linker 10793predefines Defines to be passed to the C preprocessor 10794signed_char Defines to pass to CPP to say whether @code{char} is signed 10795 by default 10796startfile Object files to include at the start of the link 10797@end smallexample 10798 10799Here is a small example of a spec file: 10800 10801@smallexample 10802%rename lib old_lib 10803 10804*lib: 10805--start-group -lgcc -lc -leval1 --end-group %(old_lib) 10806@end smallexample 10807 10808This example renames the spec called @samp{lib} to @samp{old_lib} and 10809then overrides the previous definition of @samp{lib} with a new one. 10810The new definition adds in some extra command-line options before 10811including the text of the old definition. 10812 10813@dfn{Spec strings} are a list of command-line options to be passed to their 10814corresponding program. In addition, the spec strings can contain 10815@samp{%}-prefixed sequences to substitute variable text or to 10816conditionally insert text into the command line. Using these constructs 10817it is possible to generate quite complex command lines. 10818 10819Here is a table of all defined @samp{%}-sequences for spec 10820strings. Note that spaces are not generated automatically around the 10821results of expanding these sequences. Therefore you can concatenate them 10822together or combine them with constant text in a single argument. 10823 10824@table @code 10825@item %% 10826Substitute one @samp{%} into the program name or argument. 10827 10828@item %i 10829Substitute the name of the input file being processed. 10830 10831@item %b 10832Substitute the basename of the input file being processed. 10833This is the substring up to (and not including) the last period 10834and not including the directory. 10835 10836@item %B 10837This is the same as @samp{%b}, but include the file suffix (text after 10838the last period). 10839 10840@item %d 10841Marks the argument containing or following the @samp{%d} as a 10842temporary file name, so that that file is deleted if GCC exits 10843successfully. Unlike @samp{%g}, this contributes no text to the 10844argument. 10845 10846@item %g@var{suffix} 10847Substitute a file name that has suffix @var{suffix} and is chosen 10848once per compilation, and mark the argument in the same way as 10849@samp{%d}. To reduce exposure to denial-of-service attacks, the file 10850name is now chosen in a way that is hard to predict even when previously 10851chosen file names are known. For example, @samp{%g.s @dots{} %g.o @dots{} %g.s} 10852might turn into @samp{ccUVUUAU.s ccXYAXZ12.o ccUVUUAU.s}. @var{suffix} matches 10853the regexp @samp{[.A-Za-z]*} or the special string @samp{%O}, which is 10854treated exactly as if @samp{%O} had been preprocessed. Previously, @samp{%g} 10855was simply substituted with a file name chosen once per compilation, 10856without regard to any appended suffix (which was therefore treated 10857just like ordinary text), making such attacks more likely to succeed. 10858 10859@item %u@var{suffix} 10860Like @samp{%g}, but generates a new temporary file name 10861each time it appears instead of once per compilation. 10862 10863@item %U@var{suffix} 10864Substitutes the last file name generated with @samp{%u@var{suffix}}, generating a 10865new one if there is no such last file name. In the absence of any 10866@samp{%u@var{suffix}}, this is just like @samp{%g@var{suffix}}, except they don't share 10867the same suffix @emph{space}, so @samp{%g.s @dots{} %U.s @dots{} %g.s @dots{} %U.s} 10868involves the generation of two distinct file names, one 10869for each @samp{%g.s} and another for each @samp{%U.s}. Previously, @samp{%U} was 10870simply substituted with a file name chosen for the previous @samp{%u}, 10871without regard to any appended suffix. 10872 10873@item %j@var{suffix} 10874Substitutes the name of the @code{HOST_BIT_BUCKET}, if any, and if it is 10875writable, and if @option{-save-temps} is not used; 10876otherwise, substitute the name 10877of a temporary file, just like @samp{%u}. This temporary file is not 10878meant for communication between processes, but rather as a junk 10879disposal mechanism. 10880 10881@item %|@var{suffix} 10882@itemx %m@var{suffix} 10883Like @samp{%g}, except if @option{-pipe} is in effect. In that case 10884@samp{%|} substitutes a single dash and @samp{%m} substitutes nothing at 10885all. These are the two most common ways to instruct a program that it 10886should read from standard input or write to standard output. If you 10887need something more elaborate you can use an @samp{%@{pipe:@code{X}@}} 10888construct: see for example @file{f/lang-specs.h}. 10889 10890@item %.@var{SUFFIX} 10891Substitutes @var{.SUFFIX} for the suffixes of a matched switch's args 10892when it is subsequently output with @samp{%*}. @var{SUFFIX} is 10893terminated by the next space or %. 10894 10895@item %w 10896Marks the argument containing or following the @samp{%w} as the 10897designated output file of this compilation. This puts the argument 10898into the sequence of arguments that @samp{%o} substitutes. 10899 10900@item %o 10901Substitutes the names of all the output files, with spaces 10902automatically placed around them. You should write spaces 10903around the @samp{%o} as well or the results are undefined. 10904@samp{%o} is for use in the specs for running the linker. 10905Input files whose names have no recognized suffix are not compiled 10906at all, but they are included among the output files, so they are 10907linked. 10908 10909@item %O 10910Substitutes the suffix for object files. Note that this is 10911handled specially when it immediately follows @samp{%g, %u, or %U}, 10912because of the need for those to form complete file names. The 10913handling is such that @samp{%O} is treated exactly as if it had already 10914been substituted, except that @samp{%g, %u, and %U} do not currently 10915support additional @var{suffix} characters following @samp{%O} as they do 10916following, for example, @samp{.o}. 10917 10918@item %p 10919Substitutes the standard macro predefinitions for the 10920current target machine. Use this when running @code{cpp}. 10921 10922@item %P 10923Like @samp{%p}, but puts @samp{__} before and after the name of each 10924predefined macro, except for macros that start with @samp{__} or with 10925@samp{_@var{L}}, where @var{L} is an uppercase letter. This is for ISO 10926C@. 10927 10928@item %I 10929Substitute any of @option{-iprefix} (made from @env{GCC_EXEC_PREFIX}), 10930@option{-isysroot} (made from @env{TARGET_SYSTEM_ROOT}), 10931@option{-isystem} (made from @env{COMPILER_PATH} and @option{-B} options) 10932and @option{-imultilib} as necessary. 10933 10934@item %s 10935Current argument is the name of a library or startup file of some sort. 10936Search for that file in a standard list of directories and substitute 10937the full name found. The current working directory is included in the 10938list of directories scanned. 10939 10940@item %T 10941Current argument is the name of a linker script. Search for that file 10942in the current list of directories to scan for libraries. If the file 10943is located insert a @option{--script} option into the command line 10944followed by the full path name found. If the file is not found then 10945generate an error message. Note: the current working directory is not 10946searched. 10947 10948@item %e@var{str} 10949Print @var{str} as an error message. @var{str} is terminated by a newline. 10950Use this when inconsistent options are detected. 10951 10952@item %(@var{name}) 10953Substitute the contents of spec string @var{name} at this point. 10954 10955@item %x@{@var{option}@} 10956Accumulate an option for @samp{%X}. 10957 10958@item %X 10959Output the accumulated linker options specified by @option{-Wl} or a @samp{%x} 10960spec string. 10961 10962@item %Y 10963Output the accumulated assembler options specified by @option{-Wa}. 10964 10965@item %Z 10966Output the accumulated preprocessor options specified by @option{-Wp}. 10967 10968@item %a 10969Process the @code{asm} spec. This is used to compute the 10970switches to be passed to the assembler. 10971 10972@item %A 10973Process the @code{asm_final} spec. This is a spec string for 10974passing switches to an assembler post-processor, if such a program is 10975needed. 10976 10977@item %l 10978Process the @code{link} spec. This is the spec for computing the 10979command line passed to the linker. Typically it makes use of the 10980@samp{%L %G %S %D and %E} sequences. 10981 10982@item %D 10983Dump out a @option{-L} option for each directory that GCC believes might 10984contain startup files. If the target supports multilibs then the 10985current multilib directory is prepended to each of these paths. 10986 10987@item %L 10988Process the @code{lib} spec. This is a spec string for deciding which 10989libraries are included on the command line to the linker. 10990 10991@item %G 10992Process the @code{libgcc} spec. This is a spec string for deciding 10993which GCC support library is included on the command line to the linker. 10994 10995@item %S 10996Process the @code{startfile} spec. This is a spec for deciding which 10997object files are the first ones passed to the linker. Typically 10998this might be a file named @file{crt0.o}. 10999 11000@item %E 11001Process the @code{endfile} spec. This is a spec string that specifies 11002the last object files that are passed to the linker. 11003 11004@item %C 11005Process the @code{cpp} spec. This is used to construct the arguments 11006to be passed to the C preprocessor. 11007 11008@item %1 11009Process the @code{cc1} spec. This is used to construct the options to be 11010passed to the actual C compiler (@samp{cc1}). 11011 11012@item %2 11013Process the @code{cc1plus} spec. This is used to construct the options to be 11014passed to the actual C++ compiler (@samp{cc1plus}). 11015 11016@item %* 11017Substitute the variable part of a matched option. See below. 11018Note that each comma in the substituted string is replaced by 11019a single space. 11020 11021@item %<@code{S} 11022Remove all occurrences of @code{-S} from the command line. Note---this 11023command is position dependent. @samp{%} commands in the spec string 11024before this one see @code{-S}, @samp{%} commands in the spec string 11025after this one do not. 11026 11027@item %:@var{function}(@var{args}) 11028Call the named function @var{function}, passing it @var{args}. 11029@var{args} is first processed as a nested spec string, then split 11030into an argument vector in the usual fashion. The function returns 11031a string which is processed as if it had appeared literally as part 11032of the current spec. 11033 11034The following built-in spec functions are provided: 11035 11036@table @code 11037@item @code{getenv} 11038The @code{getenv} spec function takes two arguments: an environment 11039variable name and a string. If the environment variable is not 11040defined, a fatal error is issued. Otherwise, the return value is the 11041value of the environment variable concatenated with the string. For 11042example, if @env{TOPDIR} is defined as @file{/path/to/top}, then: 11043 11044@smallexample 11045%:getenv(TOPDIR /include) 11046@end smallexample 11047 11048expands to @file{/path/to/top/include}. 11049 11050@item @code{if-exists} 11051The @code{if-exists} spec function takes one argument, an absolute 11052pathname to a file. If the file exists, @code{if-exists} returns the 11053pathname. Here is a small example of its usage: 11054 11055@smallexample 11056*startfile: 11057crt0%O%s %:if-exists(crti%O%s) crtbegin%O%s 11058@end smallexample 11059 11060@item @code{if-exists-else} 11061The @code{if-exists-else} spec function is similar to the @code{if-exists} 11062spec function, except that it takes two arguments. The first argument is 11063an absolute pathname to a file. If the file exists, @code{if-exists-else} 11064returns the pathname. If it does not exist, it returns the second argument. 11065This way, @code{if-exists-else} can be used to select one file or another, 11066based on the existence of the first. Here is a small example of its usage: 11067 11068@smallexample 11069*startfile: 11070crt0%O%s %:if-exists(crti%O%s) \ 11071%:if-exists-else(crtbeginT%O%s crtbegin%O%s) 11072@end smallexample 11073 11074@item @code{replace-outfile} 11075The @code{replace-outfile} spec function takes two arguments. It looks for the 11076first argument in the outfiles array and replaces it with the second argument. Here 11077is a small example of its usage: 11078 11079@smallexample 11080%@{fgnu-runtime:%:replace-outfile(-lobjc -lobjc-gnu)@} 11081@end smallexample 11082 11083@item @code{remove-outfile} 11084The @code{remove-outfile} spec function takes one argument. It looks for the 11085first argument in the outfiles array and removes it. Here is a small example 11086its usage: 11087 11088@smallexample 11089%:remove-outfile(-lm) 11090@end smallexample 11091 11092@item @code{pass-through-libs} 11093The @code{pass-through-libs} spec function takes any number of arguments. It 11094finds any @option{-l} options and any non-options ending in @file{.a} (which it 11095assumes are the names of linker input library archive files) and returns a 11096result containing all the found arguments each prepended by 11097@option{-plugin-opt=-pass-through=} and joined by spaces. This list is 11098intended to be passed to the LTO linker plugin. 11099 11100@smallexample 11101%:pass-through-libs(%G %L %G) 11102@end smallexample 11103 11104@item @code{print-asm-header} 11105The @code{print-asm-header} function takes no arguments and simply 11106prints a banner like: 11107 11108@smallexample 11109Assembler options 11110================= 11111 11112Use "-Wa,OPTION" to pass "OPTION" to the assembler. 11113@end smallexample 11114 11115It is used to separate compiler options from assembler options 11116in the @option{--target-help} output. 11117@end table 11118 11119@item %@{@code{S}@} 11120Substitutes the @code{-S} switch, if that switch is given to GCC@. 11121If that switch is not specified, this substitutes nothing. Note that 11122the leading dash is omitted when specifying this option, and it is 11123automatically inserted if the substitution is performed. Thus the spec 11124string @samp{%@{foo@}} matches the command-line option @option{-foo} 11125and outputs the command-line option @option{-foo}. 11126 11127@item %W@{@code{S}@} 11128Like %@{@code{S}@} but mark last argument supplied within as a file to be 11129deleted on failure. 11130 11131@item %@{@code{S}*@} 11132Substitutes all the switches specified to GCC whose names start 11133with @code{-S}, but which also take an argument. This is used for 11134switches like @option{-o}, @option{-D}, @option{-I}, etc. 11135GCC considers @option{-o foo} as being 11136one switch whose name starts with @samp{o}. %@{o*@} substitutes this 11137text, including the space. Thus two arguments are generated. 11138 11139@item %@{@code{S}*&@code{T}*@} 11140Like %@{@code{S}*@}, but preserve order of @code{S} and @code{T} options 11141(the order of @code{S} and @code{T} in the spec is not significant). 11142There can be any number of ampersand-separated variables; for each the 11143wild card is optional. Useful for CPP as @samp{%@{D*&U*&A*@}}. 11144 11145@item %@{@code{S}:@code{X}@} 11146Substitutes @code{X}, if the @option{-S} switch is given to GCC@. 11147 11148@item %@{!@code{S}:@code{X}@} 11149Substitutes @code{X}, if the @option{-S} switch is @emph{not} given to GCC@. 11150 11151@item %@{@code{S}*:@code{X}@} 11152Substitutes @code{X} if one or more switches whose names start with 11153@code{-S} are specified to GCC@. Normally @code{X} is substituted only 11154once, no matter how many such switches appeared. However, if @code{%*} 11155appears somewhere in @code{X}, then @code{X} is substituted once 11156for each matching switch, with the @code{%*} replaced by the part of 11157that switch matching the @code{*}. 11158 11159If @code{%*} appears as the last part of a spec sequence then a space 11160will be added after the end of the last substitution. If there is more 11161text in the sequence however then a space will not be generated. This 11162allows the @code{%*} substitution to be used as part of a larger 11163string. For example, a spec string like this: 11164 11165@smallexample 11166%@{mcu=*:--script=%*/memory.ld@} 11167@end smallexample 11168 11169when matching an option like @code{-mcu=newchip} will produce: 11170 11171@smallexample 11172--script=newchip/memory.ld 11173@end smallexample 11174 11175@item %@{.@code{S}:@code{X}@} 11176Substitutes @code{X}, if processing a file with suffix @code{S}. 11177 11178@item %@{!.@code{S}:@code{X}@} 11179Substitutes @code{X}, if @emph{not} processing a file with suffix @code{S}. 11180 11181@item %@{,@code{S}:@code{X}@} 11182Substitutes @code{X}, if processing a file for language @code{S}. 11183 11184@item %@{!,@code{S}:@code{X}@} 11185Substitutes @code{X}, if not processing a file for language @code{S}. 11186 11187@item %@{@code{S}|@code{P}:@code{X}@} 11188Substitutes @code{X} if either @code{-S} or @code{-P} is given to 11189GCC@. This may be combined with @samp{!}, @samp{.}, @samp{,}, and 11190@code{*} sequences as well, although they have a stronger binding than 11191the @samp{|}. If @code{%*} appears in @code{X}, all of the 11192alternatives must be starred, and only the first matching alternative 11193is substituted. 11194 11195For example, a spec string like this: 11196 11197@smallexample 11198%@{.c:-foo@} %@{!.c:-bar@} %@{.c|d:-baz@} %@{!.c|d:-boggle@} 11199@end smallexample 11200 11201@noindent 11202outputs the following command-line options from the following input 11203command-line options: 11204 11205@smallexample 11206fred.c -foo -baz 11207jim.d -bar -boggle 11208-d fred.c -foo -baz -boggle 11209-d jim.d -bar -baz -boggle 11210@end smallexample 11211 11212@item %@{S:X; T:Y; :D@} 11213 11214If @code{S} is given to GCC, substitutes @code{X}; else if @code{T} is 11215given to GCC, substitutes @code{Y}; else substitutes @code{D}. There can 11216be as many clauses as you need. This may be combined with @code{.}, 11217@code{,}, @code{!}, @code{|}, and @code{*} as needed. 11218 11219 11220@end table 11221 11222The conditional text @code{X} in a %@{@code{S}:@code{X}@} or similar 11223construct may contain other nested @samp{%} constructs or spaces, or 11224even newlines. They are processed as usual, as described above. 11225Trailing white space in @code{X} is ignored. White space may also 11226appear anywhere on the left side of the colon in these constructs, 11227except between @code{.} or @code{*} and the corresponding word. 11228 11229The @option{-O}, @option{-f}, @option{-m}, and @option{-W} switches are 11230handled specifically in these constructs. If another value of 11231@option{-O} or the negated form of a @option{-f}, @option{-m}, or 11232@option{-W} switch is found later in the command line, the earlier 11233switch value is ignored, except with @{@code{S}*@} where @code{S} is 11234just one letter, which passes all matching options. 11235 11236The character @samp{|} at the beginning of the predicate text is used to 11237indicate that a command should be piped to the following command, but 11238only if @option{-pipe} is specified. 11239 11240It is built into GCC which switches take arguments and which do not. 11241(You might think it would be useful to generalize this to allow each 11242compiler's spec to say which switches take arguments. But this cannot 11243be done in a consistent fashion. GCC cannot even decide which input 11244files have been specified without knowing which switches take arguments, 11245and it must know which input files to compile in order to tell which 11246compilers to run). 11247 11248GCC also knows implicitly that arguments starting in @option{-l} are to be 11249treated as compiler output files, and passed to the linker in their 11250proper position among the other output files. 11251 11252@c man begin OPTIONS 11253 11254@node Target Options 11255@section Specifying Target Machine and Compiler Version 11256@cindex target options 11257@cindex cross compiling 11258@cindex specifying machine version 11259@cindex specifying compiler version and target machine 11260@cindex compiler version, specifying 11261@cindex target machine, specifying 11262 11263The usual way to run GCC is to run the executable called @command{gcc}, or 11264@command{@var{machine}-gcc} when cross-compiling, or 11265@command{@var{machine}-gcc-@var{version}} to run a version other than the 11266one that was installed last. 11267 11268@node Submodel Options 11269@section Hardware Models and Configurations 11270@cindex submodel options 11271@cindex specifying hardware config 11272@cindex hardware models and configurations, specifying 11273@cindex machine dependent options 11274 11275Each target machine types can have its own 11276special options, starting with @samp{-m}, to choose among various 11277hardware models or configurations---for example, 68010 vs 68020, 11278floating coprocessor or none. A single installed version of the 11279compiler can compile for any model or configuration, according to the 11280options specified. 11281 11282Some configurations of the compiler also support additional special 11283options, usually for compatibility with other compilers on the same 11284platform. 11285 11286@c This list is ordered alphanumerically by subsection name. 11287@c It should be the same order and spelling as these options are listed 11288@c in Machine Dependent Options 11289 11290@menu 11291* AArch64 Options:: 11292* Adapteva Epiphany Options:: 11293* ARC Options:: 11294* ARM Options:: 11295* AVR Options:: 11296* Blackfin Options:: 11297* C6X Options:: 11298* CRIS Options:: 11299* CR16 Options:: 11300* Darwin Options:: 11301* DEC Alpha Options:: 11302* FR30 Options:: 11303* FRV Options:: 11304* GNU/Linux Options:: 11305* H8/300 Options:: 11306* HPPA Options:: 11307* i386 and x86-64 Options:: 11308* i386 and x86-64 Windows Options:: 11309* IA-64 Options:: 11310* LM32 Options:: 11311* M32C Options:: 11312* M32R/D Options:: 11313* M680x0 Options:: 11314* MCore Options:: 11315* MeP Options:: 11316* MicroBlaze Options:: 11317* MIPS Options:: 11318* MMIX Options:: 11319* MN10300 Options:: 11320* Moxie Options:: 11321* MSP430 Options:: 11322* NDS32 Options:: 11323* Nios II Options:: 11324* PDP-11 Options:: 11325* picoChip Options:: 11326* PowerPC Options:: 11327* RL78 Options:: 11328* RS/6000 and PowerPC Options:: 11329* RX Options:: 11330* S/390 and zSeries Options:: 11331* Score Options:: 11332* SH Options:: 11333* Solaris 2 Options:: 11334* SPARC Options:: 11335* SPU Options:: 11336* System V Options:: 11337* TILE-Gx Options:: 11338* TILEPro Options:: 11339* V850 Options:: 11340* VAX Options:: 11341* VMS Options:: 11342* VxWorks Options:: 11343* x86-64 Options:: 11344* Xstormy16 Options:: 11345* Xtensa Options:: 11346* zSeries Options:: 11347@end menu 11348 11349@node AArch64 Options 11350@subsection AArch64 Options 11351@cindex AArch64 Options 11352 11353These options are defined for AArch64 implementations: 11354 11355@table @gcctabopt 11356 11357@item -mabi=@var{name} 11358@opindex mabi 11359Generate code for the specified data model. Permissible values 11360are @samp{ilp32} for SysV-like data model where int, long int and pointer 11361are 32-bit, and @samp{lp64} for SysV-like data model where int is 32-bit, 11362but long int and pointer are 64-bit. 11363 11364The default depends on the specific target configuration. Note that 11365the LP64 and ILP32 ABIs are not link-compatible; you must compile your 11366entire program with the same ABI, and link with a compatible set of libraries. 11367 11368@item -mbig-endian 11369@opindex mbig-endian 11370Generate big-endian code. This is the default when GCC is configured for an 11371@samp{aarch64_be-*-*} target. 11372 11373@item -mgeneral-regs-only 11374@opindex mgeneral-regs-only 11375Generate code which uses only the general registers. 11376 11377@item -mlittle-endian 11378@opindex mlittle-endian 11379Generate little-endian code. This is the default when GCC is configured for an 11380@samp{aarch64-*-*} but not an @samp{aarch64_be-*-*} target. 11381 11382@item -mcmodel=tiny 11383@opindex mcmodel=tiny 11384Generate code for the tiny code model. The program and its statically defined 11385symbols must be within 1GB of each other. Pointers are 64 bits. Programs can 11386be statically or dynamically linked. This model is not fully implemented and 11387mostly treated as @samp{small}. 11388 11389@item -mcmodel=small 11390@opindex mcmodel=small 11391Generate code for the small code model. The program and its statically defined 11392symbols must be within 4GB of each other. Pointers are 64 bits. Programs can 11393be statically or dynamically linked. This is the default code model. 11394 11395@item -mcmodel=large 11396@opindex mcmodel=large 11397Generate code for the large code model. This makes no assumptions about 11398addresses and sizes of sections. Pointers are 64 bits. Programs can be 11399statically linked only. 11400 11401@item -mstrict-align 11402@opindex mstrict-align 11403Do not assume that unaligned memory references will be handled by the system. 11404 11405@item -momit-leaf-frame-pointer 11406@itemx -mno-omit-leaf-frame-pointer 11407@opindex momit-leaf-frame-pointer 11408@opindex mno-omit-leaf-frame-pointer 11409Omit or keep the frame pointer in leaf functions. The former behaviour is the 11410default. 11411 11412@item -mtls-dialect=desc 11413@opindex mtls-dialect=desc 11414Use TLS descriptors as the thread-local storage mechanism for dynamic accesses 11415of TLS variables. This is the default. 11416 11417@item -mtls-dialect=traditional 11418@opindex mtls-dialect=traditional 11419Use traditional TLS as the thread-local storage mechanism for dynamic accesses 11420of TLS variables. 11421 11422@item -mfix-cortex-a53-835769 11423@itemx -mno-fix-cortex-a53-835769 11424@opindex -mfix-cortex-a53-835769 11425@opindex -mno-fix-cortex-a53-835769 11426Enable or disable the workaround for the ARM Cortex-A53 erratum number 835769. 11427This will involve inserting a NOP instruction between memory instructions and 1142864-bit integer multiply-accumulate instructions. 11429 11430@item -march=@var{name} 11431@opindex march 11432Specify the name of the target architecture, optionally suffixed by one or 11433more feature modifiers. This option has the form 11434@option{-march=@var{arch}@r{@{}+@r{[}no@r{]}@var{feature}@r{@}*}}, where the 11435only permissible value for @var{arch} is @samp{armv8-a}. The permissible 11436values for @var{feature} are documented in the sub-section below. 11437 11438Where conflicting feature modifiers are specified, the right-most feature is 11439used. 11440 11441GCC uses this name to determine what kind of instructions it can emit when 11442generating assembly code. 11443 11444Where @option{-march} is specified without either of @option{-mtune} 11445or @option{-mcpu} also being specified, the code will be tuned to perform 11446well across a range of target processors implementing the target 11447architecture. 11448 11449@item -mtune=@var{name} 11450@opindex mtune 11451Specify the name of the target processor for which GCC should tune the 11452performance of the code. Permissible values for this option are: 11453@samp{generic}, @samp{cortex-a53}, @samp{cortex-a57}. 11454 11455Additionally, this option can specify that GCC should tune the performance 11456of the code for a big.LITTLE system. The only permissible value is 11457@samp{cortex-a57.cortex-a53}. 11458 11459Where none of @option{-mtune=}, @option{-mcpu=} or @option{-march=} 11460are specified, the code will be tuned to perform well across a range 11461of target processors. 11462 11463This option cannot be suffixed by feature modifiers. 11464 11465@item -mcpu=@var{name} 11466@opindex mcpu 11467Specify the name of the target processor, optionally suffixed by one or more 11468feature modifiers. This option has the form 11469@option{-mcpu=@var{cpu}@r{@{}+@r{[}no@r{]}@var{feature}@r{@}*}}, where the 11470permissible values for @var{cpu} are the same as those available for 11471@option{-mtune}. 11472 11473The permissible values for @var{feature} are documented in the sub-section 11474below. 11475 11476Where conflicting feature modifiers are specified, the right-most feature is 11477used. 11478 11479GCC uses this name to determine what kind of instructions it can emit when 11480generating assembly code (as if by @option{-march}) and to determine 11481the target processor for which to tune for performance (as if 11482by @option{-mtune}). Where this option is used in conjunction 11483with @option{-march} or @option{-mtune}, those options take precedence 11484over the appropriate part of this option. 11485@end table 11486 11487@subsubsection @option{-march} and @option{-mcpu} feature modifiers 11488@cindex @option{-march} feature modifiers 11489@cindex @option{-mcpu} feature modifiers 11490Feature modifiers used with @option{-march} and @option{-mcpu} can be one 11491the following: 11492 11493@table @samp 11494@item crc 11495Enable CRC extension. 11496@item crypto 11497Enable Crypto extension. This implies Advanced SIMD is enabled. 11498@item fp 11499Enable floating-point instructions. 11500@item simd 11501Enable Advanced SIMD instructions. This implies floating-point instructions 11502are enabled. This is the default for all current possible values for options 11503@option{-march} and @option{-mcpu=}. 11504@end table 11505 11506@node Adapteva Epiphany Options 11507@subsection Adapteva Epiphany Options 11508 11509These @samp{-m} options are defined for Adapteva Epiphany: 11510 11511@table @gcctabopt 11512@item -mhalf-reg-file 11513@opindex mhalf-reg-file 11514Don't allocate any register in the range @code{r32}@dots{}@code{r63}. 11515That allows code to run on hardware variants that lack these registers. 11516 11517@item -mprefer-short-insn-regs 11518@opindex mprefer-short-insn-regs 11519Preferrentially allocate registers that allow short instruction generation. 11520This can result in increased instruction count, so this may either reduce or 11521increase overall code size. 11522 11523@item -mbranch-cost=@var{num} 11524@opindex mbranch-cost 11525Set the cost of branches to roughly @var{num} ``simple'' instructions. 11526This cost is only a heuristic and is not guaranteed to produce 11527consistent results across releases. 11528 11529@item -mcmove 11530@opindex mcmove 11531Enable the generation of conditional moves. 11532 11533@item -mnops=@var{num} 11534@opindex mnops 11535Emit @var{num} NOPs before every other generated instruction. 11536 11537@item -mno-soft-cmpsf 11538@opindex mno-soft-cmpsf 11539For single-precision floating-point comparisons, emit an @code{fsub} instruction 11540and test the flags. This is faster than a software comparison, but can 11541get incorrect results in the presence of NaNs, or when two different small 11542numbers are compared such that their difference is calculated as zero. 11543The default is @option{-msoft-cmpsf}, which uses slower, but IEEE-compliant, 11544software comparisons. 11545 11546@item -mstack-offset=@var{num} 11547@opindex mstack-offset 11548Set the offset between the top of the stack and the stack pointer. 11549E.g., a value of 8 means that the eight bytes in the range @code{sp+0@dots{}sp+7} 11550can be used by leaf functions without stack allocation. 11551Values other than @samp{8} or @samp{16} are untested and unlikely to work. 11552Note also that this option changes the ABI; compiling a program with a 11553different stack offset than the libraries have been compiled with 11554generally does not work. 11555This option can be useful if you want to evaluate if a different stack 11556offset would give you better code, but to actually use a different stack 11557offset to build working programs, it is recommended to configure the 11558toolchain with the appropriate @option{--with-stack-offset=@var{num}} option. 11559 11560@item -mno-round-nearest 11561@opindex mno-round-nearest 11562Make the scheduler assume that the rounding mode has been set to 11563truncating. The default is @option{-mround-nearest}. 11564 11565@item -mlong-calls 11566@opindex mlong-calls 11567If not otherwise specified by an attribute, assume all calls might be beyond 11568the offset range of the @code{b} / @code{bl} instructions, and therefore load the 11569function address into a register before performing a (otherwise direct) call. 11570This is the default. 11571 11572@item -mshort-calls 11573@opindex short-calls 11574If not otherwise specified by an attribute, assume all direct calls are 11575in the range of the @code{b} / @code{bl} instructions, so use these instructions 11576for direct calls. The default is @option{-mlong-calls}. 11577 11578@item -msmall16 11579@opindex msmall16 11580Assume addresses can be loaded as 16-bit unsigned values. This does not 11581apply to function addresses for which @option{-mlong-calls} semantics 11582are in effect. 11583 11584@item -mfp-mode=@var{mode} 11585@opindex mfp-mode 11586Set the prevailing mode of the floating-point unit. 11587This determines the floating-point mode that is provided and expected 11588at function call and return time. Making this mode match the mode you 11589predominantly need at function start can make your programs smaller and 11590faster by avoiding unnecessary mode switches. 11591 11592@var{mode} can be set to one the following values: 11593 11594@table @samp 11595@item caller 11596Any mode at function entry is valid, and retained or restored when 11597the function returns, and when it calls other functions. 11598This mode is useful for compiling libraries or other compilation units 11599you might want to incorporate into different programs with different 11600prevailing FPU modes, and the convenience of being able to use a single 11601object file outweighs the size and speed overhead for any extra 11602mode switching that might be needed, compared with what would be needed 11603with a more specific choice of prevailing FPU mode. 11604 11605@item truncate 11606This is the mode used for floating-point calculations with 11607truncating (i.e.@: round towards zero) rounding mode. That includes 11608conversion from floating point to integer. 11609 11610@item round-nearest 11611This is the mode used for floating-point calculations with 11612round-to-nearest-or-even rounding mode. 11613 11614@item int 11615This is the mode used to perform integer calculations in the FPU, e.g.@: 11616integer multiply, or integer multiply-and-accumulate. 11617@end table 11618 11619The default is @option{-mfp-mode=caller} 11620 11621@item -mnosplit-lohi 11622@itemx -mno-postinc 11623@itemx -mno-postmodify 11624@opindex mnosplit-lohi 11625@opindex mno-postinc 11626@opindex mno-postmodify 11627Code generation tweaks that disable, respectively, splitting of 32-bit 11628loads, generation of post-increment addresses, and generation of 11629post-modify addresses. The defaults are @option{msplit-lohi}, 11630@option{-mpost-inc}, and @option{-mpost-modify}. 11631 11632@item -mnovect-double 11633@opindex mno-vect-double 11634Change the preferred SIMD mode to SImode. The default is 11635@option{-mvect-double}, which uses DImode as preferred SIMD mode. 11636 11637@item -max-vect-align=@var{num} 11638@opindex max-vect-align 11639The maximum alignment for SIMD vector mode types. 11640@var{num} may be 4 or 8. The default is 8. 11641Note that this is an ABI change, even though many library function 11642interfaces are unaffected if they don't use SIMD vector modes 11643in places that affect size and/or alignment of relevant types. 11644 11645@item -msplit-vecmove-early 11646@opindex msplit-vecmove-early 11647Split vector moves into single word moves before reload. In theory this 11648can give better register allocation, but so far the reverse seems to be 11649generally the case. 11650 11651@item -m1reg-@var{reg} 11652@opindex m1reg- 11653Specify a register to hold the constant @minus{}1, which makes loading small negative 11654constants and certain bitmasks faster. 11655Allowable values for @var{reg} are @samp{r43} and @samp{r63}, 11656which specify use of that register as a fixed register, 11657and @samp{none}, which means that no register is used for this 11658purpose. The default is @option{-m1reg-none}. 11659 11660@end table 11661 11662@node ARC Options 11663@subsection ARC Options 11664@cindex ARC options 11665 11666The following options control the architecture variant for which code 11667is being compiled: 11668 11669@c architecture variants 11670@table @gcctabopt 11671 11672@item -mbarrel-shifter 11673@opindex mbarrel-shifter 11674Generate instructions supported by barrel shifter. This is the default 11675unless @samp{-mcpu=ARC601} is in effect. 11676 11677@item -mcpu=@var{cpu} 11678@opindex mcpu 11679Set architecture type, register usage, and instruction scheduling 11680parameters for @var{cpu}. There are also shortcut alias options 11681available for backward compatibility and convenience. Supported 11682values for @var{cpu} are 11683 11684@table @samp 11685@opindex mA6 11686@opindex mARC600 11687@item ARC600 11688Compile for ARC600. Aliases: @option{-mA6}, @option{-mARC600}. 11689 11690@item ARC601 11691@opindex mARC601 11692Compile for ARC601. Alias: @option{-mARC601}. 11693 11694@item ARC700 11695@opindex mA7 11696@opindex mARC700 11697Compile for ARC700. Aliases: @option{-mA7}, @option{-mARC700}. 11698This is the default when configured with @samp{--with-cpu=arc700}@. 11699@end table 11700 11701@item -mdpfp 11702@opindex mdpfp 11703@itemx -mdpfp-compact 11704@opindex mdpfp-compact 11705FPX: Generate Double Precision FPX instructions, tuned for the compact 11706implementation. 11707 11708@item -mdpfp-fast 11709@opindex mdpfp-fast 11710FPX: Generate Double Precision FPX instructions, tuned for the fast 11711implementation. 11712 11713@item -mno-dpfp-lrsr 11714@opindex mno-dpfp-lrsr 11715Disable LR and SR instructions from using FPX extension aux registers. 11716 11717@item -mea 11718@opindex mea 11719Generate Extended arithmetic instructions. Currently only 11720@code{divaw}, @code{adds}, @code{subs}, and @code{sat16} are 11721supported. This is always enabled for @samp{-mcpu=ARC700}. 11722 11723@item -mno-mpy 11724@opindex mno-mpy 11725Do not generate mpy instructions for ARC700. 11726 11727@item -mmul32x16 11728@opindex mmul32x16 11729Generate 32x16 bit multiply and mac instructions. 11730 11731@item -mmul64 11732@opindex mmul64 11733Generate mul64 and mulu64 instructions. Only valid for @samp{-mcpu=ARC600}. 11734 11735@item -mnorm 11736@opindex mnorm 11737Generate norm instruction. This is the default if @samp{-mcpu=ARC700} 11738is in effect. 11739 11740@item -mspfp 11741@opindex mspfp 11742@itemx -mspfp-compact 11743@opindex mspfp-compact 11744FPX: Generate Single Precision FPX instructions, tuned for the compact 11745implementation. 11746 11747@item -mspfp-fast 11748@opindex mspfp-fast 11749FPX: Generate Single Precision FPX instructions, tuned for the fast 11750implementation. 11751 11752@item -msimd 11753@opindex msimd 11754Enable generation of ARC SIMD instructions via target-specific 11755builtins. Only valid for @samp{-mcpu=ARC700}. 11756 11757@item -msoft-float 11758@opindex msoft-float 11759This option ignored; it is provided for compatibility purposes only. 11760Software floating point code is emitted by default, and this default 11761can overridden by FPX options; @samp{mspfp}, @samp{mspfp-compact}, or 11762@samp{mspfp-fast} for single precision, and @samp{mdpfp}, 11763@samp{mdpfp-compact}, or @samp{mdpfp-fast} for double precision. 11764 11765@item -mswap 11766@opindex mswap 11767Generate swap instructions. 11768 11769@end table 11770 11771The following options are passed through to the assembler, and also 11772define preprocessor macro symbols. 11773 11774@c Flags used by the assembler, but for which we define preprocessor 11775@c macro symbols as well. 11776@table @gcctabopt 11777@item -mdsp-packa 11778@opindex mdsp-packa 11779Passed down to the assembler to enable the DSP Pack A extensions. 11780Also sets the preprocessor symbol @code{__Xdsp_packa}. 11781 11782@item -mdvbf 11783@opindex mdvbf 11784Passed down to the assembler to enable the dual viterbi butterfly 11785extension. Also sets the preprocessor symbol @code{__Xdvbf}. 11786 11787@c ARC700 4.10 extension instruction 11788@item -mlock 11789@opindex mlock 11790Passed down to the assembler to enable the Locked Load/Store 11791Conditional extension. Also sets the preprocessor symbol 11792@code{__Xlock}. 11793 11794@item -mmac-d16 11795@opindex mmac-d16 11796Passed down to the assembler. Also sets the preprocessor symbol 11797@code{__Xxmac_d16}. 11798 11799@item -mmac-24 11800@opindex mmac-24 11801Passed down to the assembler. Also sets the preprocessor symbol 11802@code{__Xxmac_24}. 11803 11804@c ARC700 4.10 extension instruction 11805@item -mrtsc 11806@opindex mrtsc 11807Passed down to the assembler to enable the 64-bit Time-Stamp Counter 11808extension instruction. Also sets the preprocessor symbol 11809@code{__Xrtsc}. 11810 11811@c ARC700 4.10 extension instruction 11812@item -mswape 11813@opindex mswape 11814Passed down to the assembler to enable the swap byte ordering 11815extension instruction. Also sets the preprocessor symbol 11816@code{__Xswape}. 11817 11818@item -mtelephony 11819@opindex mtelephony 11820Passed down to the assembler to enable dual and single operand 11821instructions for telephony. Also sets the preprocessor symbol 11822@code{__Xtelephony}. 11823 11824@item -mxy 11825@opindex mxy 11826Passed down to the assembler to enable the XY Memory extension. Also 11827sets the preprocessor symbol @code{__Xxy}. 11828 11829@end table 11830 11831The following options control how the assembly code is annotated: 11832 11833@c Assembly annotation options 11834@table @gcctabopt 11835@item -misize 11836@opindex misize 11837Annotate assembler instructions with estimated addresses. 11838 11839@item -mannotate-align 11840@opindex mannotate-align 11841Explain what alignment considerations lead to the decision to make an 11842instruction short or long. 11843 11844@end table 11845 11846The following options are passed through to the linker: 11847 11848@c options passed through to the linker 11849@table @gcctabopt 11850@item -marclinux 11851@opindex marclinux 11852Passed through to the linker, to specify use of the @code{arclinux} emulation. 11853This option is enabled by default in tool chains built for 11854@w{@code{arc-linux-uclibc}} and @w{@code{arceb-linux-uclibc}} targets 11855when profiling is not requested. 11856 11857@item -marclinux_prof 11858@opindex marclinux_prof 11859Passed through to the linker, to specify use of the 11860@code{arclinux_prof} emulation. This option is enabled by default in 11861tool chains built for @w{@code{arc-linux-uclibc}} and 11862@w{@code{arceb-linux-uclibc}} targets when profiling is requested. 11863 11864@end table 11865 11866The following options control the semantics of generated code: 11867 11868@c semantically relevant code generation options 11869@table @gcctabopt 11870@item -mepilogue-cfi 11871@opindex mepilogue-cfi 11872Enable generation of call frame information for epilogues. 11873 11874@item -mno-epilogue-cfi 11875@opindex mno-epilogue-cfi 11876Disable generation of call frame information for epilogues. 11877 11878@item -mlong-calls 11879@opindex mlong-calls 11880Generate call insns as register indirect calls, thus providing access 11881to the full 32-bit address range. 11882 11883@item -mmedium-calls 11884@opindex mmedium-calls 11885Don't use less than 25 bit addressing range for calls, which is the 11886offset available for an unconditional branch-and-link 11887instruction. Conditional execution of function calls is suppressed, to 11888allow use of the 25-bit range, rather than the 21-bit range with 11889conditional branch-and-link. This is the default for tool chains built 11890for @w{@code{arc-linux-uclibc}} and @w{@code{arceb-linux-uclibc}} targets. 11891 11892@item -mno-sdata 11893@opindex mno-sdata 11894Do not generate sdata references. This is the default for tool chains 11895built for @w{@code{arc-linux-uclibc}} and @w{@code{arceb-linux-uclibc}} 11896targets. 11897 11898@item -mucb-mcount 11899@opindex mucb-mcount 11900Instrument with mcount calls as used in UCB code. I.e. do the 11901counting in the callee, not the caller. By default ARC instrumentation 11902counts in the caller. 11903 11904@item -mvolatile-cache 11905@opindex mvolatile-cache 11906Use ordinarily cached memory accesses for volatile references. This is the 11907default. 11908 11909@item -mno-volatile-cache 11910@opindex mno-volatile-cache 11911Enable cache bypass for volatile references. 11912 11913@end table 11914 11915The following options fine tune code generation: 11916@c code generation tuning options 11917@table @gcctabopt 11918@item -malign-call 11919@opindex malign-call 11920Do alignment optimizations for call instructions. 11921 11922@item -mauto-modify-reg 11923@opindex mauto-modify-reg 11924Enable the use of pre/post modify with register displacement. 11925 11926@item -mbbit-peephole 11927@opindex mbbit-peephole 11928Enable bbit peephole2. 11929 11930@item -mno-brcc 11931@opindex mno-brcc 11932This option disables a target-specific pass in @file{arc_reorg} to 11933generate @code{BRcc} instructions. It has no effect on @code{BRcc} 11934generation driven by the combiner pass. 11935 11936@item -mcase-vector-pcrel 11937@opindex mcase-vector-pcrel 11938Use pc-relative switch case tables - this enables case table shortening. 11939This is the default for @option{-Os}. 11940 11941@item -mcompact-casesi 11942@opindex mcompact-casesi 11943Enable compact casesi pattern. 11944This is the default for @option{-Os}. 11945 11946@item -mno-cond-exec 11947@opindex mno-cond-exec 11948Disable ARCompact specific pass to generate conditional execution instructions. 11949Due to delay slot scheduling and interactions between operand numbers, 11950literal sizes, instruction lengths, and the support for conditional execution, 11951the target-independent pass to generate conditional execution is often lacking, 11952so the ARC port has kept a special pass around that tries to find more 11953conditional execution generating opportunities after register allocation, 11954branch shortening, and delay slot scheduling have been done. This pass 11955generally, but not always, improves performance and code size, at the cost of 11956extra compilation time, which is why there is an option to switch it off. 11957If you have a problem with call instructions exceeding their allowable 11958offset range because they are conditionalized, you should consider using 11959@option{-mmedium-calls} instead. 11960 11961@item -mearly-cbranchsi 11962@opindex mearly-cbranchsi 11963Enable pre-reload use of the cbranchsi pattern. 11964 11965@item -mexpand-adddi 11966@opindex mexpand-adddi 11967Expand @code{adddi3} and @code{subdi3} at rtl generation time into 11968@code{add.f}, @code{adc} etc. 11969 11970@item -mindexed-loads 11971@opindex mindexed-loads 11972Enable the use of indexed loads. This can be problematic because some 11973optimizers will then assume the that indexed stores exist, which is not 11974the case. 11975 11976@item -mlra 11977@opindex mlra 11978Enable Local Register Allocation. This is still experimental for ARC, 11979so by default the compiler uses standard reload 11980(i.e. @samp{-mno-lra}). 11981 11982@item -mlra-priority-none 11983@opindex mlra-priority-none 11984Don't indicate any priority for target registers. 11985 11986@item -mlra-priority-compact 11987@opindex mlra-priority-compact 11988Indicate target register priority for r0..r3 / r12..r15. 11989 11990@item -mlra-priority-noncompact 11991@opindex mlra-priority-noncompact 11992Reduce target regsiter priority for r0..r3 / r12..r15. 11993 11994@item -mno-millicode 11995@opindex mno-millicode 11996When optimizing for size (using @option{-Os}), prologues and epilogues 11997that have to save or restore a large number of registers are often 11998shortened by using call to a special function in libgcc; this is 11999referred to as a @emph{millicode} call. As these calls can pose 12000performance issues, and/or cause linking issues when linking in a 12001nonstandard way, this option is provided to turn off millicode call 12002generation. 12003 12004@item -mmixed-code 12005@opindex mmixed-code 12006Tweak register allocation to help 16-bit instruction generation. 12007This generally has the effect of decreasing the average instruction size 12008while increasing the instruction count. 12009 12010@item -mq-class 12011@opindex mq-class 12012Enable 'q' instruction alternatives. 12013This is the default for @option{-Os}. 12014 12015@item -mRcq 12016@opindex mRcq 12017Enable Rcq constraint handling - most short code generation depends on this. 12018This is the default. 12019 12020@item -mRcw 12021@opindex mRcw 12022Enable Rcw constraint handling - ccfsm condexec mostly depends on this. 12023This is the default. 12024 12025@item -msize-level=@var{level} 12026@opindex msize-level 12027Fine-tune size optimization with regards to instruction lengths and alignment. 12028The recognized values for @var{level} are: 12029@table @samp 12030@item 0 12031No size optimization. This level is deprecated and treated like @samp{1}. 12032 12033@item 1 12034Short instructions are used opportunistically. 12035 12036@item 2 12037In addition, alignment of loops and of code after barriers are dropped. 12038 12039@item 3 12040In addition, optional data alignment is dropped, and the option @option{Os} is enabled. 12041 12042@end table 12043 12044This defaults to @samp{3} when @option{-Os} is in effect. Otherwise, 12045the behavior when this is not set is equivalent to level @samp{1}. 12046 12047@item -mtune=@var{cpu} 12048@opindex mtune 12049Set instruction scheduling parameters for @var{cpu}, overriding any implied 12050by @option{-mcpu=}. 12051 12052Supported values for @var{cpu} are 12053 12054@table @samp 12055@item ARC600 12056Tune for ARC600 cpu. 12057 12058@item ARC601 12059Tune for ARC601 cpu. 12060 12061@item ARC700 12062Tune for ARC700 cpu with standard multiplier block. 12063 12064@item ARC700-xmac 12065Tune for ARC700 cpu with XMAC block. 12066 12067@item ARC725D 12068Tune for ARC725D cpu. 12069 12070@item ARC750D 12071Tune for ARC750D cpu. 12072 12073@end table 12074 12075@item -mmultcost=@var{num} 12076@opindex mmultcost 12077Cost to assume for a multiply instruction, with @samp{4} being equal to a 12078normal instruction. 12079 12080@item -munalign-prob-threshold=@var{probability} 12081@opindex munalign-prob-threshold 12082Set probability threshold for unaligning branches. 12083When tuning for @samp{ARC700} and optimizing for speed, branches without 12084filled delay slot are preferably emitted unaligned and long, unless 12085profiling indicates that the probability for the branch to be taken 12086is below @var{probability}. @xref{Cross-profiling}. 12087The default is (REG_BR_PROB_BASE/2), i.e.@: 5000. 12088 12089@end table 12090 12091The following options are maintained for backward compatibility, but 12092are now deprecated and will be removed in a future release: 12093 12094@c Deprecated options 12095@table @gcctabopt 12096 12097@item -margonaut 12098@opindex margonaut 12099Obsolete FPX. 12100 12101@item -mbig-endian 12102@opindex mbig-endian 12103@itemx -EB 12104@opindex EB 12105Compile code for big endian targets. Use of these options is now 12106deprecated. Users wanting big-endian code, should use the 12107@w{@code{arceb-elf32}} and @w{@code{arceb-linux-uclibc}} targets when 12108building the tool chain, for which big-endian is the default. 12109 12110@item -mlittle-endian 12111@opindex mlittle-endian 12112@itemx -EL 12113@opindex EL 12114Compile code for little endian targets. Use of these options is now 12115deprecated. Users wanting little-endian code should use the 12116@w{@code{arc-elf32}} and @w{@code{arc-linux-uclibc}} targets when 12117building the tool chain, for which little-endian is the default. 12118 12119@item -mbarrel_shifter 12120@opindex mbarrel_shifter 12121Replaced by @samp{-mbarrel-shifter} 12122 12123@item -mdpfp_compact 12124@opindex mdpfp_compact 12125Replaced by @samp{-mdpfp-compact} 12126 12127@item -mdpfp_fast 12128@opindex mdpfp_fast 12129Replaced by @samp{-mdpfp-fast} 12130 12131@item -mdsp_packa 12132@opindex mdsp_packa 12133Replaced by @samp{-mdsp-packa} 12134 12135@item -mEA 12136@opindex mEA 12137Replaced by @samp{-mea} 12138 12139@item -mmac_24 12140@opindex mmac_24 12141Replaced by @samp{-mmac-24} 12142 12143@item -mmac_d16 12144@opindex mmac_d16 12145Replaced by @samp{-mmac-d16} 12146 12147@item -mspfp_compact 12148@opindex mspfp_compact 12149Replaced by @samp{-mspfp-compact} 12150 12151@item -mspfp_fast 12152@opindex mspfp_fast 12153Replaced by @samp{-mspfp-fast} 12154 12155@item -mtune=@var{cpu} 12156@opindex mtune 12157Values @samp{arc600}, @samp{arc601}, @samp{arc700} and 12158@samp{arc700-xmac} for @var{cpu} are replaced by @samp{ARC600}, 12159@samp{ARC601}, @samp{ARC700} and @samp{ARC700-xmac} respectively 12160 12161@item -multcost=@var{num} 12162@opindex multcost 12163Replaced by @samp{-mmultcost}. 12164 12165@end table 12166 12167@node ARM Options 12168@subsection ARM Options 12169@cindex ARM options 12170 12171These @samp{-m} options are defined for Advanced RISC Machines (ARM) 12172architectures: 12173 12174@table @gcctabopt 12175@item -mabi=@var{name} 12176@opindex mabi 12177Generate code for the specified ABI@. Permissible values are: @samp{apcs-gnu}, 12178@samp{atpcs}, @samp{aapcs}, @samp{aapcs-linux} and @samp{iwmmxt}. 12179 12180@item -mapcs-frame 12181@opindex mapcs-frame 12182Generate a stack frame that is compliant with the ARM Procedure Call 12183Standard for all functions, even if this is not strictly necessary for 12184correct execution of the code. Specifying @option{-fomit-frame-pointer} 12185with this option causes the stack frames not to be generated for 12186leaf functions. The default is @option{-mno-apcs-frame}. 12187 12188@item -mapcs 12189@opindex mapcs 12190This is a synonym for @option{-mapcs-frame}. 12191 12192@ignore 12193@c not currently implemented 12194@item -mapcs-stack-check 12195@opindex mapcs-stack-check 12196Generate code to check the amount of stack space available upon entry to 12197every function (that actually uses some stack space). If there is 12198insufficient space available then either the function 12199@samp{__rt_stkovf_split_small} or @samp{__rt_stkovf_split_big} is 12200called, depending upon the amount of stack space required. The runtime 12201system is required to provide these functions. The default is 12202@option{-mno-apcs-stack-check}, since this produces smaller code. 12203 12204@c not currently implemented 12205@item -mapcs-float 12206@opindex mapcs-float 12207Pass floating-point arguments using the floating-point registers. This is 12208one of the variants of the APCS@. This option is recommended if the 12209target hardware has a floating-point unit or if a lot of floating-point 12210arithmetic is going to be performed by the code. The default is 12211@option{-mno-apcs-float}, since the size of integer-only code is 12212slightly increased if @option{-mapcs-float} is used. 12213 12214@c not currently implemented 12215@item -mapcs-reentrant 12216@opindex mapcs-reentrant 12217Generate reentrant, position-independent code. The default is 12218@option{-mno-apcs-reentrant}. 12219@end ignore 12220 12221@item -mthumb-interwork 12222@opindex mthumb-interwork 12223Generate code that supports calling between the ARM and Thumb 12224instruction sets. Without this option, on pre-v5 architectures, the 12225two instruction sets cannot be reliably used inside one program. The 12226default is @option{-mno-thumb-interwork}, since slightly larger code 12227is generated when @option{-mthumb-interwork} is specified. In AAPCS 12228configurations this option is meaningless. 12229 12230@item -mno-sched-prolog 12231@opindex mno-sched-prolog 12232Prevent the reordering of instructions in the function prologue, or the 12233merging of those instruction with the instructions in the function's 12234body. This means that all functions start with a recognizable set 12235of instructions (or in fact one of a choice from a small set of 12236different function prologues), and this information can be used to 12237locate the start of functions inside an executable piece of code. The 12238default is @option{-msched-prolog}. 12239 12240@item -mfloat-abi=@var{name} 12241@opindex mfloat-abi 12242Specifies which floating-point ABI to use. Permissible values 12243are: @samp{soft}, @samp{softfp} and @samp{hard}. 12244 12245Specifying @samp{soft} causes GCC to generate output containing 12246library calls for floating-point operations. 12247@samp{softfp} allows the generation of code using hardware floating-point 12248instructions, but still uses the soft-float calling conventions. 12249@samp{hard} allows generation of floating-point instructions 12250and uses FPU-specific calling conventions. 12251 12252The default depends on the specific target configuration. Note that 12253the hard-float and soft-float ABIs are not link-compatible; you must 12254compile your entire program with the same ABI, and link with a 12255compatible set of libraries. 12256 12257@item -mlittle-endian 12258@opindex mlittle-endian 12259Generate code for a processor running in little-endian mode. This is 12260the default for all standard configurations. 12261 12262@item -mbig-endian 12263@opindex mbig-endian 12264Generate code for a processor running in big-endian mode; the default is 12265to compile code for a little-endian processor. 12266 12267@item -mwords-little-endian 12268@opindex mwords-little-endian 12269This option only applies when generating code for big-endian processors. 12270Generate code for a little-endian word order but a big-endian byte 12271order. That is, a byte order of the form @samp{32107654}. Note: this 12272option should only be used if you require compatibility with code for 12273big-endian ARM processors generated by versions of the compiler prior to 122742.8. This option is now deprecated. 12275 12276@item -march=@var{name} 12277@opindex march 12278This specifies the name of the target ARM architecture. GCC uses this 12279name to determine what kind of instructions it can emit when generating 12280assembly code. This option can be used in conjunction with or instead 12281of the @option{-mcpu=} option. Permissible names are: @samp{armv2}, 12282@samp{armv2a}, @samp{armv3}, @samp{armv3m}, @samp{armv4}, @samp{armv4t}, 12283@samp{armv5}, @samp{armv5t}, @samp{armv5e}, @samp{armv5te}, 12284@samp{armv6}, @samp{armv6j}, 12285@samp{armv6t2}, @samp{armv6z}, @samp{armv6zk}, @samp{armv6-m}, 12286@samp{armv7}, @samp{armv7-a}, @samp{armv7-r}, @samp{armv7-m}, @samp{armv7e-m}, 12287@samp{armv7ve}, @samp{armv8-a}, @samp{armv8-a+crc}, 12288@samp{iwmmxt}, @samp{iwmmxt2}, @samp{ep9312}. 12289 12290@option{-march=armv7ve} is the armv7-a architecture with virtualization 12291extensions. 12292 12293@option{-march=armv8-a+crc} enables code generation for the ARMv8-A 12294architecture together with the optional CRC32 extensions. 12295 12296@option{-march=native} causes the compiler to auto-detect the architecture 12297of the build computer. At present, this feature is only supported on 12298GNU/Linux, and not all architectures are recognized. If the auto-detect 12299is unsuccessful the option has no effect. 12300 12301@item -mtune=@var{name} 12302@opindex mtune 12303This option specifies the name of the target ARM processor for 12304which GCC should tune the performance of the code. 12305For some ARM implementations better performance can be obtained by using 12306this option. 12307Permissible names are: @samp{arm2}, @samp{arm250}, 12308@samp{arm3}, @samp{arm6}, @samp{arm60}, @samp{arm600}, @samp{arm610}, 12309@samp{arm620}, @samp{arm7}, @samp{arm7m}, @samp{arm7d}, @samp{arm7dm}, 12310@samp{arm7di}, @samp{arm7dmi}, @samp{arm70}, @samp{arm700}, 12311@samp{arm700i}, @samp{arm710}, @samp{arm710c}, @samp{arm7100}, 12312@samp{arm720}, 12313@samp{arm7500}, @samp{arm7500fe}, @samp{arm7tdmi}, @samp{arm7tdmi-s}, 12314@samp{arm710t}, @samp{arm720t}, @samp{arm740t}, 12315@samp{strongarm}, @samp{strongarm110}, @samp{strongarm1100}, 12316@samp{strongarm1110}, 12317@samp{arm8}, @samp{arm810}, @samp{arm9}, @samp{arm9e}, @samp{arm920}, 12318@samp{arm920t}, @samp{arm922t}, @samp{arm946e-s}, @samp{arm966e-s}, 12319@samp{arm968e-s}, @samp{arm926ej-s}, @samp{arm940t}, @samp{arm9tdmi}, 12320@samp{arm10tdmi}, @samp{arm1020t}, @samp{arm1026ej-s}, 12321@samp{arm10e}, @samp{arm1020e}, @samp{arm1022e}, 12322@samp{arm1136j-s}, @samp{arm1136jf-s}, @samp{mpcore}, @samp{mpcorenovfp}, 12323@samp{arm1156t2-s}, @samp{arm1156t2f-s}, @samp{arm1176jz-s}, @samp{arm1176jzf-s}, 12324@samp{cortex-a5}, @samp{cortex-a7}, @samp{cortex-a8}, @samp{cortex-a9}, 12325@samp{cortex-a12}, @samp{cortex-a15}, @samp{cortex-a53}, @samp{cortex-a57}, 12326@samp{cortex-r4}, 12327@samp{cortex-r4f}, @samp{cortex-r5}, @samp{cortex-r7}, @samp{cortex-m4}, 12328@samp{cortex-m3}, 12329@samp{cortex-m1}, 12330@samp{cortex-m0}, 12331@samp{cortex-m0plus}, 12332@samp{marvell-pj4}, 12333@samp{xscale}, @samp{iwmmxt}, @samp{iwmmxt2}, @samp{ep9312}, 12334@samp{fa526}, @samp{fa626}, 12335@samp{fa606te}, @samp{fa626te}, @samp{fmp626}, @samp{fa726te}. 12336 12337Additionally, this option can specify that GCC should tune the performance 12338of the code for a big.LITTLE system. Permissible names are: 12339@samp{cortex-a15.cortex-a7}, @samp{cortex-a57.cortex-a53}. 12340 12341@option{-mtune=generic-@var{arch}} specifies that GCC should tune the 12342performance for a blend of processors within architecture @var{arch}. 12343The aim is to generate code that run well on the current most popular 12344processors, balancing between optimizations that benefit some CPUs in the 12345range, and avoiding performance pitfalls of other CPUs. The effects of 12346this option may change in future GCC versions as CPU models come and go. 12347 12348@option{-mtune=native} causes the compiler to auto-detect the CPU 12349of the build computer. At present, this feature is only supported on 12350GNU/Linux, and not all architectures are recognized. If the auto-detect is 12351unsuccessful the option has no effect. 12352 12353@item -mcpu=@var{name} 12354@opindex mcpu 12355This specifies the name of the target ARM processor. GCC uses this name 12356to derive the name of the target ARM architecture (as if specified 12357by @option{-march}) and the ARM processor type for which to tune for 12358performance (as if specified by @option{-mtune}). Where this option 12359is used in conjunction with @option{-march} or @option{-mtune}, 12360those options take precedence over the appropriate part of this option. 12361 12362Permissible names for this option are the same as those for 12363@option{-mtune}. 12364 12365@option{-mcpu=generic-@var{arch}} is also permissible, and is 12366equivalent to @option{-march=@var{arch} -mtune=generic-@var{arch}}. 