Return to internals.tex CVS log

Up to [Research Unix] / researchv10dc / cmd / gcc

Annotation of researchv10dc/cmd/gcc/internals.tex, revision 1.1

1.1     ! root        1: \input texinfo  @c -*-texinfo-*-
        !             2: 
        !             3: @settitle Internals of GNU CC
        !             4: @setfilename internals
        !             5: 
        !             6: @ifinfo
        !             7: This file documents the internals of the GNU compiler.
        !             8: 
        !             9: Copyright (C) 1988 Free Software Foundation, Inc.
        !            10: 
        !            11: Permission is granted to make and distribute verbatim copies of
        !            12: this manual provided the copyright notice and this permission notice
        !            13: are preserved on all copies.
        !            14: 
        !            15: @ignore
        !            16: Permission is granted to process this file through Tex and print the
        !            17: results, provided the printed document carries copying permission
        !            18: notice identical to this one except for the removal of this paragraph
        !            19: (this paragraph not being relevant to the printed manual).
        !            20: 
        !            21: @end ignore
        !            22: Permission is granted to copy and distribute modified versions of this
        !            23: manual under the conditions for verbatim copying, provided also that the
        !            24: section entitled ``GNU CC General Public License'' is included exactly as
        !            25: in the original, and provided that the entire resulting derived work is
        !            26: distributed under the terms of a permission notice identical to this one.
        !            27: 
        !            28: Permission is granted to copy and distribute translations of this manual
        !            29: into another language, under the above conditions for modified versions,
        !            30: except that the section entitled ``GNU CC General Public License'' and
        !            31: this permission notice may be included in translations approved by the
        !            32: Free Software Foundation instead of in the original English.
        !            33: @end ifinfo
        !            34: 
        !            35: @setchapternewpage odd
        !            36: 
        !            37: @titlepage
        !            38: @center @titlefont{Internals of GNU CC}
        !            39: @sp 2
        !            40: @center Richard M. Stallman
        !            41: @sp 3
        !            42: @center last updated 24 April 1988
        !            43: @sp 1
        !            44: @center for version 1.21
        !            45: @page
        !            46: @vskip 0pt plus 1filll
        !            47: Copyright @copyright{} 1988 Free Software Foundation, Inc.
        !            48: 
        !            49: Permission is granted to make and distribute verbatim copies of
        !            50: this manual provided the copyright notice and this permission notice
        !            51: are preserved on all copies.
        !            52: 
        !            53: Permission is granted to copy and distribute modified versions of this
        !            54: manual under the conditions for verbatim copying, provided also that the
        !            55: section entitled ``GNU CC General Public License'' is included exactly as
        !            56: in the original, and provided that the entire resulting derived work is
        !            57: distributed under the terms of a permission notice identical to this one.
        !            58: 
        !            59: Permission is granted to copy and distribute translations of this manual
        !            60: into another language, under the above conditions for modified versions,
        !            61: except that the section entitled ``GNU CC General Public License'' may be
        !            62: included in a translation approved by the author instead of in the original
        !            63: English.
        !            64: @end titlepage
        !            65: @page
        !            66: 
        !            67: @ifinfo
        !            68: @node Top, Copying,, (DIR)
        !            69: @ichapter Introduction
        !            70: 
        !            71: This manual documents how to run, install and port the GNU C compiler, as
        !            72: well as its new features and incompatibilities, and how to report bugs.
        !            73: 
        !            74: @end ifinfo
        !            75: @menu
        !            76: * Copying::         GNU CC General Public License says
        !            77:                      how you can copy and share GNU CC.
        !            78: * Contributors::    People who have contributed to GNU CC.
        !            79: * Options::         Command options supported by @samp{gcc}.
        !            80: * Installation::    How to configure, compile and install GNU CC.
        !            81: * Trouble::         If you have trouble installing GNU CC.
        !            82: * Incompatibilities:: Incompatibilities of GNU CC.
        !            83: * Extensions::      GNU extensions to the C language.
        !            84: * Bugs::            How to report bugs (if you want to get them fixed).
        !            85: * Portability::     Goals of GNU CC's portability features.
        !            86: * Interface::       Function-call interface of GNU CC output.
        !            87: * Passes::          Order of passes, what they do, and what each file is for.
        !            88: * RTL::             The intermediate representation that most passes work on.
        !            89: * Machine Desc::    How to write machine description instruction patterns.
        !            90: * Machine Macros::  How to write the machine description C macros.
        !            91: @end menu
        !            92: 
        !            93: @node Copying, Contributors, Top, Top
        !            94: @unnumbered GNU CC GENERAL PUBLIC LICENSE
        !            95: @center (Clarified 11 Feb 1988)
        !            96: 
        !            97:   The license agreements of most software companies keep you at the
        !            98: mercy of those companies.  By contrast, our general public license is
        !            99: intended to give everyone the right to share GNU CC.  To make sure that
        !           100: you get the rights we want you to have, we need to make restrictions
        !           101: that forbid anyone to deny you these rights or to ask you to surrender
        !           102: the rights.  Hence this license agreement.
        !           103: 
        !           104:   Specifically, we want to make sure that you have the right to give
        !           105: away copies of GNU CC, that you receive source code or else can get it
        !           106: if you want it, that you can change GNU CC or use pieces of it in new
        !           107: free programs, and that you know you can do these things.
        !           108: 
        !           109:   To make sure that everyone has such rights, we have to forbid you to
        !           110: deprive anyone else of these rights.  For example, if you distribute
        !           111: copies of GNU CC, you must give the recipients all the rights that you
        !           112: have.  You must make sure that they, too, receive or can get the
        !           113: source code.  And you must tell them their rights.
        !           114: 
        !           115:   Also, for our own protection, we must make certain that everyone
        !           116: finds out that there is no warranty for GNU CC.  If GNU CC is modified by
        !           117: someone else and passed on, we want its recipients to know that what
        !           118: they have is not what we distributed, so that any problems introduced
        !           119: by others will not reflect on our reputation.
        !           120: 
        !           121:   Therefore we (Richard Stallman and the Free Software Foundation,
        !           122: Inc.) make the following terms which say what you must do to be
        !           123: allowed to distribute or change GNU CC.
        !           124: 
        !           125: @unnumberedsec COPYING POLICIES
        !           126: 
        !           127: @enumerate
        !           128: @item
        !           129: You may copy and distribute verbatim copies of GNU CC source code as
        !           130: you receive it, in any medium, provided that you conspicuously and
        !           131: appropriately publish on each copy a valid copyright notice
        !           132: ``Copyright @copyright{} 1988 Free Software Foundation, Inc.'' (or
        !           133: with whatever year is appropriate); keep intact the notices on all
        !           134: files that refer to this License Agreement and to the absence of any
        !           135: warranty; and give any other recipients of the GNU CC program a copy
        !           136: of this License Agreement along with the program.  You may charge a
        !           137: distribution fee for the physical act of transferring a copy.
        !           138: 
        !           139: @item
        !           140: You may modify your copy or copies of GNU CC or any portion of it,
        !           141: and copy and distribute such modifications under the terms of
        !           142: Paragraph 1 above, provided that you also do the following:
        !           143: 
        !           144: @itemize @bullet
        !           145: @item
        !           146: cause the modified files to carry prominent notices stating
        !           147: that you changed the files and the date of any change; and
        !           148: 
        !           149: @item
        !           150: cause the whole of any work that you distribute or publish, that
        !           151: in whole or in part contains or is a derivative of GNU CC or any
        !           152: part thereof, to be licensed at no charge to all third parties on
        !           153: terms identical to those contained in this License Agreement
        !           154: (except that you may choose to grant more extensive warranty
        !           155: protection to some or all third parties, at your option).
        !           156: 
        !           157: @item
        !           158: You may charge a distribution fee for the physical act of
        !           159: transferring a copy, and you may at your option offer warranty
        !           160: protection in exchange for a fee.
        !           161: @end itemize
        !           162: 
        !           163: Mere aggregation of another unrelated program with this program (or its
        !           164: derivative) on a volume of a storage or distribution medium does not bring
        !           165: the other program under the scope of these terms.
        !           166: 
        !           167: @item
        !           168: You may copy and distribute GNU CC (or a portion or derivative of it,
        !           169: under Paragraph 2) in object code or executable form under the terms
        !           170: of Paragraphs 1 and 2 above provided that you also do one of the
        !           171: following:
        !           172: 
        !           173: @itemize @bullet
        !           174: @item
        !           175: accompany it with the complete corresponding machine-readable
        !           176: source code, which must be distributed under the terms of
        !           177: Paragraphs 1 and 2 above; or,
        !           178: 
        !           179: @item
        !           180: accompany it with a written offer, valid for at least three
        !           181: years, to give any third party free (except for a nominal
        !           182: shipping charge) a complete machine-readable copy of the
        !           183: corresponding source code, to be distributed under the terms of
        !           184: Paragraphs 1 and 2 above; or,
        !           185: 
        !           186: @item
        !           187: accompany it with the information you received as to where the
        !           188: corresponding source code may be obtained.  (This alternative is
        !           189: allowed only for noncommercial distribution and only if you
        !           190: received the program in object code or executable form alone.)
        !           191: @end itemize
        !           192: 
        !           193: For an executable file, complete source code means all the source code
        !           194: for all modules it contains; but, as a special exception, it need not
        !           195: include source code for modules which are standard libraries that
        !           196: accompany the operating system on which the executable file runs.
        !           197: 
        !           198: @item
        !           199: You may not copy, sublicense, distribute or transfer GNU CC except as
        !           200: expressly provided under this License Agreement.  Any attempt
        !           201: otherwise to copy, sublicense, distribute or transfer GNU CC is void
        !           202: and your rights to use the program under this License agreement shall
        !           203: be automatically terminated.  However, parties who have received
        !           204: computer software programs from you with this License Agreement will
        !           205: not have their licenses terminated so long as such parties remain in
        !           206: full compliance.
        !           207: 
        !           208: @item
        !           209: If you wish to incorporate parts of GNU CC into other free programs
        !           210: whose distribution conditions are different, write to the Free Software
        !           211: Foundation at 675 Mass Ave, Cambridge, MA 02139.  We have not yet worked
        !           212: out a simple rule that can be stated here, but we will often permit this.
        !           213: We will be guided by the two goals of preserving the free status of all
        !           214: derivatives of our free software and of promoting the sharing and reuse of
        !           215: software.
        !           216: @end enumerate
        !           217: 
        !           218: Your comments and suggestions about our licensing policies and our
        !           219: software are welcome!  Please contact the Free Software Foundation, Inc.,
        !           220: 675 Mass Ave, Cambridge, MA 02139, or call (617) 876-3296.
        !           221: 
        !           222: @unnumberedsec NO WARRANTY
        !           223: 
        !           224:   BECAUSE GNU CC IS LICENSED FREE OF CHARGE, WE PROVIDE ABSOLUTELY NO
        !           225: WARRANTY, TO THE EXTENT PERMITTED BY APPLICABLE STATE LAW.  EXCEPT
        !           226: WHEN OTHERWISE STATED IN WRITING, FREE SOFTWARE FOUNDATION, INC,
        !           227: RICHARD M. STALLMAN AND/OR OTHER PARTIES PROVIDE GNU CC "AS IS" WITHOUT
        !           228: WARRANTY OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT
        !           229: LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
        !           230: A PARTICULAR PURPOSE.  THE ENTIRE RISK AS TO THE QUALITY AND
        !           231: PERFORMANCE OF GNU CC IS WITH YOU.  SHOULD GNU CC PROVE DEFECTIVE, YOU
        !           232: ASSUME THE COST OF ALL NECESSARY SERVICING, REPAIR OR CORRECTION.
        !           233: 
        !           234:  IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW WILL RICHARD M.
        !           235: STALLMAN, THE FREE SOFTWARE FOUNDATION, INC., AND/OR ANY OTHER PARTY
        !           236: WHO MAY MODIFY AND REDISTRIBUTE GNU CC AS PERMITTED ABOVE, BE LIABLE TO
        !           237: YOU FOR DAMAGES, INCLUDING ANY LOST PROFITS, LOST MONIES, OR OTHER
        !           238: SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE USE OR
        !           239: INABILITY TO USE (INCLUDING BUT NOT LIMITED TO LOSS OF DATA OR DATA
        !           240: BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY THIRD PARTIES OR A
        !           241: FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS) GNU CC, EVEN
        !           242: IF YOU HAVE BEEN ADVISED OF THE POSSIBILITY OF SUCH DAMAGES, OR FOR
        !           243: ANY CLAIM BY ANY OTHER PARTY.
        !           244: 
        !           245: @node Contributors, Options, Copying, Top
        !           246: @unnumbered Contributors to GNU CC
        !           247: 
        !           248: In addition to Richard Stallman, several people have written parts
        !           249: of GNU CC.
        !           250: 
        !           251: @itemize @bullet
        !           252: @item
        !           253: The idea of using RTL and some of the optimization ideas came from the
        !           254: U. of Arizona Portable Optimizer, written by Jack Davidson and
        !           255: Christopher Fraser.  See ``Register Allocation and Exhaustive Peephole
        !           256: Optimization'', Software Practice and Experience 14 (9), Sept. 1984,
        !           257: 857-866.
        !           258: 
        !           259: @item
        !           260: Paul Rubin wrote most of the preprocessor.
        !           261: 
        !           262: @item
        !           263: Leonard Tower wrote parts of the parser, RTL generator, RTL
        !           264: definitions, and of the Vax machine description.
        !           265: 
        !           266: @item
        !           267: Ted Lemon wrote parts of the RTL reader and printer.
        !           268: 
        !           269: @item
        !           270: Nobuyuki Hikichi of Software Research Associates, Tokyo, contributed
        !           271: the support for the SONY NEWS machine.
        !           272: 
        !           273: @item
        !           274: Charles LaBrec contributed the support for the Integrated Solutions
        !           275: 68020 system.
        !           276: 
        !           277: @item
        !           278: Michael Tiemann of MCC wrote the description of the National
        !           279: Semiconductor 32000 series cpu, with some contributions from Jan Stein
        !           280: of the Chalmers Computer Club.  Tiemann also wrote the code for inline
        !           281: function integration.
        !           282: 
        !           283: @item
        !           284: Robert Brown implemented the support for Encore 32000 systems.
        !           285: 
        !           286: @item
        !           287: Michael Kashtan of SRI adapted GNU CC to the Vomit-Making System.
        !           288: 
        !           289: @item
        !           290: Alex Crain provided changes for the 3b1.
        !           291: 
        !           292: @item
        !           293: Chris Hanson and another person who should remind me of his name
        !           294: assisted in making GNU CC work on HP-UX for the 9000 series 300.
        !           295: @end itemize
        !           296: 
        !           297: @node Options, Installation, Contributors, Top
        !           298: @chapter GNU CC Command Options
        !           299: 
        !           300: The GNU C compiler uses a command syntax much like the Unix C compiler.
        !           301: The @code{gcc} program accepts options and file names as operands.
        !           302: Multiple single-letter options may @emph{not} be grouped: @samp{-dr} is
        !           303: very different from @samp{-d -r}.
        !           304: 
        !           305: When you invoke GNU CC, it normally does preprocessing, compilation,
        !           306: assembly and linking.  File names which end in @samp{.c} are taken as C
        !           307: source to be preprocessed and compiled; compiler output files plus any
        !           308: input files with names ending in @samp{.s} are assembled; then the
        !           309: resulting object files, plus any other input files, are linked together to
        !           310: produce an executable.
        !           311: 
        !           312: Command options allow you to stop this process at an intermediate stage.
        !           313: For example, the @samp{-c} option says not to run the linker.  Then the
        !           314: output consists of object files output by the assembler.
        !           315: 
        !           316: Other command options are passed on to one stage.  Some options control
        !           317: the preprocessor and others the compiler itself.  Yet other options
        !           318: control the assembler and linker; these are not documented here because the
        !           319: GNU assembler and linker are not yet released.
        !           320: 
        !           321: Here are the options to control the overall compilation process, including
        !           322: those that say whether to link, whether to assemble, and so on.
        !           323: 
        !           324: @table @samp
        !           325: @item -o @var{file}
        !           326: Place output in file @var{file}.  This applies regardless to whatever
        !           327: sort of output is being produced, whether it be an executable file,
        !           328: an object file, an assembler file or preprocessed C code.
        !           329: 
        !           330: If @samp{-o} is not specified, the default is to put an executable file
        !           331: in @file{a.out}, the object file @file{@var{source}.c} in
        !           332: @file{@var{source}.o}, an assembler file in @file{@var{source}.s}, and
        !           333: preprocessed C on standard output.@refill
        !           334: 
        !           335: @item -c
        !           336: Compile or assemble the source files, but do not link.  Produce object
        !           337: files with names made by replacing @samp{.c} or @samp{.s} with
        !           338: @samp{.o} at the end of the input file names.  Do nothing at all for
        !           339: object files specified as input.
        !           340: 
        !           341: @item -S
        !           342: Compile into assembler code but do not assemble.  The assembler output
        !           343: file name is made by replacing @samp{.c} with @samp{.s} at the end of
        !           344: the input file name.  Do nothing at all for assembler source files or
        !           345: object files specified as input.
        !           346: 
        !           347: @item -E
        !           348: Run only the C preprocessor.  Preprocess all the C source files
        !           349: specified and output the results to standard output.
        !           350: 
        !           351: @item -v
        !           352: Compiler driver program prints the commands it executes as it runs
        !           353: the preprocessor, compiler proper, assembler and linker.  Some of
        !           354: these are directed to print their own version numbers.
        !           355: 
        !           356: @item -B@var{prefix}
        !           357: Compiler driver program tries @var{prefix} as a prefix for each
        !           358: program it tries to run.  These programs are @file{cpp}, @file{cc1},
        !           359: @file{as} and @file{ld}.
        !           360: 
        !           361: For each subprogram to be run, the compiler driver first tries the
        !           362: @samp{-B} prefix, if any.  If that name is not found, or if @samp{-B}
        !           363: was not specified, the driver tries two standard prefixes, which are
        !           364: @file{/usr/lib/gcc-} and @file{/usr/local/lib/gcc-}.  If neither of
        !           365: those results in a file name that is found, the unmodified program
        !           366: name is searched for using the directories specified in your
        !           367: @samp{PATH} environment variable.
        !           368: 
        !           369: The run-time support file @file{gnulib} is also searched for using
        !           370: the @samp{-B} prefix, if needed.  If it is not found there, the two
        !           371: standard prefixes above are tried, and that is all.  The file is left
        !           372: out of the link if it is not found by those means.  Most of the time,
        !           373: on most machines, you can do without it.
        !           374: @end table
        !           375: 
        !           376: These options control the details of C compilation itself.
        !           377: 
        !           378: @table @samp
        !           379: @item -ansi
        !           380: Support all ANSI standard C programs.
        !           381: 
        !           382: This turns off certain features of GNU C that are incompatible with
        !           383: ANSI C, such as the @code{asm}, @code{inline} and @code{typeof}
        !           384: keywords, and predefined macros such as @code{unix} and @code{vax}
        !           385: that identify the type of system you are using.  It also enables the
        !           386: undesirable and rarely used ANSI trigraph feature.
        !           387: 
        !           388: The @samp{-ansi} option does not cause non-ANSI programs to be
        !           389: rejected gratuitously.  For that, @samp{-pedantic} is required in
        !           390: addition to @samp{-ansi}.
        !           391: 
        !           392: The macro @code{__STRICT_ANSI__} is predefined when the @samp{-ansi}
        !           393: option is used.  Some header files may notice this macro and refrain
        !           394: from declaring certain functions or defining certain macros that the
        !           395: ANSI standard doesn't call for; this is to avoid interfering with
        !           396: any programs that might use these names for other things.
        !           397: 
        !           398: @item -traditional
        !           399: Attempt to support some aspects of traditional C compilers.
        !           400: Specifically:
        !           401: 
        !           402: @itemize @bullet
        !           403: @item
        !           404: All @code{extern} declarations take effect globally even if they
        !           405: are written inside of a function definition.  This includes implicit
        !           406: declarations of functions.
        !           407: 
        !           408: @item
        !           409: The keywords @code{typeof}, @code{inline}, @code{signed}, @code{const}
        !           410: and @code{volatile} are not recognized.@refill
        !           411: 
        !           412: @item
        !           413: Comparisons between pointers and integers are always allowed.
        !           414: 
        !           415: @item
        !           416: Integer types @code{unsigned short} and @code{unsigned char} promote
        !           417: to @code{unsigned int}.
        !           418: 
        !           419: @item
        !           420: In the preprocessor, comments convert to nothing at all, rather than to
        !           421: a space.  This allows traditional token concatenation.
        !           422: 
        !           423: @item
        !           424: In the preprocessor, single and double quote characters are ignored
        !           425: when scanning macro definitions, so that macro arguments can be replaced
        !           426: even within a string or character constant.  Quote characters are also
        !           427: ignored when skipping text inside a failing conditional directive.
        !           428: @end itemize
        !           429: 
        !           430: @item -O
        !           431: Optimize.  Optimizing compilation takes somewhat more time, and a lot
        !           432: more memory for a large function.
        !           433: 
        !           434: Without @samp{-O}, the compiler's goal is to reduce the cost of
        !           435: compilation and to make debugging produce the expected results.
        !           436: Statements are independent: if you stop the program with a breakpoint
        !           437: between statements, you can then assign a new value to any variable or
        !           438: change the program counter to any other statement in the function and
        !           439: get exactly the results you would expect from the source code.
        !           440: 
        !           441: Without @samp{-O}, only variables declared @code{register} are
        !           442: allocated in registers.  The resulting compiled code is a little worse
        !           443: than produced by PCC without @samp{-O}.
        !           444: 
        !           445: With @samp{-O}, the compiler tries to reduce code size and execution
        !           446: time.
        !           447: 
        !           448: Some of the @samp{-f} options described below turn specific kinds of
        !           449: optimization on or off.
        !           450: 
        !           451: @item -g
        !           452: Produce debugging information in DBX format.
        !           453: 
        !           454: Unlike most other C compilers, GNU CC allows you to use @samp{-g} with
        !           455: @samp{-O}.  The shortcuts taken by optimized code may occasionally
        !           456: produce surprising results: some variables you declared may not exist
        !           457: at all; flow of control may briefly move where you did not expect it;
        !           458: some statements may not be executed because they compute constant
        !           459: results or their values were already at hand; some statements may
        !           460: execute in different places because they were moved out of loops.
        !           461: Nevertheless it proves possible to debug optimized output.  This makes
        !           462: it reasonable to use the optimizer for programs that might have bugs.
        !           463: 
        !           464: @item -gg
        !           465: Produce debugging information in GDB's own format.  This requires
        !           466: the GNU assembler and linker in order to work.
        !           467: 
        !           468: @item -w
        !           469: Inhibit all warning messages.
        !           470: 
        !           471: @item -W
        !           472: Print extra warning messages for these events:
        !           473: 
        !           474: @itemize @bullet
        !           475: @item
        !           476: An automatic variable is used without first being initialized.
        !           477: 
        !           478: These warnings are possible only in optimizing compilation,
        !           479: because they require data flow information that is computed only
        !           480: when optimizing.  They occur only for variables that are
        !           481: candidates for register allocation.  Therefore, they do not occur
        !           482: for a variable that is declared @code{volatile}, or whose address
        !           483: is taken, or whose size is other than 1, 2, 4 or 8 bytes.  Also,
        !           484: they do not occur for structures, unions or arrays, even when
        !           485: they are in registers.
        !           486: 
        !           487: Note that there may be no warning about a variable that is used
        !           488: only to compute a value that itself is never used, because such
        !           489: computations may be deleted by the flow analysis pass before the
        !           490: warnings are printed.
        !           491: 
        !           492: These warnings are made optional because GNU CC is not smart
        !           493: enough to see all the reasons why the code might be correct
        !           494: despite appearing to have an error.  Here is one example of how
        !           495: this can happen:
        !           496: 
        !           497: @example
        !           498: @{
        !           499:   int x;
        !           500:   switch (y)
        !           501:     @{
        !           502:     case 1: x = 1;
        !           503:       break;
        !           504:     case 2: x = 4;
        !           505:       break;
        !           506:     case 3: x = 5;
        !           507:     @}
        !           508:   foo (x);
        !           509: @}
        !           510: @end example
        !           511: 
        !           512: @noindent
        !           513: If the value of @code{y} is always 1, 2 or 3, then @code{x} is
        !           514: always initialized, but GNU CC doesn't know this.  Here is
        !           515: another common case:
        !           516: 
        !           517: @example
        !           518: @{
        !           519:   int save_y;
        !           520:   if (change_y) save_y = y, y = new_y;
        !           521:   @dots{}
        !           522:   if (change_y) y = save_y;
        !           523: @}
        !           524: @end example
        !           525: 
        !           526: @noindent
        !           527: This has no bug because @code{x} is used only if it is set.
        !           528: 
        !           529: @item
        !           530: A nonvolatile automatic variable might be changed by a call to
        !           531: @code{longjmp}.  These warnings as well are possible only in
        !           532: optimizing compilation.
        !           533: 
        !           534: The compiler sees only the calls to @code{setjmp}.  It cannot know
        !           535: where @code{longjmp} will be called; in fact, a signal handler could
        !           536: call it at any point in the code.  As a result, you may get a warning
        !           537: even when there is in fact no problem because @code{longjmp} cannot
        !           538: in fact be called at the place which would cause a problem.
        !           539: 
        !           540: @item
        !           541: A function can return either with or without a value.  (Falling
        !           542: off the end of the function body is considered returning without
        !           543: a value.)  For example, this function would inspire such a
        !           544: warning:
        !           545: 
        !           546: @example
        !           547: foo (a)
        !           548: @{
        !           549:   if (a > 0)
        !           550:     return a;
        !           551: @}
        !           552: @end example
        !           553: 
        !           554: Spurious warnings can occur because GNU CC does not realize that
        !           555: certain functions (including @code{abort} and @code{longjmp})
        !           556: will never return.
        !           557: @end itemize
        !           558: 
        !           559: In the future, other useful warnings may also be enabled by this
        !           560: option.
        !           561: 
        !           562: @item -Wimplicit
        !           563: Warn whenever a function is implicitly declared.
        !           564: 
        !           565: @item -Wreturn-type
        !           566: Warn whenever a function is defined with a return-type that defaults
        !           567: to @code{int}.  Also warn about any @code{return} statement with no
        !           568: return-value in a function whose return-type is not @code{void}.
        !           569: 
        !           570: @item -Wcomment
        !           571: Warn whenever a comment-start sequence @samp{/*} appears in a comment.
        !           572: 
        !           573: @item -Wall
        !           574: All of the above @samp{-W} options combined.
        !           575: 
        !           576: @item -p
        !           577: Generate extra code to write profile information suitable for the
        !           578: analysis program @code{prof}.
        !           579: 
        !           580: @item -pg
        !           581: Generate extra code to write profile information suitable for the
        !           582: analysis program @code{gprof}.
        !           583: 
        !           584: @item -l@var{library}
        !           585: Search a standard list of directories for a library named
        !           586: @var{library}, which is actually a file named
        !           587: @file{lib@var{library}.a}.  The linker uses this file as if it
        !           588: had been specified precisely by name.
        !           589: 
        !           590: The directories searched include several standard system directories
        !           591: plus any that you specify with @samp{-L}.
        !           592: 
        !           593: Normally the files found this way are library files---archive files
        !           594: whose members are object files.  The linker handles an archive file by
        !           595: through it for members which define symbols that have so far been
        !           596: referenced but not defined.  But if the file that is found is an
        !           597: ordinary object file, it is linked in the usual fashion.  The only
        !           598: difference between an @samp{-l} option and the full file name of the
        !           599: file that is found is syntactic and the fact that several directories
        !           600: are searched.
        !           601: 
        !           602: @item -L@var{dir}
        !           603: Add directory @var{dir} to the list of directories to be searched
        !           604: for @samp{-l}.
        !           605: 
        !           606: @item -nostdlib
        !           607: Don't use the standard system libraries and startup files when
        !           608: linking.  Only the files you specify (plus @file{gnulib}) will be
        !           609: passed to the linker.
        !           610: 
        !           611: @item -m@var{machinespec}
        !           612: Machine-dependent option specifying something about the type of target
        !           613: machine.  These options are defined by the macro
        !           614: @code{TARGET_SWITCHES} in the machine description.  The default for
        !           615: the options is also defined by that macro, which enables you to change
        !           616: the defaults.@refill
        !           617: 
        !           618: These are the @samp{-m} options defined in the 68000 machine
        !           619: description:
        !           620: 
        !           621: @table @samp
        !           622: @item -m68020
        !           623: Generate output for a 68020 (rather than a 68000).  This is the
        !           624: default if you use the unmodified sources.
        !           625: 
        !           626: @item -m68000
        !           627: Generate output for a 68000 (rather than a 68020).
        !           628: 
        !           629: @item -m68881
        !           630: Generate output containing 68881 instructions for floating point.
        !           631: This is the default if you use the unmodified sources.
        !           632: 
        !           633: @item -msoft-float
        !           634: Generate output containing library calls for floating point.
        !           635: 
        !           636: @item -mshort
        !           637: Consider type @code{int} to be 16 bits wide, like @code{short int}.
        !           638: 
        !           639: @item -mnobitfield
        !           640: Do not use the bit-field instructions.  @samp{-m68000} implies
        !           641: @samp{-mnobitfield}.
        !           642: 
        !           643: @item -mbitfield
        !           644: Do use the bit-field instructions.  @samp{-m68020} implies
        !           645: @samp{-mbitfield}.  This is the default if you use the unmodified
        !           646: sources.
        !           647: 
        !           648: @item -mrtd
        !           649: Use a different function-calling convention, in which functions
        !           650: that take a fixed number of arguments return with the @code{rtd}
        !           651: instruction, which pops their arguments while returning.  This
        !           652: saves one instruction in the caller since there is no need to pop
        !           653: the arguments there.
        !           654: 
        !           655: This calling convention is incompatible with the one normally
        !           656: used on Unix, so you cannot use it if you need to call libraries
        !           657: compiled with the Unix compiler.
        !           658: 
        !           659: Also, you must provide function prototypes for all functions that
        !           660: take variable numbers of arguments (including @code{printf});
        !           661: otherwise incorrect code will be generated for calls to those
        !           662: functions.
        !           663: 
        !           664: In addition, seriously incorrect code will result if you call a
        !           665: function with too many arguments.  (Normally, extra arguments are
        !           666: harmlessly ignored.)
        !           667: 
        !           668: The @code{rtd} instruction is supported by the 68010 and 68020
        !           669: processors, but not by the 68000.
        !           670: @end table
        !           671: 
        !           672: These @samp{-m} options are defined in the Vax machine description:
        !           673: 
        !           674: @table @samp
        !           675: @item -munix
        !           676: Do not output certain jump instructions (@code{aobleq} and so on)
        !           677: that the Unix assembler for the Vax cannot handle across long
        !           678: ranges.
        !           679: 
        !           680: @item -mgnu
        !           681: Do output those jump instructions, on the assumption that you
        !           682: will assemble with the GNU assembler.
        !           683: 
        !           684: @item -mg
        !           685: Output code for g-format floating point numbers instead of d-format.
        !           686: @end table
        !           687: 
        !           688: @item -f@var{flag}
        !           689: Specify machine-independent flags.  These are the flags:
        !           690: 
        !           691: @table @samp
        !           692: @item -ffloat-store
        !           693: Do not store floating-point variables in registers.  This
        !           694: prevents undesirable excess precision on machines such as the
        !           695: 68000 where the floating registers (of the 68881) keep more
        !           696: precision than a @code{double} is supposed to have.
        !           697: 
        !           698: For most programs, the excess precision does only good, but a few
        !           699: programs rely on the precise definition of IEEE floating point.
        !           700: Use @samp{-ffloat-store} for such programs.
        !           701: 
        !           702: @item -fno-asm
        !           703: Do not recognize @code{asm}, @code{inline} or @code{typeof} as a
        !           704: keyword.  These words may then be used as identifiers.
        !           705: 
        !           706: @item -fno-defer-pop
        !           707: Always pop the arguments to each function call as soon as that
        !           708: function returns.  Normally the compiler (when optimizing) lets
        !           709: arguments accumulate on the stack for several function calls and
        !           710: pops them all at once.
        !           711: 
        !           712: @item -fcombine-regs
        !           713: Allow the combine pass to combine an instruction that copies one
        !           714: register into another.  This might or might not produce better
        !           715: code when used in addition to @samp{-O}.  I am interested in
        !           716: hearing about the difference this makes.
        !           717: 
        !           718: @item -fforce-mem
        !           719: Force memory operands to be copied into registers before doing
        !           720: arithmetic on them.  This may produce better code by making all
        !           721: memory references potential common subexpressions.  When they are
        !           722: not common subexpressions, instruction combination should
        !           723: eliminate the separate register-load.  I am interested in hearing
        !           724: about the difference this makes.
        !           725: 
        !           726: @item -fforce-addr
        !           727: Force memory address constants to be copied into registers before
        !           728: doing arithmetic on them.  This may produce better code just as
        !           729: @samp{-fforce-mem} may.  I am interested in hearing about the
        !           730: difference this makes.
        !           731: 
        !           732: @item -fomit-frame-pointer
        !           733: Don't keep the frame pointer in a register for functions that
        !           734: don't need one.  This avoids the instructions to save, set up and
        !           735: restore frame pointers; it also makes an extra register available
        !           736: in many functions.  @strong{It also makes debugging impossible.}
        !           737: 
        !           738: On some machines, such as the Vax, this flag has no effect,
        !           739: because the standard calling sequence automatically handles the
        !           740: frame pointer and nothing is saved by pretending it doesn't
        !           741: exist.  The machine-description macro
        !           742: @code{FRAME_POINTER_REQUIRED} controls whether a target machine
        !           743: supports this flag.  @xref{Registers}.@refill
        !           744: 
        !           745: @item -finline-functions
        !           746: Integrate all simple functions into their callers.  The compiler
        !           747: heuristically decides which functions are simple enough to be
        !           748: worth integrating in this way.
        !           749: 
        !           750: If all calls to a given function are integrated, and the function
        !           751: is declared @code{static}, then the function is normally not
        !           752: output as assembler code in its own right.
        !           753: 
        !           754: @item -fkeep-inline-functions
        !           755: Even if all calls to a given function are integrated, and the
        !           756: function is declared @code{static}, nevertheless output a
        !           757: separate run-time callable version of the function.
        !           758: 
        !           759: @item -fwritable-strings
        !           760: Store string constants in the writable data segment and don't
        !           761: uniquize them.  This is for compatibility with old programs which
        !           762: assume they can write into string constants.  Writing into string
        !           763: constants is a very bad idea; ``constants'' should be constant.
        !           764: 
        !           765: @item -fno-function-cse
        !           766: Do not put function addresses in registers; make each instruction
        !           767: that calls a constant function contain the function's address
        !           768: explicitly.
        !           769: 
        !           770: This option results in less efficient code, but some strange
        !           771: hacks that alter the assembler output may be confused by the
        !           772: optimizations performed when this option is not used.
        !           773: 
        !           774: @item -fvolatile
        !           775: Consider all memory references through pointers to be volatile.
        !           776: 
        !           777: @item -funsigned-char
        !           778: Let the type @code{char} be the unsigned, like @code{unsigned
        !           779: char}.
        !           780: 
        !           781: Each kind of machine has a default for what @code{char} should
        !           782: be.  It is either like @code{unsigned char} by default or like
        !           783: @code{signed char} by default.  (Actually, at present, the
        !           784: default is always signed.)
        !           785: 
        !           786: The type @code{char} is always a distinct type from either
        !           787: @code{signed char} or @code{unsigned char}, even though its
        !           788: behavior is always just like one of those two.
        !           789: 
        !           790: @item -fsigned-char
        !           791: Let the type @code{char} be signed, like @code{signed char}.
        !           792: 
        !           793: @item -ffixed-@var{reg}
        !           794: Treat the register named @var{reg} as a fixed register; generated
        !           795: code should never refer to it (except perhaps as a stack pointer,
        !           796: frame pointer or in some other fixed role).
        !           797: 
        !           798: @var{reg} must be the name of a register.  The register names
        !           799: accepted are machine-specific and are defined in the
        !           800: @code{REGISTER_NAMES} macro in the machine description macro
        !           801: file.
        !           802: 
        !           803: @item -fcall-used-@var{reg}
        !           804: Treat the register named @var{reg} as an allocatable register
        !           805: that is clobbered by function calls.  It may be allocated for
        !           806: temporaries or variables that do not live across a call.
        !           807: Functions compiled this way will not save and restore the
        !           808: register @var{reg}.
        !           809: 
        !           810: Use of this flag for a register that has a fixed pervasive role
        !           811: in the machine's execution model, such as the stack pointer or
        !           812: frame pointer, will produce disastrous results.
        !           813: 
        !           814: @item -fcall-saved-@var{reg}
        !           815: Treat the register named @var{reg} as an allocatable register
        !           816: saved by functions.  It may be allocated even for temporaries or
        !           817: variables that live across a call.  Functions compiled this way
        !           818: will save and restore the register @var{reg} if they use it.
        !           819: 
        !           820: Use of this flag for a register that has a fixed pervasive role
        !           821: in the machine's execution model, such as the stack pointer or
        !           822: frame pointer, will produce disastrous results.
        !           823: 
        !           824: A different sort of disaster will result from the use of this
        !           825: flag for a register in which function values are may be returned.
        !           826: @end table
        !           827: 
        !           828: @item -d@var{letters}
        !           829: Says to make debugging dumps at times specified by @var{letters}.
        !           830: Here are the possible letters:
        !           831: 
        !           832: @table @samp
        !           833: @item r
        !           834: Dump after RTL generation.
        !           835: @item j
        !           836: Dump after first jump optimization.
        !           837: @item J
        !           838: Dump after last jump optimization.
        !           839: @item s
        !           840: Dump after CSE (including the jump optimization that sometimes
        !           841: follows CSE).
        !           842: @item L
        !           843: Dump after loop optimization.
        !           844: @item f
        !           845: Dump after flow analysis.
        !           846: @item c
        !           847: Dump after instruction combination.
        !           848: @item l
        !           849: Dump after local register allocation.
        !           850: @item g
        !           851: Dump after global register allocation.
        !           852: @item m
        !           853: Print statistics on memory usage, at the end of the run.
        !           854: @end table
        !           855: 
        !           856: @item -pedantic
        !           857: Issue all the warnings demanded by strict ANSI standard C; reject
        !           858: all programs that use forbidden extensions.
        !           859: 
        !           860: Valid ANSI standard C programs should compile properly with or without
        !           861: this option (though a rare few will require @samp{-ansi}).  However,
        !           862: without this option, certain GNU extensions and traditional C features
        !           863: are supported as well.  With this option, they are rejected.  There is
        !           864: no reason to @i{use} this option; it exists only to satisfy pedants.
        !           865: @end table
        !           866: 
        !           867: These options control the C preprocessor, which is run on each C source
        !           868: file before actual compilation.  If you use the @samp{-E} option, nothing
        !           869: is done except C preprocessing.  Some of these options make sense only
        !           870: together with @samp{-E} because they request preprocessor output that is
        !           871: not suitable for actual compilation.
        !           872: 
        !           873: @table @samp
        !           874: @item -C
        !           875: Tell the preprocessor not to discard comments.  Used with the
        !           876: @samp{-E} option.
        !           877: 
        !           878: @item -I@var{dir}
        !           879: Search directory @var{dir} for include files.
        !           880: 
        !           881: @item -I-
        !           882: Any directories specified with @samp{-I} options before the @samp{-I-}
        !           883: option are searched only for the case of @samp{#include "@var{file}"};
        !           884: they are not searched for @samp{#include <@var{file}>}.
        !           885: 
        !           886: If additional directories are specified with @samp{-I} options after
        !           887: the @samp{-I-}, these directories are searched for all @samp{#include}
        !           888: directives.  (Ordinarily @emph{all} @samp{-I} directories are used
        !           889: this way.)
        !           890: 
        !           891: In addition, the @samp{-I-} option inhibits the use of the current
        !           892: directory as the first search directory for @samp{#include
        !           893: "@var{file}"}.  Therefore, the current directory is searched only if
        !           894: it is requested explicitly with @samp{-I.}.  Specifying both
        !           895: @samp{-I-} and @samp{-I.} allows you to control precisely which
        !           896: directories are searched before the current one and which are searched
        !           897: after.
        !           898: 
        !           899: @item -nostdinc
        !           900: Do not search the standard system directories for header files.  Only
        !           901: the directories you have specified with @samp{-I} options (and the
        !           902: current directory, if appropriate) are searched.
        !           903: 
        !           904: Between @samp{-nostdinc} and @samp{-I-}, you can eliminate all
        !           905: directories from the search path except those you specify.
        !           906: 
        !           907: @item -M
        !           908: Tell the preprocessor to output a rule suitable for @code{make}
        !           909: describing the dependencies of each source file.  For each source
        !           910: file, the preprocessor outputs one @code{make}-rule whose target is
        !           911: the object file name for that source file and whose dependencies are
        !           912: all the files @samp{#include}d in it.  This rule may be a single line
        !           913: or may be continued with @samp{\}-newline if it is long.
        !           914: 
        !           915: @samp{-M} implies @samp{-E}.
        !           916: 
        !           917: @item -MM
        !           918: Like @samp{-M} but the output mentions only the user-header files
        !           919: included with @samp{#include "@var{file}"}.  System header files
        !           920: included with @samp{#include <@var{file}>} are omitted.
        !           921: 
        !           922: @samp{-MM} implies @samp{-E}.
        !           923: 
        !           924: @item -D@var{macro}
        !           925: Define macro @var{macro} with the empty string as its definition.
        !           926: 
        !           927: @item -D@var{macro}=@var{defn}
        !           928: Define macro @var{macro} as @var{defn}.
        !           929: 
        !           930: @item -U@var{macro}
        !           931: Undefine macro @var{macro}.
        !           932: 
        !           933: @item -T
        !           934: Support ANSI C trigraphs.  You don't want to know about this
        !           935: brain-damage.  The @samp{-ansi} option also has this effect.
        !           936: @end table
        !           937: 
        !           938: @node Installation, Trouble, Options, Top
        !           939: @chapter Installing GNU CC
        !           940: 
        !           941: Here is the procedure for installing GNU CC on a Unix system.
        !           942: @menu
        !           943: * VMS Install::   See below for installation on VMS.
        !           944: @end menu
        !           945: @iftex
        !           946: (See below for VMS.)
        !           947: @end iftex
        !           948: 
        !           949: @enumerate
        !           950: @item
        !           951: Edit @file{Makefile}.  If you are using HPUX, you must make a few
        !           952: changes described in comments at the beginning of the file.
        !           953: 
        !           954: @item
        !           955: Choose configuration files.
        !           956: 
        !           957: @itemize @bullet
        !           958: @item
        !           959: Make a symbolic link named @file{config.h} to the top-level
        !           960: config file for the machine you are using (@pxref{Config}).  This
        !           961: file is responsible for defining information about the host
        !           962: machine.  It includes @file{tm.h}.
        !           963: 
        !           964: The file's name should be @file{config-@var{machine}.h}.  On VMS,
        !           965: use @file{config-vms.h} rather than @file{config-vax.h}.  On the
        !           966: HP 9000 series 300, use @file{config-hp9k3.h} rather than
        !           967: @file{config-m68k.h}.@refill
        !           968: 
        !           969: If your system does not support symbolic links, you might want to
        !           970: set up @file{config.h} to contain a @samp{#include} command which
        !           971: refers to the appropriate file.
        !           972: 
        !           973: @item
        !           974: Make a symbolic link named @file{tm.h} to the machine-description
        !           975: macro file for your machine (its name should be
        !           976: @file{tm-@var{machine}.h}).
