![[BACK]](/cvsweb/icons/back.gif){kind=icon}
Return to gcc.texinfo CVS log ![[TXT]](/cvsweb/icons/text.gif){kind=icon}
![[DIR]](/cvsweb/icons/dir.gif){kind=icon}
Up to [GCC 1.x] / gcc|
Return to gcc.texinfo CVS log | Up to [GCC 1.x] / gcc |
1.1 root 1: \input texinfo @c -*-texinfo-*-
2:
3: @settitle Using and Porting GNU CC
4: @setfilename gcc.info
5:
6: @ifinfo
7: This file documents the use and the internals of the GNU compiler.
8:
1.1.1.5 root 9: Copyright (C) 1988, 1989 Free Software Foundation, Inc.
1.1 root 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
1.1.1.6 ! root 24: section entitled ``GNU General Public License'' is included exactly as
1.1 root 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,
1.1.1.6 ! root 30: except that the section entitled ``GNU General Public License'' and
1.1 root 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{Using and Porting GNU CC}
39: @sp 2
40: @center Richard M. Stallman
41: @sp 3
1.1.1.6 ! root 42: @center last updated 23 Feb 1989
1.1 root 43: @sp 1
1.1.1.6 ! root 44: @center for version 1.34
1.1 root 45: @page
46: @vskip 0pt plus 1filll
1.1.1.5 root 47: Copyright @copyright{} 1988, 1989 Free Software Foundation, Inc.
1.1 root 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
1.1.1.6 ! root 55: section entitled ``GNU General Public License'' is included exactly as
1.1 root 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,
1.1.1.6 ! root 61: <except that the section entitled ``GNU General Public License'' and
1.1 root 62: this permission notice may be included in translations approved by the
63: Free Software Foundation instead of in the original 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
1.1.1.6 ! root 76: * Copying:: GNU General Public License says
1.1 root 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
1.1.1.6 ! root 94: @unnumbered GNU GENERAL PUBLIC LICENSE
! 95: @center Version 1, February 1989
1.1 root 96:
1.1.1.6 ! root 97: @display
! 98: Copyright @copyright{} 1989 Free Software Foundation, Inc.
! 99: 675 Mass Ave, Cambridge, MA 02139, USA
! 100:
! 101: Everyone is permitted to copy and distribute verbatim copies
! 102: of this license document, but changing it is not allowed.
! 103: @end display
! 104:
! 105: @unnumberedsec Preamble
! 106:
! 107: The license agreements of most software companies try to keep users
! 108: at the mercy of those companies. By contrast, our General Public
! 109: License is intended to guarantee your freedom to share and change free
! 110: software---to make sure the software is free for all its users. The
! 111: General Public License applies to the Free Software Foundation's
! 112: software and to any other program whose authors commit to using it.
! 113: You can use it for your programs, too.
! 114:
! 115: When we speak of free software, we are referring to freedom, not
! 116: price. Specifically, the General Public License is designed to make
! 117: sure that you have the freedom to give away or sell copies of free
! 118: software, that you receive source code or can get it if you want it,
! 119: that you can change the software or use pieces of it in new free
! 120: programs; and that you know you can do these things.
! 121:
! 122: To protect your rights, we need to make restrictions that forbid
! 123: anyone to deny you these rights or to ask you to surrender the rights.
! 124: These restrictions translate to certain responsibilities for you if you
! 125: distribute copies of the software, or if you modify it.
! 126:
! 127: For example, if you distribute copies of a such a program, whether
! 128: gratis or for a fee, you must give the recipients all the rights that
! 129: you have. You must make sure that they, too, receive or can get the
1.1 root 130: source code. And you must tell them their rights.
131:
1.1.1.6 ! root 132: We protect your rights with two steps: (1) copyright the software, and
! 133: (2) offer you this license which gives you legal permission to copy,
! 134: distribute and/or modify the software.
! 135:
! 136: Also, for each author's protection and ours, we want to make certain
! 137: that everyone understands that there is no warranty for this free
! 138: software. If the software is modified by someone else and passed on, we
! 139: want its recipients to know that what they have is not the original, so
! 140: that any problems introduced by others will not reflect on the original
! 141: authors' reputations.
1.1 root 142:
1.1.1.6 ! root 143: The precise terms and conditions for copying, distribution and
! 144: modification follow.
1.1 root 145:
1.1.1.6 ! root 146: @iftex
! 147: @unnumberedsec TERMS AND CONDITIONS
! 148: @end iftex
! 149: @ifinfo
! 150: @center TERMS AND CONDITIONS
! 151: @end ifinfo
1.1 root 152:
1.1.1.6 ! root 153: @enumerate
1.1 root 154: @item
1.1.1.6 ! root 155: This License Agreement applies to any program or other work which
! 156: contains a notice placed by the copyright holder saying it may be
! 157: distributed under the terms of this General Public License. The
! 158: ``Program'', below, refers to any such program or work, and a ``work based
! 159: on the Program'' means either the Program or any work containing the
! 160: Program or a portion of it, either verbatim or with modifications. Each
! 161: licensee is addressed as ``you''.
! 162:
! 163: @item
! 164: You may copy and distribute verbatim copies of the Program's source
! 165: code as you receive it, in any medium, provided that you conspicuously and
! 166: appropriately publish on each copy an appropriate copyright notice and
! 167: disclaimer of warranty; keep intact all the notices that refer to this
! 168: General Public License and to the absence of any warranty; and give any
! 169: other recipients of the Program a copy of this General Public License
! 170: along with the Program. You may charge a fee for the physical act of
! 171: transferring a copy.
! 172:
! 173: @item
! 174: You may modify your copy or copies of the Program or any portion of
! 175: it, and copy and distribute such modifications under the terms of Paragraph
! 176: 1 above, provided that you also do the following:
1.1 root 177:
178: @itemize @bullet
179: @item
1.1.1.6 ! root 180: cause the modified files to carry prominent notices stating that
! 181: you changed the files and the date of any change; and
1.1 root 182:
183: @item
184: cause the whole of any work that you distribute or publish, that
1.1.1.6 ! root 185: in whole or in part contains the Program or any part thereof, either
! 186: with or without modifications, to be licensed at no charge to all
! 187: third parties under the terms of this General Public License (except
! 188: that you may choose to grant warranty protection to some or all
! 189: third parties, at your option).