12367See @option{-mtune} for more information. 12368 12369@option{-mcpu=native} causes the compiler to auto-detect the CPU 12370of the build computer. At present, this feature is only supported on 12371GNU/Linux, and not all architectures are recognized. If the auto-detect 12372is unsuccessful the option has no effect. 12373 12374@item -mfpu=@var{name} 12375@opindex mfpu 12376This specifies what floating-point hardware (or hardware emulation) is 12377available on the target. Permissible names are: @samp{vfp}, @samp{vfpv3}, 12378@samp{vfpv3-fp16}, @samp{vfpv3-d16}, @samp{vfpv3-d16-fp16}, @samp{vfpv3xd}, 12379@samp{vfpv3xd-fp16}, @samp{neon}, @samp{neon-fp16}, @samp{vfpv4}, 12380@samp{vfpv4-d16}, @samp{fpv4-sp-d16}, @samp{neon-vfpv4}, 12381@samp{fp-armv8}, @samp{neon-fp-armv8}, and @samp{crypto-neon-fp-armv8}. 12382 12383If @option{-msoft-float} is specified this specifies the format of 12384floating-point values. 12385 12386If the selected floating-point hardware includes the NEON extension 12387(e.g. @option{-mfpu}=@samp{neon}), note that floating-point 12388operations are not generated by GCC's auto-vectorization pass unless 12389@option{-funsafe-math-optimizations} is also specified. This is 12390because NEON hardware does not fully implement the IEEE 754 standard for 12391floating-point arithmetic (in particular denormal values are treated as 12392zero), so the use of NEON instructions may lead to a loss of precision. 12393 12394@item -mfp16-format=@var{name} 12395@opindex mfp16-format 12396Specify the format of the @code{__fp16} half-precision floating-point type. 12397Permissible names are @samp{none}, @samp{ieee}, and @samp{alternative}; 12398the default is @samp{none}, in which case the @code{__fp16} type is not 12399defined. @xref{Half-Precision}, for more information. 12400 12401@item -mstructure-size-boundary=@var{n} 12402@opindex mstructure-size-boundary 12403The sizes of all structures and unions are rounded up to a multiple 12404of the number of bits set by this option. Permissible values are 8, 32 12405and 64. The default value varies for different toolchains. For the COFF 12406targeted toolchain the default value is 8. A value of 64 is only allowed 12407if the underlying ABI supports it. 12408 12409Specifying a larger number can produce faster, more efficient code, but 12410can also increase the size of the program. Different values are potentially 12411incompatible. Code compiled with one value cannot necessarily expect to 12412work with code or libraries compiled with another value, if they exchange 12413information using structures or unions. 12414 12415@item -mabort-on-noreturn 12416@opindex mabort-on-noreturn 12417Generate a call to the function @code{abort} at the end of a 12418@code{noreturn} function. It is executed if the function tries to 12419return. 12420 12421@item -mlong-calls 12422@itemx -mno-long-calls 12423@opindex mlong-calls 12424@opindex mno-long-calls 12425Tells the compiler to perform function calls by first loading the 12426address of the function into a register and then performing a subroutine 12427call on this register. This switch is needed if the target function 12428lies outside of the 64-megabyte addressing range of the offset-based 12429version of subroutine call instruction. 12430 12431Even if this switch is enabled, not all function calls are turned 12432into long calls. The heuristic is that static functions, functions 12433that have the @samp{short-call} attribute, functions that are inside 12434the scope of a @samp{#pragma no_long_calls} directive, and functions whose 12435definitions have already been compiled within the current compilation 12436unit are not turned into long calls. The exceptions to this rule are 12437that weak function definitions, functions with the @samp{long-call} 12438attribute or the @samp{section} attribute, and functions that are within 12439the scope of a @samp{#pragma long_calls} directive are always 12440turned into long calls. 12441 12442This feature is not enabled by default. Specifying 12443@option{-mno-long-calls} restores the default behavior, as does 12444placing the function calls within the scope of a @samp{#pragma 12445long_calls_off} directive. Note these switches have no effect on how 12446the compiler generates code to handle function calls via function 12447pointers. 12448 12449@item -msingle-pic-base 12450@opindex msingle-pic-base 12451Treat the register used for PIC addressing as read-only, rather than 12452loading it in the prologue for each function. The runtime system is 12453responsible for initializing this register with an appropriate value 12454before execution begins. 12455 12456@item -mpic-register=@var{reg} 12457@opindex mpic-register 12458Specify the register to be used for PIC addressing. 12459For standard PIC base case, the default will be any suitable register 12460determined by compiler. For single PIC base case, the default is 12461@samp{R9} if target is EABI based or stack-checking is enabled, 12462otherwise the default is @samp{R10}. 12463 12464@item -mpic-data-is-text-relative 12465@opindex mpic-data-is-text-relative 12466Assume that each data segments are relative to text segment at load time. 12467Therefore, it permits addressing data using PC-relative operations. 12468This option is on by default for targets other than VxWorks RTP. 12469 12470@item -mpoke-function-name 12471@opindex mpoke-function-name 12472Write the name of each function into the text section, directly 12473preceding the function prologue. The generated code is similar to this: 12474 12475@smallexample 12476 t0 12477 .ascii "arm_poke_function_name", 0 12478 .align 12479 t1 12480 .word 0xff000000 + (t1 - t0) 12481 arm_poke_function_name 12482 mov ip, sp 12483 stmfd sp!, @{fp, ip, lr, pc@} 12484 sub fp, ip, #4 12485@end smallexample 12486 12487When performing a stack backtrace, code can inspect the value of 12488@code{pc} stored at @code{fp + 0}. If the trace function then looks at 12489location @code{pc - 12} and the top 8 bits are set, then we know that 12490there is a function name embedded immediately preceding this location 12491and has length @code{((pc[-3]) & 0xff000000)}. 12492 12493@item -mthumb 12494@itemx -marm 12495@opindex marm 12496@opindex mthumb 12497 12498Select between generating code that executes in ARM and Thumb 12499states. The default for most configurations is to generate code 12500that executes in ARM state, but the default can be changed by 12501configuring GCC with the @option{--with-mode=}@var{state} 12502configure option. 12503 12504@item -mtpcs-frame 12505@opindex mtpcs-frame 12506Generate a stack frame that is compliant with the Thumb Procedure Call 12507Standard for all non-leaf functions. (A leaf function is one that does 12508not call any other functions.) The default is @option{-mno-tpcs-frame}. 12509 12510@item -mtpcs-leaf-frame 12511@opindex mtpcs-leaf-frame 12512Generate a stack frame that is compliant with the Thumb Procedure Call 12513Standard for all leaf functions. (A leaf function is one that does 12514not call any other functions.) The default is @option{-mno-apcs-leaf-frame}. 12515 12516@item -mcallee-super-interworking 12517@opindex mcallee-super-interworking 12518Gives all externally visible functions in the file being compiled an ARM 12519instruction set header which switches to Thumb mode before executing the 12520rest of the function. This allows these functions to be called from 12521non-interworking code. This option is not valid in AAPCS configurations 12522because interworking is enabled by default. 12523 12524@item -mcaller-super-interworking 12525@opindex mcaller-super-interworking 12526Allows calls via function pointers (including virtual functions) to 12527execute correctly regardless of whether the target code has been 12528compiled for interworking or not. There is a small overhead in the cost 12529of executing a function pointer if this option is enabled. This option 12530is not valid in AAPCS configurations because interworking is enabled 12531by default. 12532 12533@item -mtp=@var{name} 12534@opindex mtp 12535Specify the access model for the thread local storage pointer. The valid 12536models are @option{soft}, which generates calls to @code{__aeabi_read_tp}, 12537@option{cp15}, which fetches the thread pointer from @code{cp15} directly 12538(supported in the arm6k architecture), and @option{auto}, which uses the 12539best available method for the selected processor. The default setting is 12540@option{auto}. 12541 12542@item -mtls-dialect=@var{dialect} 12543@opindex mtls-dialect 12544Specify the dialect to use for accessing thread local storage. Two 12545@var{dialect}s are supported---@samp{gnu} and @samp{gnu2}. The 12546@samp{gnu} dialect selects the original GNU scheme for supporting 12547local and global dynamic TLS models. The @samp{gnu2} dialect 12548selects the GNU descriptor scheme, which provides better performance 12549for shared libraries. The GNU descriptor scheme is compatible with 12550the original scheme, but does require new assembler, linker and 12551library support. Initial and local exec TLS models are unaffected by 12552this option and always use the original scheme. 12553 12554@item -mword-relocations 12555@opindex mword-relocations 12556Only generate absolute relocations on word-sized values (i.e. R_ARM_ABS32). 12557This is enabled by default on targets (uClinux, SymbianOS) where the runtime 12558loader imposes this restriction, and when @option{-fpic} or @option{-fPIC} 12559is specified. 12560 12561@item -mfix-cortex-m3-ldrd 12562@opindex mfix-cortex-m3-ldrd 12563Some Cortex-M3 cores can cause data corruption when @code{ldrd} instructions 12564with overlapping destination and base registers are used. This option avoids 12565generating these instructions. This option is enabled by default when 12566@option{-mcpu=cortex-m3} is specified. 12567 12568@item -munaligned-access 12569@itemx -mno-unaligned-access 12570@opindex munaligned-access 12571@opindex mno-unaligned-access 12572Enables (or disables) reading and writing of 16- and 32- bit values 12573from addresses that are not 16- or 32- bit aligned. By default 12574unaligned access is disabled for all pre-ARMv6 and all ARMv6-M 12575architectures, and enabled for all other architectures. If unaligned 12576access is not enabled then words in packed data structures will be 12577accessed a byte at a time. 12578 12579The ARM attribute @code{Tag_CPU_unaligned_access} will be set in the 12580generated object file to either true or false, depending upon the 12581setting of this option. If unaligned access is enabled then the 12582preprocessor symbol @code{__ARM_FEATURE_UNALIGNED} will also be 12583defined. 12584 12585@item -mneon-for-64bits 12586@opindex mneon-for-64bits 12587Enables using Neon to handle scalar 64-bits operations. This is 12588disabled by default since the cost of moving data from core registers 12589to Neon is high. 12590 12591@item -mslow-flash-data 12592@opindex mslow-flash-data 12593Assume loading data from flash is slower than fetching instruction. 12594Therefore literal load is minimized for better performance. 12595This option is only supported when compiling for ARMv7 M-profile and 12596off by default. 12597 12598@item -mrestrict-it 12599@opindex mrestrict-it 12600Restricts generation of IT blocks to conform to the rules of ARMv8. 12601IT blocks can only contain a single 16-bit instruction from a select 12602set of instructions. This option is on by default for ARMv8 Thumb mode. 12603@end table 12604 12605@node AVR Options 12606@subsection AVR Options 12607@cindex AVR Options 12608 12609These options are defined for AVR implementations: 12610 12611@table @gcctabopt 12612@item -mmcu=@var{mcu} 12613@opindex mmcu 12614Specify Atmel AVR instruction set architectures (ISA) or MCU type. 12615 12616The default for this option is@tie{}@code{avr2}. 12617 12618GCC supports the following AVR devices and ISAs: 12619 12620@include avr-mmcu.texi 12621 12622@item -maccumulate-args 12623@opindex maccumulate-args 12624Accumulate outgoing function arguments and acquire/release the needed 12625stack space for outgoing function arguments once in function 12626prologue/epilogue. Without this option, outgoing arguments are pushed 12627before calling a function and popped afterwards. 12628 12629Popping the arguments after the function call can be expensive on 12630AVR so that accumulating the stack space might lead to smaller 12631executables because arguments need not to be removed from the 12632stack after such a function call. 12633 12634This option can lead to reduced code size for functions that perform 12635several calls to functions that get their arguments on the stack like 12636calls to printf-like functions. 12637 12638@item -mbranch-cost=@var{cost} 12639@opindex mbranch-cost 12640Set the branch costs for conditional branch instructions to 12641@var{cost}. Reasonable values for @var{cost} are small, non-negative 12642integers. The default branch cost is 0. 12643 12644@item -mcall-prologues 12645@opindex mcall-prologues 12646Functions prologues/epilogues are expanded as calls to appropriate 12647subroutines. Code size is smaller. 12648 12649@item -mint8 12650@opindex mint8 12651Assume @code{int} to be 8-bit integer. This affects the sizes of all types: a 12652@code{char} is 1 byte, an @code{int} is 1 byte, a @code{long} is 2 bytes, 12653and @code{long long} is 4 bytes. Please note that this option does not 12654conform to the C standards, but it results in smaller code 12655size. 12656 12657@item -mno-interrupts 12658@opindex mno-interrupts 12659Generated code is not compatible with hardware interrupts. 12660Code size is smaller. 12661 12662@item -mrelax 12663@opindex mrelax 12664Try to replace @code{CALL} resp.@: @code{JMP} instruction by the shorter 12665@code{RCALL} resp.@: @code{RJMP} instruction if applicable. 12666Setting @code{-mrelax} just adds the @code{--relax} option to the 12667linker command line when the linker is called. 12668 12669Jump relaxing is performed by the linker because jump offsets are not 12670known before code is located. Therefore, the assembler code generated by the 12671compiler is the same, but the instructions in the executable may 12672differ from instructions in the assembler code. 12673 12674Relaxing must be turned on if linker stubs are needed, see the 12675section on @code{EIND} and linker stubs below. 12676 12677@item -msp8 12678@opindex msp8 12679Treat the stack pointer register as an 8-bit register, 12680i.e.@: assume the high byte of the stack pointer is zero. 12681In general, you don't need to set this option by hand. 12682 12683This option is used internally by the compiler to select and 12684build multilibs for architectures @code{avr2} and @code{avr25}. 12685These architectures mix devices with and without @code{SPH}. 12686For any setting other than @code{-mmcu=avr2} or @code{-mmcu=avr25} 12687the compiler driver will add or remove this option from the compiler 12688proper's command line, because the compiler then knows if the device 12689or architecture has an 8-bit stack pointer and thus no @code{SPH} 12690register or not. 12691 12692@item -mstrict-X 12693@opindex mstrict-X 12694Use address register @code{X} in a way proposed by the hardware. This means 12695that @code{X} is only used in indirect, post-increment or 12696pre-decrement addressing. 12697 12698Without this option, the @code{X} register may be used in the same way 12699as @code{Y} or @code{Z} which then is emulated by additional 12700instructions. 12701For example, loading a value with @code{X+const} addressing with a 12702small non-negative @code{const < 64} to a register @var{Rn} is 12703performed as 12704 12705@example 12706adiw r26, const ; X += const 12707ld @var{Rn}, X ; @var{Rn} = *X 12708sbiw r26, const ; X -= const 12709@end example 12710 12711@item -mtiny-stack 12712@opindex mtiny-stack 12713Only change the lower 8@tie{}bits of the stack pointer. 12714 12715@item -Waddr-space-convert 12716@opindex Waddr-space-convert 12717Warn about conversions between address spaces in the case where the 12718resulting address space is not contained in the incoming address space. 12719@end table 12720 12721@subsubsection @code{EIND} and Devices with more than 128 Ki Bytes of Flash 12722@cindex @code{EIND} 12723Pointers in the implementation are 16@tie{}bits wide. 12724The address of a function or label is represented as word address so 12725that indirect jumps and calls can target any code address in the 12726range of 64@tie{}Ki words. 12727 12728In order to facilitate indirect jump on devices with more than 128@tie{}Ki 12729bytes of program memory space, there is a special function register called 12730@code{EIND} that serves as most significant part of the target address 12731when @code{EICALL} or @code{EIJMP} instructions are used. 12732 12733Indirect jumps and calls on these devices are handled as follows by 12734the compiler and are subject to some limitations: 12735 12736@itemize @bullet 12737 12738@item 12739The compiler never sets @code{EIND}. 12740 12741@item 12742The compiler uses @code{EIND} implicitely in @code{EICALL}/@code{EIJMP} 12743instructions or might read @code{EIND} directly in order to emulate an 12744indirect call/jump by means of a @code{RET} instruction. 12745 12746@item 12747The compiler assumes that @code{EIND} never changes during the startup 12748code or during the application. In particular, @code{EIND} is not 12749saved/restored in function or interrupt service routine 12750prologue/epilogue. 12751 12752@item 12753For indirect calls to functions and computed goto, the linker 12754generates @emph{stubs}. Stubs are jump pads sometimes also called 12755@emph{trampolines}. Thus, the indirect call/jump jumps to such a stub. 12756The stub contains a direct jump to the desired address. 12757 12758@item 12759Linker relaxation must be turned on so that the linker will generate 12760the stubs correctly an all situaltion. See the compiler option 12761@code{-mrelax} and the linler option @code{--relax}. 12762There are corner cases where the linker is supposed to generate stubs 12763but aborts without relaxation and without a helpful error message. 12764 12765@item 12766The default linker script is arranged for code with @code{EIND = 0}. 12767If code is supposed to work for a setup with @code{EIND != 0}, a custom 12768linker script has to be used in order to place the sections whose 12769name start with @code{.trampolines} into the segment where @code{EIND} 12770points to. 12771 12772@item 12773The startup code from libgcc never sets @code{EIND}. 12774Notice that startup code is a blend of code from libgcc and AVR-LibC. 12775For the impact of AVR-LibC on @code{EIND}, see the 12776@w{@uref{http://nongnu.org/avr-libc/user-manual/,AVR-LibC user manual}}. 12777 12778@item 12779It is legitimate for user-specific startup code to set up @code{EIND} 12780early, for example by means of initialization code located in 12781section @code{.init3}. Such code runs prior to general startup code 12782that initializes RAM and calls constructors, but after the bit 12783of startup code from AVR-LibC that sets @code{EIND} to the segment 12784where the vector table is located. 12785@example 12786#include <avr/io.h> 12787 12788static void 12789__attribute__((section(".init3"),naked,used,no_instrument_function)) 12790init3_set_eind (void) 12791@{ 12792 __asm volatile ("ldi r24,pm_hh8(__trampolines_start)\n\t" 12793 "out %i0,r24" :: "n" (&EIND) : "r24","memory"); 12794@} 12795@end example 12796 12797@noindent 12798The @code{__trampolines_start} symbol is defined in the linker script. 12799 12800@item 12801Stubs are generated automatically by the linker if 12802the following two conditions are met: 12803@itemize @minus 12804 12805@item The address of a label is taken by means of the @code{gs} modifier 12806(short for @emph{generate stubs}) like so: 12807@example 12808LDI r24, lo8(gs(@var{func})) 12809LDI r25, hi8(gs(@var{func})) 12810@end example 12811@item The final location of that label is in a code segment 12812@emph{outside} the segment where the stubs are located. 12813@end itemize 12814 12815@item 12816The compiler emits such @code{gs} modifiers for code labels in the 12817following situations: 12818@itemize @minus 12819@item Taking address of a function or code label. 12820@item Computed goto. 12821@item If prologue-save function is used, see @option{-mcall-prologues} 12822command-line option. 12823@item Switch/case dispatch tables. If you do not want such dispatch 12824tables you can specify the @option{-fno-jump-tables} command-line option. 12825@item C and C++ constructors/destructors called during startup/shutdown. 12826@item If the tools hit a @code{gs()} modifier explained above. 12827@end itemize 12828 12829@item 12830Jumping to non-symbolic addresses like so is @emph{not} supported: 12831 12832@example 12833int main (void) 12834@{ 12835 /* Call function at word address 0x2 */ 12836 return ((int(*)(void)) 0x2)(); 12837@} 12838@end example 12839 12840Instead, a stub has to be set up, i.e.@: the function has to be called 12841through a symbol (@code{func_4} in the example): 12842 12843@example 12844int main (void) 12845@{ 12846 extern int func_4 (void); 12847 12848 /* Call function at byte address 0x4 */ 12849 return func_4(); 12850@} 12851@end example 12852 12853and the application be linked with @code{-Wl,--defsym,func_4=0x4}. 12854Alternatively, @code{func_4} can be defined in the linker script. 12855@end itemize 12856 12857@subsubsection Handling of the @code{RAMPD}, @code{RAMPX}, @code{RAMPY} and @code{RAMPZ} Special Function Registers 12858@cindex @code{RAMPD} 12859@cindex @code{RAMPX} 12860@cindex @code{RAMPY} 12861@cindex @code{RAMPZ} 12862Some AVR devices support memories larger than the 64@tie{}KiB range 12863that can be accessed with 16-bit pointers. To access memory locations 12864outside this 64@tie{}KiB range, the contentent of a @code{RAMP} 12865register is used as high part of the address: 12866The @code{X}, @code{Y}, @code{Z} address register is concatenated 12867with the @code{RAMPX}, @code{RAMPY}, @code{RAMPZ} special function 12868register, respectively, to get a wide address. Similarly, 12869@code{RAMPD} is used together with direct addressing. 12870 12871@itemize 12872@item 12873The startup code initializes the @code{RAMP} special function 12874registers with zero. 12875 12876@item 12877If a @ref{AVR Named Address Spaces,named address space} other than 12878generic or @code{__flash} is used, then @code{RAMPZ} is set 12879as needed before the operation. 12880 12881@item 12882If the device supports RAM larger than 64@tie{}KiB and the compiler 12883needs to change @code{RAMPZ} to accomplish an operation, @code{RAMPZ} 12884is reset to zero after the operation. 12885 12886@item 12887If the device comes with a specific @code{RAMP} register, the ISR 12888prologue/epilogue saves/restores that SFR and initializes it with 12889zero in case the ISR code might (implicitly) use it. 12890 12891@item 12892RAM larger than 64@tie{}KiB is not supported by GCC for AVR targets. 12893If you use inline assembler to read from locations outside the 1289416-bit address range and change one of the @code{RAMP} registers, 12895you must reset it to zero after the access. 12896 12897@end itemize 12898 12899@subsubsection AVR Built-in Macros 12900 12901GCC defines several built-in macros so that the user code can test 12902for the presence or absence of features. Almost any of the following 12903built-in macros are deduced from device capabilities and thus 12904triggered by the @code{-mmcu=} command-line option. 12905 12906For even more AVR-specific built-in macros see 12907@ref{AVR Named Address Spaces} and @ref{AVR Built-in Functions}. 12908 12909@table @code 12910 12911@item __AVR_ARCH__ 12912Build-in macro that resolves to a decimal number that identifies the 12913architecture and depends on the @code{-mmcu=@var{mcu}} option. 12914Possible values are: 12915 12916@code{2}, @code{25}, @code{3}, @code{31}, @code{35}, 12917@code{4}, @code{5}, @code{51}, @code{6}, @code{102}, @code{104}, 12918@code{105}, @code{106}, @code{107} 12919 12920for @var{mcu}=@code{avr2}, @code{avr25}, @code{avr3}, 12921@code{avr31}, @code{avr35}, @code{avr4}, @code{avr5}, @code{avr51}, 12922@code{avr6}, @code{avrxmega2}, @code{avrxmega4}, @code{avrxmega5}, 12923@code{avrxmega6}, @code{avrxmega7}, respectively. 12924If @var{mcu} specifies a device, this built-in macro is set 12925accordingly. For example, with @code{-mmcu=atmega8} the macro will be 12926defined to @code{4}. 12927 12928@item __AVR_@var{Device}__ 12929Setting @code{-mmcu=@var{device}} defines this built-in macro which reflects 12930the device's name. For example, @code{-mmcu=atmega8} defines the 12931built-in macro @code{__AVR_ATmega8__}, @code{-mmcu=attiny261a} defines 12932@code{__AVR_ATtiny261A__}, etc. 12933 12934The built-in macros' names follow 12935the scheme @code{__AVR_@var{Device}__} where @var{Device} is 12936the device name as from the AVR user manual. The difference between 12937@var{Device} in the built-in macro and @var{device} in 12938@code{-mmcu=@var{device}} is that the latter is always lowercase. 12939 12940If @var{device} is not a device but only a core architecture like 12941@code{avr51}, this macro will not be defined. 12942 12943@item __AVR_XMEGA__ 12944The device / architecture belongs to the XMEGA family of devices. 12945 12946@item __AVR_HAVE_ELPM__ 12947The device has the the @code{ELPM} instruction. 12948 12949@item __AVR_HAVE_ELPMX__ 12950The device has the @code{ELPM R@var{n},Z} and @code{ELPM 12951R@var{n},Z+} instructions. 12952 12953@item __AVR_HAVE_MOVW__ 12954The device has the @code{MOVW} instruction to perform 16-bit 12955register-register moves. 12956 12957@item __AVR_HAVE_LPMX__ 12958The device has the @code{LPM R@var{n},Z} and 12959@code{LPM R@var{n},Z+} instructions. 12960 12961@item __AVR_HAVE_MUL__ 12962The device has a hardware multiplier. 12963 12964@item __AVR_HAVE_JMP_CALL__ 12965The device has the @code{JMP} and @code{CALL} instructions. 12966This is the case for devices with at least 16@tie{}KiB of program 12967memory. 12968 12969@item __AVR_HAVE_EIJMP_EICALL__ 12970@itemx __AVR_3_BYTE_PC__ 12971The device has the @code{EIJMP} and @code{EICALL} instructions. 12972This is the case for devices with more than 128@tie{}KiB of program memory. 12973This also means that the program counter 12974(PC) is 3@tie{}bytes wide. 12975 12976@item __AVR_2_BYTE_PC__ 12977The program counter (PC) is 2@tie{}bytes wide. This is the case for devices 12978with up to 128@tie{}KiB of program memory. 12979 12980@item __AVR_HAVE_8BIT_SP__ 12981@itemx __AVR_HAVE_16BIT_SP__ 12982The stack pointer (SP) register is treated as 8-bit respectively 1298316-bit register by the compiler. 12984The definition of these macros is affected by @code{-mtiny-stack}. 12985 12986@item __AVR_HAVE_SPH__ 12987@itemx __AVR_SP8__ 12988The device has the SPH (high part of stack pointer) special function 12989register or has an 8-bit stack pointer, respectively. 12990The definition of these macros is affected by @code{-mmcu=} and 12991in the cases of @code{-mmcu=avr2} and @code{-mmcu=avr25} also 12992by @code{-msp8}. 12993 12994@item __AVR_HAVE_RAMPD__ 12995@itemx __AVR_HAVE_RAMPX__ 12996@itemx __AVR_HAVE_RAMPY__ 12997@itemx __AVR_HAVE_RAMPZ__ 12998The device has the @code{RAMPD}, @code{RAMPX}, @code{RAMPY}, 12999@code{RAMPZ} special function register, respectively. 13000 13001@item __NO_INTERRUPTS__ 13002This macro reflects the @code{-mno-interrupts} command line option. 13003 13004@item __AVR_ERRATA_SKIP__ 13005@itemx __AVR_ERRATA_SKIP_JMP_CALL__ 13006Some AVR devices (AT90S8515, ATmega103) must not skip 32-bit 13007instructions because of a hardware erratum. Skip instructions are 13008@code{SBRS}, @code{SBRC}, @code{SBIS}, @code{SBIC} and @code{CPSE}. 13009The second macro is only defined if @code{__AVR_HAVE_JMP_CALL__} is also 13010set. 13011 13012@item __AVR_ISA_RMW__ 13013The device has Read-Modify-Write instructions (XCH, LAC, LAS and LAT). 13014 13015@item __AVR_SFR_OFFSET__=@var{offset} 13016Instructions that can address I/O special function registers directly 13017like @code{IN}, @code{OUT}, @code{SBI}, etc.@: may use a different 13018address as if addressed by an instruction to access RAM like @code{LD} 13019or @code{STS}. This offset depends on the device architecture and has 13020to be subtracted from the RAM address in order to get the 13021respective I/O@tie{}address. 13022 13023@item __WITH_AVRLIBC__ 13024The compiler is configured to be used together with AVR-Libc. 13025See the @code{--with-avrlibc} configure option. 13026 13027@end table 13028 13029@node Blackfin Options 13030@subsection Blackfin Options 13031@cindex Blackfin Options 13032 13033@table @gcctabopt 13034@item -mcpu=@var{cpu}@r{[}-@var{sirevision}@r{]} 13035@opindex mcpu= 13036Specifies the name of the target Blackfin processor. Currently, @var{cpu} 13037can be one of @samp{bf512}, @samp{bf514}, @samp{bf516}, @samp{bf518}, 13038@samp{bf522}, @samp{bf523}, @samp{bf524}, @samp{bf525}, @samp{bf526}, 13039@samp{bf527}, @samp{bf531}, @samp{bf532}, @samp{bf533}, 13040@samp{bf534}, @samp{bf536}, @samp{bf537}, @samp{bf538}, @samp{bf539}, 13041@samp{bf542}, @samp{bf544}, @samp{bf547}, @samp{bf548}, @samp{bf549}, 13042@samp{bf542m}, @samp{bf544m}, @samp{bf547m}, @samp{bf548m}, @samp{bf549m}, 13043@samp{bf561}, @samp{bf592}. 13044 13045The optional @var{sirevision} specifies the silicon revision of the target 13046Blackfin processor. Any workarounds available for the targeted silicon revision 13047are enabled. If @var{sirevision} is @samp{none}, no workarounds are enabled. 13048If @var{sirevision} is @samp{any}, all workarounds for the targeted processor 13049are enabled. The @code{__SILICON_REVISION__} macro is defined to two 13050hexadecimal digits representing the major and minor numbers in the silicon 13051revision. If @var{sirevision} is @samp{none}, the @code{__SILICON_REVISION__} 13052is not defined. If @var{sirevision} is @samp{any}, the 13053@code{__SILICON_REVISION__} is defined to be @code{0xffff}. 13054If this optional @var{sirevision} is not used, GCC assumes the latest known 13055silicon revision of the targeted Blackfin processor. 13056 13057GCC defines a preprocessor macro for the specified @var{cpu}. 13058For the @samp{bfin-elf} toolchain, this option causes the hardware BSP 13059provided by libgloss to be linked in if @option{-msim} is not given. 13060 13061Without this option, @samp{bf532} is used as the processor by default. 13062 13063Note that support for @samp{bf561} is incomplete. For @samp{bf561}, 13064only the preprocessor macro is defined. 13065 13066@item -msim 13067@opindex msim 13068Specifies that the program will be run on the simulator. This causes 13069the simulator BSP provided by libgloss to be linked in. This option 13070has effect only for @samp{bfin-elf} toolchain. 13071Certain other options, such as @option{-mid-shared-library} and 13072@option{-mfdpic}, imply @option{-msim}. 13073 13074@item -momit-leaf-frame-pointer 13075@opindex momit-leaf-frame-pointer 13076Don't keep the frame pointer in a register for leaf functions. This 13077avoids the instructions to save, set up and restore frame pointers and 13078makes an extra register available in leaf functions. The option 13079@option{-fomit-frame-pointer} removes the frame pointer for all functions, 13080which might make debugging harder. 13081 13082@item -mspecld-anomaly 13083@opindex mspecld-anomaly 13084When enabled, the compiler ensures that the generated code does not 13085contain speculative loads after jump instructions. If this option is used, 13086@code{__WORKAROUND_SPECULATIVE_LOADS} is defined. 13087 13088@item -mno-specld-anomaly 13089@opindex mno-specld-anomaly 13090Don't generate extra code to prevent speculative loads from occurring. 13091 13092@item -mcsync-anomaly 13093@opindex mcsync-anomaly 13094When enabled, the compiler ensures that the generated code does not 13095contain CSYNC or SSYNC instructions too soon after conditional branches. 13096If this option is used, @code{__WORKAROUND_SPECULATIVE_SYNCS} is defined. 13097 13098@item -mno-csync-anomaly 13099@opindex mno-csync-anomaly 13100Don't generate extra code to prevent CSYNC or SSYNC instructions from 13101occurring too soon after a conditional branch. 13102 13103@item -mlow-64k 13104@opindex mlow-64k 13105When enabled, the compiler is free to take advantage of the knowledge that 13106the entire program fits into the low 64k of memory. 13107 13108@item -mno-low-64k 13109@opindex mno-low-64k 13110Assume that the program is arbitrarily large. This is the default. 13111 13112@item -mstack-check-l1 13113@opindex mstack-check-l1 13114Do stack checking using information placed into L1 scratchpad memory by the 13115uClinux kernel. 13116 13117@item -mid-shared-library 13118@opindex mid-shared-library 13119Generate code that supports shared libraries via the library ID method. 13120This allows for execute in place and shared libraries in an environment 13121without virtual memory management. This option implies @option{-fPIC}. 13122With a @samp{bfin-elf} target, this option implies @option{-msim}. 13123 13124@item -mno-id-shared-library 13125@opindex mno-id-shared-library 13126Generate code that doesn't assume ID-based shared libraries are being used. 13127This is the default. 13128 13129@item -mleaf-id-shared-library 13130@opindex mleaf-id-shared-library 13131Generate code that supports shared libraries via the library ID method, 13132but assumes that this library or executable won't link against any other 13133ID shared libraries. That allows the compiler to use faster code for jumps 13134and calls. 13135 13136@item -mno-leaf-id-shared-library 13137@opindex mno-leaf-id-shared-library 13138Do not assume that the code being compiled won't link against any ID shared 13139libraries. Slower code is generated for jump and call insns. 13140 13141@item -mshared-library-id=n 13142@opindex mshared-library-id 13143Specifies the identification number of the ID-based shared library being 13144compiled. Specifying a value of 0 generates more compact code; specifying 13145other values forces the allocation of that number to the current 13146library but is no more space- or time-efficient than omitting this option. 13147 13148@item -msep-data 13149@opindex msep-data 13150Generate code that allows the data segment to be located in a different 13151area of memory from the text segment. This allows for execute in place in 13152an environment without virtual memory management by eliminating relocations 13153against the text section. 13154 13155@item -mno-sep-data 13156@opindex mno-sep-data 13157Generate code that assumes that the data segment follows the text segment. 13158This is the default. 13159 13160@item -mlong-calls 13161@itemx -mno-long-calls 13162@opindex mlong-calls 13163@opindex mno-long-calls 13164Tells the compiler to perform function calls by first loading the 13165address of the function into a register and then performing a subroutine 13166call on this register. This switch is needed if the target function 13167lies outside of the 24-bit addressing range of the offset-based 13168version of subroutine call instruction. 13169 13170This feature is not enabled by default. Specifying 13171@option{-mno-long-calls} restores the default behavior. Note these 13172switches have no effect on how the compiler generates code to handle 13173function calls via function pointers. 13174 13175@item -mfast-fp 13176@opindex mfast-fp 13177Link with the fast floating-point library. This library relaxes some of 13178the IEEE floating-point standard's rules for checking inputs against 13179Not-a-Number (NAN), in the interest of performance. 13180 13181@item -minline-plt 13182@opindex minline-plt 13183Enable inlining of PLT entries in function calls to functions that are 13184not known to bind locally. It has no effect without @option{-mfdpic}. 13185 13186@item -mmulticore 13187@opindex mmulticore 13188Build a standalone application for multicore Blackfin processors. 13189This option causes proper start files and link scripts supporting 13190multicore to be used, and defines the macro @code{__BFIN_MULTICORE}. 13191It can only be used with @option{-mcpu=bf561@r{[}-@var{sirevision}@r{]}}. 13192 13193This option can be used with @option{-mcorea} or @option{-mcoreb}, which 13194selects the one-application-per-core programming model. Without 13195@option{-mcorea} or @option{-mcoreb}, the single-application/dual-core 13196programming model is used. In this model, the main function of Core B 13197should be named as @code{coreb_main}. 13198 13199If this option is not used, the single-core application programming 13200model is used. 13201 13202@item -mcorea 13203@opindex mcorea 13204Build a standalone application for Core A of BF561 when using 13205the one-application-per-core programming model. Proper start files 13206and link scripts are used to support Core A, and the macro 13207@code{__BFIN_COREA} is defined. 13208This option can only be used in conjunction with @option{-mmulticore}. 13209 13210@item -mcoreb 13211@opindex mcoreb 13212Build a standalone application for Core B of BF561 when using 13213the one-application-per-core programming model. Proper start files 13214and link scripts are used to support Core B, and the macro 13215@code{__BFIN_COREB} is defined. When this option is used, @code{coreb_main} 13216should be used instead of @code{main}. 13217This option can only be used in conjunction with @option{-mmulticore}. 13218 13219@item -msdram 13220@opindex msdram 13221Build a standalone application for SDRAM. Proper start files and 13222link scripts are used to put the application into SDRAM, and the macro 13223@code{__BFIN_SDRAM} is defined. 13224The loader should initialize SDRAM before loading the application. 13225 13226@item -micplb 13227@opindex micplb 13228Assume that ICPLBs are enabled at run time. This has an effect on certain 13229anomaly workarounds. For Linux targets, the default is to assume ICPLBs 13230are enabled; for standalone applications the default is off. 13231@end table 13232 13233@node C6X Options 13234@subsection C6X Options 13235@cindex C6X Options 13236 13237@table @gcctabopt 13238@item -march=@var{name} 13239@opindex march 13240This specifies the name of the target architecture. GCC uses this 13241name to determine what kind of instructions it can emit when generating 13242assembly code. Permissible names are: @samp{c62x}, 13243@samp{c64x}, @samp{c64x+}, @samp{c67x}, @samp{c67x+}, @samp{c674x}. 13244 13245@item -mbig-endian 13246@opindex mbig-endian 13247Generate code for a big-endian target. 13248 13249@item -mlittle-endian 13250@opindex mlittle-endian 13251Generate code for a little-endian target. This is the default. 13252 13253@item -msim 13254@opindex msim 13255Choose startup files and linker script suitable for the simulator. 13256 13257@item -msdata=default 13258@opindex msdata=default 13259Put small global and static data in the @samp{.neardata} section, 13260which is pointed to by register @code{B14}. Put small uninitialized 13261global and static data in the @samp{.bss} section, which is adjacent 13262to the @samp{.neardata} section. Put small read-only data into the 13263@samp{.rodata} section. The corresponding sections used for large 13264pieces of data are @samp{.fardata}, @samp{.far} and @samp{.const}. 13265 13266@item -msdata=all 13267@opindex msdata=all 13268Put all data, not just small objects, into the sections reserved for 13269small data, and use addressing relative to the @code{B14} register to 13270access them. 13271 13272@item -msdata=none 13273@opindex msdata=none 13274Make no use of the sections reserved for small data, and use absolute 13275addresses to access all data. Put all initialized global and static 13276data in the @samp{.fardata} section, and all uninitialized data in the 13277@samp{.far} section. Put all constant data into the @samp{.const} 13278section. 13279@end table 13280 13281@node CRIS Options 13282@subsection CRIS Options 13283@cindex CRIS Options 13284 13285These options are defined specifically for the CRIS ports. 13286 13287@table @gcctabopt 13288@item -march=@var{architecture-type} 13289@itemx -mcpu=@var{architecture-type} 13290@opindex march 13291@opindex mcpu 13292Generate code for the specified architecture. The choices for 13293@var{architecture-type} are @samp{v3}, @samp{v8} and @samp{v10} for 13294respectively ETRAX@w{ }4, ETRAX@w{ }100, and ETRAX@w{ }100@w{ }LX@. 13295Default is @samp{v0} except for cris-axis-linux-gnu, where the default is 13296@samp{v10}. 13297 13298@item -mtune=@var{architecture-type} 13299@opindex mtune 13300Tune to @var{architecture-type} everything applicable about the generated 13301code, except for the ABI and the set of available instructions. The 13302choices for @var{architecture-type} are the same as for 13303@option{-march=@var{architecture-type}}. 13304 13305@item -mmax-stack-frame=@var{n} 13306@opindex mmax-stack-frame 13307Warn when the stack frame of a function exceeds @var{n} bytes. 13308 13309@item -metrax4 13310@itemx -metrax100 13311@opindex metrax4 13312@opindex metrax100 13313The options @option{-metrax4} and @option{-metrax100} are synonyms for 13314@option{-march=v3} and @option{-march=v8} respectively. 13315 13316@item -mmul-bug-workaround 13317@itemx -mno-mul-bug-workaround 13318@opindex mmul-bug-workaround 13319@opindex mno-mul-bug-workaround 13320Work around a bug in the @code{muls} and @code{mulu} instructions for CPU 13321models where it applies. This option is active by default. 13322 13323@item -mpdebug 13324@opindex mpdebug 13325Enable CRIS-specific verbose debug-related information in the assembly 13326code. This option also has the effect of turning off the @samp{#NO_APP} 13327formatted-code indicator to the assembler at the beginning of the 13328assembly file. 13329 13330@item -mcc-init 13331@opindex mcc-init 13332Do not use condition-code results from previous instruction; always emit 13333compare and test instructions before use of condition codes. 13334 13335@item -mno-side-effects 13336@opindex mno-side-effects 13337Do not emit instructions with side effects in addressing modes other than 13338post-increment. 13339 13340@item -mstack-align 13341@itemx -mno-stack-align 13342@itemx -mdata-align 13343@itemx -mno-data-align 13344@itemx -mconst-align 13345@itemx -mno-const-align 13346@opindex mstack-align 13347@opindex mno-stack-align 13348@opindex mdata-align 13349@opindex mno-data-align 13350@opindex mconst-align 13351@opindex mno-const-align 13352These options (@samp{no-} options) arrange (eliminate arrangements) for the 13353stack frame, individual data and constants to be aligned for the maximum 13354single data access size for the chosen CPU model. The default is to 13355arrange for 32-bit alignment. ABI details such as structure layout are 13356not affected by these options. 13357 13358@item -m32-bit 13359@itemx -m16-bit 13360@itemx -m8-bit 13361@opindex m32-bit 13362@opindex m16-bit 13363@opindex m8-bit 13364Similar to the stack- data- and const-align options above, these options 13365arrange for stack frame, writable data and constants to all be 32-bit, 1336616-bit or 8-bit aligned. The default is 32-bit alignment. 13367 13368@item -mno-prologue-epilogue 13369@itemx -mprologue-epilogue 13370@opindex mno-prologue-epilogue 13371@opindex mprologue-epilogue 13372With @option{-mno-prologue-epilogue}, the normal function prologue and 13373epilogue which set up the stack frame are omitted and no return 13374instructions or return sequences are generated in the code. Use this 13375option only together with visual inspection of the compiled code: no 13376warnings or errors are generated when call-saved registers must be saved, 13377or storage for local variables needs to be allocated. 13378 13379@item -mno-gotplt 13380@itemx -mgotplt 13381@opindex mno-gotplt 13382@opindex mgotplt 13383With @option{-fpic} and @option{-fPIC}, don't generate (do generate) 13384instruction sequences that load addresses for functions from the PLT part 13385of the GOT rather than (traditional on other architectures) calls to the 13386PLT@. The default is @option{-mgotplt}. 13387 13388@item -melf 13389@opindex melf 13390Legacy no-op option only recognized with the cris-axis-elf and 13391cris-axis-linux-gnu targets. 13392 13393@item -mlinux 13394@opindex mlinux 13395Legacy no-op option only recognized with the cris-axis-linux-gnu target. 13396 13397@item -sim 13398@opindex sim 13399This option, recognized for the cris-axis-elf, arranges 13400to link with input-output functions from a simulator library. Code, 13401initialized data and zero-initialized data are allocated consecutively. 13402 13403@item -sim2 13404@opindex sim2 13405Like @option{-sim}, but pass linker options to locate initialized data at 134060x40000000 and zero-initialized data at 0x80000000. 13407@end table 13408 13409@node CR16 Options 13410@subsection CR16 Options 13411@cindex CR16 Options 13412 13413These options are defined specifically for the CR16 ports. 13414 13415@table @gcctabopt 13416 13417@item -mmac 13418@opindex mmac 13419Enable the use of multiply-accumulate instructions. Disabled by default. 13420 13421@item -mcr16cplus 13422@itemx -mcr16c 13423@opindex mcr16cplus 13424@opindex mcr16c 13425Generate code for CR16C or CR16C+ architecture. CR16C+ architecture 13426is default. 13427 13428@item -msim 13429@opindex msim 13430Links the library libsim.a which is in compatible with simulator. Applicable 13431to ELF compiler only. 13432 13433@item -mint32 13434@opindex mint32 13435Choose integer type as 32-bit wide. 13436 13437@item -mbit-ops 13438@opindex mbit-ops 13439Generates @code{sbit}/@code{cbit} instructions for bit manipulations. 13440 13441@item -mdata-model=@var{model} 13442@opindex mdata-model 13443Choose a data model. The choices for @var{model} are @samp{near}, 13444@samp{far} or @samp{medium}. @samp{medium} is default. 13445However, @samp{far} is not valid with @option{-mcr16c}, as the 13446CR16C architecture does not support the far data model. 13447@end table 13448 13449@node Darwin Options 13450@subsection Darwin Options 13451@cindex Darwin options 13452 13453These options are defined for all architectures running the Darwin operating 13454system. 13455 13456FSF GCC on Darwin does not create ``fat'' object files; it creates 13457an object file for the single architecture that GCC was built to 13458target. Apple's GCC on Darwin does create ``fat'' files if multiple 13459@option{-arch} options are used; it does so by running the compiler or 13460linker multiple times and joining the results together with 13461@file{lipo}. 13462 13463The subtype of the file created (like @samp{ppc7400} or @samp{ppc970} or 13464@samp{i686}) is determined by the flags that specify the ISA 13465that GCC is targeting, like @option{-mcpu} or @option{-march}. The 13466@option{-force_cpusubtype_ALL} option can be used to override this. 13467 13468The Darwin tools vary in their behavior when presented with an ISA 13469mismatch. The assembler, @file{as}, only permits instructions to 13470be used that are valid for the subtype of the file it is generating, 13471so you cannot put 64-bit instructions in a @samp{ppc750} object file. 13472The linker for shared libraries, @file{/usr/bin/libtool}, fails 13473and prints an error if asked to create a shared library with a less 13474restrictive subtype than its input files (for instance, trying to put 13475a @samp{ppc970} object file in a @samp{ppc7400} library). The linker 13476for executables, @command{ld}, quietly gives the executable the most 13477restrictive subtype of any of its input files. 13478 13479@table @gcctabopt 13480@item -F@var{dir} 13481@opindex F 13482Add the framework directory @var{dir} to the head of the list of 13483directories to be searched for header files. These directories are 13484interleaved with those specified by @option{-I} options and are 13485scanned in a left-to-right order. 13486 13487A framework directory is a directory with frameworks in it. A 13488framework is a directory with a @file{Headers} and/or 13489@file{PrivateHeaders} directory contained directly in it that ends 13490in @file{.framework}. The name of a framework is the name of this 13491directory excluding the @file{.framework}. Headers associated with 13492the framework are found in one of those two directories, with 13493@file{Headers} being searched first. A subframework is a framework 13494directory that is in a framework's @file{Frameworks} directory. 13495Includes of subframework headers can only appear in a header of a 13496framework that contains the subframework, or in a sibling subframework 13497header. Two subframeworks are siblings if they occur in the same 13498framework. A subframework should not have the same name as a 13499framework; a warning is issued if this is violated. Currently a 13500subframework cannot have subframeworks; in the future, the mechanism 13501may be extended to support this. The standard frameworks can be found 13502in @file{/System/Library/Frameworks} and 13503@file{/Library/Frameworks}. An example include looks like 13504@code{#include <Framework/header.h>}, where @file{Framework} denotes 13505the name of the framework and @file{header.h} is found in the 13506@file{PrivateHeaders} or @file{Headers} directory. 13507 13508@item -iframework@var{dir} 13509@opindex iframework 13510Like @option{-F} except the directory is a treated as a system 13511directory. The main difference between this @option{-iframework} and 13512@option{-F} is that with @option{-iframework} the compiler does not 13513warn about constructs contained within header files found via 13514@var{dir}. This option is valid only for the C family of languages. 13515 13516@item -gused 13517@opindex gused 13518Emit debugging information for symbols that are used. For stabs 13519debugging format, this enables @option{-feliminate-unused-debug-symbols}. 13520This is by default ON@. 13521 13522@item -gfull 13523@opindex gfull 13524Emit debugging information for all symbols and types. 13525 13526@item -mmacosx-version-min=@var{version} 13527The earliest version of MacOS X that this executable will run on 13528is @var{version}. Typical values of @var{version} include @code{10.1}, 13529@code{10.2}, and @code{10.3.9}. 13530 13531If the compiler was built to use the system's headers by default, 13532then the default for this option is the system version on which the 13533compiler is running, otherwise the default is to make choices that 13534are compatible with as many systems and code bases as possible. 13535 13536@item -mkernel 13537@opindex mkernel 13538Enable kernel development mode. The @option{-mkernel} option sets 13539@option{-static}, @option{-fno-common}, @option{-fno-use-cxa-atexit}, 13540@option{-fno-exceptions}, @option{-fno-non-call-exceptions}, 13541@option{-fapple-kext}, @option{-fno-weak} and @option{-fno-rtti} where 13542applicable. This mode also sets @option{-mno-altivec}, 13543@option{-msoft-float}, @option{-fno-builtin} and 13544@option{-mlong-branch} for PowerPC targets. 13545 13546@item -mone-byte-bool 13547@opindex mone-byte-bool 13548Override the defaults for @samp{bool} so that @samp{sizeof(bool)==1}. 13549By default @samp{sizeof(bool)} is @samp{4} when compiling for 13550Darwin/PowerPC and @samp{1} when compiling for Darwin/x86, so this 13551option has no effect on x86. 13552 13553@strong{Warning:} The @option{-mone-byte-bool} switch causes GCC 13554to generate code that is not binary compatible with code generated 13555without that switch. Using this switch may require recompiling all 13556other modules in a program, including system libraries. Use this 13557switch to conform to a non-default data model. 13558 13559@item -mfix-and-continue 13560@itemx -ffix-and-continue 13561@itemx -findirect-data 13562@opindex mfix-and-continue 13563@opindex ffix-and-continue 13564@opindex findirect-data 13565Generate code suitable for fast turnaround development, such as to 13566allow GDB to dynamically load @code{.o} files into already-running 13567programs. @option{-findirect-data} and @option{-ffix-and-continue} 13568are provided for backwards compatibility. 13569 13570@item -all_load 13571@opindex all_load 13572Loads all members of static archive libraries. 13573See man ld(1) for more information. 13574 13575@item -arch_errors_fatal 13576@opindex arch_errors_fatal 13577Cause the errors having to do with files that have the wrong architecture 13578to be fatal. 13579 13580@item -bind_at_load 13581@opindex bind_at_load 13582Causes the output file to be marked such that the dynamic linker will 13583bind all undefined references when the file is loaded or launched. 13584 13585@item -bundle 13586@opindex bundle 13587Produce a Mach-o bundle format file. 13588See man ld(1) for more information. 13589 13590@item -bundle_loader @var{executable} 13591@opindex bundle_loader 13592This option specifies the @var{executable} that will load the build 13593output file being linked. See man ld(1) for more information. 13594 13595@item -dynamiclib 13596@opindex dynamiclib 13597When passed this option, GCC produces a dynamic library instead of 13598an executable when linking, using the Darwin @file{libtool} command. 13599 13600@item -force_cpusubtype_ALL 13601@opindex force_cpusubtype_ALL 13602This causes GCC's output file to have the @var{ALL} subtype, instead of 13603one controlled by the @option{-mcpu} or @option{-march} option. 13604 13605@item -allowable_client @var{client_name} 13606@itemx -client_name 13607@itemx -compatibility_version 13608@itemx -current_version 13609@itemx -dead_strip 13610@itemx -dependency-file 13611@itemx -dylib_file 13612@itemx -dylinker_install_name 13613@itemx -dynamic 13614@itemx -exported_symbols_list 13615@itemx -filelist 13616@need 800 13617@itemx -flat_namespace 13618@itemx -force_flat_namespace 13619@itemx -headerpad_max_install_names 13620@itemx -image_base 13621@itemx -init 13622@itemx -install_name 13623@itemx -keep_private_externs 13624@itemx -multi_module 13625@itemx -multiply_defined 13626@itemx -multiply_defined_unused 13627@need 800 13628@itemx -noall_load 13629@itemx -no_dead_strip_inits_and_terms 13630@itemx -nofixprebinding 13631@itemx -nomultidefs 13632@itemx -noprebind 13633@itemx -noseglinkedit 13634@itemx -pagezero_size 13635@itemx -prebind 13636@itemx -prebind_all_twolevel_modules 13637@itemx -private_bundle 13638@need 800 13639@itemx -read_only_relocs 13640@itemx -sectalign 13641@itemx -sectobjectsymbols 13642@itemx -whyload 13643@itemx -seg1addr 13644@itemx -sectcreate 13645@itemx -sectobjectsymbols 13646@itemx -sectorder 13647@itemx -segaddr 13648@itemx -segs_read_only_addr 13649@need 800 13650@itemx -segs_read_write_addr 13651@itemx -seg_addr_table 13652@itemx -seg_addr_table_filename 13653@itemx -seglinkedit 13654@itemx -segprot 13655@itemx -segs_read_only_addr 13656@itemx -segs_read_write_addr 13657@itemx -single_module 13658@itemx -static 13659@itemx -sub_library 13660@need 800 13661@itemx -sub_umbrella 13662@itemx -twolevel_namespace 13663@itemx -umbrella 13664@itemx -undefined 13665@itemx -unexported_symbols_list 13666@itemx -weak_reference_mismatches 13667@itemx -whatsloaded 13668@opindex allowable_client 13669@opindex client_name 13670@opindex compatibility_version 13671@opindex current_version 13672@opindex dead_strip 13673@opindex dependency-file 13674@opindex dylib_file 13675@opindex dylinker_install_name 13676@opindex dynamic 13677@opindex exported_symbols_list 13678@opindex filelist 13679@opindex flat_namespace 13680@opindex force_flat_namespace 13681@opindex headerpad_max_install_names 13682@opindex image_base 13683@opindex init 13684@opindex install_name 13685@opindex keep_private_externs 13686@opindex multi_module 13687@opindex multiply_defined 13688@opindex multiply_defined_unused 13689@opindex noall_load 13690@opindex no_dead_strip_inits_and_terms 13691@opindex nofixprebinding 13692@opindex nomultidefs 13693@opindex noprebind 13694@opindex noseglinkedit 13695@opindex pagezero_size 13696@opindex prebind 13697@opindex prebind_all_twolevel_modules 13698@opindex private_bundle 13699@opindex read_only_relocs 13700@opindex sectalign 13701@opindex sectobjectsymbols 13702@opindex whyload 13703@opindex seg1addr 13704@opindex sectcreate 13705@opindex sectobjectsymbols 13706@opindex sectorder 13707@opindex segaddr 13708@opindex segs_read_only_addr 13709@opindex segs_read_write_addr 13710@opindex seg_addr_table 13711@opindex seg_addr_table_filename 13712@opindex seglinkedit 13713@opindex segprot 13714@opindex segs_read_only_addr 13715@opindex segs_read_write_addr 13716@opindex single_module 13717@opindex static 13718@opindex sub_library 13719@opindex sub_umbrella 13720@opindex twolevel_namespace 13721@opindex umbrella 13722@opindex undefined 13723@opindex unexported_symbols_list 13724@opindex weak_reference_mismatches 13725@opindex whatsloaded 13726These options are passed to the Darwin linker. The Darwin linker man page 13727describes them in detail. 13728@end table 13729 13730@node DEC Alpha Options 13731@subsection DEC Alpha Options 13732 13733These @samp{-m} options are defined for the DEC Alpha implementations: 13734 13735@table @gcctabopt 13736@item -mno-soft-float 13737@itemx -msoft-float 13738@opindex mno-soft-float 13739@opindex msoft-float 13740Use (do not use) the hardware floating-point instructions for 13741floating-point operations. When @option{-msoft-float} is specified, 13742functions in @file{libgcc.a} are used to perform floating-point 13743operations. Unless they are replaced by routines that emulate the 13744floating-point operations, or compiled in such a way as to call such 13745emulations routines, these routines issue floating-point 13746operations. If you are compiling for an Alpha without floating-point 13747operations, you must ensure that the library is built so as not to call 13748them. 13749 13750Note that Alpha implementations without floating-point operations are 13751required to have floating-point registers. 13752 13753@item -mfp-reg 13754@itemx -mno-fp-regs 13755@opindex mfp-reg 13756@opindex mno-fp-regs 13757Generate code that uses (does not use) the floating-point register set. 13758@option{-mno-fp-regs} implies @option{-msoft-float}. If the floating-point 13759register set is not used, floating-point operands are passed in integer 13760registers as if they were integers and floating-point results are passed 13761in @code{$0} instead of @code{$f0}. This is a non-standard calling sequence, 13762so any function with a floating-point argument or return value called by code 13763compiled with @option{-mno-fp-regs} must also be compiled with that 13764option. 13765 13766A typical use of this option is building a kernel that does not use, 13767and hence need not save and restore, any floating-point registers. 13768 13769@item -mieee 13770@opindex mieee 13771The Alpha architecture implements floating-point hardware optimized for 13772maximum performance. It is mostly compliant with the IEEE floating-point 13773standard. However, for full compliance, software assistance is 13774required. This option generates code fully IEEE-compliant code 13775@emph{except} that the @var{inexact-flag} is not maintained (see below). 13776If this option is turned on, the preprocessor macro @code{_IEEE_FP} is 13777defined during compilation. The resulting code is less efficient but is 13778able to correctly support denormalized numbers and exceptional IEEE 13779values such as not-a-number and plus/minus infinity. Other Alpha 13780compilers call this option @option{-ieee_with_no_inexact}. 13781 13782@item -mieee-with-inexact 13783@opindex mieee-with-inexact 13784This is like @option{-mieee} except the generated code also maintains 13785the IEEE @var{inexact-flag}. Turning on this option causes the 13786generated code to implement fully-compliant IEEE math. In addition to 13787@code{_IEEE_FP}, @code{_IEEE_FP_EXACT} is defined as a preprocessor 13788macro. On some Alpha implementations the resulting code may execute 13789significantly slower than the code generated by default. Since there is 13790very little code that depends on the @var{inexact-flag}, you should 13791normally not specify this option. Other Alpha compilers call this 13792option @option{-ieee_with_inexact}. 13793 13794@item -mfp-trap-mode=@var{trap-mode} 13795@opindex mfp-trap-mode 13796This option controls what floating-point related traps are enabled. 13797Other Alpha compilers call this option @option{-fptm @var{trap-mode}}. 13798The trap mode can be set to one of four values: 13799 13800@table @samp 13801@item n 13802This is the default (normal) setting. The only traps that are enabled 13803are the ones that cannot be disabled in software (e.g., division by zero 13804trap). 13805 13806@item u 13807In addition to the traps enabled by @samp{n}, underflow traps are enabled 13808as well. 13809 13810@item su 13811Like @samp{u}, but the instructions are marked to be safe for software 13812completion (see Alpha architecture manual for details). 13813 13814@item sui 13815Like @samp{su}, but inexact traps are enabled as well. 13816@end table 13817 13818@item -mfp-rounding-mode=@var{rounding-mode} 13819@opindex mfp-rounding-mode 13820Selects the IEEE rounding mode. Other Alpha compilers call this option 13821@option{-fprm @var{rounding-mode}}. The @var{rounding-mode} can be one 13822of: 13823 13824@table @samp 13825@item n 13826Normal IEEE rounding mode. Floating-point numbers are rounded towards 13827the nearest machine number or towards the even machine number in case 13828of a tie. 13829 13830@item m 13831Round towards minus infinity. 13832 13833@item c 13834Chopped rounding mode. Floating-point numbers are rounded towards zero. 13835 13836@item d 13837Dynamic rounding mode. A field in the floating-point control register 13838(@var{fpcr}, see Alpha architecture reference manual) controls the 13839rounding mode in effect. The C library initializes this register for 13840rounding towards plus infinity. Thus, unless your program modifies the 13841@var{fpcr}, @samp{d} corresponds to round towards plus infinity. 13842@end table 13843 13844@item -mtrap-precision=@var{trap-precision} 13845@opindex mtrap-precision 13846In the Alpha architecture, floating-point traps are imprecise. This 13847means without software assistance it is impossible to recover from a 13848floating trap and program execution normally needs to be terminated. 13849GCC can generate code that can assist operating system trap handlers 13850in determining the exact location that caused a floating-point trap. 13851Depending on the requirements of an application, different levels of 13852precisions can be selected: 13853 13854@table @samp 13855@item p 13856Program precision. This option is the default and means a trap handler 13857can only identify which program caused a floating-point exception. 13858 13859@item f 13860Function precision. The trap handler can determine the function that 13861caused a floating-point exception. 