        !           977: 
        !           978: For the 68000/68020, do not use @file{tm-m68k.h} directly;
        !           979: instead use one of the files @file{tm-sun3.h}, @file{tm-sun2.h},
        !           980: @file{tm-isi68.h}, @file{tm-news800.h} or @file{tm-3b1.h}.  Each
        !           981: of those files includes @file{tm-m68k.h} but sets up a few things
        !           982: differently as appropriate to the specific model of
        !           983: machine.@refill
        !           984: 
        !           985: There are two files you can use for a 680x0 running HPUX:
        !           986: @file{tm-hp9k320.h} and @file{tm-hp9k320g.h}.  Use the former if
        !           987: you are installing GNU CC alone.  The latter is for another option
        !           988: where GNU CC together with the GNU assembler, linker, debugger
        !           989: and other utilities are used to replace all of HPUX that deals
        !           990: with compilation.  Not all of the pieces of GNU software needed for
        !           991: this mode of operation are as yet in distribution; full instructions
        !           992: will appear here in the future.@refill
        !           993: 
        !           994: For the 32000, use @file{tm-sequent.h} if you are using a Sequent
        !           995: machine, or @file{tm-encore.h} for an Encore machine; otherwise,
        !           996: perhaps @file{tm-ns32k.h} will work for you.
        !           997: 
        !           998: For the vax, use @file{tm-vax.h} on Unix, or @file{tm-vms.h} on
        !           999: VMS.@refill
        !          1000: 
        !          1001: @item
        !          1002: Make a symbolic link named @file{md} to the machine description
        !          1003: pattern file (its name should be @file{@var{machine}.md}).
        !          1004: 
        !          1005: @item
        !          1006: Make a symbolic link named @file{aux-output.c} to the output
        !          1007: subroutine file for your machine (its name should be
        !          1008: @file{output-@var{machine}.c}).
        !          1009: @end itemize
        !          1010: 
        !          1011: @item
        !          1012: Make sure the Bison parser generator is installed.  (This is
        !          1013: unnecessary if the Bison output file @file{parse.tab.c} is more recent
        !          1014: than @file{parse.y} and you do not plan to change @file{parse.y}.)
        !          1015: 
        !          1016: Note that if you have an old version of Bison you may get an error
        !          1017: from the line with the @samp{%expect} directive.  If so, simply remove
        !          1018: that line from @file{parse.y} and proceed.
        !          1019: 
        !          1020: @item
        !          1021: If you are using a Sun, make sure the environment variable
        !          1022: @code{FLOAT_OPTION} is not set.  If this option were set to
        !          1023: @code{f68881} when @file{gnulib} is compiled, the resulting code would
        !          1024: demand to be linked with a special startup file and will not link
        !          1025: properly without special pains.
        !          1026: 
        !          1027: @item
        !          1028: Build the compiler.  Just type @samp{make} in the compiler directory.
        !          1029: 
        !          1030: @item
        !          1031: Move the first-stage object files and executables into a subdirectory
        !          1032: with this command:
        !          1033: 
        !          1034: @example
        !          1035: make stage1
        !          1036: @end example
        !          1037: 
        !          1038: The files are moved into a subdirectory named @file{stage1}.
        !          1039: Once installation is complete, you may wish to delete these files
        !          1040: with @code{rm -r stage1}.
        !          1041: 
        !          1042: @item
        !          1043: Recompile the compiler with itself, with this command:
        !          1044: 
        !          1045: @example
        !          1046: make CC=stage1/gcc CFLAGS="-g -O -Bstage1/"
        !          1047: @end example
        !          1048: 
        !          1049: On a 68000 or 68020 system lacking floating point hardware,
        !          1050: unless you have selected a @file{tm.h} file that expects by default
        !          1051: that there is no such hardware, do this instead:
        !          1052: 
        !          1053: @example
        !          1054: make CC=stage1/gcc CFLAGS="-g -O -Bstage1/ -msoft-float"
        !          1055: @end example
        !          1056: 
        !          1057: @item
        !          1058: If you wish to test the compiler by compiling it with itself one more
        !          1059: time, do this:
        !          1060: 
        !          1061: @example
        !          1062: make stage2
        !          1063: make CC=stage2/gcc CFLAGS="-g -O -Bstage2/"
        !          1064: foreach file (*.o)
        !          1065: cmp $file stage2/$file
        !          1066: end
        !          1067: @end example
        !          1068: 
        !          1069: This will notify you if any of these stage 3 object files differs from
        !          1070: those of stage 2.  Any difference, no matter how innocuous, indicates
        !          1071: that the stage 2 compiler has compiled GNU CC incorrectly, and is
        !          1072: therefore a potentially serious bug which you should investigate and
        !          1073: report (@pxref{Bugs}).
        !          1074: 
        !          1075: @item
        !          1076: Install the compiler driver, the compiler's passes and run-time support.
        !          1077: You can use the following command:
        !          1078: 
        !          1079: @example
        !          1080: make install
        !          1081: @end example
        !          1082: 
        !          1083: @noindent
        !          1084: This copies the files @file{cc1}, @file{cpp} and @file{gnulib} to
        !          1085: files @file{gcc-cc1}, @file{gcc-cpp} and @file{gcc-gnulib} in
        !          1086: directory @file{/usr/local/lib}, which is where the compiler driver
        !          1087: program looks for them.  It also copies the driver program @file{gcc}
        !          1088: into the directory @file{/usr/local}, so that it appears in typical
        !          1089: execution search paths.@refill
        !          1090: 
        !          1091: @strong{Warning: the GNU CPP may not work for @file{ioctl.h},
        !          1092: @file{ttychars.h} and other system header files unless the
        !          1093: @samp{-traditional} option is used.}  The bug is in the header files:
        !          1094: at least on some machines, they rely on behavior that is incompatible
        !          1095: with ANSI C.  This behavior consists of substituting for macro
        !          1096: argument names when they appear inside of character constants.  The
        !          1097: @samp{-traditional} option tells GNU CC to behave the way these
        !          1098: headers expect.
        !          1099: 
        !          1100: Because of this problem, you might prefer to configure GNU CC to use
        !          1101: the system's own C preprocessor.  To do so, make the file
        !          1102: @file{/usr/local/lib/gcc-cpp} a link to @file{/lib/cpp}.
        !          1103: 
        !          1104: Alternatively, on Sun systems and 4.3BSD at least, you can correct the
        !          1105: include files by running the shell script @file{fixincludes}.  This
        !          1106: installs modified, corrected copies of the files @file{ioctl.h} and
        !          1107: @file{ttychars.h} in a special directory where only GNU CC will
        !          1108: normally look for them.
        !          1109: 
        !          1110: The file @file{/usr/include/vaxuba/qvioctl.h} used in the X window
        !          1111: system needs a similar correction.
        !          1112: @end enumerate
        !          1113: 
        !          1114: If you cannot install the compiler's passes and run-time support in
        !          1115: @file{/usr/local/lib}, you can alternatively use the @samp{-B} option to
        !          1116: specify a prefix by which they may be found.  The compiler concatenates
        !          1117: the prefix with the names  @file{cpp}, @file{cc1} and @file{gnulib}.
        !          1118: Thus, you can put the files in a directory @file{/usr/foo/gcc} and
        !          1119: specify @samp{-B/usr/foo/gcc/} when you run GNU CC.
        !          1120: 
        !          1121: @node VMS Install,, Installation, Installation
        !          1122: @section Installing GNU CC on VMS
        !          1123: 
        !          1124: The VMS version of GNU CC is distributed in an unusual tape format which
        !          1125: consists of several tape files.  The first is a command file; the second is
        !          1126: an executable program which reads Unix tar format; the third is another
        !          1127: command file which uses this program to read the remainder of the tape.
        !          1128: 
        !          1129: To load the tape, it suffices to mount it @samp{/foreign} and then do
        !          1130: @samp{@@mta0:} to execute the command file at the beginning of the tape.
        !          1131: 
        !          1132: The tape contains executables and object files as well as sources, so no
        !          1133: compilation is necessary unless you change the sources.  (This is a good
        !          1134: thing, since you probably don't have any other C compiler.)  If you must
        !          1135: recompile, here is how:
        !          1136: 
        !          1137: @enumerate
        !          1138: @item
        !          1139: Copy the file @file{tm-vms.h} to @file{tm.h}, @file{config-vms.h} to
        !          1140: @file{config.h}, @file{vax.md} to @file{md.} and @file{output-vax.c}
        !          1141: to @file{aux-output.c}.@refill
        !          1142: 
        !          1143: @item
        !          1144: Type @samp{@@make} to do recompile everything.
        !          1145: @end enumerate
        !          1146: 
        !          1147: To install the @samp{GCC} command so you can use the compiler easily, in
        !          1148: the same manner as you use the VMS C compiler, you must install the VMS CLD
        !          1149: file for GNU CC as follows:
        !          1150: 
        !          1151: @enumerate
        !          1152: @item
        !          1153: Define the VMS logical names @samp{GNU_CC} and @samp{GNU_CC_INCLUDE}
        !          1154: to point to the directories where the GNU CC executables
        !          1155: (@samp{gcc-cpp}, @samp{gcc-cc1}, etc.) and the C include files are
        !          1156: kept.  This should be done with the commands:@refill
        !          1157: 
        !          1158: @example
        !          1159: $ assign /super /system disk:[gcc] gnu_cc
        !          1160: $ assign /super /system disk:[gcc.include] gnu_cc_include
        !          1161: @end example
        !          1162: 
        !          1163: @noindent
        !          1164: with the appropriate disk and directory names.  These commands can be
        !          1165: placed in your system startup file so they will be executed whenever
        !          1166: the machine is rebooted.
        !          1167: 
        !          1168: @item
        !          1169: Install the @samp{GCC} command with the command line:
        !          1170: 
        !          1171: @example
        !          1172: $ set command /table=sys$library:dcltables gnu_cc:gcc
        !          1173: @end example
        !          1174: 
        !          1175: @noindent
        !          1176: Now you can invoke the compiler with a command like @samp{gcc /verbose
        !          1177: file.c}, which is equivalent to the command @samp{gcc -v -c file.c} in
        !          1178: Unix.
        !          1179: @end enumerate
        !          1180: 
        !          1181: @node Trouble, Incompatibilities, Installation, Top
        !          1182: @chapter Trouble in Installation
        !          1183: 
        !          1184: Here are some of the things that have caused trouble for people installing
        !          1185: GNU CC.
        !          1186: 
        !          1187: @itemize
        !          1188: @item
        !          1189: On certain systems, defining certain environment variables such as
        !          1190: @samp{CC} can interfere with the functioning of @code{make}.
        !          1191: @end itemize
        !          1192: 
        !          1193: @node Incompatibilities, Extensions, Trouble, Top
        !          1194: @chapter Incompatibilities of GNU CC
        !          1195: 
        !          1196: There are several noteworthy incompatibilities between GNU C and most
        !          1197: existing (non-ANSI) versions of C.
        !          1198: 
        !          1199: Ultimately our intention is that the @samp{-traditional} option will
        !          1200: eliminate most of these incompatibilities by telling GNU C to behave
        !          1201: like the other C compilers.
        !          1202: 
        !          1203: @itemize @bullet
        !          1204: @item
        !          1205: GNU CC normally makes string constants read-only.  If several
        !          1206: identical-looking string constants are used, GNU CC stores only one
        !          1207: copy of the string.
        !          1208: 
        !          1209: One consequence is that you cannot call @code{mktemp} with a string
        !          1210: constant argument.  The function @code{mktemp} always alters the
        !          1211: string its argument points to.
        !          1212: 
        !          1213: Another consequence is that @code{sscanf} does not work on some
        !          1214: systems when passed a string constant as its format control string.
        !          1215: This is because @code{sscanf} incorrectly tries to write into the
        !          1216: string constant.
        !          1217: 
        !          1218: The best solution to these problems is to change the program to use
        !          1219: @code{char}-array variables with initialization strings for these
        !          1220: purposes instead of string constants.  But if this is not possible,
        !          1221: you can use the @samp{-fwritable-strings} flag, which directs GNU CC
        !          1222: to handle string constants the same way most C compilers do.
        !          1223: 
        !          1224: @item
        !          1225: GNU CC does not substitute macro arguments when they appear inside of
        !          1226: string constants.  For example, the following macro in GNU CC
        !          1227: 
        !          1228: @example
        !          1229: #define foo(a) "a"
        !          1230: @end example
        !          1231: 
        !          1232: @noindent
        !          1233: will produce output @samp{"a"} regardless of what the argument @var{a} is.
        !          1234: 
        !          1235: The @samp{-traditional} option directs GNU CC to handle such cases
        !          1236: (among others) in the old-fashioned (non-ANSI) fashion.
        !          1237: 
        !          1238: @item
        !          1239: When you use @code{setjmp} and @code{longjmp}, the only automatic
        !          1240: variables guaranteed to remain valid are those declared
        !          1241: @code{volatile}.  This is a consequence of automatic register
        !          1242: allocation.  Consider this function:
        !          1243: 
        !          1244: @example
        !          1245: jmp_buf j;
        !          1246: 
        !          1247: foo ()
        !          1248: @{
        !          1249:   int a, b;
        !          1250: 
        !          1251:   a = fun1 ();
        !          1252:   if (setjmp (j))
        !          1253:     return a;
        !          1254: 
        !          1255:   a = fun2 ();
        !          1256:   /* @r{@code{longjmp (j)} may be occur in @code{fun3}.} */
        !          1257:   return a + fun3 ();
        !          1258: @}
        !          1259: @end example
        !          1260: 
        !          1261: Here @code{a} may or may not be restored to its first value when the
        !          1262: @code{longjmp} occurs.  If @code{a} is allocated in a register, then
        !          1263: its first value is restored; otherwise, it keeps the last value stored
        !          1264: in it.
        !          1265: 
        !          1266: If you use the @samp{-W} option with the @samp{-O} option, you will
        !          1267: get a warning when GNU CC thinks such a problem might be possible.
        !          1268: 
        !          1269: @item
        !          1270: Declarations of external variables and functions within a block apply
        !          1271: only to the block containing the declaration.  In other words, they
        !          1272: have the same scope as any other declaration in the same place.
        !          1273: 
        !          1274: In some other C compilers, a @code{extern} declaration affects all the
        !          1275: rest of the file even if it happens within a block.
        !          1276: 
        !          1277: The @samp{-traditional} option directs GNU C to treat all @code{extern}
        !          1278: declarations as global, like traditional compilers.
        !          1279: 
        !          1280: @item
        !          1281: In traditional C, you can combine @code{long}, etc., with a typedef name,
        !          1282: as shown here:
        !          1283: 
        !          1284: @example
        !          1285: typedef int foo;
        !          1286: typedef long foo bar;
        !          1287: @end example
        !          1288: 
        !          1289: In ANSI C, this is not allowed: @code{long} and other type modifiers
        !          1290: require an explicit @code{int}.  Because this criterion is expressed
        !          1291: by Bison grammar rules rather than C code, the @samp{-traditional}
        !          1292: flag cannot alter it.
        !          1293: 
        !          1294: @item
        !          1295: When compiling functions that return structures or unions, GNU CC
        !          1296: output code uses a method different from that used on most versions of
        !          1297: Unix.  As a result, code compiled with GNU CC cannot call a
        !          1298: structure-returning function compiled with PCC, and vice versa.
        !          1299: 
        !          1300: The method used by GCC is as follows: a structure or union which is 1,
        !          1301: 2, 4 or 8 bytes long is returned like a scalar.  A structure or union
        !          1302: with any other size is stored into an address supplied by the caller
        !          1303: in a special, fixed register.
        !          1304: 
        !          1305: PCC usually handles all sizes of structures and unions by returning
        !          1306: the address of a block of static storage containing the value.  This
        !          1307: method is not used in GCC because it is slower and nonreentrant.
        !          1308: 
        !          1309: On systems where PCC works this way, you may be able to make GCC-compiled
        !          1310: code call such functions that were compiled with PCC by declaring them
        !          1311: to return a pointer to the structure or union instead of the structure
        !          1312: or union itself.  For example, instead of this:
        !          1313: 
        !          1314: @example
        !          1315: struct foo nextfoo ();
        !          1316: @end example
        !          1317: 
        !          1318: @noindent
        !          1319: write this:
        !          1320: 
        !          1321: @example
        !          1322: struct foo *nextfoo ();
        !          1323: #define nextfoo *nextfoo
        !          1324: @end example
        !          1325: 
        !          1326: @noindent
        !          1327: (Note that this assumes you are using the GNU preprocessor, so that
        !          1328: the ANSI antirecursion rules for macro expansions are effective.)
        !          1329: @end itemize
        !          1330: 
        !          1331: @node Extensions, Bugs, Incompatibilities, Top
        !          1332: @chapter GNU Extensions to the C Language
        !          1333: 
        !          1334: GNU C provides several language features not found in ANSI standard C.
        !          1335: (The @samp{-pedantic} option directs GNU CC to print a warning message if
        !          1336: any of these features is used.)  To test for the availability of these
        !          1337: features in conditional compilation, check for a predefined macro
        !          1338: @code{__GNUC__}, which is always defined under GNU CC.
        !          1339: 
        !          1340: @menu
        !          1341: * Statement Exprs::     Putting statements and declarations inside expressions.
        !          1342: * Naming Types::        Giving a name to the type of some expression.
        !          1343: * Typeof::             @code{typeof}: referring to the type of an expression.
        !          1344: * Lvalues::            Using @samp{?:}, @samp{,} and casts in lvalues.
        !          1345: * Conditionals::       Omitting the middle operand of a @samp{?:} expression.
        !          1346: * Zero-Length::                Zero-length arrays.
        !          1347: * Variable-Length::    Arrays whose length is computed at run time.
        !          1348: * Subscripting::       Any array can be subscripted, even if not an lvalue.
        !          1349: * Pointer Arith::      Arithmetic on @code{void}-pointers and function pointers.
        !          1350: * Constructors::       Constructor expressions give structures, unions
        !          1351:                         or arrays as values.
        !          1352: * Dollar Signs::        Dollar sign is allowed in identifiers.
        !          1353: * Alignment::           Inquiring about the alignment of a type or variable.
        !          1354: * Inline::              Defining inline functions (as fast as macros).
        !          1355: * Extended Asm::       Assembler instructions with C expressions as operands.
        !          1356:                         (With them you can define ``built-in'' functions.)
        !          1357: * Asm Labels::         Specifying the assembler name to use for a C symbol.
        !          1358: @end menu
        !          1359: 
        !          1360: @node Statement Exprs, Naming Types, Extensions, Extensions
        !          1361: @section Statements and Declarations inside of Expressions
        !          1362: 
        !          1363: A compound statement in parentheses may appear inside an expression in GNU
        !          1364: C.  This allows you to declare variables within an expression.  For
        !          1365: example:
        !          1366: 
        !          1367: @example
        !          1368: (@{ int y = foo (); int z;
        !          1369:    if (y > 0) z = y;
        !          1370:    else z = - y;
        !          1371:    z; @})
        !          1372: @end example
        !          1373: 
        !          1374: @noindent
        !          1375: is a valid (though slightly more complex than necessary) expression
        !          1376: for the absolute value of @code{foo ()}.
        !          1377: 
        !          1378: This feature is especially useful in making macro definitions ``safe'' (so
        !          1379: that they evaluate each operand exactly once).  For example, the
        !          1380: ``maximum'' function is commonly defined as a macro in standard C as
        !          1381: follows:
        !          1382: 
        !          1383: @example
        !          1384: #define max(a,b) ((a) > (b) ? (a) : (b))
        !          1385: @end example
        !          1386: 
        !          1387: @noindent
        !          1388: But this definition computes either @var{a} or @var{b} twice, with bad
        !          1389: results if the operand has side effects.  In GNU C, if you know the
        !          1390: type of the operands (here let's assume @code{int}), you can define
        !          1391: the macro safely as follows:
        !          1392: 
        !          1393: @example
        !          1394: #define maxint(a,b) \
        !          1395:   (@{int _a = (a), _b = (b); _a > _b ? _a : _b; @})
        !          1396: @end example
        !          1397: 
        !          1398: Embedded statements are not allowed in constant expressions, such as
        !          1399: the value of an enumeration constant, the width of a bit field, or
        !          1400: the initial value of a static variable.
        !          1401: 
        !          1402: If you don't know the type of the operand, you can still do this, but you
        !          1403: must use @code{typeof} (@pxref{Typeof}) or type naming (@pxref{Naming
        !          1404: Types}).
        !          1405: 
        !          1406: @node Naming Types, Typeof, Statement Exprs, Extensions
        !          1407: @section Naming an Expression's Type
        !          1408: 
        !          1409: You can give a name to the type of an expression using a @code{typedef}
        !          1410: declaration with an initializer.  Here is how to define @var{name} as a
        !          1411: type name for the type of @var{exp}:
        !          1412: 
        !          1413: @example
        !          1414: typedef @var{name} = @var{exp};
        !          1415: @end example
        !          1416: 
        !          1417: This is useful in conjunction with the statements-within-expressions
        !          1418: feature.  Here is how the two together can be used to define a safe
        !          1419: ``maximum'' macro that operates on any arithmetic type:
        !          1420: 
        !          1421: @example
        !          1422: #define max(a,b) \
        !          1423:   (@{typedef _ta = (a), _tb = (b);  \
        !          1424:     _ta _a = (a); _tb _b = (b);     \
        !          1425:     _a > _b ? _a : _b; @})
        !          1426: @end example
        !          1427: 
        !          1428: The reason for using names that start with underscores for the local
        !          1429: variables is to avoid conflicts with variable names that occur within the
        !          1430: expressions that are substituted for @code{a} and @code{b}.  Eventually we
        !          1431: hope to design a new form of declaration syntax that allows you to declare
        !          1432: variables whose scopes start only after their initializers; this will be a
        !          1433: more reliable way to prevent such conflicts.
        !          1434: 
        !          1435: @node Typeof, Lvalues, Naming Types, Extensions
        !          1436: @section Referring to a Type with @code{typeof}
        !          1437: 
        !          1438: Another way to refer to the type of an expression is with @code{typeof}.
        !          1439: The syntax of using of this keyword looks like @code{sizeof}, but the
        !          1440: construct acts semantically like a type name defined with @code{typedef}.
        !          1441: 
        !          1442: There are two ways of writing the argument to @code{typeof}: with an
        !          1443: expression or with a type.  Here is an example with an expression:
        !          1444: 
        !          1445: @example
        !          1446: typeof (x[0](1))
        !          1447: @end example
        !          1448: 
        !          1449: @noindent
        !          1450: This assumes that @code{x} is an array of functions; the type described
        !          1451: is that of the values of the functions.
        !          1452: 
        !          1453: Here is an example with a typename as the argument:
        !          1454: 
        !          1455: @example
        !          1456: typeof (int *)
        !          1457: @end example
        !          1458: 
        !          1459: @noindent
        !          1460: Here the type described is that of pointers to @code{int}.
        !          1461: 
        !          1462: A @code{typeof}-construct can be used anywhere a typedef name could be
        !          1463: used.  For example, you can use it in a declaration, in a cast, or inside
        !          1464: of @code{sizeof} or @code{typeof}.
        !          1465: 
        !          1466: @itemize @bullet
        !          1467: @item
        !          1468: This declares @code{y} with the type of what @code{x} points to.
        !          1469: 
        !          1470: @example
        !          1471: typeof (*x) y;
        !          1472: @end example
        !          1473: 
        !          1474: @item
        !          1475: This declares @code{y} as an array of such values.
        !          1476: 
        !          1477: @example
        !          1478: typeof (*x) y[4];
        !          1479: @end example
        !          1480: 
        !          1481: @item
        !          1482: This declares @code{y} as an array of pointers to characters:
        !          1483: 
        !          1484: @example
        !          1485: typeof (typeof (char *)[4]) y;
        !          1486: @end example
        !          1487: 
        !          1488: @noindent
        !          1489: It is equivalent to the following traditional C declaration:
        !          1490: 
        !          1491: @example
        !          1492: char *y[4];
        !          1493: @end example
        !          1494: 
        !          1495: To see the meaning of the declaration using @code{typeof}, and why it
        !          1496: might be a useful way to write, let's rewrite it with these macros:
        !          1497: 
        !          1498: @example
        !          1499: #define pointer(T)  typeof(T *)
        !          1500: #define array(T, N) typeof(T [N])
        !          1501: @end example
        !          1502: 
        !          1503: @noindent
        !          1504: Now the declaration can be rewritten this way:
        !          1505: 
        !          1506: @example
        !          1507: array (pointer (char), 4) y;
        !          1508: @end example
        !          1509: 
        !          1510: @noindent
        !          1511: Thus, @samp{array (pointer (char), 4)} is the type of arrays of 4
        !          1512: pointers to @code{char}.
        !          1513: @end itemize
        !          1514: 
        !          1515: @node Lvalues, Conditionals, Typeof, Extensions
        !          1516: @section Generalized Lvalues
        !          1517: 
        !          1518: Compound expressions, conditional expressions and casts are allowed as
        !          1519: lvalues provided their operands are lvalues.  This means that you can take
        !          1520: their addresses or store values into them.
        !          1521: 
        !          1522: For example, a compound expression can be assigned, provided the last
        !          1523: expression in the sequence is an lvalue.  These two expressions are
        !          1524: equivalent:
        !          1525: 
        !          1526: @example
        !          1527: (a, b) += 5
        !          1528: a, (b += 5)
        !          1529: @end example
        !          1530: 
        !          1531: Similarly, the address of the compound expression can be taken.  These two
        !          1532: expressions are equivalent:
        !          1533: 
        !          1534: @example
        !          1535: &(a, b)
        !          1536: a, &b
        !          1537: @end example
        !          1538: 
        !          1539: A conditional expression is a valid lvalue if its type is not void and the
        !          1540: true and false branches are both valid lvalues.  For example, these two
        !          1541: expressions are equivalent:
        !          1542: 
        !          1543: @example
        !          1544: (a ? b : c) = 5
        !          1545: (a ? b = 5 : (c = 5))
        !          1546: @end example
        !          1547: 
        !          1548: A cast is a valid lvalue if its operand is valid.  Taking the address of
        !          1549: the cast is the same as taking the address without a cast, except for the
        !          1550: type of the result.  For example, these two expressions are equivalent (but
        !          1551: the second may be valid when the type of @samp{a} does not permit a cast to
        !          1552: @samp{int *}).
        !          1553: 
        !          1554: @example
        !          1555: &(int *)a
        !          1556: (int **)&a
        !          1557: @end example
        !          1558: 
        !          1559: A simple assignment whose left-hand side is a cast works by converting the
        !          1560: right-hand side first to the specified type, then to the type of the inner
        !          1561: left-hand side expression.  After this is stored, the value is converter
        !          1562: back to the specified type to become the value of the assignment.  Thus, if
        !          1563: @samp{a} has type @samp{char *}, the following two expressions are
        !          1564: equivalent:
        !          1565: 
        !          1566: @example
        !          1567: (int)a = 5
        !          1568: (int)(a = (char *)5)
        !          1569: @end example
        !          1570: 
        !          1571: An assignment-with-arithmetic operation such as @samp{+=} applied to a cast
        !          1572: performs the arithmetic using the type resulting from the cast, and then
        !          1573: continues as in the previous case.  Therefore, these two expressions are
        !          1574: equivalent:
        !          1575: 
        !          1576: @example
        !          1577: (int)a += 5
        !          1578: (int)(a = (char *) ((int)a + 5))
        !          1579: @end example
        !          1580: 
        !          1581: @node Conditionals, Zero-Length, Lvalues, Extensions
        !          1582: @section Conditional Expressions with Omitted Middle-Operands
        !          1583: 
        !          1584: The middle operand in a conditional expression may be omitted.  Then
        !          1585: if the first operand is nonzero, its value is the value of the conditional
        !          1586: expression.
        !          1587: 
        !          1588: Therefore, the expression
        !          1589: 
        !          1590: @example
        !          1591: x ? : y
        !          1592: @end example
        !          1593: 
        !          1594: @noindent
        !          1595: has the value of @code{x} if that is nonzero; otherwise, the value of
        !          1596: @code{y}.
        !          1597: 
        !          1598: This example is perfectly equivalent to
        !          1599: 
        !          1600: @example
        !          1601: x ? x : y
        !          1602: @end example
        !          1603: 
        !          1604: @noindent
        !          1605: In this simple case, the ability to omit the middle operand is not
        !          1606: especially useful.  When it becomes useful is when the first operand does,
        !          1607: or may (if it is a macro argument), contain a side effect.  Then repeating
        !          1608: the operand in the middle would perform the side effect twice.  Omitting
        !          1609: the middle operand uses the value already computed without the undesirable
        !          1610: effects of recomputing it.
        !          1611: 
        !          1612: @node Zero-Length, Variable-Length, Conditionals, Extensions
        !          1613: @section Arrays of Length Zero
        !          1614: 
        !          1615: Zero-length arrays are allowed in GNU C.  They are very useful as the last
        !          1616: element of a structure which is really a header for a variable-length
        !          1617: object:
        !          1618: 
        !          1619: @example
        !          1620: struct line @{
        !          1621:   int length;
        !          1622:   char contents[0];
        !          1623: @};
        !          1624: 
        !          1625: @{
        !          1626:   struct line *thisline 
        !          1627:     = (struct line *) malloc (sizeof (struct line) + this_length);
        !          1628:   thisline->length = thislength;
        !          1629: @}
        !          1630: @end example
        !          1631: 
        !          1632: In standard C, you would have to give @code{contents} a length of 1, which
        !          1633: means either you waste space or complicate the argument to @code{malloc}.
        !          1634: 
        !          1635: @node Variable-Length, Subscripting, Zero-Length, Extensions
        !          1636: @section Arrays of Variable Length
        !          1637: 
        !          1638: Variable-length automatic arrays are allowed in GNU C.  These arrays are
        !          1639: declared like any other automatic arrays, but with a length that is not a
        !          1640: constant expression.  The storage is allocated at that time and
        !          1641: deallocated when the brace-level is exited.  For example:
        !          1642: 
        !          1643: @example
        !          1644: FILE *concat_fopen (char *s1, char *s2, char *mode)
        !          1645: @{
        !          1646:   char str[strlen (s1) + strlen (s2) + 1];
        !          1647:   strcpy (str, s1);
        !          1648:   strcat (str, s2);
        !          1649:   return fopen (str, mode);
        !          1650: @}
        !          1651: @end example
        !          1652: 
        !          1653: You can also define structure types containing variable-length arrays, and
        !          1654: use them even for arguments or function values, as shown here:
        !          1655: 
        !          1656: @example
        !          1657: int foo;
        !          1658: 
        !          1659: struct entry
        !          1660: @{
        !          1661:   char data[foo];
        !          1662: @};
        !          1663: 
        !          1664: struct entry
        !          1665: tester (struct entry arg)
        !          1666: @{
        !          1667:   struct entry new;
        !          1668:   int i;
        !          1669:   for (i = 0; i < foo; i++)
        !          1670:     new.data[i] = arg.data[i] + 1;
        !          1671:   return new;
        !          1672: @}
        !          1673: @end example
        !          1674: 
        !          1675: @noindent
        !          1676: (Eventually there will be a way to say that the size of the array is
        !          1677: another member of the same structure.)
        !          1678: 
        !          1679: The length of an array is computed on entry to the brace-level where the
        !          1680: array is declared and is remembered for the scope of the array in case you
        !          1681: access it with @code{sizeof}.
        !          1682: 
        !          1683: Jumping or breaking out of the scope of the array name will also deallocate
        !          1684: the storage.  Jumping into the scope is not allowed; you will get an error
        !          1685: message for it.
        !          1686: 
        !          1687: You can use the function @code{alloca} to get an effect much like
        !          1688: variable-length arrays.  The function @code{alloca} is available in
        !          1689: many other C implementations (but not in all).  On the other hand,
        !          1690: variable-length arrays are more elegant.
        !          1691: 
        !          1692: There are other differences between these two methods.  Space allocated
        !          1693: with @code{alloca} exists until the containing @emph{function} returns.
        !          1694: The space for a variable-length array is deallocated as soon as the array
        !          1695: name's scope ends.  (If you use both variable-length arrays and
        !          1696: @code{alloca} in the same function, deallocation of a variable-length array
        !          1697: will also deallocate anything more recently allocated with @code{alloca}.)
        !          1698: 
        !          1699: @node Subscripting, Pointer Arith, Variable-Length, Extensions
        !          1700: @section Non-Lvalue Arrays May Have Subscripts
        !          1701: 
        !          1702: Subscripting is allowed on arrays that are not lvalues, even though the
        !          1703: unary @samp{&} operator is not.  For example, this is valid in GNU C though
        !          1704: not valid in other C dialects:
        !          1705: 
        !          1706: @example
        !          1707: struct foo @{int a[4];@};
        !          1708: 
        !          1709: struct foo f();
        !          1710: 
        !          1711: bar (int index)
        !          1712: @{
        !          1713:   return f().a[index];
        !          1714: @}
        !          1715: @end example
        !          1716: 
        !          1717: @node Pointer Arith, Initializers, Subscripting, Extensions
        !          1718: @section Arithmetic on @code{void}-Pointers and Function Pointers
        !          1719: 
        !          1720: In GNU C, addition and subtraction operations are supported on pointers to
        !          1721: @code{void} and on pointers to functions.  This is done by treating the
        !          1722: size of a @code{void} or of a function as 1.
        !          1723: 
        !          1724: A consequence of this is that @code{sizeof} is also allowed on @code{void}
        !          1725: and on function types, and returns 1.
        !          1726: 
        !          1727: @node Initializers, Constructors, Pointer Arith, Extensions
        !          1728: @section Non-Constant Initializers
        !          1729: 
        !          1730: The elements of an aggregate initializer are not required to be constant
        !          1731: expressions in GNU C.  Here is an example of an initializer with run-time
        !          1732: varying elements:
        !          1733: 
        !          1734: @example
        !          1735: foo (float f, float g)
        !          1736: @{
        !          1737:   float beat_freqs[2] = @{ f-g, f+g @};
        !          1738:   @dots{}
        !          1739: @}
        !          1740: @end example
        !          1741: 
        !          1742: @node Constructors, Dollar Signs, Initializers, Extensions
        !          1743: @section Constructor Expressions
        !          1744: 
        !          1745: GNU C supports constructor expressions.  A constructor looks like a cast
        !          1746: containing an initializer.  Its value is an object of the type specified in
        !          1747: the cast, containing the elements specified in the initializer.  The type
        !          1748: must be a structure, union or array type.
        !          1749: 
        !          1750: Assume that @code{struct foo} and @code{structure} are declared as shown:
        !          1751: 
        !          1752: @example
        !          1753: struct foo @{int a; char b[2];@} structure;
        !          1754: @end example
        !          1755: 
        !          1756: @noindent
        !          1757: Here is an example of constructing a @samp{struct foo} with a constructor:
        !          1758: 
        !          1759: @example
        !          1760: structure = ((struct foo) @{x + y, 'a', 0@});
        !          1761: @end example
        !          1762: 
        !          1763: @noindent
        !          1764: This is equivalent to writing the following:
        !          1765: 
        !          1766: @example
        !          1767: @{
        !          1768:   struct foo temp = @{x + y, 'a', 0@};
        !          1769:   structure = temp;
        !          1770: @}
        !          1771: @end example
        !          1772: 
        !          1773: You can also construct an array.  If all the elements of the constructor
        !          1774: are (made up of) simple constant expressions, suitable for use in
        !          1775: initializers, then the constructor is an lvalue and can be coerced to a
        !          1776: pointer to its first element, as shown here:
        !          1777: 
        !          1778: @example
        !          1779: char **foo = (char *[]) @{ "x", "y", "z" @};
        !          1780: @end example
        !          1781: 
        !          1782: Array constructors whose elements are not simple constants are not very
        !          1783: useful, because the constructor is not an lvalue.  There are only two valid
        !          1784: ways to use it: to subscript it, or initialize an array variable with it.
        !          1785: The former is probably slower than a @code{switch} statement, while the
        !          1786: latter does the same thing an ordinary C initializer would do.
        !          1787: 
        !          1788: @example
        !          1789: output = ((int[]) @{ 2, x, 28 @}) [input];
        !          1790: @end example
        !          1791: 
        !          1792: @node Dollar Signs, Alignment, Constructors, Extensions
        !          1793: @section Dollar Signs in Identifier Names
        !          1794: 
        !          1795: In GNU C, you may use dollar signs in identifier names.  This is because
        !          1796: many traditional C implementations allow such identifiers.
        !          1797: 
        !          1798: @node Alignment, Inline, Dollar Signs, Extensions
        !          1799: @section Inquiring about the Alignment of a Type or Variable
        !          1800: 
        !          1801: The keyword @code{__alignof} allows you to inquire about how an object
        !          1802: is aligned, or the minimum alignment usually required by a type.  Its
        !          1803: syntax is just like @code{sizeof}.
        !          1804: 
        !          1805: For example, if the target machine requires a @code{double} value to be
        !          1806: aligned on an 8-byte boundary, then @code{__alignof (double)} is 8.  This
        !          1807: is true on many RISC machines.  On more traditional machine designs,
        !          1808: @code{__alignof (double)} is 4 or even 2.
        !          1809: 
        !          1810: Some machines never actually require alignment; they allow reference to any
        !          1811: data type even at an odd addresses.  For these machines, @code{__alignof}
        !          1812: reports the @emph{recommended} alignment of a type.
        !          1813: 
        !          1814: When the operand of @code{__alignof} is an lvalue rather than a type, the
        !          1815: value is the largest alignment that the lvalue is known to have.  It may
        !          1816: have this alignment as a result of its data type, or because it is part of
        !          1817: a structure and inherits alignment from that structure. For example, after
        !          1818: this declaration:
        !          1819: 
        !          1820: @example
        !          1821: struct foo @{ int x; char y; @} foo1;
        !          1822: @end example
        !          1823: 
        !          1824: @noindent
        !          1825: the value of @code{__alignof (foo1.y)} is probably 2 or 4, the same as
        !          1826: @code{__alignof (int)}, even though the data type of @code{foo1.y} does not
        !          1827: itself demand any alignment.@refill
        !          1828: 
        !          1829: @node Inline, Extended Asm, Alignment, Extensions
        !          1830: @section An Inline Function is As Fast As a Macro
        !          1831: 
        !          1832: By declaring a function @code{inline}, you can direct GNU CC to integrate
        !          1833: that function's code into the code for its callers.  This makes execution
        !          1834: faster by eliminating the function-call overhead; in addition, if any of
        !          1835: the actual argument values are constant, their known values may permit
        !          1836: simplifications at compile time so that not all of the inline function's
        !          1837: code needs to be included.
        !          1838: 
        !          1839: To declare a function inline, use the @code{inline} keyword in its
        !          1840: declaration, like this:
        !          1841: 
        !          1842: @example
        !          1843: inline int
        !          1844: inc (int *a)
        !          1845: @{
        !          1846:   (*a)++;
        !          1847: @}
        !          1848: @end example
        !          1849: 
        !          1850: You can also make all ``simple enough'' functions inline with the
        !          1851: option @samp{-finline-functions}.  Note that certain usages in a
        !          1852: function definition can make it unsuitable for inline substitution.
        !          1853: 
        !          1854: When a function is both inline and @code{static}, if all calls to the
        !          1855: function are integrated into the caller, then the function's own assembler
        !          1856: code is never referenced.  In this case, GNU CC does not actually output
        !          1857: assembler code for the function, unless you specify the option
        !          1858: @samp{-fkeep-inline-functions}.  Some calls cannot be integrated for
        !          1859: various reasons (in particular, calls that precede the function's
        !          1860: definition cannot be integrated, and neither can recursive calls within the
        !          1861: definition).  If there is a nonintegrated call, then the function is
        !          1862: compiled to assembler code as usual.
        !          1863: 
        !          1864: When an inline function is not @code{static}, then the compiler must assume
        !          1865: that there may be calls from other source files; since a global symbol can
        !          1866: be defined only once in any program, the function must not be defined in
        !          1867: the other source files, so the calls therein cannot be integrated.
        !          1868: Therefore, a non-@code{static} inline function is always compiled on its
        !          1869: own in the usual fashion.
        !          1870: 
        !          1871: @node Extended Asm, Asm Labels, Inline, Extensions
        !          1872: @section Assembler Instructions with C Expression Operands
        !          1873: 
        !          1874: In an assembler instruction using @code{asm}, you can now specify the
        !          1875: operands of the instruction using C expressions.  This means no more
        !          1876: guessing which registers or memory locations will contain the data you want
        !          1877: to use.
        !          1878: 
        !          1879: You must specify an assembler instruction template much like what appears
        !          1880: in a machine description, plus an operand constraint string for each
        !          1881: operand.
        !          1882: 
        !          1883: For example, here is how to use the 68881's @code{fsinx} instruction:
        !          1884: 
        !          1885: @example
        !          1886: asm ("fsinx %1,%0" : "=f" (result) : "f" (angle));
        !          1887: @end example
        !          1888: 
        !          1889: @noindent
        !          1890: Here @code{angle} is the C expression for the input operand while
        !          1891: @code{result} is that of the output operand.  Each has @samp{"f"} as its
        !          1892: operand constraint, saying that a floating-point register is required.  The
        !          1893: constraints use the same language used in the machine description
        !          1894: (@pxref{Constraints}).
        !          1895: 
        !          1896: Each operand is described by an operand-constraint string followed by the C
        !          1897: expression in parentheses.  A colon separates the assembler template from
        !          1898: the first output operand, and another separates the last output operand
        !          1899: from the first input, if any.  Commas separate output operands and separate
        !          1900: inputs.  The number of operands is limited to the maximum number of
        !          1901: operands in any instruction pattern in the machine description.
        !          1902: 
        !          1903: Output operand expressions must be lvalues, and there must be at least one
        !          1904: of them.  The compiler can check this.  The input operands need not be
        !          1905: lvalues, and there need not be any.  The compiler cannot check whether the
        !          1906: operands have data types that are reasonable for the instruction being
        !          1907: executed.
        !          1908: 
        !          1909: The output operands must be write-only; GNU CC will assume that the values
        !          1910: in these operands before the instruction are dead and need not be
        !          1911: generated.  For an operand that is read-write, you must logically split its
        !          1912: function into two separate operands, one input operand and one write-only
        !          1913: output operand.  The connection between them is expressed by constraints
        !          1914: which say they need to be in the same location when the instruction
        !          1915: executes.  You can use the same C expression for both operands, or
        !          1916: different expressions.  For example, here we write the (fictitious)
        !          1917: @samp{combine} instruction with @code{bar} as its read-only source operand
        !          1918: and @code{foo} as its read-write destination:
        !          1919: 
        !          1920: @example
        !          1921: asm ("combine %2,%0" : "=r" (foo) : "0" (foo), "g" (bar));
        !          1922: @end example
        !          1923: 
        !          1924: @noindent
        !          1925: The constraint @samp{"0"} for operand 1 says that it must occupy the same
        !          1926: location as operand 0.  Therefore it is not necessary to substitute operand
        !          1927: 1 into the assembler code output.
        !          1928: 
        !          1929: Usually the most convenient way to use these @code{asm} instructions is to
        !          1930: encapsulate them in macros that look like functions.  For example,
        !          1931: 
        !          1932: @example
        !          1933: #define sin(x)       \
        !          1934: (@{ double __value, __arg = (x);   \
        !          1935:    asm ("fsinx %1,%0": "=f" (__value): "f" (__arg));  \
        !          1936:    __value; @})
        !          1937: @end example
        !          1938: 
        !          1939: @noindent
        !          1940: Here the variable @code{__arg} is used to make sure that the instruction
        !          1941: operates on a proper @code{double} value, and to accept only those
        !          1942: arguments @code{x} which can convert automatically to a @code{double}.
        !          1943: 
        !          1944: Another way to make sure the instruction operates on the correct data type
        !          1945: is to use a cast in the @code{asm}.  This is different from using a
        !          1946: variable @code{__arg} in that it converts more different types.  For
        !          1947: example, if the desired type were @code{int}, casting the argument to
        !          1948: @code{int} would accept a pointer with no complaint, while assigning the
        !          1949: argument to an @code{int} variable named @code{__arg} would warn about
        !          1950: using a pointer unless the caller explicitly casts it.