! 190:
! 191: @item
! 192: If the modified program normally reads commands interactively when
! 193: run, you must cause it, when started running for such interactive use
! 194: in the simplest and most usual way, to print or display an
! 195: announcement including an appropriate copyright notice and a notice
! 196: that there is no warranty (or else, saying that you provide a
! 197: warranty) and that users may redistribute the program under these
! 198: conditions, and telling the user how to view a copy of this General
! 199: Public License.
! 200:
! 201: @item
! 202: You may charge a fee for the physical act of transferring a
! 203: copy, and you may at your option offer warranty protection in
! 204: exchange for a fee.
1.1 root 205: @end itemize
206:
1.1.1.6 ! root 207: Mere aggregation of another independent work with the Program (or its
1.1 root 208: derivative) on a volume of a storage or distribution medium does not bring
1.1.1.6 ! root 209: the other work under the scope of these terms.
1.1 root 210:
211: @item
1.1.1.6 ! root 212: You may copy and distribute the Program (or a portion or derivative of
! 213: it, under Paragraph 2) in object code or executable form under the terms of
! 214: Paragraphs 1 and 2 above provided that you also do one of the following:
1.1 root 215:
216: @itemize @bullet
217: @item
218: accompany it with the complete corresponding machine-readable
219: source code, which must be distributed under the terms of
220: Paragraphs 1 and 2 above; or,
221:
222: @item
223: accompany it with a written offer, valid for at least three
1.1.1.6 ! root 224: years, to give any third party free (except for a nominal charge
! 225: for the cost of distribution) a complete machine-readable copy of the
1.1 root 226: corresponding source code, to be distributed under the terms of
227: Paragraphs 1 and 2 above; or,
228:
229: @item
230: accompany it with the information you received as to where the
231: corresponding source code may be obtained. (This alternative is
232: allowed only for noncommercial distribution and only if you
233: received the program in object code or executable form alone.)
234: @end itemize
235:
1.1.1.6 ! root 236: Source code for a work means the preferred form of the work for making
! 237: modifications to it. For an executable file, complete source code means
! 238: all the source code for all modules it contains; but, as a special
! 239: exception, it need not include source code for modules which are standard
! 240: libraries that accompany the operating system on which the executable
! 241: file runs, or for standard header files or definitions files that
! 242: accompany that operating system.
! 243:
! 244: @item
! 245: You may not copy, modify, sublicense, distribute or transfer the
! 246: Program except as expressly provided under this General Public License.
! 247: Any attempt otherwise to copy, modify, sublicense, distribute or transfer
! 248: the Program is void, and will automatically terminate your rights to use
! 249: the Program under this License. However, parties who have received
! 250: copies, or rights to use copies, from you under this General Public
! 251: License will not have their licenses terminated so long as such parties
! 252: remain in full compliance.
! 253:
! 254: @item
! 255: By copying, distributing or modifying the Program (or any work based
! 256: on the Program) you indicate your acceptance of this license to do so,
! 257: and all its terms and conditions.
! 258:
! 259: @item
! 260: Each time you redistribute the Program (or any work based on the
! 261: Program), the recipient automatically receives a license from the original
! 262: licensor to copy, distribute or modify the Program subject to these
! 263: terms and conditions. You may not impose any further restrictions on the
! 264: recipients' exercise of the rights granted herein.
! 265:
! 266: @item
! 267: The Free Software Foundation may publish revised and/or new versions
! 268: of the General Public License from time to time. Such new versions will
! 269: be similar in spirit to the present version, but may differ in detail to
! 270: address new problems or concerns.
! 271:
! 272: Each version is given a distinguishing version number. If the Program
! 273: specifies a version number of the license which applies to it and ``any
! 274: later version'', you have the option of following the terms and conditions
! 275: either of that version or of any later version published by the Free
! 276: Software Foundation. If the Program does not specify a version number of
! 277: the license, you may choose any version ever published by the Free Software
! 278: Foundation.
! 279:
! 280: @item
! 281: If you wish to incorporate parts of the Program into other free
! 282: programs whose distribution conditions are different, write to the author
! 283: to ask for permission. For software which is copyrighted by the Free
! 284: Software Foundation, write to the Free Software Foundation; we sometimes
! 285: make exceptions for this. Our decision will be guided by the two goals
! 286: of preserving the free status of all derivatives of our free software and
! 287: of promoting the sharing and reuse of software generally.
! 288:
! 289: @iftex
! 290: @heading NO WARRANTY
! 291: @end iftex
! 292: @ifinfo
! 293: @center NO WARRANTY
! 294: @end ifinfo
! 295:
! 296: @item
! 297: BECAUSE THE PROGRAM IS LICENSED FREE OF CHARGE, THERE IS NO WARRANTY
! 298: FOR THE PROGRAM, TO THE EXTENT PERMITTED BY APPLICABLE LAW. EXCEPT WHEN
! 299: OTHERWISE STATED IN WRITING THE COPYRIGHT HOLDERS AND/OR OTHER PARTIES
! 300: PROVIDE THE PROGRAM ``AS IS'' WITHOUT WARRANTY OF ANY KIND, EITHER EXPRESSED
! 301: OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
! 302: MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. THE ENTIRE RISK AS
! 303: TO THE QUALITY AND PERFORMANCE OF THE PROGRAM IS WITH YOU. SHOULD THE
! 304: PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF ALL NECESSARY SERVICING,
! 305: REPAIR OR CORRECTION.
! 306:
! 307: @item
! 308: IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING WILL
! 309: ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MAY MODIFY AND/OR
! 310: REDISTRIBUTE THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES,
! 311: INCLUDING ANY GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES
! 312: ARISING OUT OF THE USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT
! 313: LIMITED TO LOSS OF DATA OR DATA BEING RENDERED INACCURATE OR LOSSES
! 314: SUSTAINED BY YOU OR THIRD PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE
! 315: WITH ANY OTHER PROGRAMS), EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN
! 316: ADVISED OF THE POSSIBILITY OF SUCH DAMAGES.