13862 13863@item i 13864Instruction precision. The trap handler can determine the exact 13865instruction that caused a floating-point exception. 13866@end table 13867 13868Other Alpha compilers provide the equivalent options called 13869@option{-scope_safe} and @option{-resumption_safe}. 13870 13871@item -mieee-conformant 13872@opindex mieee-conformant 13873This option marks the generated code as IEEE conformant. You must not 13874use this option unless you also specify @option{-mtrap-precision=i} and either 13875@option{-mfp-trap-mode=su} or @option{-mfp-trap-mode=sui}. Its only effect 13876is to emit the line @samp{.eflag 48} in the function prologue of the 13877generated assembly file. 13878 13879@item -mbuild-constants 13880@opindex mbuild-constants 13881Normally GCC examines a 32- or 64-bit integer constant to 13882see if it can construct it from smaller constants in two or three 13883instructions. If it cannot, it outputs the constant as a literal and 13884generates code to load it from the data segment at run time. 13885 13886Use this option to require GCC to construct @emph{all} integer constants 13887using code, even if it takes more instructions (the maximum is six). 13888 13889You typically use this option to build a shared library dynamic 13890loader. Itself a shared library, it must relocate itself in memory 13891before it can find the variables and constants in its own data segment. 13892 13893@item -mbwx 13894@itemx -mno-bwx 13895@itemx -mcix 13896@itemx -mno-cix 13897@itemx -mfix 13898@itemx -mno-fix 13899@itemx -mmax 13900@itemx -mno-max 13901@opindex mbwx 13902@opindex mno-bwx 13903@opindex mcix 13904@opindex mno-cix 13905@opindex mfix 13906@opindex mno-fix 13907@opindex mmax 13908@opindex mno-max 13909Indicate whether GCC should generate code to use the optional BWX, 13910CIX, FIX and MAX instruction sets. The default is to use the instruction 13911sets supported by the CPU type specified via @option{-mcpu=} option or that 13912of the CPU on which GCC was built if none is specified. 13913 13914@item -mfloat-vax 13915@itemx -mfloat-ieee 13916@opindex mfloat-vax 13917@opindex mfloat-ieee 13918Generate code that uses (does not use) VAX F and G floating-point 13919arithmetic instead of IEEE single and double precision. 13920 13921@item -mexplicit-relocs 13922@itemx -mno-explicit-relocs 13923@opindex mexplicit-relocs 13924@opindex mno-explicit-relocs 13925Older Alpha assemblers provided no way to generate symbol relocations 13926except via assembler macros. Use of these macros does not allow 13927optimal instruction scheduling. GNU binutils as of version 2.12 13928supports a new syntax that allows the compiler to explicitly mark 13929which relocations should apply to which instructions. This option 13930is mostly useful for debugging, as GCC detects the capabilities of 13931the assembler when it is built and sets the default accordingly. 13932 13933@item -msmall-data 13934@itemx -mlarge-data 13935@opindex msmall-data 13936@opindex mlarge-data 13937When @option{-mexplicit-relocs} is in effect, static data is 13938accessed via @dfn{gp-relative} relocations. When @option{-msmall-data} 13939is used, objects 8 bytes long or smaller are placed in a @dfn{small data area} 13940(the @code{.sdata} and @code{.sbss} sections) and are accessed via 1394116-bit relocations off of the @code{$gp} register. This limits the 13942size of the small data area to 64KB, but allows the variables to be 13943directly accessed via a single instruction. 13944 13945The default is @option{-mlarge-data}. With this option the data area 13946is limited to just below 2GB@. Programs that require more than 2GB of 13947data must use @code{malloc} or @code{mmap} to allocate the data in the 13948heap instead of in the program's data segment. 13949 13950When generating code for shared libraries, @option{-fpic} implies 13951@option{-msmall-data} and @option{-fPIC} implies @option{-mlarge-data}. 13952 13953@item -msmall-text 13954@itemx -mlarge-text 13955@opindex msmall-text 13956@opindex mlarge-text 13957When @option{-msmall-text} is used, the compiler assumes that the 13958code of the entire program (or shared library) fits in 4MB, and is 13959thus reachable with a branch instruction. When @option{-msmall-data} 13960is used, the compiler can assume that all local symbols share the 13961same @code{$gp} value, and thus reduce the number of instructions 13962required for a function call from 4 to 1. 13963 13964The default is @option{-mlarge-text}. 13965 13966@item -mcpu=@var{cpu_type} 13967@opindex mcpu 13968Set the instruction set and instruction scheduling parameters for 13969machine type @var{cpu_type}. You can specify either the @samp{EV} 13970style name or the corresponding chip number. GCC supports scheduling 13971parameters for the EV4, EV5 and EV6 family of processors and 13972chooses the default values for the instruction set from the processor 13973you specify. If you do not specify a processor type, GCC defaults 13974to the processor on which the compiler was built. 13975 13976Supported values for @var{cpu_type} are 13977 13978@table @samp 13979@item ev4 13980@itemx ev45 13981@itemx 21064 13982Schedules as an EV4 and has no instruction set extensions. 13983 13984@item ev5 13985@itemx 21164 13986Schedules as an EV5 and has no instruction set extensions. 13987 13988@item ev56 13989@itemx 21164a 13990Schedules as an EV5 and supports the BWX extension. 13991 13992@item pca56 13993@itemx 21164pc 13994@itemx 21164PC 13995Schedules as an EV5 and supports the BWX and MAX extensions. 13996 13997@item ev6 13998@itemx 21264 13999Schedules as an EV6 and supports the BWX, FIX, and MAX extensions. 14000 14001@item ev67 14002@itemx 21264a 14003Schedules as an EV6 and supports the BWX, CIX, FIX, and MAX extensions. 14004@end table 14005 14006Native toolchains also support the value @samp{native}, 14007which selects the best architecture option for the host processor. 14008@option{-mcpu=native} has no effect if GCC does not recognize 14009the processor. 14010 14011@item -mtune=@var{cpu_type} 14012@opindex mtune 14013Set only the instruction scheduling parameters for machine type 14014@var{cpu_type}. The instruction set is not changed. 14015 14016Native toolchains also support the value @samp{native}, 14017which selects the best architecture option for the host processor. 14018@option{-mtune=native} has no effect if GCC does not recognize 14019the processor. 14020 14021@item -mmemory-latency=@var{time} 14022@opindex mmemory-latency 14023Sets the latency the scheduler should assume for typical memory 14024references as seen by the application. This number is highly 14025dependent on the memory access patterns used by the application 14026and the size of the external cache on the machine. 14027 14028Valid options for @var{time} are 14029 14030@table @samp 14031@item @var{number} 14032A decimal number representing clock cycles. 14033 14034@item L1 14035@itemx L2 14036@itemx L3 14037@itemx main 14038The compiler contains estimates of the number of clock cycles for 14039``typical'' EV4 & EV5 hardware for the Level 1, 2 & 3 caches 14040(also called Dcache, Scache, and Bcache), as well as to main memory. 14041Note that L3 is only valid for EV5. 14042 14043@end table 14044@end table 14045 14046@node FR30 Options 14047@subsection FR30 Options 14048@cindex FR30 Options 14049 14050These options are defined specifically for the FR30 port. 14051 14052@table @gcctabopt 14053 14054@item -msmall-model 14055@opindex msmall-model 14056Use the small address space model. This can produce smaller code, but 14057it does assume that all symbolic values and addresses fit into a 1405820-bit range. 14059 14060@item -mno-lsim 14061@opindex mno-lsim 14062Assume that runtime support has been provided and so there is no need 14063to include the simulator library (@file{libsim.a}) on the linker 14064command line. 14065 14066@end table 14067 14068@node FRV Options 14069@subsection FRV Options 14070@cindex FRV Options 14071 14072@table @gcctabopt 14073@item -mgpr-32 14074@opindex mgpr-32 14075 14076Only use the first 32 general-purpose registers. 14077 14078@item -mgpr-64 14079@opindex mgpr-64 14080 14081Use all 64 general-purpose registers. 14082 14083@item -mfpr-32 14084@opindex mfpr-32 14085 14086Use only the first 32 floating-point registers. 14087 14088@item -mfpr-64 14089@opindex mfpr-64 14090 14091Use all 64 floating-point registers. 14092 14093@item -mhard-float 14094@opindex mhard-float 14095 14096Use hardware instructions for floating-point operations. 14097 14098@item -msoft-float 14099@opindex msoft-float 14100 14101Use library routines for floating-point operations. 14102 14103@item -malloc-cc 14104@opindex malloc-cc 14105 14106Dynamically allocate condition code registers. 14107 14108@item -mfixed-cc 14109@opindex mfixed-cc 14110 14111Do not try to dynamically allocate condition code registers, only 14112use @code{icc0} and @code{fcc0}. 14113 14114@item -mdword 14115@opindex mdword 14116 14117Change ABI to use double word insns. 14118 14119@item -mno-dword 14120@opindex mno-dword 14121 14122Do not use double word instructions. 14123 14124@item -mdouble 14125@opindex mdouble 14126 14127Use floating-point double instructions. 14128 14129@item -mno-double 14130@opindex mno-double 14131 14132Do not use floating-point double instructions. 14133 14134@item -mmedia 14135@opindex mmedia 14136 14137Use media instructions. 14138 14139@item -mno-media 14140@opindex mno-media 14141 14142Do not use media instructions. 14143 14144@item -mmuladd 14145@opindex mmuladd 14146 14147Use multiply and add/subtract instructions. 14148 14149@item -mno-muladd 14150@opindex mno-muladd 14151 14152Do not use multiply and add/subtract instructions. 14153 14154@item -mfdpic 14155@opindex mfdpic 14156 14157Select the FDPIC ABI, which uses function descriptors to represent 14158pointers to functions. Without any PIC/PIE-related options, it 14159implies @option{-fPIE}. With @option{-fpic} or @option{-fpie}, it 14160assumes GOT entries and small data are within a 12-bit range from the 14161GOT base address; with @option{-fPIC} or @option{-fPIE}, GOT offsets 14162are computed with 32 bits. 14163With a @samp{bfin-elf} target, this option implies @option{-msim}. 14164 14165@item -minline-plt 14166@opindex minline-plt 14167 14168Enable inlining of PLT entries in function calls to functions that are 14169not known to bind locally. It has no effect without @option{-mfdpic}. 14170It's enabled by default if optimizing for speed and compiling for 14171shared libraries (i.e., @option{-fPIC} or @option{-fpic}), or when an 14172optimization option such as @option{-O3} or above is present in the 14173command line. 14174 14175@item -mTLS 14176@opindex mTLS 14177 14178Assume a large TLS segment when generating thread-local code. 14179 14180@item -mtls 14181@opindex mtls 14182 14183Do not assume a large TLS segment when generating thread-local code. 14184 14185@item -mgprel-ro 14186@opindex mgprel-ro 14187 14188Enable the use of @code{GPREL} relocations in the FDPIC ABI for data 14189that is known to be in read-only sections. It's enabled by default, 14190except for @option{-fpic} or @option{-fpie}: even though it may help 14191make the global offset table smaller, it trades 1 instruction for 4. 14192With @option{-fPIC} or @option{-fPIE}, it trades 3 instructions for 4, 14193one of which may be shared by multiple symbols, and it avoids the need 14194for a GOT entry for the referenced symbol, so it's more likely to be a 14195win. If it is not, @option{-mno-gprel-ro} can be used to disable it. 14196 14197@item -multilib-library-pic 14198@opindex multilib-library-pic 14199 14200Link with the (library, not FD) pic libraries. It's implied by 14201@option{-mlibrary-pic}, as well as by @option{-fPIC} and 14202@option{-fpic} without @option{-mfdpic}. You should never have to use 14203it explicitly. 14204 14205@item -mlinked-fp 14206@opindex mlinked-fp 14207 14208Follow the EABI requirement of always creating a frame pointer whenever 14209a stack frame is allocated. This option is enabled by default and can 14210be disabled with @option{-mno-linked-fp}. 14211 14212@item -mlong-calls 14213@opindex mlong-calls 14214 14215Use indirect addressing to call functions outside the current 14216compilation unit. This allows the functions to be placed anywhere 14217within the 32-bit address space. 14218 14219@item -malign-labels 14220@opindex malign-labels 14221 14222Try to align labels to an 8-byte boundary by inserting NOPs into the 14223previous packet. This option only has an effect when VLIW packing 14224is enabled. It doesn't create new packets; it merely adds NOPs to 14225existing ones. 14226 14227@item -mlibrary-pic 14228@opindex mlibrary-pic 14229 14230Generate position-independent EABI code. 14231 14232@item -macc-4 14233@opindex macc-4 14234 14235Use only the first four media accumulator registers. 14236 14237@item -macc-8 14238@opindex macc-8 14239 14240Use all eight media accumulator registers. 14241 14242@item -mpack 14243@opindex mpack 14244 14245Pack VLIW instructions. 14246 14247@item -mno-pack 14248@opindex mno-pack 14249 14250Do not pack VLIW instructions. 14251 14252@item -mno-eflags 14253@opindex mno-eflags 14254 14255Do not mark ABI switches in e_flags. 14256 14257@item -mcond-move 14258@opindex mcond-move 14259 14260Enable the use of conditional-move instructions (default). 14261 14262This switch is mainly for debugging the compiler and will likely be removed 14263in a future version. 14264 14265@item -mno-cond-move 14266@opindex mno-cond-move 14267 14268Disable the use of conditional-move instructions. 14269 14270This switch is mainly for debugging the compiler and will likely be removed 14271in a future version. 14272 14273@item -mscc 14274@opindex mscc 14275 14276Enable the use of conditional set instructions (default). 14277 14278This switch is mainly for debugging the compiler and will likely be removed 14279in a future version. 14280 14281@item -mno-scc 14282@opindex mno-scc 14283 14284Disable the use of conditional set instructions. 14285 14286This switch is mainly for debugging the compiler and will likely be removed 14287in a future version. 14288 14289@item -mcond-exec 14290@opindex mcond-exec 14291 14292Enable the use of conditional execution (default). 14293 14294This switch is mainly for debugging the compiler and will likely be removed 14295in a future version. 14296 14297@item -mno-cond-exec 14298@opindex mno-cond-exec 14299 14300Disable the use of conditional execution. 14301 14302This switch is mainly for debugging the compiler and will likely be removed 14303in a future version. 14304 14305@item -mvliw-branch 14306@opindex mvliw-branch 14307 14308Run a pass to pack branches into VLIW instructions (default). 14309 14310This switch is mainly for debugging the compiler and will likely be removed 14311in a future version. 14312 14313@item -mno-vliw-branch 14314@opindex mno-vliw-branch 14315 14316Do not run a pass to pack branches into VLIW instructions. 14317 14318This switch is mainly for debugging the compiler and will likely be removed 14319in a future version. 14320 14321@item -mmulti-cond-exec 14322@opindex mmulti-cond-exec 14323 14324Enable optimization of @code{&&} and @code{||} in conditional execution 14325(default). 14326 14327This switch is mainly for debugging the compiler and will likely be removed 14328in a future version. 14329 14330@item -mno-multi-cond-exec 14331@opindex mno-multi-cond-exec 14332 14333Disable optimization of @code{&&} and @code{||} in conditional execution. 14334 14335This switch is mainly for debugging the compiler and will likely be removed 14336in a future version. 14337 14338@item -mnested-cond-exec 14339@opindex mnested-cond-exec 14340 14341Enable nested conditional execution optimizations (default). 14342 14343This switch is mainly for debugging the compiler and will likely be removed 14344in a future version. 14345 14346@item -mno-nested-cond-exec 14347@opindex mno-nested-cond-exec 14348 14349Disable nested conditional execution optimizations. 14350 14351This switch is mainly for debugging the compiler and will likely be removed 14352in a future version. 14353 14354@item -moptimize-membar 14355@opindex moptimize-membar 14356 14357This switch removes redundant @code{membar} instructions from the 14358compiler-generated code. It is enabled by default. 14359 14360@item -mno-optimize-membar 14361@opindex mno-optimize-membar 14362 14363This switch disables the automatic removal of redundant @code{membar} 14364instructions from the generated code. 14365 14366@item -mtomcat-stats 14367@opindex mtomcat-stats 14368 14369Cause gas to print out tomcat statistics. 14370 14371@item -mcpu=@var{cpu} 14372@opindex mcpu 14373 14374Select the processor type for which to generate code. Possible values are 14375@samp{frv}, @samp{fr550}, @samp{tomcat}, @samp{fr500}, @samp{fr450}, 14376@samp{fr405}, @samp{fr400}, @samp{fr300} and @samp{simple}. 14377 14378@end table 14379 14380@node GNU/Linux Options 14381@subsection GNU/Linux Options 14382 14383These @samp{-m} options are defined for GNU/Linux targets: 14384 14385@table @gcctabopt 14386@item -mglibc 14387@opindex mglibc 14388Use the GNU C library. This is the default except 14389on @samp{*-*-linux-*uclibc*} and @samp{*-*-linux-*android*} targets. 14390 14391@item -muclibc 14392@opindex muclibc 14393Use uClibc C library. This is the default on 14394@samp{*-*-linux-*uclibc*} targets. 14395 14396@item -mbionic 14397@opindex mbionic 14398Use Bionic C library. This is the default on 14399@samp{*-*-linux-*android*} targets. 14400 14401@item -mandroid 14402@opindex mandroid 14403Compile code compatible with Android platform. This is the default on 14404@samp{*-*-linux-*android*} targets. 14405 14406When compiling, this option enables @option{-mbionic}, @option{-fPIC}, 14407@option{-fno-exceptions} and @option{-fno-rtti} by default. When linking, 14408this option makes the GCC driver pass Android-specific options to the linker. 14409Finally, this option causes the preprocessor macro @code{__ANDROID__} 14410to be defined. 14411 14412@item -tno-android-cc 14413@opindex tno-android-cc 14414Disable compilation effects of @option{-mandroid}, i.e., do not enable 14415@option{-mbionic}, @option{-fPIC}, @option{-fno-exceptions} and 14416@option{-fno-rtti} by default. 14417 14418@item -tno-android-ld 14419@opindex tno-android-ld 14420Disable linking effects of @option{-mandroid}, i.e., pass standard Linux 14421linking options to the linker. 14422 14423@end table 14424 14425@node H8/300 Options 14426@subsection H8/300 Options 14427 14428These @samp{-m} options are defined for the H8/300 implementations: 14429 14430@table @gcctabopt 14431@item -mrelax 14432@opindex mrelax 14433Shorten some address references at link time, when possible; uses the 14434linker option @option{-relax}. @xref{H8/300,, @code{ld} and the H8/300, 14435ld, Using ld}, for a fuller description. 14436 14437@item -mh 14438@opindex mh 14439Generate code for the H8/300H@. 14440 14441@item -ms 14442@opindex ms 14443Generate code for the H8S@. 14444 14445@item -mn 14446@opindex mn 14447Generate code for the H8S and H8/300H in the normal mode. This switch 14448must be used either with @option{-mh} or @option{-ms}. 14449 14450@item -ms2600 14451@opindex ms2600 14452Generate code for the H8S/2600. This switch must be used with @option{-ms}. 14453 14454@item -mexr 14455@opindex mexr 14456Extended registers are stored on stack before execution of function 14457with monitor attribute. Default option is @option{-mexr}. 14458This option is valid only for H8S targets. 14459 14460@item -mno-exr 14461@opindex mno-exr 14462Extended registers are not stored on stack before execution of function 14463with monitor attribute. Default option is @option{-mno-exr}. 14464This option is valid only for H8S targets. 14465 14466@item -mint32 14467@opindex mint32 14468Make @code{int} data 32 bits by default. 14469 14470@item -malign-300 14471@opindex malign-300 14472On the H8/300H and H8S, use the same alignment rules as for the H8/300. 14473The default for the H8/300H and H8S is to align longs and floats on 144744-byte boundaries. 14475@option{-malign-300} causes them to be aligned on 2-byte boundaries. 14476This option has no effect on the H8/300. 14477@end table 14478 14479@node HPPA Options 14480@subsection HPPA Options 14481@cindex HPPA Options 14482 14483These @samp{-m} options are defined for the HPPA family of computers: 14484 14485@table @gcctabopt 14486@item -march=@var{architecture-type} 14487@opindex march 14488Generate code for the specified architecture. The choices for 14489@var{architecture-type} are @samp{1.0} for PA 1.0, @samp{1.1} for PA 144901.1, and @samp{2.0} for PA 2.0 processors. Refer to 14491@file{/usr/lib/sched.models} on an HP-UX system to determine the proper 14492architecture option for your machine. Code compiled for lower numbered 14493architectures runs on higher numbered architectures, but not the 14494other way around. 14495 14496@item -mpa-risc-1-0 14497@itemx -mpa-risc-1-1 14498@itemx -mpa-risc-2-0 14499@opindex mpa-risc-1-0 14500@opindex mpa-risc-1-1 14501@opindex mpa-risc-2-0 14502Synonyms for @option{-march=1.0}, @option{-march=1.1}, and @option{-march=2.0} respectively. 14503 14504@item -mjump-in-delay 14505@opindex mjump-in-delay 14506Fill delay slots of function calls with unconditional jump instructions 14507by modifying the return pointer for the function call to be the target 14508of the conditional jump. 14509 14510@item -mdisable-fpregs 14511@opindex mdisable-fpregs 14512Prevent floating-point registers from being used in any manner. This is 14513necessary for compiling kernels that perform lazy context switching of 14514floating-point registers. If you use this option and attempt to perform 14515floating-point operations, the compiler aborts. 14516 14517@item -mdisable-indexing 14518@opindex mdisable-indexing 14519Prevent the compiler from using indexing address modes. This avoids some 14520rather obscure problems when compiling MIG generated code under MACH@. 14521 14522@item -mno-space-regs 14523@opindex mno-space-regs 14524Generate code that assumes the target has no space registers. This allows 14525GCC to generate faster indirect calls and use unscaled index address modes. 14526 14527Such code is suitable for level 0 PA systems and kernels. 14528 14529@item -mfast-indirect-calls 14530@opindex mfast-indirect-calls 14531Generate code that assumes calls never cross space boundaries. This 14532allows GCC to emit code that performs faster indirect calls. 14533 14534This option does not work in the presence of shared libraries or nested 14535functions. 14536 14537@item -mfixed-range=@var{register-range} 14538@opindex mfixed-range 14539Generate code treating the given register range as fixed registers. 14540A fixed register is one that the register allocator cannot use. This is 14541useful when compiling kernel code. A register range is specified as 14542two registers separated by a dash. Multiple register ranges can be 14543specified separated by a comma. 14544 14545@item -mlong-load-store 14546@opindex mlong-load-store 14547Generate 3-instruction load and store sequences as sometimes required by 14548the HP-UX 10 linker. This is equivalent to the @samp{+k} option to 14549the HP compilers. 14550 14551@item -mportable-runtime 14552@opindex mportable-runtime 14553Use the portable calling conventions proposed by HP for ELF systems. 14554 14555@item -mgas 14556@opindex mgas 14557Enable the use of assembler directives only GAS understands. 14558 14559@item -mschedule=@var{cpu-type} 14560@opindex mschedule 14561Schedule code according to the constraints for the machine type 14562@var{cpu-type}. The choices for @var{cpu-type} are @samp{700} 14563@samp{7100}, @samp{7100LC}, @samp{7200}, @samp{7300} and @samp{8000}. Refer 14564to @file{/usr/lib/sched.models} on an HP-UX system to determine the 14565proper scheduling option for your machine. The default scheduling is 14566@samp{8000}. 14567 14568@item -mlinker-opt 14569@opindex mlinker-opt 14570Enable the optimization pass in the HP-UX linker. Note this makes symbolic 14571debugging impossible. It also triggers a bug in the HP-UX 8 and HP-UX 9 14572linkers in which they give bogus error messages when linking some programs. 14573 14574@item -msoft-float 14575@opindex msoft-float 14576Generate output containing library calls for floating point. 14577@strong{Warning:} the requisite libraries are not available for all HPPA 14578targets. Normally the facilities of the machine's usual C compiler are 14579used, but this cannot be done directly in cross-compilation. You must make 14580your own arrangements to provide suitable library functions for 14581cross-compilation. 14582 14583@option{-msoft-float} changes the calling convention in the output file; 14584therefore, it is only useful if you compile @emph{all} of a program with 14585this option. In particular, you need to compile @file{libgcc.a}, the 14586library that comes with GCC, with @option{-msoft-float} in order for 14587this to work. 14588 14589@item -msio 14590@opindex msio 14591Generate the predefine, @code{_SIO}, for server IO@. The default is 14592@option{-mwsio}. This generates the predefines, @code{__hp9000s700}, 14593@code{__hp9000s700__} and @code{_WSIO}, for workstation IO@. These 14594options are available under HP-UX and HI-UX@. 14595 14596@item -mgnu-ld 14597@opindex mgnu-ld 14598Use options specific to GNU @command{ld}. 14599This passes @option{-shared} to @command{ld} when 14600building a shared library. It is the default when GCC is configured, 14601explicitly or implicitly, with the GNU linker. This option does not 14602affect which @command{ld} is called; it only changes what parameters 14603are passed to that @command{ld}. 14604The @command{ld} that is called is determined by the 14605@option{--with-ld} configure option, GCC's program search path, and 14606finally by the user's @env{PATH}. The linker used by GCC can be printed 14607using @samp{which `gcc -print-prog-name=ld`}. This option is only available 14608on the 64-bit HP-UX GCC, i.e.@: configured with @samp{hppa*64*-*-hpux*}. 14609 14610@item -mhp-ld 14611@opindex mhp-ld 14612Use options specific to HP @command{ld}. 14613This passes @option{-b} to @command{ld} when building 14614a shared library and passes @option{+Accept TypeMismatch} to @command{ld} on all 14615links. It is the default when GCC is configured, explicitly or 14616implicitly, with the HP linker. This option does not affect 14617which @command{ld} is called; it only changes what parameters are passed to that 14618@command{ld}. 14619The @command{ld} that is called is determined by the @option{--with-ld} 14620configure option, GCC's program search path, and finally by the user's 14621@env{PATH}. The linker used by GCC can be printed using @samp{which 14622`gcc -print-prog-name=ld`}. This option is only available on the 64-bit 14623HP-UX GCC, i.e.@: configured with @samp{hppa*64*-*-hpux*}. 14624 14625@item -mlong-calls 14626@opindex mno-long-calls 14627Generate code that uses long call sequences. This ensures that a call 14628is always able to reach linker generated stubs. The default is to generate 14629long calls only when the distance from the call site to the beginning 14630of the function or translation unit, as the case may be, exceeds a 14631predefined limit set by the branch type being used. The limits for 14632normal calls are 7,600,000 and 240,000 bytes, respectively for the 14633PA 2.0 and PA 1.X architectures. Sibcalls are always limited at 14634240,000 bytes. 14635 14636Distances are measured from the beginning of functions when using the 14637@option{-ffunction-sections} option, or when using the @option{-mgas} 14638and @option{-mno-portable-runtime} options together under HP-UX with 14639the SOM linker. 14640 14641It is normally not desirable to use this option as it degrades 14642performance. However, it may be useful in large applications, 14643particularly when partial linking is used to build the application. 14644 14645The types of long calls used depends on the capabilities of the 14646assembler and linker, and the type of code being generated. The 14647impact on systems that support long absolute calls, and long pic 14648symbol-difference or pc-relative calls should be relatively small. 14649However, an indirect call is used on 32-bit ELF systems in pic code 14650and it is quite long. 14651 14652@item -munix=@var{unix-std} 14653@opindex march 14654Generate compiler predefines and select a startfile for the specified 14655UNIX standard. The choices for @var{unix-std} are @samp{93}, @samp{95} 14656and @samp{98}. @samp{93} is supported on all HP-UX versions. @samp{95} 14657is available on HP-UX 10.10 and later. @samp{98} is available on HP-UX 1465811.11 and later. The default values are @samp{93} for HP-UX 10.00, 14659@samp{95} for HP-UX 10.10 though to 11.00, and @samp{98} for HP-UX 11.11 14660and later. 14661 14662@option{-munix=93} provides the same predefines as GCC 3.3 and 3.4. 14663@option{-munix=95} provides additional predefines for @code{XOPEN_UNIX} 14664and @code{_XOPEN_SOURCE_EXTENDED}, and the startfile @file{unix95.o}. 14665@option{-munix=98} provides additional predefines for @code{_XOPEN_UNIX}, 14666@code{_XOPEN_SOURCE_EXTENDED}, @code{_INCLUDE__STDC_A1_SOURCE} and 14667@code{_INCLUDE_XOPEN_SOURCE_500}, and the startfile @file{unix98.o}. 14668 14669It is @emph{important} to note that this option changes the interfaces 14670for various library routines. It also affects the operational behavior 14671of the C library. Thus, @emph{extreme} care is needed in using this 14672option. 14673 14674Library code that is intended to operate with more than one UNIX 14675standard must test, set and restore the variable @var{__xpg4_extended_mask} 14676as appropriate. Most GNU software doesn't provide this capability. 14677 14678@item -nolibdld 14679@opindex nolibdld 14680Suppress the generation of link options to search libdld.sl when the 14681@option{-static} option is specified on HP-UX 10 and later. 14682 14683@item -static 14684@opindex static 14685The HP-UX implementation of setlocale in libc has a dependency on 14686libdld.sl. There isn't an archive version of libdld.sl. Thus, 14687when the @option{-static} option is specified, special link options 14688are needed to resolve this dependency. 14689 14690On HP-UX 10 and later, the GCC driver adds the necessary options to 14691link with libdld.sl when the @option{-static} option is specified. 14692This causes the resulting binary to be dynamic. On the 64-bit port, 14693the linkers generate dynamic binaries by default in any case. The 14694@option{-nolibdld} option can be used to prevent the GCC driver from 14695adding these link options. 14696 14697@item -threads 14698@opindex threads 14699Add support for multithreading with the @dfn{dce thread} library 14700under HP-UX@. This option sets flags for both the preprocessor and 14701linker. 14702@end table 14703 14704@node i386 and x86-64 Options 14705@subsection Intel 386 and AMD x86-64 Options 14706@cindex i386 Options 14707@cindex x86-64 Options 14708@cindex Intel 386 Options 14709@cindex AMD x86-64 Options 14710 14711These @samp{-m} options are defined for the i386 and x86-64 family of 14712computers: 14713 14714@table @gcctabopt 14715 14716@item -march=@var{cpu-type} 14717@opindex march 14718Generate instructions for the machine type @var{cpu-type}. In contrast to 14719@option{-mtune=@var{cpu-type}}, which merely tunes the generated code 14720for the specified @var{cpu-type}, @option{-march=@var{cpu-type}} allows GCC 14721to generate code that may not run at all on processors other than the one 14722indicated. Specifying @option{-march=@var{cpu-type}} implies 14723@option{-mtune=@var{cpu-type}}. 14724 14725The choices for @var{cpu-type} are: 14726 14727@table @samp 14728@item native 14729This selects the CPU to generate code for at compilation time by determining 14730the processor type of the compiling machine. Using @option{-march=native} 14731enables all instruction subsets supported by the local machine (hence 14732the result might not run on different machines). Using @option{-mtune=native} 14733produces code optimized for the local machine under the constraints 14734of the selected instruction set. 14735 14736@item i386 14737Original Intel i386 CPU@. 14738 14739@item i486 14740Intel i486 CPU@. (No scheduling is implemented for this chip.) 14741 14742@item i586 14743@itemx pentium 14744Intel Pentium CPU with no MMX support. 14745 14746@item pentium-mmx 14747Intel Pentium MMX CPU, based on Pentium core with MMX instruction set support. 14748 14749@item pentiumpro 14750Intel Pentium Pro CPU@. 14751 14752@item i686 14753When used with @option{-march}, the Pentium Pro 14754instruction set is used, so the code runs on all i686 family chips. 14755When used with @option{-mtune}, it has the same meaning as @samp{generic}. 14756 14757@item pentium2 14758Intel Pentium II CPU, based on Pentium Pro core with MMX instruction set 14759support. 14760 14761@item pentium3 14762@itemx pentium3m 14763Intel Pentium III CPU, based on Pentium Pro core with MMX and SSE instruction 14764set support. 14765 14766@item pentium-m 14767Intel Pentium M; low-power version of Intel Pentium III CPU 14768with MMX, SSE and SSE2 instruction set support. Used by Centrino notebooks. 14769 14770@item pentium4 14771@itemx pentium4m 14772Intel Pentium 4 CPU with MMX, SSE and SSE2 instruction set support. 14773 14774@item prescott 14775Improved version of Intel Pentium 4 CPU with MMX, SSE, SSE2 and SSE3 instruction 14776set support. 14777 14778@item nocona 14779Improved version of Intel Pentium 4 CPU with 64-bit extensions, MMX, SSE, 14780SSE2 and SSE3 instruction set support. 14781 14782@item core2 14783Intel Core 2 CPU with 64-bit extensions, MMX, SSE, SSE2, SSE3 and SSSE3 14784instruction set support. 14785 14786@item nehalem 14787Intel Nehalem CPU with 64-bit extensions, MMX, SSE, SSE2, SSE3, SSSE3, 14788SSE4.1, SSE4.2 and POPCNT instruction set support. 14789 14790@item westmere 14791Intel Westmere CPU with 64-bit extensions, MMX, SSE, SSE2, SSE3, SSSE3, 14792SSE4.1, SSE4.2, POPCNT, AES and PCLMUL instruction set support. 14793 14794@item sandybridge 14795Intel Sandy Bridge CPU with 64-bit extensions, MMX, SSE, SSE2, SSE3, SSSE3, 14796SSE4.1, SSE4.2, POPCNT, AVX, AES and PCLMUL instruction set support. 14797 14798@item ivybridge 14799Intel Ivy Bridge CPU with 64-bit extensions, MMX, SSE, SSE2, SSE3, SSSE3, 14800SSE4.1, SSE4.2, POPCNT, AVX, AES, PCLMUL, FSGSBASE, RDRND and F16C 14801instruction set support. 14802 14803@item haswell 14804Intel Haswell CPU with 64-bit extensions, MOVBE, MMX, SSE, SSE2, SSE3, SSSE3, 14805SSE4.1, SSE4.2, POPCNT, AVX, AVX2, AES, PCLMUL, FSGSBASE, RDRND, FMA, 14806BMI, BMI2 and F16C instruction set support. 14807 14808@item broadwell 14809Intel Broadwell CPU with 64-bit extensions, MOVBE, MMX, SSE, SSE2, SSE3, SSSE3, 14810SSE4.1, SSE4.2, POPCNT, AVX, AVX2, AES, PCLMUL, FSGSBASE, RDRND, FMA, 14811BMI, BMI2, F16C, RDSEED, ADCX and PREFETCHW instruction set support. 14812 14813@item bonnell 14814Intel Bonnell CPU with 64-bit extensions, MOVBE, MMX, SSE, SSE2, SSE3 and SSSE3 14815instruction set support. 14816 14817@item silvermont 14818Intel Silvermont CPU with 64-bit extensions, MOVBE, MMX, SSE, SSE2, SSE3, SSSE3, 14819SSE4.1, SSE4.2, POPCNT, AES, PCLMUL and RDRND instruction set support. 14820 14821@item k6 14822AMD K6 CPU with MMX instruction set support. 14823 14824@item k6-2 14825@itemx k6-3 14826Improved versions of AMD K6 CPU with MMX and 3DNow!@: instruction set support. 14827 14828@item athlon 14829@itemx athlon-tbird 14830AMD Athlon CPU with MMX, 3dNOW!, enhanced 3DNow!@: and SSE prefetch instructions 14831support. 14832 14833@item athlon-4 14834@itemx athlon-xp 14835@itemx athlon-mp 14836Improved AMD Athlon CPU with MMX, 3DNow!, enhanced 3DNow!@: and full SSE 14837instruction set support. 14838 14839@item k8 14840@itemx opteron 14841@itemx athlon64 14842@itemx athlon-fx 14843Processors based on the AMD K8 core with x86-64 instruction set support, 14844including the AMD Opteron, Athlon 64, and Athlon 64 FX processors. 14845(This supersets MMX, SSE, SSE2, 3DNow!, enhanced 3DNow!@: and 64-bit 14846instruction set extensions.) 14847 14848@item k8-sse3 14849@itemx opteron-sse3 14850@itemx athlon64-sse3 14851Improved versions of AMD K8 cores with SSE3 instruction set support. 14852 14853@item amdfam10 14854@itemx barcelona 14855CPUs based on AMD Family 10h cores with x86-64 instruction set support. (This 14856supersets MMX, SSE, SSE2, SSE3, SSE4A, 3DNow!, enhanced 3DNow!, ABM and 64-bit 14857instruction set extensions.) 14858 14859@item bdver1 14860CPUs based on AMD Family 15h cores with x86-64 instruction set support. (This 14861supersets FMA4, AVX, XOP, LWP, AES, PCL_MUL, CX16, MMX, SSE, SSE2, SSE3, SSE4A, 14862SSSE3, SSE4.1, SSE4.2, ABM and 64-bit instruction set extensions.) 14863@item bdver2 14864AMD Family 15h core based CPUs with x86-64 instruction set support. (This 14865supersets BMI, TBM, F16C, FMA, FMA4, AVX, XOP, LWP, AES, PCL_MUL, CX16, MMX, 14866SSE, SSE2, SSE3, SSE4A, SSSE3, SSE4.1, SSE4.2, ABM and 64-bit instruction set 14867extensions.) 14868@item bdver3 14869AMD Family 15h core based CPUs with x86-64 instruction set support. (This 14870supersets BMI, TBM, F16C, FMA, FMA4, FSGSBASE, AVX, XOP, LWP, AES, 14871PCL_MUL, CX16, MMX, SSE, SSE2, SSE3, SSE4A, SSSE3, SSE4.1, SSE4.2, ABM and 1487264-bit instruction set extensions. 14873@item bdver4 14874AMD Family 15h core based CPUs with x86-64 instruction set support. (This 14875supersets BMI, BMI2, TBM, F16C, FMA, FMA4, FSGSBASE, AVX, AVX2, XOP, LWP, 14876AES, PCL_MUL, CX16, MOVBE, MMX, SSE, SSE2, SSE3, SSE4A, SSSE3, SSE4.1, 14877SSE4.2, ABM and 64-bit instruction set extensions. 14878 14879@item btver1 14880CPUs based on AMD Family 14h cores with x86-64 instruction set support. (This 14881supersets MMX, SSE, SSE2, SSE3, SSSE3, SSE4A, CX16, ABM and 64-bit 14882instruction set extensions.) 14883 14884@item btver2 14885CPUs based on AMD Family 16h cores with x86-64 instruction set support. This 14886includes MOVBE, F16C, BMI, AVX, PCL_MUL, AES, SSE4.2, SSE4.1, CX16, ABM, 14887SSE4A, SSSE3, SSE3, SSE2, SSE, MMX and 64-bit instruction set extensions. 14888 14889@item winchip-c6 14890IDT WinChip C6 CPU, dealt in same way as i486 with additional MMX instruction 14891set support. 14892 14893@item winchip2 14894IDT WinChip 2 CPU, dealt in same way as i486 with additional MMX and 3DNow!@: 14895instruction set support. 14896 14897@item c3 14898VIA C3 CPU with MMX and 3DNow!@: instruction set support. (No scheduling is 14899implemented for this chip.) 14900 14901@item c3-2 14902VIA C3-2 (Nehemiah/C5XL) CPU with MMX and SSE instruction set support. 14903(No scheduling is 14904implemented for this chip.) 14905 14906@item geode 14907AMD Geode embedded processor with MMX and 3DNow!@: instruction set support. 14908@end table 14909 14910@item -mtune=@var{cpu-type} 14911@opindex mtune 14912Tune to @var{cpu-type} everything applicable about the generated code, except 14913for the ABI and the set of available instructions. 14914While picking a specific @var{cpu-type} schedules things appropriately 14915for that particular chip, the compiler does not generate any code that 14916cannot run on the default machine type unless you use a 14917@option{-march=@var{cpu-type}} option. 14918For example, if GCC is configured for i686-pc-linux-gnu 14919then @option{-mtune=pentium4} generates code that is tuned for Pentium 4 14920but still runs on i686 machines. 14921 14922The choices for @var{cpu-type} are the same as for @option{-march}. 14923In addition, @option{-mtune} supports 2 extra choices for @var{cpu-type}: 14924 14925@table @samp 14926@item generic 14927Produce code optimized for the most common IA32/@/AMD64/@/EM64T processors. 14928If you know the CPU on which your code will run, then you should use 14929the corresponding @option{-mtune} or @option{-march} option instead of 14930@option{-mtune=generic}. But, if you do not know exactly what CPU users 14931of your application will have, then you should use this option. 14932 14933As new processors are deployed in the marketplace, the behavior of this 14934option will change. Therefore, if you upgrade to a newer version of 14935GCC, code generation controlled by this option will change to reflect 14936the processors 14937that are most common at the time that version of GCC is released. 14938 14939There is no @option{-march=generic} option because @option{-march} 14940indicates the instruction set the compiler can use, and there is no 14941generic instruction set applicable to all processors. In contrast, 14942@option{-mtune} indicates the processor (or, in this case, collection of 14943processors) for which the code is optimized. 14944 14945@item intel 14946Produce code optimized for the most current Intel processors, which are 14947Haswell and Silvermont for this version of GCC. If you know the CPU 14948on which your code will run, then you should use the corresponding 14949@option{-mtune} or @option{-march} option instead of @option{-mtune=intel}. 14950But, if you want your application performs better on both Haswell and 14951Silvermont, then you should use this option. 14952 14953As new Intel processors are deployed in the marketplace, the behavior of 14954this option will change. Therefore, if you upgrade to a newer version of 14955GCC, code generation controlled by this option will change to reflect 14956the most current Intel processors at the time that version of GCC is 14957released. 14958 14959There is no @option{-march=intel} option because @option{-march} indicates 14960the instruction set the compiler can use, and there is no common 14961instruction set applicable to all processors. In contrast, 14962@option{-mtune} indicates the processor (or, in this case, collection of 14963processors) for which the code is optimized. 14964@end table 14965 14966@item -mcpu=@var{cpu-type} 14967@opindex mcpu 14968A deprecated synonym for @option{-mtune}. 14969 14970@item -mfpmath=@var{unit} 14971@opindex mfpmath 14972Generate floating-point arithmetic for selected unit @var{unit}. The choices 14973for @var{unit} are: 14974 14975@table @samp 14976@item 387 14977Use the standard 387 floating-point coprocessor present on the majority of chips and 14978emulated otherwise. Code compiled with this option runs almost everywhere. 14979The temporary results are computed in 80-bit precision instead of the precision 14980specified by the type, resulting in slightly different results compared to most 14981of other chips. See @option{-ffloat-store} for more detailed description. 14982 14983This is the default choice for i386 compiler. 14984 14985@item sse 14986Use scalar floating-point instructions present in the SSE instruction set. 14987This instruction set is supported by Pentium III and newer chips, 14988and in the AMD line 14989by Athlon-4, Athlon XP and Athlon MP chips. The earlier version of the SSE 14990instruction set supports only single-precision arithmetic, thus the double and 14991extended-precision arithmetic are still done using 387. A later version, present 14992only in Pentium 4 and AMD x86-64 chips, supports double-precision 14993arithmetic too. 14994 14995For the i386 compiler, you must use @option{-march=@var{cpu-type}}, @option{-msse} 14996or @option{-msse2} switches to enable SSE extensions and make this option 14997effective. For the x86-64 compiler, these extensions are enabled by default. 14998 14999The resulting code should be considerably faster in the majority of cases and avoid 15000the numerical instability problems of 387 code, but may break some existing 15001code that expects temporaries to be 80 bits. 15002 15003This is the default choice for the x86-64 compiler. 15004 15005@item sse,387 15006@itemx sse+387 15007@itemx both 15008Attempt to utilize both instruction sets at once. This effectively doubles the 15009amount of available registers, and on chips with separate execution units for 15010387 and SSE the execution resources too. Use this option with care, as it is 15011still experimental, because the GCC register allocator does not model separate 15012functional units well, resulting in unstable performance. 15013@end table 15014 15015@item -masm=@var{dialect} 15016@opindex masm=@var{dialect} 15017Output assembly instructions using selected @var{dialect}. Supported 15018choices are @samp{intel} or @samp{att} (the default). Darwin does 15019not support @samp{intel}. 15020 15021@item -mieee-fp 15022@itemx -mno-ieee-fp 15023@opindex mieee-fp 15024@opindex mno-ieee-fp 15025Control whether or not the compiler uses IEEE floating-point 15026comparisons. These correctly handle the case where the result of a 15027comparison is unordered. 15028 15029@item -msoft-float 15030@opindex msoft-float 15031Generate output containing library calls for floating point. 15032 15033@strong{Warning:} the requisite libraries are not part of GCC@. 15034Normally the facilities of the machine's usual C compiler are used, but 15035this can't be done directly in cross-compilation. You must make your 15036own arrangements to provide suitable library functions for 15037cross-compilation. 15038 15039On machines where a function returns floating-point results in the 80387 15040register stack, some floating-point opcodes may be emitted even if 15041@option{-msoft-float} is used. 15042 15043@item -mno-fp-ret-in-387 15044@opindex mno-fp-ret-in-387 15045Do not use the FPU registers for return values of functions. 15046 15047The usual calling convention has functions return values of types 15048@code{float} and @code{double} in an FPU register, even if there 15049is no FPU@. The idea is that the operating system should emulate 15050an FPU@. 15051 15052The option @option{-mno-fp-ret-in-387} causes such values to be returned 15053in ordinary CPU registers instead. 15054 15055@item -mno-fancy-math-387 15056@opindex mno-fancy-math-387 15057Some 387 emulators do not support the @code{sin}, @code{cos} and 15058@code{sqrt} instructions for the 387. Specify this option to avoid 15059generating those instructions. This option is the default on FreeBSD, 15060OpenBSD and NetBSD@. This option is overridden when @option{-march} 15061indicates that the target CPU always has an FPU and so the 15062instruction does not need emulation. These 15063instructions are not generated unless you also use the 15064@option{-funsafe-math-optimizations} switch. 15065 15066@item -malign-double 15067@itemx -mno-align-double 15068@opindex malign-double 15069@opindex mno-align-double 15070Control whether GCC aligns @code{double}, @code{long double}, and 15071@code{long long} variables on a two-word boundary or a one-word 15072boundary. Aligning @code{double} variables on a two-word boundary 15073produces code that runs somewhat faster on a Pentium at the 15074expense of more memory. 15075 15076On x86-64, @option{-malign-double} is enabled by default. 15077 15078@strong{Warning:} if you use the @option{-malign-double} switch, 15079structures containing the above types are aligned differently than 15080the published application binary interface specifications for the 386 15081and are not binary compatible with structures in code compiled 15082without that switch. 15083 15084@item -m96bit-long-double 15085@itemx -m128bit-long-double 15086@opindex m96bit-long-double 15087@opindex m128bit-long-double 15088These switches control the size of @code{long double} type. The i386 15089application binary interface specifies the size to be 96 bits, 15090so @option{-m96bit-long-double} is the default in 32-bit mode. 15091 15092Modern architectures (Pentium and newer) prefer @code{long double} 15093to be aligned to an 8- or 16-byte boundary. In arrays or structures 15094conforming to the ABI, this is not possible. So specifying 15095@option{-m128bit-long-double} aligns @code{long double} 15096to a 16-byte boundary by padding the @code{long double} with an additional 1509732-bit zero. 15098 15099In the x86-64 compiler, @option{-m128bit-long-double} is the default choice as 15100its ABI specifies that @code{long double} is aligned on 16-byte boundary. 15101 15102Notice that neither of these options enable any extra precision over the x87 15103standard of 80 bits for a @code{long double}. 15104 15105@strong{Warning:} if you override the default value for your target ABI, this 15106changes the size of 15107structures and arrays containing @code{long double} variables, 15108as well as modifying the function calling convention for functions taking 15109@code{long double}. Hence they are not binary-compatible 15110with code compiled without that switch. 15111 15112@item -mlong-double-64 15113@itemx -mlong-double-80 15114@itemx -mlong-double-128 15115@opindex mlong-double-64 15116@opindex mlong-double-80 15117@opindex mlong-double-128 15118These switches control the size of @code{long double} type. A size 15119of 64 bits makes the @code{long double} type equivalent to the @code{double} 15120type. This is the default for 32-bit Bionic C library. A size 15121of 128 bits makes the @code{long double} type equivalent to the 15122@code{__float128} type. This is the default for 64-bit Bionic C library. 15123 15124@strong{Warning:} if you override the default value for your target ABI, this 15125changes the size of 15126structures and arrays containing @code{long double} variables, 15127as well as modifying the function calling convention for functions taking 15128@code{long double}. Hence they are not binary-compatible 15129with code compiled without that switch. 15130 15131@item -mlarge-data-threshold=@var{threshold} 15132@opindex mlarge-data-threshold 15133When @option{-mcmodel=medium} is specified, data objects larger than 15134@var{threshold} are placed in the large data section. This value must be the 15135same across all objects linked into the binary, and defaults to 65535. 15136 15137@item -mrtd 15138@opindex mrtd 15139Use a different function-calling convention, in which functions that 15140take a fixed number of arguments return with the @code{ret @var{num}} 15141instruction, which pops their arguments while returning. This saves one 15142instruction in the caller since there is no need to pop the arguments 15143there. 15144 15145You can specify that an individual function is called with this calling 15146sequence with the function attribute @samp{stdcall}. You can also 15147override the @option{-mrtd} option by using the function attribute 15148@samp{cdecl}. @xref{Function Attributes}. 15149 15150@strong{Warning:} this calling convention is incompatible with the one 15151normally used on Unix, so you cannot use it if you need to call 15152libraries compiled with the Unix compiler. 15153 15154Also, you must provide function prototypes for all functions that 15155take variable numbers of arguments (including @code{printf}); 15156otherwise incorrect code is generated for calls to those 15157functions. 15158 15159In addition, seriously incorrect code results if you call a 15160function with too many arguments. (Normally, extra arguments are 15161harmlessly ignored.) 15162 15163@item -mregparm=@var{num} 15164@opindex mregparm 15165Control how many registers are used to pass integer arguments. By 15166default, no registers are used to pass arguments, and at most 3 15167registers can be used. You can control this behavior for a specific 15168function by using the function attribute @samp{regparm}. 15169@xref{Function Attributes}. 15170 15171@strong{Warning:} if you use this switch, and 15172@var{num} is nonzero, then you must build all modules with the same 15173value, including any libraries. This includes the system libraries and 15174startup modules. 15175 15176@item -msseregparm 15177@opindex msseregparm 15178Use SSE register passing conventions for float and double arguments 15179and return values. You can control this behavior for a specific 15180function by using the function attribute @samp{sseregparm}. 15181@xref{Function Attributes}. 15182 15183@strong{Warning:} if you use this switch then you must build all 15184modules with the same value, including any libraries. This includes 15185the system libraries and startup modules. 15186 15187@item -mvect8-ret-in-mem 15188@opindex mvect8-ret-in-mem 15189Return 8-byte vectors in memory instead of MMX registers. This is the 15190default on Solaris@tie{}8 and 9 and VxWorks to match the ABI of the Sun 15191Studio compilers until version 12. Later compiler versions (starting 15192with Studio 12 Update@tie{}1) follow the ABI used by other x86 targets, which 15193is the default on Solaris@tie{}10 and later. @emph{Only} use this option if 15194you need to remain compatible with existing code produced by those 15195previous compiler versions or older versions of GCC@. 15196 15197@item -mpc32 15198@itemx -mpc64 15199@itemx -mpc80 15200@opindex mpc32 15201@opindex mpc64 15202@opindex mpc80 15203 15204Set 80387 floating-point precision to 32, 64 or 80 bits. When @option{-mpc32} 15205is specified, the significands of results of floating-point operations are 15206rounded to 24 bits (single precision); @option{-mpc64} rounds the 15207significands of results of floating-point operations to 53 bits (double 15208precision) and @option{-mpc80} rounds the significands of results of 15209floating-point operations to 64 bits (extended double precision), which is 15210the default. When this option is used, floating-point operations in higher 15211precisions are not available to the programmer without setting the FPU 15212control word explicitly. 15213 15214Setting the rounding of floating-point operations to less than the default 1521580 bits can speed some programs by 2% or more. Note that some mathematical 15216libraries assume that extended-precision (80-bit) floating-point operations 15217are enabled by default; routines in such libraries could suffer significant 15218loss of accuracy, typically through so-called ``catastrophic cancellation'', 15219when this option is used to set the precision to less than extended precision. 15220 15221@item -mstackrealign 15222@opindex mstackrealign 15223Realign the stack at entry. On the Intel x86, the @option{-mstackrealign} 15224option generates an alternate prologue and epilogue that realigns the 15225run-time stack if necessary. This supports mixing legacy codes that keep 152264-byte stack alignment with modern codes that keep 16-byte stack alignment for 15227SSE compatibility. See also the attribute @code{force_align_arg_pointer}, 15228applicable to individual functions. 15229 15230@item -mpreferred-stack-boundary=@var{num} 15231@opindex mpreferred-stack-boundary 15232Attempt to keep the stack boundary aligned to a 2 raised to @var{num} 15233byte boundary. If @option{-mpreferred-stack-boundary} is not specified, 15234the default is 4 (16 bytes or 128 bits). 15235 15236@strong{Warning:} When generating code for the x86-64 architecture with 15237SSE extensions disabled, @option{-mpreferred-stack-boundary=3} can be 15238used to keep the stack boundary aligned to 8 byte boundary. Since 15239x86-64 ABI require 16 byte stack alignment, this is ABI incompatible and 15240intended to be used in controlled environment where stack space is 15241important limitation. This option will lead to wrong code when functions 15242compiled with 16 byte stack alignment (such as functions from a standard 15243library) are called with misaligned stack. In this case, SSE 15244instructions may lead to misaligned memory access traps. In addition, 15245variable arguments will be handled incorrectly for 16 byte aligned 15246objects (including x87 long double and __int128), leading to wrong 15247results. You must build all modules with 15248@option{-mpreferred-stack-boundary=3}, including any libraries. This 15249includes the system libraries and startup modules. 15250 15251@item -mincoming-stack-boundary=@var{num} 15252@opindex mincoming-stack-boundary 15253Assume the incoming stack is aligned to a 2 raised to @var{num} byte 15254boundary. If @option{-mincoming-stack-boundary} is not specified, 15255the one specified by @option{-mpreferred-stack-boundary} is used. 15256 15257On Pentium and Pentium Pro, @code{double} and @code{long double} values 15258should be aligned to an 8-byte boundary (see @option{-malign-double}) or 15259suffer significant run time performance penalties. On Pentium III, the 15260Streaming SIMD Extension (SSE) data type @code{__m128} may not work 15261properly if it is not 16-byte aligned. 15262 15263To ensure proper alignment of this values on the stack, the stack boundary 15264must be as aligned as that required by any value stored on the stack. 15265Further, every function must be generated such that it keeps the stack 15266aligned. Thus calling a function compiled with a higher preferred 15267stack boundary from a function compiled with a lower preferred stack 15268boundary most likely misaligns the stack. It is recommended that 15269libraries that use callbacks always use the default setting. 15270 15271This extra alignment does consume extra stack space, and generally 15272increases code size. Code that is sensitive to stack space usage, such 15273as embedded systems and operating system kernels, may want to reduce the 15274preferred alignment to @option{-mpreferred-stack-boundary=2}. 15275 15276@item -mmmx 15277@itemx -mno-mmx 15278@itemx -msse 15279@itemx -mno-sse 15280@itemx -msse2 15281@itemx -mno-sse2 15282@itemx -msse3 15283@itemx -mno-sse3 15284@itemx -mssse3 15285@itemx -mno-ssse3 15286@itemx -msse4.1 15287@need 800 15288@itemx -mno-sse4.1 15289@itemx -msse4.2 15290@itemx -mno-sse4.2 15291@itemx -msse4 15292@itemx -mno-sse4 15293@itemx -mavx 15294@itemx -mno-avx 15295@itemx -mavx2 15296@itemx -mno-avx2 15297@itemx -mavx512f 15298@itemx -mno-avx512f 15299@need 800 15300@itemx -mavx512pf 15301@itemx -mno-avx512pf 15302@itemx -mavx512er 15303@itemx -mno-avx512er 15304@itemx -mavx512cd 15305@itemx -mno-avx512cd 15306@itemx -msha 15307@itemx -mno-sha 15308@itemx -maes 15309@itemx -mno-aes 15310@itemx -mpclmul 15311@itemx -mno-pclmul 15312@need 800 15313@itemx -mfsgsbase 15314@itemx -mno-fsgsbase 15315@itemx -mrdrnd 15316@itemx -mno-rdrnd 15317@itemx -mf16c 15318@itemx -mno-f16c 15319@itemx -mfma 15320@itemx -mno-fma 15321@itemx -mprefetchwt1 15322@itemx -mno-prefetchwt1 15323@itemx -msse4a 15324@itemx -mno-sse4a 15325@itemx -mfma4 15326@itemx -mno-fma4 15327@need 800 15328@itemx -mxop 15329@itemx -mno-xop 15330@itemx -mlwp 15331@itemx -mno-lwp 15332@itemx -m3dnow 15333@itemx -mno-3dnow 15334@itemx -mpopcnt 15335@itemx -mno-popcnt 15336@itemx -mabm 15337@itemx -mno-abm 15338@itemx -mbmi 15339@itemx -mbmi2 15340@itemx -mno-bmi 15341@itemx -mno-bmi2 15342@itemx -mlzcnt 15343@itemx -mno-lzcnt 15344@itemx -mfxsr 15345@itemx -mxsave 15346@itemx -mxsaveopt 15347@itemx -mrtm 15348@itemx -mtbm 15349@itemx -mno-tbm 15350@opindex mmmx 15351@opindex mno-mmx 15352@opindex msse 15353@opindex mno-sse 15354@opindex m3dnow 15355@opindex mno-3dnow 15356These switches enable or disable the use of instructions in the MMX, SSE, 15357SSE2, SSE3, SSSE3, SSE4.1, AVX, AVX2, AVX512F, AVX512PF, AVX512ER, AVX512CD, 15358SHA, AES, PCLMUL, FSGSBASE, RDRND, F16C, FMA, SSE4A, FMA4, XOP, LWP, ABM, 15359BMI, BMI2, FXSR, XSAVE, XSAVEOPT, LZCNT, RTM, or 3DNow!@: 15360extended instruction sets. 15361These extensions are also available as built-in functions: see 15362@ref{X86 Built-in Functions}, for details of the functions enabled and 15363disabled by these switches. 15364 15365To generate SSE/SSE2 instructions automatically from floating-point 15366code (as opposed to 387 instructions), see @option{-mfpmath=sse}. 15367 15368GCC depresses SSEx instructions when @option{-mavx} is used. Instead, it 15369generates new AVX instructions or AVX equivalence for all SSEx instructions 15370when needed. 15371 15372These options enable GCC to use these extended instructions in 15373generated code, even without @option{-mfpmath=sse}. Applications that 15374perform run-time CPU detection must compile separate files for each 15375supported architecture, using the appropriate flags. In particular, 15376the file containing the CPU detection code should be compiled without 15377these options. 15378 15379@item -mdump-tune-features 15380@opindex mdump-tune-features 15381This option instructs GCC to dump the names of the x86 performance 15382tuning features and default settings. The names can be used in 15383@option{-mtune-ctrl=@var{feature-list}}. 15384 15385@item -mtune-ctrl=@var{feature-list} 15386@opindex mtune-ctrl=@var{feature-list} 15387This option is used to do fine grain control of x86 code generation features. 15388@var{feature-list} is a comma separated list of @var{feature} names. See also 15389@option{-mdump-tune-features}. When specified, the @var{feature} will be turned 15390on if it is not preceded with @code{^}, otherwise, it will be turned off. 15391@option{-mtune-ctrl=@var{feature-list}} is intended to be used by GCC 15392developers. Using it may lead to code paths not covered by testing and can 15393potentially result in compiler ICEs or runtime errors. 15394 15395@item -mno-default 15396@opindex mno-default 15397This option instructs GCC to turn off all tunable features. See also 15398@option{-mtune-ctrl=@var{feature-list}} and @option{-mdump-tune-features}. 15399 15400@item -mcld 15401@opindex mcld 15402This option instructs GCC to emit a @code{cld} instruction in the prologue 15403of functions that use string instructions. String instructions depend on 15404the DF flag to select between autoincrement or autodecrement mode. While the 15405ABI specifies the DF flag to be cleared on function entry, some operating 15406systems violate this specification by not clearing the DF flag in their 15407exception dispatchers. The exception handler can be invoked with the DF flag 15408set, which leads to wrong direction mode when string instructions are used. 15409This option can be enabled by default on 32-bit x86 targets by configuring 15410GCC with the @option{--enable-cld} configure option. Generation of @code{cld} 15411instructions can be suppressed with the @option{-mno-cld} compiler option 15412in this case. 15413 15414@item -mvzeroupper 15415@opindex mvzeroupper 15416This option instructs GCC to emit a @code{vzeroupper} instruction 15417before a transfer of control flow out of the function to minimize 15418the AVX to SSE transition penalty as well as remove unnecessary @code{zeroupper} 15419intrinsics. 15420 15421@item -mprefer-avx128 15422@opindex mprefer-avx128 15423This option instructs GCC to use 128-bit AVX instructions instead of 15424256-bit AVX instructions in the auto-vectorizer. 15425 15426@item -mcx16 15427@opindex mcx16 15428This option enables GCC to generate @code{CMPXCHG16B} instructions. 15429@code{CMPXCHG16B} allows for atomic operations on 128-bit double quadword 15430(or oword) data types. 