        !          1951: 
        !          1952: GNU CC assumes for optimization purposes that these instructions have no
        !          1953: side effects except to change the output operands.  This does not mean that
        !          1954: instructions with a side effect cannot be used, but you must be careful,
        !          1955: because the compiler may eliminate them if the output operands aren't used,
        !          1956: or move them out of loops, or replace two with one if they constitute a
        !          1957: common subexpression.  Also, if your instruction does have a side effect on
        !          1958: a variable that otherwise appears not to change, the old value of the
        !          1959: variable may be reused later if it happens to be found in a register.
        !          1960: 
        !          1961: You can prevent an @code{asm} instruction from being deleted, moved or
        !          1962: combined by writing the keyword @code{volatile} after the @code{asm}.  For
        !          1963: example:
        !          1964: 
        !          1965: @example
        !          1966: #define set_priority(x)  \
        !          1967: asm volatile ("set_priority %1":    \
        !          1968:               "=m" (*(char *)0): "g" (x))
        !          1969: @end example
        !          1970: 
        !          1971: @noindent
        !          1972: Note that we have supplied an output operand which is not actually used in
        !          1973: the instruction.  This is because @code{asm} requires at least one output
        !          1974: operand.  This requirement exists for internal implementation reasons and
        !          1975: we might be able to relax it in the future.
        !          1976: 
        !          1977: In this case output operand has the additional benefit effect of giving the
        !          1978: appearance of writing in memory.  As a result, GNU CC will assume that data
        !          1979: previously fetched from memory must be fetched again if needed again later.
        !          1980: This may be desirable if you have not employed the @code{volatile} keyword
        !          1981: on all the variable declarations that ought to have it.
        !          1982: 
        !          1983: @node Asm Labels,,Extended Asm, Extensions
        !          1984: @section Controlling Names Used in Assembler Code
        !          1985: 
        !          1986: You can specify the name to be used in the assembler code for a C function
        !          1987: or variable by writing the @code{asm} keyword after the declarator as
        !          1988: follows:
        !          1989: 
        !          1990: @example
        !          1991: int foo asm ("myfoo") = 2;
        !          1992: @end example
        !          1993: 
        !          1994: @noindent
        !          1995: This specifies that the name to be used for the variable @code{foo} in
        !          1996: the assembler code should be @samp{myfoo} rather than the usual
        !          1997: @samp{_foo}.
        !          1998: 
        !          1999: On systems where an underscore is normally prepended to the name of a C
        !          2000: function or variable, this feature allows you to define names for the
        !          2001: linker that do not start with an underscore.
        !          2002: 
        !          2003: You cannot use @code{asm} in this way in a function @emph{definition}; but
        !          2004: you can get the same effect by writing a declaration for the function
        !          2005: before its definition and putting @code{asm} there, like this:
        !          2006: 
        !          2007: @example
        !          2008: extern func () asm ("FUNC");
        !          2009: 
        !          2010: func (x, y)
        !          2011:      int x, y;
        !          2012: @dots{}
        !          2013: @end example
        !          2014: 
        !          2015: It is up to you to make sure that the assembler names you choose do not
        !          2016: conflict with any other assembler symbols.  Also, you must not use a
        !          2017: register name; that would produce completely invalid assembler code.  GNU
        !          2018: CC does not as yet have the ability to store static variables in registers.
        !          2019: Perhaps that will be added.
        !          2020: 
        !          2021: @node Bugs, Portability, Extensions, Top
        !          2022: @chapter Reporting Bugs
        !          2023: 
        !          2024: Your bug reports play an essential role in making GNU CC reliable.
        !          2025: 
        !          2026: Reporting a bug may help you by bringing a solution to your problem, or it
        !          2027: may not.  But in any case the important function of a bug report is to help
        !          2028: the entire community by making the next version of GNU CC work better.  Bug
        !          2029: reports are your contribution to the maintenance of GNU CC.
        !          2030: 
        !          2031: In order for a bug report to serve its purpose, you must include the
        !          2032: information that makes for fixing the bug.
        !          2033: 
        !          2034: @menu
        !          2035: * Criteria:  Bug Criteria.   Have you really found a bug?
        !          2036: * Reporting: Bug Reporting.  How to report a bug effectively.
        !          2037: @end menu
        !          2038: 
        !          2039: @node Bug Criteria, Bug Reporting, Bugs, Bugs
        !          2040: @section Have You Found a Bug?
        !          2041: 
        !          2042: If you are not sure whether you have found a bug, here are some guidelines:
        !          2043: 
        !          2044: @itemize @bullet
        !          2045: @item
        !          2046: If the compiler gets a fatal signal, for any input whatever, that is a
        !          2047: compiler bug.  Reliable compilers never crash.
        !          2048: 
        !          2049: @item
        !          2050: If the compiler produces invalid assembly code, for any input whatever
        !          2051: (except an @code{asm} statement), that is a compiler bug, unless the
        !          2052: compiler reports errors (not just warnings) which would ordinarily
        !          2053: prevent the assembler from being run.
        !          2054: 
        !          2055: @item
        !          2056: If the compiler produces valid assembly code that does not correctly
        !          2057: execute the input source code, that is a compiler bug.
        !          2058: 
        !          2059: However, you must double-check to make sure, because you may have run
        !          2060: into an incompatibility between GNU C and traditional C
        !          2061: (@pxref{Incompatibilities}).  These incompatibilities might be considered
        !          2062: bugs, but they are inescapable consequences of valuable features.
        !          2063: 
        !          2064: Or you may have a program whose behavior is undefined, which happened
        !          2065: by chance to give the desired results with another C compiler.
        !          2066: 
        !          2067: For example, in many nonoptimizing compilers, you can write @samp{x;}
        !          2068: at the end of a function instead of @samp{return x;}, with the same
        !          2069: results.  But the value of the function is undefined if @samp{return}
        !          2070: is omitted; it is not a bug when GNU CC produces different results.
        !          2071: 
        !          2072: Problems often result from expressions with two increment operators,
        !          2073: as in @samp{f (*p++, *p++)}.  Your previous compiler might have
        !          2074: interpreted that expression the way you intended; GNU CC might
        !          2075: interpret it another way; neither compiler is wrong.
        !          2076: 
        !          2077: After you have localized the error to a single source line, it should
        !          2078: be easy to check for these things.  If your program is correct and
        !          2079: well defined, you have found a compiler bug.
        !          2080: 
        !          2081: @item
        !          2082: If the compiler produces an error message for valid input, that is a
        !          2083: compiler bug.
        !          2084: 
        !          2085: Note that the following is not valid input, and the error message for
        !          2086: it is not a bug:
        !          2087: 
        !          2088: @example
        !          2089: int foo (char);
        !          2090: 
        !          2091: int
        !          2092: foo (x)
        !          2093:      char x;
        !          2094: @{ @dots{} @}
        !          2095: @end example
        !          2096: 
        !          2097: @noindent
        !          2098: The prototype says to pass a @code{char}, while the definition says to
        !          2099: pass an @code{int} and treat the value as a @code{char}.  This is what
        !          2100: the ANSI standard says, and it makes sense.
        !          2101: 
        !          2102: @item
        !          2103: If the compiler does not produce an error message for invalid input,
        !          2104: that is a compiler bug.  However, you should note that your idea of
        !          2105: ``invalid input'' might be my idea of ``an extension'' or ``support
        !          2106: for traditional practice''.
        !          2107: 
        !          2108: @item
        !          2109: If you are an experienced user of C compilers, your suggestions
        !          2110: for improvement of GNU CC are welcome in any case.
        !          2111: @end itemize
        !          2112: 
        !          2113: @node Bug Reporting,, Bug Criteria, Bugs
        !          2114: @section How to Report Bugs
        !          2115: 
        !          2116: Send bug reports for GNU C to one of these addresses:
        !          2117: 
        !          2118: @example
        !          2119: bug-gcc@@prep.ai.mit.edu
        !          2120: @{ucbvax|mit-eddie|uunet@}!prep.ai.mit.edu!bug-gcc
        !          2121: @end example
        !          2122: 
        !          2123: As a last resort, snail them to:
        !          2124: 
        !          2125: @example
        !          2126: GNU Compiler Bugs
        !          2127: 545 Tech Sq
        !          2128: Cambridge, MA 02139
        !          2129: @end example
        !          2130: 
        !          2131: The fundamental principle of reporting bugs usefully is this:
        !          2132: @strong{report all the facts}.  If you are not sure whether to mention a
        !          2133: fact or leave it out, mention it!
        !          2134: 
        !          2135: Often people omit facts because they think they know what causes the
        !          2136: problem and they conclude that some details don't matter.  Thus, you might
        !          2137: assume that the name of the variable you use in an example does not matter.
        !          2138: Well, probably it doesn't, but one cannot be sure.  Perhaps the bug is a
        !          2139: stray memory reference which happens to fetch from the location where that
        !          2140: name is stored in memory; perhaps, if the name were different, the contents
        !          2141: of that location would fool the compiler into doing the right thing despite
        !          2142: the bug.  Play it safe and give an exact example.
        !          2143: 
        !          2144: If you want to enable me to fix the bug, you should include all these
        !          2145: things:
        !          2146: 
        !          2147: @itemize @bullet
        !          2148: @item
        !          2149: The version of GNU CC.  You can get this by running it with the
        !          2150: @samp{-v} option.
        !          2151: 
        !          2152: Without this, I won't know whether there is any point in looking for
        !          2153: the bug in the current version of GNU CC.
        !          2154: 
        !          2155: @item
        !          2156: A complete input file that will reproduce the bug.  If the bug is in
        !          2157: the C preprocessor, send me a source file and any header files that it
        !          2158: requires.  If the bug is in the compiler proper (@file{cc1}), run your
        !          2159: source file through the C preprocessor by doing @samp{gcc -E
        !          2160: @var{sourcefile} > @var{outfile}}, then include the contents of
        !          2161: @var{outfile} in the bug report.  (Any @samp{-I}, @samp{-D} or
        !          2162: @samp{-U} options that you used in actual compilation should also be
        !          2163: used when doing this.)
        !          2164: 
        !          2165: A single statement is not enough of an example.  In order to compile
        !          2166: it, it must be embedded in a function definition; and the bug might
        !          2167: depend on the details of how this is done.
        !          2168: 
        !          2169: Without a real example I can compile, all I can do about your bug
        !          2170: report is wish you luck.  It would be futile to try to guess how to
        !          2171: provoke the bug.  For example, bugs in register allocation and
        !          2172: reloading frequently depend on every little detail of the function
        !          2173: they happen in.
        !          2174: 
        !          2175: @item
        !          2176: The command arguments you gave GNU CC to compile that example and
        !          2177: observe the bug.  For example, did you use @samp{-O}?  To guarantee
        !          2178: you won't omit something important, list them all.
        !          2179: 
        !          2180: If I were to try to guess the arguments, I would probably guess wrong
        !          2181: and then I would not encounter the bug.
        !          2182: 
        !          2183: @item
        !          2184: The names of the files that you used for @file{tm.h} and @file{md}
        !          2185: when you installed the compiler.
        !          2186: 
        !          2187: @item
        !          2188: The type of machine you are using, and the operating system name and
        !          2189: version number.
        !          2190: 
        !          2191: @item
        !          2192: A description of what behavior you observe that you believe is
        !          2193: incorrect.  For example, ``It gets a fatal signal,'' or, ``There is an
        !          2194: incorrect assembler instruction in the output.''
        !          2195: 
        !          2196: Of course, if the bug is that the compiler gets a fatal signal, then I
        !          2197: will certainly notice it.  But if the bug is incorrect output, I might
        !          2198: not notice unless it is glaringly wrong.  I won't study all the
        !          2199: assembler code from a 50-line C program just on the off chance that it
        !          2200: might be wrong.
        !          2201: 
        !          2202: Even if the problem you experience is a fatal signal, you should still
        !          2203: say so explicitly.  Suppose something strange is going on, such as,
        !          2204: your copy of the compiler is out of synch, or you have encountered a
        !          2205: bug in the C library on your system.  (This has happened!)  Your copy
        !          2206: might crash and mine would not.  If you @i{told} me to expect a crash,
        !          2207: then when mine fails to crash, I would know that the bug was not
        !          2208: happening for me.  If you had not told me to expect a crash, then I
        !          2209: would not be able to draw any conclusion from my observations.
        !          2210: 
        !          2211: In cases where GNU CC generates incorrect code, if you send me a small
        !          2212: complete sample program I will find the error myself by running the
        !          2213: program under a debugger.  If you send me a large example or a part of
        !          2214: a larger program, I cannot do this; you must debug the compiled
        !          2215: program and narrow the problem down to one source line.  Tell me which
        !          2216: source line it is, and what you believe is incorrect about the code
        !          2217: generated for that line.
        !          2218: 
        !          2219: @item
        !          2220: If you send me examples of output from GNU CC, please use @samp{-g}
        !          2221: when you make them.  The debugging information includes source line
        !          2222: numbers which are essential for correlating the output with the input.
        !          2223: @end itemize
        !          2224: 
        !          2225: Here are some things that are not necessary:
        !          2226: 
        !          2227: @itemize @bullet
        !          2228: @item
        !          2229: A description of the envelope of the bug.
        !          2230: 
        !          2231: Often people who encounter a bug spend a lot of time investigating
        !          2232: which changes to the input file will make the bug go away and which
        !          2233: changes will not affect it.
        !          2234: 
        !          2235: This is often time consuming and not very useful, because the way I
        !          2236: will find the bug is by running a single example under the debugger
        !          2237: with breakpoints, not by pure deduction from a series of examples.
        !          2238: 
        !          2239: Of course, it can't hurt if you can find a simpler example that
        !          2240: triggers the same bug.  Errors in the output will be easier to spot,
        !          2241: running under the debugger will take less time, etc.  An easy way
        !          2242: to simplify an example is to delete all the function definitions
        !          2243: except the one where the bug occurs.  Those earlier in the file
        !          2244: may be replaced by external declarations.
        !          2245: 
        !          2246: However, simplification is not necessary; if you don't want to do
        !          2247: this, report the bug anyway.
        !          2248: 
        !          2249: @item
        !          2250: A patch for the bug.
        !          2251: 
        !          2252: A patch for the bug does help me if it is a good one.  But don't omit
        !          2253: the necessary information, such as the test case, because I might see
        !          2254: problems with your patch and decide to fix the problem another way.
        !          2255: 
        !          2256: Sometimes with a program as complicated as GNU CC it is very hard to
        !          2257: construct an example that will make the program go through a certain
        !          2258: point in the code.  If you don't send me the example, I won't be able
        !          2259: to verify that the bug is fixed.
        !          2260: 
        !          2261: @item
        !          2262: A guess about what the bug is or what it depends on.
        !          2263: 
        !          2264: Such guesses are usually wrong.  Even I can't guess right about such
        !          2265: things without using the debugger to find the facts.  They also don't
        !          2266: serve a useful purpose.
        !          2267: @end itemize
        !          2268: 
        !          2269: @node Portability, Interface, Bugs, Top
        !          2270: @chapter GNU CC and Portability
        !          2271: 
        !          2272: The main goal of GNU CC was to make a good, fast compiler for machines in
        !          2273: the class that the GNU system aims to run on: 32-bit machines that address
        !          2274: 8-bit bytes and have several general registers.  Elegance, theoretical
        !          2275: power and simplicity are only secondary.
        !          2276: 
        !          2277: GNU CC gets most of the information about the target machine from a machine
        !          2278: description which gives an algebraic formula for each of the machine's
        !          2279: instructions.  This is a very clean way to describe the target.  But when
        !          2280: the compiler needs information that is difficult to express in this
        !          2281: fashion, I have not hesitated to define an ad-hoc parameter to the machine
        !          2282: description.  The purpose of portability is to reduce the total work needed
        !          2283: on the compiler; it was not of interest for its own sake.
        !          2284: 
        !          2285: GNU CC does not contain machine dependent code, but it does contain code
        !          2286: that depends on machine parameters such as endianness (whether the most
        !          2287: significant byte has the highest or lowest address of the bytes in a word)
        !          2288: and the availability of autoincrement addressing.  In the RTL-generation
        !          2289: pass, it is often necessary to have multiple strategies for generating code
        !          2290: for a particular kind of syntax tree, strategies that are usable for different
        !          2291: combinations of parameters.  Often I have not tried to address all possible
        !          2292: cases, but only the common ones or only the ones that I have encountered.
        !          2293: As a result, a new target may require additional strategies.  You will know
        !          2294: if this happens because the compiler will call @code{abort}.  Fortunately,
        !          2295: the new strategies can be added in a machine-independent fashion, and will
        !          2296: affect only the target machines that need them.
        !          2297: 
        !          2298: @node Interface, Passes, Portability, Top
        !          2299: @chapter Interfacing to GNU CC Output
        !          2300: 
        !          2301: GNU CC is normally configured to use the same function calling convention
        !          2302: normally in use on the target system.  This is done with the
        !          2303: machine-description macros described (@pxref{Machine Macros}).
        !          2304: 
        !          2305: However, returning of structure and union values is done differently.
        !          2306: As a result, functions compiled with PCC returning such types cannot
        !          2307: be called from code compiled with GNU CC, and vice versa.  This usually
        !          2308: does not cause trouble because the Unix library routines don't return
        !          2309: structures and unions.
        !          2310: 
        !          2311: Structures and unions that are 1, 2, 4 or 8 bytes long are returned in the
        !          2312: same registers used for @code{int} or @code{double} return values.  (GNU CC
        !          2313: typically allocates variables of such types in registers also.)  Structures
        !          2314: and unions of other sizes are returned by storing them into an address
        !          2315: passed by the caller in a register.  This method is faster than the one
        !          2316: normally used by PCC and is also reentrant.  The register used for passing
        !          2317: the address is specified by the machine-description macro
        !          2318: @code{STRUCT_VALUE_REGNUM}.
        !          2319: 
        !          2320: GNU CC always passes arguments on the stack.  At some point it will be
        !          2321: extended to pass arguments in registers, for machines which use that as
        !          2322: the standard calling convention.  This will make it possible to use such
        !          2323: a convention on other machines as well.  However, that would render it
        !          2324: completely incompatible with PCC.  We will probably do this once we
        !          2325: have a complete GNU system so we can compile the libraries with GNU CC.
        !          2326: 
        !          2327: If you use @code{longjmp}, beware of automatic variables.  ANSI C says that
        !          2328: automatic variables that are not declared @code{volatile} have undefined
        !          2329: values after a @code{longjmp}.  And this is all GNU CC promises to do,
        !          2330: because it is very difficult to restore register variables correctly, and
        !          2331: one of GNU CC's features is that it can put variables in registers without
        !          2332: your asking it to.
        !          2333: 
        !          2334: If you want a variable to be unaltered by @code{longjmp}, and you don't
        !          2335: want to write @code{volatile} because old C compilers don't accept it,
        !          2336: just take the address of the variable.  If a variable's address is ever
        !          2337: taken, even if just to compute it and ignore it, then the variable cannot
        !          2338: go in a register:
        !          2339: 
        !          2340: @example
        !          2341: @{
        !          2342:   int careful;
        !          2343:   &careful;
        !          2344:   @dots{}
        !          2345: @}
        !          2346: @end example
        !          2347: 
        !          2348: Code compiled with GNU CC may call certain library routines.  The routines
        !          2349: needed on the Vax and 68000 are in the file @file{gnulib.c}.  You must
        !          2350: compile this file with the standard C compiler, not with GNU CC, and then
        !          2351: link it with each program you compile with GNU CC.  (In actuality, many
        !          2352: programs will not need it.)  The usual function call interface is used
        !          2353: for calling the library routines.  Some standard parts of the C library,
        !          2354: such as @code{bcopy}, are also called automatically.
        !          2355: 
        !          2356: @node Passes, RTL, Interface, Top
        !          2357: @chapter Passes and Files of the Compiler
        !          2358: 
        !          2359: The overall control structure of the compiler is in @file{toplev.c}.  This
        !          2360: file is responsible for initialization, decoding arguments, opening and
        !          2361: closing files, and sequencing the passes.
        !          2362: 
        !          2363: The parsing pass is invoked only once, to parse the entire input.  The RTL
        !          2364: intermediate code for a function is generated as the function is parsed, a
        !          2365: statement at a time.  Each statement is read in as a syntax tree and then
        !          2366: converted to RTL; then the storage for the tree for the statement is
        !          2367: reclaimed.  Storage for types (and the expressions for their sizes),
        !          2368: declarations, and a representation of the binding contours and how they nest,
        !          2369: remains until the function is finished being compiled; these are all needed
        !          2370: to output the debugging information.
        !          2371: 
        !          2372: Each time the parsing pass reads a complete function definition or
        !          2373: top-level declaration, it calls the function
        !          2374: @code{rest_of_compilation} or @code{rest_of_decl_compilation} in
        !          2375: @file{toplev.c}, which are responsible for all further processing
        !          2376: necessary, ending with output of the assembler language.  All other
        !          2377: compiler passes run, in sequence, within @code{rest_of_compilation}.
        !          2378: When that function returns from compiling a function definition, the
        !          2379: storage used for that function definition's compilation is entirely
        !          2380: freed, unless it is an inline function (@pxref{Inline}).
        !          2381: 
        !          2382: Here is a list of all the passes of the compiler and their source files.
        !          2383: Also included is a description of where debugging dumps can be requested
        !          2384: with @samp{-d} options.
        !          2385: 
        !          2386: @itemize @bullet
        !          2387: @item
        !          2388: Parsing.  This pass reads the entire text of a function definition,
        !          2389: constructing partial syntax trees.  This and RTL generation are no longer
        !          2390: truly separate passes (formerly they were), but it is easier to think
        !          2391: of them as separate.
        !          2392: 
        !          2393: The tree representation does not entirely follow C syntax, because it is
        !          2394: intended to support other languages as well.
        !          2395: 
        !          2396: C data type analysis is also done in this pass, and every tree node
        !          2397: that represents an expression has a data type attached.  Variables are
        !          2398: represented as declaration nodes.
        !          2399: 
        !          2400: Constant folding and associative-law simplifications are also done
        !          2401: during this pass.
        !          2402: 
        !          2403: The source files for parsing are @file{parse.y}, @file{decl.c},
        !          2404: @file{typecheck.c}, @file{stor-layout.c}, @file{fold-const.c}, and
        !          2405: @file{tree.c}.  The last three are intended to be language-independent.
        !          2406: There are also header files @file{parse.h}, @file{c-tree.h},
        !          2407: @file{tree.h} and @file{tree.def}.  The last two define the format of
        !          2408: the tree representation.@refill
        !          2409: 
        !          2410: @item
        !          2411: RTL generation.  This is the conversion of syntax tree into RTL code.
        !          2412: It is actually done statement-by-statement during parsing, but for
        !          2413: most purposes it can be thought of as a separate pass.
        !          2414: 
        !          2415: This is where the bulk of target-parameter-dependent code is found,
        !          2416: since often it is necessary for strategies to apply only when certain
        !          2417: standard kinds of instructions are available.  The purpose of named
        !          2418: instruction patterns is to provide this information to the RTL
        !          2419: generation pass.
        !          2420: 
        !          2421: Optimization is done in this pass for @code{if}-conditions that are
        !          2422: comparisons, boolean operations or conditional expressions.  Tail
        !          2423: recursion is detected at this time also.  Decisions are made about how
        !          2424: best to arrange loops and how to output @code{switch} statements.
        !          2425: 
        !          2426: The source files for RTL generation are @file{stmt.c}, @file{expr.c},
        !          2427: @file{explow.c}, @file{expmed.c}, @file{optabs.c} and @file{emit-rtl.c}.
        !          2428: Also, the file @file{insn-emit.c}, generated from the machine description
        !          2429: by the program @code{genemit}, is used in this pass.  The header files
        !          2430: @file{expr.h} is used for communication within this pass.@refill
        !          2431: 
        !          2432: The header files @file{insn-flags.h} and @file{insn-codes.h},
        !          2433: generated from the machine description by the programs @code{genflags}
        !          2434: and @code{gencodes}, tell this pass which standard names are available
        !          2435: for use and which patterns correspond to them.@refill
        !          2436: 
        !          2437: Aside from debugging information output, none of the following passes
        !          2438: refers to the tree structure representation of the function (only
        !          2439: part of which is saved).
        !          2440: 
        !          2441: The decision of whether the function can and should be expanded inline
        !          2442: in its subsequent callers is made at the end of rtl generation.  The
        !          2443: function must meet certain criteria, currently related to the size of
        !          2444: the function and the types and number of parameters it has.  Note that
        !          2445: this function may contain loops, recursive calls to itself
        !          2446: (tail-recursive functions can be inlined!), gotos, in short, all
        !          2447: constructs supported by GNU CC.
        !          2448: 
        !          2449: The option @samp{-dr} causes a debugging dump of the RTL code after
        !          2450: this pass.  This dump file's name is made by appending @samp{.rtl} to
        !          2451: the input file name.
        !          2452: 
        !          2453: @item
        !          2454: Jump optimization.  This pass simplifies jumps to the following
        !          2455: instruction, jumps across jumps, and jumps to jumps.  It deletes
        !          2456: unreferenced labels and unreachable code, except that unreachable code
        !          2457: that contains a loop is not recognized as unreachable in this pass.
        !          2458: (Such loops are deleted later in the basic block analysis.)
        !          2459: 
        !          2460: Jump optimization is performed two or three times.  The first time is
        !          2461: immediately following RTL generation.  The second time is after CSE,
        !          2462: but only if CSE says repeated jump optimization is needed.  The
        !          2463: last time is right before the final pass.  That time, cross-jumping
        !          2464: and deletion of no-op move instructions are done together with the
        !          2465: optimizations described above.
        !          2466: 
        !          2467: The source file of this pass is @file{jump.c}.
        !          2468: 
        !          2469: The option @samp{-dj} causes a debugging dump of the RTL code after
        !          2470: this pass is run for the first time.  This dump file's name is made by
        !          2471: appending @samp{.jump} to the input file name.
        !          2472: 
        !          2473: @item
        !          2474: Register scan.  This pass finds the first and last use of each
        !          2475: register, as a guide for common subexpression elimination.  Its source
        !          2476: is in @file{regclass.c}.
        !          2477: 
        !          2478: @item
        !          2479: Common subexpression elimination.  This pass also does constant
        !          2480: propagation.  Its source file is @file{cse.c}.  If constant
        !          2481: propagation causes conditional jumps to become unconditional or to
        !          2482: become no-ops, jump optimization is run again when CSE is finished.
        !          2483: 
        !          2484: The option @samp{-ds} causes a debugging dump of the RTL code after
        !          2485: this pass.  This dump file's name is made by appending @samp{.cse} to
        !          2486: the input file name.
        !          2487: 
        !          2488: @item
        !          2489: Loop optimization.  This pass moves constant expressions out of loops.
        !          2490: Its source file is @file{loop.c}.
        !          2491: 
        !          2492: The option @samp{-dL} causes a debugging dump of the RTL code after
        !          2493: this pass.  This dump file's name is made by appending @samp{.loop} to
        !          2494: the input file name.
        !          2495: 
        !          2496: @item
        !          2497: Stupid register allocation is performed at this point in a
        !          2498: nonoptimizing compilation.  It does a little data flow analysis as
        !          2499: well.  When stupid register allocation is in use, the next pass
        !          2500: executed is the reloading pass; the others in between are skipped.
        !          2501: The source file is @file{stupid.c}.
        !          2502: 
        !          2503: @item
        !          2504: Data flow analysis (@file{flow.c}).  This pass divides the program
        !          2505: into basic blocks (and in the process deletes unreachable loops); then
        !          2506: it computes which pseudo-registers are live at each point in the
        !          2507: program, and makes the first instruction that uses a value point at
        !          2508: the instruction that computed the value.
        !          2509: 
        !          2510: This pass also deletes computations whose results are never used, and
        !          2511: combines memory references with add or subtract instructions to make
        !          2512: autoincrement or autodecrement addressing.
        !          2513: 
        !          2514: The option @samp{-df} causes a debugging dump of the RTL code after
        !          2515: this pass.  This dump file's name is made by appending @samp{.flow} to
        !          2516: the input file name.  If stupid register allocation is in use, this
        !          2517: dump file reflects the full results of such allocation.
        !          2518: 
        !          2519: @item
        !          2520: Instruction combination (@file{combine.c}).  This pass attempts to
        !          2521: combine groups of two or three instructions that are related by data
        !          2522: flow into single instructions.  It combines the RTL expressions for
        !          2523: the instructions by substitution, simplifies the result using algebra,
        !          2524: and then attempts to match the result against the machine description.
        !          2525: 
        !          2526: The option @samp{-dc} causes a debugging dump of the RTL code after
        !          2527: this pass.  This dump file's name is made by appending @samp{.combine}
        !          2528: to the input file name.
        !          2529: 
        !          2530: @item
        !          2531: Register class preferencing.  The RTL code is scanned to find out
        !          2532: which register class is best for each pseudo register.  The source
        !          2533: file is @file{regclass.c}.
        !          2534: 
        !          2535: @item
        !          2536: Local register allocation (@file{local-alloc.c}).  This pass allocates
        !          2537: hard registers to pseudo registers that are used only within one basic
        !          2538: block.  Because the basic block is linear, it can use fast and
        !          2539: powerful techniques to do a very good job.
        !          2540: 
        !          2541: The option @samp{-dl} causes a debugging dump of the RTL code after
        !          2542: this pass.  This dump file's name is made by appending @samp{.lreg} to
        !          2543: the input file name.
        !          2544: 
        !          2545: @item
        !          2546: Global register allocation (@file{global-alloc.c}).  This pass
        !          2547: allocates hard registers for the remaining pseudo registers (those
        !          2548: whose life spans are not contained in one basic block).
        !          2549: 
        !          2550: @item
        !          2551: Reloading.  This pass renumbers pseudo registers with the hardware
        !          2552: registers numbers they were allocated.  Pseudo registers that did not
        !          2553: get hard registers are replaced with stack slots.  Then it finds
        !          2554: instructions that are invalid because a value has failed to end up in
        !          2555: a register, or has ended up in a register of the wrong kind.  It fixes
        !          2556: up these instructions by reloading the problematical values
        !          2557: temporarily into registers.  Additional instructions are generated to
        !          2558: do the copying.
        !          2559: 
        !          2560: Source files are @file{reload.c} and @file{reload1.c}, plus the header
        !          2561: @file{reload.h} used for communication between them.
        !          2562: 
        !          2563: The option @samp{-dg} causes a debugging dump of the RTL code after
        !          2564: this pass.  This dump file's name is made by appending @samp{.greg} to
        !          2565: the input file name.
        !          2566: 
        !          2567: @item
        !          2568: Jump optimization is repeated, this time including cross-jumping
        !          2569: and deletion of no-op move instructions.  Machine-specific peephole
        !          2570: optimizations are performed at the same time.
        !          2571: 
        !          2572: The option @samp{-dJ} causes a debugging dump of the RTL code after
        !          2573: this pass.  This dump file's name is made by appending @samp{.jump2}
        !          2574: to the input file name.
        !          2575: 
        !          2576: @item
        !          2577: Final.  This pass outputs the assembler code for the function.  It is
        !          2578: also responsible for identifying spurious test and compare
        !          2579: instructions.  The function entry and exit sequences are generated
        !          2580: directly as assembler code in this pass; they never exist as RTL.
        !          2581: 
        !          2582: The source files are @file{final.c} plus @file{insn-output.c}; the
        !          2583: latter is generated automatically from the machine description by the
        !          2584: tool @file{genoutput}.  The header file @file{conditions.h} is used
        !          2585: for communication between these files.
        !          2586: 
        !          2587: @item
        !          2588: Debugging information output.  This is run after final because it must
        !          2589: output the stack slot offsets for pseudo registers that did not get
        !          2590: hard registers.  Source files are @file{dbxout.c} for DBX symbol table
        !          2591: format and @file{symout.c} for GDB's own symbol table format.
        !          2592: @end itemize
        !          2593: 
        !          2594: Some additional files are used by all or many passes:
        !          2595: 
        !          2596: @itemize @bullet
        !          2597: @item
        !          2598: Every pass uses @file{machmode.def}, which defines the machine modes.
        !          2599: 
        !          2600: @item
        !          2601: All the passes that work with RTL use the header files @file{rtl.h}
        !          2602: and @file{rtl.def}, and subroutines in file @file{rtl.c}.  The tools
        !          2603: @code{gen*} also use these files to read and work with the machine
        !          2604: description RTL.
        !          2605: 
        !          2606: @item
        !          2607: Several passes refer to the header file @file{insn-config.h} which
        !          2608: contains a few parameters (C macro definitions) generated
        !          2609: automatically from the machine description RTL by the tool
        !          2610: @code{genconfig}.
        !          2611: 
        !          2612: @item
        !          2613: Several passes use the instruction recognizer, which consists of
        !          2614: @file{recog.c} and @file{recog.h}, plus the files @file{insn-recog.c}
        !          2615: and @file{insn-extract.c} that are generated automatically from the
        !          2616: machine description by the tools @file{genrecog} and
        !          2617: @file{genextract}.@refill
        !          2618: 
        !          2619: @item
        !          2620: Several passes use the header files @file{regs.h} which defines the
        !          2621: information recorded about pseudo register usage, and @file{basic-block.h}
        !          2622: which defines the information recorded about basic blocks.
        !          2623: 
        !          2624: @item
        !          2625: @file{hard-reg-set.h} defines the type @code{HARD_REG_SET}, a bit-vector
        !          2626: with a bit for each hard register, and some macros to manipulate it.
        !          2627: This type is just @code{int} if the machine has few enough hard registers;
        !          2628: otherwise it is an array of @code{int} and some of the macros expand
        !          2629: into loops.
        !          2630: @end itemize
        !          2631: 
        !          2632: @node RTL, Machine Desc, Passes, Top
        !          2633: @chapter RTL Representation
        !          2634: 
        !          2635: Most of the work of the compiler is done on an intermediate representation
        !          2636: called register transfer language.  In this language, the instructions to be
        !          2637: output are described, pretty much one by one, in an algebraic form that
        !          2638: describes what the instruction does.
        !          2639: 
        !          2640: RTL is inspired by Lisp lists.  It has both an internal form, made up of
        !          2641: structures that point at other structures, and a textual form that is used
        !          2642: in the machine description and in printed debugging dumps.  The textual
        !          2643: form uses nested parentheses to indicate the pointers in the internal form.
        !          2644: 
        !          2645: @menu
        !          2646: * RTL Objects::       Expressions vs vectors vs strings vs integers.
        !          2647: * Accessors::         Macros to access expression operands or vector elts.
        !          2648: * Flags::             Other flags in an RTL expression.
        !          2649: * Machine Modes::     Describing the size and format of a datum.
        !          2650: * Constants::         Expressions with constant values.
        !          2651: * Regs and Memory::   Expressions representing register contents or memory.
        !          2652: * Arithmetic::        Expressions representing arithmetic on other expressions.
        !          2653: * Comparisons::       Expressions representing comparison of expressions.
        !          2654: * Bit Fields::        Expressions representing bit-fields in memory or reg.
        !          2655: * Conversions::       Extending, truncating, floating or fixing.
        !          2656: * RTL Declarations::  Declaring volatility, constancy, etc.
        !          2657: * Side Effects::      Expressions for storing in registers, etc.
        !          2658: * Incdec::            Embedded side-effects for autoincrement addressing.
        !          2659: * Assembler::        Representing @code{asm} with operands.
        !          2660: * Insns::             Expression types for entire insns.
        !          2661: * Calls::            RTL representation of function call insns.
        !          2662: * Sharing::           Some expressions are unique; others *must* be copied.
        !          2663: @end menu
        !          2664: 
        !          2665: @node RTL Objects, Accessors, RTL, RTL
        !          2666: @section RTL Object Types
        !          2667: 
        !          2668: RTL uses four kinds of objects: expressions, integers, strings and vectors.
        !          2669: Expressions are the most important ones.  An RTL expression (``RTX'', for
        !          2670: short) is a C structure, but it is usually referred to with a pointer; a
        !          2671: type that is given the typedef name @code{rtx}.
        !          2672: 
        !          2673: An integer is simply an @code{int}, and a string is a @code{char *}.
        !          2674: Within RTL code, strings appear only inside @samp{symbol_ref} expressions,
        !          2675: but they appear in other contexts in the RTL expressions that make up
        !          2676: machine descriptions.  Their written form uses decimal digits.
        !          2677: 
        !          2678: A string is a sequence of characters.  In core it is represented as a
        !          2679: @code{char *} in usual C fashion, and it is written in C syntax as well.
        !          2680: However, strings in RTL may never be null.  If you write an empty string in
        !          2681: a machine description, it is represented in core as a null pointer rather
        !          2682: than as a pointer to a null character.  In certain contexts, these null
        !          2683: pointers instead of strings are valid.
        !          2684: 
        !          2685: A vector contains an arbitrary, specified number of pointers to
        !          2686: expressions.  The number of elements in the vector is explicitly present in
        !          2687: the vector.  The written form of a vector consists of square brackets
        !          2688: (@samp{[@dots{}]}) surrounding the elements, in sequence and with
        !          2689: whitespace separating them.  Vectors of length zero are not created; null
        !          2690: pointers are used instead.
        !          2691: 
        !          2692: Expressions are classified by @dfn{expression codes} (also called RTX
        !          2693: codes).  The expression code is a name defined in @file{rtl.def}, which is
        !          2694: also (in upper case) a C enumeration constant.  The possible expression
        !          2695: codes and their meanings are machine-independent.  The code of an RTX can
        !          2696: be extracted with the macro @code{GET_CODE (@var{x})} and altered with
        !          2697: @code{PUT_CODE (@var{x}, @var{newcode})}.
        !          2698: 
        !          2699: The expression code determines how many operands the expression contains,
        !          2700: and what kinds of objects they are.  In RTL, unlike Lisp, you cannot tell
        !          2701: by looking at an operand what kind of object it is.  Instead, you must know
        !          2702: from its context---from the expression code of the containing expression.
        !          2703: For example, in an expression of code @samp{subreg}, the first operand is
        !          2704: to be regarded as an expression and the second operand as an integer.  In
        !          2705: an expression of code @samp{plus}, there are two operands, both of which
        !          2706: are to be regarded as expressions.  In a @samp{symbol_ref} expression,
        !          2707: there is one operand, which is to be regarded as a string.
        !          2708: 
        !          2709: Expressions are written as parentheses containing the name of the
        !          2710: expression type, its flags and machine mode if any, and then the operands
        !          2711: of the expression (separated by spaces).
        !          2712: 
        !          2713: Expression code names in the @samp{md} file are written in lower case,
        !          2714: but when they appear in C code they are written in upper case.  In this
        !          2715: manual, they are shown as follows: @samp{const_int}.
        !          2716: 
        !          2717: In a few contexts a null pointer is valid where an expression is normally
        !          2718: wanted.  The written form of this is @samp{(nil)}.
        !          2719: 
        !          2720: @node Accessors, Flags, RTL Objects, RTL
        !          2721: @section Access to Operands
        !          2722: 
        !          2723: For each expression type @file{rtl.def} specifies the number of contained
        !          2724: objects and their kinds, with four possibilities: @samp{e} for expression
        !          2725: (actually a pointer to an expression), @samp{i} for integer, @samp{s} for
        !          2726: string, and @samp{E} for vector of expressions.  The sequence of letters
        !          2727: for an expression code is called its @dfn{format}.  Thus, the format of
        !          2728: @samp{subreg} is @samp{ei}.@refill
        !          2729: 
        !          2730: Two other format characters are used occasionally: @samp{u} and @samp{0}.
        !          2731: @samp{u} is equivalent to @samp{e} except that it is printed differently in
        !          2732: debugging dumps, and @samp{0} means a slot whose contents do not fit any
        !          2733: normal category.  @samp{0} slots are not printed at all in dumps, and are
        !          2734: often used in special ways by small parts of the compiler.@refill
        !          2735: 
        !          2736: There are macros to get the number of operands and the format of an
        !          2737: expression code:
        !          2738: 
        !          2739: @table @code
        !          2740: @item GET_RTX_LENGTH (@var{code})
        !          2741: Number of operands of an RTX of code @var{code}.
        !          2742: 
        !          2743: @item GET_RTX_FORMAT (@var{code})
        !          2744: The format of an RTX of code @var{code}, as a C string.
        !          2745: @end table
        !          2746: 
        !          2747: Operands of expressions are accessed using the macros @code{XEXP},
        !          2748: @code{XINT} and @code{XSTR}.  Each of these macros takes two arguments: an
        !          2749: expression-pointer (RTX) and an operand number (counting from zero).
        !          2750: Thus,@refill
        !          2751: 
        !          2752: @example
        !          2753: XEXP (@var{x}, 2)
        !          2754: @end example
        !          2755: 
        !          2756: @noindent
        !          2757: accesses operand 2 of expression @var{x}, as an expression.
        !          2758: 
        !          2759: @example
        !          2760: XINT (@var{x}, 2)
        !          2761: @end example
        !          2762: 
        !          2763: @noindent
        !          2764: accesses the same operand as an integer.  @code{XSTR}, used in the same
        !          2765: fashion, would access it as a string.
        !          2766: 
        !          2767: Any operand can be accessed as an integer, as an expression or as a string.
        !          2768: You must choose the correct method of access for the kind of value actually
        !          2769: stored in the operand.  You would do this based on the expression code of
        !          2770: the containing expression.  That is also how you would know how many
        !          2771: operands there are.
        !          2772: 
        !          2773: For example, if @var{x} is a @samp{subreg} expression, you know that it has
        !          2774: two operands which can be correctly accessed as @code{XEXP (@var{x}, 0)}
        !          2775: and @code{XINT (@var{x}, 1)}.  If you did @code{XINT (@var{x}, 0)}, you
        !          2776: would get the address of the expression operand but cast as an integer;
        !          2777: that might occasionally be useful, but it would be cleaner to write
        !          2778: @code{(int) XEXP (@var{x}, 0)}.  @code{XEXP (@var{x}, 1)} would also
        !          2779: compile without error, and would return the second, integer operand cast as
        !          2780: an expression pointer, which would probably result in a crash when
        !          2781: accessed.  Nothing stops you from writing @code{XEXP (@var{x}, 28)} either,
        !          2782: but this will access memory past the end of the expression with
        !          2783: unpredictable results.@refill
        !          2784: 
        !          2785: Access to operands which are vectors is more complicated.  You can use the
        !          2786: macro @code{XVEC} to get the vector-pointer itself, or the macros
        !          2787: @code{XVECEXP} and @code{XVECLEN} to access the elements and length of a
        !          2788: vector.
        !          2789: 
        !          2790: @table @code
        !          2791: @item XVEC (@var{exp}, @var{idx})
        !          2792: Access the vector-pointer which is operand number @var{idx} in @var{exp}.
        !          2793: 
        !          2794: @item XVECLEN (@var{exp}, @var{idx})
        !          2795: Access the length (number of elements) in the vector which is
        !          2796: in operand number @var{idx} in @var{exp}.  This value is an @code{int}.
        !          2797: 
        !          2798: @item XVECEXP (@var{exp}, @var{idx}, @var{eltnum})
        !          2799: Access element number @var{eltnum} in the vector which is
        !          2800: in operand number @var{idx} in @var{exp}.  This value is an RTX.
        !          2801: 
        !          2802: It is up to you to make sure that @var{eltnum} is not negative
        !          2803: and is less than @code{XVECLEN (@var{exp}, @var{idx})}.
        !          2804: @end table
        !          2805: 
        !          2806: All the macros defined in this section expand into lvalues and therefore
        !          2807: can be used to assign the operands, lengths and vector elements as well as
        !          2808: to access them.
        !          2809: 
        !          2810: @node Flags, Machine Modes, Accessors, RTL
        !          2811: @section Flags in an RTL Expression
        !          2812: 
        !          2813: RTL expressions contain several flags (one-bit bit-fields) that are used
        !          2814: in certain types of expression.
        !          2815: 
        !          2816: @table @code
        !          2817: @item used
        !          2818: This flag is used only momentarily, at the end of RTL generation for a
        !          2819: function, to count the number of times an expression appears in insns.
        !          2820: Expressions that appear more than once are copied, according to the
        !          2821: rules for shared structure (@pxref{Sharing}).
        !          2822: 
        !          2823: @item volatil
        !          2824: This flag is used in @samp{mem} and @samp{reg} expressions and in insns.