1.1 root 317: @end enumerate
318:
1.1.1.6 ! root 319: @iftex
! 320: @heading END OF TERMS AND CONDITIONS
! 321: @end iftex
! 322: @ifinfo
! 323: @center END OF TERMS AND CONDITIONS
! 324: @end ifinfo
! 325:
! 326: @page
! 327: @unnumberedsec Appendix: How to Apply These Terms to Your New Programs
! 328:
! 329: If you develop a new program, and you want it to be of the greatest
! 330: possible use to humanity, the best way to achieve this is to make it
! 331: free software which everyone can redistribute and change under these
! 332: terms.
! 333:
! 334: To do so, attach the following notices to the program. It is safest to
! 335: attach them to the start of each source file to most effectively convey
! 336: the exclusion of warranty; and each file should have at least the
! 337: ``copyright'' line and a pointer to where the full notice is found.
! 338:
! 339: @smallexample
! 340: @var{one line to give the program's name and a brief idea of what it does.}
! 341: Copyright (C) 19@var{yy} @var{name of author}
! 342:
! 343: This program is free software; you can redistribute it and/or modify
! 344: it under the terms of the GNU General Public License as published by
! 345: the Free Software Foundation; either version 1, or (at your option)
! 346: any later version.
! 347:
! 348: This program is distributed in the hope that it will be useful,
! 349: but WITHOUT ANY WARRANTY; without even the implied warranty of
! 350: MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
! 351: GNU General Public License for more details.
! 352:
! 353: You should have received a copy of the GNU General Public License
! 354: along with this program; if not, write to the Free Software
! 355: Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
! 356: @end smallexample
! 357:
! 358: Also add information on how to contact you by electronic and paper mail.
! 359:
! 360: If the program is interactive, make it output a short notice like this
! 361: when it starts in an interactive mode:
! 362:
! 363: @smallexample
! 364: Gnomovision version 69, Copyright (C) 19@var{yy} @var{name of author}
! 365: Gnomovision comes with ABSOLUTELY NO WARRANTY; for details type `show w'.
! 366: This is free software, and you are welcome to redistribute it
! 367: under certain conditions; type `show c' for details.
! 368: @end smallexample
! 369:
! 370: The hypothetical commands `show w' and `show c' should show the
! 371: appropriate parts of the General Public License. Of course, the
! 372: commands you use may be called something other than `show w' and `show
! 373: c'; they could even be mouse-clicks or menu items---whatever suits your
! 374: program.
! 375:
! 376: You should also get your employer (if you work as a programmer) or your
! 377: school, if any, to sign a ``copyright disclaimer'' for the program, if
! 378: necessary. Here a sample; alter the names:
! 379:
! 380: @example
! 381: Yoyodyne, Inc., hereby disclaims all copyright interest in the
! 382: program `Gnomovision' (a program to direct compilers to make passes
! 383: at assemblers) written by James Hacker.
! 384:
! 385: @var{signature of Ty Coon}, 1 April 1989
! 386: Ty Coon, President of Vice
! 387: @end example
! 388:
! 389: That's all there is to it!
1.1 root 390:
391: @node Contributors, Options, Copying, Top
392: @unnumbered Contributors to GNU CC
393:
394: In addition to Richard Stallman, several people have written parts
395: of GNU CC.
396:
397: @itemize @bullet
398: @item
399: The idea of using RTL and some of the optimization ideas came from the
400: U. of Arizona Portable Optimizer, written by Jack Davidson and
401: Christopher Fraser. See ``Register Allocation and Exhaustive Peephole
402: Optimization'', Software Practice and Experience 14 (9), Sept. 1984,
403: 857-866.
404:
405: @item
406: Paul Rubin wrote most of the preprocessor.
407:
408: @item
1.1.1.6 ! root 409: Leonard Tower wrote parts of the parser, RTL generator, and RTL
1.1 root 410: definitions, and of the Vax machine description.
411:
412: @item
413: Ted Lemon wrote parts of the RTL reader and printer.
414:
415: @item
416: Jim Wilson implemented loop strength reduction and some other
417: loop optimizations.
418:
419: @item
420: Nobuyuki Hikichi of Software Research Associates, Tokyo, contributed
421: the support for the SONY NEWS machine.
422:
423: @item
424: Charles LaBrec contributed the support for the Integrated Solutions
425: 68020 system.
426:
427: @item
428: Michael Tiemann of MCC wrote most of the description of the National
429: Semiconductor 32000 series cpu. He also wrote the code for inline
430: function integration and for the SPARC cpu and Motorola 88000 cpu
431: and part of the Sun FPA support.
432:
433: @item
434: Jan Stein of the Chalmers Computer Society provided support for
435: Genix, as well as part of the 32000 machine description.
436:
437: @item
438: Randy Smith finished the Sun FPA support.
439:
440: @item
441: Robert Brown implemented the support for Encore 32000 systems.
442:
443: @item
444: David Kashtan of SRI adapted GNU CC to the Vomit-Making System.
445:
446: @item
447: Alex Crain provided changes for the 3b1.
448:
449: @item
450: Greg Satz and Chris Hanson assisted in making GNU CC work on HP-UX for
451: the 9000 series 300.
452:
453: @item
454: William Schelter did most of the work on the Intel 80386 support.
1.1.1.5 root 455:
456: @item
457: Christopher Smith did the port for Convex machines.
458:
459: @item
460: Paul Petersen wrote the machine description for the Alliant FX/8.
1.1 root 461: @end itemize
462:
463: @node Options, Installation, Contributors, Top
464: @chapter GNU CC Command Options
465:
466: The GNU C compiler uses a command syntax much like the Unix C compiler.
467: The @code{gcc} program accepts options and file names as operands.
468: Multiple single-letter options may @emph{not} be grouped: @samp{-dr} is
469: very different from @samp{-d -r}.
470:
471: When you invoke GNU CC, it normally does preprocessing, compilation,
472: assembly and linking. File names which end in @samp{.c} are taken as C
1.1.1.5 root 473: source to be preprocessed and compiled; file names ending in @samp{.i}
474: are taken as preprocessor output to be compiled; compiler output files
475: plus any input files with names ending in @samp{.s} are assembled; then
476: the resulting object files, plus any other input files, are linked
477: together to produce an executable.
1.1 root 478:
479: Command options allow you to stop this process at an intermediate stage.