15431This is useful for high-resolution counters that can be updated 15432by multiple processors (or cores). This instruction is generated as part of 15433atomic built-in functions: see @ref{__sync Builtins} or 15434@ref{__atomic Builtins} for details. 15435 15436@item -msahf 15437@opindex msahf 15438This option enables generation of @code{SAHF} instructions in 64-bit code. 15439Early Intel Pentium 4 CPUs with Intel 64 support, 15440prior to the introduction of Pentium 4 G1 step in December 2005, 15441lacked the @code{LAHF} and @code{SAHF} instructions 15442which were supported by AMD64. 15443These are load and store instructions, respectively, for certain status flags. 15444In 64-bit mode, the @code{SAHF} instruction is used to optimize @code{fmod}, 15445@code{drem}, and @code{remainder} built-in functions; 15446see @ref{Other Builtins} for details. 15447 15448@item -mmovbe 15449@opindex mmovbe 15450This option enables use of the @code{movbe} instruction to implement 15451@code{__builtin_bswap32} and @code{__builtin_bswap64}. 15452 15453@item -mcrc32 15454@opindex mcrc32 15455This option enables built-in functions @code{__builtin_ia32_crc32qi}, 15456@code{__builtin_ia32_crc32hi}, @code{__builtin_ia32_crc32si} and 15457@code{__builtin_ia32_crc32di} to generate the @code{crc32} machine instruction. 15458 15459@item -mrecip 15460@opindex mrecip 15461This option enables use of @code{RCPSS} and @code{RSQRTSS} instructions 15462(and their vectorized variants @code{RCPPS} and @code{RSQRTPS}) 15463with an additional Newton-Raphson step 15464to increase precision instead of @code{DIVSS} and @code{SQRTSS} 15465(and their vectorized 15466variants) for single-precision floating-point arguments. These instructions 15467are generated only when @option{-funsafe-math-optimizations} is enabled 15468together with @option{-finite-math-only} and @option{-fno-trapping-math}. 15469Note that while the throughput of the sequence is higher than the throughput 15470of the non-reciprocal instruction, the precision of the sequence can be 15471decreased by up to 2 ulp (i.e. the inverse of 1.0 equals 0.99999994). 15472 15473Note that GCC implements @code{1.0f/sqrtf(@var{x})} in terms of @code{RSQRTSS} 15474(or @code{RSQRTPS}) already with @option{-ffast-math} (or the above option 15475combination), and doesn't need @option{-mrecip}. 15476 15477Also note that GCC emits the above sequence with additional Newton-Raphson step 15478for vectorized single-float division and vectorized @code{sqrtf(@var{x})} 15479already with @option{-ffast-math} (or the above option combination), and 15480doesn't need @option{-mrecip}. 15481 15482@item -mrecip=@var{opt} 15483@opindex mrecip=opt 15484This option controls which reciprocal estimate instructions 15485may be used. @var{opt} is a comma-separated list of options, which may 15486be preceded by a @samp{!} to invert the option: 15487 15488@table @samp 15489@item all 15490Enable all estimate instructions. 15491 15492@item default 15493Enable the default instructions, equivalent to @option{-mrecip}. 15494 15495@item none 15496Disable all estimate instructions, equivalent to @option{-mno-recip}. 15497 15498@item div 15499Enable the approximation for scalar division. 15500 15501@item vec-div 15502Enable the approximation for vectorized division. 15503 15504@item sqrt 15505Enable the approximation for scalar square root. 15506 15507@item vec-sqrt 15508Enable the approximation for vectorized square root. 15509@end table 15510 15511So, for example, @option{-mrecip=all,!sqrt} enables 15512all of the reciprocal approximations, except for square root. 15513 15514@item -mveclibabi=@var{type} 15515@opindex mveclibabi 15516Specifies the ABI type to use for vectorizing intrinsics using an 15517external library. Supported values for @var{type} are @samp{svml} 15518for the Intel short 15519vector math library and @samp{acml} for the AMD math core library. 15520To use this option, both @option{-ftree-vectorize} and 15521@option{-funsafe-math-optimizations} have to be enabled, and an SVML or ACML 15522ABI-compatible library must be specified at link time. 15523 15524GCC currently emits calls to @code{vmldExp2}, 15525@code{vmldLn2}, @code{vmldLog102}, @code{vmldLog102}, @code{vmldPow2}, 15526@code{vmldTanh2}, @code{vmldTan2}, @code{vmldAtan2}, @code{vmldAtanh2}, 15527@code{vmldCbrt2}, @code{vmldSinh2}, @code{vmldSin2}, @code{vmldAsinh2}, 15528@code{vmldAsin2}, @code{vmldCosh2}, @code{vmldCos2}, @code{vmldAcosh2}, 15529@code{vmldAcos2}, @code{vmlsExp4}, @code{vmlsLn4}, @code{vmlsLog104}, 15530@code{vmlsLog104}, @code{vmlsPow4}, @code{vmlsTanh4}, @code{vmlsTan4}, 15531@code{vmlsAtan4}, @code{vmlsAtanh4}, @code{vmlsCbrt4}, @code{vmlsSinh4}, 15532@code{vmlsSin4}, @code{vmlsAsinh4}, @code{vmlsAsin4}, @code{vmlsCosh4}, 15533@code{vmlsCos4}, @code{vmlsAcosh4} and @code{vmlsAcos4} for corresponding 15534function type when @option{-mveclibabi=svml} is used, and @code{__vrd2_sin}, 15535@code{__vrd2_cos}, @code{__vrd2_exp}, @code{__vrd2_log}, @code{__vrd2_log2}, 15536@code{__vrd2_log10}, @code{__vrs4_sinf}, @code{__vrs4_cosf}, 15537@code{__vrs4_expf}, @code{__vrs4_logf}, @code{__vrs4_log2f}, 15538@code{__vrs4_log10f} and @code{__vrs4_powf} for the corresponding function type 15539when @option{-mveclibabi=acml} is used. 15540 15541@item -mabi=@var{name} 15542@opindex mabi 15543Generate code for the specified calling convention. Permissible values 15544are @samp{sysv} for the ABI used on GNU/Linux and other systems, and 15545@samp{ms} for the Microsoft ABI. The default is to use the Microsoft 15546ABI when targeting Microsoft Windows and the SysV ABI on all other systems. 15547You can control this behavior for a specific function by 15548using the function attribute @samp{ms_abi}/@samp{sysv_abi}. 15549@xref{Function Attributes}. 15550 15551@item -mtls-dialect=@var{type} 15552@opindex mtls-dialect 15553Generate code to access thread-local storage using the @samp{gnu} or 15554@samp{gnu2} conventions. @samp{gnu} is the conservative default; 15555@samp{gnu2} is more efficient, but it may add compile- and run-time 15556requirements that cannot be satisfied on all systems. 15557 15558@item -mpush-args 15559@itemx -mno-push-args 15560@opindex mpush-args 15561@opindex mno-push-args 15562Use PUSH operations to store outgoing parameters. This method is shorter 15563and usually equally fast as method using SUB/MOV operations and is enabled 15564by default. In some cases disabling it may improve performance because of 15565improved scheduling and reduced dependencies. 15566 15567@item -maccumulate-outgoing-args 15568@opindex maccumulate-outgoing-args 15569If enabled, the maximum amount of space required for outgoing arguments is 15570computed in the function prologue. This is faster on most modern CPUs 15571because of reduced dependencies, improved scheduling and reduced stack usage 15572when the preferred stack boundary is not equal to 2. The drawback is a notable 15573increase in code size. This switch implies @option{-mno-push-args}. 15574 15575@item -mthreads 15576@opindex mthreads 15577Support thread-safe exception handling on MinGW. Programs that rely 15578on thread-safe exception handling must compile and link all code with the 15579@option{-mthreads} option. When compiling, @option{-mthreads} defines 15580@code{-D_MT}; when linking, it links in a special thread helper library 15581@option{-lmingwthrd} which cleans up per-thread exception-handling data. 15582 15583@item -mno-align-stringops 15584@opindex mno-align-stringops 15585Do not align the destination of inlined string operations. This switch reduces 15586code size and improves performance in case the destination is already aligned, 15587but GCC doesn't know about it. 15588 15589@item -minline-all-stringops 15590@opindex minline-all-stringops 15591By default GCC inlines string operations only when the destination is 15592known to be aligned to least a 4-byte boundary. 15593This enables more inlining and increases code 15594size, but may improve performance of code that depends on fast 15595@code{memcpy}, @code{strlen}, 15596and @code{memset} for short lengths. 15597 15598@item -minline-stringops-dynamically 15599@opindex minline-stringops-dynamically 15600For string operations of unknown size, use run-time checks with 15601inline code for small blocks and a library call for large blocks. 15602 15603@item -mstringop-strategy=@var{alg} 15604@opindex mstringop-strategy=@var{alg} 15605Override the internal decision heuristic for the particular algorithm to use 15606for inlining string operations. The allowed values for @var{alg} are: 15607 15608@table @samp 15609@item rep_byte 15610@itemx rep_4byte 15611@itemx rep_8byte 15612Expand using i386 @code{rep} prefix of the specified size. 15613 15614@item byte_loop 15615@itemx loop 15616@itemx unrolled_loop 15617Expand into an inline loop. 15618 15619@item libcall 15620Always use a library call. 15621@end table 15622 15623@item -mmemcpy-strategy=@var{strategy} 15624@opindex mmemcpy-strategy=@var{strategy} 15625Override the internal decision heuristic to decide if @code{__builtin_memcpy} 15626should be inlined and what inline algorithm to use when the expected size 15627of the copy operation is known. @var{strategy} 15628is a comma-separated list of @var{alg}:@var{max_size}:@var{dest_align} triplets. 15629@var{alg} is specified in @option{-mstringop-strategy}, @var{max_size} specifies 15630the max byte size with which inline algorithm @var{alg} is allowed. For the last 15631triplet, the @var{max_size} must be @code{-1}. The @var{max_size} of the triplets 15632in the list must be specified in increasing order. The minimal byte size for 15633@var{alg} is @code{0} for the first triplet and @code{@var{max_size} + 1} of the 15634preceding range. 15635 15636@item -mmemset-strategy=@var{strategy} 15637@opindex mmemset-strategy=@var{strategy} 15638The option is similar to @option{-mmemcpy-strategy=} except that it is to control 15639@code{__builtin_memset} expansion. 15640 15641@item -momit-leaf-frame-pointer 15642@opindex momit-leaf-frame-pointer 15643Don't keep the frame pointer in a register for leaf functions. This 15644avoids the instructions to save, set up, and restore frame pointers and 15645makes an extra register available in leaf functions. The option 15646@option{-fomit-leaf-frame-pointer} removes the frame pointer for leaf functions, 15647which might make debugging harder. 15648 15649@item -mtls-direct-seg-refs 15650@itemx -mno-tls-direct-seg-refs 15651@opindex mtls-direct-seg-refs 15652Controls whether TLS variables may be accessed with offsets from the 15653TLS segment register (@code{%gs} for 32-bit, @code{%fs} for 64-bit), 15654or whether the thread base pointer must be added. Whether or not this 15655is valid depends on the operating system, and whether it maps the 15656segment to cover the entire TLS area. 15657 15658For systems that use the GNU C Library, the default is on. 15659 15660@item -msse2avx 15661@itemx -mno-sse2avx 15662@opindex msse2avx 15663Specify that the assembler should encode SSE instructions with VEX 15664prefix. The option @option{-mavx} turns this on by default. 15665 15666@item -mfentry 15667@itemx -mno-fentry 15668@opindex mfentry 15669If profiling is active (@option{-pg}), put the profiling 15670counter call before the prologue. 15671Note: On x86 architectures the attribute @code{ms_hook_prologue} 15672isn't possible at the moment for @option{-mfentry} and @option{-pg}. 15673 15674@item -m8bit-idiv 15675@itemx -mno-8bit-idiv 15676@opindex 8bit-idiv 15677On some processors, like Intel Atom, 8-bit unsigned integer divide is 15678much faster than 32-bit/64-bit integer divide. This option generates a 15679run-time check. If both dividend and divisor are within range of 0 15680to 255, 8-bit unsigned integer divide is used instead of 1568132-bit/64-bit integer divide. 15682 15683@item -mavx256-split-unaligned-load 15684@itemx -mavx256-split-unaligned-store 15685@opindex avx256-split-unaligned-load 15686@opindex avx256-split-unaligned-store 15687Split 32-byte AVX unaligned load and store. 15688 15689@item -mstack-protector-guard=@var{guard} 15690@opindex mstack-protector-guard=@var{guard} 15691Generate stack protection code using canary at @var{guard}. Supported 15692locations are @samp{global} for global canary or @samp{tls} for per-thread 15693canary in the TLS block (the default). This option has effect only when 15694@option{-fstack-protector} or @option{-fstack-protector-all} is specified. 15695 15696@end table 15697 15698These @samp{-m} switches are supported in addition to the above 15699on x86-64 processors in 64-bit environments. 15700 15701@table @gcctabopt 15702@item -m32 15703@itemx -m64 15704@itemx -mx32 15705@itemx -m16 15706@opindex m32 15707@opindex m64 15708@opindex mx32 15709@opindex m16 15710Generate code for a 16-bit, 32-bit or 64-bit environment. 15711The @option{-m32} option sets @code{int}, @code{long}, and pointer types 15712to 32 bits, and 15713generates code that runs on any i386 system. 15714 15715The @option{-m64} option sets @code{int} to 32 bits and @code{long} and pointer 15716types to 64 bits, and generates code for the x86-64 architecture. 15717For Darwin only the @option{-m64} option also turns off the @option{-fno-pic} 15718and @option{-mdynamic-no-pic} options. 15719 15720The @option{-mx32} option sets @code{int}, @code{long}, and pointer types 15721to 32 bits, and 15722generates code for the x86-64 architecture. 15723 15724The @option{-m16} option is the same as @option{-m32}, except for that 15725it outputs the @code{.code16gcc} assembly directive at the beginning of 15726the assembly output so that the binary can run in 16-bit mode. 15727 15728@item -mno-red-zone 15729@opindex mno-red-zone 15730Do not use a so-called ``red zone'' for x86-64 code. The red zone is mandated 15731by the x86-64 ABI; it is a 128-byte area beyond the location of the 15732stack pointer that is not modified by signal or interrupt handlers 15733and therefore can be used for temporary data without adjusting the stack 15734pointer. The flag @option{-mno-red-zone} disables this red zone. 15735 15736@item -mcmodel=small 15737@opindex mcmodel=small 15738Generate code for the small code model: the program and its symbols must 15739be linked in the lower 2 GB of the address space. Pointers are 64 bits. 15740Programs can be statically or dynamically linked. This is the default 15741code model. 15742 15743@item -mcmodel=kernel 15744@opindex mcmodel=kernel 15745Generate code for the kernel code model. The kernel runs in the 15746negative 2 GB of the address space. 15747This model has to be used for Linux kernel code. 15748 15749@item -mcmodel=medium 15750@opindex mcmodel=medium 15751Generate code for the medium model: the program is linked in the lower 2 15752GB of the address space. Small symbols are also placed there. Symbols 15753with sizes larger than @option{-mlarge-data-threshold} are put into 15754large data or BSS sections and can be located above 2GB. Programs can 15755be statically or dynamically linked. 15756 15757@item -mcmodel=large 15758@opindex mcmodel=large 15759Generate code for the large model. This model makes no assumptions 15760about addresses and sizes of sections. 15761 15762@item -maddress-mode=long 15763@opindex maddress-mode=long 15764Generate code for long address mode. This is only supported for 64-bit 15765and x32 environments. It is the default address mode for 64-bit 15766environments. 15767 15768@item -maddress-mode=short 15769@opindex maddress-mode=short 15770Generate code for short address mode. This is only supported for 32-bit 15771and x32 environments. It is the default address mode for 32-bit and 15772x32 environments. 15773@end table 15774 15775@node i386 and x86-64 Windows Options 15776@subsection i386 and x86-64 Windows Options 15777@cindex i386 and x86-64 Windows Options 15778 15779These additional options are available for Microsoft Windows targets: 15780 15781@table @gcctabopt 15782@item -mconsole 15783@opindex mconsole 15784This option 15785specifies that a console application is to be generated, by 15786instructing the linker to set the PE header subsystem type 15787required for console applications. 15788This option is available for Cygwin and MinGW targets and is 15789enabled by default on those targets. 15790 15791@item -mdll 15792@opindex mdll 15793This option is available for Cygwin and MinGW targets. It 15794specifies that a DLL---a dynamic link library---is to be 15795generated, enabling the selection of the required runtime 15796startup object and entry point. 15797 15798@item -mnop-fun-dllimport 15799@opindex mnop-fun-dllimport 15800This option is available for Cygwin and MinGW targets. It 15801specifies that the @code{dllimport} attribute should be ignored. 15802 15803@item -mthread 15804@opindex mthread 15805This option is available for MinGW targets. It specifies 15806that MinGW-specific thread support is to be used. 15807 15808@item -municode 15809@opindex municode 15810This option is available for MinGW-w64 targets. It causes 15811the @code{UNICODE} preprocessor macro to be predefined, and 15812chooses Unicode-capable runtime startup code. 15813 15814@item -mwin32 15815@opindex mwin32 15816This option is available for Cygwin and MinGW targets. It 15817specifies that the typical Microsoft Windows predefined macros are to 15818be set in the pre-processor, but does not influence the choice 15819of runtime library/startup code. 15820 15821@item -mwindows 15822@opindex mwindows 15823This option is available for Cygwin and MinGW targets. It 15824specifies that a GUI application is to be generated by 15825instructing the linker to set the PE header subsystem type 15826appropriately. 15827 15828@item -fno-set-stack-executable 15829@opindex fno-set-stack-executable 15830This option is available for MinGW targets. It specifies that 15831the executable flag for the stack used by nested functions isn't 15832set. This is necessary for binaries running in kernel mode of 15833Microsoft Windows, as there the User32 API, which is used to set executable 15834privileges, isn't available. 15835 15836@item -fwritable-relocated-rdata 15837@opindex fno-writable-relocated-rdata 15838This option is available for MinGW and Cygwin targets. It specifies 15839that relocated-data in read-only section is put into .data 15840section. This is a necessary for older runtimes not supporting 15841modification of .rdata sections for pseudo-relocation. 15842 15843@item -mpe-aligned-commons 15844@opindex mpe-aligned-commons 15845This option is available for Cygwin and MinGW targets. It 15846specifies that the GNU extension to the PE file format that 15847permits the correct alignment of COMMON variables should be 15848used when generating code. It is enabled by default if 15849GCC detects that the target assembler found during configuration 15850supports the feature. 15851@end table 15852 15853See also under @ref{i386 and x86-64 Options} for standard options. 15854 15855@node IA-64 Options 15856@subsection IA-64 Options 15857@cindex IA-64 Options 15858 15859These are the @samp{-m} options defined for the Intel IA-64 architecture. 15860 15861@table @gcctabopt 15862@item -mbig-endian 15863@opindex mbig-endian 15864Generate code for a big-endian target. This is the default for HP-UX@. 15865 15866@item -mlittle-endian 15867@opindex mlittle-endian 15868Generate code for a little-endian target. This is the default for AIX5 15869and GNU/Linux. 15870 15871@item -mgnu-as 15872@itemx -mno-gnu-as 15873@opindex mgnu-as 15874@opindex mno-gnu-as 15875Generate (or don't) code for the GNU assembler. This is the default. 15876@c Also, this is the default if the configure option @option{--with-gnu-as} 15877@c is used. 15878 15879@item -mgnu-ld 15880@itemx -mno-gnu-ld 15881@opindex mgnu-ld 15882@opindex mno-gnu-ld 15883Generate (or don't) code for the GNU linker. This is the default. 15884@c Also, this is the default if the configure option @option{--with-gnu-ld} 15885@c is used. 15886 15887@item -mno-pic 15888@opindex mno-pic 15889Generate code that does not use a global pointer register. The result 15890is not position independent code, and violates the IA-64 ABI@. 15891 15892@item -mvolatile-asm-stop 15893@itemx -mno-volatile-asm-stop 15894@opindex mvolatile-asm-stop 15895@opindex mno-volatile-asm-stop 15896Generate (or don't) a stop bit immediately before and after volatile asm 15897statements. 15898 15899@item -mregister-names 15900@itemx -mno-register-names 15901@opindex mregister-names 15902@opindex mno-register-names 15903Generate (or don't) @samp{in}, @samp{loc}, and @samp{out} register names for 15904the stacked registers. This may make assembler output more readable. 15905 15906@item -mno-sdata 15907@itemx -msdata 15908@opindex mno-sdata 15909@opindex msdata 15910Disable (or enable) optimizations that use the small data section. This may 15911be useful for working around optimizer bugs. 15912 15913@item -mconstant-gp 15914@opindex mconstant-gp 15915Generate code that uses a single constant global pointer value. This is 15916useful when compiling kernel code. 15917 15918@item -mauto-pic 15919@opindex mauto-pic 15920Generate code that is self-relocatable. This implies @option{-mconstant-gp}. 15921This is useful when compiling firmware code. 15922 15923@item -minline-float-divide-min-latency 15924@opindex minline-float-divide-min-latency 15925Generate code for inline divides of floating-point values 15926using the minimum latency algorithm. 15927 15928@item -minline-float-divide-max-throughput 15929@opindex minline-float-divide-max-throughput 15930Generate code for inline divides of floating-point values 15931using the maximum throughput algorithm. 15932 15933@item -mno-inline-float-divide 15934@opindex mno-inline-float-divide 15935Do not generate inline code for divides of floating-point values. 15936 15937@item -minline-int-divide-min-latency 15938@opindex minline-int-divide-min-latency 15939Generate code for inline divides of integer values 15940using the minimum latency algorithm. 15941 15942@item -minline-int-divide-max-throughput 15943@opindex minline-int-divide-max-throughput 15944Generate code for inline divides of integer values 15945using the maximum throughput algorithm. 15946 15947@item -mno-inline-int-divide 15948@opindex mno-inline-int-divide 15949Do not generate inline code for divides of integer values. 15950 15951@item -minline-sqrt-min-latency 15952@opindex minline-sqrt-min-latency 15953Generate code for inline square roots 15954using the minimum latency algorithm. 15955 15956@item -minline-sqrt-max-throughput 15957@opindex minline-sqrt-max-throughput 15958Generate code for inline square roots 15959using the maximum throughput algorithm. 15960 15961@item -mno-inline-sqrt 15962@opindex mno-inline-sqrt 15963Do not generate inline code for @code{sqrt}. 15964 15965@item -mfused-madd 15966@itemx -mno-fused-madd 15967@opindex mfused-madd 15968@opindex mno-fused-madd 15969Do (don't) generate code that uses the fused multiply/add or multiply/subtract 15970instructions. The default is to use these instructions. 15971 15972@item -mno-dwarf2-asm 15973@itemx -mdwarf2-asm 15974@opindex mno-dwarf2-asm 15975@opindex mdwarf2-asm 15976Don't (or do) generate assembler code for the DWARF 2 line number debugging 15977info. This may be useful when not using the GNU assembler. 15978 15979@item -mearly-stop-bits 15980@itemx -mno-early-stop-bits 15981@opindex mearly-stop-bits 15982@opindex mno-early-stop-bits 15983Allow stop bits to be placed earlier than immediately preceding the 15984instruction that triggered the stop bit. This can improve instruction 15985scheduling, but does not always do so. 15986 15987@item -mfixed-range=@var{register-range} 15988@opindex mfixed-range 15989Generate code treating the given register range as fixed registers. 15990A fixed register is one that the register allocator cannot use. This is 15991useful when compiling kernel code. A register range is specified as 15992two registers separated by a dash. Multiple register ranges can be 15993specified separated by a comma. 15994 15995@item -mtls-size=@var{tls-size} 15996@opindex mtls-size 15997Specify bit size of immediate TLS offsets. Valid values are 14, 22, and 1599864. 15999 16000@item -mtune=@var{cpu-type} 16001@opindex mtune 16002Tune the instruction scheduling for a particular CPU, Valid values are 16003@samp{itanium}, @samp{itanium1}, @samp{merced}, @samp{itanium2}, 16004and @samp{mckinley}. 16005 16006@item -milp32 16007@itemx -mlp64 16008@opindex milp32 16009@opindex mlp64 16010Generate code for a 32-bit or 64-bit environment. 16011The 32-bit environment sets int, long and pointer to 32 bits. 16012The 64-bit environment sets int to 32 bits and long and pointer 16013to 64 bits. These are HP-UX specific flags. 16014 16015@item -mno-sched-br-data-spec 16016@itemx -msched-br-data-spec 16017@opindex mno-sched-br-data-spec 16018@opindex msched-br-data-spec 16019(Dis/En)able data speculative scheduling before reload. 16020This results in generation of @code{ld.a} instructions and 16021the corresponding check instructions (@code{ld.c} / @code{chk.a}). 16022The default is 'disable'. 16023 16024@item -msched-ar-data-spec 16025@itemx -mno-sched-ar-data-spec 16026@opindex msched-ar-data-spec 16027@opindex mno-sched-ar-data-spec 16028(En/Dis)able data speculative scheduling after reload. 16029This results in generation of @code{ld.a} instructions and 16030the corresponding check instructions (@code{ld.c} / @code{chk.a}). 16031The default is 'enable'. 16032 16033@item -mno-sched-control-spec 16034@itemx -msched-control-spec 16035@opindex mno-sched-control-spec 16036@opindex msched-control-spec 16037(Dis/En)able control speculative scheduling. This feature is 16038available only during region scheduling (i.e.@: before reload). 16039This results in generation of the @code{ld.s} instructions and 16040the corresponding check instructions @code{chk.s}. 16041The default is 'disable'. 16042 16043@item -msched-br-in-data-spec 16044@itemx -mno-sched-br-in-data-spec 16045@opindex msched-br-in-data-spec 16046@opindex mno-sched-br-in-data-spec 16047(En/Dis)able speculative scheduling of the instructions that 16048are dependent on the data speculative loads before reload. 16049This is effective only with @option{-msched-br-data-spec} enabled. 16050The default is 'enable'. 16051 16052@item -msched-ar-in-data-spec 16053@itemx -mno-sched-ar-in-data-spec 16054@opindex msched-ar-in-data-spec 16055@opindex mno-sched-ar-in-data-spec 16056(En/Dis)able speculative scheduling of the instructions that 16057are dependent on the data speculative loads after reload. 16058This is effective only with @option{-msched-ar-data-spec} enabled. 16059The default is 'enable'. 16060 16061@item -msched-in-control-spec 16062@itemx -mno-sched-in-control-spec 16063@opindex msched-in-control-spec 16064@opindex mno-sched-in-control-spec 16065(En/Dis)able speculative scheduling of the instructions that 16066are dependent on the control speculative loads. 16067This is effective only with @option{-msched-control-spec} enabled. 16068The default is 'enable'. 16069 16070@item -mno-sched-prefer-non-data-spec-insns 16071@itemx -msched-prefer-non-data-spec-insns 16072@opindex mno-sched-prefer-non-data-spec-insns 16073@opindex msched-prefer-non-data-spec-insns 16074If enabled, data-speculative instructions are chosen for schedule 16075only if there are no other choices at the moment. This makes 16076the use of the data speculation much more conservative. 16077The default is 'disable'. 16078 16079@item -mno-sched-prefer-non-control-spec-insns 16080@itemx -msched-prefer-non-control-spec-insns 16081@opindex mno-sched-prefer-non-control-spec-insns 16082@opindex msched-prefer-non-control-spec-insns 16083If enabled, control-speculative instructions are chosen for schedule 16084only if there are no other choices at the moment. This makes 16085the use of the control speculation much more conservative. 16086The default is 'disable'. 16087 16088@item -mno-sched-count-spec-in-critical-path 16089@itemx -msched-count-spec-in-critical-path 16090@opindex mno-sched-count-spec-in-critical-path 16091@opindex msched-count-spec-in-critical-path 16092If enabled, speculative dependencies are considered during 16093computation of the instructions priorities. This makes the use of the 16094speculation a bit more conservative. 16095The default is 'disable'. 16096 16097@item -msched-spec-ldc 16098@opindex msched-spec-ldc 16099Use a simple data speculation check. This option is on by default. 16100 16101@item -msched-control-spec-ldc 16102@opindex msched-spec-ldc 16103Use a simple check for control speculation. This option is on by default. 16104 16105@item -msched-stop-bits-after-every-cycle 16106@opindex msched-stop-bits-after-every-cycle 16107Place a stop bit after every cycle when scheduling. This option is on 16108by default. 16109 16110@item -msched-fp-mem-deps-zero-cost 16111@opindex msched-fp-mem-deps-zero-cost 16112Assume that floating-point stores and loads are not likely to cause a conflict 16113when placed into the same instruction group. This option is disabled by 16114default. 16115 16116@item -msel-sched-dont-check-control-spec 16117@opindex msel-sched-dont-check-control-spec 16118Generate checks for control speculation in selective scheduling. 16119This flag is disabled by default. 16120 16121@item -msched-max-memory-insns=@var{max-insns} 16122@opindex msched-max-memory-insns 16123Limit on the number of memory insns per instruction group, giving lower 16124priority to subsequent memory insns attempting to schedule in the same 16125instruction group. Frequently useful to prevent cache bank conflicts. 16126The default value is 1. 16127 16128@item -msched-max-memory-insns-hard-limit 16129@opindex msched-max-memory-insns-hard-limit 16130Makes the limit specified by @option{msched-max-memory-insns} a hard limit, 16131disallowing more than that number in an instruction group. 16132Otherwise, the limit is ``soft'', meaning that non-memory operations 16133are preferred when the limit is reached, but memory operations may still 16134be scheduled. 16135 16136@end table 16137 16138@node LM32 Options 16139@subsection LM32 Options 16140@cindex LM32 options 16141 16142These @option{-m} options are defined for the LatticeMico32 architecture: 16143 16144@table @gcctabopt 16145@item -mbarrel-shift-enabled 16146@opindex mbarrel-shift-enabled 16147Enable barrel-shift instructions. 16148 16149@item -mdivide-enabled 16150@opindex mdivide-enabled 16151Enable divide and modulus instructions. 16152 16153@item -mmultiply-enabled 16154@opindex multiply-enabled 16155Enable multiply instructions. 16156 16157@item -msign-extend-enabled 16158@opindex msign-extend-enabled 16159Enable sign extend instructions. 16160 16161@item -muser-enabled 16162@opindex muser-enabled 16163Enable user-defined instructions. 16164 16165@end table 16166 16167@node M32C Options 16168@subsection M32C Options 16169@cindex M32C options 16170 16171@table @gcctabopt 16172@item -mcpu=@var{name} 16173@opindex mcpu= 16174Select the CPU for which code is generated. @var{name} may be one of 16175@samp{r8c} for the R8C/Tiny series, @samp{m16c} for the M16C (up to 16176/60) series, @samp{m32cm} for the M16C/80 series, or @samp{m32c} for 16177the M32C/80 series. 16178 16179@item -msim 16180@opindex msim 16181Specifies that the program will be run on the simulator. This causes 16182an alternate runtime library to be linked in which supports, for 16183example, file I/O@. You must not use this option when generating 16184programs that will run on real hardware; you must provide your own 16185runtime library for whatever I/O functions are needed. 16186 16187@item -memregs=@var{number} 16188@opindex memregs= 16189Specifies the number of memory-based pseudo-registers GCC uses 16190during code generation. These pseudo-registers are used like real 16191registers, so there is a tradeoff between GCC's ability to fit the 16192code into available registers, and the performance penalty of using 16193memory instead of registers. Note that all modules in a program must 16194be compiled with the same value for this option. Because of that, you 16195must not use this option with GCC's default runtime libraries. 16196 16197@end table 16198 16199@node M32R/D Options 16200@subsection M32R/D Options 16201@cindex M32R/D options 16202 16203These @option{-m} options are defined for Renesas M32R/D architectures: 16204 16205@table @gcctabopt 16206@item -m32r2 16207@opindex m32r2 16208Generate code for the M32R/2@. 16209 16210@item -m32rx 16211@opindex m32rx 16212Generate code for the M32R/X@. 16213 16214@item -m32r 16215@opindex m32r 16216Generate code for the M32R@. This is the default. 16217 16218@item -mmodel=small 16219@opindex mmodel=small 16220Assume all objects live in the lower 16MB of memory (so that their addresses 16221can be loaded with the @code{ld24} instruction), and assume all subroutines 16222are reachable with the @code{bl} instruction. 16223This is the default. 16224 16225The addressability of a particular object can be set with the 16226@code{model} attribute. 16227 16228@item -mmodel=medium 16229@opindex mmodel=medium 16230Assume objects may be anywhere in the 32-bit address space (the compiler 16231generates @code{seth/add3} instructions to load their addresses), and 16232assume all subroutines are reachable with the @code{bl} instruction. 16233 16234@item -mmodel=large 16235@opindex mmodel=large 16236Assume objects may be anywhere in the 32-bit address space (the compiler 16237generates @code{seth/add3} instructions to load their addresses), and 16238assume subroutines may not be reachable with the @code{bl} instruction 16239(the compiler generates the much slower @code{seth/add3/jl} 16240instruction sequence). 16241 16242@item -msdata=none 16243@opindex msdata=none 16244Disable use of the small data area. Variables are put into 16245one of @samp{.data}, @samp{.bss}, or @samp{.rodata} (unless the 16246@code{section} attribute has been specified). 16247This is the default. 16248 16249The small data area consists of sections @samp{.sdata} and @samp{.sbss}. 16250Objects may be explicitly put in the small data area with the 16251@code{section} attribute using one of these sections. 16252 16253@item -msdata=sdata 16254@opindex msdata=sdata 16255Put small global and static data in the small data area, but do not 16256generate special code to reference them. 16257 16258@item -msdata=use 16259@opindex msdata=use 16260Put small global and static data in the small data area, and generate 16261special instructions to reference them. 16262 16263@item -G @var{num} 16264@opindex G 16265@cindex smaller data references 16266Put global and static objects less than or equal to @var{num} bytes 16267into the small data or BSS sections instead of the normal data or BSS 16268sections. The default value of @var{num} is 8. 16269The @option{-msdata} option must be set to one of @samp{sdata} or @samp{use} 16270for this option to have any effect. 16271 16272All modules should be compiled with the same @option{-G @var{num}} value. 16273Compiling with different values of @var{num} may or may not work; if it 16274doesn't the linker gives an error message---incorrect code is not 16275generated. 16276 16277@item -mdebug 16278@opindex mdebug 16279Makes the M32R-specific code in the compiler display some statistics 16280that might help in debugging programs. 16281 16282@item -malign-loops 16283@opindex malign-loops 16284Align all loops to a 32-byte boundary. 16285 16286@item -mno-align-loops 16287@opindex mno-align-loops 16288Do not enforce a 32-byte alignment for loops. This is the default. 16289 16290@item -missue-rate=@var{number} 16291@opindex missue-rate=@var{number} 16292Issue @var{number} instructions per cycle. @var{number} can only be 1 16293or 2. 16294 16295@item -mbranch-cost=@var{number} 16296@opindex mbranch-cost=@var{number} 16297@var{number} can only be 1 or 2. If it is 1 then branches are 16298preferred over conditional code, if it is 2, then the opposite applies. 16299 16300@item -mflush-trap=@var{number} 16301@opindex mflush-trap=@var{number} 16302Specifies the trap number to use to flush the cache. The default is 1630312. Valid numbers are between 0 and 15 inclusive. 16304 16305@item -mno-flush-trap 16306@opindex mno-flush-trap 16307Specifies that the cache cannot be flushed by using a trap. 16308 16309@item -mflush-func=@var{name} 16310@opindex mflush-func=@var{name} 16311Specifies the name of the operating system function to call to flush 16312the cache. The default is @emph{_flush_cache}, but a function call 16313is only used if a trap is not available. 16314 16315@item -mno-flush-func 16316@opindex mno-flush-func 16317Indicates that there is no OS function for flushing the cache. 16318 16319@end table 16320 16321@node M680x0 Options 16322@subsection M680x0 Options 16323@cindex M680x0 options 16324 16325These are the @samp{-m} options defined for M680x0 and ColdFire processors. 16326The default settings depend on which architecture was selected when 16327the compiler was configured; the defaults for the most common choices 16328are given below. 16329 16330@table @gcctabopt 16331@item -march=@var{arch} 16332@opindex march 16333Generate code for a specific M680x0 or ColdFire instruction set 16334architecture. Permissible values of @var{arch} for M680x0 16335architectures are: @samp{68000}, @samp{68010}, @samp{68020}, 16336@samp{68030}, @samp{68040}, @samp{68060} and @samp{cpu32}. ColdFire 16337architectures are selected according to Freescale's ISA classification 16338and the permissible values are: @samp{isaa}, @samp{isaaplus}, 16339@samp{isab} and @samp{isac}. 16340 16341GCC defines a macro @samp{__mcf@var{arch}__} whenever it is generating 16342code for a ColdFire target. The @var{arch} in this macro is one of the 16343@option{-march} arguments given above. 16344 16345When used together, @option{-march} and @option{-mtune} select code 16346that runs on a family of similar processors but that is optimized 16347for a particular microarchitecture. 16348 16349@item -mcpu=@var{cpu} 16350@opindex mcpu 16351Generate code for a specific M680x0 or ColdFire processor. 16352The M680x0 @var{cpu}s are: @samp{68000}, @samp{68010}, @samp{68020}, 16353@samp{68030}, @samp{68040}, @samp{68060}, @samp{68302}, @samp{68332} 16354and @samp{cpu32}. The ColdFire @var{cpu}s are given by the table 16355below, which also classifies the CPUs into families: 16356 16357@multitable @columnfractions 0.20 0.80 16358@item @strong{Family} @tab @strong{@samp{-mcpu} arguments} 16359@item @samp{51} @tab @samp{51} @samp{51ac} @samp{51ag} @samp{51cn} @samp{51em} @samp{51je} @samp{51jf} @samp{51jg} @samp{51jm} @samp{51mm} @samp{51qe} @samp{51qm} 16360@item @samp{5206} @tab @samp{5202} @samp{5204} @samp{5206} 16361@item @samp{5206e} @tab @samp{5206e} 16362@item @samp{5208} @tab @samp{5207} @samp{5208} 16363@item @samp{5211a} @tab @samp{5210a} @samp{5211a} 16364@item @samp{5213} @tab @samp{5211} @samp{5212} @samp{5213} 16365@item @samp{5216} @tab @samp{5214} @samp{5216} 16366@item @samp{52235} @tab @samp{52230} @samp{52231} @samp{52232} @samp{52233} @samp{52234} @samp{52235} 16367@item @samp{5225} @tab @samp{5224} @samp{5225} 16368@item @samp{52259} @tab @samp{52252} @samp{52254} @samp{52255} @samp{52256} @samp{52258} @samp{52259} 16369@item @samp{5235} @tab @samp{5232} @samp{5233} @samp{5234} @samp{5235} @samp{523x} 16370@item @samp{5249} @tab @samp{5249} 16371@item @samp{5250} @tab @samp{5250} 16372@item @samp{5271} @tab @samp{5270} @samp{5271} 16373@item @samp{5272} @tab @samp{5272} 16374@item @samp{5275} @tab @samp{5274} @samp{5275} 16375@item @samp{5282} @tab @samp{5280} @samp{5281} @samp{5282} @samp{528x} 16376@item @samp{53017} @tab @samp{53011} @samp{53012} @samp{53013} @samp{53014} @samp{53015} @samp{53016} @samp{53017} 16377@item @samp{5307} @tab @samp{5307} 16378@item @samp{5329} @tab @samp{5327} @samp{5328} @samp{5329} @samp{532x} 16379@item @samp{5373} @tab @samp{5372} @samp{5373} @samp{537x} 16380@item @samp{5407} @tab @samp{5407} 16381@item @samp{5475} @tab @samp{5470} @samp{5471} @samp{5472} @samp{5473} @samp{5474} @samp{5475} @samp{547x} @samp{5480} @samp{5481} @samp{5482} @samp{5483} @samp{5484} @samp{5485} 16382@end multitable 16383 16384@option{-mcpu=@var{cpu}} overrides @option{-march=@var{arch}} if 16385@var{arch} is compatible with @var{cpu}. Other combinations of 16386@option{-mcpu} and @option{-march} are rejected. 16387 16388GCC defines the macro @samp{__mcf_cpu_@var{cpu}} when ColdFire target 16389@var{cpu} is selected. It also defines @samp{__mcf_family_@var{family}}, 16390where the value of @var{family} is given by the table above. 16391 16392@item -mtune=@var{tune} 16393@opindex mtune 16394Tune the code for a particular microarchitecture within the 16395constraints set by @option{-march} and @option{-mcpu}. 16396The M680x0 microarchitectures are: @samp{68000}, @samp{68010}, 16397@samp{68020}, @samp{68030}, @samp{68040}, @samp{68060} 16398and @samp{cpu32}. The ColdFire microarchitectures 16399are: @samp{cfv1}, @samp{cfv2}, @samp{cfv3}, @samp{cfv4} and @samp{cfv4e}. 16400 16401You can also use @option{-mtune=68020-40} for code that needs 16402to run relatively well on 68020, 68030 and 68040 targets. 16403@option{-mtune=68020-60} is similar but includes 68060 targets 16404as well. These two options select the same tuning decisions as 16405@option{-m68020-40} and @option{-m68020-60} respectively. 16406 16407GCC defines the macros @samp{__mc@var{arch}} and @samp{__mc@var{arch}__} 16408when tuning for 680x0 architecture @var{arch}. It also defines 16409@samp{mc@var{arch}} unless either @option{-ansi} or a non-GNU @option{-std} 16410option is used. If GCC is tuning for a range of architectures, 16411as selected by @option{-mtune=68020-40} or @option{-mtune=68020-60}, 16412it defines the macros for every architecture in the range. 16413 16414GCC also defines the macro @samp{__m@var{uarch}__} when tuning for 16415ColdFire microarchitecture @var{uarch}, where @var{uarch} is one 16416of the arguments given above. 16417 16418@item -m68000 16419@itemx -mc68000 16420@opindex m68000 16421@opindex mc68000 16422Generate output for a 68000. This is the default 16423when the compiler is configured for 68000-based systems. 16424It is equivalent to @option{-march=68000}. 16425 16426Use this option for microcontrollers with a 68000 or EC000 core, 16427including the 68008, 68302, 68306, 68307, 68322, 68328 and 68356. 16428 16429@item -m68010 16430@opindex m68010 16431Generate output for a 68010. This is the default 16432when the compiler is configured for 68010-based systems. 16433It is equivalent to @option{-march=68010}. 16434 16435@item -m68020 16436@itemx -mc68020 16437@opindex m68020 16438@opindex mc68020 16439Generate output for a 68020. This is the default 16440when the compiler is configured for 68020-based systems. 16441It is equivalent to @option{-march=68020}. 16442 16443@item -m68030 16444@opindex m68030 16445Generate output for a 68030. This is the default when the compiler is 16446configured for 68030-based systems. It is equivalent to 16447@option{-march=68030}. 16448 16449@item -m68040 16450@opindex m68040 16451Generate output for a 68040. This is the default when the compiler is 16452configured for 68040-based systems. It is equivalent to 16453@option{-march=68040}. 16454 16455This option inhibits the use of 68881/68882 instructions that have to be 16456emulated by software on the 68040. Use this option if your 68040 does not 16457have code to emulate those instructions. 16458 16459@item -m68060 16460@opindex m68060 16461Generate output for a 68060. This is the default when the compiler is 16462configured for 68060-based systems. It is equivalent to 16463@option{-march=68060}. 16464 16465This option inhibits the use of 68020 and 68881/68882 instructions that 16466have to be emulated by software on the 68060. Use this option if your 68060 16467does not have code to emulate those instructions. 16468 16469@item -mcpu32 16470@opindex mcpu32 16471Generate output for a CPU32. This is the default 16472when the compiler is configured for CPU32-based systems. 16473It is equivalent to @option{-march=cpu32}. 16474 16475Use this option for microcontrollers with a 16476CPU32 or CPU32+ core, including the 68330, 68331, 68332, 68333, 68334, 1647768336, 68340, 68341, 68349 and 68360. 16478 16479@item -m5200 16480@opindex m5200 16481Generate output for a 520X ColdFire CPU@. This is the default 16482when the compiler is configured for 520X-based systems. 16483It is equivalent to @option{-mcpu=5206}, and is now deprecated 16484in favor of that option. 16485 16486Use this option for microcontroller with a 5200 core, including 16487the MCF5202, MCF5203, MCF5204 and MCF5206. 16488 16489@item -m5206e 16490@opindex m5206e 16491Generate output for a 5206e ColdFire CPU@. The option is now 16492deprecated in favor of the equivalent @option{-mcpu=5206e}. 16493 16494@item -m528x 16495@opindex m528x 16496Generate output for a member of the ColdFire 528X family. 16497The option is now deprecated in favor of the equivalent 16498@option{-mcpu=528x}. 16499 16500@item -m5307 16501@opindex m5307 16502Generate output for a ColdFire 5307 CPU@. The option is now deprecated 16503in favor of the equivalent @option{-mcpu=5307}. 16504 16505@item -m5407 16506@opindex m5407 16507Generate output for a ColdFire 5407 CPU@. The option is now deprecated 16508in favor of the equivalent @option{-mcpu=5407}. 16509 16510@item -mcfv4e 16511@opindex mcfv4e 16512Generate output for a ColdFire V4e family CPU (e.g.@: 547x/548x). 16513This includes use of hardware floating-point instructions. 16514The option is equivalent to @option{-mcpu=547x}, and is now 16515deprecated in favor of that option. 16516 16517@item -m68020-40 16518@opindex m68020-40 16519Generate output for a 68040, without using any of the new instructions. 16520This results in code that can run relatively efficiently on either a 1652168020/68881 or a 68030 or a 68040. The generated code does use the 1652268881 instructions that are emulated on the 68040. 16523 16524The option is equivalent to @option{-march=68020} @option{-mtune=68020-40}. 16525 16526@item -m68020-60 16527@opindex m68020-60 16528Generate output for a 68060, without using any of the new instructions. 16529This results in code that can run relatively efficiently on either a 1653068020/68881 or a 68030 or a 68040. The generated code does use the 1653168881 instructions that are emulated on the 68060. 16532 16533The option is equivalent to @option{-march=68020} @option{-mtune=68020-60}. 16534 16535@item -mhard-float 16536@itemx -m68881 16537@opindex mhard-float 16538@opindex m68881 16539Generate floating-point instructions. This is the default for 68020 16540and above, and for ColdFire devices that have an FPU@. It defines the 16541macro @samp{__HAVE_68881__} on M680x0 targets and @samp{__mcffpu__} 16542on ColdFire targets. 16543 16544@item -msoft-float 16545@opindex msoft-float 16546Do not generate floating-point instructions; use library calls instead. 16547This is the default for 68000, 68010, and 68832 targets. It is also 16548the default for ColdFire devices that have no FPU. 16549 16550@item -mdiv 16551@itemx -mno-div 16552@opindex mdiv 16553@opindex mno-div 16554Generate (do not generate) ColdFire hardware divide and remainder 16555instructions. If @option{-march} is used without @option{-mcpu}, 16556the default is ``on'' for ColdFire architectures and ``off'' for M680x0 16557architectures. Otherwise, the default is taken from the target CPU 16558(either the default CPU, or the one specified by @option{-mcpu}). For 16559example, the default is ``off'' for @option{-mcpu=5206} and ``on'' for 16560@option{-mcpu=5206e}. 16561 16562GCC defines the macro @samp{__mcfhwdiv__} when this option is enabled. 16563 16564@item -mshort 16565@opindex mshort 16566Consider type @code{int} to be 16 bits wide, like @code{short int}. 16567Additionally, parameters passed on the stack are also aligned to a 1656816-bit boundary even on targets whose API mandates promotion to 32-bit. 16569 16570@item -mno-short 16571@opindex mno-short 16572Do not consider type @code{int} to be 16 bits wide. This is the default. 16573 16574@item -mnobitfield 16575@itemx -mno-bitfield 16576@opindex mnobitfield 16577@opindex mno-bitfield 16578Do not use the bit-field instructions. The @option{-m68000}, @option{-mcpu32} 16579and @option{-m5200} options imply @w{@option{-mnobitfield}}. 16580 16581@item -mbitfield 16582@opindex mbitfield 16583Do use the bit-field instructions. The @option{-m68020} option implies 16584@option{-mbitfield}. This is the default if you use a configuration 16585designed for a 68020. 16586 16587@item -mrtd 16588@opindex mrtd 16589Use a different function-calling convention, in which functions 16590that take a fixed number of arguments return with the @code{rtd} 16591instruction, which pops their arguments while returning. This 16592saves one instruction in the caller since there is no need to pop 16593the arguments there. 16594 16595This calling convention is incompatible with the one normally 16596used on Unix, so you cannot use it if you need to call libraries 16597compiled with the Unix compiler. 16598 16599Also, you must provide function prototypes for all functions that 16600take variable numbers of arguments (including @code{printf}); 16601otherwise incorrect code is generated for calls to those 16602functions. 16603 16604In addition, seriously incorrect code results if you call a 16605function with too many arguments. (Normally, extra arguments are 16606harmlessly ignored.) 16607 16608The @code{rtd} instruction is supported by the 68010, 68020, 68030, 1660968040, 68060 and CPU32 processors, but not by the 68000 or 5200. 16610 16611@item -mno-rtd 16612@opindex mno-rtd 16613Do not use the calling conventions selected by @option{-mrtd}. 16614This is the default. 16615 16616@item -malign-int 16617@itemx -mno-align-int 16618@opindex malign-int 16619@opindex mno-align-int 16620Control whether GCC aligns @code{int}, @code{long}, @code{long long}, 16621@code{float}, @code{double}, and @code{long double} variables on a 32-bit 16622boundary (@option{-malign-int}) or a 16-bit boundary (@option{-mno-align-int}). 16623Aligning variables on 32-bit boundaries produces code that runs somewhat 16624faster on processors with 32-bit busses at the expense of more memory. 16625 16626@strong{Warning:} if you use the @option{-malign-int} switch, GCC 16627aligns structures containing the above types differently than 16628most published application binary interface specifications for the m68k. 16629 16630@item -mpcrel 16631@opindex mpcrel 16632Use the pc-relative addressing mode of the 68000 directly, instead of 16633using a global offset table. At present, this option implies @option{-fpic}, 16634allowing at most a 16-bit offset for pc-relative addressing. @option{-fPIC} is 16635not presently supported with @option{-mpcrel}, though this could be supported for 1663668020 and higher processors. 16637 16638@item -mno-strict-align 16639@itemx -mstrict-align 16640@opindex mno-strict-align 16641@opindex mstrict-align 16642Do not (do) assume that unaligned memory references are handled by 16643the system. 16644 16645@item -msep-data 16646Generate code that allows the data segment to be located in a different 16647area of memory from the text segment. This allows for execute-in-place in 16648an environment without virtual memory management. This option implies 16649@option{-fPIC}. 16650 16651@item -mno-sep-data 16652Generate code that assumes that the data segment follows the text segment. 16653This is the default. 16654 16655@item -mid-shared-library 16656Generate code that supports shared libraries via the library ID method. 16657This allows for execute-in-place and shared libraries in an environment 16658without virtual memory management. This option implies @option{-fPIC}. 16659 16660@item -mno-id-shared-library 16661Generate code that doesn't assume ID-based shared libraries are being used. 16662This is the default. 16663 16664@item -mshared-library-id=n 16665Specifies the identification number of the ID-based shared library being 16666compiled. Specifying a value of 0 generates more compact code; specifying 16667other values forces the allocation of that number to the current 16668library, but is no more space- or time-efficient than omitting this option. 16669 16670@item -mxgot 16671@itemx -mno-xgot 16672@opindex mxgot 16673@opindex mno-xgot 16674When generating position-independent code for ColdFire, generate code 16675that works if the GOT has more than 8192 entries. This code is 16676larger and slower than code generated without this option. On M680x0 16677processors, this option is not needed; @option{-fPIC} suffices. 16678 16679GCC normally uses a single instruction to load values from the GOT@. 16680While this is relatively efficient, it only works if the GOT 16681is smaller than about 64k. Anything larger causes the linker 16682to report an error such as: 16683 16684@cindex relocation truncated to fit (ColdFire) 16685@smallexample 16686relocation truncated to fit: R_68K_GOT16O foobar 16687@end smallexample 16688 16689If this happens, you should recompile your code with @option{-mxgot}. 16690It should then work with very large GOTs. However, code generated with 16691@option{-mxgot} is less efficient, since it takes 4 instructions to fetch 16692the value of a global symbol. 16693 16694Note that some linkers, including newer versions of the GNU linker, 16695can create multiple GOTs and sort GOT entries. If you have such a linker, 16696you should only need to use @option{-mxgot} when compiling a single 16697object file that accesses more than 8192 GOT entries. Very few do. 16698 16699These options have no effect unless GCC is generating 16700position-independent code. 16701 16702@end table 16703 16704@node MCore Options 16705@subsection MCore Options 16706@cindex MCore options 16707 16708These are the @samp{-m} options defined for the Motorola M*Core 16709processors. 16710 16711@table @gcctabopt 16712 16713@item -mhardlit 16714@itemx -mno-hardlit 16715@opindex mhardlit 16716@opindex mno-hardlit 16717Inline constants into the code stream if it can be done in two 16718instructions or less. 16719 16720@item -mdiv 16721@itemx -mno-div 16722@opindex mdiv 16723@opindex mno-div 16724Use the divide instruction. (Enabled by default). 16725 16726@item -mrelax-immediate 16727@itemx -mno-relax-immediate 16728@opindex mrelax-immediate 16729@opindex mno-relax-immediate 16730Allow arbitrary-sized immediates in bit operations. 16731 16732@item -mwide-bitfields 16733@itemx -mno-wide-bitfields 16734@opindex mwide-bitfields 16735@opindex mno-wide-bitfields 16736Always treat bit-fields as @code{int}-sized. 16737 16738@item -m4byte-functions 16739@itemx -mno-4byte-functions 16740@opindex m4byte-functions 16741@opindex mno-4byte-functions 16742Force all functions to be aligned to a 4-byte boundary. 16743 16744@item -mcallgraph-data 16745@itemx -mno-callgraph-data 16746@opindex mcallgraph-data 16747@opindex mno-callgraph-data 16748Emit callgraph information. 16749 16750@item -mslow-bytes 16751@itemx -mno-slow-bytes 16752@opindex mslow-bytes 16753@opindex mno-slow-bytes 16754Prefer word access when reading byte quantities. 16755 16756@item -mlittle-endian 16757@itemx -mbig-endian 16758@opindex mlittle-endian 16759@opindex mbig-endian 16760Generate code for a little-endian target. 16761 16762@item -m210 16763@itemx -m340 16764@opindex m210 16765@opindex m340 16766Generate code for the 210 processor. 16767 16768@item -mno-lsim 16769@opindex mno-lsim 16770Assume that runtime support has been provided and so omit the 16771simulator library (@file{libsim.a)} from the linker command line. 16772 16773@item -mstack-increment=@var{size} 16774@opindex mstack-increment 16775Set the maximum amount for a single stack increment operation. Large 16776values can increase the speed of programs that contain functions 16777that need a large amount of stack space, but they can also trigger a 16778segmentation fault if the stack is extended too much. The default 16779value is 0x1000. 16780 16781@end table 16782 16783@node MeP Options 16784@subsection MeP Options 16785@cindex MeP options 16786 16787@table @gcctabopt 16788 16789@item -mabsdiff 16790@opindex mabsdiff 16791Enables the @code{abs} instruction, which is the absolute difference 16792between two registers. 16793 16794@item -mall-opts 16795@opindex mall-opts 16796Enables all the optional instructions---average, multiply, divide, bit 16797operations, leading zero, absolute difference, min/max, clip, and 16798saturation. 16799 16800 16801@item -maverage 16802@opindex maverage 16803Enables the @code{ave} instruction, which computes the average of two 16804registers. 16805 16806@item -mbased=@var{n} 16807@opindex mbased= 16808Variables of size @var{n} bytes or smaller are placed in the 16809@code{.based} section by default. Based variables use the @code{$tp} 16810register as a base register, and there is a 128-byte limit to the 16811@code{.based} section. 16812 16813@item -mbitops 16814@opindex mbitops 16815Enables the bit operation instructions---bit test (@code{btstm}), set 16816(@code{bsetm}), clear (@code{bclrm}), invert (@code{bnotm}), and 16817test-and-set (@code{tas}). 16818 16819@item -mc=@var{name} 16820@opindex mc= 16821Selects which section constant data is placed in. @var{name} may 16822be @code{tiny}, @code{near}, or @code{far}. 16823 16824@item -mclip 16825@opindex mclip 16826Enables the @code{clip} instruction. Note that @code{-mclip} is not 16827useful unless you also provide @code{-mminmax}. 16828 16829@item -mconfig=@var{name} 16830@opindex mconfig= 16831Selects one of the built-in core configurations. Each MeP chip has 16832one or more modules in it; each module has a core CPU and a variety of 16833coprocessors, optional instructions, and peripherals. The 16834@code{MeP-Integrator} tool, not part of GCC, provides these 16835configurations through this option; using this option is the same as 16836using all the corresponding command-line options. The default 16837configuration is @code{default}. 16838 16839@item -mcop 16840@opindex mcop 16841Enables the coprocessor instructions. By default, this is a 32-bit 16842coprocessor. Note that the coprocessor is normally enabled via the 16843@code{-mconfig=} option. 16844 16845@item -mcop32 16846@opindex mcop32 16847Enables the 32-bit coprocessor's instructions. 16848 16849@item -mcop64 16850@opindex mcop64 16851Enables the 64-bit coprocessor's instructions. 16852 16853@item -mivc2 16854@opindex mivc2 16855Enables IVC2 scheduling. IVC2 is a 64-bit VLIW coprocessor. 16856 16857@item -mdc 16858@opindex mdc 16859Causes constant variables to be placed in the @code{.near} section. 16860 16861@item -mdiv 16862@opindex mdiv 16863Enables the @code{div} and @code{divu} instructions. 16864 16865@item -meb 16866@opindex meb 16867Generate big-endian code. 16868 16869@item -mel 16870@opindex mel 16871Generate little-endian code. 16872 16873@item -mio-volatile 16874@opindex mio-volatile 16875Tells the compiler that any variable marked with the @code{io} 16876attribute is to be considered volatile. 16877 16878@item -ml 16879@opindex ml 16880Causes variables to be assigned to the @code{.far} section by default. 16881 16882@item -mleadz 16883@opindex mleadz 16884Enables the @code{leadz} (leading zero) instruction. 16885 16886@item -mm 16887@opindex mm 16888Causes variables to be assigned to the @code{.near} section by default. 16889 16890@item -mminmax 16891@opindex mminmax 16892Enables the @code{min} and @code{max} instructions. 16893 16894@item -mmult 16895@opindex mmult 16896Enables the multiplication and multiply-accumulate instructions. 16897 16898@item -mno-opts 16899@opindex mno-opts 16900Disables all the optional instructions enabled by @code{-mall-opts}. 16901 16902@item -mrepeat 16903@opindex mrepeat 16904Enables the @code{repeat} and @code{erepeat} instructions, used for 16905low-overhead looping. 16906 16907@item -ms 16908@opindex ms 16909Causes all variables to default to the @code{.tiny} section. Note 16910that there is a 65536-byte limit to this section. Accesses to these 16911variables use the @code{%gp} base register. 16912 16913@item -msatur 16914@opindex msatur 16915Enables the saturation instructions. Note that the compiler does not 16916currently generate these itself, but this option is included for 16917compatibility with other tools, like @code{as}. 16918 16919@item -msdram 16920@opindex msdram 16921Link the SDRAM-based runtime instead of the default ROM-based runtime. 16922 16923@item -msim 16924@opindex msim 16925Link the simulator run-time libraries. 16926 16927@item -msimnovec 16928@opindex msimnovec 16929Link the simulator runtime libraries, excluding built-in support 16930for reset and exception vectors and tables. 16931 16932@item -mtf 16933@opindex mtf 16934Causes all functions to default to the @code{.far} section. Without 16935this option, functions default to the @code{.near} section. 16936 16937@item -mtiny=@var{n} 16938@opindex mtiny= 16939Variables that are @var{n} bytes or smaller are allocated to the 16940@code{.tiny} section. These variables use the @code{$gp} base 16941register. The default for this option is 4, but note that there's a 1694265536-byte limit to the @code{.tiny} section. 16943 16944@end table 16945 16946@node MicroBlaze Options 16947@subsection MicroBlaze Options 16948@cindex MicroBlaze Options 16949 16950@table @gcctabopt 16951 16952@item -msoft-float 16953@opindex msoft-float 16954Use software emulation for floating point (default). 16955 16956@item -mhard-float 16957@opindex mhard-float 16958Use hardware floating-point instructions. 16959 16960@item -mmemcpy 16961@opindex mmemcpy 16962Do not optimize block moves, use @code{memcpy}. 16963 16964@item -mno-clearbss 16965@opindex mno-clearbss 16966This option is deprecated. Use @option{-fno-zero-initialized-in-bss} instead. 16967 16968@item -mcpu=@var{cpu-type} 16969@opindex mcpu= 16970Use features of, and schedule code for, the given CPU. 16971Supported values are in the format @samp{v@var{X}.@var{YY}.@var{Z}}, 16972where @var{X} is a major version, @var{YY} is the minor version, and 16973@var{Z} is compatibility code. Example values are @samp{v3.00.a}, 16974@samp{v4.00.b}, @samp{v5.00.a}, @samp{v5.00.b}, @samp{v5.00.b}, @samp{v6.00.a}. 16975 16976@item -mxl-soft-mul 16977@opindex mxl-soft-mul 16978Use software multiply emulation (default). 16979 16980@item -mxl-soft-div 16981@opindex mxl-soft-div 16982Use software emulation for divides (default). 16983 16984@item -mxl-barrel-shift 16985@opindex mxl-barrel-shift 16986Use the hardware barrel shifter. 16987 16988@item -mxl-pattern-compare 16989@opindex mxl-pattern-compare 16990Use pattern compare instructions. 16991 16992@item -msmall-divides 16993@opindex msmall-divides 16994Use table lookup optimization for small signed integer divisions. 16995 16996@item -mxl-stack-check 16997@opindex mxl-stack-check 16998This option is deprecated. Use @option{-fstack-check} instead. 