        !          2825: In RTL dump files, it is printed as @samp{/v}.
        !          2826: 
        !          2827: In a @samp{mem} expression, it is 1 if the memory reference is volatile.
        !          2828: Volatile memory references may not be deleted, reordered or combined.
        !          2829: 
        !          2830: In a @samp{reg} expression, it is 1 if the value is a user-level variable.
        !          2831: 0 indicates an internal compiler temporary.
        !          2832: 
        !          2833: In an insn, 1 means the insn has been deleted.
        !          2834: 
        !          2835: @item in_struct
        !          2836: This flag is used in @samp{mem} expressions.  It is 1 if the memory
        !          2837: datum referred to is all or part of a structure or array; 0 if it is (or
        !          2838: might be) a scalar variable.  A reference through a C pointer has 0
        !          2839: because the pointer might point to a scalar variable.
        !          2840: 
        !          2841: This information allows the compiler to determine something about possible
        !          2842: cases of aliasing.
        !          2843: 
        !          2844: In an RTL dump, this flag is represented as @samp{/s}.
        !          2845: 
        !          2846: @item unchanging
        !          2847: This flag is used in @samp{reg} and @samp{mem} expressions.  1 means
        !          2848: that the value of the expression never changes (at least within the
        !          2849: current function).
        !          2850: 
        !          2851: In an RTL dump, this flag is represented as @samp{/u}.
        !          2852: @end table
        !          2853: 
        !          2854: @node Machine Modes, Constants, Flags, RTL
        !          2855: @section Machine Modes
        !          2856: 
        !          2857: A machine mode describes a size of data object and the representation used
        !          2858: for it.  In the C code, machine modes are represented by an enumeration
        !          2859: type, @code{enum machine_mode}, defined in @file{machmode.def}.  Each RTL
        !          2860: expression has room for a machine mode and so do certain kinds of tree
        !          2861: expressions (declarations and types, to be precise).
        !          2862: 
        !          2863: In debugging dumps and machine descriptions, the machine mode of an RTL
        !          2864: expression is written after the expression code with a colon to separate
        !          2865: them.  The letters @samp{mode} which appear at the end of each machine mode
        !          2866: name are omitted.  For example, @code{(reg:SI 38)} is a @samp{reg}
        !          2867: expression with machine mode @code{SImode}.  If the mode is
        !          2868: @code{VOIDmode}, it is not written at all.
        !          2869: 
        !          2870: Here is a table of machine modes.
        !          2871: 
        !          2872: @table @code
        !          2873: @item QImode
        !          2874: ``Quarter-Integer'' mode represents a single byte treated as an integer.
        !          2875: 
        !          2876: @item HImode
        !          2877: ``Half-Integer'' mode represents a two-byte integer.
        !          2878: 
        !          2879: @item SImode
        !          2880: ``Single Integer'' mode represents a four-byte integer.
        !          2881: 
        !          2882: @item DImode
        !          2883: ``Double Integer'' mode represents an eight-byte integer.
        !          2884: 
        !          2885: @item TImode
        !          2886: ``Tetra Integer'' (?) mode represents a sixteen-byte integer.
        !          2887: 
        !          2888: @item SFmode
        !          2889: ``Single Floating'' mode represents a single-precision (four byte) floating
        !          2890: point number.
        !          2891: 
        !          2892: @item DFmode
        !          2893: ``Double Floating'' mode represents a double-precision (eight byte) floating
        !          2894: point number.
        !          2895: 
        !          2896: @item TFmode
        !          2897: ``Tetra Floating'' mode represents a quadruple-precision (sixteen byte)
        !          2898: floating point number.
        !          2899: 
        !          2900: @item BLKmode
        !          2901: ``Block'' mode represents values that are aggregates to which none of
        !          2902: the other modes apply.  In RTL, only memory references can have this mode,
        !          2903: and only if they appear in string-move or vector instructions.  On machines
        !          2904: which have no such instructions, @code{BLKmode} will not appear in RTL.
        !          2905: 
        !          2906: @item VOIDmode
        !          2907: Void mode means the absence of a mode or an unspecified mode.
        !          2908: For example, RTL expressions of code @samp{const_int} have mode
        !          2909: @code{VOIDmode} because they can be taken to have whatever mode the context
        !          2910: requires.  In debugging dumps of RTL, @code{VOIDmode} is expressed by
        !          2911: the absence of any mode.
        !          2912: 
        !          2913: @item EPmode
        !          2914: ``Entry Pointer'' mode is intended to be used for function variables in
        !          2915: Pascal and other block structured languages.  Such values contain
        !          2916: both a function address and a static chain pointer for access to
        !          2917: automatic variables of outer levels.  This mode is only partially
        !          2918: implemented since C does not use it.
        !          2919: 
        !          2920: @item CSImode@r{, @dots{}}
        !          2921: ``Complex Single Integer'' mode stands for a complex number represented
        !          2922: as a pair of @code{SImode} integers.  Any of the integer and floating modes
        !          2923: may have @samp{C} prefixed to its name to obtain a complex number mode.
        !          2924: For example, there are @code{CQImode}, @code{CSFmode}, and @code{CDFmode}.
        !          2925: Since C does not support complex numbers, these machine modes are only
        !          2926: partially implemented.
        !          2927: 
        !          2928: @item BImode
        !          2929: This is the machine mode of a bit-field in a structure.  It is used
        !          2930: only in the syntax tree, never in RTL, and in the syntax tree it appears
        !          2931: only in declaration nodes.  In C, it appears only in @code{FIELD_DECL}
        !          2932: nodes for structure fields defined with a bit size.
        !          2933: @end table
        !          2934: 
        !          2935: The machine description defines @code{Pmode} as a C macro which expands
        !          2936: into the machine mode used for addresses.  Normally this is @code{SImode}.
        !          2937: 
        !          2938: The only modes which a machine description @i{must} support are
        !          2939: @code{QImode}, @code{SImode}, @code{SFmode} and @code{DFmode}.  The
        !          2940: compiler will attempt to use @code{DImode} for two-word structures and
        !          2941: unions, but it would not be hard to program it to avoid this.  Likewise,
        !          2942: you can arrange for the C type @code{short int} to avoid using
        !          2943: @code{HImode}.  In the long term it would be desirable to make the set of
        !          2944: available machine modes machine-dependent and eliminate all assumptions
        !          2945: about specific machine modes or their uses from the machine-independent
        !          2946: code of the compiler.
        !          2947: 
        !          2948: Here are some C macros that relate to machine modes:
        !          2949: 
        !          2950: @table @code
        !          2951: @item GET_MODE (@var{x})
        !          2952: Returns the machine mode of the RTX @var{x}.
        !          2953: 
        !          2954: @item PUT_MODE (@var{x}, @var{newmode})
        !          2955: Alters the machine mode of the RTX @var{x} to be @var{newmode}.
        !          2956: 
        !          2957: @item GET_MODE_SIZE (@var{m})
        !          2958: Returns the size in bytes of a datum of mode @var{m}.
        !          2959: 
        !          2960: @item GET_MODE_BITSIZE (@var{m})
        !          2961: Returns the size in bits of a datum of mode @var{m}.
        !          2962: 
        !          2963: @item GET_MODE_UNIT_SIZE (@var{m})
        !          2964: Returns the size in bits of the subunits of a datum of mode @var{m}.
        !          2965: This is the same as @code{GET_MODE_SIZE} except in the case of
        !          2966: complex modes and @code{EPmode}.  For them, the unit size is the
        !          2967: size of the real or imaginary part, or the size of the function
        !          2968: pointer or the context pointer.
        !          2969: @end table
        !          2970: 
        !          2971: @node Constants, Regs and Memory, Machine Modes, RTL
        !          2972: @section Constant Expression Types
        !          2973: 
        !          2974: The simplest RTL expressions are those that represent constant values.
        !          2975: 
        !          2976: @table @code
        !          2977: @item (const_int @var{i})
        !          2978: This type of expression represents the integer value @var{i}.  @var{i}
        !          2979: is customarily accessed with the macro @code{INTVAL} as in
        !          2980: @code{INTVAL (@var{exp})}, which is equivalent to @code{XINT (@var{exp}, 0)}.
        !          2981: 
        !          2982: There is only one expression object for the integer value zero;
        !          2983: it is the value of the variable @code{const0_rtx}.  Likewise, the
        !          2984: only expression for integer value one is found in @code{const1_rtx}.
        !          2985: Any attempt to create an expression of code @samp{const_int} and
        !          2986: value zero or one will return @code{const0_rtx} or @code{const1_rtx}
        !          2987: as appropriate.
        !          2988: 
        !          2989: @item (const_double:@var{m} @var{i0} @var{i1})
        !          2990: Represents a floating point constant value of mode @var{m}.  The two
        !          2991: integers @var{i0} and @var{i1} together contain the bits of a
        !          2992: @code{double} value.  To convert them to a @code{double}, do
        !          2993: 
        !          2994: @example
        !          2995: union @{ double d; int i[2];@} u;
        !          2996: u.i[0] = XINT (x, 0);
        !          2997: u.i[1] = XINT (x, 1);
        !          2998: @end example
        !          2999: 
        !          3000: @noindent
        !          3001: and then refer to @code{u.d}.  The value of the constant is
        !          3002: represented as a double in this fashion even if the value represented
        !          3003: is single-precision.
        !          3004: 
        !          3005: The global variables @code{dconst0_rtx} and @code{fconst0_rtx} hold
        !          3006: @samp{const_double} expressions with value 0, in modes @code{DFmode} and
        !          3007: @code{SFmode}, respectively.
        !          3008: 
        !          3009: @item (symbol_ref @var{symbol})
        !          3010: Represents the value of an assembler label for data.  @var{symbol} is
        !          3011: a string that describes the name of the assembler label.  If it starts
        !          3012: with a @samp{*}, the label is the rest of @var{symbol} not including
        !          3013: the @samp{*}.  Otherwise, the label is @var{symbol}, prefixed with
        !          3014: @samp{_}.
        !          3015: 
        !          3016: @item (label_ref @var{label})
        !          3017: Represents the value of an assembler label for code.  It contains one
        !          3018: operand, an expression, which must be a @samp{code_label} that appears
        !          3019: in the instruction sequence to identify the place where the label
        !          3020: should go.
        !          3021: 
        !          3022: The reason for using a distinct expression type for code label
        !          3023: references is so that jump optimization can distinguish them.
        !          3024: 
        !          3025: @item (const @var{exp})
        !          3026: Represents a constant that is the result of an assembly-time
        !          3027: arithmetic computation.  The operand, @var{exp}, is an expression that
        !          3028: contains only constants (@samp{const_int}, @samp{symbol_ref} and
        !          3029: @samp{label_ref} expressions) combined with @samp{plus} and
        !          3030: @samp{minus}.  However, not all combinations are valid, since the
        !          3031: assembler cannot do arbitrary arithmetic on relocatable symbols.
        !          3032: @end table
        !          3033: 
        !          3034: @node Regs and Memory, Arithmetic, Constants, RTL
        !          3035: @section Registers and Memory
        !          3036: 
        !          3037: Here are the RTL expression types for describing access to machine
        !          3038: registers and to main memory.
        !          3039: 
        !          3040: @table @code
        !          3041: @item (reg:@var{m} @var{n})
        !          3042: For small values of the integer @var{n} (less than
        !          3043: @code{FIRST_PSEUDO_REGISTER}), this stands for a reference to machine
        !          3044: register number @var{n}: a @dfn{hard register}.  For larger values of
        !          3045: @var{n}, it stands for a temporary value or @dfn{pseudo register}.
        !          3046: The compiler's strategy is to generate code assuming an unlimited
        !          3047: number of such pseudo registers, and later convert them into hard
        !          3048: registers or into memory references.
        !          3049: 
        !          3050: The symbol @code{FIRST_PSEUDO_REGISTER} is defined by the machine
        !          3051: description, since the number of hard registers on the machine is an
        !          3052: invariant characteristic of the machine.  Note, however, that not
        !          3053: all of the machine registers must be general registers.  All the
        !          3054: machine registers that can be used for storage of data are given
        !          3055: hard register numbers, even those that can be used only in certain
        !          3056: instructions or can hold only certain types of data.
        !          3057: 
        !          3058: Each pseudo register number used in a function's RTL code is
        !          3059: represented by a unique @samp{reg} expression.
        !          3060: 
        !          3061: @var{m} is the machine mode of the reference.  It is necessary because
        !          3062: machines can generally refer to each register in more than one mode.
        !          3063: For example, a register may contain a full word but there may be
        !          3064: instructions to refer to it as a half word or as a single byte, as
        !          3065: well as instructions to refer to it as a floating point number of
        !          3066: various precisions.
        !          3067: 
        !          3068: Even for a register that the machine can access in only one mode,
        !          3069: the mode must always be specified.
        !          3070: 
        !          3071: A hard register may be accessed in various modes throughout one
        !          3072: function, but each pseudo register is given a natural mode
        !          3073: and is accessed only in that mode.  When it is necessary to describe
        !          3074: an access to a pseudo register using a nonnatural mode, a @samp{subreg}
        !          3075: expression is used.
        !          3076: 
        !          3077: A @samp{reg} expression with a machine mode that specifies more than
        !          3078: one word of data may actually stand for several consecutive registers.
        !          3079: If in addition the register number specifies a hardware register, then
        !          3080: it actually represents several consecutive hardware registers starting
        !          3081: with the specified one.
        !          3082: 
        !          3083: Such multi-word hardware register @samp{reg} expressions may not be live
        !          3084: across the boundary of a basic block.  The lifetime analysis pass does not
        !          3085: know how to record properly that several consecutive registers are
        !          3086: actually live there, and therefore register allocation would be confused.
        !          3087: The CSE pass must go out of its way to make sure the situation does
        !          3088: not arise.
        !          3089: 
        !          3090: @item (subreg:@var{m} @var{reg} @var{wordnum})
        !          3091: @samp{subreg} expressions are used to refer to a register in a machine
        !          3092: mode other than its natural one, or to refer to one register of
        !          3093: a multi-word @samp{reg} that actually refers to several registers.
        !          3094: 
        !          3095: Each pseudo-register has a natural mode.  If it is necessary to
        !          3096: operate on it in a different mode---for example, to perform a fullword
        !          3097: move instruction on a pseudo-register that contains a single byte---
        !          3098: the pseudo-register must be enclosed in a @samp{subreg}.  In such
        !          3099: a case, @var{wordnum} is zero.
        !          3100: 
        !          3101: The other use of @samp{subreg} is to extract the individual registers
        !          3102: of a multi-register value.  Machine modes such as @code{DImode} and
        !          3103: @code{EPmode} indicate values longer than a word, values which usually
        !          3104: require two consecutive registers.  To access one of the registers,
        !          3105: use a @samp{subreg} with mode @code{SImode} and a @var{wordnum} that
        !          3106: says which register.
        !          3107: 
        !          3108: The compilation parameter @code{WORDS_BIG_ENDIAN}, if defined, says
        !          3109: that word number zero is the most significant part; otherwise, it is
        !          3110: the least significant part.
        !          3111: 
        !          3112: Note that it is not valid to access a @code{DFmode} value in @code{SFmode}
        !          3113: using a @samp{subreg}.  On some machines the most significant part of a
        !          3114: @code{DFmode} value does not have the same format as a single-precision
        !          3115: floating value.
        !          3116: 
        !          3117: @item (cc0)
        !          3118: This refers to the machine's condition code register.  It has no
        !          3119: operands and may not have a machine mode.  It may be validly used in
        !          3120: only two contexts: as the destination of an assignment (in test and
        !          3121: compare instructions) and in comparison operators comparing against
        !          3122: zero (@samp{const_int} with value zero; that is to say,
        !          3123: @code{const0_rtx}).
        !          3124: 
        !          3125: There is only one expression object of code @samp{cc0}; it is the
        !          3126: value of the variable @code{cc0_rtx}.  Any attempt to create an
        !          3127: expression of code @samp{cc0} will return @code{cc0_rtx}.
        !          3128: 
        !          3129: One special thing about the condition code register is that
        !          3130: instructions can set it implicitly.  On many machines, nearly all
        !          3131: instructions set the condition code based on the value that they
        !          3132: compute or store.  It is not necessary to record these actions
        !          3133: explicitly in the RTL because the machine description includes a
        !          3134: prescription for recognizing the instructions that do so (by means of
        !          3135: the macro @code{NOTICE_UPDATE_CC}).  Only instructions whose sole
        !          3136: purpose is to set the condition code, and instructions that use the
        !          3137: condition code, need mention @code{(cc0)}.
        !          3138: 
        !          3139: @item (pc)
        !          3140: This represents the machine's program counter.  It has no operands and
        !          3141: may not have a machine mode.  @code{(pc)} may be validly used only in
        !          3142: certain specific contexts in jump instructions.
        !          3143: 
        !          3144: There is only one expression object of code @samp{pc}; it is the value
        !          3145: of the variable @code{pc_rtx}.  Any attempt to create an expression of
        !          3146: code @samp{pc} will return @code{pc_rtx}.
        !          3147: 
        !          3148: All instructions that do not jump alter the program counter implicitly
        !          3149: by incrementing it, but there is no need to mention this in the RTL.
        !          3150: 
        !          3151: @item (mem:@var{m} @var{addr})
        !          3152: This RTX represents a reference to main memory at an address
        !          3153: represented by the expression @var{addr}.  @var{m} specifies how large
        !          3154: a unit of memory is accessed.
        !          3155: @end table
        !          3156: 
        !          3157: @node Arithmetic, Comparisons, Regs and Memory, RTL
        !          3158: @section RTL Expressions for Arithmetic
        !          3159: 
        !          3160: @table @code
        !          3161: @item (plus:@var{m} @var{x} @var{y})
        !          3162: Represents the sum of the values represented by @var{x} and @var{y}
        !          3163: carried out in machine mode @var{m}.  This is valid only if
        !          3164: @var{x} and @var{y} both are valid for mode @var{m}.
        !          3165: 
        !          3166: @item (minus:@var{m} @var{x} @var{y})
        !          3167: Like @samp{plus} but represents subtraction.
        !          3168: 
        !          3169: @item (minus @var{x} @var{y})
        !          3170: Represents the result of subtracting @var{y} from @var{x}
        !          3171: for purposes of comparison.  The absence of a machine mode
        !          3172: in the @samp{minus} expression indicates that the result is
        !          3173: computed without overflow, as if with infinite precision.
        !          3174: 
        !          3175: Of course, machines can't really subtract with infinite precision.
        !          3176: However, they can pretend to do so when only the sign of the
        !          3177: result will be used, which is the case when the result is stored
        !          3178: in @code{(cc0)}.  And that is the only way this kind of expression
        !          3179: may validly be used: as a value to be stored in the condition codes.
        !          3180: 
        !          3181: @item (neg:@var{m} @var{x})
        !          3182: Represents the negation (subtraction from zero) of the value
        !          3183: represented by @var{x}, carried out in mode @var{m}.  @var{x} must be
        !          3184: valid for mode @var{m}.
        !          3185: 
        !          3186: @item (mult:@var{m} @var{x} @var{y})
        !          3187: Represents the signed product of the values represented by @var{x} and
        !          3188: @var{y} carried out in machine mode @var{m}.  If
        !          3189: @var{x} and @var{y} are both valid for mode @var{m}, this is ordinary
        !          3190: size-preserving multiplication.  Alternatively, both @var{x} and @var{y}
        !          3191: may be valid for a different, narrower mode.  This represents the
        !          3192: kind of multiplication that generates a product wider than the operands.
        !          3193: Widening multiplication and same-size multiplication are completely
        !          3194: distinct and supported by different machine instructions; machines may
        !          3195: support one but not the other.@refill
        !          3196: 
        !          3197: @samp{mult} may be used for floating point division as well.
        !          3198: Then @var{m} is a floating point machine mode.
        !          3199: 
        !          3200: @item (umult:@var{m} @var{x} @var{y})
        !          3201: Like @samp{mult} but represents unsigned multiplication.  It may be
        !          3202: used in both same-size and widening forms, like @samp{mult}.
        !          3203: @samp{umult} is used only for fixed-point multiplication.
        !          3204: 
        !          3205: @item (div:@var{m} @var{x} @var{y})
        !          3206: Represents the quotient in signed division of @var{x} by @var{y},
        !          3207: carried out in machine mode @var{m}.  If @var{m} is a floating-point
        !          3208: mode, it represents the exact quotient; otherwise, the integerized
        !          3209: quotient.  If @var{x} and @var{y} are both valid for mode @var{m},
        !          3210: this is ordinary size-preserving division.  Some machines have
        !          3211: division instructions in which the operands and quotient widths are
        !          3212: not all the same; such instructions are represented by @samp{div}
        !          3213: expressions in which the machine modes are not all the same.
        !          3214: 
        !          3215: @item (udiv:@var{m} @var{x} @var{y})
        !          3216: Like @samp{div} but represents unsigned division.
        !          3217: 
        !          3218: @item (mod:@var{m} @var{x} @var{y})
        !          3219: @itemx (umod:@var{m} @var{x} @var{y})
        !          3220: Like @samp{div} and @samp{udiv} but represent the remainder instead of
        !          3221: the quotient.
        !          3222: 
        !          3223: @item (not:@var{m} @var{x})
        !          3224: Represents the bitwise complement of the value represented by @var{x},
        !          3225: carried out in mode @var{m}, which must be a fixed-point machine mode.
        !          3226: @var{x} must be valid for mode @var{m}, which must be a fixed-point mode.
        !          3227: 
        !          3228: @item (and:@var{m} @var{x} @var{y})
        !          3229: Represents the bitwise logical-and of the values represented by
        !          3230: @var{x} and @var{y}, carried out in machine mode @var{m}.  This is
        !          3231: valid only if @var{x} and @var{y} both are valid for mode @var{m},
        !          3232: which must be a fixed-point mode.
        !          3233: 
        !          3234: @item (ior:@var{m} @var{x} @var{y})
        !          3235: Represents the bitwise inclusive-or of the values represented by
        !          3236: @var{x} and @var{y}, carried out in machine mode @var{m}.  This is
        !          3237: valid only if @var{x} and @var{y} both are valid for mode @var{m},
        !          3238: which must be a fixed-point mode.
        !          3239: 
        !          3240: @item (xor:@var{m} @var{x} @var{y})
        !          3241: Represents the bitwise exclusive-or of the values represented by
        !          3242: @var{x} and @var{y}, carried out in machine mode @var{m}.  This is
        !          3243: valid only if @var{x} and @var{y} both are valid for mode @var{m},
        !          3244: which must be a fixed-point mode.
        !          3245: 
        !          3246: @item (lshift:@var{m} @var{x} @var{c})
        !          3247: Represents the result of logically shifting @var{x} left by @var{c}
        !          3248: places.  @var{x} must be valid for the mode @var{m}, a fixed-point
        !          3249: machine mode.  @var{c} must be valid for a fixed-point mode;
        !          3250: which mode is determined by the mode called for in the machine
        !          3251: description entry for the left-shift instruction.  For example,
        !          3252: on the Vax, the mode of @var{c} is @code{QImode} regardless of @var{m}.
        !          3253: 
        !          3254: On some machines, negative values of @var{c} may be meaningful; this
        !          3255: is why logical left shift and arithmetic left shift are distinguished.
        !          3256: For example, Vaxes have no right-shift instructions, and right shifts
        !          3257: are represented as left-shift instructions whose counts happen
        !          3258: to be negative constants or else computed (in a previous instruction)
        !          3259: by negation.
        !          3260: 
        !          3261: @item (ashift:@var{m} @var{x} @var{c})
        !          3262: Like @samp{lshift} but for arithmetic left shift.
        !          3263: 
        !          3264: @item (lshiftrt:@var{m} @var{x} @var{c})
        !          3265: @itemx (ashiftrt:@var{m} @var{x} @var{c})
        !          3266: Like @samp{lshift} and @samp{ashift} but for right shift.
        !          3267: 
        !          3268: @item (rotate:@var{m} @var{x} @var{c})
        !          3269: @itemx (rotatert:@var{m} @var{x} @var{c})
        !          3270: Similar but represent left and right rotate.
        !          3271: 
        !          3272: @item (abs:@var{m} @var{x})
        !          3273: Represents the absolute value of @var{x}, computed in mode @var{m}.
        !          3274: @var{x} must be valid for @var{m}.
        !          3275: 
        !          3276: @item (sqrt:@var{m} @var{x})
        !          3277: Represents the square root of @var{x}, computed in mode @var{m}.
        !          3278: @var{x} must be valid for @var{m}.  Most often @var{m} will be
        !          3279: a floating point mode.
        !          3280: 
        !          3281: @item (ffs:@var{m} @var{x})
        !          3282: Represents the one plus the index of the least significant 1-bit in
        !          3283: @var{x}, represented as an integer of mode @var{m}.  (The value is
        !          3284: zero if @var{x} is zero.)  The mode of @var{x} need not be @var{m};
        !          3285: depending on the target machine, various mode combinations may be
        !          3286: valid.
        !          3287: @end table
        !          3288: 
        !          3289: @node Comparisons, Bit Fields, Arithmetic, RTL
        !          3290: @section Comparison Operations
        !          3291: 
        !          3292: Comparison operators test a relation on two operands and are considered to
        !          3293: represent the value 1 if the relation holds, or zero if it does not.  The
        !          3294: mode of the comparison is determined by the operands; they must both be
        !          3295: valid for a common machine mode.  A comparison with both operands constant
        !          3296: would be invalid as the machine mode could not be deduced from it, but such
        !          3297: a comparison should never exist in RTL due to constant folding.
        !          3298: 
        !          3299: Inequality comparisons come in two flavors, signed and unsigned.  Thus,
        !          3300: there are distinct expression codes @samp{gt} and @samp{gtu} for signed and
        !          3301: unsigned greater-than.  These can produce different results for the same
        !          3302: pair of integer values: for example, 1 is signed greater-than -1 but not
        !          3303: unsigned greater-than, because -1 when regarded as unsigned is actually
        !          3304: @code{0xffffffff} which is greater than 1.
        !          3305: 
        !          3306: The signed comparisons are also used for floating point values.  Floating
        !          3307: point comparisons are distinguished by the machine modes of the operands.
        !          3308: 
        !          3309: The comparison operators may be used to compare the condition codes
        !          3310: @code{(cc0)} against zero, as in @code{(eq (cc0) (const_int 0))}.  Such a
        !          3311: construct actually refers to the result of the preceding instruction in
        !          3312: which the condition codes were set.  The above example stands for 1 if the
        !          3313: condition codes were set to say ``zero'' or ``equal'', 0 otherwise.
        !          3314: Although the same comparison operators are used for this as may be used in
        !          3315: other contexts on actual data, no confusion can result since the machine
        !          3316: description would never allow both kinds of uses in the same context.
        !          3317: 
        !          3318: @table @code
        !          3319: @item (eq @var{x} @var{y})
        !          3320: 1 if the values represented by @var{x} and @var{y} are equal,
        !          3321: otherwise 0.
        !          3322: 
        !          3323: @item (ne @var{x} @var{y})
        !          3324: 1 if the values represented by @var{x} and @var{y} are not equal,
        !          3325: otherwise 0.
        !          3326: 
        !          3327: @item (gt @var{x} @var{y})
        !          3328: 1 if the @var{x} is greater than @var{y}.  If they are fixed-point,
        !          3329: the comparison is done in a signed sense.
        !          3330: 
        !          3331: @item (gtu @var{x} @var{y})
        !          3332: Like @samp{gt} but does unsigned comparison, on fixed-point numbers only.
        !          3333: 
        !          3334: @item (lt @var{x} @var{y})
        !          3335: @item (ltu @var{x} @var{y})
        !          3336: Like @samp{gt} and @samp{gtu} but test for ``less than''.
        !          3337: 
        !          3338: @item (ge @var{x} @var{y})
        !          3339: @item (geu @var{x} @var{y})
        !          3340: Like @samp{gt} and @samp{gtu} but test for ``greater than or equal''.
        !          3341: 
        !          3342: @item (le @var{x} @var{y})
        !          3343: @item (leu @var{x} @var{y})
        !          3344: Like @samp{gt} and @samp{gtu} but test for ``less than or equal''.
        !          3345: 
        !          3346: @item (if_then_else @var{cond} @var{then} @var{else})
        !          3347: This is not a comparison operation but is listed here because it is
        !          3348: always used in conjunction with a comparison operation.  To be
        !          3349: precise, @var{cond} is a comparison expression.  This expression
        !          3350: represents a choice, according to @var{cond}, between the value
        !          3351: represented by @var{then} and the one represented by @var{else}.
        !          3352: 
        !          3353: On most machines, @samp{if_then_else} expressions are valid only
        !          3354: to express conditional jumps.
        !          3355: @end table
        !          3356: 
        !          3357: @node Bit Fields, Conversions, Comparisons, RTL
        !          3358: @section Bit-fields
        !          3359: 
        !          3360: Special expression codes exist to represent bit-field instructions.
        !          3361: These types of expressions are lvalues in RTL; they may appear
        !          3362: on the left side of a assignment, indicating insertion of a value
        !          3363: into the specified bit field.
        !          3364: 
        !          3365: @table @code
        !          3366: @item (sign_extract:SI @var{loc} @var{size} @var{pos})
        !          3367: This represents a reference to a sign-extended bit-field contained or
        !          3368: starting in @var{loc} (a memory or register reference).  The bit field
        !          3369: is @var{size} bits wide and starts at bit @var{pos}.  The compilation
        !          3370: option @code{BITS_BIG_ENDIAN} says which end of the memory unit
        !          3371: @var{pos} counts from.
        !          3372: 
        !          3373: Which machine modes are valid for @var{loc} depends on the machine,
        !          3374: but typically @var{loc} should be a single byte when in memory
        !          3375: or a full word in a register.
        !          3376: 
        !          3377: @item (zero_extract:SI @var{loc} @var{size} @var{pos})
        !          3378: Like @samp{sign_extract} but refers to an unsigned or zero-extended
        !          3379: bit field.  The same sequence of bits are extracted, but they
        !          3380: are filled to an entire word with zeros instead of by sign-extension.
        !          3381: @end table
        !          3382: 
        !          3383: @node Conversions, RTL Declarations, Bit Fields, RTL
        !          3384: @section Conversions
        !          3385: 
        !          3386: All conversions between machine modes must be represented by
        !          3387: explicit conversion operations.  For example, an expression
        !          3388: which is the sum of a byte and a full word cannot be written as
        !          3389: @code{(plus:SI (reg:QI 34) (reg:SI 80))} because the @samp{plus}
        !          3390: operation requires two operands of the same machine mode.
        !          3391: Therefore, the byte-sized operand is enclosed in a conversion
        !          3392: operation, as in
        !          3393: 
        !          3394: @example
        !          3395: (plus:SI (sign_extend:SI (reg:QI 34)) (reg:SI 80))
        !          3396: @end example
        !          3397: 
        !          3398: The conversion operation is not a mere placeholder, because there
        !          3399: may be more than one way of converting from a given starting mode
        !          3400: to the desired final mode.  The conversion operation code says how
        !          3401: to do it.
        !          3402: 
        !          3403: @table @code
        !          3404: @item (sign_extend:@var{m} @var{x})
        !          3405: Represents the result of sign-extending the value @var{x}
        !          3406: to machine mode @var{m}.  @var{m} must be a fixed-point mode
        !          3407: and @var{x} a fixed-point value of a mode narrower than @var{m}.
        !          3408: 
        !          3409: @item (zero_extend:@var{m} @var{x})
        !          3410: Represents the result of zero-extending the value @var{x}
        !          3411: to machine mode @var{m}.  @var{m} must be a fixed-point mode
        !          3412: and @var{x} a fixed-point value of a mode narrower than @var{m}.
        !          3413: 
        !          3414: @item (float_extend:@var{m} @var{x})
        !          3415: Represents the result of extending the value @var{x}
        !          3416: to machine mode @var{m}.  @var{m} must be a floating point mode
        !          3417: and @var{x} a floating point value of a mode narrower than @var{m}.
        !          3418: 
        !          3419: @item (truncate:@var{m} @var{x})
        !          3420: Represents the result of truncating the value @var{x}
        !          3421: to machine mode @var{m}.  @var{m} must be a fixed-point mode
        !          3422: and @var{x} a fixed-point value of a mode wider than @var{m}.
        !          3423: 
        !          3424: @item (float_truncate:@var{m} @var{x})
        !          3425: Represents the result of truncating the value @var{x}
        !          3426: to machine mode @var{m}.  @var{m} must be a floating point mode
        !          3427: and @var{x} a floating point value of a mode wider than @var{m}.
        !          3428: 
        !          3429: @item (float:@var{m} @var{x})
        !          3430: Represents the result of converting fixed point value @var{x},
        !          3431: regarded as signed, to floating point mode @var{m}.
        !          3432: 
        !          3433: @item (unsigned_float:@var{m} @var{x})
        !          3434: Represents the result of converting fixed point value @var{x},
        !          3435: regarded as unsigned, to floating point mode @var{m}.
        !          3436: 
        !          3437: @item (fix:@var{m} @var{x})
        !          3438: When @var{m} is a fixed point mode, represents the result of
        !          3439: converting floating point value @var{x} to mode @var{m}, regarded as
        !          3440: signed.  How rounding is done is not specified, so this operation may
        !          3441: be used validly in compiling C code only for integer-valued operands.
        !          3442: 
        !          3443: @item (unsigned_fix:@var{m} @var{x})
        !          3444: Represents the result of converting floating point value @var{x} to
        !          3445: fixed point mode @var{m}, regarded as unsigned.  How rounding is done
        !          3446: is not specified.
        !          3447: 
        !          3448: @item (fix:@var{m} @var{x})
        !          3449: When @var{m} is a floating point mode, represents the result of
        !          3450: converting floating point value @var{x} (valid for mode @var{m}) to an
        !          3451: integer, still represented in floating point mode @var{m}, by rounding
        !          3452: towards zero.
        !          3453: @end table
        !          3454: 
        !          3455: @node RTL Declarations, Side Effects, Conversions, RTL
        !          3456: @section Declarations
        !          3457: 
        !          3458: Declaration expression codes do not represent arithmetic operations
        !          3459: but rather state assertions about their operands.
        !          3460: 
        !          3461: @table @code
        !          3462: @item (strict_low_part (subreg:@var{m} (reg:@var{n} @var{r}) 0))
        !          3463: This expression code is used in only one context: operand 0 of a
        !          3464: @samp{set} expression.  In addition, the operand of this expression
        !          3465: must be a @samp{subreg} expression.
        !          3466: 
        !          3467: The presence of @samp{strict_low_part} says that the part of the
        !          3468: register which is meaningful in mode @var{n}, but is not part of
        !          3469: mode @var{m}, is not to be altered.  Normally, an assignment to such
        !          3470: a subreg is allowed to have undefined effects on the rest of the
        !          3471: register when @var{m} is less than a word.
        !          3472: @end table
        !          3473: 
        !          3474: @node Side Effects, Incdec, RTL Declarations, RTL
        !          3475: @section Side Effect Expressions
        !          3476: 
        !          3477: The expression codes described so far represent values, not actions.
        !          3478: But machine instructions never produce values; they are meaningful
        !          3479: only for their side effects on the state of the machine.  Special
        !          3480: expression codes are used to represent side effects.
        !          3481: 
        !          3482: The body of an instruction is always one of these side effect codes;
        !          3483: the codes described above, which represent values, appear only as
        !          3484: the operands of these.
        !          3485: 
        !          3486: @table @code
        !          3487: @item (set @var{lval} @var{x})
        !          3488: Represents the action of storing the value of @var{x} into the place
        !          3489: represented by @var{lval}.  @var{lval} must be an expression
        !          3490: representing a place that can be stored in: @samp{reg} (or
        !          3491: @samp{subreg} or @samp{strict_low_part}), @samp{mem}, @samp{pc} or
        !          3492: @samp{cc0}.@refill
        !          3493: 
        !          3494: If @var{lval} is a @samp{reg}, @samp{subreg} or @samp{mem}, it has a
        !          3495: machine mode; then @var{x} must be valid for that mode.@refill
        !          3496: 
        !          3497: If @var{lval} is a @samp{reg} whose machine mode is less than the full
        !          3498: width of the register, then it means that the part of the register
        !          3499: specified by the machine mode is given the specified value and the
        !          3500: rest of the register receives an undefined value.  Likewise, if
        !          3501: @var{lval} is a @samp{subreg} whose machine mode is narrower than
        !          3502: @code{SImode}, the rest of the register can be changed in an undefined way.
        !          3503: 
        !          3504: If @var{lval} is a @samp{strict_low_part} of a @samp{subreg}, then the
        !          3505: part of the register specified by the machine mode of the
        !          3506: @samp{subreg} is given the value @var{x} and the rest of the register
        !          3507: is not changed.@refill
        !          3508: 
        !          3509: If @var{lval} is @code{(cc0)}, it has no machine mode, and @var{x} may
        !          3510: have any mode.  This represents a ``test'' or ``compare'' instruction.@refill
        !          3511: 
        !          3512: If @var{lval} is @code{(pc)}, we have a jump instruction, and the
        !          3513: possibilities for @var{x} are very limited.  It may be a
        !          3514: @samp{label_ref} expression (unconditional jump).  It may be an
        !          3515: @samp{if_then_else} (conditional jump), in which case either the
        !          3516: second or the third operand must be @code{(pc)} (for the case which
        !          3517: does not jump) and the other of the two must be a @samp{label_ref}
        !          3518: (for the case which does jump).  @var{x} may also be a @samp{mem} or
        !          3519: @code{(plus:SI (pc) @var{y})}, where @var{y} may be a @samp{reg} or a
        !          3520: @samp{mem}; these unusual patterns are used to represent jumps through
        !          3521: branch tables.@refill
        !          3522: 
        !          3523: @item (return)
        !          3524: Represents a return from the current function, on machines where this
        !          3525: can be done with one instruction, such as Vaxes.  On machines where a
        !          3526: multi-instruction ``epilogue'' must be executed in order to return
        !          3527: from the function, returning is done by jumping to a label which
        !          3528: precedes the epilogue, and the @samp{return} expression code is never
        !          3529: used.
        !          3530: 
        !          3531: @item (call @var{function} @var{nargs})
        !          3532: Represents a function call.  @var{function} is a @samp{mem} expression
        !          3533: whose address is the address of the function to be called.  @var{nargs}
        !          3534: is an expression representing the number of words of argument.
        !          3535: 
        !          3536: Each machine has a standard machine mode which @var{function} must
        !          3537: have.  The machine description defines macro @code{FUNCTION_MODE} to
        !          3538: expand into the requisite mode name.  The purpose of this mode is to
        !          3539: specify what kind of addressing is allowed, on machines where the
        !          3540: allowed kinds of addressing depend on the machine mode being
        !          3541: addressed.
        !          3542: 
        !          3543: @item (clobber @var{x})
        !          3544: Represents the storing or possible storing of an unpredictable,
        !          3545: undescribed value into @var{x}, which must be a @samp{reg} or
        !          3546: @samp{mem} expression.
        !          3547: 
        !          3548: One place this is used is in string instructions that store standard
        !          3549: values into particular hard registers.  It may not be worth the
        !          3550: trouble to describe the values that are stored, but it is essential to
        !          3551: inform the compiler that the registers will be altered, lest it
        !          3552: attempt to keep data in them across the string instruction.
        !          3553: 
        !          3554: @var{x} may also be null---a null C pointer, no expression at all.
        !          3555: Such a @code{(clobber (null))} expression means that all memory
        !          3556: locations must be presumed clobbered.
        !          3557: 
        !          3558: Note that the machine description classifies certain hard registers as
        !          3559: ``call-clobbered''.  All function call instructions are assumed by
        !          3560: default to clobber these registers, so there is no need to use
        !          3561: @samp{clobber} expressions to indicate this fact.  Also, each function
        !          3562: call is assumed to have the potential to alter any memory location.
        !          3563: 
        !          3564: @item (use @var{x})
        !          3565: Represents the use of the value of @var{x}.  It indicates that the
        !          3566: value in @var{x} at this point in the program is needed, even though
        !          3567: it may not be apparent why this is so.  Therefore, the compiler will
        !          3568: not attempt to delete instructions whose only effect is to store a
        !          3569: value in @var{x}.  @var{x} must be a @samp{reg} expression.
        !          3570: 
        !          3571: @item (parallel [@var{x0} @var{x1} @dots{}])
        !          3572: Represents several side effects performed in parallel.  The square
        !          3573: brackets stand for a vector; the operand of @samp{parallel} is a
        !          3574: vector of expressions.  @var{x0}, @var{x1} and so on are individual
        !          3575: side effects---expressions of code @samp{set}, @samp{call},
        !          3576: @samp{return}, @samp{clobber} or @samp{use}.@refill
        !          3577: 
        !          3578: ``In parallel'' means that first all the values used in the individual
        !          3579: side-effects are computed, and second all the actual side-effects are
        !          3580: performed.  For example,
        !          3581: 
        !          3582: @example
        !          3583: (parallel [(set (reg:SI 1) (mem:SI (reg:SI 1)))
        !          3584:            (set (mem:SI (reg:SI 1)) (reg:SI 1))])
        !          3585: @end example
        !          3586: 
        !          3587: @noindent
        !          3588: says unambiguously that the values of hard register 1 and the memory
        !          3589: location addressed by it are interchanged.  In both places where
        !          3590: @code{(reg:SI 1)} appears as a memory address it refers to the value
        !          3591: in register 1 @emph{before} the execution of the instruction.
        !          3592: 
        !          3593: @item (sequence [@var{insns} @dots{}])
        !          3594: Represents a sequence of insns.  Each of the @var{insns} that appears
        !          3595: in the vector is suitable for appearing in the chain of insns, so it
        !          3596: must be an @samp{insn}, @samp{jump_insn}, @samp{call_insn},
        !          3597: @samp{code_label}, @samp{barrier} or @samp{note}.
        !          3598: 
        !          3599: A @samp{sequence} RTX never appears in an actual insn.  It represents
        !          3600: the sequence of insns that result from a @samp{define_expand}
        !          3601: @emph{before} those insns are passed to @code{emit_insn} to insert
        !          3602: them in the chain of insns.  When actually inserted, the individual
        !          3603: sub-insns are separated out and the @samp{sequence} is forgotten.
        !          3604: @end table
        !          3605: 
        !          3606: Three expression codes appear in place of a side effect, as the body of an
        !          3607: insn, though strictly speaking they do not describe side effects as such:
        !          3608: 
        !          3609: @table @code
        !          3610: @item (asm_input @var{s})
        !          3611: Represents literal assembler code as described by the string @var{s}.
        !          3612: 
        !          3613: @item (addr_vec:@var{m} [@var{lr0} @var{lr1} @dots{}])
        !          3614: Represents a table of jump addresses.  The vector elements @var{lr0},
        !          3615: etc., are @samp{label_ref} expressions.  The mode @var{m} specifies
        !          3616: how much space is given to each address; normally @var{m} would be
        !          3617: @code{Pmode}.
        !          3618: 
        !          3619: @item (addr_diff_vec:@var{m} @var{base} [@var{lr0} @var{lr1} @dots{}])
        !          3620: Represents a table of jump addresses expressed as offsets from
        !          3621: @var{base}.  The vector elements @var{lr0}, etc., are @samp{label_ref}
        !          3622: expressions and so is @var{base}.  The mode @var{m} specifies how much
        !          3623: space is given to each address-difference.@refill
        !          3624: @end table
        !          3625: 
        !          3626: @node Incdec, Assembler, Side Effects, RTL
        !          3627: @section Embedded Side-Effects on Addresses
        !          3628: 
        !          3629: Four special side-effect expression codes appear as memory addresses.
        !          3630: 
        !          3631: @table @code
        !          3632: @item (pre_dec:@var{m} @var{x})
        !          3633: Represents the side effect of decrementing @var{x} by a standard
        !          3634: amount and represents also the value that @var{x} has after being
        !          3635: decremented.  @var{x} must be a @samp{reg} or @samp{mem}, but most
        !          3636: machines allow only a @samp{reg}.  @var{m} must be the machine mode
        !          3637: for pointers on the machine in use.  The amount @var{x} is decremented
        !          3638: by is the length in bytes of the machine mode of the containing memory
        !          3639: reference of which this expression serves as the address.  Here is an
        !          3640: example of its use:@refill
        !          3641: 
        !          3642: @example
        !          3643: (mem:DF (pre_dec:SI (reg:SI 39)))
        !          3644: @end example
        !          3645: 
        !          3646: @noindent
        !          3647: This says to decrement pseudo register 39 by the length of a @code{DFmode}
        !          3648: value and use the result to address a @code{DFmode} value.