480: For example, the @samp{-c} option says not to run the linker. Then the
481: output consists of object files output by the assembler.
482:
1.1.1.5 root 483: Other command options are passed on to one stage of processing. Some
484: options control the preprocessor and others the compiler itself. Yet
485: other options control the assembler and linker; these are not documented
486: here, but you rarely need to use any of them.
1.1 root 487:
488: Here are the options to control the overall compilation process, including
489: those that say whether to link, whether to assemble, and so on.
490:
491: @table @samp
492: @item -o @var{file}
493: Place output in file @var{file}. This applies regardless to whatever
494: sort of output is being produced, whether it be an executable file,
495: an object file, an assembler file or preprocessed C code.
496:
497: If @samp{-o} is not specified, the default is to put an executable file
498: in @file{a.out}, the object file @file{@var{source}.c} in
499: @file{@var{source}.o}, an assembler file in @file{@var{source}.s}, and
500: preprocessed C on standard output.@refill
501:
502: @item -c
503: Compile or assemble the source files, but do not link. Produce object
504: files with names made by replacing @samp{.c} or @samp{.s} with
505: @samp{.o} at the end of the input file names. Do nothing at all for
506: object files specified as input.
507:
508: @item -S
509: Compile into assembler code but do not assemble. The assembler output
510: file name is made by replacing @samp{.c} with @samp{.s} at the end of
511: the input file name. Do nothing at all for assembler source files or
512: object files specified as input.
513:
514: @item -E
515: Run only the C preprocessor. Preprocess all the C source files
516: specified and output the results to standard output.
517:
518: @item -v
519: Compiler driver program prints the commands it executes as it runs
520: the preprocessor, compiler proper, assembler and linker. Some of
521: these are directed to print their own version numbers.
522:
1.1.1.5 root 523: @item -pipe
524: Use pipes rather than temporary files for communication between the
525: various stages of compilation. This fails to work on some systems
526: where the assembler is unable to read from a pipe; but the GNU
527: assembler has no trouble.
528:
1.1 root 529: @item -B@var{prefix}
530: Compiler driver program tries @var{prefix} as a prefix for each
531: program it tries to run. These programs are @file{cpp}, @file{cc1},
532: @file{as} and @file{ld}.
533:
534: For each subprogram to be run, the compiler driver first tries the
535: @samp{-B} prefix, if any. If that name is not found, or if @samp{-B}
536: was not specified, the driver tries two standard prefixes, which are
537: @file{/usr/lib/gcc-} and @file{/usr/local/lib/gcc-}. If neither of
538: those results in a file name that is found, the unmodified program
539: name is searched for using the directories specified in your
540: @samp{PATH} environment variable.
541:
542: The run-time support file @file{gnulib} is also searched for using
543: the @samp{-B} prefix, if needed. If it is not found there, the two
544: standard prefixes above are tried, and that is all. The file is left
545: out of the link if it is not found by those means. Most of the time,
546: on most machines, you can do without it.
1.1.1.5 root 547:
548: You can get a similar result from the environment variable;
549: @code{GCC_EXEC_PREFIX} if it is defined, its value is used as a prefix
550: in the same way. If both the @samp{-B} option and the
551: @code{GCC_EXEC_PREFIX} variable are present, the @samp{-B} option is
552: used first and the environment variable value second.
1.1 root 553: @end table
554:
555: These options control the details of C compilation itself.
556:
557: @table @samp
558: @item -ansi
559: Support all ANSI standard C programs.
560:
561: This turns off certain features of GNU C that are incompatible with
562: ANSI C, such as the @code{asm}, @code{inline} and @code{typeof}
563: keywords, and predefined macros such as @code{unix} and @code{vax}
564: that identify the type of system you are using. It also enables the
565: undesirable and rarely used ANSI trigraph feature.
566:
567: The @samp{-ansi} option does not cause non-ANSI programs to be
568: rejected gratuitously. For that, @samp{-pedantic} is required in
569: addition to @samp{-ansi}.
570:
571: The macro @code{__STRICT_ANSI__} is predefined when the @samp{-ansi}
572: option is used. Some header files may notice this macro and refrain
573: from declaring certain functions or defining certain macros that the
574: ANSI standard doesn't call for; this is to avoid interfering with
575: any programs that might use these names for other things.
576:
577: @item -traditional
578: Attempt to support some aspects of traditional C compilers.
579: Specifically:
580:
581: @itemize @bullet
582: @item
583: All @code{extern} declarations take effect globally even if they
584: are written inside of a function definition. This includes implicit
585: declarations of functions.
586:
587: @item
588: The keywords @code{typeof}, @code{inline}, @code{signed}, @code{const}
589: and @code{volatile} are not recognized.@refill
590:
591: @item
592: Comparisons between pointers and integers are always allowed.
593:
594: @item
595: Integer types @code{unsigned short} and @code{unsigned char} promote
596: to @code{unsigned int}.
597:
598: @item
599: Out-of-range floating point literals are not an error.
600:
601: @item
1.1.1.2 root 602: All automatic variables not declared @code{register} are preserved by
603: @code{longjmp}. Ordinarily, GNU C follows ANSI C: automatic variables
604: not declared @code{volatile} may be clobbered.
605:
606: @item
1.1 root 607: In the preprocessor, comments convert to nothing at all, rather than
608: to a space. This allows traditional token concatenation.
609:
610: @item
611: In the preprocessor, macro arguments are recognized within string
612: constants in a macro definition (and their values are stringified,
613: though without additional quote marks, when they appear in such a
614: context). The preprocessor always considers a string constant to end
615: at a newline.
616:
617: @item
618: The predefined macro @code{__STDC__} is not defined when you use
619: @samp{-traditional}, but @code{__GNUC__} is (since the GNU extensions
620: which @code{__GNUC__} indicates are not affected by
621: @samp{-traditional}). If you need to write header files that work
622: differently depending on whether @samp{-traditional} is in use, by
623: testing both of these predefined macros you can distinguish four
624: situations: GNU C, traditional GNU C, other ANSI C compilers, and
625: other old C compilers.
626: @end itemize
627:
628: @item -O
629: Optimize. Optimizing compilation takes somewhat more time, and a lot
630: more memory for a large function.