16999 17000@item -mxl-gp-opt 17001@opindex mxl-gp-opt 17002Use GP-relative @code{.sdata}/@code{.sbss} sections. 17003 17004@item -mxl-multiply-high 17005@opindex mxl-multiply-high 17006Use multiply high instructions for high part of 32x32 multiply. 17007 17008@item -mxl-float-convert 17009@opindex mxl-float-convert 17010Use hardware floating-point conversion instructions. 17011 17012@item -mxl-float-sqrt 17013@opindex mxl-float-sqrt 17014Use hardware floating-point square root instruction. 17015 17016@item -mbig-endian 17017@opindex mbig-endian 17018Generate code for a big-endian target. 17019 17020@item -mlittle-endian 17021@opindex mlittle-endian 17022Generate code for a little-endian target. 17023 17024@item -mxl-reorder 17025@opindex mxl-reorder 17026Use reorder instructions (swap and byte reversed load/store). 17027 17028@item -mxl-mode-@var{app-model} 17029Select application model @var{app-model}. Valid models are 17030@table @samp 17031@item executable 17032normal executable (default), uses startup code @file{crt0.o}. 17033 17034@item xmdstub 17035for use with Xilinx Microprocessor Debugger (XMD) based 17036software intrusive debug agent called xmdstub. This uses startup file 17037@file{crt1.o} and sets the start address of the program to 0x800. 17038 17039@item bootstrap 17040for applications that are loaded using a bootloader. 17041This model uses startup file @file{crt2.o} which does not contain a processor 17042reset vector handler. This is suitable for transferring control on a 17043processor reset to the bootloader rather than the application. 17044 17045@item novectors 17046for applications that do not require any of the 17047MicroBlaze vectors. This option may be useful for applications running 17048within a monitoring application. This model uses @file{crt3.o} as a startup file. 17049@end table 17050 17051Option @option{-xl-mode-@var{app-model}} is a deprecated alias for 17052@option{-mxl-mode-@var{app-model}}. 17053 17054@end table 17055 17056@node MIPS Options 17057@subsection MIPS Options 17058@cindex MIPS options 17059 17060@table @gcctabopt 17061 17062@item -EB 17063@opindex EB 17064Generate big-endian code. 17065 17066@item -EL 17067@opindex EL 17068Generate little-endian code. This is the default for @samp{mips*el-*-*} 17069configurations. 17070 17071@item -march=@var{arch} 17072@opindex march 17073Generate code that runs on @var{arch}, which can be the name of a 17074generic MIPS ISA, or the name of a particular processor. 17075The ISA names are: 17076@samp{mips1}, @samp{mips2}, @samp{mips3}, @samp{mips4}, 17077@samp{mips32}, @samp{mips32r2}, @samp{mips64} and @samp{mips64r2}. 17078The processor names are: 17079@samp{4kc}, @samp{4km}, @samp{4kp}, @samp{4ksc}, 17080@samp{4kec}, @samp{4kem}, @samp{4kep}, @samp{4ksd}, 17081@samp{5kc}, @samp{5kf}, 17082@samp{20kc}, 17083@samp{24kc}, @samp{24kf2_1}, @samp{24kf1_1}, 17084@samp{24kec}, @samp{24kef2_1}, @samp{24kef1_1}, 17085@samp{34kc}, @samp{34kf2_1}, @samp{34kf1_1}, @samp{34kn}, 17086@samp{74kc}, @samp{74kf2_1}, @samp{74kf1_1}, @samp{74kf3_2}, 17087@samp{1004kc}, @samp{1004kf2_1}, @samp{1004kf1_1}, 17088@samp{loongson2e}, @samp{loongson2f}, @samp{loongson3a}, 17089@samp{m4k}, 17090@samp{m14k}, @samp{m14kc}, @samp{m14ke}, @samp{m14kec}, 17091@samp{octeon}, @samp{octeon+}, @samp{octeon2}, 17092@samp{orion}, 17093@samp{r2000}, @samp{r3000}, @samp{r3900}, @samp{r4000}, @samp{r4400}, 17094@samp{r4600}, @samp{r4650}, @samp{r4700}, @samp{r6000}, @samp{r8000}, 17095@samp{rm7000}, @samp{rm9000}, 17096@samp{r10000}, @samp{r12000}, @samp{r14000}, @samp{r16000}, 17097@samp{sb1}, 17098@samp{sr71000}, 17099@samp{vr4100}, @samp{vr4111}, @samp{vr4120}, @samp{vr4130}, @samp{vr4300}, 17100@samp{vr5000}, @samp{vr5400}, @samp{vr5500}, 17101@samp{xlr} and @samp{xlp}. 17102The special value @samp{from-abi} selects the 17103most compatible architecture for the selected ABI (that is, 17104@samp{mips1} for 32-bit ABIs and @samp{mips3} for 64-bit ABIs)@. 17105 17106The native Linux/GNU toolchain also supports the value @samp{native}, 17107which selects the best architecture option for the host processor. 17108@option{-march=native} has no effect if GCC does not recognize 17109the processor. 17110 17111In processor names, a final @samp{000} can be abbreviated as @samp{k} 17112(for example, @option{-march=r2k}). Prefixes are optional, and 17113@samp{vr} may be written @samp{r}. 17114 17115Names of the form @samp{@var{n}f2_1} refer to processors with 17116FPUs clocked at half the rate of the core, names of the form 17117@samp{@var{n}f1_1} refer to processors with FPUs clocked at the same 17118rate as the core, and names of the form @samp{@var{n}f3_2} refer to 17119processors with FPUs clocked a ratio of 3:2 with respect to the core. 17120For compatibility reasons, @samp{@var{n}f} is accepted as a synonym 17121for @samp{@var{n}f2_1} while @samp{@var{n}x} and @samp{@var{b}fx} are 17122accepted as synonyms for @samp{@var{n}f1_1}. 17123 17124GCC defines two macros based on the value of this option. The first 17125is @samp{_MIPS_ARCH}, which gives the name of target architecture, as 17126a string. The second has the form @samp{_MIPS_ARCH_@var{foo}}, 17127where @var{foo} is the capitalized value of @samp{_MIPS_ARCH}@. 17128For example, @option{-march=r2000} sets @samp{_MIPS_ARCH} 17129to @samp{"r2000"} and defines the macro @samp{_MIPS_ARCH_R2000}. 17130 17131Note that the @samp{_MIPS_ARCH} macro uses the processor names given 17132above. In other words, it has the full prefix and does not 17133abbreviate @samp{000} as @samp{k}. In the case of @samp{from-abi}, 17134the macro names the resolved architecture (either @samp{"mips1"} or 17135@samp{"mips3"}). It names the default architecture when no 17136@option{-march} option is given. 17137 17138@item -mtune=@var{arch} 17139@opindex mtune 17140Optimize for @var{arch}. Among other things, this option controls 17141the way instructions are scheduled, and the perceived cost of arithmetic 17142operations. The list of @var{arch} values is the same as for 17143@option{-march}. 17144 17145When this option is not used, GCC optimizes for the processor 17146specified by @option{-march}. By using @option{-march} and 17147@option{-mtune} together, it is possible to generate code that 17148runs on a family of processors, but optimize the code for one 17149particular member of that family. 17150 17151@option{-mtune} defines the macros @samp{_MIPS_TUNE} and 17152@samp{_MIPS_TUNE_@var{foo}}, which work in the same way as the 17153@option{-march} ones described above. 17154 17155@item -mips1 17156@opindex mips1 17157Equivalent to @option{-march=mips1}. 17158 17159@item -mips2 17160@opindex mips2 17161Equivalent to @option{-march=mips2}. 17162 17163@item -mips3 17164@opindex mips3 17165Equivalent to @option{-march=mips3}. 17166 17167@item -mips4 17168@opindex mips4 17169Equivalent to @option{-march=mips4}. 17170 17171@item -mips32 17172@opindex mips32 17173Equivalent to @option{-march=mips32}. 17174 17175@item -mips32r2 17176@opindex mips32r2 17177Equivalent to @option{-march=mips32r2}. 17178 17179@item -mips64 17180@opindex mips64 17181Equivalent to @option{-march=mips64}. 17182 17183@item -mips64r2 17184@opindex mips64r2 17185Equivalent to @option{-march=mips64r2}. 17186 17187@item -mips16 17188@itemx -mno-mips16 17189@opindex mips16 17190@opindex mno-mips16 17191Generate (do not generate) MIPS16 code. If GCC is targeting a 17192MIPS32 or MIPS64 architecture, it makes use of the MIPS16e ASE@. 17193 17194MIPS16 code generation can also be controlled on a per-function basis 17195by means of @code{mips16} and @code{nomips16} attributes. 17196@xref{Function Attributes}, for more information. 17197 17198@item -mflip-mips16 17199@opindex mflip-mips16 17200Generate MIPS16 code on alternating functions. This option is provided 17201for regression testing of mixed MIPS16/non-MIPS16 code generation, and is 17202not intended for ordinary use in compiling user code. 17203 17204@item -minterlink-compressed 17205@item -mno-interlink-compressed 17206@opindex minterlink-compressed 17207@opindex mno-interlink-compressed 17208Require (do not require) that code using the standard (uncompressed) MIPS ISA 17209be link-compatible with MIPS16 and microMIPS code, and vice versa. 17210 17211For example, code using the standard ISA encoding cannot jump directly 17212to MIPS16 or microMIPS code; it must either use a call or an indirect jump. 17213@option{-minterlink-compressed} therefore disables direct jumps unless GCC 17214knows that the target of the jump is not compressed. 17215 17216@item -minterlink-mips16 17217@itemx -mno-interlink-mips16 17218@opindex minterlink-mips16 17219@opindex mno-interlink-mips16 17220Aliases of @option{-minterlink-compressed} and 17221@option{-mno-interlink-compressed}. These options predate the microMIPS ASE 17222and are retained for backwards compatibility. 17223 17224@item -mabi=32 17225@itemx -mabi=o64 17226@itemx -mabi=n32 17227@itemx -mabi=64 17228@itemx -mabi=eabi 17229@opindex mabi=32 17230@opindex mabi=o64 17231@opindex mabi=n32 17232@opindex mabi=64 17233@opindex mabi=eabi 17234Generate code for the given ABI@. 17235 17236Note that the EABI has a 32-bit and a 64-bit variant. GCC normally 17237generates 64-bit code when you select a 64-bit architecture, but you 17238can use @option{-mgp32} to get 32-bit code instead. 17239 17240For information about the O64 ABI, see 17241@uref{http://gcc.gnu.org/@/projects/@/mipso64-abi.html}. 17242 17243GCC supports a variant of the o32 ABI in which floating-point registers 17244are 64 rather than 32 bits wide. You can select this combination with 17245@option{-mabi=32} @option{-mfp64}. This ABI relies on the @code{mthc1} 17246and @code{mfhc1} instructions and is therefore only supported for 17247MIPS32R2 processors. 17248 17249The register assignments for arguments and return values remain the 17250same, but each scalar value is passed in a single 64-bit register 17251rather than a pair of 32-bit registers. For example, scalar 17252floating-point values are returned in @samp{$f0} only, not a 17253@samp{$f0}/@samp{$f1} pair. The set of call-saved registers also 17254remains the same, but all 64 bits are saved. 17255 17256@item -mabicalls 17257@itemx -mno-abicalls 17258@opindex mabicalls 17259@opindex mno-abicalls 17260Generate (do not generate) code that is suitable for SVR4-style 17261dynamic objects. @option{-mabicalls} is the default for SVR4-based 17262systems. 17263 17264@item -mshared 17265@itemx -mno-shared 17266Generate (do not generate) code that is fully position-independent, 17267and that can therefore be linked into shared libraries. This option 17268only affects @option{-mabicalls}. 17269 17270All @option{-mabicalls} code has traditionally been position-independent, 17271regardless of options like @option{-fPIC} and @option{-fpic}. However, 17272as an extension, the GNU toolchain allows executables to use absolute 17273accesses for locally-binding symbols. It can also use shorter GP 17274initialization sequences and generate direct calls to locally-defined 17275functions. This mode is selected by @option{-mno-shared}. 17276 17277@option{-mno-shared} depends on binutils 2.16 or higher and generates 17278objects that can only be linked by the GNU linker. However, the option 17279does not affect the ABI of the final executable; it only affects the ABI 17280of relocatable objects. Using @option{-mno-shared} generally makes 17281executables both smaller and quicker. 17282 17283@option{-mshared} is the default. 17284 17285@item -mplt 17286@itemx -mno-plt 17287@opindex mplt 17288@opindex mno-plt 17289Assume (do not assume) that the static and dynamic linkers 17290support PLTs and copy relocations. This option only affects 17291@option{-mno-shared -mabicalls}. For the n64 ABI, this option 17292has no effect without @option{-msym32}. 17293 17294You can make @option{-mplt} the default by configuring 17295GCC with @option{--with-mips-plt}. The default is 17296@option{-mno-plt} otherwise. 17297 17298@item -mxgot 17299@itemx -mno-xgot 17300@opindex mxgot 17301@opindex mno-xgot 17302Lift (do not lift) the usual restrictions on the size of the global 17303offset table. 17304 17305GCC normally uses a single instruction to load values from the GOT@. 17306While this is relatively efficient, it only works if the GOT 17307is smaller than about 64k. Anything larger causes the linker 17308to report an error such as: 17309 17310@cindex relocation truncated to fit (MIPS) 17311@smallexample 17312relocation truncated to fit: R_MIPS_GOT16 foobar 17313@end smallexample 17314 17315If this happens, you should recompile your code with @option{-mxgot}. 17316This works with very large GOTs, although the code is also 17317less efficient, since it takes three instructions to fetch the 17318value of a global symbol. 17319 17320Note that some linkers can create multiple GOTs. If you have such a 17321linker, you should only need to use @option{-mxgot} when a single object 17322file accesses more than 64k's worth of GOT entries. Very few do. 17323 17324These options have no effect unless GCC is generating position 17325independent code. 17326 17327@item -mgp32 17328@opindex mgp32 17329Assume that general-purpose registers are 32 bits wide. 17330 17331@item -mgp64 17332@opindex mgp64 17333Assume that general-purpose registers are 64 bits wide. 17334 17335@item -mfp32 17336@opindex mfp32 17337Assume that floating-point registers are 32 bits wide. 17338 17339@item -mfp64 17340@opindex mfp64 17341Assume that floating-point registers are 64 bits wide. 17342 17343@item -mhard-float 17344@opindex mhard-float 17345Use floating-point coprocessor instructions. 17346 17347@item -msoft-float 17348@opindex msoft-float 17349Do not use floating-point coprocessor instructions. Implement 17350floating-point calculations using library calls instead. 17351 17352@item -mno-float 17353@opindex mno-float 17354Equivalent to @option{-msoft-float}, but additionally asserts that the 17355program being compiled does not perform any floating-point operations. 17356This option is presently supported only by some bare-metal MIPS 17357configurations, where it may select a special set of libraries 17358that lack all floating-point support (including, for example, the 17359floating-point @code{printf} formats). 17360If code compiled with @code{-mno-float} accidentally contains 17361floating-point operations, it is likely to suffer a link-time 17362or run-time failure. 17363 17364@item -msingle-float 17365@opindex msingle-float 17366Assume that the floating-point coprocessor only supports single-precision 17367operations. 17368 17369@item -mdouble-float 17370@opindex mdouble-float 17371Assume that the floating-point coprocessor supports double-precision 17372operations. This is the default. 17373 17374@item -mabs=2008 17375@itemx -mabs=legacy 17376@opindex mabs=2008 17377@opindex mabs=legacy 17378These options control the treatment of the special not-a-number (NaN) 17379IEEE 754 floating-point data with the @code{abs.@i{fmt}} and 17380@code{neg.@i{fmt}} machine instructions. 17381 17382By default or when the @option{-mabs=legacy} is used the legacy 17383treatment is selected. In this case these instructions are considered 17384arithmetic and avoided where correct operation is required and the 17385input operand might be a NaN. A longer sequence of instructions that 17386manipulate the sign bit of floating-point datum manually is used 17387instead unless the @option{-ffinite-math-only} option has also been 17388specified. 17389 17390The @option{-mabs=2008} option selects the IEEE 754-2008 treatment. In 17391this case these instructions are considered non-arithmetic and therefore 17392operating correctly in all cases, including in particular where the 17393input operand is a NaN. These instructions are therefore always used 17394for the respective operations. 17395 17396@item -mnan=2008 17397@itemx -mnan=legacy 17398@opindex mnan=2008 17399@opindex mnan=legacy 17400These options control the encoding of the special not-a-number (NaN) 17401IEEE 754 floating-point data. 17402 17403The @option{-mnan=legacy} option selects the legacy encoding. In this 17404case quiet NaNs (qNaNs) are denoted by the first bit of their trailing 17405significand field being 0, whereas signalling NaNs (sNaNs) are denoted 17406by the first bit of their trailing significand field being 1. 17407 17408The @option{-mnan=2008} option selects the IEEE 754-2008 encoding. In 17409this case qNaNs are denoted by the first bit of their trailing 17410significand field being 1, whereas sNaNs are denoted by the first bit of 17411their trailing significand field being 0. 17412 17413The default is @option{-mnan=legacy} unless GCC has been configured with 17414@option{--with-nan=2008}. 17415 17416@item -mllsc 17417@itemx -mno-llsc 17418@opindex mllsc 17419@opindex mno-llsc 17420Use (do not use) @samp{ll}, @samp{sc}, and @samp{sync} instructions to 17421implement atomic memory built-in functions. When neither option is 17422specified, GCC uses the instructions if the target architecture 17423supports them. 17424 17425@option{-mllsc} is useful if the runtime environment can emulate the 17426instructions and @option{-mno-llsc} can be useful when compiling for 17427nonstandard ISAs. You can make either option the default by 17428configuring GCC with @option{--with-llsc} and @option{--without-llsc} 17429respectively. @option{--with-llsc} is the default for some 17430configurations; see the installation documentation for details. 17431 17432@item -mdsp 17433@itemx -mno-dsp 17434@opindex mdsp 17435@opindex mno-dsp 17436Use (do not use) revision 1 of the MIPS DSP ASE@. 17437@xref{MIPS DSP Built-in Functions}. This option defines the 17438preprocessor macro @samp{__mips_dsp}. It also defines 17439@samp{__mips_dsp_rev} to 1. 17440 17441@item -mdspr2 17442@itemx -mno-dspr2 17443@opindex mdspr2 17444@opindex mno-dspr2 17445Use (do not use) revision 2 of the MIPS DSP ASE@. 17446@xref{MIPS DSP Built-in Functions}. This option defines the 17447preprocessor macros @samp{__mips_dsp} and @samp{__mips_dspr2}. 17448It also defines @samp{__mips_dsp_rev} to 2. 17449 17450@item -msmartmips 17451@itemx -mno-smartmips 17452@opindex msmartmips 17453@opindex mno-smartmips 17454Use (do not use) the MIPS SmartMIPS ASE. 17455 17456@item -mpaired-single 17457@itemx -mno-paired-single 17458@opindex mpaired-single 17459@opindex mno-paired-single 17460Use (do not use) paired-single floating-point instructions. 17461@xref{MIPS Paired-Single Support}. This option requires 17462hardware floating-point support to be enabled. 17463 17464@item -mdmx 17465@itemx -mno-mdmx 17466@opindex mdmx 17467@opindex mno-mdmx 17468Use (do not use) MIPS Digital Media Extension instructions. 17469This option can only be used when generating 64-bit code and requires 17470hardware floating-point support to be enabled. 17471 17472@item -mips3d 17473@itemx -mno-mips3d 17474@opindex mips3d 17475@opindex mno-mips3d 17476Use (do not use) the MIPS-3D ASE@. @xref{MIPS-3D Built-in Functions}. 17477The option @option{-mips3d} implies @option{-mpaired-single}. 17478 17479@item -mmicromips 17480@itemx -mno-micromips 17481@opindex mmicromips 17482@opindex mno-mmicromips 17483Generate (do not generate) microMIPS code. 17484 17485MicroMIPS code generation can also be controlled on a per-function basis 17486by means of @code{micromips} and @code{nomicromips} attributes. 17487@xref{Function Attributes}, for more information. 17488 17489@item -mmt 17490@itemx -mno-mt 17491@opindex mmt 17492@opindex mno-mt 17493Use (do not use) MT Multithreading instructions. 17494 17495@item -mmcu 17496@itemx -mno-mcu 17497@opindex mmcu 17498@opindex mno-mcu 17499Use (do not use) the MIPS MCU ASE instructions. 17500 17501@item -meva 17502@itemx -mno-eva 17503@opindex meva 17504@opindex mno-eva 17505Use (do not use) the MIPS Enhanced Virtual Addressing instructions. 17506 17507@item -mvirt 17508@itemx -mno-virt 17509@opindex mvirt 17510@opindex mno-virt 17511Use (do not use) the MIPS Virtualization Application Specific instructions. 17512 17513@item -mlong64 17514@opindex mlong64 17515Force @code{long} types to be 64 bits wide. See @option{-mlong32} for 17516an explanation of the default and the way that the pointer size is 17517determined. 17518 17519@item -mlong32 17520@opindex mlong32 17521Force @code{long}, @code{int}, and pointer types to be 32 bits wide. 17522 17523The default size of @code{int}s, @code{long}s and pointers depends on 17524the ABI@. All the supported ABIs use 32-bit @code{int}s. The n64 ABI 17525uses 64-bit @code{long}s, as does the 64-bit EABI; the others use 1752632-bit @code{long}s. Pointers are the same size as @code{long}s, 17527or the same size as integer registers, whichever is smaller. 17528 17529@item -msym32 17530@itemx -mno-sym32 17531@opindex msym32 17532@opindex mno-sym32 17533Assume (do not assume) that all symbols have 32-bit values, regardless 17534of the selected ABI@. This option is useful in combination with 17535@option{-mabi=64} and @option{-mno-abicalls} because it allows GCC 17536to generate shorter and faster references to symbolic addresses. 17537 17538@item -G @var{num} 17539@opindex G 17540Put definitions of externally-visible data in a small data section 17541if that data is no bigger than @var{num} bytes. GCC can then generate 17542more efficient accesses to the data; see @option{-mgpopt} for details. 17543 17544The default @option{-G} option depends on the configuration. 17545 17546@item -mlocal-sdata 17547@itemx -mno-local-sdata 17548@opindex mlocal-sdata 17549@opindex mno-local-sdata 17550Extend (do not extend) the @option{-G} behavior to local data too, 17551such as to static variables in C@. @option{-mlocal-sdata} is the 17552default for all configurations. 17553 17554If the linker complains that an application is using too much small data, 17555you might want to try rebuilding the less performance-critical parts with 17556@option{-mno-local-sdata}. You might also want to build large 17557libraries with @option{-mno-local-sdata}, so that the libraries leave 17558more room for the main program. 17559 17560@item -mextern-sdata 17561@itemx -mno-extern-sdata 17562@opindex mextern-sdata 17563@opindex mno-extern-sdata 17564Assume (do not assume) that externally-defined data is in 17565a small data section if the size of that data is within the @option{-G} limit. 17566@option{-mextern-sdata} is the default for all configurations. 17567 17568If you compile a module @var{Mod} with @option{-mextern-sdata} @option{-G 17569@var{num}} @option{-mgpopt}, and @var{Mod} references a variable @var{Var} 17570that is no bigger than @var{num} bytes, you must make sure that @var{Var} 17571is placed in a small data section. If @var{Var} is defined by another 17572module, you must either compile that module with a high-enough 17573@option{-G} setting or attach a @code{section} attribute to @var{Var}'s 17574definition. If @var{Var} is common, you must link the application 17575with a high-enough @option{-G} setting. 17576 17577The easiest way of satisfying these restrictions is to compile 17578and link every module with the same @option{-G} option. However, 17579you may wish to build a library that supports several different 17580small data limits. You can do this by compiling the library with 17581the highest supported @option{-G} setting and additionally using 17582@option{-mno-extern-sdata} to stop the library from making assumptions 17583about externally-defined data. 17584 17585@item -mgpopt 17586@itemx -mno-gpopt 17587@opindex mgpopt 17588@opindex mno-gpopt 17589Use (do not use) GP-relative accesses for symbols that are known to be 17590in a small data section; see @option{-G}, @option{-mlocal-sdata} and 17591@option{-mextern-sdata}. @option{-mgpopt} is the default for all 17592configurations. 17593 17594@option{-mno-gpopt} is useful for cases where the @code{$gp} register 17595might not hold the value of @code{_gp}. For example, if the code is 17596part of a library that might be used in a boot monitor, programs that 17597call boot monitor routines pass an unknown value in @code{$gp}. 17598(In such situations, the boot monitor itself is usually compiled 17599with @option{-G0}.) 17600 17601@option{-mno-gpopt} implies @option{-mno-local-sdata} and 17602@option{-mno-extern-sdata}. 17603 17604@item -membedded-data 17605@itemx -mno-embedded-data 17606@opindex membedded-data 17607@opindex mno-embedded-data 17608Allocate variables to the read-only data section first if possible, then 17609next in the small data section if possible, otherwise in data. This gives 17610slightly slower code than the default, but reduces the amount of RAM required 17611when executing, and thus may be preferred for some embedded systems. 17612 17613@item -muninit-const-in-rodata 17614@itemx -mno-uninit-const-in-rodata 17615@opindex muninit-const-in-rodata 17616@opindex mno-uninit-const-in-rodata 17617Put uninitialized @code{const} variables in the read-only data section. 17618This option is only meaningful in conjunction with @option{-membedded-data}. 17619 17620@item -mcode-readable=@var{setting} 17621@opindex mcode-readable 17622Specify whether GCC may generate code that reads from executable sections. 17623There are three possible settings: 17624 17625@table @gcctabopt 17626@item -mcode-readable=yes 17627Instructions may freely access executable sections. This is the 17628default setting. 17629 17630@item -mcode-readable=pcrel 17631MIPS16 PC-relative load instructions can access executable sections, 17632but other instructions must not do so. This option is useful on 4KSc 17633and 4KSd processors when the code TLBs have the Read Inhibit bit set. 17634It is also useful on processors that can be configured to have a dual 17635instruction/data SRAM interface and that, like the M4K, automatically 17636redirect PC-relative loads to the instruction RAM. 17637 17638@item -mcode-readable=no 17639Instructions must not access executable sections. This option can be 17640useful on targets that are configured to have a dual instruction/data 17641SRAM interface but that (unlike the M4K) do not automatically redirect 17642PC-relative loads to the instruction RAM. 17643@end table 17644 17645@item -msplit-addresses 17646@itemx -mno-split-addresses 17647@opindex msplit-addresses 17648@opindex mno-split-addresses 17649Enable (disable) use of the @code{%hi()} and @code{%lo()} assembler 17650relocation operators. This option has been superseded by 17651@option{-mexplicit-relocs} but is retained for backwards compatibility. 17652 17653@item -mexplicit-relocs 17654@itemx -mno-explicit-relocs 17655@opindex mexplicit-relocs 17656@opindex mno-explicit-relocs 17657Use (do not use) assembler relocation operators when dealing with symbolic 17658addresses. The alternative, selected by @option{-mno-explicit-relocs}, 17659is to use assembler macros instead. 17660 17661@option{-mexplicit-relocs} is the default if GCC was configured 17662to use an assembler that supports relocation operators. 17663 17664@item -mcheck-zero-division 17665@itemx -mno-check-zero-division 17666@opindex mcheck-zero-division 17667@opindex mno-check-zero-division 17668Trap (do not trap) on integer division by zero. 17669 17670The default is @option{-mcheck-zero-division}. 17671 17672@item -mdivide-traps 17673@itemx -mdivide-breaks 17674@opindex mdivide-traps 17675@opindex mdivide-breaks 17676MIPS systems check for division by zero by generating either a 17677conditional trap or a break instruction. Using traps results in 17678smaller code, but is only supported on MIPS II and later. Also, some 17679versions of the Linux kernel have a bug that prevents trap from 17680generating the proper signal (@code{SIGFPE}). Use @option{-mdivide-traps} to 17681allow conditional traps on architectures that support them and 17682@option{-mdivide-breaks} to force the use of breaks. 17683 17684The default is usually @option{-mdivide-traps}, but this can be 17685overridden at configure time using @option{--with-divide=breaks}. 17686Divide-by-zero checks can be completely disabled using 17687@option{-mno-check-zero-division}. 17688 17689@item -mmemcpy 17690@itemx -mno-memcpy 17691@opindex mmemcpy 17692@opindex mno-memcpy 17693Force (do not force) the use of @code{memcpy()} for non-trivial block 17694moves. The default is @option{-mno-memcpy}, which allows GCC to inline 17695most constant-sized copies. 17696 17697@item -mlong-calls 17698@itemx -mno-long-calls 17699@opindex mlong-calls 17700@opindex mno-long-calls 17701Disable (do not disable) use of the @code{jal} instruction. Calling 17702functions using @code{jal} is more efficient but requires the caller 17703and callee to be in the same 256 megabyte segment. 17704 17705This option has no effect on abicalls code. The default is 17706@option{-mno-long-calls}. 17707 17708@item -mmad 17709@itemx -mno-mad 17710@opindex mmad 17711@opindex mno-mad 17712Enable (disable) use of the @code{mad}, @code{madu} and @code{mul} 17713instructions, as provided by the R4650 ISA@. 17714 17715@item -mimadd 17716@itemx -mno-imadd 17717@opindex mimadd 17718@opindex mno-imadd 17719Enable (disable) use of the @code{madd} and @code{msub} integer 17720instructions. The default is @option{-mimadd} on architectures 17721that support @code{madd} and @code{msub} except for the 74k 17722architecture where it was found to generate slower code. 17723 17724@item -mfused-madd 17725@itemx -mno-fused-madd 17726@opindex mfused-madd 17727@opindex mno-fused-madd 17728Enable (disable) use of the floating-point multiply-accumulate 17729instructions, when they are available. The default is 17730@option{-mfused-madd}. 17731 17732On the R8000 CPU when multiply-accumulate instructions are used, 17733the intermediate product is calculated to infinite precision 17734and is not subject to the FCSR Flush to Zero bit. This may be 17735undesirable in some circumstances. On other processors the result 17736is numerically identical to the equivalent computation using 17737separate multiply, add, subtract and negate instructions. 17738 17739@item -nocpp 17740@opindex nocpp 17741Tell the MIPS assembler to not run its preprocessor over user 17742assembler files (with a @samp{.s} suffix) when assembling them. 17743 17744@item -mfix-24k 17745@item -mno-fix-24k 17746@opindex mfix-24k 17747@opindex mno-fix-24k 17748Work around the 24K E48 (lost data on stores during refill) errata. 17749The workarounds are implemented by the assembler rather than by GCC@. 17750 17751@item -mfix-r4000 17752@itemx -mno-fix-r4000 17753@opindex mfix-r4000 17754@opindex mno-fix-r4000 17755Work around certain R4000 CPU errata: 17756@itemize @minus 17757@item 17758A double-word or a variable shift may give an incorrect result if executed 17759immediately after starting an integer division. 17760@item 17761A double-word or a variable shift may give an incorrect result if executed 17762while an integer multiplication is in progress. 17763@item 17764An integer division may give an incorrect result if started in a delay slot 17765of a taken branch or a jump. 17766@end itemize 17767 17768@item -mfix-r4400 17769@itemx -mno-fix-r4400 17770@opindex mfix-r4400 17771@opindex mno-fix-r4400 17772Work around certain R4400 CPU errata: 17773@itemize @minus 17774@item 17775A double-word or a variable shift may give an incorrect result if executed 17776immediately after starting an integer division. 17777@end itemize 17778 17779@item -mfix-r10000 17780@itemx -mno-fix-r10000 17781@opindex mfix-r10000 17782@opindex mno-fix-r10000 17783Work around certain R10000 errata: 17784@itemize @minus 17785@item 17786@code{ll}/@code{sc} sequences may not behave atomically on revisions 17787prior to 3.0. They may deadlock on revisions 2.6 and earlier. 17788@end itemize 17789 17790This option can only be used if the target architecture supports 17791branch-likely instructions. @option{-mfix-r10000} is the default when 17792@option{-march=r10000} is used; @option{-mno-fix-r10000} is the default 17793otherwise. 17794 17795@item -mfix-rm7000 17796@itemx -mno-fix-rm7000 17797@opindex mfix-rm7000 17798Work around the RM7000 @code{dmult}/@code{dmultu} errata. The 17799workarounds are implemented by the assembler rather than by GCC@. 17800 17801@item -mfix-vr4120 17802@itemx -mno-fix-vr4120 17803@opindex mfix-vr4120 17804Work around certain VR4120 errata: 17805@itemize @minus 17806@item 17807@code{dmultu} does not always produce the correct result. 17808@item 17809@code{div} and @code{ddiv} do not always produce the correct result if one 17810of the operands is negative. 17811@end itemize 17812The workarounds for the division errata rely on special functions in 17813@file{libgcc.a}. At present, these functions are only provided by 17814the @code{mips64vr*-elf} configurations. 17815 17816Other VR4120 errata require a NOP to be inserted between certain pairs of 17817instructions. These errata are handled by the assembler, not by GCC itself. 17818 17819@item -mfix-vr4130 17820@opindex mfix-vr4130 17821Work around the VR4130 @code{mflo}/@code{mfhi} errata. The 17822workarounds are implemented by the assembler rather than by GCC, 17823although GCC avoids using @code{mflo} and @code{mfhi} if the 17824VR4130 @code{macc}, @code{macchi}, @code{dmacc} and @code{dmacchi} 17825instructions are available instead. 17826 17827@item -mfix-sb1 17828@itemx -mno-fix-sb1 17829@opindex mfix-sb1 17830Work around certain SB-1 CPU core errata. 17831(This flag currently works around the SB-1 revision 2 17832``F1'' and ``F2'' floating-point errata.) 17833 17834@item -mr10k-cache-barrier=@var{setting} 17835@opindex mr10k-cache-barrier 17836Specify whether GCC should insert cache barriers to avoid the 17837side-effects of speculation on R10K processors. 17838 17839In common with many processors, the R10K tries to predict the outcome 17840of a conditional branch and speculatively executes instructions from 17841the ``taken'' branch. It later aborts these instructions if the 17842predicted outcome is wrong. However, on the R10K, even aborted 17843instructions can have side effects. 17844 17845This problem only affects kernel stores and, depending on the system, 17846kernel loads. As an example, a speculatively-executed store may load 17847the target memory into cache and mark the cache line as dirty, even if 17848the store itself is later aborted. If a DMA operation writes to the 17849same area of memory before the ``dirty'' line is flushed, the cached 17850data overwrites the DMA-ed data. See the R10K processor manual 17851for a full description, including other potential problems. 17852 17853One workaround is to insert cache barrier instructions before every memory 17854access that might be speculatively executed and that might have side 17855effects even if aborted. @option{-mr10k-cache-barrier=@var{setting}} 17856controls GCC's implementation of this workaround. It assumes that 17857aborted accesses to any byte in the following regions does not have 17858side effects: 17859 17860@enumerate 17861@item 17862the memory occupied by the current function's stack frame; 17863 17864@item 17865the memory occupied by an incoming stack argument; 17866 17867@item 17868the memory occupied by an object with a link-time-constant address. 17869@end enumerate 17870 17871It is the kernel's responsibility to ensure that speculative 17872accesses to these regions are indeed safe. 17873 17874If the input program contains a function declaration such as: 17875 17876@smallexample 17877void foo (void); 17878@end smallexample 17879 17880then the implementation of @code{foo} must allow @code{j foo} and 17881@code{jal foo} to be executed speculatively. GCC honors this 17882restriction for functions it compiles itself. It expects non-GCC 17883functions (such as hand-written assembly code) to do the same. 17884 17885The option has three forms: 17886 17887@table @gcctabopt 17888@item -mr10k-cache-barrier=load-store 17889Insert a cache barrier before a load or store that might be 17890speculatively executed and that might have side effects even 17891if aborted. 17892 17893@item -mr10k-cache-barrier=store 17894Insert a cache barrier before a store that might be speculatively 17895executed and that might have side effects even if aborted. 17896 17897@item -mr10k-cache-barrier=none 17898Disable the insertion of cache barriers. This is the default setting. 17899@end table 17900 17901@item -mflush-func=@var{func} 17902@itemx -mno-flush-func 17903@opindex mflush-func 17904Specifies the function to call to flush the I and D caches, or to not 17905call any such function. If called, the function must take the same 17906arguments as the common @code{_flush_func()}, that is, the address of the 17907memory range for which the cache is being flushed, the size of the 17908memory range, and the number 3 (to flush both caches). The default 17909depends on the target GCC was configured for, but commonly is either 17910@samp{_flush_func} or @samp{__cpu_flush}. 17911 17912@item mbranch-cost=@var{num} 17913@opindex mbranch-cost 17914Set the cost of branches to roughly @var{num} ``simple'' instructions. 17915This cost is only a heuristic and is not guaranteed to produce 17916consistent results across releases. A zero cost redundantly selects 17917the default, which is based on the @option{-mtune} setting. 17918 17919@item -mbranch-likely 17920@itemx -mno-branch-likely 17921@opindex mbranch-likely 17922@opindex mno-branch-likely 17923Enable or disable use of Branch Likely instructions, regardless of the 17924default for the selected architecture. By default, Branch Likely 17925instructions may be generated if they are supported by the selected 17926architecture. An exception is for the MIPS32 and MIPS64 architectures 17927and processors that implement those architectures; for those, Branch 17928Likely instructions are not be generated by default because the MIPS32 17929and MIPS64 architectures specifically deprecate their use. 17930 17931@item -mfp-exceptions 17932@itemx -mno-fp-exceptions 17933@opindex mfp-exceptions 17934Specifies whether FP exceptions are enabled. This affects how 17935FP instructions are scheduled for some processors. 17936The default is that FP exceptions are 17937enabled. 17938 17939For instance, on the SB-1, if FP exceptions are disabled, and we are emitting 1794064-bit code, then we can use both FP pipes. Otherwise, we can only use one 17941FP pipe. 17942 17943@item -mvr4130-align 17944@itemx -mno-vr4130-align 17945@opindex mvr4130-align 17946The VR4130 pipeline is two-way superscalar, but can only issue two 17947instructions together if the first one is 8-byte aligned. When this 17948option is enabled, GCC aligns pairs of instructions that it 17949thinks should execute in parallel. 17950 17951This option only has an effect when optimizing for the VR4130. 17952It normally makes code faster, but at the expense of making it bigger. 17953It is enabled by default at optimization level @option{-O3}. 17954 17955@item -msynci 17956@itemx -mno-synci 17957@opindex msynci 17958Enable (disable) generation of @code{synci} instructions on 17959architectures that support it. The @code{synci} instructions (if 17960enabled) are generated when @code{__builtin___clear_cache()} is 17961compiled. 17962 17963This option defaults to @code{-mno-synci}, but the default can be 17964overridden by configuring with @code{--with-synci}. 17965 17966When compiling code for single processor systems, it is generally safe 17967to use @code{synci}. However, on many multi-core (SMP) systems, it 17968does not invalidate the instruction caches on all cores and may lead 17969to undefined behavior. 17970 17971@item -mrelax-pic-calls 17972@itemx -mno-relax-pic-calls 17973@opindex mrelax-pic-calls 17974Try to turn PIC calls that are normally dispatched via register 17975@code{$25} into direct calls. This is only possible if the linker can 17976resolve the destination at link-time and if the destination is within 17977range for a direct call. 17978 17979@option{-mrelax-pic-calls} is the default if GCC was configured to use 17980an assembler and a linker that support the @code{.reloc} assembly 17981directive and @code{-mexplicit-relocs} is in effect. With 17982@code{-mno-explicit-relocs}, this optimization can be performed by the 17983assembler and the linker alone without help from the compiler. 17984 17985@item -mmcount-ra-address 17986@itemx -mno-mcount-ra-address 17987@opindex mmcount-ra-address 17988@opindex mno-mcount-ra-address 17989Emit (do not emit) code that allows @code{_mcount} to modify the 17990calling function's return address. When enabled, this option extends 17991the usual @code{_mcount} interface with a new @var{ra-address} 17992parameter, which has type @code{intptr_t *} and is passed in register 17993@code{$12}. @code{_mcount} can then modify the return address by 17994doing both of the following: 17995@itemize 17996@item 17997Returning the new address in register @code{$31}. 17998@item 17999Storing the new address in @code{*@var{ra-address}}, 18000if @var{ra-address} is nonnull. 18001@end itemize 18002 18003The default is @option{-mno-mcount-ra-address}. 18004 18005@end table 18006 18007@node MMIX Options 18008@subsection MMIX Options 18009@cindex MMIX Options 18010 18011These options are defined for the MMIX: 18012 18013@table @gcctabopt 18014@item -mlibfuncs 18015@itemx -mno-libfuncs 18016@opindex mlibfuncs 18017@opindex mno-libfuncs 18018Specify that intrinsic library functions are being compiled, passing all 18019values in registers, no matter the size. 18020 18021@item -mepsilon 18022@itemx -mno-epsilon 18023@opindex mepsilon 18024@opindex mno-epsilon 18025Generate floating-point comparison instructions that compare with respect 18026to the @code{rE} epsilon register. 18027 18028@item -mabi=mmixware 18029@itemx -mabi=gnu 18030@opindex mabi=mmixware 18031@opindex mabi=gnu 18032Generate code that passes function parameters and return values that (in 18033the called function) are seen as registers @code{$0} and up, as opposed to 18034the GNU ABI which uses global registers @code{$231} and up. 18035 18036@item -mzero-extend 18037@itemx -mno-zero-extend 18038@opindex mzero-extend 18039@opindex mno-zero-extend 18040When reading data from memory in sizes shorter than 64 bits, use (do not 18041use) zero-extending load instructions by default, rather than 18042sign-extending ones. 18043 18044@item -mknuthdiv 18045@itemx -mno-knuthdiv 18046@opindex mknuthdiv 18047@opindex mno-knuthdiv 18048Make the result of a division yielding a remainder have the same sign as 18049the divisor. With the default, @option{-mno-knuthdiv}, the sign of the 18050remainder follows the sign of the dividend. Both methods are 18051arithmetically valid, the latter being almost exclusively used. 18052 18053@item -mtoplevel-symbols 18054@itemx -mno-toplevel-symbols 18055@opindex mtoplevel-symbols 18056@opindex mno-toplevel-symbols 18057Prepend (do not prepend) a @samp{:} to all global symbols, so the assembly 18058code can be used with the @code{PREFIX} assembly directive. 18059 18060@item -melf 18061@opindex melf 18062Generate an executable in the ELF format, rather than the default 18063@samp{mmo} format used by the @command{mmix} simulator. 18064 18065@item -mbranch-predict 18066@itemx -mno-branch-predict 18067@opindex mbranch-predict 18068@opindex mno-branch-predict 18069Use (do not use) the probable-branch instructions, when static branch 18070prediction indicates a probable branch. 18071 18072@item -mbase-addresses 18073@itemx -mno-base-addresses 18074@opindex mbase-addresses 18075@opindex mno-base-addresses 18076Generate (do not generate) code that uses @emph{base addresses}. Using a 18077base address automatically generates a request (handled by the assembler 18078and the linker) for a constant to be set up in a global register. The 18079register is used for one or more base address requests within the range 0 18080to 255 from the value held in the register. The generally leads to short 18081and fast code, but the number of different data items that can be 18082addressed is limited. This means that a program that uses lots of static 18083data may require @option{-mno-base-addresses}. 18084 18085@item -msingle-exit 18086@itemx -mno-single-exit 18087@opindex msingle-exit 18088@opindex mno-single-exit 18089Force (do not force) generated code to have a single exit point in each 18090function. 18091@end table 18092 18093@node MN10300 Options 18094@subsection MN10300 Options 18095@cindex MN10300 options 18096 18097These @option{-m} options are defined for Matsushita MN10300 architectures: 18098 18099@table @gcctabopt 18100@item -mmult-bug 18101@opindex mmult-bug 18102Generate code to avoid bugs in the multiply instructions for the MN10300 18103processors. This is the default. 18104 18105@item -mno-mult-bug 18106@opindex mno-mult-bug 18107Do not generate code to avoid bugs in the multiply instructions for the 18108MN10300 processors. 18109 18110@item -mam33 18111@opindex mam33 18112Generate code using features specific to the AM33 processor. 18113 18114@item -mno-am33 18115@opindex mno-am33 18116Do not generate code using features specific to the AM33 processor. This 18117is the default. 18118 18119@item -mam33-2 18120@opindex mam33-2 18121Generate code using features specific to the AM33/2.0 processor. 18122 18123@item -mam34 18124@opindex mam34 18125Generate code using features specific to the AM34 processor. 18126 18127@item -mtune=@var{cpu-type} 18128@opindex mtune 18129Use the timing characteristics of the indicated CPU type when 18130scheduling instructions. This does not change the targeted processor 18131type. The CPU type must be one of @samp{mn10300}, @samp{am33}, 18132@samp{am33-2} or @samp{am34}. 18133 18134@item -mreturn-pointer-on-d0 18135@opindex mreturn-pointer-on-d0 18136When generating a function that returns a pointer, return the pointer 18137in both @code{a0} and @code{d0}. Otherwise, the pointer is returned 18138only in @code{a0}, and attempts to call such functions without a prototype 18139result in errors. Note that this option is on by default; use 18140@option{-mno-return-pointer-on-d0} to disable it. 18141 18142@item -mno-crt0 18143@opindex mno-crt0 18144Do not link in the C run-time initialization object file. 18145 18146@item -mrelax 18147@opindex mrelax 18148Indicate to the linker that it should perform a relaxation optimization pass 18149to shorten branches, calls and absolute memory addresses. This option only 18150has an effect when used on the command line for the final link step. 18151 18152This option makes symbolic debugging impossible. 18153 18154@item -mliw 18155@opindex mliw 18156Allow the compiler to generate @emph{Long Instruction Word} 18157instructions if the target is the @samp{AM33} or later. This is the 18158default. This option defines the preprocessor macro @samp{__LIW__}. 18159 18160@item -mnoliw 18161@opindex mnoliw 18162Do not allow the compiler to generate @emph{Long Instruction Word} 18163instructions. This option defines the preprocessor macro 18164@samp{__NO_LIW__}. 18165 18166@item -msetlb 18167@opindex msetlb 18168Allow the compiler to generate the @emph{SETLB} and @emph{Lcc} 18169instructions if the target is the @samp{AM33} or later. This is the 18170default. This option defines the preprocessor macro @samp{__SETLB__}. 18171 18172@item -mnosetlb 18173@opindex mnosetlb 18174Do not allow the compiler to generate @emph{SETLB} or @emph{Lcc} 18175instructions. This option defines the preprocessor macro 18176@samp{__NO_SETLB__}. 18177 18178@end table 18179 18180@node Moxie Options 18181@subsection Moxie Options 18182@cindex Moxie Options 18183 18184@table @gcctabopt 18185 18186@item -meb 18187@opindex meb 18188Generate big-endian code. This is the default for @samp{moxie-*-*} 18189configurations. 18190 18191@item -mel 18192@opindex mel 18193Generate little-endian code. 18194 18195@item -mno-crt0 18196@opindex mno-crt0 18197Do not link in the C run-time initialization object file. 18198 18199@end table 18200 18201@node MSP430 Options 18202@subsection MSP430 Options 18203@cindex MSP430 Options 18204 18205These options are defined for the MSP430: 18206 18207@table @gcctabopt 18208 18209@item -masm-hex 18210@opindex masm-hex 18211Force assembly output to always use hex constants. Normally such 18212constants are signed decimals, but this option is available for 18213testsuite and/or aesthetic purposes. 18214 18215@item -mmcu= 18216@opindex mmcu= 18217Select the MCU to target. This is used to create a C preprocessor 18218symbol based upon the MCU name, converted to upper case and pre- and 18219post- fixed with @code{__}. This in turn will be used by the 18220@code{msp430.h} header file to select an MCU specific supplimentary 18221header file. 18222 18223The option also sets the ISA to use. If the MCU name is one that is 18224known to only support the 430 ISA then that is selected, otherwise the 18225430X ISA is selected. A generic MCU name of @code{msp430} can also be 18226used to select the 430 ISA. Similarly the generic @code{msp430x} MCU 18227name will select the 430X ISA. 18228 18229In addition an MCU specific linker script will be added to the linker 18230command line. The script's name is the name of the MCU with 18231@code{.ld} appended. Thus specifying @option{-mmcu=xxx} on the gcc 18232command line will define the C preprocessor symbol @code{__XXX__} and 18233cause the linker to search for a script called @file{xxx.ld}. 18234 18235This option is also passed on to the assembler. 18236 18237@item -mcpu= 18238@opindex -mcpu= 18239Specifies the ISA to use. Accepted values are @code{msp430}, 18240@code{msp430x} and @code{msp430xv2}. This option is deprecated. The 18241@option{-mmcu=} option should be used to select the ISA. 18242 18243@item -msim 18244@opindex msim 18245Link to the simulator runtime libraries and linker script. Overrides 18246any scripts that would be selected by the @option{-mmcu=} option. 18247 18248@item -mlarge 18249@opindex mlarge 18250Use large-model addressing (20-bit pointers, 32-bit @code{size_t}). 18251 18252@item -msmall 18253@opindex msmall 18254Use small-model addressing (16-bit pointers, 16-bit @code{size_t}). 18255 18256@item -mrelax 18257@opindex mrelax 18258This option is passed to the assembler and linker, and allows the 18259linker to perform certain optimizations that cannot be done until 18260the final link. 18261 18262@item mhwmult= 18263@opindex mhwmult= 18264Describes the type of hardware multiply supported by the target. 18265Accepted values are @code{none} for no hardware multiply, @code{16bit} 18266for the original 16-bit-only multiply supported by early MCUs. 18267@code{32bit} for the 16/32-bit multiply supported by later MCUs and 18268@code{f5series} for the 16/32-bit multiply supported by F5-series MCUs. 18269A value of @code{auto} can also be given. This tells GCC to deduce 18270the hardware multiply support based upon the MCU name provided by the 18271@option{-mmcu} option. If no @option{-mmcu} option is specified then 18272@code{32bit} hardware multiply support is assumed. @code{auto} is the 18273default setting. 18274 18275Hardware multiplies are normally performed by calling a library 18276routine. This saves space in the generated code. When compiling at 18277@code{-O3} or higher however the hardware multiplier is invoked 18278inline. This makes for bigger, but faster code. 18279 18280The hardware multiply routines disable interrupts whilst running and 18281restore the previous interrupt state when they finish. This makes 18282them safe to use inside interrupt handlers as well as in normal code. 18283 18284@item -minrt 18285@opindex minrt 18286Enable the use of a minimum runtime environment - no static 18287initializers or constructors. This is intended for memory-constrained 18288devices. The compiler will include special symbols in some objects 18289that tell the linker and runtime which code fragments are required. 18290 18291@end table 18292 18293@node NDS32 Options 18294@subsection NDS32 Options 18295@cindex NDS32 Options 18296 18297These options are defined for NDS32 implementations: 18298 18299@table @gcctabopt 18300 18301@item -mbig-endian 18302@opindex mbig-endian 18303Generate code in big-endian mode. 18304 18305@item -mlittle-endian 18306@opindex mlittle-endian 18307Generate code in little-endian mode. 18308 18309@item -mreduced-regs 18310@opindex mreduced-regs 18311Use reduced-set registers for register allocation. 18312 18313@item -mfull-regs 18314@opindex mfull-regs 18315Use full-set registers for register allocation. 18316 18317@item -mcmov 18318@opindex mcmov 18319Generate conditional move instructions. 18320 18321@item -mno-cmov 18322@opindex mno-cmov 18323Do not generate conditional move instructions. 18324 18325@item -mperf-ext 18326@opindex mperf-ext 18327Generate performance extension instructions. 18328 18329@item -mno-perf-ext 18330@opindex mno-perf-ext 18331Do not generate performance extension instructions. 18332 18333@item -mv3push 18334@opindex mv3push 18335Generate v3 push25/pop25 instructions. 18336 18337@item -mno-v3push 18338@opindex mno-v3push 18339Do not generate v3 push25/pop25 instructions. 18340 18341@item -m16-bit 18342@opindex m16-bit 18343Generate 16-bit instructions. 18344 18345@item -mno-16-bit 18346@opindex mno-16-bit 18347Do not generate 16-bit instructions. 18348 18349@item -mgp-direct 18350@opindex mgp-direct 18351Generate GP base instructions directly. 18352 18353@item -mno-gp-direct 18354@opindex mno-gp-direct 18355Do no generate GP base instructions directly. 18356 18357@item -misr-vector-size=@var{num} 18358@opindex misr-vector-size 18359Specify the size of each interrupt vector, which must be 4 or 16. 18360 18361@item -mcache-block-size=@var{num} 18362@opindex mcache-block-size 18363Specify the size of each cache block, 18364which must be a power of 2 between 4 and 512. 18365 18366@item -march=@var{arch} 18367@opindex march 18368Specify the name of the target architecture. 18369 18370@item -mforce-fp-as-gp 18371@opindex mforce-fp-as-gp 18372Prevent $fp being allocated during register allocation so that compiler 18373is able to force performing fp-as-gp optimization. 18374 18375@item -mforbid-fp-as-gp 18376@opindex mforbid-fp-as-gp 18377Forbid using $fp to access static and global variables. 18378This option strictly forbids fp-as-gp optimization 18379regardless of @option{-mforce-fp-as-gp}. 18380 18381@item -mex9 18382@opindex mex9 18383Use special directives to guide linker doing ex9 optimization. 18384 18385@item -mctor-dtor 18386@opindex mctor-dtor 18387Enable constructor/destructor feature. 18388 18389@item -mrelax 18390@opindex mrelax 18391Guide linker to relax instructions. 18392 18393@end table 18394 18395@node Nios II Options 18396@subsection Nios II Options 18397@cindex Nios II options 18398@cindex Altera Nios II options 18399 18400These are the options defined for the Altera Nios II processor. 18401 18402@table @gcctabopt 18403 18404@item -G @var{num} 18405@opindex G 18406@cindex smaller data references 18407Put global and static objects less than or equal to @var{num} bytes 18408into the small data or BSS sections instead of the normal data or BSS 18409sections. The default value of @var{num} is 8. 18410 18411@item -mgpopt 18412@itemx -mno-gpopt 18413@opindex mgpopt 18414@opindex mno-gpopt 18415Generate (do not generate) GP-relative accesses for objects in the 18416small data or BSS sections. The default is @option{-mgpopt} except 18417when @option{-fpic} or @option{-fPIC} is specified to generate 18418position-independent code. Note that the Nios II ABI does not permit 18419GP-relative accesses from shared libraries. 18420 18421You may need to specify @option{-mno-gpopt} explicitly when building 18422programs that include large amounts of small data, including large 18423GOT data sections. In this case, the 16-bit offset for GP-relative 18424addressing may not be large enough to allow access to the entire 18425small data section. 18426 18427@item -mel 18428@itemx -meb 18429@opindex mel 18430@opindex meb 18431Generate little-endian (default) or big-endian (experimental) code, 18432respectively. 18433 18434@item -mbypass-cache 18435@itemx -mno-bypass-cache 18436@opindex mno-bypass-cache 18437@opindex mbypass-cache 18438Force all load and store instructions to always bypass cache by 18439using I/O variants of the instructions. The default is not to 18440bypass the cache. 18441 18442@item -mno-cache-volatile 18443@itemx -mcache-volatile 18444@opindex mcache-volatile 18445@opindex mno-cache-volatile 18446Volatile memory access bypass the cache using the I/O variants of 18447the load and store instructions. The default is not to bypass the cache. 18448 18449@item -mno-fast-sw-div 18450@itemx -mfast-sw-div 18451@opindex mno-fast-sw-div 18452@opindex mfast-sw-div 18453Do not use table-based fast divide for small numbers. The default 18454is to use the fast divide at @option{-O3} and above. 18455 18456@item -mno-hw-mul 18457@itemx -mhw-mul 18458@itemx -mno-hw-mulx 18459@itemx -mhw-mulx 18460@itemx -mno-hw-div 18461@itemx -mhw-div 18462@opindex mno-hw-mul 18463@opindex mhw-mul 18464@opindex mno-hw-mulx 18465@opindex mhw-mulx 18466@opindex mno-hw-div 18467@opindex mhw-div 18468Enable or disable emitting @code{mul}, @code{mulx} and @code{div} family of 18469instructions by the compiler. The default is to emit @code{mul} 18470and not emit @code{div} and @code{mulx}. 18471 18472@item -mcustom-@var{insn}=@var{N} 18473@itemx -mno-custom-@var{insn} 18474@opindex mcustom-@var{insn} 18475@opindex mno-custom-@var{insn} 18476Each @option{-mcustom-@var{insn}=@var{N}} option enables use of a 18477custom instruction with encoding @var{N} when generating code that uses 18478@var{insn}. For example, @code{-mcustom-fadds=253} generates custom 18479instruction 253 for single-precision floating-point add operations instead 18480of the default behavior of using a library call. 18481 18482The following values of @var{insn} are supported. Except as otherwise 18483noted, floating-point operations are expected to be implemented with 18484normal IEEE 754 semantics and correspond directly to the C operators or the 18485equivalent GCC built-in functions (@pxref{Other Builtins}). 18486 18487Single-precision floating point: 18488@table @asis 18489 18490@item @samp{fadds}, @samp{fsubs}, @samp{fdivs}, @samp{fmuls} 18491Binary arithmetic operations. 18492 18493@item @samp{fnegs} 18494Unary negation. 18495 18496@item @samp{fabss} 18497Unary absolute value. 18498 18499@item @samp{fcmpeqs}, @samp{fcmpges}, @samp{fcmpgts}, @samp{fcmples}, @samp{fcmplts}, @samp{fcmpnes} 18500Comparison operations. 18501 18502@item @samp{fmins}, @samp{fmaxs} 18503Floating-point minimum and maximum. These instructions are only 18504generated if @option{-ffinite-math-only} is specified. 18505 18506@item @samp{fsqrts} 18507Unary square root operation. 18508 18509@item @samp{fcoss}, @samp{fsins}, @samp{ftans}, @samp{fatans}, @samp{fexps}, @samp{flogs} 18510Floating-point trigonometric and exponential functions. These instructions 18511are only generated if @option{-funsafe-math-optimizations} is also specified. 18512 18513@end table 18514 18515Double-precision floating point: 18516@table @asis 18517 18518@item @samp{faddd}, @samp{fsubd}, @samp{fdivd}, @samp{fmuld} 18519Binary arithmetic operations. 18520 18521@item @samp{fnegd} 18522Unary negation. 18523 18524@item @samp{fabsd} 18525Unary absolute value. 18526 18527@item @samp{fcmpeqd}, @samp{fcmpged}, @samp{fcmpgtd}, @samp{fcmpled}, @samp{fcmpltd}, @samp{fcmpned} 18528Comparison operations. 18529 18530@item @samp{fmind}, @samp{fmaxd} 18531Double-precision minimum and maximum. These instructions are only 18532generated if @option{-ffinite-math-only} is specified. 18533 18534@item @samp{fsqrtd} 18535Unary square root operation. 18536 18537@item @samp{fcosd}, @samp{fsind}, @samp{ftand}, @samp{fatand}, @samp{fexpd}, @samp{flogd} 18538Double-precision trigonometric and exponential functions. These instructions 18539are only generated if @option{-funsafe-math-optimizations} is also specified. 18540 18541@end table 18542 18543Conversions: 18544@table @asis 18545@item @samp{fextsd} 18546Conversion from single precision to double precision. 18547 18548@item @samp{ftruncds} 18549Conversion from double precision to single precision. 18550 18551@item @samp{fixsi}, @samp{fixsu}, @samp{fixdi}, @samp{fixdu} 18552Conversion from floating point to signed or unsigned integer types, with 18553truncation towards zero. 18554 18555@item @samp{floatis}, @samp{floatus}, @samp{floatid}, @samp{floatud} 18556Conversion from signed or unsigned integer types to floating-point types. 18557 18558@end table 18559 18560In addition, all of the following transfer instructions for internal 18561registers X and Y must be provided to use any of the double-precision 18562floating-point instructions. Custom instructions taking two 18563double-precision source operands expect the first operand in the 1856464-bit register X. The other operand (or only operand of a unary 18565operation) is given to the custom arithmetic instruction with the 18566least significant half in source register @var{src1} and the most 18567significant half in @var{src2}. A custom instruction that returns a 18568double-precision result returns the most significant 32 bits in the 18569destination register and the other half in 32-bit register Y. 18570GCC automatically generates the necessary code sequences to write 18571register X and/or read register Y when double-precision floating-point 18572instructions are used. 18573 18574@table @asis 18575 18576@item @samp{fwrx} 18577Write @var{src1} into the least significant half of X and @var{src2} into 18578the most significant half of X. 18579 18580@item @samp{fwry} 18581Write @var{src1} into Y. 18582 18583@item @samp{frdxhi}, @samp{frdxlo} 18584Read the most or least (respectively) significant half of X and store it in 18585@var{dest}. 18586 18587@item @samp{frdy} 18588Read the value of Y and store it into @var{dest}. 18589@end table 18590 18591Note that you can gain more local control over generation of Nios II custom 18592instructions by using the @code{target("custom-@var{insn}=@var{N}")} 18593and @code{target("no-custom-@var{insn}")} function attributes 18594(@pxref{Function Attributes}) 18595or pragmas (@pxref{Function Specific Option Pragmas}). 18596 18597@item -mcustom-fpu-cfg=@var{name} 18598@opindex mcustom-fpu-cfg 18599 18600This option enables a predefined, named set of custom instruction encodings 18601(see @option{-mcustom-@var{insn}} above). 