        !          3649: 
        !          3650: @item (pre_inc:@var{m} @var{x})
        !          3651: Similar, but specifies incrementing @var{x} instead of decrementing it.
        !          3652: 
        !          3653: @item (post_dec:@var{m} @var{x})
        !          3654: Represents the same side effect as @samp{pre_decrement} but a different
        !          3655: value.  The value represented here is the value @var{x} has @i{before}
        !          3656: being decremented.
        !          3657: 
        !          3658: @item (post_inc:@var{m} @var{x})
        !          3659: Similar, but specifies incrementing @var{x} instead of decrementing it.
        !          3660: @end table
        !          3661: 
        !          3662: These embedded side effect expressions must be used with care.  Instruction
        !          3663: patterns may not use them.  Until the @samp{flow} pass of the compiler,
        !          3664: they may occur only to represent pushes onto the stack.  The @samp{flow}
        !          3665: pass finds cases where registers are incremented or decremented in one
        !          3666: instruction and used as an address shortly before or after; these cases are
        !          3667: then transformed to use pre- or post-increment or -decrement.
        !          3668: 
        !          3669: Explicit popping of the stack could be represented with these embedded
        !          3670: side effect operators, but that would not be safe; the instruction
        !          3671: combination pass could move the popping past pushes, thus changing
        !          3672: the meaning of the code.
        !          3673: 
        !          3674: An instruction that can be represented with an embedded side effect
        !          3675: could also be represented using @samp{parallel} containing an additional
        !          3676: @samp{set} to describe how the address register is altered.  This is not
        !          3677: done because machines that allow these operations at all typically
        !          3678: allow them wherever a memory address is called for.  Describing them as
        !          3679: additional parallel stores would require doubling the number of entries
        !          3680: in the machine description.
        !          3681: 
        !          3682: @node Assembler, Insns, IncDec, RTL
        !          3683: @section Assembler Instructions as Expressions
        !          3684: 
        !          3685: The RTX code @samp{asm_operands} represents a value produced by a
        !          3686: user-specified assembler instruction.  It is used to represent
        !          3687: an @code{asm} statement with arguments.  An @code{asm} statement with
        !          3688: a single output operand, like this:
        !          3689: 
        !          3690: @example
        !          3691: asm ("foo %1,%2,%0" : "a" (outputvar) : "g" (x + y), "di" (*z));
        !          3692: @end example
        !          3693: 
        !          3694: @noindent
        !          3695: is represented using a single @samp{asm_operands} RTX which represents
        !          3696: the value that is stored in @code{outputvar}:
        !          3697: 
        !          3698: @example
        !          3699: (set @var{rtx-for-outputvar}
        !          3700:      (asm_operands "foo %1,%2,%0" "a" 0
        !          3701:                    [@var{rtx-for-addition-result} @var{rtx-for-*z}]
        !          3702:                    [(asm_input:@var{m1} "g")
        !          3703:                     (asm_input:@var{m2} "di")]))
        !          3704: @end example
        !          3705: 
        !          3706: @noindent
        !          3707: Here the operands of the @samp{asm_operands} RTX are the assembler
        !          3708: template string, the output-operand's constraint, the index-number of the
        !          3709: output operand among the output operands specified, a vector of input
        !          3710: operand RTX's, and a vector of input-operand modes and constraints.  The
        !          3711: mode @var{m1} is the mode of the sum @code{x+y}; @var{m2} is that of
        !          3712: @code{*z}.
        !          3713: 
        !          3714: When an @code{asm} statement has multiple output values, its insn has
        !          3715: several such @samp{set} RTX's inside of a @samp{parallel}.  Each @samp{set}
        !          3716: contains a @samp{asm_operands}; all of these share the same assembler
        !          3717: template and vectors, but each contains the constraint for the respective
        !          3718: output operand.  They are also distinguished by the output-operand index
        !          3719: number, which is 0, 1, @dots{} for successive output operands.
        !          3720: 
        !          3721: @node Insns, Calls, Assembler, RTL
        !          3722: @section Insns
        !          3723: 
        !          3724: The RTL representation of the code for a function is a doubly-linked
        !          3725: chain of objects called @dfn{insns}.  Insns are expressions with
        !          3726: special codes that are used for no other purpose.  Some insns are
        !          3727: actual instructions; others represent dispatch tables for @code{switch}
        !          3728: statements; others represent labels to jump to or various sorts of
        !          3729: declarative information.
        !          3730: 
        !          3731: In addition to its own specific data, each insn must have a unique id-number
        !          3732: that distinguishes it from all other insns in the current function, and
        !          3733: chain pointers to the preceding and following insns.  These three fields
        !          3734: occupy the same position in every insn, independent of the expression code
        !          3735: of the insn.  They could be accessed with @code{XEXP} and @code{XINT},
        !          3736: but instead three special macros are always used:
        !          3737: 
        !          3738: @table @code
        !          3739: @item INSN_UID (@var{i})
        !          3740: Accesses the unique id of insn @var{i}.
        !          3741: 
        !          3742: @item PREV_INSN (@var{i})
        !          3743: Accesses the chain pointer to the insn preceding @var{i}.
        !          3744: If @var{i} is the first insn, this is a null pointer.
        !          3745: 
        !          3746: @item NEXT_INSN (@var{i})
        !          3747: Accesses the chain pointer to the insn following @var{i}.
        !          3748: If @var{i} is the last insn, this is a null pointer.
        !          3749: @end table
        !          3750: 
        !          3751: The @code{NEXT_INSN} and @code{PREV_INSN} pointers must always
        !          3752: correspond: if @var{i} is not the first insn,
        !          3753: 
        !          3754: @example
        !          3755: NEXT_INSN (PREV_INSN (@var{insn})) == @var{insn}
        !          3756: @end example
        !          3757: 
        !          3758: @noindent
        !          3759: is always true.
        !          3760: 
        !          3761: Every insn has one of the following six expression codes:
        !          3762: 
        !          3763: @table @samp
        !          3764: @item insn
        !          3765: The expression code @samp{insn} is used for instructions that do not jump
        !          3766: and do not do function calls.  Insns with code @samp{insn} have four
        !          3767: additional fields beyond the three mandatory ones listed above.
        !          3768: These four are described in a table below.
        !          3769: 
        !          3770: @item jump_insn
        !          3771: The expression code @samp{jump_insn} is used for instructions that may jump
        !          3772: (or, more generally, may contain @samp{label_ref} expressions).
        !          3773: @samp{jump_insn} insns have the same extra fields as @samp{insn} insns,
        !          3774: accessed in the same way.
        !          3775: 
        !          3776: @item call_insn
        !          3777: The expression code @samp{call_insn} is used for instructions that may do
        !          3778: function calls.  It is important to distinguish these instructions because
        !          3779: they imply that certain registers and memory locations may be altered
        !          3780: unpredictably.
        !          3781: 
        !          3782: @samp{call_insn} insns have the same extra fields as @samp{insn} insns,
        !          3783: accessed in the same way.
        !          3784: 
        !          3785: @item code_label
        !          3786: A @samp{code_label} insn represents a label that a jump insn can jump to.
        !          3787: It contains one special field of data in addition to the three standard ones.
        !          3788: It is used to hold the @dfn{label number}, a number that identifies this
        !          3789: label uniquely among all the labels in the compilation (not just in the
        !          3790: current function).  Ultimately, the label is represented in the assembler
        !          3791: output as an assembler label @samp{L@var{n}} where @var{n} is the label number.
        !          3792: 
        !          3793: @item barrier
        !          3794: Barriers are placed in the instruction stream after unconditional
        !          3795: jump instructions to indicate that the jumps are unconditional.
        !          3796: They contain no information beyond the three standard fields.
        !          3797: 
        !          3798: @item note
        !          3799: @samp{note} insns are used to represent additional debugging and
        !          3800: declarative information.  They contain two nonstandard fields, an
        !          3801: integer which is accessed with the macro @code{NOTE_LINE_NUMBER} and a
        !          3802: string accessed with @code{NOTE_SOURCE_FILE}.
        !          3803: 
        !          3804: If @code{NOTE_LINE_NUMBER} is positive, the note represents the
        !          3805: position of a source line and @code{NOTE_SOURCE_FILE} is the source file name
        !          3806: that the line came from.  These notes control generation of line
        !          3807: number data in the assembler output.
        !          3808: 
        !          3809: Otherwise, @code{NOTE_LINE_NUMBER} is not really a line number but a
        !          3810: code with one of the following values (and @code{NOTE_SOURCE_FILE}
        !          3811: must contain a null pointer):
        !          3812: 
        !          3813: @table @code
        !          3814: @item NOTE_INSN_DELETED
        !          3815: Such a note is completely ignorable.  Some passes of the compiler
        !          3816: delete insns by altering them into notes of this kind.
        !          3817: 
        !          3818: @item NOTE_INSN_BLOCK_BEG
        !          3819: @itemx NOTE_INSN_BLOCK_END
        !          3820: These types of notes indicate the position of the beginning and end
        !          3821: of a level of scoping of variable names.  They control the output
        !          3822: of debugging information.
        !          3823: 
        !          3824: @item NOTE_INSN_LOOP_BEG
        !          3825: @itemx NOTE_INSN_LOOP_END
        !          3826: These types of notes indicate the position of the beginning and end
        !          3827: of a @code{while} or @code{for} loop.  They enable the loop optimizer
        !          3828: to find loops quickly.
        !          3829: @end table
        !          3830: @end table
        !          3831: 
        !          3832: Here is a table of the extra fields of @samp{insn}, @samp{jump_insn}
        !          3833: and @samp{call_insn} insns:
        !          3834: 
        !          3835: @table @code
        !          3836: @item PATTERN (@var{i})
        !          3837: An expression for the side effect performed by this insn.
        !          3838: 
        !          3839: @item REG_NOTES (@var{i})
        !          3840: A list (chain of @samp{expr_list} expressions) giving information
        !          3841: about the usage of registers in this insn.  This list is set up by the
        !          3842: flow analysis pass; it is a null pointer until then.
        !          3843: 
        !          3844: @item LOG_LINKS (@var{i})
        !          3845: A list (chain of @samp{insn_list} expressions) of previous ``related''
        !          3846: insns: insns which store into registers values that are used for the
        !          3847: first time in this insn.  (An additional constraint is that neither a
        !          3848: jump nor a label may come between the related insns).  This list is
        !          3849: set up by the flow analysis pass; it is a null pointer until then.
        !          3850: 
        !          3851: @item INSN_CODE (@var{i})
        !          3852: An integer that says which pattern in the machine description matches
        !          3853: this insn, or -1 if the matching has not yet been attempted.
        !          3854: 
        !          3855: Such matching is never attempted and this field is not used on an insn
        !          3856: whose pattern consists of a single @samp{use}, @samp{clobber},
        !          3857: @samp{asm}, @samp{addr_vec} or @samp{addr_diff_vec} expression.
        !          3858: @end table
        !          3859: 
        !          3860: The @code{LOG_LINKS} field of an insn is a chain of @samp{insn_list}
        !          3861: expressions.  Each of these has two operands: the first is an insn,
        !          3862: and the second is another @samp{insn_list} expression (the next one in
        !          3863: the chain).  The last @samp{insn_list} in the chain has a null pointer
        !          3864: as second operand.  The significant thing about the chain is which
        !          3865: insns appear in it (as first operands of @samp{insn_list}
        !          3866: expressions).  Their order is not significant.
        !          3867: 
        !          3868: The @code{REG_NOTES} field of an insn is a similar chain but of
        !          3869: @samp{expr_list} expressions instead of @samp{insn_list}.  There are four
        !          3870: kinds of register notes, which are distinguished by the machine mode of the
        !          3871: @samp{expr_list}, which a register note is really understood as being an
        !          3872: @code{enum reg_note}.  The first operand @var{op} of the @samp{expr_list}
        !          3873: is data whose meaning depends on the kind of note.  Here are the four
        !          3874: kinds:
        !          3875: 
        !          3876: @table @code
        !          3877: @item REG_DEAD
        !          3878: The register @var{op} dies in this insn; that is to say, altering the
        !          3879: value immediately after this insn would not affect the future behavior
        !          3880: of the program.
        !          3881: 
        !          3882: @item REG_INC
        !          3883: The register @var{op} is incremented (or decremented; at this level
        !          3884: there is no distinction) by an embedded side effect inside this insn.
        !          3885: This means it appears in a @code{POST_INC}, @code{PRE_INC},
        !          3886: @code{POST_DEC} or @code{PRE_DEC} RTX.
        !          3887: 
        !          3888: @item REG_EQUIV
        !          3889: The register that is set by this insn will be equal to @var{op} at run
        !          3890: time, and could validly be replaced in all its occurrences by
        !          3891: @var{op}.  (``Validly'' here refers to the data flow of the program;
        !          3892: simple replacement may make some insns invalid.)
        !          3893: 
        !          3894: The value which the insn explicitly copies into the register may look
        !          3895: different from @var{op}, but they will be equal at run time.
        !          3896: 
        !          3897: For example, when a constant is loaded into a register that is never
        !          3898: assigned any other value, this kind of note is used.
        !          3899: 
        !          3900: When a parameter is copied into a pseudo-register at entry to a function,
        !          3901: a note of this kind records that the register is equivalent to the stack
        !          3902: slot where the parameter was passed.  Although in this case the register
        !          3903: may be set by other insns, it is still valid to replace the register
        !          3904: by the stack slot throughout the function.
        !          3905: 
        !          3906: @item REG_EQUAL
        !          3907: The register that is set by this insn will be equal to @var{op} at run
        !          3908: time at the end of this insn (but not necessarily elsewhere in the
        !          3909: function).
        !          3910: 
        !          3911: The RTX @var{op} is typically an arithmetic expression.  For example,
        !          3912: when a sequence of insns such as a library call is used to perform an
        !          3913: arithmetic operation, this kind of note is attached to the insn that
        !          3914: produces or copies the final value.  It tells the CSE pass how to
        !          3915: think of that value.
        !          3916: 
        !          3917: @item REG_RETVAL
        !          3918: This insn copies the value of a library call, and @var{op} is the
        !          3919: first insn that was generated to set up the arguments for the library
        !          3920: call.
        !          3921: 
        !          3922: Flow analysis uses this note to delete all of a library call whose
        !          3923: result is dead.
        !          3924: 
        !          3925: @item REG_WAS_0
        !          3926: The register @var{op} contained zero before this insn.  You can rely
        !          3927: on this note if it is present; its absence implies nothing.
        !          3928: @end table
        !          3929: 
        !          3930: (The only difference between the expression codes @samp{insn_list} and
        !          3931: @samp{expr_list} is that the first operand of an @samp{insn_list} is
        !          3932: assumed to be an insn and is printed in debugging dumps as the insn's
        !          3933: unique id; the first operand of an @samp{expr_list} is printed in the
        !          3934: ordinary way as an expression.)
        !          3935: 
        !          3936: @node Calls, Sharing, Insns, RTL
        !          3937: @section RTL Representation of Function-Call Insns
        !          3938: 
        !          3939: Insns that call subroutines have the RTL expression code @samp{call_insn}.
        !          3940: These insns must satisfy special rules, and their bodies must use a special
        !          3941: RTL expression code, @samp{call}.
        !          3942: 
        !          3943: A @samp{call} expression has two operands, as follows:
        !          3944: 
        !          3945: @example
        !          3946: (call @var{nbytes} (mem:@var{fm} @var{addr}))
        !          3947: @end example
        !          3948: 
        !          3949: @noindent
        !          3950: Here @var{nbytes} is an operand that represents the number of bytes of
        !          3951: argument data being passed to the subroutine, @var{fm} is a machine mode
        !          3952: (which must equal as the definition of the @code{FUNCTION_MODE} macro in
        !          3953: the machine description) and @var{addr} represents the address of the
        !          3954: subroutine.
        !          3955: 
        !          3956: For a subroutine that returns no value, the @samp{call} RTX as shown above
        !          3957: is the entire body of the insn.
        !          3958: 
        !          3959: For a subroutine that returns a value whose mode is not @code{BLKmode},
        !          3960: the value is returned in a hard register.  If this register's number is
        !          3961: @var{r}, then the body of the call insn looks like this:
        !          3962: 
        !          3963: @example
        !          3964: (set (reg:@var{m} @var{r})
        !          3965:      (call @var{nbytes} (mem:@var{fm} @var{addr})))
        !          3966: @end example
        !          3967: 
        !          3968: @noindent
        !          3969: This RTL expression makes it clear (to the optimizer passes) that the
        !          3970: appropriate register receives a useful value in this insn.
        !          3971: 
        !          3972: Immediately after RTL generation, if the value of the subroutine is
        !          3973: actually used, this call insn is always followed closely by an insn which
        !          3974: refers to the register @var{r}.  This remains true through all the
        !          3975: optimizer passes until cross jumping occurs.
        !          3976: 
        !          3977: The following insn has one of two forms.  Either it copies the value into a
        !          3978: pseudo-register, like this:
        !          3979: 
        !          3980: @example
        !          3981: (set (reg:@var{m} @var{p}) (reg:@var{m} @var{r}))
        !          3982: @end example
        !          3983: 
        !          3984: @noindent
        !          3985: or (in the case where the calling function will simply return whatever
        !          3986: value the call produced, and no operation is needed to do this):
        !          3987: 
        !          3988: @example
        !          3989: (use (reg:@var{m} @var{r}))
        !          3990: @end example
        !          3991: 
        !          3992: @noindent
        !          3993: Between the call insn and this following insn there may intervene only a
        !          3994: stack-adjustment insn (and perhaps some @samp{note} insns).
        !          3995: 
        !          3996: When a subroutine returns a @code{BLKmode} value, it is handled by
        !          3997: passing to the subroutine the address of a place to store the value.
        !          3998: So the call insn itself does not ``return'' any value, and it has the
        !          3999: same RTL form as a call that returns nothing.
        !          4000: 
        !          4001: @node Sharing,, Calls, RTL
        !          4002: @section Structure Sharing Assumptions
        !          4003: 
        !          4004: The compiler assumes that certain kinds of RTL expressions are unique;
        !          4005: there do not exist two distinct objects representing the same value.
        !          4006: In other cases, it makes an opposite assumption: that no RTL expression
        !          4007: object of a certain kind appears in more than one place in the
        !          4008: containing structure.
        !          4009: 
        !          4010: These assumptions refer to a single function; except for the RTL
        !          4011: objects that describe global variables and external functions,
        !          4012: no RTL objects are common to two functions.
        !          4013: 
        !          4014: @itemize @bullet
        !          4015: @item
        !          4016: Each pseudo-register has only a single @samp{reg} object to represent it,
        !          4017: and therefore only a single machine mode.
        !          4018: 
        !          4019: @item
        !          4020: For any symbolic label, there is only one @samp{symbol_ref} object
        !          4021: referring to it.
        !          4022: 
        !          4023: @item
        !          4024: There is only one @samp{const_int} expression with value zero,
        !          4025: and only one with value one.
        !          4026: 
        !          4027: @item
        !          4028: There is only one @samp{pc} expression.
        !          4029: 
        !          4030: @item
        !          4031: There is only one @samp{cc0} expression.
        !          4032: 
        !          4033: @item
        !          4034: There is only one @samp{const_double} expression with mode
        !          4035: @code{SFmode} and value zero, and only one with mode @code{DFmode} and
        !          4036: value zero.
        !          4037: 
        !          4038: @item
        !          4039: No @samp{label_ref} appears in more than one place in the RTL
        !          4040: structure; in other words, it is safe to do a tree-walk of all the
        !          4041: insns in the function and assume that each time a @samp{label_ref} is
        !          4042: seen it is distinct from all others that are seen.
        !          4043: 
        !          4044: @item
        !          4045: Only one @samp{mem} object is normally created for each static
        !          4046: variable or stack slot, so these objects are frequently shared in all
        !          4047: the places they appear.  However, separate but equal objects for these
        !          4048: variables are occasionally made.
        !          4049: 
        !          4050: @item
        !          4051: No RTL object appears in more than one place in the RTL structure
        !          4052: except as described above.  Many passes of the compiler rely on this
        !          4053: by assuming that they can modify RTL objects in place without unwanted
        !          4054: side-effects on other insns.
        !          4055: 
        !          4056: @item
        !          4057: During initial RTL generation, shared structure is freely introduced.
        !          4058: After all the RTL for a function has been generated, all shared
        !          4059: structure is copied by @code{unshare_all_rtl} in @file{emit-rtl.c},
        !          4060: after which the above rules are guaranteed to be followed.
        !          4061: 
        !          4062: @item
        !          4063: During the combiner pass, shared structure with an insn can exist
        !          4064: temporarily.  However, the shared structure is copied before the
        !          4065: combiner is finished with the insn.  This is done by
        !          4066: @code{copy_substitutions} in @samp{combine.c}.
        !          4067: @end itemize
        !          4068: 
        !          4069: @node Machine Desc, Machine Macros, RTL, Top
        !          4070: @chapter Machine Descriptions
        !          4071: 
        !          4072: A machine description has two parts: a file of instruction patterns
        !          4073: (@file{.md} file) and a C header file of macro definitions.
        !          4074: 
        !          4075: The @file{.md} file for a target machine contains a pattern for each
        !          4076: instruction that the target machine supports (or at least each instruction
        !          4077: that is worth telling the compiler about).  It may also contain comments.
        !          4078: A semicolon causes the rest of the line to be a comment, unless the semicolon
        !          4079: is inside a quoted string.
        !          4080: 
        !          4081: See the next chapter for information on the C header file.
        !          4082: 
        !          4083: @menu
        !          4084: * Patterns::            How to write instruction patterns.
        !          4085: * Example::             An explained example of a @samp{define_insn} pattern.
        !          4086: * RTL Template::        The RTL template defines what insns match a pattern.
        !          4087: * Output Template::     The output template says how to make assembler code
        !          4088:                           from such an insn.
        !          4089: * Output Statement::    For more generality, write C code to output 
        !          4090:                           the assembler code.
        !          4091: * Constraints::         When not all operands are general operands.
        !          4092: * Standard Names::      Names mark patterns to use for code generation.
        !          4093: * Pattern Ordering::    When the order of patterns makes a difference.
        !          4094: * Dependent Patterns::  Having one pattern may make you need another.
        !          4095: * Jump Patterns::       Special considerations for patterns for jump insns.
        !          4096: * Peephole Definitions::Defining machine-specific peephole optimizations.
        !          4097: * Expander Definitions::Generating a sequence of several RTL insns
        !          4098:                          for a standard operation.
        !          4099: @end menu
        !          4100: 
        !          4101: @node Patterns, Example, Machine Desc, Machine Desc
        !          4102: @section Everything about Instruction Patterns
        !          4103: 
        !          4104: Each instruction pattern contains an incomplete RTL expression, with pieces
        !          4105: to be filled in later, operand constraints that restrict how the pieces can
        !          4106: be filled in, and an output pattern or C code to generate the assembler
        !          4107: output, all wrapped up in a @samp{define_insn} expression.
        !          4108: 
        !          4109: A @samp{define_insn} is an RTL expression containing four operands:
        !          4110: 
        !          4111: @enumerate
        !          4112: @item
        !          4113: An optional name.  The presence of a name indicate that this instruction
        !          4114: pattern can perform a certain standard job for the RTL-generation
        !          4115: pass of the compiler.  This pass knows certain names and will use
        !          4116: the instruction patterns with those names, if the names are defined
        !          4117: in the machine description.
        !          4118: 
        !          4119: The absence of a name is indicated by writing an empty string
        !          4120: where the name should go.  Nameless instruction patterns are never
        !          4121: used for generating RTL code, but they may permit several simpler insns
        !          4122: to be combined later on.
        !          4123: 
        !          4124: Names that are not thus known and used in RTL-generation have no
        !          4125: effect; they are equivalent to no name at all.
        !          4126: 
        !          4127: @item
        !          4128: The @dfn{RTL template} (@pxref{RTL Template}) is a vector of
        !          4129: incomplete RTL expressions which show what the instruction should look
        !          4130: like.  It is incomplete because it may contain @samp{match_operand}
        !          4131: and @samp{match_dup} expressions that stand for operands of the
        !          4132: instruction.
        !          4133: 
        !          4134: If the vector has only one element, that element is what the
        !          4135: instruction should look like.  If the vector has multiple elements,
        !          4136: then the instruction looks like a @samp{parallel} expression
        !          4137: containing that many elements as described.
        !          4138: 
        !          4139: @item
        !          4140: A condition.  This is a string which contains a C expression that is
        !          4141: the final test to decide whether an insn body matches this pattern.
        !          4142: 
        !          4143: For a named pattern, the condition (if present) may not depend on
        !          4144: the data in the insn being matched, but only the target-machine-type
        !          4145: flags.  The compiler needs to test these conditions during
        !          4146: initialization in order to learn exactly which named instructions are
        !          4147: available in a particular run.
        !          4148: 
        !          4149: For nameless patterns, the condition is applied only when matching an
        !          4150: individual insn, and only after the insn has matched the pattern's
        !          4151: recognition template.  The insn's operands may be found in the vector
        !          4152: @code{operands}.
        !          4153: 
        !          4154: @item
        !          4155: The @dfn{output template}: a string that says how to output matching
        !          4156: insns as assembler code.  @samp{%} in this string specifies where
        !          4157: to substitute the value of an operand.  @xref{Output Template}.
        !          4158: 
        !          4159: When simple substitution isn't general enough, you can specify a piece
        !          4160: of C code to compute the output.  @xref{Output Statement}.
        !          4161: @end enumerate
        !          4162: 
        !          4163: @node Example, RTL Template, Patterns, Machine Desc
        !          4164: @section Example of @samp{define_insn}
        !          4165: 
        !          4166: Here is an actual example of an instruction pattern, for the 68000/68020.
        !          4167: 
        !          4168: @example
        !          4169: (define_insn "tstsi"
        !          4170:   [(set (cc0)
        !          4171:         (match_operand:SI 0 "general_operand" "rm"))]
        !          4172:   ""
        !          4173:   "*
        !          4174: @{ if (TARGET_68020 || ! ADDRESS_REG_P (operands[0]))
        !          4175:     return \"tstl %0\";
        !          4176:   return \"cmpl #0,%0\"; @}")
        !          4177: @end example
        !          4178: 
        !          4179: This is an instruction that sets the condition codes based on the value of
        !          4180: a general operand.  It has no condition, so any insn whose RTL description
        !          4181: has the form shown may be handled according to this pattern.  The name
        !          4182: @samp{tstsi} means ``test a @code{SImode} value'' and tells the RTL generation
        !          4183: pass that, when it is necessary to test such a value, an insn to do so
        !          4184: can be constructed using this pattern.
        !          4185: 
        !          4186: The output control string is a piece of C code which chooses which
        !          4187: output template to return based on the kind of operand and the specific
        !          4188: type of CPU for which code is being generated.
        !          4189: 
        !          4190: @samp{"rm"} is an operand constraint.  Its meaning is explained below.
        !          4191: 
        !          4192: @node RTL Template, Output Template, Example, Machine Desc
        !          4193: @section RTL Template for Generating and Recognizing Insns
        !          4194: 
        !          4195: The RTL template is used to define which insns match the particular pattern
        !          4196: and how to find their operands.  For named patterns, the RTL template also
        !          4197: says how to construct an insn from specified operands.
        !          4198: 
        !          4199: Construction involves substituting specified operands into a copy of the
        !          4200: template.  Matching involves determining the values that serve as the
        !          4201: operands in the insn being matched.  Both of these activities are
        !          4202: controlled by special expression types that direct matching and
        !          4203: substitution of the operands.
        !          4204: 
        !          4205: @table @code
        !          4206: @item (match_operand:@var{m} @var{n} @var{testfn} @var{constraint})
        !          4207: This expression is a placeholder for operand number @var{n} of
        !          4208: the insn.  When constructing an insn, operand number @var{n}
        !          4209: will be substituted at this point.  When matching an insn, whatever
        !          4210: appears at this position in the insn will be taken as operand
        !          4211: number @var{n}; but it must satisfy @var{testfn} or this instruction
        !          4212: pattern will not match at all.
        !          4213: 
        !          4214: Operand numbers must be chosen consecutively counting from zero in
        !          4215: each instruction pattern.  There may be only one @samp{match_operand}
        !          4216: expression in the pattern for each expression number, and they must
        !          4217: appear in order of increasing expression number.
        !          4218: 
        !          4219: @var{testfn} is a string that is the name of a C function that accepts
        !          4220: two arguments, a machine mode and an expression.  During matching,
        !          4221: the function will be called with @var{m} as the mode argument
        !          4222: and the putative operand as the other argument.  If it returns zero,
        !          4223: this instruction pattern fails to match.  @var{testfn} may be
        !          4224: an empty string; then it means no test is to be done on the operand.
        !          4225: 
        !          4226: Most often, @var{testfn} is @code{"general_operand"}.  It checks
        !          4227: that the putative operand is either a constant, a register or a
        !          4228: memory reference, and that it is valid for mode @var{m}.
        !          4229: 
        !          4230: For an operand that must be a register, @var{testfn} should be
        !          4231: @code{"register_operand"}.  This prevents GNU CC from creating insns
        !          4232: that have memory references in these operands, insns which would only
        !          4233: have to be taken apart in the reload pass.
        !          4234: 
        !          4235: For an operand that must be a constant, either @var{testfn} should be
        !          4236: @code{"immediate_operand"}, or the instruction pattern's extra condition
        !          4237: should check for constants, or both.
        !          4238: 
        !          4239: @var{constraint} is explained later (@pxref{Constraints}).
        !          4240: 
        !          4241: @item (match_dup @var{n})
        !          4242: This expression is also a placeholder for operand number @var{n}.
        !          4243: It is used when the operand needs to appear more than once in the
        !          4244: insn.
        !          4245: 
        !          4246: In construction, @samp{match_dup} behaves exactly like
        !          4247: @samp{match_operand}: the operand is substituted into the insn being
        !          4248: constructed.  But in matching, @samp{match_dup} behaves differently.
        !          4249: It assumes that operand number @var{n} has already been determined by
        !          4250: a @samp{match_operand} appearing earlier in the recognition template,
        !          4251: and it matches only an identical-looking expression.
        !          4252: 
        !          4253: @item (address (match_operand:@var{m} @var{n} "address_operand" ""))
        !          4254: This complex of expressions is a placeholder for an operand number
        !          4255: @var{n} in a ``load address'' instruction: an operand which specifies
        !          4256: a memory location in the usual way, but for which the actual operand
        !          4257: value used is the address of the location, not the contents of the
        !          4258: location.
        !          4259: 
        !          4260: @samp{address} expressions never appear in RTL code, only in machine
        !          4261: descriptions.  And they are used only in machine descriptions that do
        !          4262: not use the operand constraint feature.  When operand constraints are
        !          4263: in use, the letter @samp{p} in the constraint serves this purpose.
        !          4264: 
        !          4265: @var{m} is the machine mode of the @emph{memory location being
        !          4266: addressed}, not the machine mode of the address itself.  That mode is
        !          4267: always the same on a given target machine (it is @code{Pmode}, which
        !          4268: normally is @code{SImode}), so there is no point in mentioning it;
        !          4269: thus, no machine mode is written in the @samp{address} expression.  If
        !          4270: some day support is added for machines in which addresses of different
        !          4271: kinds of objects appear differently or are used differently (such as
        !          4272: the PDP-10), different formats would perhaps need different machine
        !          4273: modes and these modes might be written in the @samp{address}
        !          4274: expression.
        !          4275: @end table
        !          4276: 
        !          4277: @node Output Template, Output Statement, RTL Template, Machine Desc
        !          4278: @section Output Templates and Operand Substitution
        !          4279: 
        !          4280: The @dfn{output template} is a string which specifies how to output
        !          4281: the assembler code for an instruction pattern.  Most of the template
        !          4282: is a fixed string which is output literally.  The character @samp{%}
        !          4283: is used to specify where to substitute an operand; it can also be
        !          4284: used to identify places different variants of the assembler require
        !          4285: different syntax.
        !          4286: 
        !          4287: In the simplest case, a @samp{%} followed by a digit @var{n} says to output
        !          4288: operand @var{n} at that point in the string.
        !          4289: 
        !          4290: @samp{%} followed by a letter and a digit says to output an operand in an
        !          4291: alternate fashion.  Four letters have standard, built-in meanings described
        !          4292: below.  The machine description macro @code{PRINT_OPERAND} can define
        !          4293: additional letters with nonstandard meanings.
        !          4294: 
        !          4295: @samp{%c@var{digit}} can be used to substitute an operand that is a
        !          4296: constant value without the syntax that normally indicates an immediate
        !          4297: operand.
        !          4298: 
        !          4299: @samp{%n@var{digit}} is like @samp{%c@var{digit}} except that the value of
        !          4300: the constant is negated before printing.
        !          4301: 
        !          4302: @samp{%a@var{digit}} can be used to substitute an operand as if it were a
        !          4303: memory reference, with the actual operand treated as the address.  This may
        !          4304: be useful when outputting a ``load address'' instruction, because often the
        !          4305: assembler syntax for such an instruction requires you to write the operand
        !          4306: as if it were a memory reference.
        !          4307: 
        !          4308: @samp{%l@var{digit}} is used to substitute a @code{label_ref} into a jump
        !          4309: instruction.
        !          4310: 
        !          4311: @samp{%} followed by a punctuation character specifies a substitution that
        !          4312: does not use an operand.  Only one case is standard: @samp{%%} outputs a
        !          4313: @samp{%} into the assembler code.  Other nonstandard cases can be
        !          4314: defined in the @code{PRINT_OPERAND} macro.
        !          4315: 
        !          4316: The template may generate multiple assembler instructions.  Write the text
        !          4317: for the instructions, with @samp{\;} between them.
        !          4318: 
        !          4319: When the RTL contains two operand which are required by constraint to match
        !          4320: each other, the output template must refer only to the lower-numbered operand.
        !          4321: Matching operands are not always identical, and the rest of the compiler
        !          4322: arranges to put the proper RTL expression for printing into the lower-numbered
        !          4323: operand.
        !          4324: 
        !          4325: One use of nonstandard letters or punctuation following @samp{%} is to
        !          4326: distinguish between different assembler languages for the same machine; for
        !          4327: example, Motorola syntax versus MIT syntax for the 68000.  Motorola syntax
        !          4328: requires periods in most opcode names, while MIT syntax does not.  For
        !          4329: example, the opcode @samp{movel} in MIT syntax is @samp{move.l} in Motorola
        !          4330: syntax.  The same file of patterns is used for both kinds of output syntax,
        !          4331: but the character sequence @samp{%.} is used in each place where Motorola
        !          4332: syntax wants a period.  The @code{PRINT_OPERAND} macro for Motorola syntax
        !          4333: defines the sequence to output a period; the macro for MIT syntax defines
        !          4334: it to do nothing.
        !          4335: 
        !          4336: @node Output Statement, Constraints, Output Template, Machine Desc
        !          4337: @section C Statements for Generating Assembler Output
        !          4338: 
        !          4339: Often a single fixed template string cannot produce correct and efficient
        !          4340: assembler code for all the cases that are recognized by a single
        !          4341: instruction pattern.  For example, the opcodes may depend on the kinds of
        !          4342: operands; or some unfortunate combinations of operands may require extra
        !          4343: machine instructions.
        !          4344: 
        !          4345: If the output control string starts with a @samp{*}, then it is not an
        !          4346: output template but rather a piece of C program that should compute a
        !          4347: template.  It should execute a @code{return} statement to return the
        !          4348: template-string you want.  Most such templates use C string literals, which
        !          4349: require doublequote characters to delimit them.  To include these
        !          4350: doublequote characters in the string, prefix each one with @samp{\}.
        !          4351: 
        !          4352: The operands may be found in the array @code{operands}, whose C data type
        !          4353: is @code{rtx []}.
        !          4354: 
        !          4355: It is possible to output an assembler instruction and then go on to output
        !          4356: or compute more of them, using the subroutine @code{output_asm_insn}.  This
        !          4357: receives two arguments: a template-string and a vector of operands.  The
        !          4358: vector may be @code{operands}, or it may be another array of @code{rtx}
        !          4359: that you declare locally and initialize yourself.
        !          4360: 
        !          4361: When an insn pattern has multiple alternatives in its constraints, often
        !          4362: the appearance of the assembler code determined mostly by which alternative
        !          4363: was matched.  When this is so, the C code can test the variable
        !          4364: @code{which_alternative}, which is the ordinal number of the alternative
        !          4365: that was actually satisfied (0 for the first, 1 for the second alternative,
        !          4366: etc.).
        !          4367: 
        !          4368: For example, suppose there are two opcodes for storing zero, @samp{clrreg}
        !          4369: for registers and @samp{clrmem} for memory locations.  Here is how
        !          4370: a pattern could use @code{which_alternative} to choose between them:
        !          4371: 
        !          4372: @example
        !          4373: (define_insn ""
        !          4374:   [(set (match_operand:SI 0 "general_operand" "r,m")
        !          4375:         (const_int 0))]
        !          4376:   ""
        !          4377:   "*
        !          4378:   return (which_alternative == 0
        !          4379:           ? \"clrreg %0\" : \"clrmem %0\");
        !          4380:   ")
        !          4381: @end example
        !          4382: 
        !          4383: @node Constraints, Standard Names, Output Statement, Machine Desc
        !          4384: @section Operand Constraints
        !          4385: 
        !          4386: Each @samp{match_operand} in an instruction pattern can specify a
        !          4387: constraint for the type of operands allowed.  Constraints can say whether
        !          4388: an operand may be in a register, and which kinds of register; whether the
        !          4389: operand can be a memory reference, and which kinds of address; whether the
        !          4390: operand may be an immediate constant, and which possible values it may
        !          4391: have.  Constraints can also require two operands to match.
        !          4392: 
        !          4393: @menu
        !          4394: * Simple Constraints::  Basic use of constraints.
        !          4395: * Multi-Alternative::   When an insn has two alternative constraint-patterns.
        !          4396: * Class Preferences::   Constraints guide which hard register to put things in.
        !          4397: * Modifiers::           More precise control over effects of constraints.
        !          4398: * No Constraints::      Describing a clean machine without constraints.
        !          4399: @end menu
        !          4400: 
        !          4401: @node Simple Constraints, Multi-Alternative, Constraints, Constraints
        !          4402: @subsection Simple Constraints
        !          4403: 
        !          4404: The simplest kind of constraint is a string full of letters, each of
        !          4405: which describes one kind of operand that is permitted.  Here are
        !          4406: the letters that are allowed:
        !          4407: 
        !          4408: @table @asis
        !          4409: @item @samp{m}
        !          4410: A memory operand is allowed, with any kind of address that the machine
        !          4411: supports in general.
        !          4412: 
        !          4413: @item @samp{o}
        !          4414: A memory operand is allowed, but only if the address is
        !          4415: @dfn{offsetable}.  This means that adding a small integer (actually,
        !          4416: the width in bytes of the operand, as determined by its machine mode)
        !          4417: may be added to the address and the result is also a valid memory
        !          4418: address.
        !          4419: 
        !          4420: For example, an address which is constant is offsetable; so is an
        !          4421: address that is the sum of a register and a constant (as long as a
        !          4422: slightly larger constant is also within the range of address-offsets
        !          4423: supported by the machine); but an autoincrement or autodecrement
        !          4424: address is not offsetable.  More complicated indirect/indexed
        !          4425: addresses may or may not be offsetable depending on the other
        !          4426: addressing modes that the machine supports.
        !          4427: 
        !          4428: Note that in an output operand which can be matched by another
        !          4429: operand, the constraint letter @samp{o} is valid only when accompanied
        !          4430: by both @samp{<} (if the target machine has predecrement addressing)
        !          4431: and @samp{>} (if the target machine has preincrement addressing).
        !          4432: 
        !          4433: @item @samp{<}
        !          4434: A memory operand with autodecrement addressing (either predecrement or
        !          4435: postdecrement) is allowed.
        !          4436: 
        !          4437: @item @samp{>}
        !          4438: A memory operand with autoincrement addressing (either preincrement or
        !          4439: postincrement) is allowed.
        !          4440: 
        !          4441: @item @samp{r}
        !          4442: A register operand is allowed provided that it is in a general
        !          4443: register.
        !          4444: 
        !          4445: @item @samp{d}, @samp{a}, @samp{f}, @dots{}
        !          4446: Other letters can be defined in machine-dependent fashion to stand for
        !          4447: particular classes of registers.  @samp{d}, @samp{a} and @samp{f} are
        !          4448: defined on the 68000/68020 to stand for data, address and floating
        !          4449: point registers.
        !          4450: 
        !          4451: @item @samp{i}
        !          4452: An immediate integer operand (one with constant value) is allowed.
        !          4453: This includes symbolic constants whose values will be known only at
        !          4454: assembly time.
        !          4455: 
        !          4456: @item @samp{n}
        !          4457: An immediate integer operand with a known numeric value is allowed.
        !          4458: Many systems cannot support assembly-time constants for operands less
        !          4459: than a word wide.  Constraints for these operands should use @samp{n}
        !          4460: rather than @samp{i}.
        !          4461: 
        !          4462: @item @samp{I}, @samp{J}, @samp{K}, @dots{}
        !          4463: Other letters in the range @samp{I} through @samp{M} may be defined in
        !          4464: a machine-dependent fashion to permit immediate integer operands with
        !          4465: explicit integer values in specified ranges.  For example, on the
        !          4466: 68000, @samp{I} is defined to stand for the range of values 1 to 8.
        !          4467: This is the range permitted as a shift count in the shift
        !          4468: instructions.
        !          4469: 
        !          4470: @item @samp{F}
        !          4471: An immediate floating operand (expression code @samp{const_double}) is
        !          4472: allowed.
        !          4473: 
        !          4474: @item @samp{G}, @samp{H}
        !          4475: @samp{G} and @samp{H} may be defined in a machine-dependent fashion to
        !          4476: permit immediate floating operands in particular ranges of values.
        !          4477: 
        !          4478: @item @samp{s}
        !          4479: An immediate integer operand whose value is not an explicit integer is
        !          4480: allowed.
        !          4481: 
        !          4482: This might appear strange; if an insn allows a constant operand with a
        !          4483: value not known at compile time, it certainly must allow any known
        !          4484: value.  So why use @samp{s} instead of @samp{i}?  Sometimes it allows
        !          4485: better code to be generated.
        !          4486: 
        !          4487: For example, on the 68000 in a fullword instruction it is possible to
        !          4488: use an immediate operand; but if the immediate value is between -32
        !          4489: and 31, better code results from loading the value into a register and
        !          4490: using the register.  This is because the load into the register can be
        !          4491: done with a @samp{moveq} instruction.  We arrange for this to happen
        !          4492: by defining the letter @samp{K} to mean ``any integer outside the
        !          4493: range -32 to 31'', and then specifying @samp{Ks} in the operand
        !          4494: constraints.
        !          4495: 
        !          4496: @item @samp{g}
        !          4497: Any register, memory or immediate integer operand is allowed, except for
        !          4498: registers that are not general registers.
        !          4499: 
        !          4500: @item @samp{@var{n}} (a digit)
        !          4501: An operand that matches operand number @var{n} is allowed.
        !          4502: If a digit is used together with letters, the digit should come last.
        !          4503: 
        !          4504: This is called a @dfn{matching constraint} and what it really means is
        !          4505: that the assembler has only a single operand that fills two roles
        !          4506: considered separate in the RTL insn.  For example, an add insn has two
        !          4507: input operands and one output operand in the RTL, but on most machines
        !          4508: an add instruction really has only two operands, one of them an
        !          4509: input-output operand.
        !          4510: 
        !          4511: Matching constraints work only in circumstances like that add insn.