631:
632: Without @samp{-O}, the compiler's goal is to reduce the cost of
633: compilation and to make debugging produce the expected results.
634: Statements are independent: if you stop the program with a breakpoint
635: between statements, you can then assign a new value to any variable or
636: change the program counter to any other statement in the function and
637: get exactly the results you would expect from the source code.
638:
639: Without @samp{-O}, only variables declared @code{register} are
640: allocated in registers. The resulting compiled code is a little worse
641: than produced by PCC without @samp{-O}.
642:
643: With @samp{-O}, the compiler tries to reduce code size and execution
644: time.
645:
646: Some of the @samp{-f} options described below turn specific kinds of
647: optimization on or off.
648:
649: @item -g
650: Produce debugging information in the operating system's native format
651: (for DBX or SDB). GDB also can work with this debugging information.
652:
653: Unlike most other C compilers, GNU CC allows you to use @samp{-g} with
654: @samp{-O}. The shortcuts taken by optimized code may occasionally
655: produce surprising results: some variables you declared may not exist
656: at all; flow of control may briefly move where you did not expect it;
657: some statements may not be executed because they compute constant
658: results or their values were already at hand; some statements may
659: execute in different places because they were moved out of loops.
660: Nevertheless it proves possible to debug optimized output. This makes
661: it reasonable to use the optimizer for programs that might have bugs.
662:
663: @item -gg
664: Produce debugging information in GDB's own format. This requires the
665: GNU assembler and linker in order to work.
666:
667: This feature will probably be eliminated. It was intended to enable
668: GDB to read the symbol table faster, but it doesn't result in enough
669: of a speedup to be worth the larger object files and executables. We
1.1.1.2 root 670: are working on other ways of making GDB start even faster, which work
671: with DBX format debugging information and could be made to work with
672: SDB format.
1.1 root 673:
674: @item -w
675: Inhibit all warning messages.
676:
677: @item -W
678: Print extra warning messages for these events:
679:
680: @itemize @bullet
681: @item
682: An automatic variable is used without first being initialized.
683:
684: These warnings are possible only in optimizing compilation,
685: because they require data flow information that is computed only
1.1.1.6 ! root 686: when optimizing. If you don't specify @samp{-O}, you simply won't
! 687: get these warnings.
! 688:
! 689: These warnings occur only for variables that are candidates for
! 690: register allocation. Therefore, they do not occur for a variable that
! 691: is declared @code{volatile}, or whose address is taken, or whose size
! 692: is other than 1, 2, 4 or 8 bytes. Also, they do not occur for
! 693: structures, unions or arrays, even when they are in registers.
! 694:
! 695: Note that there may be no warning about a variable that is used only
! 696: to compute a value that itself is never used, because such
! 697: computations may be deleted by data flow analysis before the warnings
! 698: are printed.
1.1 root 699:
700: These warnings are made optional because GNU CC is not smart
701: enough to see all the reasons why the code might be correct
702: despite appearing to have an error. Here is one example of how
703: this can happen:
704:
705: @example
706: @{
707: int x;
708: switch (y)
709: @{
710: case 1: x = 1;
711: break;
712: case 2: x = 4;
713: break;
714: case 3: x = 5;
715: @}
716: foo (x);
717: @}
718: @end example
719:
720: @noindent
721: If the value of @code{y} is always 1, 2 or 3, then @code{x} is
722: always initialized, but GNU CC doesn't know this. Here is
723: another common case:
724:
725: @example
726: @{
727: int save_y;
728: if (change_y) save_y = y, y = new_y;
729: @dots{}
730: if (change_y) y = save_y;
731: @}
732: @end example
733:
734: @noindent
735: This has no bug because @code{save_y} is used only if it is set.
736:
1.1.1.5 root 737: Some spurious warnings can be avoided if you declare as
738: @code{volatile} all the functions you use that never return.
739: @xref{Function Attributes}.
740:
1.1 root 741: @item
742: A nonvolatile automatic variable might be changed by a call to
743: @code{longjmp}. These warnings as well are possible only in
744: optimizing compilation.
745:
746: The compiler sees only the calls to @code{setjmp}. It cannot know
747: where @code{longjmp} will be called; in fact, a signal handler could
748: call it at any point in the code. As a result, you may get a warning
749: even when there is in fact no problem because @code{longjmp} cannot
750: in fact be called at the place which would cause a problem.
751:
752: @item
753: A function can return either with or without a value. (Falling
754: off the end of the function body is considered returning without
1.1.1.6 ! root 755: a value.) For example, this function would evoke such a
1.1 root 756: warning:
757:
758: @example
759: foo (a)
760: @{
761: if (a > 0)
762: return a;
763: @}
764: @end example
765:
766: Spurious warnings can occur because GNU CC does not realize that
767: certain functions (including @code{abort} and @code{longjmp})
768: will never return.
1.1.1.4 root 769:
770: @item
771: An expression-statement contains no side effects.
1.1 root 772: @end itemize
773:
774: In the future, other useful warnings may also be enabled by this
775: option.
776:
777: @item -Wimplicit
778: Warn whenever a function is implicitly declared.
779:
780: @item -Wreturn-type
781: Warn whenever a function is defined with a return-type that defaults
782: to @code{int}. Also warn about any @code{return} statement with no
783: return-value in a function whose return-type is not @code{void}.
784:
785: @item -Wunused
1.1.1.5 root 786: Warn whenever a local variable is unused aside from its declaration,
787: and whenever a function is declared static but never defined.
1.1 root 788:
789: @item -Wcomment
790: Warn whenever a comment-start sequence @samp{/*} appears in a comment.
791:
792: @item -Wall
793: All of the above @samp{-W} options combined.
794:
1.1.1.6 ! root 795: @item -Wcast-qual
! 796: Warn whenever a pointer is cast so as to remove a type qualifier from
! 797: the target type. For example, warn if a @code{const char *} is cast
! 798: to an ordinary @code{char *}.
! 799:
1.1 root 800: @item -Wwrite-strings
801: Give string constants the type @code{const char[@var{length}]} so that
802: copying the address of one into a non-@code{const} @code{char *}
803: pointer will get a warning. These warnings will help you find at
804: compile time code that can try to write into a string constant, but
805: only if you have been very careful about using @code{const} in
806: declarations and prototypes. Otherwise, it will just be a nuisance;
807: this is why we did not make @samp{-Wall} request these warnings.