18602Currently, the following sets are defined: 18603 18604@option{-mcustom-fpu-cfg=60-1} is equivalent to: 18605@gccoptlist{-mcustom-fmuls=252 @gol 18606-mcustom-fadds=253 @gol 18607-mcustom-fsubs=254 @gol 18608-fsingle-precision-constant} 18609 18610@option{-mcustom-fpu-cfg=60-2} is equivalent to: 18611@gccoptlist{-mcustom-fmuls=252 @gol 18612-mcustom-fadds=253 @gol 18613-mcustom-fsubs=254 @gol 18614-mcustom-fdivs=255 @gol 18615-fsingle-precision-constant} 18616 18617@option{-mcustom-fpu-cfg=72-3} is equivalent to: 18618@gccoptlist{-mcustom-floatus=243 @gol 18619-mcustom-fixsi=244 @gol 18620-mcustom-floatis=245 @gol 18621-mcustom-fcmpgts=246 @gol 18622-mcustom-fcmples=249 @gol 18623-mcustom-fcmpeqs=250 @gol 18624-mcustom-fcmpnes=251 @gol 18625-mcustom-fmuls=252 @gol 18626-mcustom-fadds=253 @gol 18627-mcustom-fsubs=254 @gol 18628-mcustom-fdivs=255 @gol 18629-fsingle-precision-constant} 18630 18631Custom instruction assignments given by individual 18632@option{-mcustom-@var{insn}=} options override those given by 18633@option{-mcustom-fpu-cfg=}, regardless of the 18634order of the options on the command line. 18635 18636Note that you can gain more local control over selection of a FPU 18637configuration by using the @code{target("custom-fpu-cfg=@var{name}")} 18638function attribute (@pxref{Function Attributes}) 18639or pragma (@pxref{Function Specific Option Pragmas}). 18640 18641@end table 18642 18643These additional @samp{-m} options are available for the Altera Nios II 18644ELF (bare-metal) target: 18645 18646@table @gcctabopt 18647 18648@item -mhal 18649@opindex mhal 18650Link with HAL BSP. This suppresses linking with the GCC-provided C runtime 18651startup and termination code, and is typically used in conjunction with 18652@option{-msys-crt0=} to specify the location of the alternate startup code 18653provided by the HAL BSP. 18654 18655@item -msmallc 18656@opindex msmallc 18657Link with a limited version of the C library, @option{-lsmallc}, rather than 18658Newlib. 18659 18660@item -msys-crt0=@var{startfile} 18661@opindex msys-crt0 18662@var{startfile} is the file name of the startfile (crt0) to use 18663when linking. This option is only useful in conjunction with @option{-mhal}. 18664 18665@item -msys-lib=@var{systemlib} 18666@opindex msys-lib 18667@var{systemlib} is the library name of the library that provides 18668low-level system calls required by the C library, 18669e.g. @code{read} and @code{write}. 18670This option is typically used to link with a library provided by a HAL BSP. 18671 18672@end table 18673 18674@node PDP-11 Options 18675@subsection PDP-11 Options 18676@cindex PDP-11 Options 18677 18678These options are defined for the PDP-11: 18679 18680@table @gcctabopt 18681@item -mfpu 18682@opindex mfpu 18683Use hardware FPP floating point. This is the default. (FIS floating 18684point on the PDP-11/40 is not supported.) 18685 18686@item -msoft-float 18687@opindex msoft-float 18688Do not use hardware floating point. 18689 18690@item -mac0 18691@opindex mac0 18692Return floating-point results in ac0 (fr0 in Unix assembler syntax). 18693 18694@item -mno-ac0 18695@opindex mno-ac0 18696Return floating-point results in memory. This is the default. 18697 18698@item -m40 18699@opindex m40 18700Generate code for a PDP-11/40. 18701 18702@item -m45 18703@opindex m45 18704Generate code for a PDP-11/45. This is the default. 18705 18706@item -m10 18707@opindex m10 18708Generate code for a PDP-11/10. 18709 18710@item -mbcopy-builtin 18711@opindex mbcopy-builtin 18712Use inline @code{movmemhi} patterns for copying memory. This is the 18713default. 18714 18715@item -mbcopy 18716@opindex mbcopy 18717Do not use inline @code{movmemhi} patterns for copying memory. 18718 18719@item -mint16 18720@itemx -mno-int32 18721@opindex mint16 18722@opindex mno-int32 18723Use 16-bit @code{int}. This is the default. 18724 18725@item -mint32 18726@itemx -mno-int16 18727@opindex mint32 18728@opindex mno-int16 18729Use 32-bit @code{int}. 18730 18731@item -mfloat64 18732@itemx -mno-float32 18733@opindex mfloat64 18734@opindex mno-float32 18735Use 64-bit @code{float}. This is the default. 18736 18737@item -mfloat32 18738@itemx -mno-float64 18739@opindex mfloat32 18740@opindex mno-float64 18741Use 32-bit @code{float}. 18742 18743@item -mabshi 18744@opindex mabshi 18745Use @code{abshi2} pattern. This is the default. 18746 18747@item -mno-abshi 18748@opindex mno-abshi 18749Do not use @code{abshi2} pattern. 18750 18751@item -mbranch-expensive 18752@opindex mbranch-expensive 18753Pretend that branches are expensive. This is for experimenting with 18754code generation only. 18755 18756@item -mbranch-cheap 18757@opindex mbranch-cheap 18758Do not pretend that branches are expensive. This is the default. 18759 18760@item -munix-asm 18761@opindex munix-asm 18762Use Unix assembler syntax. This is the default when configured for 18763@samp{pdp11-*-bsd}. 18764 18765@item -mdec-asm 18766@opindex mdec-asm 18767Use DEC assembler syntax. This is the default when configured for any 18768PDP-11 target other than @samp{pdp11-*-bsd}. 18769@end table 18770 18771@node picoChip Options 18772@subsection picoChip Options 18773@cindex picoChip options 18774 18775These @samp{-m} options are defined for picoChip implementations: 18776 18777@table @gcctabopt 18778 18779@item -mae=@var{ae_type} 18780@opindex mcpu 18781Set the instruction set, register set, and instruction scheduling 18782parameters for array element type @var{ae_type}. Supported values 18783for @var{ae_type} are @samp{ANY}, @samp{MUL}, and @samp{MAC}. 18784 18785@option{-mae=ANY} selects a completely generic AE type. Code 18786generated with this option runs on any of the other AE types. The 18787code is not as efficient as it would be if compiled for a specific 18788AE type, and some types of operation (e.g., multiplication) do not 18789work properly on all types of AE. 18790 18791@option{-mae=MUL} selects a MUL AE type. This is the most useful AE type 18792for compiled code, and is the default. 18793 18794@option{-mae=MAC} selects a DSP-style MAC AE. Code compiled with this 18795option may suffer from poor performance of byte (char) manipulation, 18796since the DSP AE does not provide hardware support for byte load/stores. 18797 18798@item -msymbol-as-address 18799Enable the compiler to directly use a symbol name as an address in a 18800load/store instruction, without first loading it into a 18801register. Typically, the use of this option generates larger 18802programs, which run faster than when the option isn't used. However, the 18803results vary from program to program, so it is left as a user option, 18804rather than being permanently enabled. 18805 18806@item -mno-inefficient-warnings 18807Disables warnings about the generation of inefficient code. These 18808warnings can be generated, for example, when compiling code that 18809performs byte-level memory operations on the MAC AE type. The MAC AE has 18810no hardware support for byte-level memory operations, so all byte 18811load/stores must be synthesized from word load/store operations. This is 18812inefficient and a warning is generated to indicate 18813that you should rewrite the code to avoid byte operations, or to target 18814an AE type that has the necessary hardware support. This option disables 18815these warnings. 18816 18817@end table 18818 18819@node PowerPC Options 18820@subsection PowerPC Options 18821@cindex PowerPC options 18822 18823These are listed under @xref{RS/6000 and PowerPC Options}. 18824 18825@node RL78 Options 18826@subsection RL78 Options 18827@cindex RL78 Options 18828 18829@table @gcctabopt 18830 18831@item -msim 18832@opindex msim 18833Links in additional target libraries to support operation within a 18834simulator. 18835 18836@item -mmul=none 18837@itemx -mmul=g13 18838@itemx -mmul=rl78 18839@opindex mmul 18840Specifies the type of hardware multiplication support to be used. The 18841default is @code{none}, which uses software multiplication functions. 18842The @code{g13} option is for the hardware multiply/divide peripheral 18843only on the RL78/G13 targets. The @code{rl78} option is for the 18844standard hardware multiplication defined in the RL78 software manual. 18845 18846@end table 18847 18848@node RS/6000 and PowerPC Options 18849@subsection IBM RS/6000 and PowerPC Options 18850@cindex RS/6000 and PowerPC Options 18851@cindex IBM RS/6000 and PowerPC Options 18852 18853These @samp{-m} options are defined for the IBM RS/6000 and PowerPC: 18854@table @gcctabopt 18855@item -mpowerpc-gpopt 18856@itemx -mno-powerpc-gpopt 18857@itemx -mpowerpc-gfxopt 18858@itemx -mno-powerpc-gfxopt 18859@need 800 18860@itemx -mpowerpc64 18861@itemx -mno-powerpc64 18862@itemx -mmfcrf 18863@itemx -mno-mfcrf 18864@itemx -mpopcntb 18865@itemx -mno-popcntb 18866@itemx -mpopcntd 18867@itemx -mno-popcntd 18868@itemx -mfprnd 18869@itemx -mno-fprnd 18870@need 800 18871@itemx -mcmpb 18872@itemx -mno-cmpb 18873@itemx -mmfpgpr 18874@itemx -mno-mfpgpr 18875@itemx -mhard-dfp 18876@itemx -mno-hard-dfp 18877@opindex mpowerpc-gpopt 18878@opindex mno-powerpc-gpopt 18879@opindex mpowerpc-gfxopt 18880@opindex mno-powerpc-gfxopt 18881@opindex mpowerpc64 18882@opindex mno-powerpc64 18883@opindex mmfcrf 18884@opindex mno-mfcrf 18885@opindex mpopcntb 18886@opindex mno-popcntb 18887@opindex mpopcntd 18888@opindex mno-popcntd 18889@opindex mfprnd 18890@opindex mno-fprnd 18891@opindex mcmpb 18892@opindex mno-cmpb 18893@opindex mmfpgpr 18894@opindex mno-mfpgpr 18895@opindex mhard-dfp 18896@opindex mno-hard-dfp 18897You use these options to specify which instructions are available on the 18898processor you are using. The default value of these options is 18899determined when configuring GCC@. Specifying the 18900@option{-mcpu=@var{cpu_type}} overrides the specification of these 18901options. We recommend you use the @option{-mcpu=@var{cpu_type}} option 18902rather than the options listed above. 18903 18904Specifying @option{-mpowerpc-gpopt} allows 18905GCC to use the optional PowerPC architecture instructions in the 18906General Purpose group, including floating-point square root. Specifying 18907@option{-mpowerpc-gfxopt} allows GCC to 18908use the optional PowerPC architecture instructions in the Graphics 18909group, including floating-point select. 18910 18911The @option{-mmfcrf} option allows GCC to generate the move from 18912condition register field instruction implemented on the POWER4 18913processor and other processors that support the PowerPC V2.01 18914architecture. 18915The @option{-mpopcntb} option allows GCC to generate the popcount and 18916double-precision FP reciprocal estimate instruction implemented on the 18917POWER5 processor and other processors that support the PowerPC V2.02 18918architecture. 18919The @option{-mpopcntd} option allows GCC to generate the popcount 18920instruction implemented on the POWER7 processor and other processors 18921that support the PowerPC V2.06 architecture. 18922The @option{-mfprnd} option allows GCC to generate the FP round to 18923integer instructions implemented on the POWER5+ processor and other 18924processors that support the PowerPC V2.03 architecture. 18925The @option{-mcmpb} option allows GCC to generate the compare bytes 18926instruction implemented on the POWER6 processor and other processors 18927that support the PowerPC V2.05 architecture. 18928The @option{-mmfpgpr} option allows GCC to generate the FP move to/from 18929general-purpose register instructions implemented on the POWER6X 18930processor and other processors that support the extended PowerPC V2.05 18931architecture. 18932The @option{-mhard-dfp} option allows GCC to generate the decimal 18933floating-point instructions implemented on some POWER processors. 18934 18935The @option{-mpowerpc64} option allows GCC to generate the additional 1893664-bit instructions that are found in the full PowerPC64 architecture 18937and to treat GPRs as 64-bit, doubleword quantities. GCC defaults to 18938@option{-mno-powerpc64}. 18939 18940@item -mcpu=@var{cpu_type} 18941@opindex mcpu 18942Set architecture type, register usage, and 18943instruction scheduling parameters for machine type @var{cpu_type}. 18944Supported values for @var{cpu_type} are @samp{401}, @samp{403}, 18945@samp{405}, @samp{405fp}, @samp{440}, @samp{440fp}, @samp{464}, @samp{464fp}, 18946@samp{476}, @samp{476fp}, @samp{505}, @samp{601}, @samp{602}, @samp{603}, 18947@samp{603e}, @samp{604}, @samp{604e}, @samp{620}, @samp{630}, @samp{740}, 18948@samp{7400}, @samp{7450}, @samp{750}, @samp{801}, @samp{821}, @samp{823}, 18949@samp{860}, @samp{970}, @samp{8540}, @samp{a2}, @samp{e300c2}, 18950@samp{e300c3}, @samp{e500mc}, @samp{e500mc64}, @samp{e5500}, 18951@samp{e6500}, @samp{ec603e}, @samp{G3}, @samp{G4}, @samp{G5}, 18952@samp{titan}, @samp{power3}, @samp{power4}, @samp{power5}, @samp{power5+}, 18953@samp{power6}, @samp{power6x}, @samp{power7}, @samp{power8}, @samp{powerpc}, 18954@samp{powerpc64}, and @samp{rs64}. 18955 18956@option{-mcpu=powerpc}, and @option{-mcpu=powerpc64} specify pure 32-bit 18957PowerPC and 64-bit PowerPC architecture machine 18958types, with an appropriate, generic processor model assumed for 18959scheduling purposes. 18960 18961The other options specify a specific processor. Code generated under 18962those options runs best on that processor, and may not run at all on 18963others. 18964 18965The @option{-mcpu} options automatically enable or disable the 18966following options: 18967 18968@gccoptlist{-maltivec -mfprnd -mhard-float -mmfcrf -mmultiple @gol 18969-mpopcntb -mpopcntd -mpowerpc64 @gol 18970-mpowerpc-gpopt -mpowerpc-gfxopt -msingle-float -mdouble-float @gol 18971-msimple-fpu -mstring -mmulhw -mdlmzb -mmfpgpr -mvsx @gol 18972-mcrypto -mdirect-move -mpower8-fusion -mpower8-vector @gol 18973-mquad-memory -mquad-memory-atomic} 18974 18975The particular options set for any particular CPU varies between 18976compiler versions, depending on what setting seems to produce optimal 18977code for that CPU; it doesn't necessarily reflect the actual hardware's 18978capabilities. If you wish to set an individual option to a particular 18979value, you may specify it after the @option{-mcpu} option, like 18980@option{-mcpu=970 -mno-altivec}. 18981 18982On AIX, the @option{-maltivec} and @option{-mpowerpc64} options are 18983not enabled or disabled by the @option{-mcpu} option at present because 18984AIX does not have full support for these options. You may still 18985enable or disable them individually if you're sure it'll work in your 18986environment. 18987 18988@item -mtune=@var{cpu_type} 18989@opindex mtune 18990Set the instruction scheduling parameters for machine type 18991@var{cpu_type}, but do not set the architecture type or register usage, 18992as @option{-mcpu=@var{cpu_type}} does. The same 18993values for @var{cpu_type} are used for @option{-mtune} as for 18994@option{-mcpu}. If both are specified, the code generated uses the 18995architecture and registers set by @option{-mcpu}, but the 18996scheduling parameters set by @option{-mtune}. 18997 18998@item -mcmodel=small 18999@opindex mcmodel=small 19000Generate PowerPC64 code for the small model: The TOC is limited to 1900164k. 19002 19003@item -mcmodel=medium 19004@opindex mcmodel=medium 19005Generate PowerPC64 code for the medium model: The TOC and other static 19006data may be up to a total of 4G in size. 19007 19008@item -mcmodel=large 19009@opindex mcmodel=large 19010Generate PowerPC64 code for the large model: The TOC may be up to 4G 19011in size. Other data and code is only limited by the 64-bit address 19012space. 19013 19014@item -maltivec 19015@itemx -mno-altivec 19016@opindex maltivec 19017@opindex mno-altivec 19018Generate code that uses (does not use) AltiVec instructions, and also 19019enable the use of built-in functions that allow more direct access to 19020the AltiVec instruction set. You may also need to set 19021@option{-mabi=altivec} to adjust the current ABI with AltiVec ABI 19022enhancements. 19023 19024When @option{-maltivec} is used, rather than @option{-maltivec=le} or 19025@option{-maltivec=be}, the element order for Altivec intrinsics such 19026as @code{vec_splat}, @code{vec_extract}, and @code{vec_insert} will 19027match array element order corresponding to the endianness of the 19028target. That is, element zero identifies the leftmost element in a 19029vector register when targeting a big-endian platform, and identifies 19030the rightmost element in a vector register when targeting a 19031little-endian platform. 19032 19033@item -maltivec=be 19034@opindex maltivec=be 19035Generate Altivec instructions using big-endian element order, 19036regardless of whether the target is big- or little-endian. This is 19037the default when targeting a big-endian platform. 19038 19039The element order is used to interpret element numbers in Altivec 19040intrinsics such as @code{vec_splat}, @code{vec_extract}, and 19041@code{vec_insert}. By default, these will match array element order 19042corresponding to the endianness for the target. 19043 19044@item -maltivec=le 19045@opindex maltivec=le 19046Generate Altivec instructions using little-endian element order, 19047regardless of whether the target is big- or little-endian. This is 19048the default when targeting a little-endian platform. This option is 19049currently ignored when targeting a big-endian platform. 19050 19051The element order is used to interpret element numbers in Altivec 19052intrinsics such as @code{vec_splat}, @code{vec_extract}, and 19053@code{vec_insert}. By default, these will match array element order 19054corresponding to the endianness for the target. 19055 19056@item -mvrsave 19057@itemx -mno-vrsave 19058@opindex mvrsave 19059@opindex mno-vrsave 19060Generate VRSAVE instructions when generating AltiVec code. 19061 19062@item -mgen-cell-microcode 19063@opindex mgen-cell-microcode 19064Generate Cell microcode instructions. 19065 19066@item -mwarn-cell-microcode 19067@opindex mwarn-cell-microcode 19068Warn when a Cell microcode instruction is emitted. An example 19069of a Cell microcode instruction is a variable shift. 19070 19071@item -msecure-plt 19072@opindex msecure-plt 19073Generate code that allows @command{ld} and @command{ld.so} 19074to build executables and shared 19075libraries with non-executable @code{.plt} and @code{.got} sections. 19076This is a PowerPC 1907732-bit SYSV ABI option. 19078 19079@item -mbss-plt 19080@opindex mbss-plt 19081Generate code that uses a BSS @code{.plt} section that @command{ld.so} 19082fills in, and 19083requires @code{.plt} and @code{.got} 19084sections that are both writable and executable. 19085This is a PowerPC 32-bit SYSV ABI option. 19086 19087@item -misel 19088@itemx -mno-isel 19089@opindex misel 19090@opindex mno-isel 19091This switch enables or disables the generation of ISEL instructions. 19092 19093@item -misel=@var{yes/no} 19094This switch has been deprecated. Use @option{-misel} and 19095@option{-mno-isel} instead. 19096 19097@item -mspe 19098@itemx -mno-spe 19099@opindex mspe 19100@opindex mno-spe 19101This switch enables or disables the generation of SPE simd 19102instructions. 19103 19104@item -mpaired 19105@itemx -mno-paired 19106@opindex mpaired 19107@opindex mno-paired 19108This switch enables or disables the generation of PAIRED simd 19109instructions. 19110 19111@item -mspe=@var{yes/no} 19112This option has been deprecated. Use @option{-mspe} and 19113@option{-mno-spe} instead. 19114 19115@item -mvsx 19116@itemx -mno-vsx 19117@opindex mvsx 19118@opindex mno-vsx 19119Generate code that uses (does not use) vector/scalar (VSX) 19120instructions, and also enable the use of built-in functions that allow 19121more direct access to the VSX instruction set. 19122 19123@item -mcrypto 19124@itemx -mno-crypto 19125@opindex mcrypto 19126@opindex mno-crypto 19127Enable the use (disable) of the built-in functions that allow direct 19128access to the cryptographic instructions that were added in version 191292.07 of the PowerPC ISA. 19130 19131@item -mdirect-move 19132@itemx -mno-direct-move 19133@opindex mdirect-move 19134@opindex mno-direct-move 19135Generate code that uses (does not use) the instructions to move data 19136between the general purpose registers and the vector/scalar (VSX) 19137registers that were added in version 2.07 of the PowerPC ISA. 19138 19139@item -mpower8-fusion 19140@itemx -mno-power8-fusion 19141@opindex mpower8-fusion 19142@opindex mno-power8-fusion 19143Generate code that keeps (does not keeps) some integer operations 19144adjacent so that the instructions can be fused together on power8 and 19145later processors. 19146 19147@item -mpower8-vector 19148@itemx -mno-power8-vector 19149@opindex mpower8-vector 19150@opindex mno-power8-vector 19151Generate code that uses (does not use) the vector and scalar 19152instructions that were added in version 2.07 of the PowerPC ISA. Also 19153enable the use of built-in functions that allow more direct access to 19154the vector instructions. 19155 19156@item -mquad-memory 19157@itemx -mno-quad-memory 19158@opindex mquad-memory 19159@opindex mno-quad-memory 19160Generate code that uses (does not use) the non-atomic quad word memory 19161instructions. The @option{-mquad-memory} option requires use of 1916264-bit mode. 19163 19164@item -mquad-memory-atomic 19165@itemx -mno-quad-memory-atomic 19166@opindex mquad-memory-atomic 19167@opindex mno-quad-memory-atomic 19168Generate code that uses (does not use) the atomic quad word memory 19169instructions. The @option{-mquad-memory-atomic} option requires use of 1917064-bit mode. 19171 19172@item -mfloat-gprs=@var{yes/single/double/no} 19173@itemx -mfloat-gprs 19174@opindex mfloat-gprs 19175This switch enables or disables the generation of floating-point 19176operations on the general-purpose registers for architectures that 19177support it. 19178 19179The argument @var{yes} or @var{single} enables the use of 19180single-precision floating-point operations. 19181 19182The argument @var{double} enables the use of single and 19183double-precision floating-point operations. 19184 19185The argument @var{no} disables floating-point operations on the 19186general-purpose registers. 19187 19188This option is currently only available on the MPC854x. 19189 19190@item -m32 19191@itemx -m64 19192@opindex m32 19193@opindex m64 19194Generate code for 32-bit or 64-bit environments of Darwin and SVR4 19195targets (including GNU/Linux). The 32-bit environment sets int, long 19196and pointer to 32 bits and generates code that runs on any PowerPC 19197variant. The 64-bit environment sets int to 32 bits and long and 19198pointer to 64 bits, and generates code for PowerPC64, as for 19199@option{-mpowerpc64}. 19200 19201@item -mfull-toc 19202@itemx -mno-fp-in-toc 19203@itemx -mno-sum-in-toc 19204@itemx -mminimal-toc 19205@opindex mfull-toc 19206@opindex mno-fp-in-toc 19207@opindex mno-sum-in-toc 19208@opindex mminimal-toc 19209Modify generation of the TOC (Table Of Contents), which is created for 19210every executable file. The @option{-mfull-toc} option is selected by 19211default. In that case, GCC allocates at least one TOC entry for 19212each unique non-automatic variable reference in your program. GCC 19213also places floating-point constants in the TOC@. However, only 1921416,384 entries are available in the TOC@. 19215 19216If you receive a linker error message that saying you have overflowed 19217the available TOC space, you can reduce the amount of TOC space used 19218with the @option{-mno-fp-in-toc} and @option{-mno-sum-in-toc} options. 19219@option{-mno-fp-in-toc} prevents GCC from putting floating-point 19220constants in the TOC and @option{-mno-sum-in-toc} forces GCC to 19221generate code to calculate the sum of an address and a constant at 19222run time instead of putting that sum into the TOC@. You may specify one 19223or both of these options. Each causes GCC to produce very slightly 19224slower and larger code at the expense of conserving TOC space. 19225 19226If you still run out of space in the TOC even when you specify both of 19227these options, specify @option{-mminimal-toc} instead. This option causes 19228GCC to make only one TOC entry for every file. When you specify this 19229option, GCC produces code that is slower and larger but which 19230uses extremely little TOC space. You may wish to use this option 19231only on files that contain less frequently-executed code. 19232 19233@item -maix64 19234@itemx -maix32 19235@opindex maix64 19236@opindex maix32 19237Enable 64-bit AIX ABI and calling convention: 64-bit pointers, 64-bit 19238@code{long} type, and the infrastructure needed to support them. 19239Specifying @option{-maix64} implies @option{-mpowerpc64}, 19240while @option{-maix32} disables the 64-bit ABI and 19241implies @option{-mno-powerpc64}. GCC defaults to @option{-maix32}. 19242 19243@item -mxl-compat 19244@itemx -mno-xl-compat 19245@opindex mxl-compat 19246@opindex mno-xl-compat 19247Produce code that conforms more closely to IBM XL compiler semantics 19248when using AIX-compatible ABI@. Pass floating-point arguments to 19249prototyped functions beyond the register save area (RSA) on the stack 19250in addition to argument FPRs. Do not assume that most significant 19251double in 128-bit long double value is properly rounded when comparing 19252values and converting to double. Use XL symbol names for long double 19253support routines. 19254 19255The AIX calling convention was extended but not initially documented to 19256handle an obscure K&R C case of calling a function that takes the 19257address of its arguments with fewer arguments than declared. IBM XL 19258compilers access floating-point arguments that do not fit in the 19259RSA from the stack when a subroutine is compiled without 19260optimization. Because always storing floating-point arguments on the 19261stack is inefficient and rarely needed, this option is not enabled by 19262default and only is necessary when calling subroutines compiled by IBM 19263XL compilers without optimization. 19264 19265@item -mpe 19266@opindex mpe 19267Support @dfn{IBM RS/6000 SP} @dfn{Parallel Environment} (PE)@. Link an 19268application written to use message passing with special startup code to 19269enable the application to run. The system must have PE installed in the 19270standard location (@file{/usr/lpp/ppe.poe/}), or the @file{specs} file 19271must be overridden with the @option{-specs=} option to specify the 19272appropriate directory location. The Parallel Environment does not 19273support threads, so the @option{-mpe} option and the @option{-pthread} 19274option are incompatible. 19275 19276@item -malign-natural 19277@itemx -malign-power 19278@opindex malign-natural 19279@opindex malign-power 19280On AIX, 32-bit Darwin, and 64-bit PowerPC GNU/Linux, the option 19281@option{-malign-natural} overrides the ABI-defined alignment of larger 19282types, such as floating-point doubles, on their natural size-based boundary. 19283The option @option{-malign-power} instructs GCC to follow the ABI-specified 19284alignment rules. GCC defaults to the standard alignment defined in the ABI@. 19285 19286On 64-bit Darwin, natural alignment is the default, and @option{-malign-power} 19287is not supported. 19288 19289@item -msoft-float 19290@itemx -mhard-float 19291@opindex msoft-float 19292@opindex mhard-float 19293Generate code that does not use (uses) the floating-point register set. 19294Software floating-point emulation is provided if you use the 19295@option{-msoft-float} option, and pass the option to GCC when linking. 19296 19297@item -msingle-float 19298@itemx -mdouble-float 19299@opindex msingle-float 19300@opindex mdouble-float 19301Generate code for single- or double-precision floating-point operations. 19302@option{-mdouble-float} implies @option{-msingle-float}. 19303 19304@item -msimple-fpu 19305@opindex msimple-fpu 19306Do not generate @code{sqrt} and @code{div} instructions for hardware 19307floating-point unit. 19308 19309@item -mfpu=@var{name} 19310@opindex mfpu 19311Specify type of floating-point unit. Valid values for @var{name} are 19312@samp{sp_lite} (equivalent to @option{-msingle-float -msimple-fpu}), 19313@samp{dp_lite} (equivalent to @option{-mdouble-float -msimple-fpu}), 19314@samp{sp_full} (equivalent to @option{-msingle-float}), 19315and @samp{dp_full} (equivalent to @option{-mdouble-float}). 19316 19317@item -mxilinx-fpu 19318@opindex mxilinx-fpu 19319Perform optimizations for the floating-point unit on Xilinx PPC 405/440. 19320 19321@item -mmultiple 19322@itemx -mno-multiple 19323@opindex mmultiple 19324@opindex mno-multiple 19325Generate code that uses (does not use) the load multiple word 19326instructions and the store multiple word instructions. These 19327instructions are generated by default on POWER systems, and not 19328generated on PowerPC systems. Do not use @option{-mmultiple} on little-endian 19329PowerPC systems, since those instructions do not work when the 19330processor is in little-endian mode. The exceptions are PPC740 and 19331PPC750 which permit these instructions in little-endian mode. 19332 19333@item -mstring 19334@itemx -mno-string 19335@opindex mstring 19336@opindex mno-string 19337Generate code that uses (does not use) the load string instructions 19338and the store string word instructions to save multiple registers and 19339do small block moves. These instructions are generated by default on 19340POWER systems, and not generated on PowerPC systems. Do not use 19341@option{-mstring} on little-endian PowerPC systems, since those 19342instructions do not work when the processor is in little-endian mode. 19343The exceptions are PPC740 and PPC750 which permit these instructions 19344in little-endian mode. 19345 19346@item -mupdate 19347@itemx -mno-update 19348@opindex mupdate 19349@opindex mno-update 19350Generate code that uses (does not use) the load or store instructions 19351that update the base register to the address of the calculated memory 19352location. These instructions are generated by default. If you use 19353@option{-mno-update}, there is a small window between the time that the 19354stack pointer is updated and the address of the previous frame is 19355stored, which means code that walks the stack frame across interrupts or 19356signals may get corrupted data. 19357 19358@item -mavoid-indexed-addresses 19359@itemx -mno-avoid-indexed-addresses 19360@opindex mavoid-indexed-addresses 19361@opindex mno-avoid-indexed-addresses 19362Generate code that tries to avoid (not avoid) the use of indexed load 19363or store instructions. These instructions can incur a performance 19364penalty on Power6 processors in certain situations, such as when 19365stepping through large arrays that cross a 16M boundary. This option 19366is enabled by default when targeting Power6 and disabled otherwise. 19367 19368@item -mfused-madd 19369@itemx -mno-fused-madd 19370@opindex mfused-madd 19371@opindex mno-fused-madd 19372Generate code that uses (does not use) the floating-point multiply and 19373accumulate instructions. These instructions are generated by default 19374if hardware floating point is used. The machine-dependent 19375@option{-mfused-madd} option is now mapped to the machine-independent 19376@option{-ffp-contract=fast} option, and @option{-mno-fused-madd} is 19377mapped to @option{-ffp-contract=off}. 19378 19379@item -mmulhw 19380@itemx -mno-mulhw 19381@opindex mmulhw 19382@opindex mno-mulhw 19383Generate code that uses (does not use) the half-word multiply and 19384multiply-accumulate instructions on the IBM 405, 440, 464 and 476 processors. 19385These instructions are generated by default when targeting those 19386processors. 19387 19388@item -mdlmzb 19389@itemx -mno-dlmzb 19390@opindex mdlmzb 19391@opindex mno-dlmzb 19392Generate code that uses (does not use) the string-search @samp{dlmzb} 19393instruction on the IBM 405, 440, 464 and 476 processors. This instruction is 19394generated by default when targeting those processors. 19395 19396@item -mno-bit-align 19397@itemx -mbit-align 19398@opindex mno-bit-align 19399@opindex mbit-align 19400On System V.4 and embedded PowerPC systems do not (do) force structures 19401and unions that contain bit-fields to be aligned to the base type of the 19402bit-field. 19403 19404For example, by default a structure containing nothing but 8 19405@code{unsigned} bit-fields of length 1 is aligned to a 4-byte 19406boundary and has a size of 4 bytes. By using @option{-mno-bit-align}, 19407the structure is aligned to a 1-byte boundary and is 1 byte in 19408size. 19409 19410@item -mno-strict-align 19411@itemx -mstrict-align 19412@opindex mno-strict-align 19413@opindex mstrict-align 19414On System V.4 and embedded PowerPC systems do not (do) assume that 19415unaligned memory references are handled by the system. 19416 19417@item -mrelocatable 19418@itemx -mno-relocatable 19419@opindex mrelocatable 19420@opindex mno-relocatable 19421Generate code that allows (does not allow) a static executable to be 19422relocated to a different address at run time. A simple embedded 19423PowerPC system loader should relocate the entire contents of 19424@code{.got2} and 4-byte locations listed in the @code{.fixup} section, 19425a table of 32-bit addresses generated by this option. For this to 19426work, all objects linked together must be compiled with 19427@option{-mrelocatable} or @option{-mrelocatable-lib}. 19428@option{-mrelocatable} code aligns the stack to an 8-byte boundary. 19429 19430@item -mrelocatable-lib 19431@itemx -mno-relocatable-lib 19432@opindex mrelocatable-lib 19433@opindex mno-relocatable-lib 19434Like @option{-mrelocatable}, @option{-mrelocatable-lib} generates a 19435@code{.fixup} section to allow static executables to be relocated at 19436run time, but @option{-mrelocatable-lib} does not use the smaller stack 19437alignment of @option{-mrelocatable}. Objects compiled with 19438@option{-mrelocatable-lib} may be linked with objects compiled with 19439any combination of the @option{-mrelocatable} options. 19440 19441@item -mno-toc 19442@itemx -mtoc 19443@opindex mno-toc 19444@opindex mtoc 19445On System V.4 and embedded PowerPC systems do not (do) assume that 19446register 2 contains a pointer to a global area pointing to the addresses 19447used in the program. 19448 19449@item -mlittle 19450@itemx -mlittle-endian 19451@opindex mlittle 19452@opindex mlittle-endian 19453On System V.4 and embedded PowerPC systems compile code for the 19454processor in little-endian mode. The @option{-mlittle-endian} option is 19455the same as @option{-mlittle}. 19456 19457@item -mbig 19458@itemx -mbig-endian 19459@opindex mbig 19460@opindex mbig-endian 19461On System V.4 and embedded PowerPC systems compile code for the 19462processor in big-endian mode. The @option{-mbig-endian} option is 19463the same as @option{-mbig}. 19464 19465@item -mdynamic-no-pic 19466@opindex mdynamic-no-pic 19467On Darwin and Mac OS X systems, compile code so that it is not 19468relocatable, but that its external references are relocatable. The 19469resulting code is suitable for applications, but not shared 19470libraries. 19471 19472@item -msingle-pic-base 19473@opindex msingle-pic-base 19474Treat the register used for PIC addressing as read-only, rather than 19475loading it in the prologue for each function. The runtime system is 19476responsible for initializing this register with an appropriate value 19477before execution begins. 19478 19479@item -mprioritize-restricted-insns=@var{priority} 19480@opindex mprioritize-restricted-insns 19481This option controls the priority that is assigned to 19482dispatch-slot restricted instructions during the second scheduling 19483pass. The argument @var{priority} takes the value @samp{0}, @samp{1}, 19484or @samp{2} to assign no, highest, or second-highest (respectively) 19485priority to dispatch-slot restricted 19486instructions. 19487 19488@item -msched-costly-dep=@var{dependence_type} 19489@opindex msched-costly-dep 19490This option controls which dependences are considered costly 19491by the target during instruction scheduling. The argument 19492@var{dependence_type} takes one of the following values: 19493 19494@table @asis 19495@item @samp{no} 19496No dependence is costly. 19497 19498@item @samp{all} 19499All dependences are costly. 19500 19501@item @samp{true_store_to_load} 19502A true dependence from store to load is costly. 19503 19504@item @samp{store_to_load} 19505Any dependence from store to load is costly. 19506 19507@item @var{number} 19508Any dependence for which the latency is greater than or equal to 19509@var{number} is costly. 19510@end table 19511 19512@item -minsert-sched-nops=@var{scheme} 19513@opindex minsert-sched-nops 19514This option controls which NOP insertion scheme is used during 19515the second scheduling pass. The argument @var{scheme} takes one of the 19516following values: 19517 19518@table @asis 19519@item @samp{no} 19520Don't insert NOPs. 19521 19522@item @samp{pad} 19523Pad with NOPs any dispatch group that has vacant issue slots, 19524according to the scheduler's grouping. 19525 19526@item @samp{regroup_exact} 19527Insert NOPs to force costly dependent insns into 19528separate groups. Insert exactly as many NOPs as needed to force an insn 19529to a new group, according to the estimated processor grouping. 19530 19531@item @var{number} 19532Insert NOPs to force costly dependent insns into 19533separate groups. Insert @var{number} NOPs to force an insn to a new group. 19534@end table 19535 19536@item -mcall-sysv 19537@opindex mcall-sysv 19538On System V.4 and embedded PowerPC systems compile code using calling 19539conventions that adhere to the March 1995 draft of the System V 19540Application Binary Interface, PowerPC processor supplement. This is the 19541default unless you configured GCC using @samp{powerpc-*-eabiaix}. 19542 19543@item -mcall-sysv-eabi 19544@itemx -mcall-eabi 19545@opindex mcall-sysv-eabi 19546@opindex mcall-eabi 19547Specify both @option{-mcall-sysv} and @option{-meabi} options. 19548 19549@item -mcall-sysv-noeabi 19550@opindex mcall-sysv-noeabi 19551Specify both @option{-mcall-sysv} and @option{-mno-eabi} options. 19552 19553@item -mcall-aixdesc 19554@opindex m 19555On System V.4 and embedded PowerPC systems compile code for the AIX 19556operating system. 19557 19558@item -mcall-linux 19559@opindex mcall-linux 19560On System V.4 and embedded PowerPC systems compile code for the 19561Linux-based GNU system. 19562 19563@item -mcall-freebsd 19564@opindex mcall-freebsd 19565On System V.4 and embedded PowerPC systems compile code for the 19566FreeBSD operating system. 19567 19568@item -mcall-netbsd 19569@opindex mcall-netbsd 19570On System V.4 and embedded PowerPC systems compile code for the 19571NetBSD operating system. 19572 19573@item -mcall-openbsd 19574@opindex mcall-netbsd 19575On System V.4 and embedded PowerPC systems compile code for the 19576OpenBSD operating system. 19577 19578@item -maix-struct-return 19579@opindex maix-struct-return 19580Return all structures in memory (as specified by the AIX ABI)@. 19581 19582@item -msvr4-struct-return 19583@opindex msvr4-struct-return 19584Return structures smaller than 8 bytes in registers (as specified by the 19585SVR4 ABI)@. 19586 19587@item -mabi=@var{abi-type} 19588@opindex mabi 19589Extend the current ABI with a particular extension, or remove such extension. 19590Valid values are @var{altivec}, @var{no-altivec}, @var{spe}, 19591@var{no-spe}, @var{ibmlongdouble}, @var{ieeelongdouble}, 19592@var{elfv1}, @var{elfv2}@. 19593 19594@item -mabi=spe 19595@opindex mabi=spe 19596Extend the current ABI with SPE ABI extensions. This does not change 19597the default ABI, instead it adds the SPE ABI extensions to the current 19598ABI@. 19599 19600@item -mabi=no-spe 19601@opindex mabi=no-spe 19602Disable Book-E SPE ABI extensions for the current ABI@. 19603 19604@item -mabi=ibmlongdouble 19605@opindex mabi=ibmlongdouble 19606Change the current ABI to use IBM extended-precision long double. 19607This is a PowerPC 32-bit SYSV ABI option. 19608 19609@item -mabi=ieeelongdouble 19610@opindex mabi=ieeelongdouble 19611Change the current ABI to use IEEE extended-precision long double. 19612This is a PowerPC 32-bit Linux ABI option. 19613 19614@item -mabi=elfv1 19615@opindex mabi=elfv1 19616Change the current ABI to use the ELFv1 ABI. 19617This is the default ABI for big-endian PowerPC 64-bit Linux. 19618Overriding the default ABI requires special system support and is 19619likely to fail in spectacular ways. 19620 19621@item -mabi=elfv2 19622@opindex mabi=elfv2 19623Change the current ABI to use the ELFv2 ABI. 19624This is the default ABI for little-endian PowerPC 64-bit Linux. 19625Overriding the default ABI requires special system support and is 19626likely to fail in spectacular ways. 19627 19628@item -mprototype 19629@itemx -mno-prototype 19630@opindex mprototype 19631@opindex mno-prototype 19632On System V.4 and embedded PowerPC systems assume that all calls to 19633variable argument functions are properly prototyped. Otherwise, the 19634compiler must insert an instruction before every non-prototyped call to 19635set or clear bit 6 of the condition code register (@var{CR}) to 19636indicate whether floating-point values are passed in the floating-point 19637registers in case the function takes variable arguments. With 19638@option{-mprototype}, only calls to prototyped variable argument functions 19639set or clear the bit. 19640 19641@item -msim 19642@opindex msim 19643On embedded PowerPC systems, assume that the startup module is called 19644@file{sim-crt0.o} and that the standard C libraries are @file{libsim.a} and 19645@file{libc.a}. This is the default for @samp{powerpc-*-eabisim} 19646configurations. 19647 19648@item -mmvme 19649@opindex mmvme 19650On embedded PowerPC systems, assume that the startup module is called 19651@file{crt0.o} and the standard C libraries are @file{libmvme.a} and 19652@file{libc.a}. 19653 19654@item -mads 19655@opindex mads 19656On embedded PowerPC systems, assume that the startup module is called 19657@file{crt0.o} and the standard C libraries are @file{libads.a} and 19658@file{libc.a}. 19659 19660@item -myellowknife 19661@opindex myellowknife 19662On embedded PowerPC systems, assume that the startup module is called 19663@file{crt0.o} and the standard C libraries are @file{libyk.a} and 19664@file{libc.a}. 19665 19666@item -mvxworks 19667@opindex mvxworks 19668On System V.4 and embedded PowerPC systems, specify that you are 19669compiling for a VxWorks system. 19670 19671@item -memb 19672@opindex memb 19673On embedded PowerPC systems, set the @var{PPC_EMB} bit in the ELF flags 19674header to indicate that @samp{eabi} extended relocations are used. 19675 19676@item -meabi 19677@itemx -mno-eabi 19678@opindex meabi 19679@opindex mno-eabi 19680On System V.4 and embedded PowerPC systems do (do not) adhere to the 19681Embedded Applications Binary Interface (EABI), which is a set of 19682modifications to the System V.4 specifications. Selecting @option{-meabi} 19683means that the stack is aligned to an 8-byte boundary, a function 19684@code{__eabi} is called from @code{main} to set up the EABI 19685environment, and the @option{-msdata} option can use both @code{r2} and 19686@code{r13} to point to two separate small data areas. Selecting 19687@option{-mno-eabi} means that the stack is aligned to a 16-byte boundary, 19688no EABI initialization function is called from @code{main}, and the 19689@option{-msdata} option only uses @code{r13} to point to a single 19690small data area. The @option{-meabi} option is on by default if you 19691configured GCC using one of the @samp{powerpc*-*-eabi*} options. 19692 19693@item -msdata=eabi 19694@opindex msdata=eabi 19695On System V.4 and embedded PowerPC systems, put small initialized 19696@code{const} global and static data in the @samp{.sdata2} section, which 19697is pointed to by register @code{r2}. Put small initialized 19698non-@code{const} global and static data in the @samp{.sdata} section, 19699which is pointed to by register @code{r13}. Put small uninitialized 19700global and static data in the @samp{.sbss} section, which is adjacent to 19701the @samp{.sdata} section. The @option{-msdata=eabi} option is 19702incompatible with the @option{-mrelocatable} option. The 19703@option{-msdata=eabi} option also sets the @option{-memb} option. 19704 19705@item -msdata=sysv 19706@opindex msdata=sysv 19707On System V.4 and embedded PowerPC systems, put small global and static 19708data in the @samp{.sdata} section, which is pointed to by register 19709@code{r13}. Put small uninitialized global and static data in the 19710@samp{.sbss} section, which is adjacent to the @samp{.sdata} section. 19711The @option{-msdata=sysv} option is incompatible with the 19712@option{-mrelocatable} option. 19713 19714@item -msdata=default 19715@itemx -msdata 19716@opindex msdata=default 19717@opindex msdata 19718On System V.4 and embedded PowerPC systems, if @option{-meabi} is used, 19719compile code the same as @option{-msdata=eabi}, otherwise compile code the 19720same as @option{-msdata=sysv}. 19721 19722@item -msdata=data 19723@opindex msdata=data 19724On System V.4 and embedded PowerPC systems, put small global 19725data in the @samp{.sdata} section. Put small uninitialized global 19726data in the @samp{.sbss} section. Do not use register @code{r13} 19727to address small data however. This is the default behavior unless 19728other @option{-msdata} options are used. 19729 19730@item -msdata=none 19731@itemx -mno-sdata 19732@opindex msdata=none 19733@opindex mno-sdata 19734On embedded PowerPC systems, put all initialized global and static data 19735in the @samp{.data} section, and all uninitialized data in the 19736@samp{.bss} section. 19737 19738@item -mblock-move-inline-limit=@var{num} 19739@opindex mblock-move-inline-limit 19740Inline all block moves (such as calls to @code{memcpy} or structure 19741copies) less than or equal to @var{num} bytes. The minimum value for 19742@var{num} is 32 bytes on 32-bit targets and 64 bytes on 64-bit 19743targets. The default value is target-specific. 19744 19745@item -G @var{num} 19746@opindex G 19747@cindex smaller data references (PowerPC) 19748@cindex .sdata/.sdata2 references (PowerPC) 19749On embedded PowerPC systems, put global and static items less than or 19750equal to @var{num} bytes into the small data or BSS sections instead of 19751the normal data or BSS section. By default, @var{num} is 8. The 19752@option{-G @var{num}} switch is also passed to the linker. 19753All modules should be compiled with the same @option{-G @var{num}} value. 19754 19755@item -mregnames 19756@itemx -mno-regnames 19757@opindex mregnames 19758@opindex mno-regnames 19759On System V.4 and embedded PowerPC systems do (do not) emit register 19760names in the assembly language output using symbolic forms. 19761 19762@item -mlongcall 19763@itemx -mno-longcall 19764@opindex mlongcall 19765@opindex mno-longcall 19766By default assume that all calls are far away so that a longer and more 19767expensive calling sequence is required. This is required for calls 19768farther than 32 megabytes (33,554,432 bytes) from the current location. 19769A short call is generated if the compiler knows 19770the call cannot be that far away. This setting can be overridden by 19771the @code{shortcall} function attribute, or by @code{#pragma 19772longcall(0)}. 19773 19774Some linkers are capable of detecting out-of-range calls and generating 19775glue code on the fly. On these systems, long calls are unnecessary and 19776generate slower code. As of this writing, the AIX linker can do this, 19777as can the GNU linker for PowerPC/64. It is planned to add this feature 19778to the GNU linker for 32-bit PowerPC systems as well. 19779 19780On Darwin/PPC systems, @code{#pragma longcall} generates @code{jbsr 19781callee, L42}, plus a @dfn{branch island} (glue code). The two target 19782addresses represent the callee and the branch island. The 19783Darwin/PPC linker prefers the first address and generates a @code{bl 19784callee} if the PPC @code{bl} instruction reaches the callee directly; 19785otherwise, the linker generates @code{bl L42} to call the branch 19786island. The branch island is appended to the body of the 19787calling function; it computes the full 32-bit address of the callee 19788and jumps to it. 19789 19790On Mach-O (Darwin) systems, this option directs the compiler emit to 19791the glue for every direct call, and the Darwin linker decides whether 19792to use or discard it. 19793 19794In the future, GCC may ignore all longcall specifications 19795when the linker is known to generate glue. 19796 19797@item -mtls-markers 19798@itemx -mno-tls-markers 19799@opindex mtls-markers 19800@opindex mno-tls-markers 19801Mark (do not mark) calls to @code{__tls_get_addr} with a relocation 19802specifying the function argument. The relocation allows the linker to 19803reliably associate function call with argument setup instructions for 19804TLS optimization, which in turn allows GCC to better schedule the 19805sequence. 19806 19807@item -pthread 19808@opindex pthread 19809Adds support for multithreading with the @dfn{pthreads} library. 19810This option sets flags for both the preprocessor and linker. 19811 19812@item -mrecip 19813@itemx -mno-recip 19814@opindex mrecip 19815This option enables use of the reciprocal estimate and 19816reciprocal square root estimate instructions with additional 19817Newton-Raphson steps to increase precision instead of doing a divide or 19818square root and divide for floating-point arguments. You should use 19819the @option{-ffast-math} option when using @option{-mrecip} (or at 19820least @option{-funsafe-math-optimizations}, 19821@option{-finite-math-only}, @option{-freciprocal-math} and 19822@option{-fno-trapping-math}). Note that while the throughput of the 19823sequence is generally higher than the throughput of the non-reciprocal 19824instruction, the precision of the sequence can be decreased by up to 2 19825ulp (i.e.@: the inverse of 1.0 equals 0.99999994) for reciprocal square 19826roots. 19827 19828@item -mrecip=@var{opt} 19829@opindex mrecip=opt 19830This option controls which reciprocal estimate instructions 19831may be used. @var{opt} is a comma-separated list of options, which may 19832be preceded by a @code{!} to invert the option: 19833@code{all}: enable all estimate instructions, 19834@code{default}: enable the default instructions, equivalent to @option{-mrecip}, 19835@code{none}: disable all estimate instructions, equivalent to @option{-mno-recip}; 19836@code{div}: enable the reciprocal approximation instructions for both single and double precision; 19837@code{divf}: enable the single-precision reciprocal approximation instructions; 19838@code{divd}: enable the double-precision reciprocal approximation instructions; 19839@code{rsqrt}: enable the reciprocal square root approximation instructions for both single and double precision; 19840@code{rsqrtf}: enable the single-precision reciprocal square root approximation instructions; 19841@code{rsqrtd}: enable the double-precision reciprocal square root approximation instructions; 19842 19843So, for example, @option{-mrecip=all,!rsqrtd} enables 19844all of the reciprocal estimate instructions, except for the 19845@code{FRSQRTE}, @code{XSRSQRTEDP}, and @code{XVRSQRTEDP} instructions 19846which handle the double-precision reciprocal square root calculations. 19847 19848@item -mrecip-precision 19849@itemx -mno-recip-precision 19850@opindex mrecip-precision 19851Assume (do not assume) that the reciprocal estimate instructions 19852provide higher-precision estimates than is mandated by the PowerPC 19853ABI. Selecting @option{-mcpu=power6}, @option{-mcpu=power7} or 19854@option{-mcpu=power8} automatically selects @option{-mrecip-precision}. 19855The double-precision square root estimate instructions are not generated by 19856default on low-precision machines, since they do not provide an 19857estimate that converges after three steps. 19858 19859@item -mveclibabi=@var{type} 19860@opindex mveclibabi 19861Specifies the ABI type to use for vectorizing intrinsics using an 19862external library. The only type supported at present is @code{mass}, 19863which specifies to use IBM's Mathematical Acceleration Subsystem 19864(MASS) libraries for vectorizing intrinsics using external libraries. 19865GCC currently emits calls to @code{acosd2}, @code{acosf4}, 19866@code{acoshd2}, @code{acoshf4}, @code{asind2}, @code{asinf4}, 19867@code{asinhd2}, @code{asinhf4}, @code{atan2d2}, @code{atan2f4}, 19868@code{atand2}, @code{atanf4}, @code{atanhd2}, @code{atanhf4}, 19869@code{cbrtd2}, @code{cbrtf4}, @code{cosd2}, @code{cosf4}, 19870@code{coshd2}, @code{coshf4}, @code{erfcd2}, @code{erfcf4}, 19871@code{erfd2}, @code{erff4}, @code{exp2d2}, @code{exp2f4}, 19872@code{expd2}, @code{expf4}, @code{expm1d2}, @code{expm1f4}, 19873@code{hypotd2}, @code{hypotf4}, @code{lgammad2}, @code{lgammaf4}, 19874@code{log10d2}, @code{log10f4}, @code{log1pd2}, @code{log1pf4}, 19875@code{log2d2}, @code{log2f4}, @code{logd2}, @code{logf4}, 19876@code{powd2}, @code{powf4}, @code{sind2}, @code{sinf4}, @code{sinhd2}, 19877@code{sinhf4}, @code{sqrtd2}, @code{sqrtf4}, @code{tand2}, 19878@code{tanf4}, @code{tanhd2}, and @code{tanhf4} when generating code 19879for power7. Both @option{-ftree-vectorize} and 19880@option{-funsafe-math-optimizations} must also be enabled. The MASS 19881libraries must be specified at link time. 19882 19883@item -mfriz 19884@itemx -mno-friz 19885@opindex mfriz 19886Generate (do not generate) the @code{friz} instruction when the 19887@option{-funsafe-math-optimizations} option is used to optimize 19888rounding of floating-point values to 64-bit integer and back to floating 19889point. The @code{friz} instruction does not return the same value if 19890the floating-point number is too large to fit in an integer. 19891 19892@item -mpointers-to-nested-functions 19893@itemx -mno-pointers-to-nested-functions 19894@opindex mpointers-to-nested-functions 19895Generate (do not generate) code to load up the static chain register 19896(@var{r11}) when calling through a pointer on AIX and 64-bit Linux 19897systems where a function pointer points to a 3-word descriptor giving 19898the function address, TOC value to be loaded in register @var{r2}, and 19899static chain value to be loaded in register @var{r11}. The 19900@option{-mpointers-to-nested-functions} is on by default. You cannot 19901call through pointers to nested functions or pointers 19902to functions compiled in other languages that use the static chain if 19903you use the @option{-mno-pointers-to-nested-functions}. 19904 19905@item -msave-toc-indirect 19906@itemx -mno-save-toc-indirect 19907@opindex msave-toc-indirect 19908Generate (do not generate) code to save the TOC value in the reserved 19909stack location in the function prologue if the function calls through 19910a pointer on AIX and 64-bit Linux systems. If the TOC value is not 19911saved in the prologue, it is saved just before the call through the 19912pointer. The @option{-mno-save-toc-indirect} option is the default. 19913 19914@item -mcompat-align-parm 19915@itemx -mno-compat-align-parm 19916@opindex mcompat-align-parm 19917Generate (do not generate) code to pass structure parameters with a 19918maximum alignment of 64 bits, for compatibility with older versions 19919of GCC. 19920 19921Older versions of GCC (prior to 4.9.0) incorrectly did not align a 19922structure parameter on a 128-bit boundary when that structure contained 19923a member requiring 128-bit alignment. This is corrected in more 19924recent versions of GCC. This option may be used to generate code 19925that is compatible with functions compiled with older versions of 19926GCC. 19927 19928The @option{-mno-compat-align-parm} option is the default. 19929@end table 19930 19931@node RX Options 19932@subsection RX Options 19933@cindex RX Options 19934 19935These command-line options are defined for RX targets: 19936 19937@table @gcctabopt 19938@item -m64bit-doubles 19939@itemx -m32bit-doubles 19940@opindex m64bit-doubles 19941@opindex m32bit-doubles 19942Make the @code{double} data type be 64 bits (@option{-m64bit-doubles}) 19943or 32 bits (@option{-m32bit-doubles}) in size. The default is 19944@option{-m32bit-doubles}. @emph{Note} RX floating-point hardware only 19945works on 32-bit values, which is why the default is 19946@option{-m32bit-doubles}. 19947 19948@item -fpu 19949@itemx -nofpu 19950@opindex fpu 19951@opindex nofpu 19952Enables (@option{-fpu}) or disables (@option{-nofpu}) the use of RX 19953floating-point hardware. The default is enabled for the @var{RX600} 19954series and disabled for the @var{RX200} series. 19955 19956Floating-point instructions are only generated for 32-bit floating-point 19957values, however, so the FPU hardware is not used for doubles if the 19958@option{-m64bit-doubles} option is used. 19959 19960@emph{Note} If the @option{-fpu} option is enabled then 19961@option{-funsafe-math-optimizations} is also enabled automatically. 19962This is because the RX FPU instructions are themselves unsafe. 19963 19964@item -mcpu=@var{name} 19965@opindex -mcpu 19966Selects the type of RX CPU to be targeted. Currently three types are 19967supported, the generic @var{RX600} and @var{RX200} series hardware and 19968the specific @var{RX610} CPU. The default is @var{RX600}. 19969 19970The only difference between @var{RX600} and @var{RX610} is that the 19971@var{RX610} does not support the @code{MVTIPL} instruction. 19972 19973The @var{RX200} series does not have a hardware floating-point unit 19974and so @option{-nofpu} is enabled by default when this type is 19975selected. 19976 19977@item -mbig-endian-data 19978@itemx -mlittle-endian-data 19979@opindex mbig-endian-data 19980@opindex mlittle-endian-data 19981Store data (but not code) in the big-endian format. The default is 19982@option{-mlittle-endian-data}, i.e.@: to store data in the little-endian 19983format. 19984 19985@item -msmall-data-limit=@var{N} 19986@opindex msmall-data-limit 19987Specifies the maximum size in bytes of global and static variables 19988which can be placed into the small data area. Using the small data 19989area can lead to smaller and faster code, but the size of area is 19990limited and it is up to the programmer to ensure that the area does 19991not overflow. Also when the small data area is used one of the RX's 19992registers (usually @code{r13}) is reserved for use pointing to this 19993area, so it is no longer available for use by the compiler. This 19994could result in slower and/or larger code if variables are pushed onto 19995the stack instead of being held in this register. 19996 19997Note, common variables (variables that have not been initialized) and 19998constants are not placed into the small data area as they are assigned 19999to other sections in the output executable. 20000 20001The default value is zero, which disables this feature. Note, this 20002feature is not enabled by default with higher optimization levels 20003(@option{-O2} etc) because of the potentially detrimental effects of 20004reserving a register. It is up to the programmer to experiment and 20005discover whether this feature is of benefit to their program. See the 20006description of the @option{-mpid} option for a description of how the 20007actual register to hold the small data area pointer is chosen. 20008 20009@item -msim 20010@itemx -mno-sim 20011@opindex msim 20012@opindex mno-sim 20013Use the simulator runtime. The default is to use the libgloss 20014board-specific runtime. 20015 20016@item -mas100-syntax 20017@itemx -mno-as100-syntax 20018@opindex mas100-syntax 20019@opindex mno-as100-syntax 20020When generating assembler output use a syntax that is compatible with 20021Renesas's AS100 assembler. This syntax can also be handled by the GAS 20022assembler, but it has some restrictions so it is not generated by default. 20023 20024@item -mmax-constant-size=@var{N} 20025@opindex mmax-constant-size 20026Specifies the maximum size, in bytes, of a constant that can be used as 20027an operand in a RX instruction. Although the RX instruction set does 20028allow constants of up to 4 bytes in length to be used in instructions, 20029a longer value equates to a longer instruction. Thus in some 20030circumstances it can be beneficial to restrict the size of constants 20031that are used in instructions. Constants that are too big are instead 20032placed into a constant pool and referenced via register indirection. 20033 20034The value @var{N} can be between 0 and 4. A value of 0 (the default) 20035or 4 means that constants of any size are allowed. 20036 20037@item -mrelax 20038@opindex mrelax 20039Enable linker relaxation. Linker relaxation is a process whereby the 20040linker attempts to reduce the size of a program by finding shorter 20041versions of various instructions. Disabled by default. 20042 20043@item -mint-register=@var{N} 20044@opindex mint-register 20045Specify the number of registers to reserve for fast interrupt handler 20046functions. The value @var{N} can be between 0 and 4. A value of 1 20047means that register @code{r13} is reserved for the exclusive use 20048of fast interrupt handlers. A value of 2 reserves @code{r13} and 20049@code{r12}. A value of 3 reserves @code{r13}, @code{r12} and 20050@code{r11}, and a value of 4 reserves @code{r13} through @code{r10}. 20051A value of 0, the default, does not reserve any registers. 20052 20053@item -msave-acc-in-interrupts 20054@opindex msave-acc-in-interrupts 20055Specifies that interrupt handler functions should preserve the 20056accumulator register. This is only necessary if normal code might use 20057the accumulator register, for example because it performs 64-bit 20058multiplications. The default is to ignore the accumulator as this 20059makes the interrupt handlers faster. 20060 20061@item -mpid 20062@itemx -mno-pid 20063@opindex mpid 20064@opindex mno-pid 20065Enables the generation of position independent data. When enabled any 20066access to constant data is done via an offset from a base address 20067held in a register. This allows the location of constant data to be 20068determined at run time without requiring the executable to be 20069relocated, which is a benefit to embedded applications with tight 20070memory constraints. Data that can be modified is not affected by this 20071option. 20072 20073Note, using this feature reserves a register, usually @code{r13}, for 20074the constant data base address. This can result in slower and/or 20075larger code, especially in complicated functions. 20076 20077The actual register chosen to hold the constant data base address 20078depends upon whether the @option{-msmall-data-limit} and/or the 20079@option{-mint-register} command-line options are enabled. Starting 20080with register @code{r13} and proceeding downwards, registers are 20081allocated first to satisfy the requirements of @option{-mint-register}, 20082then @option{-mpid} and finally @option{-msmall-data-limit}. Thus it 20083is possible for the small data area register to be @code{r8} if both 20084@option{-mint-register=4} and @option{-mpid} are specified on the 20085command line. 20086 20087By default this feature is not enabled. The default can be restored 20088via the @option{-mno-pid} command-line option. 20089 20090@item -mno-warn-multiple-fast-interrupts 20091@itemx -mwarn-multiple-fast-interrupts 20092@opindex mno-warn-multiple-fast-interrupts 20093@opindex mwarn-multiple-fast-interrupts 20094Prevents GCC from issuing a warning message if it finds more than one 20095fast interrupt handler when it is compiling a file. The default is to 20096issue a warning for each extra fast interrupt handler found, as the RX 20097only supports one such interrupt. 