        !          4512: More precisely, the matching constraint must appear in an input-only
        !          4513: operand and the operand that it matches must be an output-only operand
        !          4514: with a lower number.
        !          4515: 
        !          4516: For operands to match in a particular case usually means that they
        !          4517: are identical-looking RTL expressions.  But in a few special cases
        !          4518: specific kinds of dissimilarity are allowed.  For example, @code{*x}
        !          4519: as an input operand will match @code{*x++} as an output operand.
        !          4520: For proper results in such cases, the output template should always
        !          4521: use the output-operand's number when printing the operand.
        !          4522: 
        !          4523: @item @samp{p}
        !          4524: An operand that is a valid memory address is allowed.  This is
        !          4525: for ``load address'' and ``push address'' instructions.
        !          4526: 
        !          4527: If @samp{p} is used in the constraint, the test-function in the
        !          4528: @samp{match_operand} must be @code{address_operand}.
        !          4529: @end table
        !          4530: 
        !          4531: In order to have valid assembler code, each operand must satisfy
        !          4532: its constraint.  But a failure to do so does not prevent the pattern
        !          4533: from applying to an insn.  Instead, it directs the compiler to modify
        !          4534: the code so that the constraint will be satisfied.  Usually this is
        !          4535: done by copying an operand into a register.
        !          4536: 
        !          4537: Contrast, therefore, the two instruction patterns that follow:
        !          4538: 
        !          4539: @example
        !          4540: (define_insn ""
        !          4541:   [(set (match_operand:SI 0 "general_operand" "r")
        !          4542:         (plus:SI (match_dup 0)
        !          4543:                  (match_operand:SI 1 "general_operand" "r")))]
        !          4544:   ""
        !          4545:   "@dots{}")
        !          4546: @end example
        !          4547: 
        !          4548: @noindent
        !          4549: which has two operands, one of which must appear in two places, and
        !          4550: 
        !          4551: @example
        !          4552: (define_insn ""
        !          4553:   [(set (match_operand:SI 0 "general_operand" "r")
        !          4554:         (plus:SI (match_operand:SI 1 "general_operand" "0")
        !          4555:                  (match_operand:SI 2 "general_operand" "r")))]
        !          4556:   ""
        !          4557:   "@dots{}")
        !          4558: @end example
        !          4559: 
        !          4560: @noindent
        !          4561: which has three operands, two of which are required by a constraint to be
        !          4562: identical.  If we are considering an insn of the form
        !          4563: 
        !          4564: @example
        !          4565: (insn @var{n} @var{prev} @var{next}
        !          4566:   (set (reg:SI 3)
        !          4567:        (plus:SI (reg:SI 6) (reg:SI 109)))
        !          4568:   @dots{})
        !          4569: @end example
        !          4570: 
        !          4571: @noindent
        !          4572: the first pattern would not apply at all, because this insn does not
        !          4573: contain two identical subexpressions in the right place.  The pattern would
        !          4574: say, ``That does not look like an add instruction; try other patterns.''
        !          4575: The second pattern would say, ``Yes, that's an add instruction, but there
        !          4576: is something wrong with it.''  It would direct the reload pass of the
        !          4577: compiler to generate additional insns to make the constraint true.  The
        !          4578: results might look like this:
        !          4579: 
        !          4580: @example
        !          4581: (insn @var{n2} @var{prev} @var{n}
        !          4582:   (set (reg:SI 3) (reg:SI 6))
        !          4583:   @dots{})
        !          4584: 
        !          4585: (insn @var{n} @var{n2} @var{next}
        !          4586:   (set (reg:SI 3)
        !          4587:        (plus:SI (reg:SI 3) (reg:SI 109)))
        !          4588:   @dots{})
        !          4589: @end example
        !          4590: 
        !          4591: Because insns that don't fit the constraints are fixed up by loading
        !          4592: operands into registers, every instruction pattern's constraints must
        !          4593: permit the case where all the operands are in registers.  It need not
        !          4594: permit all classes of registers; the compiler knows how to copy registers
        !          4595: into other registers of the proper class in order to make an instruction
        !          4596: valid.  But if no registers are permitted, the compiler will be stymied: it
        !          4597: does not know how to save a register in memory in order to make an
        !          4598: instruction valid.  Instruction patterns that reject registers can be
        !          4599: made valid by attaching a condition-expression that refuses to match
        !          4600: an insn at all if the crucial operand is a register.
        !          4601: 
        !          4602: @node Multi-Alternative, Class Preferences, Simple Constraints, Constraints
        !          4603: @subsection Multiple Alternative Constraints
        !          4604: 
        !          4605: Sometimes a single instruction has multiple alternative sets of possible
        !          4606: operands.  For example, on the 68000, a logical-or instruction can combine
        !          4607: register or an immediate value into memory, or it can combine any kind of
        !          4608: operand into a register; but it cannot combine one memory location into
        !          4609: another.
        !          4610: 
        !          4611: These constraints are represented as multiple alternatives.  An alternative
        !          4612: can be described by a series of letters for each operand.  The overall
        !          4613: constraint for an operand is made from the letters for this operand
        !          4614: from the first alternative, a comma, the letters for this operand from
        !          4615: the second alternative, a comma, and so on until the last alternative.
        !          4616: Here is how it is done for fullword logical-or on the 68000:
        !          4617: 
        !          4618: @example
        !          4619: (define_insn "iorsi3"
        !          4620:   [(set (match_operand:SI 0 "general_operand" "=%m,d")
        !          4621:         (ior:SI (match_operand:SI 1 "general_operand" "0,0")
        !          4622:                 (match_operand:SI 2 "general_operand" "dKs,dmKs")))]
        !          4623:   @dots{})
        !          4624: @end example
        !          4625: 
        !          4626: The first alternative has @samp{m} (memory) for operand 0, @samp{0} for
        !          4627: operand 1 (meaning it must match operand 0), and @samp{dKs} for operand 2.
        !          4628: The second alternative has @samp{d} (data register) for operand 0, @samp{0}
        !          4629: for operand 1, and @samp{dmKs} for operand 2.  The @samp{=} and @samp{%} in
        !          4630: the constraint for operand 0 are not part of any alternative; their meaning
        !          4631: is explained in the next section.
        !          4632: 
        !          4633: If all the operands fit any one alternative, the instruction is valid.
        !          4634: Otherwise, for each alternative, the compiler counts how many instructions
        !          4635: must be added to copy the operands so that that alternative applies.
        !          4636: The alternative requiring the least copying is chosen.  If two alternatives
        !          4637: need the same amount of copying, the one that comes first is chosen.
        !          4638: These choices can be altered with the @samp{?} and @samp{!} characters:
        !          4639: 
        !          4640: @table @samp
        !          4641: @item ?
        !          4642: Disparage slightly the alternative that the @samp{?} appears in,
        !          4643: as a choice when no alternative applies exactly.  The compiler regards
        !          4644: this alternative as one unit more costly for each @samp{?} that appears
        !          4645: in it.
        !          4646: 
        !          4647: @item !
        !          4648: Disparage severely the alternative that the @samp{!} appears in.
        !          4649: When operands must be copied into registers, the compiler will
        !          4650: never choose this alternative as the one to strive for.
        !          4651: @end table
        !          4652: 
        !          4653: When an insn pattern has multiple alternatives in its constraints,
        !          4654: often the appearance of the assembler code determined mostly by which
        !          4655: alternative was matched.  When this is so, the C code for writing the
        !          4656: assembler code can use the variable @code{which_alternative}, which is
        !          4657: the ordinal number of the alternative that was actually satisfied
        !          4658: (0 for the first, 1 for the second alternative, etc.).  For example:
        !          4659: 
        !          4660: @example
        !          4661: (define_insn ""
        !          4662:   [(set (match_operand:SI 0 "general_operand" "r,m")
        !          4663:         (const_int 0))]
        !          4664:   ""
        !          4665:   "*
        !          4666:   return (which_alternative == 0
        !          4667:           ? \"clrreg %0\" : \"clrmem %0\");
        !          4668:   ")
        !          4669: @end example
        !          4670: 
        !          4671: @node Class Preferences, Modifiers, Multi-Alternative, Constraints
        !          4672: @subsection Register Class Preferences
        !          4673: 
        !          4674: The operand constraints have another function: they enable the compiler
        !          4675: to decide which kind of hardware register a pseudo register is best
        !          4676: allocated to.  The compiler examines the constraints that apply to the
        !          4677: insns that use the pseudo register, looking for the machine-dependent
        !          4678: letters such as @samp{d} and @samp{a} that specify classes of registers.
        !          4679: The pseudo register is put in whichever class gets the most ``votes''.
        !          4680: The constraint letters @samp{g} and @samp{r} also vote: they vote in
        !          4681: favor of a general register.  The machine description says which registers
        !          4682: are considered general.
        !          4683: 
        !          4684: Of course, on some machines all registers are equivalent, and no register
        !          4685: classes are defined.  Then none of this complexity is relevant.
        !          4686: 
        !          4687: @node Modifiers, No Constraints, Class Preferences, Constraints
        !          4688: @subsection Constraint Modifier Characters
        !          4689: 
        !          4690: @table @samp
        !          4691: @item =
        !          4692: Means that this operand is write-only for this instruction: the previous
        !          4693: value is discarded and replaced by output data.
        !          4694: 
        !          4695: @item +
        !          4696: Means that this operand is both read and written by the instruction.
        !          4697: 
        !          4698: When the compiler fixes up the operands to satisfy the constraints,
        !          4699: it needs to know which operands are inputs to the instruction and
        !          4700: which are outputs from it.  @samp{=} identifies an output; @samp{+}
        !          4701: identifies an operand that is both input and output; all other operands
        !          4702: are assumed to be input only.
        !          4703: 
        !          4704: @item &
        !          4705: Means (in a particular alternative) that this operand is written
        !          4706: before the instruction is finished using the input operands.
        !          4707: Therefore, this operand may not lie in a register that is used as an
        !          4708: input operand or as part of any memory address.
        !          4709: 
        !          4710: @samp{&} applies only to the alternative in which it is written.  In
        !          4711: constraints with multiple alternatives, sometimes one alternative
        !          4712: requires @samp{&} while others do not.  See, for example, the
        !          4713: @samp{movdf} insn of the 68000.
        !          4714: 
        !          4715: @samp{&} does not obviate the need to write @samp{=}.
        !          4716: 
        !          4717: @item %
        !          4718: Declares the instruction to be commutative for this operand and the
        !          4719: following operand.  This means that the compiler may interchange the
        !          4720: two operands if that is the cheapest way to make all operands fit the
        !          4721: constraints.  This is often used in patterns for addition instructions
        !          4722: that really have only two operands: the result must go in one of the
        !          4723: arguments.  Here for example, is how the 68000 halfword-add
        !          4724: instruction is defined:
        !          4725: 
        !          4726: @example
        !          4727: (define_insn "addhi3"
        !          4728:   [(set (match_operand:HI 0 "general_operand" "=m,r")
        !          4729:      (plus:HI (match_operand:HI 1 "general_operand" "%0,0")
        !          4730:               (match_operand:HI 2 "general_operand" "di,g")))]
        !          4731:   @dots{})
        !          4732: @end example
        !          4733: 
        !          4734: Note that in previous versions of GNU CC the @samp{%} constraint
        !          4735: modifier always applied to operands 1 and 2 regardless of which
        !          4736: operand it was written in.  The usual custom was to write it in
        !          4737: operand 0.  Now it must be in operand 1 if the operands to be
        !          4738: exchanged are 1 and 2.
        !          4739: 
        !          4740: @item #
        !          4741: Says that all following characters, up to the next comma, are to be
        !          4742: ignored as a constraint.  They are significant only for choosing
        !          4743: register preferences.
        !          4744: 
        !          4745: @item *
        !          4746: Says that the following character should be ignored when choosing
        !          4747: register preferences.  @samp{*} has no effect on the meaning of the
        !          4748: constraint as a constraint.
        !          4749: 
        !          4750: Here is an example: the 68000 has an instruction to sign-extend a
        !          4751: halfword in a data register, and can also sign-extend a value by
        !          4752: copying it into an address register.  While either kind of register is
        !          4753: acceptable, the constraints on an address-register destination are
        !          4754: less strict, so it is best if register allocation makes an address
        !          4755: register its goal.  Therefore, @samp{*} is used so that the @samp{d}
        !          4756: constraint letter (for data register) is ignored when computing
        !          4757: register preferences.
        !          4758: 
        !          4759: @example
        !          4760: (define_insn "extendhisi2"
        !          4761:   [(set (match_operand:SI 0 "general_operand" "=*d,a")
        !          4762:         (sign_extend:SI
        !          4763:          (match_operand:HI 1 "general_operand" "0,g")))]
        !          4764:   @dots{})
        !          4765: @end example
        !          4766: @end table
        !          4767: 
        !          4768: @node No Constraints,, Modifiers, Constraints
        !          4769: @subsection Not Using Constraints
        !          4770: 
        !          4771: Some machines are so clean that operand constraints are not required.  For
        !          4772: example, on the Vax, an operand valid in one context is valid in any other
        !          4773: context.  On such a machine, every operand constraint would be @samp{g},
        !          4774: excepting only operands of ``load address'' instructions which are
        !          4775: written as if they referred to a memory location's contents but actual
        !          4776: refer to its address.  They would have constraint @samp{p}.
        !          4777: 
        !          4778: For such machines, instead of writing @samp{g} and @samp{p} for all
        !          4779: the constraints, you can choose to write a description with empty constraints.
        !          4780: Then you write @samp{""} for the constraint in every @samp{match_operand}.
        !          4781: Address operands are identified by writing an @samp{address} expression
        !          4782: around the @samp{match_operand}, not by their constraints.
        !          4783: 
        !          4784: When the machine description has just empty constraints, certain parts
        !          4785: of compilation are skipped, making the compiler faster.
        !          4786: 
        !          4787: @node Standard Names, Pattern Ordering, Constraints, Machine Desc
        !          4788: @section Standard Names for Patterns Used in Generation
        !          4789: 
        !          4790: Here is a table of the instruction names that are meaningful in the RTL
        !          4791: generation pass of the compiler.  Giving one of these names to an
        !          4792: instruction pattern tells the RTL generation pass that it can use the
        !          4793: pattern in to accomplish a certain task.
        !          4794: 
        !          4795: @table @asis
        !          4796: @item @samp{mov@var{m}}
        !          4797: Here @var{m} is a two-letter machine mode name, in lower case.  This
        !          4798: instruction pattern moves data with that machine mode from operand 1 to
        !          4799: operand 0.  For example, @samp{movsi} moves full-word data.
        !          4800: 
        !          4801: If operand 0 is a @samp{subreg} with mode @var{m} of a register whose
        !          4802: natural mode is wider than @var{m}, the effect of this instruction is
        !          4803: to store the specified value in the part of the register that corresponds
        !          4804: to mode @var{m}.  The effect on the rest of the register is undefined.
        !          4805: 
        !          4806: This class of patterns is special in several ways.  First of all, each
        !          4807: of these names @emph{must} be defined, because there is no other way
        !          4808: to copy a datum from one place to another.
        !          4809: 
        !          4810: Second, these patterns are not used solely in the RTL generation pass.
        !          4811: Even the reload pass can generate move insns to copy values from stack
        !          4812: slots into temporary registers.  When it does so, one of the operands
        !          4813: is a hard register and the other is an operand that can have a reload.
        !          4814: 
        !          4815: Therefore, when given such a pair of operands, the pattern must
        !          4816: generate RTL which needs no temporary registers---no registers other
        !          4817: than the operands.  For example, if you support the pattern with a
        !          4818: @code{define_expand}, then in such a case you mustn't call
        !          4819: @code{force_reg} or any other such function which might generate new
        !          4820: pseudo registers.
        !          4821: 
        !          4822: This requirement exists even for subword modes on a RISC machine where
        !          4823: fetching those modes from memory normally requires several insns and
        !          4824: some temporary registers.  Look in @file{spur.md} to see how the
        !          4825: requirement is satisfied.
        !          4826: 
        !          4827: The variety of operands that have reloads depends on the rest of the
        !          4828: machine description, but typically on a RISC machine these can only be
        !          4829: pseudo registers that did not get hard registers, while on other
        !          4830: machines explicit memory references will get optional reloads.
        !          4831: 
        !          4832: @item @samp{movstrict@var{m}}
        !          4833: Like @samp{mov@var{m}} except that if operand 0 is a @samp{subreg}
        !          4834: with mode @var{m} of a register whose natural mode is wider,
        !          4835: the @samp{movstrict@var{m}} instruction is guaranteed not to alter
        !          4836: any of the register except the part which belongs to mode @var{m}.
        !          4837: 
        !          4838: @item @samp{add@var{m}3}
        !          4839: Add operand 2 and operand 1, storing the result in operand 0.  All operands
        !          4840: must have mode @var{m}.  This can be used even on two-address machines, by
        !          4841: means of constraints requiring operands 1 and 0 to be the same location.
        !          4842: 
        !          4843: @item @samp{sub@var{m}3}, @samp{mul@var{m}3}, @samp{umul@var{m}3}, @samp{div@var{m}3}, @samp{udiv@var{m}3}, @samp{mod@var{m}3}, @samp{umod@var{m}3}, @samp{and@var{m}3}, @samp{ior@var{m}3}, @samp{xor@var{m}3}
        !          4844: Similar, for other arithmetic operations.
        !          4845: 
        !          4846: @item @samp{andcb@var{m}3}
        !          4847: Bitwise logical-and operand 1 with the complement of operand 2
        !          4848: and store the result in operand 0.
        !          4849: 
        !          4850: @item @samp{mulhisi3}
        !          4851: Multiply operands 1 and 2, which have mode @code{HImode}, and store
        !          4852: a @code{SImode} product in operand 0.
        !          4853: 
        !          4854: @item @samp{mulqihi3}, @samp{mulsidi3}
        !          4855: Similar widening-multiplication instructions of other widths.
        !          4856: 
        !          4857: @item @samp{umulqihi3}, @samp{umulhisi3}, @samp{umulsidi3}
        !          4858: Similar widening-multiplication instructions that do unsigned
        !          4859: multiplication.
        !          4860: 
        !          4861: @item @samp{divmod@var{m}4}
        !          4862: Signed division that produces both a quotient and a remainder.
        !          4863: Operand 1 is divided by operand 2 to produce a quotient stored
        !          4864: in operand 0 and a remainder stored in operand 3.
        !          4865: 
        !          4866: @item @samp{udivmod@var{m}4}
        !          4867: Similar, but does unsigned division.
        !          4868: 
        !          4869: @item @samp{divmod@var{m}@var{n}4}
        !          4870: Like @samp{divmod@var{m}4} except that only the dividend has mode
        !          4871: @var{m}; the divisor, quotient and remainder have mode @var{n}.
        !          4872: For example, the Vax has a @samp{divmoddisi4} instruction
        !          4873: (but it is omitted from the machine description, because it
        !          4874: is so slow that it is faster to compute remainders by the
        !          4875: circumlocution that the compiler will use if this instruction is
        !          4876: not available).
        !          4877: 
        !          4878: @item @samp{ashl@var{m}3}
        !          4879: Arithmetic-shift operand 1 left by a number of bits specified by
        !          4880: operand 2, and store the result in operand 0.  Operand 2 has
        !          4881: mode @code{SImode}, not mode @var{m}.
        !          4882: 
        !          4883: @item @samp{ashr@var{m}3}, @samp{lshl@var{m}3}, @samp{lshr@var{m}3}, @samp{rotl@var{m}3}, @samp{rotr@var{m}3}
        !          4884: Other shift and rotate instructions.
        !          4885: 
        !          4886: Logical and arithmetic left shift are the same.  Machines that do not
        !          4887: allow negative shift counts often have only one instruction for
        !          4888: shifting left.  On such machines, you should define a pattern named
        !          4889: @samp{ashl@var{m}3} and leave @samp{lshl@var{m}3} undefined.
        !          4890: 
        !          4891: @item @samp{neg@var{m}2}
        !          4892: Negate operand 1 and store the result in operand 0.
        !          4893: 
        !          4894: @item @samp{abs@var{m}2}
        !          4895: Store the absolute value of operand 1 into operand 0.
        !          4896: 
        !          4897: @item @samp{sqrt@var{m}2}
        !          4898: Store the square root of operand 1 into operand 0.
        !          4899: 
        !          4900: @item @samp{ffs@var{m}2}
        !          4901: Store into operand 0 one plus the index of the least significant 1-bit
        !          4902: of operand 1.  If operand 1 is zero, store zero.  @var{m} is the mode
        !          4903: of operand 0; operand 1's mode is specified by the instruction
        !          4904: pattern, and the compiler will convert the operand to that mode before
        !          4905: generating the instruction.
        !          4906: 
        !          4907: @item @samp{one_cmpl@var{m}2}
        !          4908: Store the bitwise-complement of operand 1 into operand 0.
        !          4909: 
        !          4910: @item @samp{cmp@var{m}}
        !          4911: Compare operand 0 and operand 1, and set the condition codes.
        !          4912: The RTL pattern should look like this:
        !          4913: 
        !          4914: @example
        !          4915: (set (cc0) (minus (match_operand:@var{m} 0 @dots{})
        !          4916:                   (match_operand:@var{m} 1 @dots{})))
        !          4917: @end example
        !          4918: 
        !          4919: Each such definition in the machine description, for integer mode
        !          4920: @var{m}, must have a corresponding @samp{tst@var{m}} pattern, because
        !          4921: optimization can simplify the compare into a test when operand 1 is
        !          4922: zero.
        !          4923: 
        !          4924: @item @samp{tst@var{m}}
        !          4925: Compare operand 0 against zero, and set the condition codes.
        !          4926: The RTL pattern should look like this:
        !          4927: 
        !          4928: @example
        !          4929: (set (cc0) (match_operand:@var{m} 0 @dots{}))
        !          4930: @end example
        !          4931: 
        !          4932: @item @samp{movstr@var{m}}
        !          4933: Block move instruction.  The addresses of the destination and source
        !          4934: strings are the first two operands, and both are in mode @code{Pmode}.
        !          4935: The number of bytes to move is the third operand, in mode @var{m}.
        !          4936: 
        !          4937: @item @samp{cmpstr@var{m}}
        !          4938: Block compare instruction, with operands like @samp{movstr@var{m}}
        !          4939: except that the two memory blocks are compared byte by byte
        !          4940: in lexicographic order.  The effect of the instruction is to set
        !          4941: the condition codes.
        !          4942: 
        !          4943: @item @samp{float@var{m}@var{n}2}
        !          4944: Convert operand 1 (valid for fixed point mode @var{m}) to floating
        !          4945: point mode @var{n} and store in operand 0 (which has mode @var{n}).
        !          4946: 
        !          4947: @item @samp{fix@var{m}@var{n}2}
        !          4948: Convert operand 1 (valid for floating point mode @var{m}) to fixed
        !          4949: point mode @var{n} as a signed number and store in operand 0 (which
        !          4950: has mode @var{n}).  This instruction's result is defined only when
        !          4951: the value of operand 1 is an integer.
        !          4952: 
        !          4953: @item @samp{fixuns@var{m}@var{n}2}
        !          4954: Convert operand 1 (valid for floating point mode @var{m}) to fixed
        !          4955: point mode @var{n} as an unsigned number and store in operand 0 (which
        !          4956: has mode @var{n}).  This instruction's result is defined only when the
        !          4957: value of operand 1 is an integer.
        !          4958: 
        !          4959: @item @samp{ftrunc@var{m}2}
        !          4960: Convert operand 1 (valid for floating point mode @var{m}) to an
        !          4961: integer value, still represented in floating point mode @var{m}, and
        !          4962: store it in operand 0 (valid for floating point mode @var{m}).
        !          4963: 
        !          4964: @item @samp{fix_trunc@var{m}@var{n}2}
        !          4965: Like @samp{fix@var{m}@var{n}2} but works for any floating point value
        !          4966: of mode @var{m} by converting the value to an integer.
        !          4967: 
        !          4968: @item @samp{fixuns_trunc@var{m}@var{n}2}
        !          4969: Like @samp{fixuns@var{m}@var{n}2} but works for any floating point
        !          4970: value of mode @var{m} by converting the value to an integer.
        !          4971: 
        !          4972: @item @samp{trunc@var{m}@var{n}}
        !          4973: Truncate operand 1 (valid for mode @var{m}) to mode @var{n} and
        !          4974: store in operand 0 (which has mode @var{n}).  Both modes must be fixed
        !          4975: point or both floating point.
        !          4976: 
        !          4977: @item @samp{extend@var{m}@var{n}}
        !          4978: Sign-extend operand 1 (valid for mode @var{m}) to mode @var{n} and
        !          4979: store in operand 0 (which has mode @var{n}).  Both modes must be fixed
        !          4980: point or both floating point.
        !          4981: 
        !          4982: @item @samp{zero_extend@var{m}@var{n}}
        !          4983: Zero-extend operand 1 (valid for mode @var{m}) to mode @var{n} and
        !          4984: store in operand 0 (which has mode @var{n}).  Both modes must be fixed
        !          4985: point.
        !          4986: 
        !          4987: @item @samp{extv}
        !          4988: Extract a bit-field from operand 1 (a register or memory operand),
        !          4989: where operand 2 specifies the width in bits and operand 3 the starting
        !          4990: bit, and store it in operand 0.  Operand 0 must have @code{Simode}.
        !          4991: Operand 1 may have mode @code{QImode} or @code{SImode}; often
        !          4992: @code{SImode} is allowed only for registers.  Operands 2 and 3 must be
        !          4993: valid for @code{SImode}.
        !          4994: 
        !          4995: The RTL generation pass generates this instruction only with constants
        !          4996: for operands 2 and 3.
        !          4997: 
        !          4998: The bit-field value is sign-extended to a full word integer
        !          4999: before it is stored in operand 0.
        !          5000: 
        !          5001: @item @samp{extzv}
        !          5002: Like @samp{extv} except that the bit-field value is zero-extended.
        !          5003: 
        !          5004: @item @samp{insv}
        !          5005: Store operand 3 (which must be valid for @code{SImode}) into a
        !          5006: bit-field in operand 0, where operand 1 specifies the width in bits
        !          5007: and operand 2 the starting bit.  Operand 0 may have mode @code{QImode}
        !          5008: or @code{SImode}; often @code{SImode} is allowed only for registers.
        !          5009: Operands 1 and 2 must be valid for @code{SImode}.
        !          5010: 
        !          5011: The RTL generation pass generates this instruction only with constants
        !          5012: for operands 1 and 2.
        !          5013: 
        !          5014: @item @samp{s@var{cond}}
        !          5015: Store zero or nonzero in the operand according to the condition codes.
        !          5016: Value stored is nonzero iff the condition @var{cond} is true.
        !          5017: @var{cond} is the name of a comparison operation expression code, such
        !          5018: as @samp{eq}, @samp{lt} or @samp{leu}.
        !          5019: 
        !          5020: You specify the mode that the operand must have when you write the
        !          5021: @code{match_operand} expression.  The compiler automatically sees
        !          5022: which mode you have used and supplies an operand of that mode.
        !          5023: 
        !          5024: The value stored for a true condition must have 1 as its low bit.
        !          5025: Otherwise the instruction is not suitable and must be omitted from the
        !          5026: machine description.  You must tell the compiler exactly which value
        !          5027: is stored by defining the macro @code{STORE_FLAG_VALUE}.
        !          5028: 
        !          5029: @item @samp{b@var{cond}}
        !          5030: Conditional branch instruction.  Operand 0 is a @samp{label_ref}
        !          5031: that refers to the label to jump to.  Jump if the condition codes
        !          5032: meet condition @var{cond}.
        !          5033: 
        !          5034: @item @samp{call}
        !          5035: Subroutine call instruction.  Operand 1 is the number of bytes of
        !          5036: arguments pushed (in mode @code{SImode}), and operand 0 is the
        !          5037: function to call.  Operand 0 should be a @samp{mem} RTX whose address
        !          5038: is the address of the function.
        !          5039: 
        !          5040: @item @samp{return}
        !          5041: Subroutine return instruction.  This instruction pattern name should be
        !          5042: defined only if a single instruction can do all the work of returning
        !          5043: from a function.
        !          5044: 
        !          5045: @item @samp{casesi}
        !          5046: Instruction to jump through a dispatch table, including bounds checking.
        !          5047: This instruction takes five operands:
        !          5048: 
        !          5049: @enumerate
        !          5050: @item
        !          5051: The index to dispatch on, which has mode @code{SImode}.
        !          5052: 
        !          5053: @item
        !          5054: The lower bound for indices in the table, an integer constant.
        !          5055: 
        !          5056: @item
        !          5057: The upper bound for indices in the table, an integer constant.
        !          5058: 
        !          5059: @item
        !          5060: A label to jump to if the index has a value outside the bounds.
        !          5061: (If the machine-description macro @code{CASE_DROPS_THROUGH} is defined,
        !          5062: then an out-of-bounds index drops through to the code following
        !          5063: the jump table instead of jumping to this label.  In that case,
        !          5064: this label is not actually used by the @samp{casesi} instruction,
        !          5065: but it is always provided as an operand.)
        !          5066: 
        !          5067: @item
        !          5068: A label that precedes the table itself.
        !          5069: @end enumerate
        !          5070: 
        !          5071: The table is a @samp{addr_vec} or @samp{addr_diff_vec} inside of a
        !          5072: @samp{jump_insn}.  The number of elements in the table is one plus the
        !          5073: difference between the upper bound and the lower bound.
        !          5074: 
        !          5075: @item @samp{tablejump}
        !          5076: Instruction to jump to a variable address.  This is a low-level
        !          5077: capability which can be used to implement a dispatch table when there
        !          5078: is no @samp{casesi} pattern.
        !          5079: 
        !          5080: This pattern requires two operands: the address or offset, and a label
        !          5081: which should immediately precede the jump table.  If the macro
        !          5082: @code{CASE_VECTOR_PC_RELATIVE} is defined then the first operand is an
        !          5083: absolute address to jump to; otherwise, it is an offset which counts
        !          5084: from the address of the table.
        !          5085: 
        !          5086: The @samp{tablejump} insn is always the last insn before the jump
        !          5087: table it uses.  Its assembler code normally has no need to use the
        !          5088: second operand, but you should incorporate it in the RTL pattern so
        !          5089: that the jump optimizer will not delete the table as unreachable code.
        !          5090: @end table
        !          5091: 
        !          5092: @node Pattern Ordering, Dependent Patterns, Standard Names, Machine Desc
        !          5093: @section When the Order of Patterns Matters
        !          5094: 
        !          5095: Sometimes an insn can match more than one instruction pattern.  Then the
        !          5096: pattern that appears first in the machine description is the one used.
        !          5097: Therefore, more specific patterns (patterns that will match fewer things)
        !          5098: and faster instructions (those that will produce better code when they
        !          5099: do match) should usually go first in the description.
        !          5100: 
        !          5101: In some cases the effect of ordering the patterns can be used to hide
        !          5102: a pattern when it is not valid.  For example, the 68000 has an
        !          5103: instruction for converting a fullword to floating point and another
        !          5104: for converting a byte to floating point.  An instruction converting
        !          5105: an integer to floating point could match either one.  We put the
        !          5106: pattern to convert the fullword first to make sure that one will
        !          5107: be used rather than the other.  (Otherwise a large integer might
        !          5108: be generated as a single-byte immediate quantity, which would not work.)
        !          5109: Instead of using this pattern ordering it would be possible to make the
        !          5110: pattern for convert-a-byte smart enough to deal properly with any
        !          5111: constant value.
        !          5112: 
        !          5113: @node Dependent Patterns, Jump Patterns, Pattern Ordering, Machine Desc
        !          5114: @section Interdependence of Patterns
        !          5115: 
        !          5116: Every machine description must have a named pattern for each of the
        !          5117: conditional branch names @samp{b@var{cond}}.  The recognition template
        !          5118: must always have the form
        !          5119: 
        !          5120: @example
        !          5121: (set (pc)
        !          5122:      (if_then_else (@var{cond} (cc0) (const_int 0))
        !          5123:                    (label_ref (match_operand 0 "" ""))
        !          5124:                    (pc)))
        !          5125: @end example
        !          5126: 
        !          5127: @noindent
        !          5128: In addition, every machine description must have an anonymous pattern
        !          5129: for each of the possible reverse-conditional branches.  These patterns
        !          5130: look like
        !          5131: 
        !          5132: @example
        !          5133: (set (pc)
        !          5134:      (if_then_else (@var{cond} (cc0) (const_int 0))
        !          5135:                    (pc)
        !          5136:                    (label_ref (match_operand 0 "" ""))))
        !          5137: @end example
        !          5138: 
        !          5139: @noindent
        !          5140: They are necessary because jump optimization can turn direct-conditional
        !          5141: branches into reverse-conditional branches.
        !          5142: 
        !          5143: The compiler does more with RTL than just create it from patterns
        !          5144: and recognize the patterns: it can perform arithmetic expression codes
        !          5145: when constant values for their operands can be determined.  As a result,
        !          5146: sometimes having one pattern can require other patterns.  For example, the
        !          5147: Vax has no `and' instruction, but it has `and not' instructions.  Here
        !          5148: is the definition of one of them:
        !          5149: 
        !          5150: @example
        !          5151: (define_insn "andcbsi2"
        !          5152:   [(set (match_operand:SI 0 "general_operand" "")
        !          5153:         (and:SI (match_dup 0)
        !          5154:                 (not:SI (match_operand:SI
        !          5155:                           1 "general_operand" ""))))]
        !          5156:   ""
        !          5157:   "bicl2 %1,%0")
        !          5158: @end example
        !          5159: 
        !          5160: @noindent
        !          5161: If operand 1 is an explicit integer constant, an instruction constructed
        !          5162: using that pattern can be simplified into an `and' like this:
        !          5163: 
        !          5164: @example
        !          5165: (set (reg:SI 41)
        !          5166:      (and:SI (reg:SI 41)
        !          5167:              (const_int 0xffff7fff)))
        !          5168: @end example
        !          5169: 
        !          5170: @noindent
        !          5171: (where the integer constant is the one's complement of what
        !          5172: appeared in the original instruction).
        !          5173: 
        !          5174: To avoid a fatal error, the compiler must have a pattern that recognizes
        !          5175: such an instruction.  Here is what is used:
        !          5176: 
        !          5177: @example
        !          5178: (define_insn ""
        !          5179:   [(set (match_operand:SI 0 "general_operand" "")
        !          5180:         (and:SI (match_dup 0)
        !          5181:                 (match_operand:SI 1 "general_operand" "")))]
        !          5182:   "GET_CODE (operands[1]) == CONST_INT"
        !          5183:   "*
        !          5184: @{ operands[1]
        !          5185:     = gen_rtx (CONST_INT, VOIDmode, ~INTVAL (operands[1]));
        !          5186:   return \"bicl2 %1,%0\";
        !          5187: @}")
        !          5188: @end example
        !          5189: 
        !          5190: @noindent
        !          5191: Whereas a pattern to match a general `and' instruction is impossible to
        !          5192: support on the Vax, this pattern is possible because it matches only a
        !          5193: constant second argument: a special case that can be output as an `and not'
        !          5194: instruction.
        !          5195: 
        !          5196: A ``compare'' instruction whose RTL looks like this:
        !          5197: 
        !          5198: @example
        !          5199: (set (cc0) (minus @var{operand} (const_int 0)))
        !          5200: @end example
        !          5201: 
        !          5202: @noindent
        !          5203: may be simplified by optimization into a ``test'' like this:
        !          5204: 
        !          5205: @example
        !          5206: (set (cc0) @var{operand})
        !          5207: @end example
        !          5208: 
        !          5209: @noindent
        !          5210: So in the machine description, each ``compare'' pattern for an integer
        !          5211: mode must have a corresponding ``test'' pattern that will match the
        !          5212: result of such simplification.
        !          5213: 
        !          5214: In some cases machines support instructions identical except for the
        !          5215: machine mode of one or more operands.  For example, there may be
        !          5216: ``sign-extend halfword'' and ``sign-extend byte'' instructions whose
        !          5217: patterns are
        !          5218: 
        !          5219: @example
        !          5220: (set (match_operand:SI 0 @dots{})
        !          5221:      (extend:SI (match_operand:HI 1 @dots{})))
        !          5222: 
        !          5223: (set (match_operand:SI 0 @dots{})
        !          5224:      (extend:SI (match_operand:QI 1 @dots{})))
        !          5225: @end example
        !          5226: 
        !          5227: @noindent
        !          5228: Constant integers do not specify a machine mode, so an instruction to
        !          5229: extend a constant value could match either pattern.  The pattern it
        !          5230: actually will match is the one that appears first in the file.  For correct
        !          5231: results, this must be the one for the widest possible mode (@code{HImode},
        !          5232: here).  If the pattern matches the @code{QImode} instruction, the results
        !          5233: will be incorrect if the constant value does not actually fit that mode.
        !          5234: 
        !          5235: Such instructions to extend constants are rarely generated because they are
        !          5236: optimized away, but they do occasionally happen in nonoptimized
        !          5237: compilations.
        !          5238: 
        !          5239: @node Jump Patterns, Peephole Definitions, Dependent Patterns, Machine Desc
        !          5240: @section Defining Jump Instruction Patterns
        !          5241: 
        !          5242: GNU CC assumes that the machine has a condition code.  A comparison insn
        !          5243: sets the condition code, recording the results of both signed and unsigned
        !          5244: comparison of the given operands.  A separate branch insn tests the
        !          5245: condition code and branches or not according its value.  The branch insns
        !          5246: come in distinct signed and unsigned flavors.  Many common machines, such
        !          5247: as the Vax, the 68000 and the 32000, work this way.
        !          5248: 
        !          5249: Some machines have distinct signed and unsigned compare instructions, and
        !          5250: only one set of conditional branch instructions.  The easiest way to handle
        !          5251: these machines is to treat them just like the others until the final stage
        !          5252: where assembly code is written.  At this time, when outputting code for the
        !          5253: compare instruction, peek ahead at the following branch using
        !          5254: @code{NEXT_INSN (insn)}.  (The variable @code{insn} refers to the insn
        !          5255: being output, in the output-writing code in an instruction pattern.)  If
        !          5256: the RTL says that is an unsigned branch, output an unsigned compare;
        !          5257: otherwise output a signed compare.  When the branch itself is output, you
        !          5258: can treat signed and unsigned branches identically.
        !          5259: 
        !          5260: The reason you can do this is that GNU CC always generates a pair of
        !          5261: consecutive RTL insns, one to set the condition code and one to test it,
        !          5262: and keeps the pair inviolate until the end.
        !          5263: 
        !          5264: To go with this technique, you must define the machine-description macro
        !          5265: @code{NOTICE_UPDATE_CC} to do @code{CC_STATUS_INIT}; in other words, no
        !          5266: compare instruction is superfluous.
        !          5267: 
        !          5268: Some machines have compare-and-branch instructions and no condition code.
        !          5269: A similar technique works for them.  When it is time to ``output'' a
        !          5270: compare instruction, record its operands in two static variables.  When
        !          5271: outputting the branch-on-condition-code instruction that follows, actually
        !          5272: output a compare-and-branch instruction that uses the remembered operands.
        !          5273: 
        !          5274: It also works to define patterns for compare-and-branch instructions.
        !          5275: In optimizing compilation, the pair of compare and branch instructions
        !          5276: will be combined accoprding to these patterns.  But this does not happen
        !          5277: if optimization is not requested.  So you must use one of the solutions
        !          5278: above in addition to any special patterns you define.
        !          5279: 
        !          5280: @node Peephole Definitions, Expander Definitions, Jump Patterns, Machine Desc
        !          5281: @section Defining Machine-Specific Peephole Optimizers
        !          5282: 
        !          5283: In addition to instruction patterns the @file{md} file may contain
        !          5284: definitions of machine-specific peephole optimizations.
        !          5285: 
        !          5286: The combiner does not notice certain peephole optimizations when the data
        !          5287: flow in the program does not suggest that it should try them.  For example,
        !          5288: sometimes two consecutive insns related in purpose can be combined even
        !          5289: though the second one does not appear to use a register computed in the
        !          5290: first one.  A machine-specific peephole optimizer can detect such
        !          5291: opportunities.
        !          5292: 
        !          5293: A definition looks like this:
        !          5294: 
        !          5295: @example
        !          5296: (define_peephole
        !          5297:   [@var{insn-pattern-1}
        !          5298:    @var{insn-pattern-2}
        !          5299:    @dots{}]
        !          5300:   "@var{condition}"
        !          5301:   "@var{template}")
        !          5302: @end example
        !          5303: 
        !          5304: In this skeleton, @var{insn-pattern-1} and so on are patterns to match
        !          5305: consecutive instructions.  The optimization applies to a sequence of
        !          5306: instructions when @var{insn-pattern-1} matches the first one,
        !          5307: @var{insn-pattern-2} matches the next, and so on.@refill
        !          5308: 
        !          5309: @var{insn-pattern-1} and so on look @emph{almost} like the second operand
        !          5310: of @code{define_insn}.  There is one important difference: this pattern is
        !          5311: an RTX, not a vector.  If the @code{define_insn} pattern would be a vector
        !          5312: of one element, the @var{insn-pattern} should be just that element, no
        !          5313: vector.  If the @code{define_insn} pattern would have multiple elements
        !          5314: then the @var{insn-pattern} must place the vector inside an explicit
        !          5315: @code{parallel} RTX.@refill
        !          5316: 
        !          5317: The operands of the instructions are matched with @code{match_operands} and
        !          5318: @code{match_dup}, as usual).  What is not usual is that the operand numbers
        !          5319: apply to all the instruction patterns in the definition.  So, you can check
        !          5320: for identical operands in two instructions by using @code{match_operand}
        !          5321: in one instruction and @code{match_dup} in the other.
        !          5322: 
        !          5323: The operand constraints used in @code{match_operand} patterns do not have
        !          5324: any direct effect on the applicability of the optimization, but they will
        !          5325: be validated afterward, so write constraints that are sure to fit whenever
        !          5326: the optimization is applied.  It is safe to use @code{"g"} for each
        !          5327: operand.
        !          5328: 
        !          5329: Once a sequence of instructions matches the patterns, the @var{condition}
        !          5330: is checked.  This is a C expression which makes the final decision whether
        !          5331: to perform the optimization (do so if the expression is nonzero).  If
        !          5332: @var{condition} is omitted (in other words, the string is empty) then the
        !          5333: optimization is applied to every sequence of instructions that matches the
        !          5334: patterns.
        !          5335: 
        !          5336: The defined peephole optimizations are applied after register allocation is
        !          5337: complete.  Therefore, the optimizer can check which operands have ended up
        !          5338: in which kinds of registers, just by looking at the operands.
        !          5339: 
        !          5340: The way to refer to the operands in @var{condition} is to write
        !          5341: @code{operands[@var{i}]} for operand number @var{i} (as matched by
        !          5342: @code{(match_operand @var{i} @dots{})}).  Use the variable @code{insn} to
        !          5343: refer to the last of the insns being matched; use @code{PREV_INSN} to find
        !          5344: the preceding insns (but be careful to skip over any @samp{note} insns that
        !          5345: intervene).@refill
        !          5346: 
        !          5347: When optimizing computations with intermediate results, you can use
        !          5348: @var{condition} to match only when the intermediate results are not used
        !          5349: elsewhere.  Use the C expression @code{dead_or_set_p (@var{insn},
        !          5350: @var{op})}, where @var{insn} is the insn in which you expect the value to
        !          5351: be used for the last time (from the value of @code{insn}, together with use
        !          5352: of @code{PREV_INSN}), and @var{op} is the intermediate value (from
        !          5353: @code{operands[@var{i}]}).@refill
        !          5354: 
        !          5355: Applying the optimization means replacing the sequence of instructions with
        !          5356: one new instruction.  The @var{template} controls ultimate output of
        !          5357: assembler code for this combined instruction.  It works exactly like the
        !          5358: template of a @code{define_insn}.  Operand numbers in this template are the
        !          5359: same ones used in matching the original sequence of instructions.
        !          5360: 
        !          5361: The result of a defined peephole optimizer does not need to match any of
        !          5362: the instruction patterns, and it does not have an opportunity to match
        !          5363: them.  The peephole optimizer definition itself serves as the instruction
        !          5364: pattern to control how the instruction is output.