808:
809: @item -p
810: Generate extra code to write profile information suitable for the
811: analysis program @code{prof}.
812:
813: @item -pg
814: Generate extra code to write profile information suitable for the
815: analysis program @code{gprof}.
816:
1.1.1.6 ! root 817: @item -a
! 818: Generate extra code to write profile information for basic blocks,
! 819: suitable for the analysis program @code{tcov}. Eventually GNU
! 820: @code{gprof} should be extended to process this data.
! 821:
1.1 root 822: @item -l@var{library}
823: Search a standard list of directories for a library named
824: @var{library}, which is actually a file named
825: @file{lib@var{library}.a}. The linker uses this file as if it
826: had been specified precisely by name.
827:
828: The directories searched include several standard system directories
829: plus any that you specify with @samp{-L}.
830:
831: Normally the files found this way are library files---archive files
832: whose members are object files. The linker handles an archive file by
833: scanning through it for members which define symbols that have so far
834: been referenced but not defined. But if the file that is found is an
835: ordinary object file, it is linked in the usual fashion. The only
836: difference between using an @samp{-l} option and specifying a file name
837: is that @samp{-l} searches several directories.
838:
839: @item -L@var{dir}
840: Add directory @var{dir} to the list of directories to be searched
841: for @samp{-l}.
842:
843: @item -nostdlib
844: Don't use the standard system libraries and startup files when
845: linking. Only the files you specify (plus @file{gnulib}) will be
846: passed to the linker.
847:
848: @item -m@var{machinespec}
849: Machine-dependent option specifying something about the type of target
850: machine. These options are defined by the macro
851: @code{TARGET_SWITCHES} in the machine description. The default for
852: the options is also defined by that macro, which enables you to change
853: the defaults.@refill
854:
855: These are the @samp{-m} options defined in the 68000 machine
856: description:
857:
858: @table @samp
859: @item -m68020
860: @itemx -mc68020
861: Generate output for a 68020 (rather than a 68000). This is the
862: default if you use the unmodified sources.
863:
864: @item -m68000
865: @item -mc68000
866: Generate output for a 68000 (rather than a 68020).
867:
868: @item -m68881
869: Generate output containing 68881 instructions for floating point.
870: This is the default if you use the unmodified sources.
871:
872: @item -mfpa
873: Generate output containing Sun FPA instructions for floating point.
874:
875: @item -msoft-float
876: Generate output containing library calls for floating point.
877:
878: @item -mshort
879: Consider type @code{int} to be 16 bits wide, like @code{short int}.
880:
881: @item -mnobitfield
882: Do not use the bit-field instructions. @samp{-m68000} implies
883: @samp{-mnobitfield}.
884:
885: @item -mbitfield
886: Do use the bit-field instructions. @samp{-m68020} implies
887: @samp{-mbitfield}. This is the default if you use the unmodified
888: sources.
889:
890: @item -mrtd
891: Use a different function-calling convention, in which functions
892: that take a fixed number of arguments return with the @code{rtd}
893: instruction, which pops their arguments while returning. This
894: saves one instruction in the caller since there is no need to pop
895: the arguments there.
896:
897: This calling convention is incompatible with the one normally
898: used on Unix, so you cannot use it if you need to call libraries
899: compiled with the Unix compiler.
900:
901: Also, you must provide function prototypes for all functions that
902: take variable numbers of arguments (including @code{printf});
903: otherwise incorrect code will be generated for calls to those
904: functions.
905:
906: In addition, seriously incorrect code will result if you call a
907: function with too many arguments. (Normally, extra arguments are
908: harmlessly ignored.)
909:
910: The @code{rtd} instruction is supported by the 68010 and 68020
911: processors, but not by the 68000.
912: @end table
913:
914: These @samp{-m} options are defined in the Vax machine description:
915:
916: @table @samp
917: @item -munix
918: Do not output certain jump instructions (@code{aobleq} and so on)
919: that the Unix assembler for the Vax cannot handle across long
920: ranges.
921:
922: @item -mgnu
923: Do output those jump instructions, on the assumption that you
924: will assemble with the GNU assembler.
925:
926: @item -mg
927: Output code for g-format floating point numbers instead of d-format.
928: @end table
929:
1.1.1.5 root 930: These @samp{-m} switches are supported on the Sparc:
931:
932: @table @samp
933: @item -mfpu
934: Generate output containing floating point instructions. This is the
935: default if you use the unmodified sources.
936:
937: @item -msoft-float
938: Generate output containing library calls for floating point.
939:
940: @item -mno-epilogue
1.1.1.6 ! root 941: Generate separate return instructions for @code{return} statements.
! 942: This has both advantages and disadvantages; I don't recall what they
! 943: are.
1.1.1.5 root 944:
945: @item -meager
946: Do eager conditional branch scheduling to fill no-op slots. This
947: optimization is new, so we suspect it has bugs; some day it will be
948: done by default, but it is optional now so you can test it when you
949: are ready.
950:
951: @emph{Test it now}, and report the bugs; otherwise we won't find them,
952: and this option may become the default with bugs still in it!
953: @end table
954:
955: These @samp{-m} options are defined in the Convex machine description:
956:
957: @table @samp
958: @item -mc1
959: Generate output for a C1. This is the default when the compiler is
960: configured for a C1.
961:
962: @item -mc2
963: Generate output for a C2. This is the default when the compiler is
964: configured for a C2.
965:
966: @item -margcount
967: Generate code which puts an argument count in the word preceding each
968: argument list. Some nonportable Convex and Vax programs need this
969: word. (Debuggers don't; this info is in the symbol table.)
970:
971: @item -mnoargcount
972: Omit the argument count word. This is the default if you use the
973: unmodified sources.
974: @end table
975:
1.1 root 976: @item -f@var{flag}
1.1.1.4 root 977: Specify machine-independent flags. Most flags have both positive and
978: negative forms; the negative form of @samp{-ffoo} would be
979: @samp{-fno-foo}. In the table below, only one of the forms is
980: listed---the one which is not the default. You can figure out the
981: other form by either removing @samp{no-} or adding it.