20098 20099@end table 20100 20101@emph{Note:} The generic GCC command-line option @option{-ffixed-@var{reg}} 20102has special significance to the RX port when used with the 20103@code{interrupt} function attribute. This attribute indicates a 20104function intended to process fast interrupts. GCC ensures 20105that it only uses the registers @code{r10}, @code{r11}, @code{r12} 20106and/or @code{r13} and only provided that the normal use of the 20107corresponding registers have been restricted via the 20108@option{-ffixed-@var{reg}} or @option{-mint-register} command-line 20109options. 20110 20111@node S/390 and zSeries Options 20112@subsection S/390 and zSeries Options 20113@cindex S/390 and zSeries Options 20114 20115These are the @samp{-m} options defined for the S/390 and zSeries architecture. 20116 20117@table @gcctabopt 20118@item -mhard-float 20119@itemx -msoft-float 20120@opindex mhard-float 20121@opindex msoft-float 20122Use (do not use) the hardware floating-point instructions and registers 20123for floating-point operations. When @option{-msoft-float} is specified, 20124functions in @file{libgcc.a} are used to perform floating-point 20125operations. When @option{-mhard-float} is specified, the compiler 20126generates IEEE floating-point instructions. This is the default. 20127 20128@item -mhard-dfp 20129@itemx -mno-hard-dfp 20130@opindex mhard-dfp 20131@opindex mno-hard-dfp 20132Use (do not use) the hardware decimal-floating-point instructions for 20133decimal-floating-point operations. When @option{-mno-hard-dfp} is 20134specified, functions in @file{libgcc.a} are used to perform 20135decimal-floating-point operations. When @option{-mhard-dfp} is 20136specified, the compiler generates decimal-floating-point hardware 20137instructions. This is the default for @option{-march=z9-ec} or higher. 20138 20139@item -mlong-double-64 20140@itemx -mlong-double-128 20141@opindex mlong-double-64 20142@opindex mlong-double-128 20143These switches control the size of @code{long double} type. A size 20144of 64 bits makes the @code{long double} type equivalent to the @code{double} 20145type. This is the default. 20146 20147@item -mbackchain 20148@itemx -mno-backchain 20149@opindex mbackchain 20150@opindex mno-backchain 20151Store (do not store) the address of the caller's frame as backchain pointer 20152into the callee's stack frame. 20153A backchain may be needed to allow debugging using tools that do not understand 20154DWARF 2 call frame information. 20155When @option{-mno-packed-stack} is in effect, the backchain pointer is stored 20156at the bottom of the stack frame; when @option{-mpacked-stack} is in effect, 20157the backchain is placed into the topmost word of the 96/160 byte register 20158save area. 20159 20160In general, code compiled with @option{-mbackchain} is call-compatible with 20161code compiled with @option{-mmo-backchain}; however, use of the backchain 20162for debugging purposes usually requires that the whole binary is built with 20163@option{-mbackchain}. Note that the combination of @option{-mbackchain}, 20164@option{-mpacked-stack} and @option{-mhard-float} is not supported. In order 20165to build a linux kernel use @option{-msoft-float}. 20166 20167The default is to not maintain the backchain. 20168 20169@item -mpacked-stack 20170@itemx -mno-packed-stack 20171@opindex mpacked-stack 20172@opindex mno-packed-stack 20173Use (do not use) the packed stack layout. When @option{-mno-packed-stack} is 20174specified, the compiler uses the all fields of the 96/160 byte register save 20175area only for their default purpose; unused fields still take up stack space. 20176When @option{-mpacked-stack} is specified, register save slots are densely 20177packed at the top of the register save area; unused space is reused for other 20178purposes, allowing for more efficient use of the available stack space. 20179However, when @option{-mbackchain} is also in effect, the topmost word of 20180the save area is always used to store the backchain, and the return address 20181register is always saved two words below the backchain. 20182 20183As long as the stack frame backchain is not used, code generated with 20184@option{-mpacked-stack} is call-compatible with code generated with 20185@option{-mno-packed-stack}. Note that some non-FSF releases of GCC 2.95 for 20186S/390 or zSeries generated code that uses the stack frame backchain at run 20187time, not just for debugging purposes. Such code is not call-compatible 20188with code compiled with @option{-mpacked-stack}. Also, note that the 20189combination of @option{-mbackchain}, 20190@option{-mpacked-stack} and @option{-mhard-float} is not supported. In order 20191to build a linux kernel use @option{-msoft-float}. 20192 20193The default is to not use the packed stack layout. 20194 20195@item -msmall-exec 20196@itemx -mno-small-exec 20197@opindex msmall-exec 20198@opindex mno-small-exec 20199Generate (or do not generate) code using the @code{bras} instruction 20200to do subroutine calls. 20201This only works reliably if the total executable size does not 20202exceed 64k. The default is to use the @code{basr} instruction instead, 20203which does not have this limitation. 20204 20205@item -m64 20206@itemx -m31 20207@opindex m64 20208@opindex m31 20209When @option{-m31} is specified, generate code compliant to the 20210GNU/Linux for S/390 ABI@. When @option{-m64} is specified, generate 20211code compliant to the GNU/Linux for zSeries ABI@. This allows GCC in 20212particular to generate 64-bit instructions. For the @samp{s390} 20213targets, the default is @option{-m31}, while the @samp{s390x} 20214targets default to @option{-m64}. 20215 20216@item -mzarch 20217@itemx -mesa 20218@opindex mzarch 20219@opindex mesa 20220When @option{-mzarch} is specified, generate code using the 20221instructions available on z/Architecture. 20222When @option{-mesa} is specified, generate code using the 20223instructions available on ESA/390. Note that @option{-mesa} is 20224not possible with @option{-m64}. 20225When generating code compliant to the GNU/Linux for S/390 ABI, 20226the default is @option{-mesa}. When generating code compliant 20227to the GNU/Linux for zSeries ABI, the default is @option{-mzarch}. 20228 20229@item -mmvcle 20230@itemx -mno-mvcle 20231@opindex mmvcle 20232@opindex mno-mvcle 20233Generate (or do not generate) code using the @code{mvcle} instruction 20234to perform block moves. When @option{-mno-mvcle} is specified, 20235use a @code{mvc} loop instead. This is the default unless optimizing for 20236size. 20237 20238@item -mdebug 20239@itemx -mno-debug 20240@opindex mdebug 20241@opindex mno-debug 20242Print (or do not print) additional debug information when compiling. 20243The default is to not print debug information. 20244 20245@item -march=@var{cpu-type} 20246@opindex march 20247Generate code that runs on @var{cpu-type}, which is the name of a system 20248representing a certain processor type. Possible values for 20249@var{cpu-type} are @samp{g5}, @samp{g6}, @samp{z900}, @samp{z990}, 20250@samp{z9-109}, @samp{z9-ec} and @samp{z10}. 20251When generating code using the instructions available on z/Architecture, 20252the default is @option{-march=z900}. Otherwise, the default is 20253@option{-march=g5}. 20254 20255@item -mtune=@var{cpu-type} 20256@opindex mtune 20257Tune to @var{cpu-type} everything applicable about the generated code, 20258except for the ABI and the set of available instructions. 20259The list of @var{cpu-type} values is the same as for @option{-march}. 20260The default is the value used for @option{-march}. 20261 20262@item -mtpf-trace 20263@itemx -mno-tpf-trace 20264@opindex mtpf-trace 20265@opindex mno-tpf-trace 20266Generate code that adds (does not add) in TPF OS specific branches to trace 20267routines in the operating system. This option is off by default, even 20268when compiling for the TPF OS@. 20269 20270@item -mfused-madd 20271@itemx -mno-fused-madd 20272@opindex mfused-madd 20273@opindex mno-fused-madd 20274Generate code that uses (does not use) the floating-point multiply and 20275accumulate instructions. These instructions are generated by default if 20276hardware floating point is used. 20277 20278@item -mwarn-framesize=@var{framesize} 20279@opindex mwarn-framesize 20280Emit a warning if the current function exceeds the given frame size. Because 20281this is a compile-time check it doesn't need to be a real problem when the program 20282runs. It is intended to identify functions that most probably cause 20283a stack overflow. It is useful to be used in an environment with limited stack 20284size e.g.@: the linux kernel. 20285 20286@item -mwarn-dynamicstack 20287@opindex mwarn-dynamicstack 20288Emit a warning if the function calls @code{alloca} or uses dynamically-sized 20289arrays. This is generally a bad idea with a limited stack size. 20290 20291@item -mstack-guard=@var{stack-guard} 20292@itemx -mstack-size=@var{stack-size} 20293@opindex mstack-guard 20294@opindex mstack-size 20295If these options are provided the S/390 back end emits additional instructions in 20296the function prologue that trigger a trap if the stack size is @var{stack-guard} 20297bytes above the @var{stack-size} (remember that the stack on S/390 grows downward). 20298If the @var{stack-guard} option is omitted the smallest power of 2 larger than 20299the frame size of the compiled function is chosen. 20300These options are intended to be used to help debugging stack overflow problems. 20301The additionally emitted code causes only little overhead and hence can also be 20302used in production-like systems without greater performance degradation. The given 20303values have to be exact powers of 2 and @var{stack-size} has to be greater than 20304@var{stack-guard} without exceeding 64k. 20305In order to be efficient the extra code makes the assumption that the stack starts 20306at an address aligned to the value given by @var{stack-size}. 20307The @var{stack-guard} option can only be used in conjunction with @var{stack-size}. 20308 20309@item -mhotpatch[=@var{halfwords}] 20310@itemx -mno-hotpatch 20311@opindex mhotpatch 20312If the hotpatch option is enabled, a ``hot-patching'' function 20313prologue is generated for all functions in the compilation unit. 20314The funtion label is prepended with the given number of two-byte 20315Nop instructions (@var{halfwords}, maximum 1000000) or 12 Nop 20316instructions if no argument is present. Functions with a 20317hot-patching prologue are never inlined automatically, and a 20318hot-patching prologue is never generated for functions functions 20319that are explicitly inline. 20320 20321This option can be overridden for individual functions with the 20322@code{hotpatch} attribute. 20323@end table 20324 20325@node Score Options 20326@subsection Score Options 20327@cindex Score Options 20328 20329These options are defined for Score implementations: 20330 20331@table @gcctabopt 20332@item -meb 20333@opindex meb 20334Compile code for big-endian mode. This is the default. 20335 20336@item -mel 20337@opindex mel 20338Compile code for little-endian mode. 20339 20340@item -mnhwloop 20341@opindex mnhwloop 20342Disable generation of @code{bcnz} instructions. 20343 20344@item -muls 20345@opindex muls 20346Enable generation of unaligned load and store instructions. 20347 20348@item -mmac 20349@opindex mmac 20350Enable the use of multiply-accumulate instructions. Disabled by default. 20351 20352@item -mscore5 20353@opindex mscore5 20354Specify the SCORE5 as the target architecture. 20355 20356@item -mscore5u 20357@opindex mscore5u 20358Specify the SCORE5U of the target architecture. 20359 20360@item -mscore7 20361@opindex mscore7 20362Specify the SCORE7 as the target architecture. This is the default. 20363 20364@item -mscore7d 20365@opindex mscore7d 20366Specify the SCORE7D as the target architecture. 20367@end table 20368 20369@node SH Options 20370@subsection SH Options 20371 20372These @samp{-m} options are defined for the SH implementations: 20373 20374@table @gcctabopt 20375@item -m1 20376@opindex m1 20377Generate code for the SH1. 20378 20379@item -m2 20380@opindex m2 20381Generate code for the SH2. 20382 20383@item -m2e 20384Generate code for the SH2e. 20385 20386@item -m2a-nofpu 20387@opindex m2a-nofpu 20388Generate code for the SH2a without FPU, or for a SH2a-FPU in such a way 20389that the floating-point unit is not used. 20390 20391@item -m2a-single-only 20392@opindex m2a-single-only 20393Generate code for the SH2a-FPU, in such a way that no double-precision 20394floating-point operations are used. 20395 20396@item -m2a-single 20397@opindex m2a-single 20398Generate code for the SH2a-FPU assuming the floating-point unit is in 20399single-precision mode by default. 20400 20401@item -m2a 20402@opindex m2a 20403Generate code for the SH2a-FPU assuming the floating-point unit is in 20404double-precision mode by default. 20405 20406@item -m3 20407@opindex m3 20408Generate code for the SH3. 20409 20410@item -m3e 20411@opindex m3e 20412Generate code for the SH3e. 20413 20414@item -m4-nofpu 20415@opindex m4-nofpu 20416Generate code for the SH4 without a floating-point unit. 20417 20418@item -m4-single-only 20419@opindex m4-single-only 20420Generate code for the SH4 with a floating-point unit that only 20421supports single-precision arithmetic. 20422 20423@item -m4-single 20424@opindex m4-single 20425Generate code for the SH4 assuming the floating-point unit is in 20426single-precision mode by default. 20427 20428@item -m4 20429@opindex m4 20430Generate code for the SH4. 20431 20432@item -m4-100 20433@opindex m4-100 20434Generate code for SH4-100. 20435 20436@item -m4-100-nofpu 20437@opindex m4-100-nofpu 20438Generate code for SH4-100 in such a way that the 20439floating-point unit is not used. 20440 20441@item -m4-100-single 20442@opindex m4-100-single 20443Generate code for SH4-100 assuming the floating-point unit is in 20444single-precision mode by default. 20445 20446@item -m4-100-single-only 20447@opindex m4-100-single-only 20448Generate code for SH4-100 in such a way that no double-precision 20449floating-point operations are used. 20450 20451@item -m4-200 20452@opindex m4-200 20453Generate code for SH4-200. 20454 20455@item -m4-200-nofpu 20456@opindex m4-200-nofpu 20457Generate code for SH4-200 without in such a way that the 20458floating-point unit is not used. 20459 20460@item -m4-200-single 20461@opindex m4-200-single 20462Generate code for SH4-200 assuming the floating-point unit is in 20463single-precision mode by default. 20464 20465@item -m4-200-single-only 20466@opindex m4-200-single-only 20467Generate code for SH4-200 in such a way that no double-precision 20468floating-point operations are used. 20469 20470@item -m4-300 20471@opindex m4-300 20472Generate code for SH4-300. 20473 20474@item -m4-300-nofpu 20475@opindex m4-300-nofpu 20476Generate code for SH4-300 without in such a way that the 20477floating-point unit is not used. 20478 20479@item -m4-300-single 20480@opindex m4-300-single 20481Generate code for SH4-300 in such a way that no double-precision 20482floating-point operations are used. 20483 20484@item -m4-300-single-only 20485@opindex m4-300-single-only 20486Generate code for SH4-300 in such a way that no double-precision 20487floating-point operations are used. 20488 20489@item -m4-340 20490@opindex m4-340 20491Generate code for SH4-340 (no MMU, no FPU). 20492 20493@item -m4-500 20494@opindex m4-500 20495Generate code for SH4-500 (no FPU). Passes @option{-isa=sh4-nofpu} to the 20496assembler. 20497 20498@item -m4a-nofpu 20499@opindex m4a-nofpu 20500Generate code for the SH4al-dsp, or for a SH4a in such a way that the 20501floating-point unit is not used. 20502 20503@item -m4a-single-only 20504@opindex m4a-single-only 20505Generate code for the SH4a, in such a way that no double-precision 20506floating-point operations are used. 20507 20508@item -m4a-single 20509@opindex m4a-single 20510Generate code for the SH4a assuming the floating-point unit is in 20511single-precision mode by default. 20512 20513@item -m4a 20514@opindex m4a 20515Generate code for the SH4a. 20516 20517@item -m4al 20518@opindex m4al 20519Same as @option{-m4a-nofpu}, except that it implicitly passes 20520@option{-dsp} to the assembler. GCC doesn't generate any DSP 20521instructions at the moment. 20522 20523@item -m5-32media 20524@opindex m5-32media 20525Generate 32-bit code for SHmedia. 20526 20527@item -m5-32media-nofpu 20528@opindex m5-32media-nofpu 20529Generate 32-bit code for SHmedia in such a way that the 20530floating-point unit is not used. 20531 20532@item -m5-64media 20533@opindex m5-64media 20534Generate 64-bit code for SHmedia. 20535 20536@item -m5-64media-nofpu 20537@opindex m5-64media-nofpu 20538Generate 64-bit code for SHmedia in such a way that the 20539floating-point unit is not used. 20540 20541@item -m5-compact 20542@opindex m5-compact 20543Generate code for SHcompact. 20544 20545@item -m5-compact-nofpu 20546@opindex m5-compact-nofpu 20547Generate code for SHcompact in such a way that the 20548floating-point unit is not used. 20549 20550@item -mb 20551@opindex mb 20552Compile code for the processor in big-endian mode. 20553 20554@item -ml 20555@opindex ml 20556Compile code for the processor in little-endian mode. 20557 20558@item -mdalign 20559@opindex mdalign 20560Align doubles at 64-bit boundaries. Note that this changes the calling 20561conventions, and thus some functions from the standard C library do 20562not work unless you recompile it first with @option{-mdalign}. 20563 20564@item -mrelax 20565@opindex mrelax 20566Shorten some address references at link time, when possible; uses the 20567linker option @option{-relax}. 20568 20569@item -mbigtable 20570@opindex mbigtable 20571Use 32-bit offsets in @code{switch} tables. The default is to use 2057216-bit offsets. 20573 20574@item -mbitops 20575@opindex mbitops 20576Enable the use of bit manipulation instructions on SH2A. 20577 20578@item -mfmovd 20579@opindex mfmovd 20580Enable the use of the instruction @code{fmovd}. Check @option{-mdalign} for 20581alignment constraints. 20582 20583@item -mrenesas 20584@opindex mrenesas 20585Comply with the calling conventions defined by Renesas. 20586 20587@item -mno-renesas 20588@opindex mno-renesas 20589Comply with the calling conventions defined for GCC before the Renesas 20590conventions were available. This option is the default for all 20591targets of the SH toolchain. 20592 20593@item -mnomacsave 20594@opindex mnomacsave 20595Mark the @code{MAC} register as call-clobbered, even if 20596@option{-mrenesas} is given. 20597 20598@item -mieee 20599@itemx -mno-ieee 20600@opindex mieee 20601@opindex mno-ieee 20602Control the IEEE compliance of floating-point comparisons, which affects the 20603handling of cases where the result of a comparison is unordered. By default 20604@option{-mieee} is implicitly enabled. If @option{-ffinite-math-only} is 20605enabled @option{-mno-ieee} is implicitly set, which results in faster 20606floating-point greater-equal and less-equal comparisons. The implcit settings 20607can be overridden by specifying either @option{-mieee} or @option{-mno-ieee}. 20608 20609@item -minline-ic_invalidate 20610@opindex minline-ic_invalidate 20611Inline code to invalidate instruction cache entries after setting up 20612nested function trampolines. 20613This option has no effect if @option{-musermode} is in effect and the selected 20614code generation option (e.g. @option{-m4}) does not allow the use of the @code{icbi} 20615instruction. 20616If the selected code generation option does not allow the use of the @code{icbi} 20617instruction, and @option{-musermode} is not in effect, the inlined code 20618manipulates the instruction cache address array directly with an associative 20619write. This not only requires privileged mode at run time, but it also 20620fails if the cache line had been mapped via the TLB and has become unmapped. 20621 20622@item -misize 20623@opindex misize 20624Dump instruction size and location in the assembly code. 20625 20626@item -mpadstruct 20627@opindex mpadstruct 20628This option is deprecated. It pads structures to multiple of 4 bytes, 20629which is incompatible with the SH ABI@. 20630 20631@item -matomic-model=@var{model} 20632@opindex matomic-model=@var{model} 20633Sets the model of atomic operations and additional parameters as a comma 20634separated list. For details on the atomic built-in functions see 20635@ref{__atomic Builtins}. The following models and parameters are supported: 20636 20637@table @samp 20638 20639@item none 20640Disable compiler generated atomic sequences and emit library calls for atomic 20641operations. This is the default if the target is not @code{sh*-*-linux*}. 20642 20643@item soft-gusa 20644Generate GNU/Linux compatible gUSA software atomic sequences for the atomic 20645built-in functions. The generated atomic sequences require additional support 20646from the interrupt/exception handling code of the system and are only suitable 20647for SH3* and SH4* single-core systems. This option is enabled by default when 20648the target is @code{sh*-*-linux*} and SH3* or SH4*. When the target is SH4A, 20649this option will also partially utilize the hardware atomic instructions 20650@code{movli.l} and @code{movco.l} to create more efficient code, unless 20651@samp{strict} is specified. 20652 20653@item soft-tcb 20654Generate software atomic sequences that use a variable in the thread control 20655block. This is a variation of the gUSA sequences which can also be used on 20656SH1* and SH2* targets. The generated atomic sequences require additional 20657support from the interrupt/exception handling code of the system and are only 20658suitable for single-core systems. When using this model, the @samp{gbr-offset=} 20659parameter has to be specified as well. 20660 20661@item soft-imask 20662Generate software atomic sequences that temporarily disable interrupts by 20663setting @code{SR.IMASK = 1111}. This model works only when the program runs 20664in privileged mode and is only suitable for single-core systems. Additional 20665support from the interrupt/exception handling code of the system is not 20666required. This model is enabled by default when the target is 20667@code{sh*-*-linux*} and SH1* or SH2*. 20668 20669@item hard-llcs 20670Generate hardware atomic sequences using the @code{movli.l} and @code{movco.l} 20671instructions only. This is only available on SH4A and is suitable for 20672multi-core systems. Since the hardware instructions support only 32 bit atomic 20673variables access to 8 or 16 bit variables is emulated with 32 bit accesses. 20674Code compiled with this option will also be compatible with other software 20675atomic model interrupt/exception handling systems if executed on an SH4A 20676system. Additional support from the interrupt/exception handling code of the 20677system is not required for this model. 20678 20679@item gbr-offset= 20680This parameter specifies the offset in bytes of the variable in the thread 20681control block structure that should be used by the generated atomic sequences 20682when the @samp{soft-tcb} model has been selected. For other models this 20683parameter is ignored. The specified value must be an integer multiple of four 20684and in the range 0-1020. 20685 20686@item strict 20687This parameter prevents mixed usage of multiple atomic models, even though they 20688would be compatible, and will make the compiler generate atomic sequences of the 20689specified model only. 20690 20691@end table 20692 20693@item -mtas 20694@opindex mtas 20695Generate the @code{tas.b} opcode for @code{__atomic_test_and_set}. 20696Notice that depending on the particular hardware and software configuration 20697this can degrade overall performance due to the operand cache line flushes 20698that are implied by the @code{tas.b} instruction. On multi-core SH4A 20699processors the @code{tas.b} instruction must be used with caution since it 20700can result in data corruption for certain cache configurations. 20701 20702@item -mprefergot 20703@opindex mprefergot 20704When generating position-independent code, emit function calls using 20705the Global Offset Table instead of the Procedure Linkage Table. 20706 20707@item -musermode 20708@itemx -mno-usermode 20709@opindex musermode 20710@opindex mno-usermode 20711Don't allow (allow) the compiler generating privileged mode code. Specifying 20712@option{-musermode} also implies @option{-mno-inline-ic_invalidate} if the 20713inlined code would not work in user mode. @option{-musermode} is the default 20714when the target is @code{sh*-*-linux*}. If the target is SH1* or SH2* 20715@option{-musermode} has no effect, since there is no user mode. 20716 20717@item -multcost=@var{number} 20718@opindex multcost=@var{number} 20719Set the cost to assume for a multiply insn. 20720 20721@item -mdiv=@var{strategy} 20722@opindex mdiv=@var{strategy} 20723Set the division strategy to be used for integer division operations. 20724For SHmedia @var{strategy} can be one of: 20725 20726@table @samp 20727 20728@item fp 20729Performs the operation in floating point. This has a very high latency, 20730but needs only a few instructions, so it might be a good choice if 20731your code has enough easily-exploitable ILP to allow the compiler to 20732schedule the floating-point instructions together with other instructions. 20733Division by zero causes a floating-point exception. 20734 20735@item inv 20736Uses integer operations to calculate the inverse of the divisor, 20737and then multiplies the dividend with the inverse. This strategy allows 20738CSE and hoisting of the inverse calculation. Division by zero calculates 20739an unspecified result, but does not trap. 20740 20741@item inv:minlat 20742A variant of @samp{inv} where, if no CSE or hoisting opportunities 20743have been found, or if the entire operation has been hoisted to the same 20744place, the last stages of the inverse calculation are intertwined with the 20745final multiply to reduce the overall latency, at the expense of using a few 20746more instructions, and thus offering fewer scheduling opportunities with 20747other code. 20748 20749@item call 20750Calls a library function that usually implements the @samp{inv:minlat} 20751strategy. 20752This gives high code density for @code{m5-*media-nofpu} compilations. 20753 20754@item call2 20755Uses a different entry point of the same library function, where it 20756assumes that a pointer to a lookup table has already been set up, which 20757exposes the pointer load to CSE and code hoisting optimizations. 20758 20759@item inv:call 20760@itemx inv:call2 20761@itemx inv:fp 20762Use the @samp{inv} algorithm for initial 20763code generation, but if the code stays unoptimized, revert to the @samp{call}, 20764@samp{call2}, or @samp{fp} strategies, respectively. Note that the 20765potentially-trapping side effect of division by zero is carried by a 20766separate instruction, so it is possible that all the integer instructions 20767are hoisted out, but the marker for the side effect stays where it is. 20768A recombination to floating-point operations or a call is not possible 20769in that case. 20770 20771@item inv20u 20772@itemx inv20l 20773Variants of the @samp{inv:minlat} strategy. In the case 20774that the inverse calculation is not separated from the multiply, they speed 20775up division where the dividend fits into 20 bits (plus sign where applicable) 20776by inserting a test to skip a number of operations in this case; this test 20777slows down the case of larger dividends. @samp{inv20u} assumes the case of a such 20778a small dividend to be unlikely, and @samp{inv20l} assumes it to be likely. 20779 20780@end table 20781 20782For targets other than SHmedia @var{strategy} can be one of: 20783 20784@table @samp 20785 20786@item call-div1 20787Calls a library function that uses the single-step division instruction 20788@code{div1} to perform the operation. Division by zero calculates an 20789unspecified result and does not trap. This is the default except for SH4, 20790SH2A and SHcompact. 20791 20792@item call-fp 20793Calls a library function that performs the operation in double precision 20794floating point. Division by zero causes a floating-point exception. This is 20795the default for SHcompact with FPU. Specifying this for targets that do not 20796have a double precision FPU will default to @code{call-div1}. 20797 20798@item call-table 20799Calls a library function that uses a lookup table for small divisors and 20800the @code{div1} instruction with case distinction for larger divisors. Division 20801by zero calculates an unspecified result and does not trap. This is the default 20802for SH4. Specifying this for targets that do not have dynamic shift 20803instructions will default to @code{call-div1}. 20804 20805@end table 20806 20807When a division strategy has not been specified the default strategy will be 20808selected based on the current target. For SH2A the default strategy is to 20809use the @code{divs} and @code{divu} instructions instead of library function 20810calls. 20811 20812@item -maccumulate-outgoing-args 20813@opindex maccumulate-outgoing-args 20814Reserve space once for outgoing arguments in the function prologue rather 20815than around each call. Generally beneficial for performance and size. Also 20816needed for unwinding to avoid changing the stack frame around conditional code. 20817 20818@item -mdivsi3_libfunc=@var{name} 20819@opindex mdivsi3_libfunc=@var{name} 20820Set the name of the library function used for 32-bit signed division to 20821@var{name}. 20822This only affects the name used in the @samp{call} and @samp{inv:call} 20823division strategies, and the compiler still expects the same 20824sets of input/output/clobbered registers as if this option were not present. 20825 20826@item -mfixed-range=@var{register-range} 20827@opindex mfixed-range 20828Generate code treating the given register range as fixed registers. 20829A fixed register is one that the register allocator can not use. This is 20830useful when compiling kernel code. A register range is specified as 20831two registers separated by a dash. Multiple register ranges can be 20832specified separated by a comma. 20833 20834@item -mindexed-addressing 20835@opindex mindexed-addressing 20836Enable the use of the indexed addressing mode for SHmedia32/SHcompact. 20837This is only safe if the hardware and/or OS implement 32-bit wrap-around 20838semantics for the indexed addressing mode. The architecture allows the 20839implementation of processors with 64-bit MMU, which the OS could use to 20840get 32-bit addressing, but since no current hardware implementation supports 20841this or any other way to make the indexed addressing mode safe to use in 20842the 32-bit ABI, the default is @option{-mno-indexed-addressing}. 20843 20844@item -mgettrcost=@var{number} 20845@opindex mgettrcost=@var{number} 20846Set the cost assumed for the @code{gettr} instruction to @var{number}. 20847The default is 2 if @option{-mpt-fixed} is in effect, 100 otherwise. 20848 20849@item -mpt-fixed 20850@opindex mpt-fixed 20851Assume @code{pt*} instructions won't trap. This generally generates 20852better-scheduled code, but is unsafe on current hardware. 20853The current architecture 20854definition says that @code{ptabs} and @code{ptrel} trap when the target 20855anded with 3 is 3. 20856This has the unintentional effect of making it unsafe to schedule these 20857instructions before a branch, or hoist them out of a loop. For example, 20858@code{__do_global_ctors}, a part of @file{libgcc} 20859that runs constructors at program 20860startup, calls functions in a list which is delimited by @minus{}1. With the 20861@option{-mpt-fixed} option, the @code{ptabs} is done before testing against @minus{}1. 20862That means that all the constructors run a bit more quickly, but when 20863the loop comes to the end of the list, the program crashes because @code{ptabs} 20864loads @minus{}1 into a target register. 20865 20866Since this option is unsafe for any 20867hardware implementing the current architecture specification, the default 20868is @option{-mno-pt-fixed}. Unless specified explicitly with 20869@option{-mgettrcost}, @option{-mno-pt-fixed} also implies @option{-mgettrcost=100}; 20870this deters register allocation from using target registers for storing 20871ordinary integers. 20872 20873@item -minvalid-symbols 20874@opindex minvalid-symbols 20875Assume symbols might be invalid. Ordinary function symbols generated by 20876the compiler are always valid to load with 20877@code{movi}/@code{shori}/@code{ptabs} or 20878@code{movi}/@code{shori}/@code{ptrel}, 20879but with assembler and/or linker tricks it is possible 20880to generate symbols that cause @code{ptabs} or @code{ptrel} to trap. 20881This option is only meaningful when @option{-mno-pt-fixed} is in effect. 20882It prevents cross-basic-block CSE, hoisting and most scheduling 20883of symbol loads. The default is @option{-mno-invalid-symbols}. 20884 20885@item -mbranch-cost=@var{num} 20886@opindex mbranch-cost=@var{num} 20887Assume @var{num} to be the cost for a branch instruction. Higher numbers 20888make the compiler try to generate more branch-free code if possible. 20889If not specified the value is selected depending on the processor type that 20890is being compiled for. 20891 20892@item -mzdcbranch 20893@itemx -mno-zdcbranch 20894@opindex mzdcbranch 20895@opindex mno-zdcbranch 20896Assume (do not assume) that zero displacement conditional branch instructions 20897@code{bt} and @code{bf} are fast. If @option{-mzdcbranch} is specified, the 20898compiler will try to prefer zero displacement branch code sequences. This is 20899enabled by default when generating code for SH4 and SH4A. It can be explicitly 20900disabled by specifying @option{-mno-zdcbranch}. 20901 20902@item -mfused-madd 20903@itemx -mno-fused-madd 20904@opindex mfused-madd 20905@opindex mno-fused-madd 20906Generate code that uses (does not use) the floating-point multiply and 20907accumulate instructions. These instructions are generated by default 20908if hardware floating point is used. The machine-dependent 20909@option{-mfused-madd} option is now mapped to the machine-independent 20910@option{-ffp-contract=fast} option, and @option{-mno-fused-madd} is 20911mapped to @option{-ffp-contract=off}. 20912 20913@item -mfsca 20914@itemx -mno-fsca 20915@opindex mfsca 20916@opindex mno-fsca 20917Allow or disallow the compiler to emit the @code{fsca} instruction for sine 20918and cosine approximations. The option @code{-mfsca} must be used in 20919combination with @code{-funsafe-math-optimizations}. It is enabled by default 20920when generating code for SH4A. Using @code{-mno-fsca} disables sine and cosine 20921approximations even if @code{-funsafe-math-optimizations} is in effect. 20922 20923@item -mfsrra 20924@itemx -mno-fsrra 20925@opindex mfsrra 20926@opindex mno-fsrra 20927Allow or disallow the compiler to emit the @code{fsrra} instruction for 20928reciprocal square root approximations. The option @code{-mfsrra} must be used 20929in combination with @code{-funsafe-math-optimizations} and 20930@code{-ffinite-math-only}. It is enabled by default when generating code for 20931SH4A. Using @code{-mno-fsrra} disables reciprocal square root approximations 20932even if @code{-funsafe-math-optimizations} and @code{-ffinite-math-only} are 20933in effect. 20934 20935@item -mpretend-cmove 20936@opindex mpretend-cmove 20937Prefer zero-displacement conditional branches for conditional move instruction 20938patterns. This can result in faster code on the SH4 processor. 20939 20940@end table 20941 20942@node Solaris 2 Options 20943@subsection Solaris 2 Options 20944@cindex Solaris 2 options 20945 20946These @samp{-m} options are supported on Solaris 2: 20947 20948@table @gcctabopt 20949@item -mclear-hwcap 20950@opindex mclear-hwcap 20951@option{-mclear-hwcap} tells the compiler to remove the hardware 20952capabilities generated by the Solaris assembler. This is only necessary 20953when object files use ISA extensions not supported by the current 20954machine, but check at runtime whether or not to use them. 20955 20956@item -mimpure-text 20957@opindex mimpure-text 20958@option{-mimpure-text}, used in addition to @option{-shared}, tells 20959the compiler to not pass @option{-z text} to the linker when linking a 20960shared object. Using this option, you can link position-dependent 20961code into a shared object. 20962 20963@option{-mimpure-text} suppresses the ``relocations remain against 20964allocatable but non-writable sections'' linker error message. 20965However, the necessary relocations trigger copy-on-write, and the 20966shared object is not actually shared across processes. Instead of 20967using @option{-mimpure-text}, you should compile all source code with 20968@option{-fpic} or @option{-fPIC}. 20969 20970@end table 20971 20972These switches are supported in addition to the above on Solaris 2: 20973 20974@table @gcctabopt 20975@item -pthreads 20976@opindex pthreads 20977Add support for multithreading using the POSIX threads library. This 20978option sets flags for both the preprocessor and linker. This option does 20979not affect the thread safety of object code produced by the compiler or 20980that of libraries supplied with it. 20981 20982@item -pthread 20983@opindex pthread 20984This is a synonym for @option{-pthreads}. 20985@end table 20986 20987@node SPARC Options 20988@subsection SPARC Options 20989@cindex SPARC options 20990 20991These @samp{-m} options are supported on the SPARC: 20992 20993@table @gcctabopt 20994@item -mno-app-regs 20995@itemx -mapp-regs 20996@opindex mno-app-regs 20997@opindex mapp-regs 20998Specify @option{-mapp-regs} to generate output using the global registers 209992 through 4, which the SPARC SVR4 ABI reserves for applications. Like the 21000global register 1, each global register 2 through 4 is then treated as an 21001allocable register that is clobbered by function calls. This is the default. 21002 21003To be fully SVR4 ABI-compliant at the cost of some performance loss, 21004specify @option{-mno-app-regs}. You should compile libraries and system 21005software with this option. 21006 21007@item -mflat 21008@itemx -mno-flat 21009@opindex mflat 21010@opindex mno-flat 21011With @option{-mflat}, the compiler does not generate save/restore instructions 21012and uses a ``flat'' or single register window model. This model is compatible 21013with the regular register window model. The local registers and the input 21014registers (0--5) are still treated as ``call-saved'' registers and are 21015saved on the stack as needed. 21016 21017With @option{-mno-flat} (the default), the compiler generates save/restore 21018instructions (except for leaf functions). This is the normal operating mode. 21019 21020@item -mfpu 21021@itemx -mhard-float 21022@opindex mfpu 21023@opindex mhard-float 21024Generate output containing floating-point instructions. This is the 21025default. 21026 21027@item -mno-fpu 21028@itemx -msoft-float 21029@opindex mno-fpu 21030@opindex msoft-float 21031Generate output containing library calls for floating point. 21032@strong{Warning:} the requisite libraries are not available for all SPARC 21033targets. Normally the facilities of the machine's usual C compiler are 21034used, but this cannot be done directly in cross-compilation. You must make 21035your own arrangements to provide suitable library functions for 21036cross-compilation. The embedded targets @samp{sparc-*-aout} and 21037@samp{sparclite-*-*} do provide software floating-point support. 21038 21039@option{-msoft-float} changes the calling convention in the output file; 21040therefore, it is only useful if you compile @emph{all} of a program with 21041this option. In particular, you need to compile @file{libgcc.a}, the 21042library that comes with GCC, with @option{-msoft-float} in order for 21043this to work. 21044 21045@item -mhard-quad-float 21046@opindex mhard-quad-float 21047Generate output containing quad-word (long double) floating-point 21048instructions. 21049 21050@item -msoft-quad-float 21051@opindex msoft-quad-float 21052Generate output containing library calls for quad-word (long double) 21053floating-point instructions. The functions called are those specified 21054in the SPARC ABI@. This is the default. 21055 21056As of this writing, there are no SPARC implementations that have hardware 21057support for the quad-word floating-point instructions. They all invoke 21058a trap handler for one of these instructions, and then the trap handler 21059emulates the effect of the instruction. Because of the trap handler overhead, 21060this is much slower than calling the ABI library routines. Thus the 21061@option{-msoft-quad-float} option is the default. 21062 21063@item -mno-unaligned-doubles 21064@itemx -munaligned-doubles 21065@opindex mno-unaligned-doubles 21066@opindex munaligned-doubles 21067Assume that doubles have 8-byte alignment. This is the default. 21068 21069With @option{-munaligned-doubles}, GCC assumes that doubles have 8-byte 21070alignment only if they are contained in another type, or if they have an 21071absolute address. Otherwise, it assumes they have 4-byte alignment. 21072Specifying this option avoids some rare compatibility problems with code 21073generated by other compilers. It is not the default because it results 21074in a performance loss, especially for floating-point code. 21075 21076@item -muser-mode 21077@itemx -mno-user-mode 21078@opindex muser-mode 21079@opindex mno-user-mode 21080Do not generate code that can only run in supervisor mode. This is relevant 21081only for the @code{casa} instruction emitted for the LEON3 processor. The 21082default is @option{-mno-user-mode}. 21083 21084@item -mno-faster-structs 21085@itemx -mfaster-structs 21086@opindex mno-faster-structs 21087@opindex mfaster-structs 21088With @option{-mfaster-structs}, the compiler assumes that structures 21089should have 8-byte alignment. This enables the use of pairs of 21090@code{ldd} and @code{std} instructions for copies in structure 21091assignment, in place of twice as many @code{ld} and @code{st} pairs. 21092However, the use of this changed alignment directly violates the SPARC 21093ABI@. Thus, it's intended only for use on targets where the developer 21094acknowledges that their resulting code is not directly in line with 21095the rules of the ABI@. 21096 21097@item -mcpu=@var{cpu_type} 21098@opindex mcpu 21099Set the instruction set, register set, and instruction scheduling parameters 21100for machine type @var{cpu_type}. Supported values for @var{cpu_type} are 21101@samp{v7}, @samp{cypress}, @samp{v8}, @samp{supersparc}, @samp{hypersparc}, 21102@samp{leon}, @samp{leon3}, @samp{sparclite}, @samp{f930}, @samp{f934}, 21103@samp{sparclite86x}, @samp{sparclet}, @samp{tsc701}, @samp{v9}, 21104@samp{ultrasparc}, @samp{ultrasparc3}, @samp{niagara}, @samp{niagara2}, 21105@samp{niagara3} and @samp{niagara4}. 21106 21107Native Solaris and GNU/Linux toolchains also support the value @samp{native}, 21108which selects the best architecture option for the host processor. 21109@option{-mcpu=native} has no effect if GCC does not recognize 21110the processor. 21111 21112Default instruction scheduling parameters are used for values that select 21113an architecture and not an implementation. These are @samp{v7}, @samp{v8}, 21114@samp{sparclite}, @samp{sparclet}, @samp{v9}. 21115 21116Here is a list of each supported architecture and their supported 21117implementations. 21118 21119@table @asis 21120@item v7 21121cypress 21122 21123@item v8 21124supersparc, hypersparc, leon, leon3 21125 21126@item sparclite 21127f930, f934, sparclite86x 21128 21129@item sparclet 21130tsc701 21131 21132@item v9 21133ultrasparc, ultrasparc3, niagara, niagara2, niagara3, niagara4 21134@end table 21135 21136By default (unless configured otherwise), GCC generates code for the V7 21137variant of the SPARC architecture. With @option{-mcpu=cypress}, the compiler 21138additionally optimizes it for the Cypress CY7C602 chip, as used in the 21139SPARCStation/SPARCServer 3xx series. This is also appropriate for the older 21140SPARCStation 1, 2, IPX etc. 21141 21142With @option{-mcpu=v8}, GCC generates code for the V8 variant of the SPARC 21143architecture. The only difference from V7 code is that the compiler emits 21144the integer multiply and integer divide instructions which exist in SPARC-V8 21145but not in SPARC-V7. With @option{-mcpu=supersparc}, the compiler additionally 21146optimizes it for the SuperSPARC chip, as used in the SPARCStation 10, 1000 and 211472000 series. 21148 21149With @option{-mcpu=sparclite}, GCC generates code for the SPARClite variant of 21150the SPARC architecture. This adds the integer multiply, integer divide step 21151and scan (@code{ffs}) instructions which exist in SPARClite but not in SPARC-V7. 21152With @option{-mcpu=f930}, the compiler additionally optimizes it for the 21153Fujitsu MB86930 chip, which is the original SPARClite, with no FPU@. With 21154@option{-mcpu=f934}, the compiler additionally optimizes it for the Fujitsu 21155MB86934 chip, which is the more recent SPARClite with FPU@. 21156 21157With @option{-mcpu=sparclet}, GCC generates code for the SPARClet variant of 21158the SPARC architecture. This adds the integer multiply, multiply/accumulate, 21159integer divide step and scan (@code{ffs}) instructions which exist in SPARClet 21160but not in SPARC-V7. With @option{-mcpu=tsc701}, the compiler additionally 21161optimizes it for the TEMIC SPARClet chip. 21162 21163With @option{-mcpu=v9}, GCC generates code for the V9 variant of the SPARC 21164architecture. This adds 64-bit integer and floating-point move instructions, 211653 additional floating-point condition code registers and conditional move 21166instructions. With @option{-mcpu=ultrasparc}, the compiler additionally 21167optimizes it for the Sun UltraSPARC I/II/IIi chips. With 21168@option{-mcpu=ultrasparc3}, the compiler additionally optimizes it for the 21169Sun UltraSPARC III/III+/IIIi/IIIi+/IV/IV+ chips. With 21170@option{-mcpu=niagara}, the compiler additionally optimizes it for 21171Sun UltraSPARC T1 chips. With @option{-mcpu=niagara2}, the compiler 21172additionally optimizes it for Sun UltraSPARC T2 chips. With 21173@option{-mcpu=niagara3}, the compiler additionally optimizes it for Sun 21174UltraSPARC T3 chips. With @option{-mcpu=niagara4}, the compiler 21175additionally optimizes it for Sun UltraSPARC T4 chips. 21176 21177@item -mtune=@var{cpu_type} 21178@opindex mtune 21179Set the instruction scheduling parameters for machine type 21180@var{cpu_type}, but do not set the instruction set or register set that the 21181option @option{-mcpu=@var{cpu_type}} does. 21182 21183The same values for @option{-mcpu=@var{cpu_type}} can be used for 21184@option{-mtune=@var{cpu_type}}, but the only useful values are those 21185that select a particular CPU implementation. Those are @samp{cypress}, 21186@samp{supersparc}, @samp{hypersparc}, @samp{leon}, @samp{leon3}, @samp{f930}, 21187@samp{f934}, @samp{sparclite86x}, @samp{tsc701}, @samp{ultrasparc}, 21188@samp{ultrasparc3}, @samp{niagara}, @samp{niagara2}, @samp{niagara3} and 21189@samp{niagara4}. With native Solaris and GNU/Linux toolchains, @samp{native} 21190can also be used. 21191 21192@item -mv8plus 21193@itemx -mno-v8plus 21194@opindex mv8plus 21195@opindex mno-v8plus 21196With @option{-mv8plus}, GCC generates code for the SPARC-V8+ ABI@. The 21197difference from the V8 ABI is that the global and out registers are 21198considered 64 bits wide. This is enabled by default on Solaris in 32-bit 21199mode for all SPARC-V9 processors. 21200 21201@item -mvis 21202@itemx -mno-vis 21203@opindex mvis 21204@opindex mno-vis 21205With @option{-mvis}, GCC generates code that takes advantage of the UltraSPARC 21206Visual Instruction Set extensions. The default is @option{-mno-vis}. 21207 21208@item -mvis2 21209@itemx -mno-vis2 21210@opindex mvis2 21211@opindex mno-vis2 21212With @option{-mvis2}, GCC generates code that takes advantage of 21213version 2.0 of the UltraSPARC Visual Instruction Set extensions. The 21214default is @option{-mvis2} when targeting a cpu that supports such 21215instructions, such as UltraSPARC-III and later. Setting @option{-mvis2} 21216also sets @option{-mvis}. 21217 21218@item -mvis3 21219@itemx -mno-vis3 21220@opindex mvis3 21221@opindex mno-vis3 21222With @option{-mvis3}, GCC generates code that takes advantage of 21223version 3.0 of the UltraSPARC Visual Instruction Set extensions. The 21224default is @option{-mvis3} when targeting a cpu that supports such 21225instructions, such as niagara-3 and later. Setting @option{-mvis3} 21226also sets @option{-mvis2} and @option{-mvis}. 21227 21228@item -mcbcond 21229@itemx -mno-cbcond 21230@opindex mcbcond 21231@opindex mno-cbcond 21232With @option{-mcbcond}, GCC generates code that takes advantage of 21233compare-and-branch instructions, as defined in the Sparc Architecture 2011. 21234The default is @option{-mcbcond} when targeting a cpu that supports such 21235instructions, such as niagara-4 and later. 21236 21237@item -mpopc 21238@itemx -mno-popc 21239@opindex mpopc 21240@opindex mno-popc 21241With @option{-mpopc}, GCC generates code that takes advantage of the UltraSPARC 21242population count instruction. The default is @option{-mpopc} 21243when targeting a cpu that supports such instructions, such as Niagara-2 and 21244later. 21245 21246@item -mfmaf 21247@itemx -mno-fmaf 21248@opindex mfmaf 21249@opindex mno-fmaf 21250With @option{-mfmaf}, GCC generates code that takes advantage of the UltraSPARC 21251Fused Multiply-Add Floating-point extensions. The default is @option{-mfmaf} 21252when targeting a cpu that supports such instructions, such as Niagara-3 and 21253later. 21254 21255@item -mfix-at697f 21256@opindex mfix-at697f 21257Enable the documented workaround for the single erratum of the Atmel AT697F 21258processor (which corresponds to erratum #13 of the AT697E processor). 21259 21260@item -mfix-ut699 21261@opindex mfix-ut699 21262Enable the documented workarounds for the floating-point errata and the data 21263cache nullify errata of the UT699 processor. 21264@end table 21265 21266These @samp{-m} options are supported in addition to the above 21267on SPARC-V9 processors in 64-bit environments: 21268 21269@table @gcctabopt 21270@item -m32 21271@itemx -m64 21272@opindex m32 21273@opindex m64 21274Generate code for a 32-bit or 64-bit environment. 21275The 32-bit environment sets int, long and pointer to 32 bits. 21276The 64-bit environment sets int to 32 bits and long and pointer 21277to 64 bits. 21278 21279@item -mcmodel=@var{which} 21280@opindex mcmodel 21281Set the code model to one of 21282 21283@table @samp 21284@item medlow 21285The Medium/Low code model: 64-bit addresses, programs 21286must be linked in the low 32 bits of memory. Programs can be statically 21287or dynamically linked. 21288 21289@item medmid 21290The Medium/Middle code model: 64-bit addresses, programs 21291must be linked in the low 44 bits of memory, the text and data segments must 21292be less than 2GB in size and the data segment must be located within 2GB of 21293the text segment. 21294 21295@item medany 21296The Medium/Anywhere code model: 64-bit addresses, programs 21297may be linked anywhere in memory, the text and data segments must be less 21298than 2GB in size and the data segment must be located within 2GB of the 21299text segment. 21300 21301@item embmedany 21302The Medium/Anywhere code model for embedded systems: 2130364-bit addresses, the text and data segments must be less than 2GB in 21304size, both starting anywhere in memory (determined at link time). The 21305global register %g4 points to the base of the data segment. Programs 21306are statically linked and PIC is not supported. 21307@end table 21308 21309@item -mmemory-model=@var{mem-model} 21310@opindex mmemory-model 21311Set the memory model in force on the processor to one of 21312 21313@table @samp 21314@item default 21315The default memory model for the processor and operating system. 21316 21317@item rmo 21318Relaxed Memory Order 21319 21320@item pso 21321Partial Store Order 21322 21323@item tso 21324Total Store Order 21325 21326@item sc 21327Sequential Consistency 21328@end table 21329 21330These memory models are formally defined in Appendix D of the Sparc V9 21331architecture manual, as set in the processor's @code{PSTATE.MM} field. 21332 21333@item -mstack-bias 21334@itemx -mno-stack-bias 21335@opindex mstack-bias 21336@opindex mno-stack-bias 21337With @option{-mstack-bias}, GCC assumes that the stack pointer, and 21338frame pointer if present, are offset by @minus{}2047 which must be added back 21339when making stack frame references. This is the default in 64-bit mode. 21340Otherwise, assume no such offset is present. 21341@end table 21342 21343@node SPU Options 21344@subsection SPU Options 21345@cindex SPU options 21346 21347These @samp{-m} options are supported on the SPU: 21348 21349@table @gcctabopt 21350@item -mwarn-reloc 21351@itemx -merror-reloc 21352@opindex mwarn-reloc 21353@opindex merror-reloc 21354 21355The loader for SPU does not handle dynamic relocations. By default, GCC 21356gives an error when it generates code that requires a dynamic 21357relocation. @option{-mno-error-reloc} disables the error, 21358@option{-mwarn-reloc} generates a warning instead. 21359 21360@item -msafe-dma 21361@itemx -munsafe-dma 21362@opindex msafe-dma 21363@opindex munsafe-dma 21364 21365Instructions that initiate or test completion of DMA must not be 21366reordered with respect to loads and stores of the memory that is being 21367accessed. 21368With @option{-munsafe-dma} you must use the @code{volatile} keyword to protect 21369memory accesses, but that can lead to inefficient code in places where the 21370memory is known to not change. Rather than mark the memory as volatile, 21371you can use @option{-msafe-dma} to tell the compiler to treat 21372the DMA instructions as potentially affecting all memory. 21373 21374@item -mbranch-hints 21375@opindex mbranch-hints 21376 21377By default, GCC generates a branch hint instruction to avoid 21378pipeline stalls for always-taken or probably-taken branches. A hint 21379is not generated closer than 8 instructions away from its branch. 21380There is little reason to disable them, except for debugging purposes, 21381or to make an object a little bit smaller. 21382 21383@item -msmall-mem 21384@itemx -mlarge-mem 21385@opindex msmall-mem 21386@opindex mlarge-mem 21387 21388By default, GCC generates code assuming that addresses are never larger 21389than 18 bits. With @option{-mlarge-mem} code is generated that assumes 21390a full 32-bit address. 21391 21392@item -mstdmain 21393@opindex mstdmain 21394 21395By default, GCC links against startup code that assumes the SPU-style 21396main function interface (which has an unconventional parameter list). 21397With @option{-mstdmain}, GCC links your program against startup 21398code that assumes a C99-style interface to @code{main}, including a 21399local copy of @code{argv} strings. 21400 21401@item -mfixed-range=@var{register-range} 21402@opindex mfixed-range 21403Generate code treating the given register range as fixed registers. 21404A fixed register is one that the register allocator cannot use. This is 21405useful when compiling kernel code. A register range is specified as 21406two registers separated by a dash. Multiple register ranges can be 21407specified separated by a comma. 21408 21409@item -mea32 21410@itemx -mea64 21411@opindex mea32 21412@opindex mea64 21413Compile code assuming that pointers to the PPU address space accessed 21414via the @code{__ea} named address space qualifier are either 32 or 64 21415bits wide. The default is 32 bits. As this is an ABI-changing option, 21416all object code in an executable must be compiled with the same setting. 21417 21418@item -maddress-space-conversion 21419@itemx -mno-address-space-conversion 21420@opindex maddress-space-conversion 21421@opindex mno-address-space-conversion 21422Allow/disallow treating the @code{__ea} address space as superset 21423of the generic address space. This enables explicit type casts 21424between @code{__ea} and generic pointer as well as implicit 21425conversions of generic pointers to @code{__ea} pointers. The 21426default is to allow address space pointer conversions. 21427 21428@item -mcache-size=@var{cache-size} 21429@opindex mcache-size 21430This option controls the version of libgcc that the compiler links to an 21431executable and selects a software-managed cache for accessing variables 21432in the @code{__ea} address space with a particular cache size. Possible 21433options for @var{cache-size} are @samp{8}, @samp{16}, @samp{32}, @samp{64} 21434and @samp{128}. The default cache size is 64KB. 21435 21436@item -matomic-updates 21437@itemx -mno-atomic-updates 21438@opindex matomic-updates 21439@opindex mno-atomic-updates 21440This option controls the version of libgcc that the compiler links to an 21441executable and selects whether atomic updates to the software-managed 21442cache of PPU-side variables are used. If you use atomic updates, changes 21443to a PPU variable from SPU code using the @code{__ea} named address space 21444qualifier do not interfere with changes to other PPU variables residing 21445in the same cache line from PPU code. If you do not use atomic updates, 21446such interference may occur; however, writing back cache lines is 21447more efficient. The default behavior is to use atomic updates. 21448 21449@item -mdual-nops 21450@itemx -mdual-nops=@var{n} 21451@opindex mdual-nops 21452By default, GCC inserts nops to increase dual issue when it expects 21453it to increase performance. @var{n} can be a value from 0 to 10. A 21454smaller @var{n} inserts fewer nops. 10 is the default, 0 is the 21455same as @option{-mno-dual-nops}. Disabled with @option{-Os}. 21456 21457@item -mhint-max-nops=@var{n} 21458@opindex mhint-max-nops 21459Maximum number of nops to insert for a branch hint. A branch hint must 21460be at least 8 instructions away from the branch it is affecting. GCC 21461inserts up to @var{n} nops to enforce this, otherwise it does not 21462generate the branch hint. 21463 21464@item -mhint-max-distance=@var{n} 21465@opindex mhint-max-distance 21466The encoding of the branch hint instruction limits the hint to be within 21467256 instructions of the branch it is affecting. By default, GCC makes 21468sure it is within 125. 21469 21470@item -msafe-hints 21471@opindex msafe-hints 21472Work around a hardware bug that causes the SPU to stall indefinitely. 21473By default, GCC inserts the @code{hbrp} instruction to make sure 21474this stall won't happen. 21475 21476@end table 21477 21478@node System V Options 21479@subsection Options for System V 21480 21481These additional options are available on System V Release 4 for 21482compatibility with other compilers on those systems: 21483 21484@table @gcctabopt 21485@item -G 21486@opindex G 21487Create a shared object. 21488It is recommended that @option{-symbolic} or @option{-shared} be used instead. 21489 21490@item -Qy 21491@opindex Qy 21492Identify the versions of each tool used by the compiler, in a 21493@code{.ident} assembler directive in the output. 21494 21495@item -Qn 21496@opindex Qn 21497Refrain from adding @code{.ident} directives to the output file (this is 21498the default). 21499 21500@item -YP,@var{dirs} 21501@opindex YP 21502Search the directories @var{dirs}, and no others, for libraries 21503specified with @option{-l}. 21504 21505@item -Ym,@var{dir} 21506@opindex Ym 21507Look in the directory @var{dir} to find the M4 preprocessor. 21508The assembler uses this option. 21509@c This is supposed to go with a -Yd for predefined M4 macro files, but 21510@c the generic assembler that comes with Solaris takes just -Ym. 21511@end table 21512 21513@node TILE-Gx Options 21514@subsection TILE-Gx Options 21515@cindex TILE-Gx options 21516 21517These @samp{-m} options are supported on the TILE-Gx: 21518 21519@table @gcctabopt 21520@item -mcmodel=small 21521@opindex mcmodel=small 21522Generate code for the small model. The distance for direct calls is 21523limited to 500M in either direction. PC-relative addresses are 32 21524bits. Absolute addresses support the full address range. 21525 21526@item -mcmodel=large 21527@opindex mcmodel=large 21528Generate code for the large model. There is no limitation on call 21529distance, pc-relative addresses, or absolute addresses. 21530 21531@item -mcpu=@var{name} 21532@opindex mcpu 21533Selects the type of CPU to be targeted. Currently the only supported 21534type is @samp{tilegx}. 21535 21536@item -m32 21537@itemx -m64 21538@opindex m32 21539@opindex m64 21540Generate code for a 32-bit or 64-bit environment. The 32-bit 21541environment sets int, long, and pointer to 32 bits. The 64-bit 21542environment sets int to 32 bits and long and pointer to 64 bits. 21543 21544@item -mbig-endian 21545@itemx -mlittle-endian 21546@opindex mbig-endian 21547@opindex mlittle-endian 21548Generate code in big/little endian mode, respectively. 21549@end table 21550 21551@node TILEPro Options 21552@subsection TILEPro Options 21553@cindex TILEPro options 21554 21555These @samp{-m} options are supported on the TILEPro: 21556 21557@table @gcctabopt 21558@item -mcpu=@var{name} 21559@opindex mcpu 21560Selects the type of CPU to be targeted. Currently the only supported 21561type is @samp{tilepro}. 21562 21563@item -m32 21564@opindex m32 21565Generate code for a 32-bit environment, which sets int, long, and 21566pointer to 32 bits. This is the only supported behavior so the flag 21567is essentially ignored. 21568@end table 21569 21570@node V850 Options 21571@subsection V850 Options 21572@cindex V850 Options 21573 21574These @samp{-m} options are defined for V850 implementations: 21575 21576@table @gcctabopt 21577@item -mlong-calls 21578@itemx -mno-long-calls 21579@opindex mlong-calls 21580@opindex mno-long-calls 21581Treat all calls as being far away (near). If calls are assumed to be 21582far away, the compiler always loads the function's address into a 21583register, and calls indirect through the pointer. 21584 21585@item -mno-ep 21586@itemx -mep 21587@opindex mno-ep 21588@opindex mep 21589Do not optimize (do optimize) basic blocks that use the same index 21590pointer 4 or more times to copy pointer into the @code{ep} register, and 21591use the shorter @code{sld} and @code{sst} instructions. The @option{-mep} 21592option is on by default if you optimize. 