        !          5365: 
        !          5366: Defined peephole optimizers are run in the last jump optimization pass, so
        !          5367: the instructions they produce are never combined or rearranged
        !          5368: automatically in any way.
        !          5369: 
        !          5370: Here is an example, taken from the 68000 machine description:
        !          5371: 
        !          5372: @example
        !          5373: (define_peephole
        !          5374:   [(set (reg:SI 15) (plus:SI (reg:SI 15) (const_int 4)))
        !          5375:    (set (match_operand:DF 0 "register_operand" "f")
        !          5376:         (match_operand:DF 1 "register_operand" "ad"))]
        !          5377:   "FP_REG_P (operands[0]) && ! FP_REG_P (operands[1])"
        !          5378:   "*
        !          5379: @{
        !          5380:   rtx xoperands[2];
        !          5381:   xoperands[1] = gen_rtx (REG, SImode, REGNO (operands[1]) + 1);
        !          5382: #ifdef MOTOROLA
        !          5383:   output_asm_insn (\"move.l %1,(sp)\", xoperands);
        !          5384:   output_asm_insn (\"move.l %1,-(sp)\", operands);
        !          5385:   return \"fmove.d (sp)+,%0\";
        !          5386: #else
        !          5387:   output_asm_insn (\"movel %1,sp@@\", xoperands);
        !          5388:   output_asm_insn (\"movel %1,sp@@-\", operands);
        !          5389:   return \"fmoved sp@@+,%0\";
        !          5390: #endif
        !          5391: @}
        !          5392: ")
        !          5393: @end example
        !          5394: 
        !          5395: The effect of this optimization is to change
        !          5396: 
        !          5397: @example
        !          5398: jbsr _foobar
        !          5399: addql #4,sp
        !          5400: movel d1,sp@@-
        !          5401: movel d0,sp@@-
        !          5402: fmoved sp@@+,fp0
        !          5403: @end example
        !          5404: 
        !          5405: @noindent
        !          5406: into
        !          5407: 
        !          5408: @example
        !          5409: jbsr _foobar
        !          5410: movel d1,sp@@
        !          5411: movel d0,sp@@-
        !          5412: fmoved sp@@+,fp0
        !          5413: @end example
        !          5414: 
        !          5415: @node Expander Definitions,, Peephole Definitions, Machine Desc
        !          5416: @section Defining RTL Sequences for Code Generation
        !          5417: 
        !          5418: On some target machines, some standard pattern names for RTL generation
        !          5419: cannot be handled with single insn, but a sequence of RTL insns can
        !          5420: represent them.  For these target machines, you can write a
        !          5421: @samp{define_expand} to specify how to generate the sequence of RTL.
        !          5422: 
        !          5423: A @samp{define_expand} is an RTL expression that looks almost like a
        !          5424: @samp{define_insn}; but, unlike the latter, a @samp{define_expand} is used
        !          5425: only for RTL generation and it can produce more than one RTL insn.
        !          5426: 
        !          5427: A @samp{define_expand} RTX has four operands:
        !          5428: 
        !          5429: @itemize @bullet
        !          5430: @item
        !          5431: The name.  Each @samp{define_expand} must have a name, since the only
        !          5432: use for it is to refer to it by name.
        !          5433: 
        !          5434: @item
        !          5435: The RTL template.  This is just like the RTL template for a
        !          5436: @samp{define_peephole} in that it is a vector of RTL expressions
        !          5437: each being one insn.
        !          5438: 
        !          5439: @item
        !          5440: The condition, a string containing a C expression.  This expression is
        !          5441: used to express how the availability of this pattern depends on
        !          5442: subclasses of target machine, selected by command-line options when
        !          5443: GNU CC is run.  This is just like the condition of a
        !          5444: @samp{define_insn} that has a standard name.
        !          5445: 
        !          5446: @item
        !          5447: The preparation statements, a string containing zero or more C
        !          5448: statements which are to be executed before RTL code is generated from
        !          5449: the RTL template.
        !          5450: 
        !          5451: Usually these statements prepare temporary registers for use as
        !          5452: internal operands in the RTL template, but they can also generate RTL
        !          5453: insns directly by calling routines such as @samp{emit_insn}, etc.
        !          5454: Any such insns precede the ones that come from the RTL template.
        !          5455: @end itemize
        !          5456: 
        !          5457: The RTL template, in addition to controlling generation of RTL insns,
        !          5458: also describes the operands that need to be specified when this pattern
        !          5459: is used.  In particular, it gives a predicate for each operand.
        !          5460: 
        !          5461: A true operand, which need to be specified in order to generate RTL from
        !          5462: the pattern, should be described with a @samp{match_operand} in its first
        !          5463: occurrence in the RTL template.  This enters information on the operand's
        !          5464: predicate into the tables that record such things.  GNU CC uses the
        !          5465: information to preload the operand into a register if that is required for
        !          5466: valid RTL code.  If the operand is referred to more than once, subsequent
        !          5467: references should use @samp{match_dup}.
        !          5468: 
        !          5469: The RTL template may also refer to internal ``operands'' which are
        !          5470: temporary registers or labels used only within the sequence made by the
        !          5471: @samp{define_expand}.  Internal operands are substituted into the RTL
        !          5472: template with @samp{match_dup}, never with @samp{match_operand}.  The
        !          5473: values of the internal operands are not passed in as arguments by the
        !          5474: compiler when it requests use of this pattern.  Instead, they are computed
        !          5475: within the pattern, in the preparation statements.  These statements
        !          5476: compute the values and store them into the appropriate elements of
        !          5477: @code{operands} so that @samp{match_dup} can find them.
        !          5478: 
        !          5479: There are two special macros defined for use in the preparation statements:
        !          5480: @code{DONE} and @code{FAIL}.  Use them with a following semicolon,
        !          5481: as a statement.
        !          5482: 
        !          5483: @table @code
        !          5484: @item DONE
        !          5485: Use the @code{DONE} macro to end RTL generation for the pattern.  The
        !          5486: only RTL insns resulting from the pattern on this occasion will be
        !          5487: those already emitted by explicit calls to @code{emit_insn} within the
        !          5488: preparation statements; the RTL template will not be generated.
        !          5489: 
        !          5490: @item FAIL
        !          5491: Make the pattern fail on this occasion.  When a pattern fails, it means
        !          5492: that the pattern was not truly available.  The calling routines in the
        !          5493: compiler will try other strategies for code generation using other patterns.
        !          5494: 
        !          5495: Failure is currently supported only for binary operations (addition,
        !          5496: multiplication, shifting, etc.).
        !          5497: 
        !          5498: Do not emit any insns explicitly with @code{emit_insn} before failing.
        !          5499: @end table
        !          5500: 
        !          5501: Here is an example, the definition of left-shift for the SPUR chip:
        !          5502: 
        !          5503: @example
        !          5504: (define_expand "ashlsi3"
        !          5505:   [(set (match_operand:SI 0 "register_operand" "")
        !          5506:         (ashift:SI
        !          5507:           (match_operand:SI 1 "register_operand" "")
        !          5508:           (match_operand:SI 2 "nonmemory_operand" "")))]
        !          5509:   ""
        !          5510:   "
        !          5511: @{
        !          5512:   if (GET_CODE (operands[2]) != CONST_INT
        !          5513:       || (unsigned) INTVAL (operands[2]) > 3)
        !          5514:     FAIL;
        !          5515: @}")
        !          5516: @end example
        !          5517: 
        !          5518: @noindent
        !          5519: This example uses @samp{define_expand} so that it can generate an RTL insn
        !          5520: for shifting when the shift-count is in the supported range of 0 to 3 but
        !          5521: fail in other cases where machine insns aren't available.  When it fails,
        !          5522: the compiler tries another strategy using different patterns (such as, a
        !          5523: library call).
        !          5524: 
        !          5525: If the compiler were able to handle nontrivial condition-strings in
        !          5526: patterns with names, then there would be possible to use a
        !          5527: @samp{define_insn} in that case.  Here is another case (zero-extension on
        !          5528: the 68000) which makes more use of the power of @samp{define_expand}:
        !          5529: 
        !          5530: @example
        !          5531: (define_expand "zero_extendhisi2"
        !          5532:   [(set (match_operand:SI 0 "general_operand" "")
        !          5533:         (const_int 0))
        !          5534:    (set (strict_low_part 
        !          5535:           (subreg:HI
        !          5536:             (match_operand:SI 0 "general_operand" "")
        !          5537:             0))
        !          5538:         (match_operand:HI 1 "general_operand" ""))]
        !          5539:   ""
        !          5540:   "operands[1] = make_safe_from (operands[1], operands[0]);")
        !          5541: @end example
        !          5542: 
        !          5543: @noindent
        !          5544: Here two RTL insns are generated, one to clear the entire output operand
        !          5545: and the other to copy the input operand into its low half.  This sequence
        !          5546: is incorrect if the input operand refers to [the old value of] the output
        !          5547: operand, so the preparation statement makes sure this isn't so.  The
        !          5548: function @code{make_safe_from} copies the @code{operands[1]} into a
        !          5549: temporary register if it refers to @code{operands[0]}.  It does this
        !          5550: by emitting another RTL insn.
        !          5551: 
        !          5552: Finally, a third example shows the use of an internal operand.
        !          5553: Zero-extension on the SPUR chip is done by @samp{and}-ing the result
        !          5554: against a halfword mask.  But this mask cannot be represented by a
        !          5555: @samp{const_int} because the constant value is too large to be legitimate
        !          5556: on this machine.  So it must be copied into a register with
        !          5557: @code{force_reg} and then the register used in the @samp{and}.
        !          5558: 
        !          5559: @example
        !          5560: (define_expand "zero_extendhisi2"
        !          5561:   [(set (match_operand:SI 0 "register_operand" "")
        !          5562:         (and:SI (subreg:SI
        !          5563:                   (match_operand:HI 1 "register_operand" "")
        !          5564:                   0)
        !          5565:                 (match_dup 2)))]
        !          5566:   ""
        !          5567:   "operands[2]
        !          5568:      = force_reg (SImode, gen_rtx (CONST_INT,
        !          5569:                                    VOIDmode, 65535)); ")
        !          5570: @end example
        !          5571: 
        !          5572: @node Machine Macros, Config, Machine Desc, Top
        !          5573: @chapter Machine Description Macros
        !          5574: 
        !          5575: The other half of the machine description is a C header file conventionally
        !          5576: given the name @file{tm-@var{machine}.h}.  The file @file{tm.h} should be a
        !          5577: link to it.  The header file @file{config.h} includes @file{tm.h} and most
        !          5578: compiler source files include @file{config.h}.
        !          5579: 
        !          5580: @menu
        !          5581: * Run-time Target::     Defining -m options like -m68000 and -m68020.
        !          5582: * Storage Layout::      Defining sizes and alignments of data types.
        !          5583: * Registers::           Naming and describing the hardware registers.
        !          5584: * Register Classes::    Defining the classes of hardware registers.
        !          5585: * Stack Layout::        Defining which way the stack grows and by how much.
        !          5586: * Library Names::       Specifying names of subroutines to call automatically.
        !          5587: * Addressing Modes::    Defining addressing modes valid for memory operands.
        !          5588: * Condition Code::      Defining how insns update the condition code.
        !          5589: * Assembler Format::    Defining how to write insns and pseudo-ops to output.
        !          5590: * Misc::                Everything else.
        !          5591: @end menu
        !          5592: 
        !          5593: @node Run-time Target, Storage Layout, Machine Macros, Machine Macros
        !          5594: @section Run-time Target Specification
        !          5595: 
        !          5596: @table @code
        !          5597: @item CPP_PREDEFINES
        !          5598: Define this to be a string constant containing @samp{-D} options
        !          5599: to define the predefined macros that identify this machine and system.
        !          5600: 
        !          5601: For example, on the Sun, one can use the value
        !          5602: 
        !          5603: @example
        !          5604: "-Dmc68000 -Dsun -Dunix"
        !          5605: @end example
        !          5606: 
        !          5607: @item extern int target_flags;
        !          5608: This declaration should be present.
        !          5609: 
        !          5610: @item TARGET_@dots{}
        !          5611: This series of macros is to allow compiler command arguments to
        !          5612: enable or disable the use of optional features of the target machine.
        !          5613: For example, one machine description serves both the 68000 and
        !          5614: the 68020; a command argument tells the compiler whether it should
        !          5615: use 68020-only instructions or not.  This command argument works
        !          5616: by means of a macro @code{TARGET_68020} that tests a bit in
        !          5617: @code{target_flags}.
        !          5618: 
        !          5619: Define a macro @code{TARGET_@var{featurename}} for each such option.
        !          5620: Its definition should test a bit in @code{target_flags}; for example:
        !          5621: 
        !          5622: @example
        !          5623: #define TARGET_68020 (target_flags & 1)
        !          5624: @end example
        !          5625: 
        !          5626: One place where these macros are used is in the condition-expressions
        !          5627: of instruction patterns.  Note how @code{TARGET_68020} appears
        !          5628: frequently in the 68000 machine description file, @file{m68k.md}.
        !          5629: Another place they are used is in the definitions of the other
        !          5630: macros in the @file{tm-@var{machine}.h} file.
        !          5631: 
        !          5632: @item TARGET_SWITCHES
        !          5633: This macro defines names of command options to set and clear
        !          5634: bits in @code{target_flags}.  Its definition is an initializer
        !          5635: with a subgrouping for each command option.
        !          5636: 
        !          5637: Each subgrouping contains a string constant, that defines the option
        !          5638: name, and a number, which contains the bits to set in
        !          5639: @code{target_flags}.  A negative number says to clear bits instead;
        !          5640: the negative of the number is which bits to clear.  The actual option
        !          5641: name is made by appending @samp{-m} to the specified name.
        !          5642: 
        !          5643: One of the subgroupings should have a null string.  The number in
        !          5644: this grouping is the default value for @code{target_flags}.  Any
        !          5645: target options act starting with that value.
        !          5646: 
        !          5647: Here is an example which defines @samp{-m68000} and @samp{-m68020}
        !          5648: with opposite meanings, and picks the latter as the default:
        !          5649: 
        !          5650: @example
        !          5651: #define TARGET_SWITCHES \
        !          5652:   @{ @{ "68020", 1@},      \
        !          5653:     @{ "68000", -1@},     \
        !          5654:     @{ "", 1@}@}
        !          5655: @end example
        !          5656: @end table
        !          5657: 
        !          5658: Sometimes certain combinations of command options do not make sense on a
        !          5659: particular target machine.  You can define a macro @code{OVERRIDE_OPTIONS}
        !          5660: to take account of this.  This macro, if defined, is executed once
        !          5661: just after all the command options have been parsed.
        !          5662: 
        !          5663: @node Storage Layout, Registers, Run-time Target, Machine Macros
        !          5664: @section Storage Layout
        !          5665: 
        !          5666: Note that the definitions of the macros in this table which are sizes or
        !          5667: alignments measured in bits do not need to be constant.  They can be C
        !          5668: expressions that refer to static variables, such as the @code{target_flags}.
        !          5669: @xref{Run-time Target}.
        !          5670: 
        !          5671: @table @code
        !          5672: @item BITS_BIG_ENDIAN
        !          5673: Define this macro if the most significant bit in a byte has the lowest
        !          5674: number.  This means that bit-field instructions count from the most
        !          5675: significant bit.  If the machine has no bit-field instructions, this
        !          5676: macro is irrelevant.
        !          5677: 
        !          5678: @item BYTES_BIG_ENDIAN
        !          5679: Define this macro if the most significant byte in a word has the
        !          5680: lowest number.
        !          5681: 
        !          5682: @item WORDS_BIG_ENDIAN
        !          5683: Define this macro if, in a multiword object, the most significant
        !          5684: word has the lowest number.
        !          5685: 
        !          5686: @item BITS_PER_UNIT
        !          5687: Number of bits in an addressable storage unit (byte); normally 8.
        !          5688: 
        !          5689: @item BITS_PER_WORD
        !          5690: Number of bits in a word; normally 32.
        !          5691: 
        !          5692: @item UNITS_PER_WORD
        !          5693: Number of storage units in a word; normally 4.
        !          5694: 
        !          5695: @item POINTER_SIZE
        !          5696: Width of a pointer, in bits.
        !          5697: 
        !          5698: @item PARM_BOUNDARY
        !          5699: Alignment required for function parameters on the stack, in bits.
        !          5700: 
        !          5701: @item STACK_BOUNDARY
        !          5702: Define this macro if you wish to preserve a certain alignment for
        !          5703: the stack pointer at all times.  The definition is a C expression
        !          5704: for the desired alignment (measured in bits).
        !          5705: 
        !          5706: @item FUNCTION_BOUNDARY
        !          5707: Alignment required for a function entry point, in bits.
        !          5708: 
        !          5709: @item BIGGEST_ALIGNMENT
        !          5710: Biggest alignment that any data type can require on this machine, in bits.
        !          5711: 
        !          5712: @item EMPTY_FIELD_ALIGNMENT
        !          5713: Alignment in bits to be given to a structure bit field that follows an
        !          5714: empty field such as @code{int : 0;}.
        !          5715: 
        !          5716: @item STRUCTURE_SIZE_BOUNDARY
        !          5717: Number of bits which any structure or union's size must be a multiple of.
        !          5718: Each structure or union's size is rounded up to a multiple of this.
        !          5719: 
        !          5720: If you do not define this macro, the default is the same as
        !          5721: @code{BITS_PER_UNIT}.
        !          5722: 
        !          5723: @item STRICT_ALIGNMENT
        !          5724: Define this if instructions will fail to work if given data not
        !          5725: on the nominal alignment.  If instructions will merely go slower
        !          5726: in that case, do not define this macro.
        !          5727: @end table
        !          5728: 
        !          5729: @node Registers, Register Classes, Storage Layout, Machine Macros
        !          5730: @section Register Usage
        !          5731: 
        !          5732: @table @code
        !          5733: @item FIRST_PSEUDO_REGISTER
        !          5734: Number of hardware registers known to the compiler.  They receive
        !          5735: numbers 0 through @code{FIRST_PSEUDO_REGISTER-1}; thus, the first
        !          5736: pseudo register's number really is assigned the number
        !          5737: @code{FIRST_PSEUDO_REGISTER}.
        !          5738: 
        !          5739: @item FIXED_REGISTERS
        !          5740: An initializer that says which registers are used for fixed purposes
        !          5741: all throughout the compiled code and are therefore not available for
        !          5742: general allocation.  These would include the stack pointer, the frame
        !          5743: pointer, the program counter on machines where that is considered one
        !          5744: of the addressable registers, and any other numbered register with a
        !          5745: standard use.
        !          5746: 
        !          5747: This information is expressed as a sequence of numbers, separated by
        !          5748: commas and surrounded by braces.  The @var{n}th number is 1 if
        !          5749: register @var{n} is fixed, 0 otherwise.
        !          5750: 
        !          5751: The table initialized from this macro, and the table initialized by
        !          5752: the following one, may be overridden at run time either automatically,
        !          5753: by the actions of the macro @code{CONDITIONAL_REGISTER_USAGE}, or by
        !          5754: the user with the command options @samp{-ffixed-@var{reg}},
        !          5755: @samp{-fcall-used-@var{reg}} and @samp{-fcall-saved-@var{reg}}.
        !          5756: 
        !          5757: @item CALL_USED_REGISTERS
        !          5758: Like @code{FIXED_REGISTERS} but has 1 for each register that is
        !          5759: clobbered (in general) by function calls as well as for fixed
        !          5760: registers.  This macro therefore identifies the registers that are not
        !          5761: available for general allocation of values that must live across
        !          5762: function calls.
        !          5763: 
        !          5764: If a register has 0 in @code{CALL_USED_REGISTERS}, the compiler
        !          5765: automatically saves it on function entry and restores it on function
        !          5766: exit, if the register is used within the function.
        !          5767: 
        !          5768: @item CONDITIONAL_REGISTER_USAGE
        !          5769: Zero or more C statements that may conditionally modify two variables
        !          5770: @code{fixed_regs} and @code{call_used_regs} (both of type @code{char
        !          5771: []}) after they have been initialized from the two preceding macros.
        !          5772: 
        !          5773: This is necessary in case the fixed or call-clobbered registers depend
        !          5774: on target flags.
        !          5775: 
        !          5776: You need not define this macro if it has no work to do.
        !          5777: 
        !          5778: @item HARD_REGNO_REGS (@var{regno}, @var{mode})
        !          5779: A C expression for the number of consecutive hard registers, starting
        !          5780: at register number @var{regno}, required to hold a value of mode
        !          5781: @var{mode}.
        !          5782: 
        !          5783: On a machine where all registers are exactly one word, a suitable
        !          5784: definition of this macro is
        !          5785: 
        !          5786: @example
        !          5787: #define HARD_REGNO_NREGS(REGNO, MODE)            \
        !          5788:    ((GET_MODE_SIZE (MODE) + UNITS_PER_WORD - 1)  \
        !          5789:     / UNITS_PER_WORD))
        !          5790: @end example
        !          5791: 
        !          5792: @item HARD_REGNO_MODE_OK (@var{regno}, @var{mode})
        !          5793: A C expression that is nonzero if it is permissible to store a value
        !          5794: of mode @var{mode} in hard register number @var{regno} (or in several
        !          5795: registers starting with that one).  For a machine where all registers
        !          5796: are equivalent, a suitable definition is
        !          5797: 
        !          5798: @example
        !          5799: #define HARD_REGNO_MODE_OK(REGNO, MODE) 1
        !          5800: @end example
        !          5801: 
        !          5802: It is not necessary for this macro to check for fixed register numbers
        !          5803: because the allocation mechanism considers them to be always occupied.
        !          5804: 
        !          5805: Many machines have special registers for floating point arithmetic.
        !          5806: Often people assume that floating point machine modes are allowed only
        !          5807: in floating point registers.  This is not true.  Any registers that
        !          5808: can hold integers can safely @emph{hold} a floating point machine
        !          5809: mode, whether or not floating arithmetic can be done on it in those
        !          5810: registers.
        !          5811: 
        !          5812: The true significance of special floating registers is rather than
        !          5813: non-floating-point machine modes @emph{may not} go in those registers.
        !          5814: This is true if the floating registers normalize any value stored in
        !          5815: them, because storing a non-floating value there would garble it.  If
        !          5816: the floating registers do not automatically normalize, if you can
        !          5817: store any bit pattern in one and retrieve it unchanged without a trap,
        !          5818: then any machine mode may go in a floating register and this macro
        !          5819: should say so.
        !          5820: 
        !          5821: Sometimes there are floating registers that are especially slow to
        !          5822: access, so that it is better to store a value in a stack frame than in
        !          5823: such a register if floating point arithmetic is not being done.  As long
        !          5824: as the floating registers are not in class @code{GENERAL_REGS}, they
        !          5825: will not be used unless some insn's constraint asks for one.
        !          5826: 
        !          5827: It is obligatory to support floating point `move' instructions into
        !          5828: and out of general registers, because unions and structures (which
        !          5829: have modes @samp{SImode} or @samp{DImode}) can be in those registers
        !          5830: and they may have floating point members.
        !          5831: 
        !          5832: @item MODES_TIEABLE_P (@var{mode1}, @var{mode2})
        !          5833: A C expression that is nonzero if it is desirable to choose register
        !          5834: allocation so as to avoid move instructions between a value of mode
        !          5835: @var{mode1} and a value of mode @var{mode2}.
        !          5836: 
        !          5837: If @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode1})} and
        !          5838: @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode2})} are ever different
        !          5839: for any @var{r}, then @code{MODES_TIEABLE_P (@var{mode1},
        !          5840: @var{mode2})} must be zero.
        !          5841: 
        !          5842: @item PC_REGNUM
        !          5843: If the program counter has a register number, define this as that
        !          5844: register number.  Otherwise, do not define it.
        !          5845: 
        !          5846: @item STACK_POINTER_REGNUM
        !          5847: The register number of the stack pointer register, which must also be
        !          5848: a fixed register according to @code{FIXED_REGISTERS}.  On many
        !          5849: machines, the hardware determines which register this is.
        !          5850: 
        !          5851: @item FRAME_POINTER_REGNUM
        !          5852: The register number of the frame pointer register, which is used to
        !          5853: access automatic variables in the stack frame.  On some machines, the
        !          5854: hardware determines which register this is.  On other machines, you
        !          5855: can choose any register you wish for this purpose.
        !          5856: 
        !          5857: @item FRAME_POINTER_REQUIRED
        !          5858: A C expression which is nonzero if a function must have and use a
        !          5859: frame pointer.  This expression is evaluated in the reload pass, in
        !          5860: the function @code{reload}, and it can in principle examine the
        !          5861: current function and decide according to the facts, but on most
        !          5862: machines the constant 0 or the constant 1 suffices.  Use 0 when the
        !          5863: machine allows code to be generated with no frame pointer, and doing
        !          5864: so saves some time or space.  Use 1 when there is no possible
        !          5865: advantage to avoiding a frame pointer.
        !          5866: 
        !          5867: In certain cases, the compiler does not know how to do without a frame
        !          5868: pointer.  The compiler recognizes those cases and automatically gives
        !          5869: the function a frame pointer regardless of what
        !          5870: @code{FRAME_POINTER_REQUIRED} says.  You don't need to worry about
        !          5871: them.@refill
        !          5872: 
        !          5873: In a function that does not require a frame pointer, the frame pointer
        !          5874: register can be allocated for ordinary usage, provided it is not
        !          5875: marked as a fixed register.  See @code{FIXED_REGISTERS} for more
        !          5876: information.
        !          5877: 
        !          5878: @item ARG_POINTER_REGNUM
        !          5879: The register number of the arg pointer register, which is used to
        !          5880: access the function's argument list.  On some machines, this is the
        !          5881: same as the frame pointer register.  On some machines, the hardware
        !          5882: determines which register this is.  On other machines, you can choose
        !          5883: any register you wish for this purpose.  It must in any case be a
        !          5884: fixed register according to @code{FIXED_REGISTERS}.
        !          5885: 
        !          5886: @item STATIC_CHAIN_REGNUM
        !          5887: The register number used for passing a function's static chain
        !          5888: pointer.  This is needed for languages such as Pascal and Algol where
        !          5889: functions defined within other functions can access the local
        !          5890: variables of the outer functions; it is not currently used because C
        !          5891: does not provide this feature.
        !          5892: 
        !          5893: The static chain register need not be a fixed register.
        !          5894: 
        !          5895: @item STRUCT_VALUE_REGNUM
        !          5896: When a function's value's mode is @code{BLKmode}, the value is not
        !          5897: returned according to @code{FUNCTION_VALUE}.  Instead, the caller
        !          5898: passes the address of a block of memory in which the value should be
        !          5899: stored.  @code{STRUCT_VALUE_REGNUM} is the register in which this
        !          5900: address is passed.
        !          5901: @end table
        !          5902: 
        !          5903: @node Register Classes, Stack Layout, Registers, Machine Macros
        !          5904: @section Register Classes
        !          5905: 
        !          5906: On many machines, the numbered registers are not all equivalent.
        !          5907: For example, certain registers may not be allowed for indexed addressing;
        !          5908: certain registers may not be allowed in some instructions.  These machine
        !          5909: restrictions are described to the compiler using @dfn{register classes}.
        !          5910: 
        !          5911: You define a number of register classes, giving each one a name and saying
        !          5912: which of the registers belong to it.  Then you can specify register classes
        !          5913: that are allowed as operands to particular instruction patterns.
        !          5914: 
        !          5915: In general, each register will belong to several classes.  In fact, one
        !          5916: class must be named @code{ALL_REGS} and contain all the registers.  Another
        !          5917: class must be named @code{NO_REGS} and contain no registers.  Often the
        !          5918: union of two classes will be another class; however, this is not required.
        !          5919: 
        !          5920: One of the classes must be named @code{GENERAL_REGS}.  There is nothing
        !          5921: terribly special about the name, but the operand constraint letters
        !          5922: @samp{r} and @samp{g} specify this class.  If @code{GENERAL_REGS} is
        !          5923: the same as @code{ALL_REGS}, just define it as a macro which expands
        !          5924: to @code{ALL_REGS}.
        !          5925: 
        !          5926: The way classes other than @code{GENERAL_REGS} are specified in operand
        !          5927: constraints is through machine-dependent operand constraint letters.
        !          5928: You can define such letters to correspond to various classes, then use
        !          5929: them in operand constraints.
        !          5930: 
        !          5931: You should define a class for the union of two classes whenever some
        !          5932: instruction allows both classes.  For example, if an instruction allows
        !          5933: either a floating-point (coprocessor) register or a general register for a
        !          5934: certain operand, you should define a class @code{FLOAT_OR_GENERAL_REGS}
        !          5935: which includes both of them.  Otherwise you will get suboptimal code.
        !          5936: 
        !          5937: You must also specify certain redundant information about the register
        !          5938: classes: for each class, which classes contain it and which ones are
        !          5939: contained in it; for each pair of classes, the largest class contained
        !          5940: in their union.
        !          5941: 
        !          5942: @table @code
        !          5943: @item enum reg_class
        !          5944: An enumeral type that must be defined with all the register class names
        !          5945: as enumeral values.  @code{NO_REGS} must be first.  @code{ALL_REGS}
        !          5946: must be the last register class, followed by one more enumeral value,
        !          5947: @code{LIM_REG_CLASSES}, which is not a register class but rather
        !          5948: tells how many classes there are.
        !          5949: 
        !          5950: Each register class has a number, which is the value of casting
        !          5951: the class name to type @code{int}.  The number serves as an index
        !          5952: in many of the tables described below.
        !          5953: 
        !          5954: @item REG_CLASS_NAMES
        !          5955: An initializer containing the names of the register classes as C string
        !          5956: constants.  These names are used in writing some of the debugging dumps.
        !          5957: 
        !          5958: @item REG_CLASS_CONTENTS
        !          5959: An initializer containing the contents of the register classes, as integers
        !          5960: which are bit masks.  The @var{n}th integer specifies the contents of class
        !          5961: @var{n}.  The way the integer @var{mask} is interpreted is that
        !          5962: register @var{r} is in the class if @code{@var{mask} & (1 << @var{r})} is 1.
        !          5963: 
        !          5964: When the machine has more than 32 registers, an integer does not suffice.
        !          5965: Then the integers are replaced by sub-initializers, braced groupings containing
        !          5966: several integers.  Each sub-initializer must be suitable as an initializer
        !          5967: for the type @code{HARD_REG_SET} which is defined in @file{hard-reg-set.h}.
        !          5968: 
        !          5969: @item REGNO_REG_CLASS (@var{regno})
        !          5970: A C expression whose value is a register class containing hard register
        !          5971: @var{regno}.  In general there is more that one such class; choose a class
        !          5972: which is @dfn{minimal}, meaning that no smaller class also contains the
        !          5973: register.
        !          5974: 
        !          5975: @item INDEX_REG_CLASS
        !          5976: A macro whose definition is the name of the class to which a valid index
        !          5977: register must belong.
        !          5978: 
        !          5979: @item REG_CLASS_FROM_LETTER (@var{char})
        !          5980: A C expression which defines the machine-dependent operand constraint
        !          5981: letters for register classes.  If @var{char} is such a letter, the value
        !          5982: should be the register class corresponding to it.  Otherwise, the value
        !          5983: should be @code{NO_REGS}.
        !          5984: 
        !          5985: @item REGNO_OK_FOR_BASE_P (@var{num})
        !          5986: A C expression which is nonzero if register number @var{num} is
        !          5987: suitable for use as a base register in operand addresses.  It may be
        !          5988: either a suitable hard register or a pseudo register that has been
        !          5989: allocated such a hard register.
        !          5990: 
        !          5991: @item REGNO_OK_FOR_INDEX_P (@var{num})
        !          5992: A C expression which is nonzero if register number @var{num} is
        !          5993: suitable for use as an index register in operand addresses.  It may be
        !          5994: either a suitable hard register or a pseudo register that has been
        !          5995: allocated such a hard register.
        !          5996: 
        !          5997: The difference between an index register and a base register is that
        !          5998: the index register may be scaled.  If an address involves the sum of
        !          5999: two registers, neither one of them scaled, then either one may be
        !          6000: labeled the ``base'' and the other the ``index''; but whichever
        !          6001: labeling is used must fit the machine's constraints of which registers
        !          6002: may serve in each capacity.  The compiler will try both labelings,
        !          6003: looking for one that is valid, and reload one or both registers only
        !          6004: if neither labeling works.
        !          6005: 
        !          6006: @item PREFERRED_RELOAD_CLASS (@var{x}, @var{class})
        !          6007: A C expression that places additional restrictions on the register class
        !          6008: to use when it is necessary to copy value @var{x} into a register in class
        !          6009: @var{class}.  The value is a register class; perhaps @var{class}, or perhaps
        !          6010: another, smaller class.  @var{class} is always safe as a value.  In fact,
        !          6011: the definition
        !          6012: 
        !          6013: @example
        !          6014: #define PREFERRED_RELOAD_CLASS(X,CLASS) CLASS
        !          6015: @end example
        !          6016: 
        !          6017: @noindent
        !          6018: is always safe.  However, sometimes returning a more restrictive class
        !          6019: makes better code.  For example, on the 68000, when @var{x} is an
        !          6020: integer constant that is in range for a @samp{moveq} instruction,
        !          6021: the value of this macro is always @code{DATA_REGS} as long as
        !          6022: @var{class} includes the data registers.  Requiring a data register
        !          6023: guarantees that a @samp{moveq} will be used.
        !          6024: 
        !          6025: @item CLASS_MAX_NREGS (@var{class}, @var{mode})
        !          6026: A C expression for the maximum number of consecutive registers
        !          6027: of class @var{class} needed to hold a value of mode @var{mode}.
        !          6028: 
        !          6029: This is closely related to the macro @code{HARD_REGNO_NREGS}.
        !          6030: In fact, the value of the macro @code{CLASS_MAX_NREGS (@var{class}, @var{mode})}
        !          6031: should be the maximum value of @code{HARD_REGNO_NREGS (@var{regno}, @var{mode})}
        !          6032: for all @var{regno} values in the class @var{class}.
        !          6033: 
        !          6034: This macro helps control the handling of multiple-word values
        !          6035: in the reload pass.
        !          6036: @end table
        !          6037: 
        !          6038: Two other special macros describe which constants fit which constraint
        !          6039: letters.
        !          6040: 
        !          6041: @table @code
        !          6042: @item CONST_OK_FOR_LETTER_P (@var{value}, @var{c})
        !          6043: A C expression that defines the machine-dependent operand constraint letters
        !          6044: that specify particular ranges of integer values.  If @var{c} is one
        !          6045: of those letters, the expression should check that @var{value}, an integer,
        !          6046: is in the appropriate range and return 1 if so, 0 otherwise.  If @var{c} is
        !          6047: not one of those letters, the value should be 0 regardless of @var{value}.
        !          6048: 
        !          6049: @item CONST_DOUBLE_OK_FOR_LETTER_P (@var{value}, @var{c})
        !          6050: A C expression that defines the machine-dependent operand constraint
        !          6051: letters that specify particular ranges of floating values.  If @var{c} is
        !          6052: one of those letters, the expression should check that @var{value}, an RTX
        !          6053: of code @samp{const_double}, is in the appropriate range and return 1 if
        !          6054: so, 0 otherwise.  If @var{c} is not one of those letters, the value should
        !          6055: be 0 regardless of @var{value}.
        !          6056: @end table
        !          6057: 
        !          6058: @node Stack Layout, Library Names, Register Classes, Machine Macros
        !          6059: @section Describing Stack Layout
        !          6060: 
        !          6061: @table @code
        !          6062: @item STACK_GROWS_DOWNWARD
        !          6063: Define this macro if pushing a word onto the stack moves the stack
        !          6064: pointer to a smaller address.
        !          6065: 
        !          6066: When we say, ``define this macro if @dots{},'' it means that the
        !          6067: compiler checks this macro only with @code{#ifdef} so the precise
        !          6068: definition used does not matter.
        !          6069: 
        !          6070: @item FRAME_GROWS_DOWNWARD
        !          6071: Define this macro if the addresses of local variable slots are at negative
        !          6072: offsets from the frame pointer.
        !          6073: 
        !          6074: @item STARTING_FRAME_OFFSET
        !          6075: Offset from the frame pointer to the first local variable slot to be allocated.
        !          6076: 
        !          6077: If @code{FRAME_GROWS_DOWNWARD}, the next slot's offset is found by
        !          6078: subtracting the length of the first slot from @code{STARTING_FRAME_OFFSET}.
        !          6079: Otherwise, it is found by adding the length of the first slot to
        !          6080: the value @code{STARTING_FRAME_OFFSET}.
        !          6081: 
        !          6082: @item PUSH_ROUNDING (@var{npushed})
        !          6083: A C expression that is the number of bytes actually pushed onto the
        !          6084: stack when an instruction attempts to push @var{npushed} bytes.
        !          6085: 
        !          6086: If the target machine does not have a push instruction, do not define
        !          6087: this macro.  That directs GNU CC to use an alternate strategy: to
        !          6088: allocate the entire argument block and then store the arguments into
        !          6089: it.
        !          6090: 
        !          6091: On some machines, the definition
        !          6092: 
        !          6093: @example
        !          6094: #define PUSH_ROUNDING(BYTES) (BYTES)
        !          6095: @end example
        !          6096: 
        !          6097: @noindent
        !          6098: will suffice.  But on other machines, instructions that appear
        !          6099: to push one byte actually push two bytes in an attempt to maintain
        !          6100: alignment.  Then the definition should be
        !          6101: 
        !          6102: @example
        !          6103: #define PUSH_ROUNDING(BYTES) (((BYTES) + 1) & ~1)
        !          6104: @end example
        !          6105: 
        !          6106: @item FIRST_PARM_OFFSET
        !          6107: Offset from the argument pointer register to the first argument's address.
        !          6108: 
        !          6109: @item RETURN_POPS_ARGS (@var{funtype})
        !          6110: A C expression that should be 1 if a function pops its own arguments
        !          6111: on returning, or 0 if the function pops no arguments and the caller
        !          6112: must therefore pop them all after the function returns.
        !          6113: 
        !          6114: @var{funtype} is a C variable whose value is a tree node that
        !          6115: describes the function in question.  Normally it is a node of type
        !          6116: @code{FUNCTION_TYPE} that describes the data type of the function.
        !          6117: From this it is possible to obtain the data types of the value and
        !          6118: arguments (if known).
        !          6119: 
        !          6120: When a call to a library function is being considered, @var{funtype}
        !          6121: will contain an identifier node for the library function.  Thus, if
        !          6122: you need to distinguish among various library functions, you can do so
        !          6123: by their names.  Note that ``library function'' in this context means
        !          6124: a function used to perform arithmetic, whose name is known specially
        !          6125: in the compiler and was not mentioned in the C code being compiled.
        !          6126: 
        !          6127: On the Vax, all functions always pop their arguments, so the
        !          6128: definition of this macro is 1.  On the 68000, using the standard
        !          6129: calling convention, no functions pop their arguments, so the value of
        !          6130: the macro is always 0 in this case.  But an alternative calling
        !          6131: convention is available in which functions that take a fixed number of
        !          6132: arguments pop them but other functions (such as @code{printf}) pop
        !          6133: nothing (the caller pops all).  When this convention is in use,
        !          6134: @var{funtype} is examined to determine whether a function takes a
        !          6135: fixed number of arguments.
        !          6136: 
        !          6137: @item FUNCTION_VALUE (@var{valtype}, @var{func})
        !          6138: A C expression to create an RTX representing the place where a
        !          6139: function returns a value of data type @var{valtype}.  @var{valtype} is
        !          6140: a tree node representing a data type.  Write @code{TYPE_MODE
        !          6141: (@var{valtype})} to get the machine mode used to represent that type.
        !          6142: On many machines, only the mode is relevant.  (Actually, on most
        !          6143: machines, scalar values are returned in the same place regardless of
        !          6144: mode).@refill
        !          6145: 
        !          6146: If the precise function being called is known, @var{func} is a tree
        !          6147: node (@code{FUNCTION_DECL}) for it; otherwise, @var{func} is a null
        !          6148: pointer.  This makes it possible to use a different value-returning
        !          6149: convention for specific functions when all their calls are
        !          6150: known.@refill
        !          6151: 
        !          6152: @item FUNCTION_OUTGOING_VALUE (@var{valtype}, @var{func})
        !          6153: Define this macro if the target machine has ``register windows''
        !          6154: so that the register in which a function returns its value is not
        !          6155: the same as the one in which the caller sees the value.
        !          6156: 
        !          6157: For such machines, @code{FUNCTION_VALUE} computes the register in
        !          6158: which the caller will see the value, and
        !          6159: @code{FUNCTION_OUTGOING_VALUE} should be defined in a similar fashion
        !          6160: to tell the function where to put the value.@refill
        !          6161: 
        !          6162: If @code{FUNCTION_OUTGOING_VALUE} is not defined,
        !          6163: @code{FUNCTION_VALUE} serves both purposes.@refill
        !          6164: 
        !          6165: @item LIBCALL_VALUE (@var{mode})
        !          6166: A C expression to create an RTX representing the place where a library
        !          6167: function returns a value of mode @var{mode}.  If the precise function
        !          6168: being called is known, @var{func} is a tree node
        !          6169: (@code{FUNCTION_DECL}) for it; otherwise, @var{func} is a null
        !          6170: pointer.  This makes it possible to use a different value-returning
        !          6171: convention for specific functions when all their calls are
        !          6172: known.@refill
        !          6173: 
        !          6174: Note that ``library function'' in this context means a compiler
        !          6175: support routine, used to perform arithmetic, whose name is known
        !          6176: specially by the compiler and was not mentioned in the C code being
        !          6177: compiled.
        !          6178: 
        !          6179: @item FUNCTION_VALUE_REGNO_P (@var{regno})
        !          6180: A C expression that is nonzero if @var{regno} is the number of a hard
        !          6181: register in which function values are sometimes returned.
        !          6182: 
        !          6183: A register whose use for returning values is limited to serving as the
        !          6184: second of a pair (for a value of type @code{double}, say) need not be
        !          6185: recognized by this macro.  So for most machines, this definition
        !          6186: suffices:
        !          6187: 
        !          6188: @example
        !          6189: #define FUNCTION_VALUE_REGNO_P(N) ((N) == 0)
        !          6190: @end example
        !          6191: 
        !          6192: @item FUNCTION_ARG (@var{cum}, @var{mode}, @var{type}, @var{named})
        !          6193: A C expression that controls whether a function argument is passed
        !          6194: in a register, and which register.
        !          6195: 
        !          6196: The arguments are @var{cum}, which summarizes all the previous
        !          6197: arguments; @var{mode}, the machine mode of the argument; @var{type},
        !          6198: the data type of the argument as a tree node or 0 if that is not known
        !          6199: (which happens for C support library functions); and @var{named},
        !          6200: which is 1 for an ordinary argument and 0 for nameless arguments that
        !          6201: correspond to @samp{...} in the called function's prototype.
        !          6202: 
        !          6203: The value of the expression should either be a @samp{reg} RTX for the
        !          6204: hard register in which to pass the argument, or zero to pass the
        !          6205: argument on the stack.
        !          6206: 
        !          6207: For the Vax and 68000, where normally all arguments are pushed, zero
        !          6208: suffices as a definition.
        !          6209: 
        !          6210: @item FUNCTION_INCOMING_ARG (@var{cum}, @var{mode}, @var{type}, @var{named})
        !          6211: Define this macro if the target machine has ``register windows'', so
        !          6212: that the register in which a function sees an arguments is not
        !          6213: necessarily the same as the one in which the caller passed the
        !          6214: argument.
        !          6215: 
        !          6216: For such machines, @code{FUNCTION_ARG} computes the register in which
        !          6217: the caller passes the value, and @code{FUNCTION_INCOMING_ARG} should
        !          6218: be defined in a similar fashion to tell the function being called
        !          6219: where the arguments will arrive.