1.1 root 982:
983: @table @samp
1.1.1.6 ! root 984: @item -fpcc-struct-return
! 985: Use the same convention for returning @code{struct} and @code{union}
! 986: values that is used by the usual C compiler on your system. This
! 987: convention is less efficient for small structures, and on many
! 988: machines it fails to be reentrant; but it has the advantage of
! 989: allowing intercallability between GCC-compiled code and PCC-compiled
! 990: code.
! 991:
1.1 root 992: @item -ffloat-store
993: Do not store floating-point variables in registers. This
994: prevents undesirable excess precision on machines such as the
995: 68000 where the floating registers (of the 68881) keep more
996: precision than a @code{double} is supposed to have.
997:
998: For most programs, the excess precision does only good, but a few
999: programs rely on the precise definition of IEEE floating point.
1000: Use @samp{-ffloat-store} for such programs.
1001:
1002: @item -fno-asm
1003: Do not recognize @code{asm}, @code{inline} or @code{typeof} as a
1004: keyword. These words may then be used as identifiers.
1005:
1006: @item -fno-defer-pop
1007: Always pop the arguments to each function call as soon as that
1008: function returns. Normally the compiler (when optimizing) lets
1009: arguments accumulate on the stack for several function calls and
1010: pops them all at once.
1011:
1012: @item -fstrength-reduce
1013: Perform the optimizations of loop strength reduction and
1014: elimination of iteration variables.
1015:
1016: @item -fcombine-regs
1017: Allow the combine pass to combine an instruction that copies one
1018: register into another. This might or might not produce better
1019: code when used in addition to @samp{-O}. I am interested in
1020: hearing about the difference this makes.
1021:
1022: @item -fforce-mem
1023: Force memory operands to be copied into registers before doing
1024: arithmetic on them. This may produce better code by making all
1025: memory references potential common subexpressions. When they are
1026: not common subexpressions, instruction combination should
1027: eliminate the separate register-load. I am interested in hearing
1028: about the difference this makes.
1029:
1030: @item -fforce-addr
1031: Force memory address constants to be copied into registers before
1032: doing arithmetic on them. This may produce better code just as
1033: @samp{-fforce-mem} may. I am interested in hearing about the
1034: difference this makes.
1035:
1036: @item -fomit-frame-pointer
1037: Don't keep the frame pointer in a register for functions that
1038: don't need one. This avoids the instructions to save, set up and
1039: restore frame pointers; it also makes an extra register available
1040: in many functions. @strong{It also makes debugging impossible.}
1041:
1042: On some machines, such as the Vax, this flag has no effect,
1043: because the standard calling sequence automatically handles the
1044: frame pointer and nothing is saved by pretending it doesn't
1045: exist. The machine-description macro
1046: @code{FRAME_POINTER_REQUIRED} controls whether a target machine
1047: supports this flag. @xref{Registers}.@refill
1048:
1049: @item -finline-functions
1050: Integrate all simple functions into their callers. The compiler
1051: heuristically decides which functions are simple enough to be
1052: worth integrating in this way.
1053:
1054: If all calls to a given function are integrated, and the function
1055: is declared @code{static}, then the function is normally not
1056: output as assembler code in its own right.
1057:
1.1.1.6 ! root 1058: @item -fcaller-saves
! 1059: Enable values to be allocated in registers that will be clobbered by
! 1060: function calls, by emitting extra instructions to save and restore the
! 1061: registers around such calls. Such allocation is done only when it
! 1062: seems to result in better code than would otherwise be produced.
! 1063:
! 1064: This option is enabled by default on certain machines, usually those
! 1065: which have no call-preserved registers to use instead.
! 1066:
1.1 root 1067: @item -fkeep-inline-functions
1068: Even if all calls to a given function are integrated, and the
1069: function is declared @code{static}, nevertheless output a
1070: separate run-time callable version of the function.
1071:
1072: @item -fwritable-strings
1073: Store string constants in the writable data segment and don't
1074: uniquize them. This is for compatibility with old programs which
1075: assume they can write into string constants. Writing into string
1076: constants is a very bad idea; ``constants'' should be constant.
1077:
1.1.1.4 root 1078: @item -fcond-mismatch
1079: Allow conditional expressions with mismatched types in the second and
1080: third arguments. The value of such an expression is void.
1081:
1.1 root 1082: @item -fno-function-cse
1083: Do not put function addresses in registers; make each instruction
1084: that calls a constant function contain the function's address
1085: explicitly.
1086:
1087: This option results in less efficient code, but some strange
1088: hacks that alter the assembler output may be confused by the
1089: optimizations performed when this option is not used.
1090:
1091: @item -fvolatile
1092: Consider all memory references through pointers to be volatile.
1093:
1.1.1.4 root 1094: @item -fshared-data
1095: Requests that the data and non-@code{const} variables of this
1096: compilation be shared data rather than private data. The distinction
1097: makes sense only on certain operating systems, where shared data is
1098: shared between processes running the same program, while private data
1099: exists in one copy per process.
1100:
1.1 root 1101: @item -funsigned-char
1.1.1.4 root 1102: Let the type @code{char} be the unsigned, like @code{unsigned char}.
1.1 root 1103:
1104: Each kind of machine has a default for what @code{char} should
1105: be. It is either like @code{unsigned char} by default or like
1106: @code{signed char} by default. (Actually, at present, the
1107: default is always signed.)
1108:
1109: The type @code{char} is always a distinct type from either
1110: @code{signed char} or @code{unsigned char}, even though its
1111: behavior is always just like one of those two.
1112:
1.1.1.4 root 1113: Note that this is equivalent to @samp{-fno-signed-char}, which is the
1114: negative form of @samp{-fsigned-char}.
1115:
1.1 root 1116: @item -fsigned-char
1117: Let the type @code{char} be signed, like @code{signed char}.
1118:
1.1.1.4 root 1119: Note that this is equivalent to @samp{-fno-unsigned-char}, which is
1120: the negative form of @samp{-funsigned-char}.
1121:
1.1 root 1122: @item -ffixed-@var{reg}
1123: Treat the register named @var{reg} as a fixed register; generated
1124: code should never refer to it (except perhaps as a stack pointer,
1125: frame pointer or in some other fixed role).