21593 21594@item -mno-prolog-function 21595@itemx -mprolog-function 21596@opindex mno-prolog-function 21597@opindex mprolog-function 21598Do not use (do use) external functions to save and restore registers 21599at the prologue and epilogue of a function. The external functions 21600are slower, but use less code space if more than one function saves 21601the same number of registers. The @option{-mprolog-function} option 21602is on by default if you optimize. 21603 21604@item -mspace 21605@opindex mspace 21606Try to make the code as small as possible. At present, this just turns 21607on the @option{-mep} and @option{-mprolog-function} options. 21608 21609@item -mtda=@var{n} 21610@opindex mtda 21611Put static or global variables whose size is @var{n} bytes or less into 21612the tiny data area that register @code{ep} points to. The tiny data 21613area can hold up to 256 bytes in total (128 bytes for byte references). 21614 21615@item -msda=@var{n} 21616@opindex msda 21617Put static or global variables whose size is @var{n} bytes or less into 21618the small data area that register @code{gp} points to. The small data 21619area can hold up to 64 kilobytes. 21620 21621@item -mzda=@var{n} 21622@opindex mzda 21623Put static or global variables whose size is @var{n} bytes or less into 21624the first 32 kilobytes of memory. 21625 21626@item -mv850 21627@opindex mv850 21628Specify that the target processor is the V850. 21629 21630@item -mv850e3v5 21631@opindex mv850e3v5 21632Specify that the target processor is the V850E3V5. The preprocessor 21633constant @samp{__v850e3v5__} is defined if this option is used. 21634 21635@item -mv850e2v4 21636@opindex mv850e2v4 21637Specify that the target processor is the V850E3V5. This is an alias for 21638the @option{-mv850e3v5} option. 21639 21640@item -mv850e2v3 21641@opindex mv850e2v3 21642Specify that the target processor is the V850E2V3. The preprocessor 21643constant @samp{__v850e2v3__} is defined if this option is used. 21644 21645@item -mv850e2 21646@opindex mv850e2 21647Specify that the target processor is the V850E2. The preprocessor 21648constant @samp{__v850e2__} is defined if this option is used. 21649 21650@item -mv850e1 21651@opindex mv850e1 21652Specify that the target processor is the V850E1. The preprocessor 21653constants @samp{__v850e1__} and @samp{__v850e__} are defined if 21654this option is used. 21655 21656@item -mv850es 21657@opindex mv850es 21658Specify that the target processor is the V850ES. This is an alias for 21659the @option{-mv850e1} option. 21660 21661@item -mv850e 21662@opindex mv850e 21663Specify that the target processor is the V850E@. The preprocessor 21664constant @samp{__v850e__} is defined if this option is used. 21665 21666If neither @option{-mv850} nor @option{-mv850e} nor @option{-mv850e1} 21667nor @option{-mv850e2} nor @option{-mv850e2v3} nor @option{-mv850e3v5} 21668are defined then a default target processor is chosen and the 21669relevant @samp{__v850*__} preprocessor constant is defined. 21670 21671The preprocessor constants @samp{__v850} and @samp{__v851__} are always 21672defined, regardless of which processor variant is the target. 21673 21674@item -mdisable-callt 21675@itemx -mno-disable-callt 21676@opindex mdisable-callt 21677@opindex mno-disable-callt 21678This option suppresses generation of the @code{CALLT} instruction for the 21679v850e, v850e1, v850e2, v850e2v3 and v850e3v5 flavors of the v850 21680architecture. 21681 21682This option is enabled by default when the RH850 ABI is 21683in use (see @option{-mrh850-abi}), and disabled by default when the 21684GCC ABI is in use. If @code{CALLT} instructions are being generated 21685then the C preprocessor symbol @code{__V850_CALLT__} will be defined. 21686 21687@item -mrelax 21688@itemx -mno-relax 21689@opindex mrelax 21690@opindex mno-relax 21691Pass on (or do not pass on) the @option{-mrelax} command line option 21692to the assembler. 21693 21694@item -mlong-jumps 21695@itemx -mno-long-jumps 21696@opindex mlong-jumps 21697@opindex mno-long-jumps 21698Disable (or re-enable) the generation of PC-relative jump instructions. 21699 21700@item -msoft-float 21701@itemx -mhard-float 21702@opindex msoft-float 21703@opindex mhard-float 21704Disable (or re-enable) the generation of hardware floating point 21705instructions. This option is only significant when the target 21706architecture is @samp{V850E2V3} or higher. If hardware floating point 21707instructions are being generated then the C preprocessor symbol 21708@code{__FPU_OK__} will be defined, otherwise the symbol 21709@code{__NO_FPU__} will be defined. 21710 21711@item -mloop 21712@opindex mloop 21713Enables the use of the e3v5 LOOP instruction. The use of this 21714instruction is not enabled by default when the e3v5 architecture is 21715selected because its use is still experimental. 21716 21717@item -mrh850-abi 21718@itemx -mghs 21719@opindex mrh850-abi 21720@opindex mghs 21721Enables support for the RH850 version of the V850 ABI. This is the 21722default. With this version of the ABI the following rules apply: 21723 21724@itemize 21725@item 21726Integer sized structures and unions are returned via a memory pointer 21727rather than a register. 21728 21729@item 21730Large structures and unions (more than 8 bytes in size) are passed by 21731value. 21732 21733@item 21734Functions are aligned to 16-bit boundaries. 21735 21736@item 21737The @option{-m8byte-align} command line option is supported. 21738 21739@item 21740The @option{-mdisable-callt} command line option is enabled by 21741default. The @option{-mno-disable-callt} command line option is not 21742supported. 21743@end itemize 21744 21745When this version of the ABI is enabled the C preprocessor symbol 21746@code{__V850_RH850_ABI__} is defined. 21747 21748@item -mgcc-abi 21749@opindex mgcc-abi 21750Enables support for the old GCC version of the V850 ABI. With this 21751version of the ABI the following rules apply: 21752 21753@itemize 21754@item 21755Integer sized structures and unions are returned in register @code{r10}. 21756 21757@item 21758Large structures and unions (more than 8 bytes in size) are passed by 21759reference. 21760 21761@item 21762Functions are aligned to 32-bit boundaries, unless optimizing for 21763size. 21764 21765@item 21766The @option{-m8byte-align} command line option is not supported. 21767 21768@item 21769The @option{-mdisable-callt} command line option is supported but not 21770enabled by default. 21771@end itemize 21772 21773When this version of the ABI is enabled the C preprocessor symbol 21774@code{__V850_GCC_ABI__} is defined. 21775 21776@item -m8byte-align 21777@itemx -mno-8byte-align 21778@opindex m8byte-align 21779@opindex mno-8byte-align 21780Enables support for @code{doubles} and @code{long long} types to be 21781aligned on 8-byte boundaries. The default is to restrict the 21782alignment of all objects to at most 4-bytes. When 21783@option{-m8byte-align} is in effect the C preprocessor symbol 21784@code{__V850_8BYTE_ALIGN__} will be defined. 21785 21786@item -mbig-switch 21787@opindex mbig-switch 21788Generate code suitable for big switch tables. Use this option only if 21789the assembler/linker complain about out of range branches within a switch 21790table. 21791 21792@item -mapp-regs 21793@opindex mapp-regs 21794This option causes r2 and r5 to be used in the code generated by 21795the compiler. This setting is the default. 21796 21797@item -mno-app-regs 21798@opindex mno-app-regs 21799This option causes r2 and r5 to be treated as fixed registers. 21800 21801@end table 21802 21803@node VAX Options 21804@subsection VAX Options 21805@cindex VAX options 21806 21807These @samp{-m} options are defined for the VAX: 21808 21809@table @gcctabopt 21810@item -munix 21811@opindex munix 21812Do not output certain jump instructions (@code{aobleq} and so on) 21813that the Unix assembler for the VAX cannot handle across long 21814ranges. 21815 21816@item -mgnu 21817@opindex mgnu 21818Do output those jump instructions, on the assumption that the 21819GNU assembler is being used. 21820 21821@item -mg 21822@opindex mg 21823Output code for G-format floating-point numbers instead of D-format. 21824@end table 21825 21826@node VMS Options 21827@subsection VMS Options 21828 21829These @samp{-m} options are defined for the VMS implementations: 21830 21831@table @gcctabopt 21832@item -mvms-return-codes 21833@opindex mvms-return-codes 21834Return VMS condition codes from @code{main}. The default is to return POSIX-style 21835condition (e.g.@ error) codes. 21836 21837@item -mdebug-main=@var{prefix} 21838@opindex mdebug-main=@var{prefix} 21839Flag the first routine whose name starts with @var{prefix} as the main 21840routine for the debugger. 21841 21842@item -mmalloc64 21843@opindex mmalloc64 21844Default to 64-bit memory allocation routines. 21845 21846@item -mpointer-size=@var{size} 21847@opindex -mpointer-size=@var{size} 21848Set the default size of pointers. Possible options for @var{size} are 21849@samp{32} or @samp{short} for 32 bit pointers, @samp{64} or @samp{long} 21850for 64 bit pointers, and @samp{no} for supporting only 32 bit pointers. 21851The later option disables @code{pragma pointer_size}. 21852@end table 21853 21854@node VxWorks Options 21855@subsection VxWorks Options 21856@cindex VxWorks Options 21857 21858The options in this section are defined for all VxWorks targets. 21859Options specific to the target hardware are listed with the other 21860options for that target. 21861 21862@table @gcctabopt 21863@item -mrtp 21864@opindex mrtp 21865GCC can generate code for both VxWorks kernels and real time processes 21866(RTPs). This option switches from the former to the latter. It also 21867defines the preprocessor macro @code{__RTP__}. 21868 21869@item -non-static 21870@opindex non-static 21871Link an RTP executable against shared libraries rather than static 21872libraries. The options @option{-static} and @option{-shared} can 21873also be used for RTPs (@pxref{Link Options}); @option{-static} 21874is the default. 21875 21876@item -Bstatic 21877@itemx -Bdynamic 21878@opindex Bstatic 21879@opindex Bdynamic 21880These options are passed down to the linker. They are defined for 21881compatibility with Diab. 21882 21883@item -Xbind-lazy 21884@opindex Xbind-lazy 21885Enable lazy binding of function calls. This option is equivalent to 21886@option{-Wl,-z,now} and is defined for compatibility with Diab. 21887 21888@item -Xbind-now 21889@opindex Xbind-now 21890Disable lazy binding of function calls. This option is the default and 21891is defined for compatibility with Diab. 21892@end table 21893 21894@node x86-64 Options 21895@subsection x86-64 Options 21896@cindex x86-64 options 21897 21898These are listed under @xref{i386 and x86-64 Options}. 21899 21900@node Xstormy16 Options 21901@subsection Xstormy16 Options 21902@cindex Xstormy16 Options 21903 21904These options are defined for Xstormy16: 21905 21906@table @gcctabopt 21907@item -msim 21908@opindex msim 21909Choose startup files and linker script suitable for the simulator. 21910@end table 21911 21912@node Xtensa Options 21913@subsection Xtensa Options 21914@cindex Xtensa Options 21915 21916These options are supported for Xtensa targets: 21917 21918@table @gcctabopt 21919@item -mconst16 21920@itemx -mno-const16 21921@opindex mconst16 21922@opindex mno-const16 21923Enable or disable use of @code{CONST16} instructions for loading 21924constant values. The @code{CONST16} instruction is currently not a 21925standard option from Tensilica. When enabled, @code{CONST16} 21926instructions are always used in place of the standard @code{L32R} 21927instructions. The use of @code{CONST16} is enabled by default only if 21928the @code{L32R} instruction is not available. 21929 21930@item -mfused-madd 21931@itemx -mno-fused-madd 21932@opindex mfused-madd 21933@opindex mno-fused-madd 21934Enable or disable use of fused multiply/add and multiply/subtract 21935instructions in the floating-point option. This has no effect if the 21936floating-point option is not also enabled. Disabling fused multiply/add 21937and multiply/subtract instructions forces the compiler to use separate 21938instructions for the multiply and add/subtract operations. This may be 21939desirable in some cases where strict IEEE 754-compliant results are 21940required: the fused multiply add/subtract instructions do not round the 21941intermediate result, thereby producing results with @emph{more} bits of 21942precision than specified by the IEEE standard. Disabling fused multiply 21943add/subtract instructions also ensures that the program output is not 21944sensitive to the compiler's ability to combine multiply and add/subtract 21945operations. 21946 21947@item -mserialize-volatile 21948@itemx -mno-serialize-volatile 21949@opindex mserialize-volatile 21950@opindex mno-serialize-volatile 21951When this option is enabled, GCC inserts @code{MEMW} instructions before 21952@code{volatile} memory references to guarantee sequential consistency. 21953The default is @option{-mserialize-volatile}. Use 21954@option{-mno-serialize-volatile} to omit the @code{MEMW} instructions. 21955 21956@item -mforce-no-pic 21957@opindex mforce-no-pic 21958For targets, like GNU/Linux, where all user-mode Xtensa code must be 21959position-independent code (PIC), this option disables PIC for compiling 21960kernel code. 21961 21962@item -mtext-section-literals 21963@itemx -mno-text-section-literals 21964@opindex mtext-section-literals 21965@opindex mno-text-section-literals 21966Control the treatment of literal pools. The default is 21967@option{-mno-text-section-literals}, which places literals in a separate 21968section in the output file. This allows the literal pool to be placed 21969in a data RAM/ROM, and it also allows the linker to combine literal 21970pools from separate object files to remove redundant literals and 21971improve code size. With @option{-mtext-section-literals}, the literals 21972are interspersed in the text section in order to keep them as close as 21973possible to their references. This may be necessary for large assembly 21974files. 21975 21976@item -mtarget-align 21977@itemx -mno-target-align 21978@opindex mtarget-align 21979@opindex mno-target-align 21980When this option is enabled, GCC instructs the assembler to 21981automatically align instructions to reduce branch penalties at the 21982expense of some code density. The assembler attempts to widen density 21983instructions to align branch targets and the instructions following call 21984instructions. If there are not enough preceding safe density 21985instructions to align a target, no widening is performed. The 21986default is @option{-mtarget-align}. These options do not affect the 21987treatment of auto-aligned instructions like @code{LOOP}, which the 21988assembler always aligns, either by widening density instructions or 21989by inserting NOP instructions. 21990 21991@item -mlongcalls 21992@itemx -mno-longcalls 21993@opindex mlongcalls 21994@opindex mno-longcalls 21995When this option is enabled, GCC instructs the assembler to translate 21996direct calls to indirect calls unless it can determine that the target 21997of a direct call is in the range allowed by the call instruction. This 21998translation typically occurs for calls to functions in other source 21999files. Specifically, the assembler translates a direct @code{CALL} 22000instruction into an @code{L32R} followed by a @code{CALLX} instruction. 22001The default is @option{-mno-longcalls}. This option should be used in 22002programs where the call target can potentially be out of range. This 22003option is implemented in the assembler, not the compiler, so the 22004assembly code generated by GCC still shows direct call 22005instructions---look at the disassembled object code to see the actual 22006instructions. Note that the assembler uses an indirect call for 22007every cross-file call, not just those that really are out of range. 22008@end table 22009 22010@node zSeries Options 22011@subsection zSeries Options 22012@cindex zSeries options 22013 22014These are listed under @xref{S/390 and zSeries Options}. 22015 22016@node Code Gen Options 22017@section Options for Code Generation Conventions 22018@cindex code generation conventions 22019@cindex options, code generation 22020@cindex run-time options 22021 22022These machine-independent options control the interface conventions 22023used in code generation. 22024 22025Most of them have both positive and negative forms; the negative form 22026of @option{-ffoo} is @option{-fno-foo}. In the table below, only 22027one of the forms is listed---the one that is not the default. You 22028can figure out the other form by either removing @samp{no-} or adding 22029it. 22030 22031@table @gcctabopt 22032@item -fbounds-check 22033@opindex fbounds-check 22034For front ends that support it, generate additional code to check that 22035indices used to access arrays are within the declared range. This is 22036currently only supported by the Java and Fortran front ends, where 22037this option defaults to true and false respectively. 22038 22039@item -fstack-reuse=@var{reuse-level} 22040@opindex fstack_reuse 22041This option controls stack space reuse for user declared local/auto variables 22042and compiler generated temporaries. @var{reuse_level} can be @samp{all}, 22043@samp{named_vars}, or @samp{none}. @samp{all} enables stack reuse for all 22044local variables and temporaries, @samp{named_vars} enables the reuse only for 22045user defined local variables with names, and @samp{none} disables stack reuse 22046completely. The default value is @samp{all}. The option is needed when the 22047program extends the lifetime of a scoped local variable or a compiler generated 22048temporary beyond the end point defined by the language. When a lifetime of 22049a variable ends, and if the variable lives in memory, the optimizing compiler 22050has the freedom to reuse its stack space with other temporaries or scoped 22051local variables whose live range does not overlap with it. Legacy code extending 22052local lifetime will likely to break with the stack reuse optimization. 22053 22054For example, 22055 22056@smallexample 22057 int *p; 22058 @{ 22059 int local1; 22060 22061 p = &local1; 22062 local1 = 10; 22063 .... 22064 @} 22065 @{ 22066 int local2; 22067 local2 = 20; 22068 ... 22069 @} 22070 22071 if (*p == 10) // out of scope use of local1 22072 @{ 22073 22074 @} 22075@end smallexample 22076 22077Another example: 22078@smallexample 22079 22080 struct A 22081 @{ 22082 A(int k) : i(k), j(k) @{ @} 22083 int i; 22084 int j; 22085 @}; 22086 22087 A *ap; 22088 22089 void foo(const A& ar) 22090 @{ 22091 ap = &ar; 22092 @} 22093 22094 void bar() 22095 @{ 22096 foo(A(10)); // temp object's lifetime ends when foo returns 22097 22098 @{ 22099 A a(20); 22100 .... 22101 @} 22102 ap->i+= 10; // ap references out of scope temp whose space 22103 // is reused with a. What is the value of ap->i? 22104 @} 22105 22106@end smallexample 22107 22108The lifetime of a compiler generated temporary is well defined by the C++ 22109standard. When a lifetime of a temporary ends, and if the temporary lives 22110in memory, the optimizing compiler has the freedom to reuse its stack 22111space with other temporaries or scoped local variables whose live range 22112does not overlap with it. However some of the legacy code relies on 22113the behavior of older compilers in which temporaries' stack space is 22114not reused, the aggressive stack reuse can lead to runtime errors. This 22115option is used to control the temporary stack reuse optimization. 22116 22117@item -ftrapv 22118@opindex ftrapv 22119This option generates traps for signed overflow on addition, subtraction, 22120multiplication operations. 22121 22122@item -fwrapv 22123@opindex fwrapv 22124This option instructs the compiler to assume that signed arithmetic 22125overflow of addition, subtraction and multiplication wraps around 22126using twos-complement representation. This flag enables some optimizations 22127and disables others. This option is enabled by default for the Java 22128front end, as required by the Java language specification. 22129 22130@item -fexceptions 22131@opindex fexceptions 22132Enable exception handling. Generates extra code needed to propagate 22133exceptions. For some targets, this implies GCC generates frame 22134unwind information for all functions, which can produce significant data 22135size overhead, although it does not affect execution. If you do not 22136specify this option, GCC enables it by default for languages like 22137C++ that normally require exception handling, and disables it for 22138languages like C that do not normally require it. However, you may need 22139to enable this option when compiling C code that needs to interoperate 22140properly with exception handlers written in C++. You may also wish to 22141disable this option if you are compiling older C++ programs that don't 22142use exception handling. 22143 22144@item -fnon-call-exceptions 22145@opindex fnon-call-exceptions 22146Generate code that allows trapping instructions to throw exceptions. 22147Note that this requires platform-specific runtime support that does 22148not exist everywhere. Moreover, it only allows @emph{trapping} 22149instructions to throw exceptions, i.e.@: memory references or floating-point 22150instructions. It does not allow exceptions to be thrown from 22151arbitrary signal handlers such as @code{SIGALRM}. 22152 22153@item -fdelete-dead-exceptions 22154@opindex fdelete-dead-exceptions 22155Consider that instructions that may throw exceptions but don't otherwise 22156contribute to the execution of the program can be optimized away. 22157This option is enabled by default for the Ada front end, as permitted by 22158the Ada language specification. 22159Optimization passes that cause dead exceptions to be removed are enabled independently at different optimization levels. 22160 22161@item -funwind-tables 22162@opindex funwind-tables 22163Similar to @option{-fexceptions}, except that it just generates any needed 22164static data, but does not affect the generated code in any other way. 22165You normally do not need to enable this option; instead, a language processor 22166that needs this handling enables it on your behalf. 22167 22168@item -fasynchronous-unwind-tables 22169@opindex fasynchronous-unwind-tables 22170Generate unwind table in DWARF 2 format, if supported by target machine. The 22171table is exact at each instruction boundary, so it can be used for stack 22172unwinding from asynchronous events (such as debugger or garbage collector). 22173 22174@item -fno-gnu-unique 22175@opindex fno-gnu-unique 22176On systems with recent GNU assembler and C library, the C++ compiler 22177uses the @code{STB_GNU_UNIQUE} binding to make sure that definitions 22178of template static data members and static local variables in inline 22179functions are unique even in the presence of @code{RTLD_LOCAL}; this 22180is necessary to avoid problems with a library used by two different 22181@code{RTLD_LOCAL} plugins depending on a definition in one of them and 22182therefore disagreeing with the other one about the binding of the 22183symbol. But this causes @code{dlclose} to be ignored for affected 22184DSOs; if your program relies on reinitialization of a DSO via 22185@code{dlclose} and @code{dlopen}, you can use 22186@option{-fno-gnu-unique}. 22187 22188@item -fpcc-struct-return 22189@opindex fpcc-struct-return 22190Return ``short'' @code{struct} and @code{union} values in memory like 22191longer ones, rather than in registers. This convention is less 22192efficient, but it has the advantage of allowing intercallability between 22193GCC-compiled files and files compiled with other compilers, particularly 22194the Portable C Compiler (pcc). 22195 22196The precise convention for returning structures in memory depends 22197on the target configuration macros. 22198 22199Short structures and unions are those whose size and alignment match 22200that of some integer type. 22201 22202@strong{Warning:} code compiled with the @option{-fpcc-struct-return} 22203switch is not binary compatible with code compiled with the 22204@option{-freg-struct-return} switch. 22205Use it to conform to a non-default application binary interface. 22206 22207@item -freg-struct-return 22208@opindex freg-struct-return 22209Return @code{struct} and @code{union} values in registers when possible. 22210This is more efficient for small structures than 22211@option{-fpcc-struct-return}. 22212 22213If you specify neither @option{-fpcc-struct-return} nor 22214@option{-freg-struct-return}, GCC defaults to whichever convention is 22215standard for the target. If there is no standard convention, GCC 22216defaults to @option{-fpcc-struct-return}, except on targets where GCC is 22217the principal compiler. In those cases, we can choose the standard, and 22218we chose the more efficient register return alternative. 22219 22220@strong{Warning:} code compiled with the @option{-freg-struct-return} 22221switch is not binary compatible with code compiled with the 22222@option{-fpcc-struct-return} switch. 22223Use it to conform to a non-default application binary interface. 22224 22225@item -fshort-enums 22226@opindex fshort-enums 22227Allocate to an @code{enum} type only as many bytes as it needs for the 22228declared range of possible values. Specifically, the @code{enum} type 22229is equivalent to the smallest integer type that has enough room. 22230 22231@strong{Warning:} the @option{-fshort-enums} switch causes GCC to generate 22232code that is not binary compatible with code generated without that switch. 22233Use it to conform to a non-default application binary interface. 22234 22235@item -fshort-double 22236@opindex fshort-double 22237Use the same size for @code{double} as for @code{float}. 22238 22239@strong{Warning:} the @option{-fshort-double} switch causes GCC to generate 22240code that is not binary compatible with code generated without that switch. 22241Use it to conform to a non-default application binary interface. 22242 22243@item -fshort-wchar 22244@opindex fshort-wchar 22245Override the underlying type for @samp{wchar_t} to be @samp{short 22246unsigned int} instead of the default for the target. This option is 22247useful for building programs to run under WINE@. 22248 22249@strong{Warning:} the @option{-fshort-wchar} switch causes GCC to generate 22250code that is not binary compatible with code generated without that switch. 22251Use it to conform to a non-default application binary interface. 22252 22253@item -fno-common 22254@opindex fno-common 22255In C code, controls the placement of uninitialized global variables. 22256Unix C compilers have traditionally permitted multiple definitions of 22257such variables in different compilation units by placing the variables 22258in a common block. 22259This is the behavior specified by @option{-fcommon}, and is the default 22260for GCC on most targets. 22261On the other hand, this behavior is not required by ISO C, and on some 22262targets may carry a speed or code size penalty on variable references. 22263The @option{-fno-common} option specifies that the compiler should place 22264uninitialized global variables in the data section of the object file, 22265rather than generating them as common blocks. 22266This has the effect that if the same variable is declared 22267(without @code{extern}) in two different compilations, 22268you get a multiple-definition error when you link them. 22269In this case, you must compile with @option{-fcommon} instead. 22270Compiling with @option{-fno-common} is useful on targets for which 22271it provides better performance, or if you wish to verify that the 22272program will work on other systems that always treat uninitialized 22273variable declarations this way. 22274 22275@item -fno-ident 22276@opindex fno-ident 22277Ignore the @samp{#ident} directive. 22278 22279@item -finhibit-size-directive 22280@opindex finhibit-size-directive 22281Don't output a @code{.size} assembler directive, or anything else that 22282would cause trouble if the function is split in the middle, and the 22283two halves are placed at locations far apart in memory. This option is 22284used when compiling @file{crtstuff.c}; you should not need to use it 22285for anything else. 22286 22287@item -fverbose-asm 22288@opindex fverbose-asm 22289Put extra commentary information in the generated assembly code to 22290make it more readable. This option is generally only of use to those 22291who actually need to read the generated assembly code (perhaps while 22292debugging the compiler itself). 22293 22294@option{-fno-verbose-asm}, the default, causes the 22295extra information to be omitted and is useful when comparing two assembler 22296files. 22297 22298@item -frecord-gcc-switches 22299@opindex frecord-gcc-switches 22300This switch causes the command line used to invoke the 22301compiler to be recorded into the object file that is being created. 22302This switch is only implemented on some targets and the exact format 22303of the recording is target and binary file format dependent, but it 22304usually takes the form of a section containing ASCII text. This 22305switch is related to the @option{-fverbose-asm} switch, but that 22306switch only records information in the assembler output file as 22307comments, so it never reaches the object file. 22308See also @option{-grecord-gcc-switches} for another 22309way of storing compiler options into the object file. 22310 22311@item -fpic 22312@opindex fpic 22313@cindex global offset table 22314@cindex PIC 22315Generate position-independent code (PIC) suitable for use in a shared 22316library, if supported for the target machine. Such code accesses all 22317constant addresses through a global offset table (GOT)@. The dynamic 22318loader resolves the GOT entries when the program starts (the dynamic 22319loader is not part of GCC; it is part of the operating system). If 22320the GOT size for the linked executable exceeds a machine-specific 22321maximum size, you get an error message from the linker indicating that 22322@option{-fpic} does not work; in that case, recompile with @option{-fPIC} 22323instead. (These maximums are 8k on the SPARC and 32k 22324on the m68k and RS/6000. The 386 has no such limit.) 22325 22326Position-independent code requires special support, and therefore works 22327only on certain machines. For the 386, GCC supports PIC for System V 22328but not for the Sun 386i. Code generated for the IBM RS/6000 is always 22329position-independent. 22330 22331When this flag is set, the macros @code{__pic__} and @code{__PIC__} 22332are defined to 1. 22333 22334@item -fPIC 22335@opindex fPIC 22336If supported for the target machine, emit position-independent code, 22337suitable for dynamic linking and avoiding any limit on the size of the 22338global offset table. This option makes a difference on the m68k, 22339PowerPC and SPARC@. 22340 22341Position-independent code requires special support, and therefore works 22342only on certain machines. 22343 22344When this flag is set, the macros @code{__pic__} and @code{__PIC__} 22345are defined to 2. 22346 22347@item -fpie 22348@itemx -fPIE 22349@opindex fpie 22350@opindex fPIE 22351These options are similar to @option{-fpic} and @option{-fPIC}, but 22352generated position independent code can be only linked into executables. 22353Usually these options are used when @option{-pie} GCC option is 22354used during linking. 22355 22356@option{-fpie} and @option{-fPIE} both define the macros 22357@code{__pie__} and @code{__PIE__}. The macros have the value 1 22358for @option{-fpie} and 2 for @option{-fPIE}. 22359 22360@item -fno-jump-tables 22361@opindex fno-jump-tables 22362Do not use jump tables for switch statements even where it would be 22363more efficient than other code generation strategies. This option is 22364of use in conjunction with @option{-fpic} or @option{-fPIC} for 22365building code that forms part of a dynamic linker and cannot 22366reference the address of a jump table. On some targets, jump tables 22367do not require a GOT and this option is not needed. 22368 22369@item -ffixed-@var{reg} 22370@opindex ffixed 22371Treat the register named @var{reg} as a fixed register; generated code 22372should never refer to it (except perhaps as a stack pointer, frame 22373pointer or in some other fixed role). 22374 22375@var{reg} must be the name of a register. The register names accepted 22376are machine-specific and are defined in the @code{REGISTER_NAMES} 22377macro in the machine description macro file. 22378 22379This flag does not have a negative form, because it specifies a 22380three-way choice. 22381 22382@item -fcall-used-@var{reg} 22383@opindex fcall-used 22384Treat the register named @var{reg} as an allocable register that is 22385clobbered by function calls. It may be allocated for temporaries or 22386variables that do not live across a call. Functions compiled this way 22387do not save and restore the register @var{reg}. 22388 22389It is an error to use this flag with the frame pointer or stack pointer. 22390Use of this flag for other registers that have fixed pervasive roles in 22391the machine's execution model produces disastrous results. 22392 22393This flag does not have a negative form, because it specifies a 22394three-way choice. 22395 22396@item -fcall-saved-@var{reg} 22397@opindex fcall-saved 22398Treat the register named @var{reg} as an allocable register saved by 22399functions. It may be allocated even for temporaries or variables that 22400live across a call. Functions compiled this way save and restore 22401the register @var{reg} if they use it. 22402 22403It is an error to use this flag with the frame pointer or stack pointer. 22404Use of this flag for other registers that have fixed pervasive roles in 22405the machine's execution model produces disastrous results. 22406 22407A different sort of disaster results from the use of this flag for 22408a register in which function values may be returned. 22409 22410This flag does not have a negative form, because it specifies a 22411three-way choice. 22412 22413@item -fpack-struct[=@var{n}] 22414@opindex fpack-struct 22415Without a value specified, pack all structure members together without 22416holes. When a value is specified (which must be a small power of two), pack 22417structure members according to this value, representing the maximum 22418alignment (that is, objects with default alignment requirements larger than 22419this are output potentially unaligned at the next fitting location. 22420 22421@strong{Warning:} the @option{-fpack-struct} switch causes GCC to generate 22422code that is not binary compatible with code generated without that switch. 22423Additionally, it makes the code suboptimal. 22424Use it to conform to a non-default application binary interface. 22425 22426@item -finstrument-functions 22427@opindex finstrument-functions 22428Generate instrumentation calls for entry and exit to functions. Just 22429after function entry and just before function exit, the following 22430profiling functions are called with the address of the current 22431function and its call site. (On some platforms, 22432@code{__builtin_return_address} does not work beyond the current 22433function, so the call site information may not be available to the 22434profiling functions otherwise.) 22435 22436@smallexample 22437void __cyg_profile_func_enter (void *this_fn, 22438 void *call_site); 22439void __cyg_profile_func_exit (void *this_fn, 22440 void *call_site); 22441@end smallexample 22442 22443The first argument is the address of the start of the current function, 22444which may be looked up exactly in the symbol table. 22445 22446This instrumentation is also done for functions expanded inline in other 22447functions. The profiling calls indicate where, conceptually, the 22448inline function is entered and exited. This means that addressable 22449versions of such functions must be available. If all your uses of a 22450function are expanded inline, this may mean an additional expansion of 22451code size. If you use @samp{extern inline} in your C code, an 22452addressable version of such functions must be provided. (This is 22453normally the case anyway, but if you get lucky and the optimizer always 22454expands the functions inline, you might have gotten away without 22455providing static copies.) 22456 22457A function may be given the attribute @code{no_instrument_function}, in 22458which case this instrumentation is not done. This can be used, for 22459example, for the profiling functions listed above, high-priority 22460interrupt routines, and any functions from which the profiling functions 22461cannot safely be called (perhaps signal handlers, if the profiling 22462routines generate output or allocate memory). 22463 22464@item -finstrument-functions-exclude-file-list=@var{file},@var{file},@dots{} 22465@opindex finstrument-functions-exclude-file-list 22466 22467Set the list of functions that are excluded from instrumentation (see 22468the description of @code{-finstrument-functions}). If the file that 22469contains a function definition matches with one of @var{file}, then 22470that function is not instrumented. The match is done on substrings: 22471if the @var{file} parameter is a substring of the file name, it is 22472considered to be a match. 22473 22474For example: 22475 22476@smallexample 22477-finstrument-functions-exclude-file-list=/bits/stl,include/sys 22478@end smallexample 22479 22480@noindent 22481excludes any inline function defined in files whose pathnames 22482contain @code{/bits/stl} or @code{include/sys}. 22483 22484If, for some reason, you want to include letter @code{','} in one of 22485@var{sym}, write @code{'\,'}. For example, 22486@code{-finstrument-functions-exclude-file-list='\,\,tmp'} 22487(note the single quote surrounding the option). 22488 22489@item -finstrument-functions-exclude-function-list=@var{sym},@var{sym},@dots{} 22490@opindex finstrument-functions-exclude-function-list 22491 22492This is similar to @code{-finstrument-functions-exclude-file-list}, 22493but this option sets the list of function names to be excluded from 22494instrumentation. The function name to be matched is its user-visible 22495name, such as @code{vector<int> blah(const vector<int> &)}, not the 22496internal mangled name (e.g., @code{_Z4blahRSt6vectorIiSaIiEE}). The 22497match is done on substrings: if the @var{sym} parameter is a substring 22498of the function name, it is considered to be a match. For C99 and C++ 22499extended identifiers, the function name must be given in UTF-8, not 22500using universal character names. 22501 22502@item -fstack-check 22503@opindex fstack-check 22504Generate code to verify that you do not go beyond the boundary of the 22505stack. You should specify this flag if you are running in an 22506environment with multiple threads, but you only rarely need to specify it in 22507a single-threaded environment since stack overflow is automatically 22508detected on nearly all systems if there is only one stack. 22509 22510Note that this switch does not actually cause checking to be done; the 22511operating system or the language runtime must do that. The switch causes 22512generation of code to ensure that they see the stack being extended. 22513 22514You can additionally specify a string parameter: @code{no} means no 22515checking, @code{generic} means force the use of old-style checking, 22516@code{specific} means use the best checking method and is equivalent 22517to bare @option{-fstack-check}. 22518 22519Old-style checking is a generic mechanism that requires no specific 22520target support in the compiler but comes with the following drawbacks: 22521 22522@enumerate 22523@item 22524Modified allocation strategy for large objects: they are always 22525allocated dynamically if their size exceeds a fixed threshold. 22526 22527@item 22528Fixed limit on the size of the static frame of functions: when it is 22529topped by a particular function, stack checking is not reliable and 22530a warning is issued by the compiler. 22531 22532@item 22533Inefficiency: because of both the modified allocation strategy and the 22534generic implementation, code performance is hampered. 22535@end enumerate 22536 22537Note that old-style stack checking is also the fallback method for 22538@code{specific} if no target support has been added in the compiler. 22539 22540@item -fstack-limit-register=@var{reg} 22541@itemx -fstack-limit-symbol=@var{sym} 22542@itemx -fno-stack-limit 22543@opindex fstack-limit-register 22544@opindex fstack-limit-symbol 22545@opindex fno-stack-limit 22546Generate code to ensure that the stack does not grow beyond a certain value, 22547either the value of a register or the address of a symbol. If a larger 22548stack is required, a signal is raised at run time. For most targets, 22549the signal is raised before the stack overruns the boundary, so 22550it is possible to catch the signal without taking special precautions. 22551 22552For instance, if the stack starts at absolute address @samp{0x80000000} 22553and grows downwards, you can use the flags 22554@option{-fstack-limit-symbol=__stack_limit} and 22555@option{-Wl,--defsym,__stack_limit=0x7ffe0000} to enforce a stack limit 22556of 128KB@. Note that this may only work with the GNU linker. 22557 22558@item -fsplit-stack 22559@opindex fsplit-stack 22560Generate code to automatically split the stack before it overflows. 22561The resulting program has a discontiguous stack which can only 22562overflow if the program is unable to allocate any more memory. This 22563is most useful when running threaded programs, as it is no longer 22564necessary to calculate a good stack size to use for each thread. This 22565is currently only implemented for the i386 and x86_64 back ends running 22566GNU/Linux. 22567 22568When code compiled with @option{-fsplit-stack} calls code compiled 22569without @option{-fsplit-stack}, there may not be much stack space 22570available for the latter code to run. If compiling all code, 22571including library code, with @option{-fsplit-stack} is not an option, 22572then the linker can fix up these calls so that the code compiled 22573without @option{-fsplit-stack} always has a large stack. Support for 22574this is implemented in the gold linker in GNU binutils release 2.21 22575and later. 22576 22577@item -fleading-underscore 22578@opindex fleading-underscore 22579This option and its counterpart, @option{-fno-leading-underscore}, forcibly 22580change the way C symbols are represented in the object file. One use 22581is to help link with legacy assembly code. 22582 22583@strong{Warning:} the @option{-fleading-underscore} switch causes GCC to 22584generate code that is not binary compatible with code generated without that 22585switch. Use it to conform to a non-default application binary interface. 22586Not all targets provide complete support for this switch. 22587 22588@item -ftls-model=@var{model} 22589@opindex ftls-model 22590Alter the thread-local storage model to be used (@pxref{Thread-Local}). 22591The @var{model} argument should be one of @code{global-dynamic}, 22592@code{local-dynamic}, @code{initial-exec} or @code{local-exec}. 22593Note that the choice is subject to optimization: the compiler may use 22594a more efficient model for symbols not visible outside of the translation 22595unit, or if @option{-fpic} is not given on the command line. 22596 22597The default without @option{-fpic} is @code{initial-exec}; with 22598@option{-fpic} the default is @code{global-dynamic}. 22599 22600@item -fvisibility=@var{default|internal|hidden|protected} 22601@opindex fvisibility 22602Set the default ELF image symbol visibility to the specified option---all 22603symbols are marked with this unless overridden within the code. 22604Using this feature can very substantially improve linking and 22605load times of shared object libraries, produce more optimized 22606code, provide near-perfect API export and prevent symbol clashes. 22607It is @strong{strongly} recommended that you use this in any shared objects 22608you distribute. 22609 22610Despite the nomenclature, @code{default} always means public; i.e., 22611available to be linked against from outside the shared object. 22612@code{protected} and @code{internal} are pretty useless in real-world 22613usage so the only other commonly used option is @code{hidden}. 22614The default if @option{-fvisibility} isn't specified is 22615@code{default}, i.e., make every 22616symbol public---this causes the same behavior as previous versions of 22617GCC@. 22618 22619A good explanation of the benefits offered by ensuring ELF 22620symbols have the correct visibility is given by ``How To Write 22621Shared Libraries'' by Ulrich Drepper (which can be found at 22622@w{@uref{http://people.redhat.com/~drepper/}})---however a superior 22623solution made possible by this option to marking things hidden when 22624the default is public is to make the default hidden and mark things 22625public. This is the norm with DLLs on Windows and with @option{-fvisibility=hidden} 22626and @code{__attribute__ ((visibility("default")))} instead of 22627@code{__declspec(dllexport)} you get almost identical semantics with 22628identical syntax. This is a great boon to those working with 22629cross-platform projects. 22630 22631For those adding visibility support to existing code, you may find 22632@samp{#pragma GCC visibility} of use. This works by you enclosing 22633the declarations you wish to set visibility for with (for example) 22634@samp{#pragma GCC visibility push(hidden)} and 22635@samp{#pragma GCC visibility pop}. 22636Bear in mind that symbol visibility should be viewed @strong{as 22637part of the API interface contract} and thus all new code should 22638always specify visibility when it is not the default; i.e., declarations 22639only for use within the local DSO should @strong{always} be marked explicitly 22640as hidden as so to avoid PLT indirection overheads---making this 22641abundantly clear also aids readability and self-documentation of the code. 22642Note that due to ISO C++ specification requirements, @code{operator new} and 22643@code{operator delete} must always be of default visibility. 22644 22645Be aware that headers from outside your project, in particular system 22646headers and headers from any other library you use, may not be 22647expecting to be compiled with visibility other than the default. You 22648may need to explicitly say @samp{#pragma GCC visibility push(default)} 22649before including any such headers. 22650 22651@samp{extern} declarations are not affected by @option{-fvisibility}, so 22652a lot of code can be recompiled with @option{-fvisibility=hidden} with 22653no modifications. However, this means that calls to @code{extern} 22654functions with no explicit visibility use the PLT, so it is more 22655effective to use @code{__attribute ((visibility))} and/or 22656@code{#pragma GCC visibility} to tell the compiler which @code{extern} 22657declarations should be treated as hidden. 22658 22659Note that @option{-fvisibility} does affect C++ vague linkage 22660entities. This means that, for instance, an exception class that is 22661be thrown between DSOs must be explicitly marked with default 22662visibility so that the @samp{type_info} nodes are unified between 22663the DSOs. 22664 22665An overview of these techniques, their benefits and how to use them 22666is at @uref{http://gcc.gnu.org/@/wiki/@/Visibility}. 22667 22668@item -fstrict-volatile-bitfields 22669@opindex fstrict-volatile-bitfields 22670This option should be used if accesses to volatile bit-fields (or other 22671structure fields, although the compiler usually honors those types 22672anyway) should use a single access of the width of the 22673field's type, aligned to a natural alignment if possible. For 22674example, targets with memory-mapped peripheral registers might require 22675all such accesses to be 16 bits wide; with this flag you can 22676declare all peripheral bit-fields as @code{unsigned short} (assuming short 22677is 16 bits on these targets) to force GCC to use 16-bit accesses 22678instead of, perhaps, a more efficient 32-bit access. 22679 22680If this option is disabled, the compiler uses the most efficient 22681instruction. In the previous example, that might be a 32-bit load 22682instruction, even though that accesses bytes that do not contain 22683any portion of the bit-field, or memory-mapped registers unrelated to 22684the one being updated. 22685 22686In some cases, such as when the @code{packed} attribute is applied to a 22687structure field, it may not be possible to access the field with a single 22688read or write that is correctly aligned for the target machine. In this 22689case GCC falls back to generating multiple accesses rather than code that 22690will fault or truncate the result at run time. 22691 22692Note: Due to restrictions of the C/C++11 memory model, write accesses are 22693not allowed to touch non bit-field members. It is therefore recommended 22694to define all bits of the field's type as bit-field members. 22695 22696The default value of this option is determined by the application binary 22697interface for the target processor. 22698 22699@item -fsync-libcalls 22700@opindex fsync-libcalls 22701This option controls whether any out-of-line instance of the @code{__sync} 22702family of functions may be used to implement the C++11 @code{__atomic} 22703family of functions. 22704 22705The default value of this option is enabled, thus the only useful form 22706of the option is @option{-fno-sync-libcalls}. This option is used in 22707the implementation of the @file{libatomic} runtime library. 22708 22709@end table 22710 22711@c man end 22712 22713@node Environment Variables 22714@section Environment Variables Affecting GCC 22715@cindex environment variables 22716 22717@c man begin ENVIRONMENT 22718This section describes several environment variables that affect how GCC 22719operates. Some of them work by specifying directories or prefixes to use 22720when searching for various kinds of files. Some are used to specify other 22721aspects of the compilation environment. 22722 22723Note that you can also specify places to search using options such as 22724@option{-B}, @option{-I} and @option{-L} (@pxref{Directory Options}). These 22725take precedence over places specified using environment variables, which 22726in turn take precedence over those specified by the configuration of GCC@. 22727@xref{Driver,, Controlling the Compilation Driver @file{gcc}, gccint, 22728GNU Compiler Collection (GCC) Internals}. 22729 22730@table @env 22731@item LANG 22732@itemx LC_CTYPE 22733@c @itemx LC_COLLATE 22734@itemx LC_MESSAGES 22735@c @itemx LC_MONETARY 22736@c @itemx LC_NUMERIC 22737@c @itemx LC_TIME 22738@itemx LC_ALL 22739@findex LANG 22740@findex LC_CTYPE 22741@c @findex LC_COLLATE 22742@findex LC_MESSAGES 22743@c @findex LC_MONETARY 22744@c @findex LC_NUMERIC 22745@c @findex LC_TIME 22746@findex LC_ALL 22747@cindex locale 22748These environment variables control the way that GCC uses 22749localization information which allows GCC to work with different 22750national conventions. GCC inspects the locale categories 22751@env{LC_CTYPE} and @env{LC_MESSAGES} if it has been configured to do 22752so. These locale categories can be set to any value supported by your 22753installation. A typical value is @samp{en_GB.UTF-8} for English in the United 22754Kingdom encoded in UTF-8. 22755 22756The @env{LC_CTYPE} environment variable specifies character 22757classification. GCC uses it to determine the character boundaries in 22758a string; this is needed for some multibyte encodings that contain quote 22759and escape characters that are otherwise interpreted as a string 22760end or escape. 22761 22762The @env{LC_MESSAGES} environment variable specifies the language to 22763use in diagnostic messages. 22764 22765If the @env{LC_ALL} environment variable is set, it overrides the value 22766of @env{LC_CTYPE} and @env{LC_MESSAGES}; otherwise, @env{LC_CTYPE} 22767and @env{LC_MESSAGES} default to the value of the @env{LANG} 22768environment variable. If none of these variables are set, GCC 22769defaults to traditional C English behavior. 22770 22771@item TMPDIR 22772@findex TMPDIR 22773If @env{TMPDIR} is set, it specifies the directory to use for temporary 22774files. GCC uses temporary files to hold the output of one stage of 22775compilation which is to be used as input to the next stage: for example, 22776the output of the preprocessor, which is the input to the compiler 22777proper. 22778 22779@item GCC_COMPARE_DEBUG 22780@findex GCC_COMPARE_DEBUG 22781Setting @env{GCC_COMPARE_DEBUG} is nearly equivalent to passing 22782@option{-fcompare-debug} to the compiler driver. See the documentation 22783of this option for more details. 22784 22785@item GCC_EXEC_PREFIX 22786@findex GCC_EXEC_PREFIX 22787If @env{GCC_EXEC_PREFIX} is set, it specifies a prefix to use in the 22788names of the subprograms executed by the compiler. No slash is added 22789when this prefix is combined with the name of a subprogram, but you can 22790specify a prefix that ends with a slash if you wish. 22791 22792If @env{GCC_EXEC_PREFIX} is not set, GCC attempts to figure out 22793an appropriate prefix to use based on the pathname it is invoked with. 22794 22795If GCC cannot find the subprogram using the specified prefix, it 22796tries looking in the usual places for the subprogram. 22797 22798The default value of @env{GCC_EXEC_PREFIX} is 22799@file{@var{prefix}/lib/gcc/} where @var{prefix} is the prefix to 22800the installed compiler. In many cases @var{prefix} is the value 22801of @code{prefix} when you ran the @file{configure} script. 22802 22803Other prefixes specified with @option{-B} take precedence over this prefix. 22804 22805This prefix is also used for finding files such as @file{crt0.o} that are 22806used for linking. 22807 22808In addition, the prefix is used in an unusual way in finding the 22809directories to search for header files. For each of the standard 22810directories whose name normally begins with @samp{/usr/local/lib/gcc} 22811(more precisely, with the value of @env{GCC_INCLUDE_DIR}), GCC tries 22812replacing that beginning with the specified prefix to produce an 22813alternate directory name. Thus, with @option{-Bfoo/}, GCC searches 22814@file{foo/bar} just before it searches the standard directory 22815@file{/usr/local/lib/bar}. 22816If a standard directory begins with the configured 22817@var{prefix} then the value of @var{prefix} is replaced by 22818@env{GCC_EXEC_PREFIX} when looking for header files. 22819 22820@item COMPILER_PATH 22821@findex COMPILER_PATH 22822The value of @env{COMPILER_PATH} is a colon-separated list of 22823directories, much like @env{PATH}. GCC tries the directories thus 22824specified when searching for subprograms, if it can't find the 22825subprograms using @env{GCC_EXEC_PREFIX}. 22826 22827@item LIBRARY_PATH 22828@findex LIBRARY_PATH 22829The value of @env{LIBRARY_PATH} is a colon-separated list of 22830directories, much like @env{PATH}. When configured as a native compiler, 22831GCC tries the directories thus specified when searching for special 22832linker files, if it can't find them using @env{GCC_EXEC_PREFIX}. Linking 22833using GCC also uses these directories when searching for ordinary 22834libraries for the @option{-l} option (but directories specified with 22835@option{-L} come first). 22836 22837@item LANG 22838@findex LANG 22839@cindex locale definition 22840This variable is used to pass locale information to the compiler. One way in 22841which this information is used is to determine the character set to be used 22842when character literals, string literals and comments are parsed in C and C++. 22843When the compiler is configured to allow multibyte characters, 22844the following values for @env{LANG} are recognized: 22845 22846@table @samp 22847@item C-JIS 22848Recognize JIS characters. 22849@item C-SJIS 22850Recognize SJIS characters. 22851@item C-EUCJP 22852Recognize EUCJP characters. 22853@end table 22854 22855If @env{LANG} is not defined, or if it has some other value, then the 22856compiler uses @code{mblen} and @code{mbtowc} as defined by the default locale to 22857recognize and translate multibyte characters. 22858@end table 22859 22860@noindent 22861Some additional environment variables affect the behavior of the 22862preprocessor. 22863 22864@include cppenv.texi 22865 22866@c man end 22867 22868@node Precompiled Headers 22869@section Using Precompiled Headers 22870@cindex precompiled headers 22871@cindex speed of compilation 22872 22873Often large projects have many header files that are included in every 22874source file. The time the compiler takes to process these header files 22875over and over again can account for nearly all of the time required to 22876build the project. To make builds faster, GCC allows you to 22877@dfn{precompile} a header file. 22878 22879To create a precompiled header file, simply compile it as you would any 22880other file, if necessary using the @option{-x} option to make the driver 22881treat it as a C or C++ header file. You may want to use a 22882tool like @command{make} to keep the precompiled header up-to-date when 22883the headers it contains change. 22884 22885A precompiled header file is searched for when @code{#include} is 22886seen in the compilation. As it searches for the included file 22887(@pxref{Search Path,,Search Path,cpp,The C Preprocessor}) the 22888compiler looks for a precompiled header in each directory just before it 22889looks for the include file in that directory. The name searched for is 22890the name specified in the @code{#include} with @samp{.gch} appended. If 22891the precompiled header file can't be used, it is ignored. 22892 22893For instance, if you have @code{#include "all.h"}, and you have 22894@file{all.h.gch} in the same directory as @file{all.h}, then the 22895precompiled header file is used if possible, and the original 22896header is used otherwise. 22897 22898Alternatively, you might decide to put the precompiled header file in a 22899directory and use @option{-I} to ensure that directory is searched 22900before (or instead of) the directory containing the original header. 22901Then, if you want to check that the precompiled header file is always 22902used, you can put a file of the same name as the original header in this 22903directory containing an @code{#error} command. 22904 22905This also works with @option{-include}. So yet another way to use 22906precompiled headers, good for projects not designed with precompiled 22907header files in mind, is to simply take most of the header files used by 22908a project, include them from another header file, precompile that header 22909file, and @option{-include} the precompiled header. If the header files 22910have guards against multiple inclusion, they are skipped because 22911they've already been included (in the precompiled header). 22912 22913If you need to precompile the same header file for different 22914languages, targets, or compiler options, you can instead make a 22915@emph{directory} named like @file{all.h.gch}, and put each precompiled 22916header in the directory, perhaps using @option{-o}. It doesn't matter 22917what you call the files in the directory; every precompiled header in 22918the directory is considered. The first precompiled header 22919encountered in the directory that is valid for this compilation is 22920used; they're searched in no particular order. 22921 22922There are many other possibilities, limited only by your imagination, 22923good sense, and the constraints of your build system. 22924 22925A precompiled header file can be used only when these conditions apply: 22926 22927@itemize 22928@item 22929Only one precompiled header can be used in a particular compilation. 22930 22931@item 22932A precompiled header can't be used once the first C token is seen. You 22933can have preprocessor directives before a precompiled header; you cannot 22934include a precompiled header from inside another header. 22935 22936@item 22937The precompiled header file must be produced for the same language as 22938the current compilation. You can't use a C precompiled header for a C++ 22939compilation. 22940 22941@item 22942The precompiled header file must have been produced by the same compiler 22943binary as the current compilation is using. 22944 22945@item 22946Any macros defined before the precompiled header is included must 22947either be defined in the same way as when the precompiled header was 22948generated, or must not affect the precompiled header, which usually 22949means that they don't appear in the precompiled header at all. 22950 22951The @option{-D} option is one way to define a macro before a 22952precompiled header is included; using a @code{#define} can also do it. 22953There are also some options that define macros implicitly, like 22954@option{-O} and @option{-Wdeprecated}; the same rule applies to macros 22955defined this way. 22956 22957@item If debugging information is output when using the precompiled 22958header, using @option{-g} or similar, the same kind of debugging information 22959must have been output when building the precompiled header. However, 22960a precompiled header built using @option{-g} can be used in a compilation 22961when no debugging information is being output. 22962 22963@item The same @option{-m} options must generally be used when building 22964and using the precompiled header. @xref{Submodel Options}, 22965for any cases where this rule is relaxed. 22966 22967@item Each of the following options must be the same when building and using 22968the precompiled header: 22969 22970@gccoptlist{-fexceptions} 22971 22972@item 22973Some other command-line options starting with @option{-f}, 22974@option{-p}, or @option{-O} must be defined in the same way as when 22975the precompiled header was generated. At present, it's not clear 22976which options are safe to change and which are not; the safest choice 22977is to use exactly the same options when generating and using the 22978precompiled header. The following are known to be safe: 22979 22980@gccoptlist{-fmessage-length= -fpreprocessed -fsched-interblock @gol 22981-fsched-spec -fsched-spec-load -fsched-spec-load-dangerous @gol 22982-fsched-verbose=@var{number} -fschedule-insns -fvisibility= @gol 22983-pedantic-errors} 22984 22985@end itemize 22986 22987For all of these except the last, the compiler automatically 22988ignores the precompiled header if the conditions aren't met. If you 22989find an option combination that doesn't work and doesn't cause the 22990precompiled header to be ignored, please consider filing a bug report, 22991see @ref{Bugs}. 22992 22993If you do use differing options when generating and using the 22994precompiled header, the actual behavior is a mixture of the 22995behavior for the options. For instance, if you use @option{-g} to 22996generate the precompiled header but not when using it, you may or may 22997not get debugging information for routines in the precompiled header. 22998