        !          6220: 
        !          6221: If @code{FUNCTION_INCOMING_ARG} is not defined, @code{FUNCTION_ARG}
        !          6222: serves both purposes.@refill
        !          6223: 
        !          6224: @item FUNCTION_ARG_PARTIAL_NREGS (@var{cum}, @var{mode}, @var{type}, @var{named})
        !          6225: A C expression for the number of words, at the beginning of an
        !          6226: argument, must be put in registers.  The value must be zero for
        !          6227: arguments that are passed entirely in registers or that are entirely
        !          6228: pushed on the stack.
        !          6229: 
        !          6230: On some machines, certain arguments must be passed partially in
        !          6231: registers and partially in memory.  On these machines, typically the
        !          6232: first @var{n} words of arguments are passed in registers, and the rest
        !          6233: on the stack.  If a multi-word argument (a @code{double} or a
        !          6234: structure) crosses that boundary, its first few words must be passed
        !          6235: in registers and the rest must be pushed.  This macro tells the
        !          6236: compiler when this occurs, and how many of the words should go in
        !          6237: registers.
        !          6238: 
        !          6239: @code{FUNCTION_ARG} for these arguments should return the first
        !          6240: register to be used by the caller for this argument; likewise
        !          6241: @code{FUNCTION_INCOMING_ARG}, for the called function.
        !          6242: 
        !          6243: @item CUMULATIVE_ARGS
        !          6244: A C type for declaring a variable that is used as the first argument
        !          6245: of @code{FUNCTION_ARG} and other related values.  For some target
        !          6246: machines, the type @code{int} suffices and can hold the number of
        !          6247: bytes of argument so far.
        !          6248: 
        !          6249: @item INIT_CUMULATIVE_ARGS (@var{cum})
        !          6250: A C statement (sans semicolon) for initializing the variable @var{cum}
        !          6251: for the state at the beginning of the argument list.  The variable has
        !          6252: type @code{CUMULATIVE_ARGS}.
        !          6253: 
        !          6254: @item FUNCTION_ARG_ADVANCE (@var{cum}, @var{mode}, @var{type}, @var{named})
        !          6255: Update the summarizer variable @var{cum} to advance past an argument
        !          6256: in the argument list.  The values @var{mode}, @var{type} and
        !          6257: @var{named} describe that argument.  Once this is done, the variable
        !          6258: @var{cum} is suitable for analyzing the @emph{following} argument
        !          6259: with @code{FUNCTION_ARG}, etc.@refill
        !          6260: 
        !          6261: @item FUNCTION_ARG_REGNO_P (@var{regno})
        !          6262: A C expression that is nonzero if @var{regno} is the number of a hard
        !          6263: register in which function arguments are sometimes passed.  This does
        !          6264: @emph{not} include implicit arguments such as the static chain and
        !          6265: the structure-value address.  On many machines, no registers can be
        !          6266: used for this purpose since all function arguments are pushed on the
        !          6267: stack.
        !          6268: 
        !          6269: @item FUNCTION_PROLOGUE (@var{file}, @var{size})
        !          6270: A C compound statement that outputs the assembler code for entry to a
        !          6271: function.  The prologue is responsible for setting up the stack frame,
        !          6272: initializing the frame pointer register, saving registers that must be
        !          6273: saved, and allocating @var{size} additional bytes of storage for the
        !          6274: local variables.  @var{size} is an integer.  @var{file} is a stdio
        !          6275: stream to which the assembler code should be output.
        !          6276: 
        !          6277: The label for the beginning of the function need not be output by this
        !          6278: macro.  That has already been done when the macro is run.
        !          6279: 
        !          6280: To determine which registers to save, the macro can refer to the array
        !          6281: @code{regs_ever_live}: element @var{r} is nonzero if hard register
        !          6282: @var{r} is used anywhere within the function.  This implies the
        !          6283: function prologue should save register @var{r}, but not if it is one
        !          6284: of the call-used registers.
        !          6285: 
        !          6286: On machines where functions may or may not have frame-pointers, the
        !          6287: function entry code must vary accordingly; it must set up the frame
        !          6288: pointer if one is wanted, and not otherwise.  To determine whether a
        !          6289: frame pointer is in wanted, the macro can refer to the variable
        !          6290: @code{frame_pointer_needed}.  The variable's value will be 1 at run
        !          6291: time in a function that needs a frame pointer.
        !          6292: 
        !          6293: @item FUNCTION_PROFILER (@var{file}, @var{labelno})
        !          6294: A C statement or compound statement to output to @var{file} some
        !          6295: assembler code to call the profiling subroutine @code{mcount}.
        !          6296: Before calling, the assembler code must load the address of a
        !          6297: counter variable into a register where @code{mcount} expects to
        !          6298: find the address.  The name of this variable is @samp{LP} followed
        !          6299: by the number @var{labelno}, so you would generate the name using
        !          6300: @samp{LP%d} in a @code{fprintf}.
        !          6301: 
        !          6302: The details of how the address should be passed to @code{mcount} are
        !          6303: determined by your operating system environment, not by GNU CC.  To
        !          6304: figure them out, compile a small program for profiling using the
        !          6305: system's installed C compiler and look at the assembler code that
        !          6306: results.
        !          6307: 
        !          6308: @item EXIT_IGNORES_STACK
        !          6309: Define this macro as a C expression that is nonzero if the return
        !          6310: instruction or the function epilogue ignores the value of the stack
        !          6311: pointer; in other words, if it is safe to delete an instruction to
        !          6312: adjust the stack pointer before a return from the function.
        !          6313: 
        !          6314: Note that this macro's value is relevant only for for which frame
        !          6315: pointers are maintained.  It is never possible to delete a final stack
        !          6316: adjustment in a function that has no frame pointer, and the compiler
        !          6317: knows this regardless of @code{EXIT_IGNORES_STACK}.
        !          6318: 
        !          6319: @item FUNCTION_EPILOGUE (@var{file}, @var{size})
        !          6320: A C compound statement that outputs the assembler code for exit from a
        !          6321: function.  The epilogue is responsible for restoring the saved
        !          6322: registers and stack pointer to their values when the function was
        !          6323: called, and returning control to the caller.  This macro takes the
        !          6324: same arguments as the macro @code{FUNCTION_PROLOGUE}, and the
        !          6325: registers to restore are determined from @code{regs_ever_live} and
        !          6326: @code{CALL_USED_REGISTERS} in the same way.
        !          6327: 
        !          6328: On some machines, there is a single instruction that does all the work
        !          6329: of returning from the function.  On these machines, give that
        !          6330: instruction the name @samp{return} and do not define the macro
        !          6331: @code{FUNCTION_EPILOGUE} at all.
        !          6332: 
        !          6333: On machines where functions may or may not have frame-pointers, the
        !          6334: function exit code must vary accordingly.  Sometimes the code for
        !          6335: these two cases is completely different.  To determine whether a frame
        !          6336: pointer is in wanted, the macro can refer to the variable
        !          6337: @code{frame_pointer_needed}.  The variable's value will be 1 at run
        !          6338: time in a function that needs a frame pointer.
        !          6339: 
        !          6340: On some machines, some functions pop their arguments on exit while
        !          6341: others leave that for the caller to do.  For example, the 68020 when
        !          6342: given @samp{-mrtd} pops arguments in functions that take a fixed
        !          6343: number of arguments.
        !          6344: 
        !          6345: Your definition of the macro @code{RETURN_POPS_ARGS} decides which
        !          6346: functions pop their own arguments.  @code{FUNCTION_EPILOGUE} needs to
        !          6347: know what was decided.  The variable @code{current_function_pops_args}
        !          6348: is nonzero if the function should pop its own arguments.  If so, use
        !          6349: the variable @code{current_function_args_size} as the number of bytes
        !          6350: to pop.
        !          6351: 
        !          6352: @item FIX_FRAME_POINTER_ADDRESS (@var{addr}, @var{depth})
        !          6353: A C compound statement to alter a memory address that uses the frame
        !          6354: pointer register so that it uses the stack pointer register instead.
        !          6355: This must be done in the instructions that load parameter values into
        !          6356: registers, when the reload pass determines that a frame pointer is not
        !          6357: necessary for the function.  @var{addr} will be a C variable name, and
        !          6358: the updated address should be stored in that variable.  @var{depth}
        !          6359: will be the current depth of stack temporaries (number of bytes of
        !          6360: arguments currently pushed).  The change in offset between a
        !          6361: frame-pointer-relative address and a stack-pointer-relative address
        !          6362: must include @var{depth}.
        !          6363: 
        !          6364: Even if your machine description specifies there will always be a
        !          6365: frame pointer in the frame pointer register, you must still define
        !          6366: @code{FIX_FRAME_POINTER_ADDRESS}, but the definition will never be
        !          6367: executed at run time, so it may be empty.
        !          6368: @end table
        !          6369: 
        !          6370: @node Library Names, Addressing Modes, Stack Layout, Machine Macros
        !          6371: @section Library Subroutine Names
        !          6372: 
        !          6373: @table @code
        !          6374: @item UDIVSI3_LIBCALL
        !          6375: A C string constant giving the name of the function to call for
        !          6376: division of a full-word by a full-word.  If you do not define this
        !          6377: macro, the default name is used, which is @code{_udivsi3}, a function
        !          6378: defined in @file{gnulib}.
        !          6379: 
        !          6380: @item UMODSI3_LIBCALL
        !          6381: A C string constant giving the name of the function to call for the
        !          6382: remainder in division of a full-word by a full-word.  If you do not
        !          6383: define this macro, the default name is used, which is @code{_umodsi3},
        !          6384: a function defined in @file{gnulib}.
        !          6385: 
        !          6386: @item TARGET_MEM_FUNCTIONS
        !          6387: Define this macro if GNU CC should generate calls to the System V
        !          6388: (and ANSI C) library functions @code{memcpy} and @code{memset}
        !          6389: rather than the BSD functions @code{bcopy} and @code{bzero}.
        !          6390: @end table
        !          6391: 
        !          6392: @node Addressing Modes, Misc, Library Names, Machine Macros
        !          6393: @section Addressing Modes
        !          6394: 
        !          6395: @table @code
        !          6396: @item HAVE_POST_INCREMENT
        !          6397: Define this macro if the machine supports post-increment addressing.
        !          6398: 
        !          6399: @item HAVE_PRE_INCREMENT
        !          6400: @itemx HAVE_POST_DECREMENT
        !          6401: @itemx HAVE_PRE_DECREMENT
        !          6402: Similar for other kinds of addressing.
        !          6403: 
        !          6404: @item CONSTANT_ADDRESS_P (@var{x})
        !          6405: A C expression that is 1 if the RTX @var{x} is a constant whose value
        !          6406: is an integer.  This includes integers whose values are not explicitly
        !          6407: known, such as @samp{symbol_ref} and @samp{label_ref} expressions and
        !          6408: @samp{const} arithmetic expressions.
        !          6409: 
        !          6410: On most machines, this can be defined as @code{CONSTANT_P (@var{x})},
        !          6411: but a few machines are more restrictive in which constant addresses
        !          6412: are supported.
        !          6413: 
        !          6414: @item MAX_REGS_PER_ADDRESS
        !          6415: A number, the maximum number of registers that can appear in a valid
        !          6416: memory address.
        !          6417: 
        !          6418: @item GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{label})
        !          6419: A C compound statement with a conditional @code{goto @var{label};}
        !          6420: executed if @var{x} (an RTX) is a legitimate memory address on the
        !          6421: target machine for a memory operand of mode @var{mode}.
        !          6422: 
        !          6423: It usually pays to define several simpler macros to serve as
        !          6424: subroutines for this one.  Otherwise it may be too complicated to
        !          6425: understand.
        !          6426: 
        !          6427: This macro must exist in two variants: a strict variant and a
        !          6428: non-strict one.  The strict variant is used in the reload pass.  It
        !          6429: must be defined so that any pseudo-register that has not been
        !          6430: allocated a hard register is considered a memory reference.  In
        !          6431: contexts where some kind of register is required, a pseudo-register
        !          6432: with no hard register must be rejected.
        !          6433: 
        !          6434: The non-strict variant is used in other passes.  It must be defined to
        !          6435: accept all pseudo-registers in every context where some kind of
        !          6436: register is required.
        !          6437: 
        !          6438: Compiler source files that want to use the strict variant of this
        !          6439: macro define the macro @code{REG_OK_STRICT}.  You should use an
        !          6440: @code{#ifdef REG_OK_STRICT} conditional to define the strict variant
        !          6441: in that case and the non-strict variant otherwise.
        !          6442: 
        !          6443: Typically among the subroutines used to define
        !          6444: @code{GO_IF_LEGITIMATE_ADDRESS} are subroutines to check for
        !          6445: acceptable registers for various purposes (one for base registers, one
        !          6446: for index registers, and so on).  Then only these subroutine macros
        !          6447: need have two variants; the higher levels of macros may be the same
        !          6448: whether strict or not.@refill
        !          6449: 
        !          6450: @item LEGITIMIZE_ADDRESS (@var{x}, @var{oldx}, @var{mode}, @var{win})
        !          6451: A C compound statement that attempts to replace @var{x} with a valid
        !          6452: memory address for an operand of mode @var{mode}.  @var{win} will be a
        !          6453: C statement label elsewhere in the code; the macro definition may use
        !          6454: 
        !          6455: @example
        !          6456: GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{win});
        !          6457: @end example
        !          6458: 
        !          6459: @noindent
        !          6460: to avoid further processing if the address has become legitimate.
        !          6461: 
        !          6462: @var{x} will always be the result of a call to @code{break_out_memory_refs},
        !          6463: and @var{oldx} will be the operand that was given to that function to produce
        !          6464: @var{x}.
        !          6465: 
        !          6466: The code generated by this macro should not alter the substructure of
        !          6467: @var{x}.  If it transforms @var{x} into a more legitimate form, it
        !          6468: should assign @var{x} (which will always be a C variable) a new value.
        !          6469: 
        !          6470: It is not necessary for this macro to come up with a legitimate
        !          6471: address.  The compiler has standard ways of doing so in all cases.  In
        !          6472: fact, it is safe for this macro to do nothing.  But often a
        !          6473: machine-dependent strategy can generate better code.
        !          6474: 
        !          6475: @item GO_IF_MODE_DEPENDENT_ADDRESS (@var{addr}, @var{label})
        !          6476: A C statement or compound statement with a conditional @code{goto
        !          6477: @var{label};} executed if memory address @var{x} (an RTX) can have
        !          6478: different meanings depending on the machine mode of the memory
        !          6479: reference it is used for.
        !          6480: 
        !          6481: Autoincrement and autodecrement addresses typically have mode-dependent
        !          6482: effects because the amount of the increment or decrement is the size
        !          6483: of the operand being addressed.  Some machines have other mode-dependent
        !          6484: addresses.  Many RISC machines have no mode-dependent addresses.
        !          6485: 
        !          6486: You may assume that @var{addr} is a valid address for the machine.
        !          6487: 
        !          6488: @item LEGITIMATE_CONSTANT_P (@var{x})
        !          6489: A C expression that is nonzero if @var{x} is a legitimate constant for
        !          6490: an immediate operand on the target machine.  You can assume that
        !          6491: either @var{x} is a @samp{const_double} or it satisfies
        !          6492: @code{CONSTANT_P}, so you need not check these things.  In fact,
        !          6493: @samp{1} is a suitable definition for this macro on machines where any
        !          6494: @samp{const_double} is valid and anything @code{CONSTANT_P} is valid.@refill
        !          6495: @end table
        !          6496: 
        !          6497: @node Misc, Condition Code, Addressing Modes, Machine Macros
        !          6498: @section Miscellaneous Parameters
        !          6499: 
        !          6500: @table @code
        !          6501: @item CASE_VECTOR_MODE
        !          6502: An alias for a machine mode name.  This is the machine mode that
        !          6503: elements of a jump-table should have.
        !          6504: 
        !          6505: @item CASE_VECTOR_PC_RELATIVE
        !          6506: Define this macro if jump-tables should contain relative addresses.
        !          6507: 
        !          6508: @item CASE_DROPS_THROUGH
        !          6509: Define this if control falls through a @code{case} insn when the index
        !          6510: value is out of range.  This means the specified default-label is
        !          6511: actually ignored by the @code{case} insn proper.
        !          6512: 
        !          6513: @item IMPLICIT_FIX_EXPR
        !          6514: An alias for a tree code that should be used by default for conversion
        !          6515: of floating point values to fixed point.  Normally,
        !          6516: @code{FIX_ROUND_EXPR} is used.@refill
        !          6517: 
        !          6518: @item FIXUNS_TRUNC_LIKE_FIX_TRUNC
        !          6519: Define this macro if the same instructions that convert a floating
        !          6520: point number to a signed fixed point number also convert validly to an
        !          6521: unsigned one.
        !          6522: 
        !          6523: @item EASY_DIV_EXPR
        !          6524: An alias for a tree code that is the easiest kind of division to
        !          6525: compile code for in the general case.  It may be
        !          6526: @code{TRUNC_DIV_EXPR}, @code{FLOOR_DIV_EXPR}, @code{CEIL_DIV_EXPR} or
        !          6527: @code{ROUND_DIV_EXPR}.  These four division operators differ in how
        !          6528: they round the result to an integer.  @code{EASY_DIV_EXPR} is used
        !          6529: when it is permissible to use any of those kinds of division and the
        !          6530: choice should be made on the basis of efficiency.@refill
        !          6531: 
        !          6532: @item DEFAULT_SIGNED_CHAR
        !          6533: An expression whose value is 1 or 0, according to whether the type
        !          6534: @code{char} should be signed or unsigned by default.  The user can
        !          6535: always override this default with the options @samp{-fsigned-char}
        !          6536: and @samp{-funsigned-char}.
        !          6537: 
        !          6538: @item SCCS_DIRECTIVE
        !          6539: Define this if the preprocessor should ignore @code{#sccs} directives
        !          6540: with no error message.
        !          6541: 
        !          6542: @item MOVE_MAX
        !          6543: The maximum number of bytes that a single instruction can move quickly
        !          6544: from memory to memory.
        !          6545: 
        !          6546: @item INT_TYPE_SIZE
        !          6547: A C expression for the size in bits of the type @code{int} on the
        !          6548: target machine.
        !          6549: 
        !          6550: @item SLOW_BYTE_ACCESS
        !          6551: Define this macro as a C expression which is nonzero if accessing less
        !          6552: than a word of memory (i.e. a @code{char} or a @code{short}) is slow
        !          6553: (requires more than one instruction).
        !          6554: 
        !          6555: @item SLOW_ZERO_EXTEND
        !          6556: Define this macro if zero-extension (of a @code{char} or @code{short}
        !          6557: to an @code{int}) can be done faster if the destination is a register
        !          6558: that is known to be zero.
        !          6559: 
        !          6560: If you define this macro, you must have instruction patterns that
        !          6561: recognize RTL structures like this:
        !          6562: 
        !          6563: @example
        !          6564: (set (strict-low-part (subreg:QI (reg:SI @dots{}) 0)) @dots{})
        !          6565: @end example
        !          6566: 
        !          6567: @noindent
        !          6568: and likewise for @code{HImode}.
        !          6569: 
        !          6570: @item SHIFT_COUNT_TRUNCATED
        !          6571: Define this macro if shift instructions ignore all but the lowest few
        !          6572: bits of the shift count.  It implies that a sign-extend or zero-extend
        !          6573: instruction for the shift count can be omitted.
        !          6574: 
        !          6575: @item TRULY_NOOP_TRUNCATION (@var{outprec}, @var{inprec})
        !          6576: A C expression which is nonzero if on this machine it is safe to
        !          6577: ``convert'' an integer of @var{inprec} bits to one of @var{outprec}
        !          6578: bits (where @var{outprec} is smaller than @var{inprec}) by merely
        !          6579: operating on it as if it had only @var{outprec} bits.
        !          6580: 
        !          6581: On many machines, this expression can be 1.
        !          6582: 
        !          6583: @item NO_FUNCTION_CSE
        !          6584: Define this macro if it is as good or better to call a constant
        !          6585: function address than to call an address kept in a register.
        !          6586: 
        !          6587: @item STORE_FLAG_VALUE
        !          6588: A C expression for the value stored by a store-flag instruction
        !          6589: (@code{s@var{cond}}) when the condition is true.  This is usually 1 or
        !          6590: -1; it is required to be an odd number.
        !          6591: 
        !          6592: Do not define @code{STORE_FLAG_VALUE} if the machine has no store-flag
        !          6593: instructions.
        !          6594: 
        !          6595: @item Pmode
        !          6596: An alias for the machine mode for pointers.  Normally the definition
        !          6597: can be
        !          6598: 
        !          6599: @example
        !          6600: #define Pmode SImode
        !          6601: @end example
        !          6602: 
        !          6603: @item FUNCTION_MODE
        !          6604: An alias for the machine mode used for memory references to functions
        !          6605: being called, in @samp{call} RTL expressions.  On most machines this
        !          6606: should be @code{QImode}.
        !          6607: 
        !          6608: @item CONST_COST (@var{x}, @var{code})
        !          6609: A part of a C @code{switch} statement that describes the relative
        !          6610: costs of constant RTL expressions.  It must contain @code{case} labels
        !          6611: for expression codes @samp{const_int}, @samp{const}, @samp{symbol_ref}, @samp{label_ref}
        !          6612: and @samp{const_double}.  Each case must ultimately reach a
        !          6613: @code{return} statement to return the relative cost of the use of that
        !          6614: kind of constant value in an expression.  The cost may depend on the
        !          6615: precise value of the constant, which is available for examination in
        !          6616: @var{x}.
        !          6617: 
        !          6618: @var{code} is the expression code---redundant, since it can be
        !          6619: obtained with @code{GET_CODE (@var{x})}.
        !          6620: 
        !          6621: @item DOLLARS_IN_IDENTIFIERS
        !          6622: Define this if the character @samp{$} should be allowed in identifier
        !          6623: names.
        !          6624: @end table
        !          6625: 
        !          6626: @node Condition Code, Assembler Format, Misc, Machine Macros
        !          6627: @section Condition Code Information
        !          6628: 
        !          6629: The file @file{conditions.h} defines a variable @code{cc_status} to
        !          6630: describe how the condition code was computed (in case the interpretation of
        !          6631: the condition code depends on the instruction that it was set by).  This
        !          6632: variable contains the RTL expressions on which the condition code is
        !          6633: currently based, and several standard flags.
        !          6634: 
        !          6635: Sometimes additional machine-specific flags must be defined in the machine
        !          6636: description header file.  It can also add additional machine-specific
        !          6637: information by defining @code{CC_STATUS_MDEP}.
        !          6638: 
        !          6639: @table @code
        !          6640: @item CC_STATUS_MDEP
        !          6641: C code for a data type which is used for declaring the @code{mdep}
        !          6642: component of @code{cc_status}.  It defaults to @code{int}.
        !          6643: 
        !          6644: @item CC_STATUS_MDEP_INIT
        !          6645: A C expression for the initial value of the @code{mdep} field.  It
        !          6646: defaults to 0.
        !          6647: 
        !          6648: @item NOTICE_UPDATE_CC (@var{exp})
        !          6649: A C compound statement to set the components of @code{cc_status}
        !          6650: appropriately for an insn whose body is @var{exp}.  It is this macro's
        !          6651: responsibility to recognize insns that set the condition code as a
        !          6652: byproduct of other activity as well as those that explicitly set
        !          6653: @code{(cc0)}.
        !          6654: 
        !          6655: If there are insn that do not set the condition code but do alter
        !          6656: other machine registers, this macro must check to see whether they
        !          6657: invalidate the expressions that the condition code is recorded as
        !          6658: reflecting.  For example, on the 68000, insns that store in address
        !          6659: registers do not set the condition code, which means that usually
        !          6660: @code{NOTICE_UPDATE_CC} can leave @code{cc_status} unaltered for such
        !          6661: insns.  But suppose that the previous insn set the condition code
        !          6662: based on location @samp{a4@@(102)} and the current insn stores a new
        !          6663: value in @samp{a4}.  Although the condition code is not changed by
        !          6664: this, it will no longer be true that it reflects the contents of
        !          6665: @samp{a4@@(102)}.  Therefore, @code{NOTICE_UPDATE_CC} must alter
        !          6666: @code{cc_status} in this case to say that nothing is known about the
        !          6667: condition code value.
        !          6668: @end table
        !          6669: 
        !          6670: @node Assembler Format,, Condition Code, Machine Macros
        !          6671: @section Output of Assembler Code
        !          6672: 
        !          6673: @table @code
        !          6674: @item ASM_SPEC
        !          6675: A C string constant that tells the GNU CC driver program options to
        !          6676: pass to the assembler.  It can also specify how to translate options
        !          6677: you give to GNU CC into options for GNU CC to pass to the assembler.
        !          6678: See the file @file{tm-sun3.h} for an example of this.
        !          6679: 
        !          6680: Do not define this macro if it does not need to do anything.
        !          6681: 
        !          6682: @item LINK_SPEC
        !          6683: A C string constant that tells the GNU CC driver program options to
        !          6684: pass to the linker.  It can also specify how to translate options you
        !          6685: give to GNU CC into options for GNU CC to pass to the linker.
        !          6686: 
        !          6687: Do not define this macro if it does not need to do anything.
        !          6688: 
        !          6689: @item ASM_FILE_START
        !          6690: A C string constant for text to be output at the start of each
        !          6691: assembler output file.  Normally this is @code{"#NO_APP"}, which is a
        !          6692: comment that has no effect on most assemblers but tells the GNU
        !          6693: assembler that it can save time by not checking for certain assembler
        !          6694: constructs.
        !          6695: 
        !          6696: @item ASM_APP_ON
        !          6697: A C string constant for text to be output before each @code{asm}
        !          6698: statement or group of consecutive ones.  Normally this is
        !          6699: @code{"#APP"}, which is a comment that has no effect on most
        !          6700: assemblers but tells the GNU assembler that it must check the lines
        !          6701: that follow for all valid assembler constructs.
        !          6702: 
        !          6703: @item ASM_APP_OFF
        !          6704: A C string constant for text to be output after each @code{asm}
        !          6705: statement or group of consecutive ones.  Normally this is
        !          6706: @code{"#NO_APP"}, which tells the GNU assembler to resume making the
        !          6707: time-saving assumptions that are valid for ordinary compiler output.
        !          6708: 
        !          6709: @item TEXT_SECTION_ASM_OP
        !          6710: A C string constant for the assembler operation that should precede
        !          6711: instructions and read-only data.  Normally @code{".text"} is right.
        !          6712: 
        !          6713: @item DATA_SECTION_ASM_OP
        !          6714: A C string constant for the assembler operation to identify the
        !          6715: following data as writable initialized data.  Normally @code{".data"}
        !          6716: is right.
        !          6717: 
        !          6718: @item REGISTER_NAMES
        !          6719: A C initializer containing the assembler's names for the machine
        !          6720: registers, each one as a C string constant.  This is what translates
        !          6721: register numbers in the compiler into assembler language.
        !          6722: 
        !          6723: @item DBX_REGISTER_NUMBER (@var{regno})
        !          6724: A C expression that returns the DBX register number for the compiler
        !          6725: register number @var{regno}.  In simple cases, the value of this
        !          6726: expression may be @var{regno} itself.  But sometimes there are some
        !          6727: registers that the compiler knows about and DBX does not, or vice
        !          6728: versa.  In such cases, some register may need to have one number in
        !          6729: the compiler and another for DBX.
        !          6730: 
        !          6731: @item DBX_NO_XREFS
        !          6732: Define this macro if DBX on your system does not support the construct
        !          6733: @samp{xs@var{tagname}}.  On some systems, this construct is used to
        !          6734: describe a forward reference to a structure named @var{tagname}.
        !          6735: On other systems, this construct is not supported at all.
        !          6736: 
        !          6737: @item DBX_CONTIN_LENGTH
        !          6738: A symbol name in DBX-format debugging information is normally
        !          6739: continued (split into two separate @code{.stabs} directives) when it
        !          6740: exceeds a certain length (by default, 80 characters).  On some
        !          6741: operating systems, DBX requires this splitting; on others, splitting
        !          6742: must not be done.  You can inhibit splitting by defining this macro
        !          6743: with the value zero.  You can override the default splitting-length by
        !          6744: defining this macro as an expression for the length you desire.
        !          6745: 
        !          6746: @item DBX_CONTIN_CHAR
        !          6747: Normally continuation is indicated by adding a @samp{\} character to
        !          6748: the end of a @code{.stabs} string when a continuation follows.  To use
        !          6749: a different character instead, define this macro as a character
        !          6750: constant for the character you want to use.  Do not define this macro
        !          6751: if backslash is correct for your system.
        !          6752: 
        !          6753: @item ASM_OUTPUT_LABEL (@var{file}, @var{name})
        !          6754: A C statement (sans semicolon) to output to the stdio stream
        !          6755: @var{file} the assembler definition of a label named @var{name}.  Use
        !          6756: the expression @code{assemble_name (@var{file}, @var{name})} to output
        !          6757: the name itself; before and after that, output the additional
        !          6758: assembler syntax for defining the name, and a newline.
        !          6759: 
        !          6760: @item ASM_DECLARE_FUNCTION_NAME (@var{file}, @var{name})
        !          6761: A C statement (sans semicolon) to output to the stdio stream
        !          6762: @var{file} any text necessary for declaring the name of a function
        !          6763: which is being defined.  This macro is responsible for outputting
        !          6764: the label definition (perhaps using @code{ASM_OUTPUT_LABEL}).
        !          6765: 
        !          6766: If this macro is not defined, then the function name is defined in the
        !          6767: usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}).
        !          6768: 
        !          6769: @item ASM_GLOBALIZE_LABEL (@var{file}, @var{name})
        !          6770: A C statement (sans semicolon) to output to the stdio stream
        !          6771: @var{file} some commands that will make the label @var{name} global;
        !          6772: that is, available for reference from other files.  Use the expression
        !          6773: @code{assemble_name (@var{file}, @var{name})} to output the name
        !          6774: itself; before and after that, output the additional assembler syntax
        !          6775: for making that name global, and a newline.
        !          6776: 
        !          6777: @item ASM_OUTPUT_EXTERNAL (@var{file}, @var{name})
        !          6778: A C statement (sans semicolon) to output to the stdio stream
        !          6779: @var{file} any text necessary for declaring the name of an external
        !          6780: symbol which is referenced in this compilation but not defined.
        !          6781: 
        !          6782: This macro need not be defined if it does not need to output anything.
        !          6783: The GNU assembler and most Unix assemblers don't require anything.
        !          6784: 
        !          6785: @item ASM_OUTPUT_LABELREF (@var{file}, @var{name})
        !          6786: A C statement to output to the stdio stream @var{file} a reference in
        !          6787: assembler syntax to a label named @var{name}.  The character @samp{_}
        !          6788: should be added to the front of the name, if that is customary on your
        !          6789: operating system, as it is in most Berkeley Unix systems.  This macro
        !          6790: is used in @code{assemble_name}.
        !          6791: 
        !          6792: @item ASM_OUTPUT_INTERNAL_LABEL (@var{file}, @var{prefix}, @var{num})
        !          6793: A C statement to output to the stdio stream @var{file} a label whose
        !          6794: name is made from the string @var{prefix} and the number @var{num}.
        !          6795: These labels are used for internal purposes, and there is no reason
        !          6796: for them to appear in the symbol table of the object file.  On many
        !          6797: systems, the letter @samp{L} at the beginning of a label has this
        !          6798: effect.  The usual definition of this macro is as follows:
        !          6799: 
        !          6800: @example
        !          6801: fprintf (@var{file}, "L%s%d:\n", @var{prefix}, @var{num})
        !          6802: @end example
        !          6803: 
        !          6804: @item ASM_OUTPUT_CASE_LABEL (@var{file}, @var{prefix}, @var{num}, @var{table})
        !          6805: Define this if the label before a jump-table needs to be output
        !          6806: specially.  The first three arguments are the same as for
        !          6807: @code{ASM_OUTPUT_INTERNAL_LABEL}; the fourth argument is the
        !          6808: jump-table which follows (a @samp{jump_insn} containing an
        !          6809: @samp{addr_vec} or @samp{addr_diff_vec}).
        !          6810: 
        !          6811: This feature is used on system V to output a @code{swbeg} statement
        !          6812: for the table.
        !          6813: 
        !          6814: If this macro is not defined, these labels are output with
        !          6815: @code{ASM_OUTPUT_INTERNAL_LABEL}.
        !          6816: 
        !          6817: @item ASM_FORMAT_PRIVATE_NAME (@var{outvar}, @var{name}, @var{number})
        !          6818: A C expression to assign to @var{outvar} (which is a variable of type
        !          6819: @code{char *}) a newly allocated string made from the string
        !          6820: @var{name} and the number @var{number}, with some suitable punctuation
        !          6821: added.  Use @code{alloca} to get space for the string.
        !          6822: 
        !          6823: This string will be used as the argument to @code{ASM_OUTPUT_LABELREF}
        !          6824: to produce an assembler label for an internal static variable whose
        !          6825: name is @var{name}.  Therefore, the string must be such as to result
        !          6826: in valid assembler code.  The argument @var{number} is different each
        !          6827: time this macro is executed; it prevents conflicts between
        !          6828: similarly-named internal static variables in different scopes.
        !          6829: 
        !          6830: Ideally this string should not be a valid C identifier, to prevent any
        !          6831: conflict with the user's own symbols.  Most assemblers allow periods
        !          6832: or percent signs in assembler symbols; putting at least one of these
        !          6833: between the name and the number will suffice.
        !          6834: 
        !          6835: @item ASM_OUTPUT_ADDR_DIFF_ELT (@var{file}, @var{value}, @var{rel})
        !          6836: This macro should be provided on machines where the addresses
        !          6837: in a dispatch table are relative to the table's own address.
        !          6838: 
        !          6839: The definition should be a C statement to output to the stdio stream
        !          6840: @var{file} an assembler pseudo-instruction to generate a difference
        !          6841: between two labels.  @var{value} and @var{rel} are the numbers of two
        !          6842: internal labels.  The definitions of these labels are output using
        !          6843: @code{ASM_OUTPUT_INTERNAL_LABEL}, and they must be printed in the same
        !          6844: way here.  For example,
        !          6845: 
        !          6846: @example
        !          6847: fprintf (@var{file}, "\t.word L%d-L%d\n",
        !          6848:          @var{value}, @var{rel})
        !          6849: @end example
        !          6850: 
        !          6851: @item ASM_OUTPUT_ADDR_VEC_ELT (@var{file}, @var{value})
        !          6852: This macro should be provided on machines where the addresses
        !          6853: in a dispatch table are absolute.
        !          6854: 
        !          6855: The definition should be a C statement to output to the stdio stream
        !          6856: @var{file} an assembler pseudo-instruction to generate a reference to
        !          6857: a label.  @var{value} is the number of an internal label whose
        !          6858: definition is output using @code{ASM_OUTPUT_INTERNAL_LABEL}.
        !          6859: For example,
        !          6860: 
        !          6861: @example
        !          6862: fprintf (@var{file}, "\t.word L%d\n", @var{value})
        !          6863: @end example
        !          6864: 
        !          6865: @item ASM_OUTPUT_DOUBLE (@var{file}, @var{value})
        !          6866: A C statement to output to the stdio stream @var{file} an assembler
        !          6867: instruction to assemble a @code{double} constant whose value is
        !          6868: @var{value}.  @var{value} will be a C expression of type
        !          6869: @code{double}.
        !          6870: 
        !          6871: @item ASM_OUTPUT_FLOAT (@var{file}, @var{value})
        !          6872: A C statement to output to the stdio stream @var{file} an assembler
        !          6873: instruction to assemble a @code{float} constant whose value is
        !          6874: @var{value}.  @var{value} will be a C expression of type @code{float}.
        !          6875: 
        !          6876: @item ASM_OUTPUT_INT (@var{file}, @var{exp})
        !          6877: @itemx ASM_OUTPUT_SHORT (@var{file}, @var{exp})
        !          6878: @itemx ASM_OUTPUT_CHAR (@var{file}, @var{exp})
        !          6879: A C statement to output to the stdio stream @var{file} an assembler
        !          6880: instruction to assemble a @code{int}, @code{short} or @code{char}
        !          6881: constant whose value is @var{value}.  The argument @var{exp} will be
        !          6882: an RTL expression which represents a constant value.  Use
        !          6883: @samp{output_addr_const (@var{exp})} to output this value as an
        !          6884: assembler expression.@refill
        !          6885: 
        !          6886: @item ASM_OUTPUT_BYTE (@var{file}, @var{value})
        !          6887: A C statement to output to the stdio stream @var{file} an assembler
        !          6888: instruction to assemble a single byte containing the number @var{value}.
        !          6889: 
        !          6890: @item ASM_OUTPUT_ASCII (@var{file}, @var{ptr}, @var{len})
        !          6891: A C statement to output to the stdio stream @var{file} an assembler
        !          6892: instruction to assemble a string constant containing the @var{len}
        !          6893: bytes at @var{ptr}.  @var{ptr} will be a C expression of type
        !          6894: @code{char *} and @var{len} a C expression of type @code{int}.
        !          6895: 
        !          6896: If the assembler has a @code{.ascii} pseudo-op as found in the
        !          6897: Berkeley Unix assembler, do not define the macro
        !          6898: @code{ASM_OUTPUT_ASCII}.
        !          6899: 
        !          6900: @item ASM_OUTPUT_SKIP (@var{file}, @var{nbytes})
        !          6901: A C statement to output to the stdio stream @var{file} an assembler
        !          6902: instruction to advance the location counter by @var{nbytes} bytes.
        !          6903: @var{nbytes} will be a C expression of type @code{int}.
        !          6904: 
        !          6905: @item ASM_OUTPUT_ALIGN (@var{file}, @var{power})
        !          6906: A C statement to output to the stdio stream @var{file} an assembler
        !          6907: instruction to advance the location counter to a multiple of 2 to the
        !          6908: @var{power} bytes.  @var{power} will be a C expression of type @code{int}.
        !          6909: 
        !          6910: @item ASM_OUTPUT_COMMON (@var{file}, @var{name}, @var{size})
        !          6911: A C statement (sans semicolon) to output to the stdio stream
        !          6912: @var{file} the assembler definition of a common-label named @var{name}
        !          6913: whose size is @var{size} bytes.  Use the expression
        !          6914: @code{assemble_name (@var{file}, @var{name})} to output the name
        !          6915: itself; before and after that, output the additional assembler syntax
        !          6916: for defining the name, and a newline.
        !          6917: 
        !          6918: This macro controls how the assembler definitions of uninitialized
        !          6919: global variables are output.
        !          6920: 
        !          6921: @item ASM_OUTPUT_LOCAL (@var{file}, @var{name}, @var{size})
        !          6922: A C statement (sans semicolon) to output to the stdio stream
        !          6923: @var{file} the assembler definition of a local-common-label named
        !          6924: @var{name} whose size is @var{size} bytes.  Use the expression
        !          6925: @code{assemble_name (@var{file}, @var{name})} to output the name
        !          6926: itself; before and after that, output the additional assembler syntax
        !          6927: for defining the name, and a newline.
        !          6928: 
        !          6929: This macro controls how the assembler definitions of uninitialized
        !          6930: static variables are output.
        !          6931: 
        !          6932: @item TARGET_BELL
        !          6933: A C constant expression for the integer value for escape sequence
        !          6934: @samp{\a}.
        !          6935: 
        !          6936: @item TARGET_BS
        !          6937: @itemx TARGET_TAB
        !          6938: @itemx TARGET_NEWLINE
        !          6939: C constant expressions for the integer values for escape sequences
        !          6940: @samp{\b}, @samp{\t} and @samp{\n}.
        !          6941: 
        !          6942: @item TARGET_VT
        !          6943: @itemx TARGET_FF
        !          6944: @itemx TARGET_CR
        !          6945: C constant expressions for the integer values for escape sequences
        !          6946: @samp{\v}, @samp{\f} and @samp{\r}.
        !          6947: 
        !          6948: @item ASM_OUTPUT_OPCODE (@var{file}, @var{ptr})
        !          6949: Define this macro if you are using an unusual assembler that
        !          6950: requires different names for the machine instructions.
        !          6951: 
        !          6952: The definition is a C statement or statements which output an
        !          6953: assembler instruction opcode to the stdio stream @var{file}.  The
        !          6954: macro-operand @var{ptr} is a variable of type @code{char *} which
        !          6955: points to the opcode name in its ``internal'' form---the form that is
        !          6956: written in the machine description.  The definition should output the
        !          6957: opcode name to @var{file}, performing any translation you desire, and
        !          6958: increment the variable @var{ptr} to point at the end of the opcode
        !          6959: so that it will not be output twice.
        !          6960: 
        !          6961: In fact, your macro definition may process less than the entire opcode
        !          6962: name, or more than the opcode name; but if you want to process text
        !          6963: that includes @samp{%}-sequences to substitute operands, you must take
        !          6964: care of the substitution yourself.  Just be sure to increment
        !          6965: @var{ptr} over whatever text should not be output normally.
        !          6966: 
        !          6967: If the macro definition does nothing, the instruction is output
        !          6968: in the usual way.
        !          6969: 
        !          6970: @item PRINT_OPERAND (@var{file}, @var{x}, @var{code})
        !          6971: A C compound statement to output to stdio stream @var{file} the
        !          6972: assembler syntax for an instruction operand @var{x}.  @var{x} is an
        !          6973: RTL expression.
        !          6974: 
        !          6975: @var{code} is a value that can be used to specify one of several ways
        !          6976: of printing the operand.  It is used when identical operands must be
        !          6977: printed differently depending on the context.  @var{code} comes from
        !          6978: the @samp{%} specification that was used to request printing of the
        !          6979: operand.  If the specification was just @samp{%@var{digit}} then
        !          6980: @var{code} is 0; if the specification was @samp{%@var{ltr}
        !          6981: @var{digit}} then @var{code} is the ASCII code for @var{ltr}.
        !          6982: 
        !          6983: If @var{x} is a register, this macro should print the register's name.
        !          6984: The names can be found in an array @code{reg_names} whose type is
        !          6985: @code{char *[]}.  @code{reg_names} is initialized from
        !          6986: @code{REGISTER_NAMES}.
        !          6987: 
        !          6988: When the machine description has a specification @samp{%@var{punct}}
        !          6989: (a @samp{%} followed by a punctuation character), this macro is called
        !          6990: with a null pointer for @var{x} and the punctuation character for
        !          6991: @var{code}.
        !          6992: 
        !          6993: @item PRINT_OPERAND_ADDRESS (@var{file}, @var{x})
        !          6994: A C compound statement to output to stdio stream @var{file} the
        !          6995: assembler syntax for an instruction operand that is a memory reference
        !          6996: whose address is @var{x}.  @var{x} is an RTL expression.
        !          6997: 
        !          6998: @item ASM_OPEN_PAREN
        !          6999: @itemx ASM_CLOSE_PAREN
        !          7000: These macros are defined as C string constant, describing the syntax
        !          7001: in the assembler for grouping arithmetic expressions.  The following
        !          7002: definitions are correct for most assemblers:
        !          7003: 
        !          7004: @example
        !          7005: #define ASM_OPEN_PAREN "("
        !          7006: #define ASM_CLOSE_PAREN ")"
        !          7007: @end example
        !          7008: @end table
        !          7009: 
        !          7010: @node Config,, Machine Macros, Top
        !          7011: @chapter The Configuration File
        !          7012: 
        !          7013: The configuration file @file{config-@var{machine}.h} contains macro
        !          7014: definitions that describe the machine and system on which the compiler is
        !          7015: running.  Most of the values in it are actually the same on all machines
        !          7016: that GNU CC runs on, so most all configuration files are identical.  But
        !          7017: there are some macros that vary:
        !          7018: 
        !          7019: @table @code
        !          7020: @item FAILURE_EXIT_CODE
        !          7021: A C expression for the status code to be returned when the compiler
        !          7022: exits after serious errors.
        !          7023: 
        !          7024: @item SUCCESS_EXIT_CODE
        !          7025: A C expression for the status code to be returned when the compiler
        !          7026: exits without serious errors.
        !          7027: @end table
        !          7028: 
        !          7029: @contents
        !          7030: @bye

This archive runs on limited infrastructure. Preserving old code on modern bandwidth. Automated agents are requested to crawl responsibly.