1126:
1127: @var{reg} must be the name of a register. The register names
1128: accepted are machine-specific and are defined in the
1129: @code{REGISTER_NAMES} macro in the machine description macro
1130: file.
1131:
1.1.1.4 root 1132: This flag does not have a negative form, because it specifies a
1133: three-way choice.
1134:
1.1 root 1135: @item -fcall-used-@var{reg}
1136: Treat the register named @var{reg} as an allocatable register
1137: that is clobbered by function calls. It may be allocated for
1138: temporaries or variables that do not live across a call.
1139: Functions compiled this way will not save and restore the
1140: register @var{reg}.
1141:
1142: Use of this flag for a register that has a fixed pervasive role
1143: in the machine's execution model, such as the stack pointer or
1144: frame pointer, will produce disastrous results.
1145:
1.1.1.4 root 1146: This flag does not have a negative form, because it specifies a
1147: three-way choice.
1148:
1.1 root 1149: @item -fcall-saved-@var{reg}
1150: Treat the register named @var{reg} as an allocatable register
1151: saved by functions. It may be allocated even for temporaries or
1152: variables that live across a call. Functions compiled this way
1153: will save and restore the register @var{reg} if they use it.
1154:
1155: Use of this flag for a register that has a fixed pervasive role
1156: in the machine's execution model, such as the stack pointer or
1157: frame pointer, will produce disastrous results.
1158:
1159: A different sort of disaster will result from the use of this
1160: flag for a register in which function values may be returned.
1.1.1.4 root 1161:
1162: This flag does not have a negative form, because it specifies a
1163: three-way choice.
1.1 root 1164: @end table
1165:
1166: @item -d@var{letters}
1167: Says to make debugging dumps at times specified by @var{letters}.
1168: Here are the possible letters:
1169:
1170: @table @samp
1171: @item r
1172: Dump after RTL generation.
1173: @item j
1174: Dump after first jump optimization.
1175: @item J
1176: Dump after last jump optimization.
1177: @item s
1178: Dump after CSE (including the jump optimization that sometimes
1179: follows CSE).
1180: @item L
1181: Dump after loop optimization.
1182: @item f
1183: Dump after flow analysis.
1184: @item c
1185: Dump after instruction combination.
1186: @item l
1187: Dump after local register allocation.
1188: @item g
1189: Dump after global register allocation.
1190: @item m
1191: Print statistics on memory usage, at the end of the run.
1192: @end table
1193:
1194: @item -pedantic
1195: Issue all the warnings demanded by strict ANSI standard C; reject
1196: all programs that use forbidden extensions.
1197:
1198: Valid ANSI standard C programs should compile properly with or without
1199: this option (though a rare few will require @samp{-ansi}). However,
1200: without this option, certain GNU extensions and traditional C features
1201: are supported as well. With this option, they are rejected. There is
1202: no reason to @i{use} this option; it exists only to satisfy pedants.
1.1.1.5 root 1203:
1204: @item -static
1205: On Suns running version 4, this prevents linking with the shared
1206: libraries. (@samp{-g} has the same effect.)
1.1 root 1207: @end table
1208:
1209: These options control the C preprocessor, which is run on each C source
1210: file before actual compilation. If you use the @samp{-E} option, nothing
1211: is done except C preprocessing. Some of these options make sense only
1212: together with @samp{-E} because they request preprocessor output that is
1213: not suitable for actual compilation.
1214:
1215: @table @samp
1216: @item -C
1217: Tell the preprocessor not to discard comments. Used with the
1218: @samp{-E} option.
1219:
1220: @item -I@var{dir}
1221: Search directory @var{dir} for include files.
1222:
1223: @item -I-
1224: Any directories specified with @samp{-I} options before the @samp{-I-}
1225: option are searched only for the case of @samp{#include "@var{file}"};
1226: they are not searched for @samp{#include <@var{file}>}.
1227:
1228: If additional directories are specified with @samp{-I} options after
1229: the @samp{-I-}, these directories are searched for all @samp{#include}
1230: directives. (Ordinarily @emph{all} @samp{-I} directories are used
1231: this way.)
1232:
1233: In addition, the @samp{-I-} option inhibits the use of the current
1234: directory as the first search directory for @samp{#include
1235: "@var{file}"}. Therefore, the current directory is searched only if
1236: it is requested explicitly with @samp{-I.}. Specifying both
1237: @samp{-I-} and @samp{-I.} allows you to control precisely which
1238: directories are searched before the current one and which are searched
1239: after.
1240:
1241: @item -nostdinc
1242: Do not search the standard system directories for header files. Only
1243: the directories you have specified with @samp{-I} options (and the
1244: current directory, if appropriate) are searched.
1245:
1246: Between @samp{-nostdinc} and @samp{-I-}, you can eliminate all
1247: directories from the search path except those you specify.
1248:
1249: @item -M
1250: Tell the preprocessor to output a rule suitable for @code{make}
1251: describing the dependencies of each source file. For each source
1252: file, the preprocessor outputs one @code{make}-rule whose target is
1253: the object file name for that source file and whose dependencies are
1254: all the files @samp{#include}d in it. This rule may be a single line
1255: or may be continued with @samp{\}-newline if it is long.
1256:
1257: @samp{-M} implies @samp{-E}.
1258:
1259: @item -MM
1260: Like @samp{-M} but the output mentions only the user-header files
1261: included with @samp{#include "@var{file}"}. System header files
1262: included with @samp{#include <@var{file}>} are omitted.
1263:
1264: @samp{-MM} implies @samp{-E}.
1265:
1266: @item -D@var{macro}
1267: Define macro @var{macro} with the empty string as its definition.
1268:
1269: @item -D@var{macro}=@var{defn}
1270: Define macro @var{macro} as @var{defn}.
1271:
1272: @item -U@var{macro}
1273: Undefine macro @var{macro}.
1274:
1275: @item -T
1276: Support ANSI C trigraphs. You don't want to know about this
1277: brain-damage. The @samp{-ansi} option also has this effect.
1278: @end table
1279:
1280: @node Installation, Trouble, Options, Top
1281: @chapter Installing GNU CC
1282:
1283: Here is the procedure for installing GNU CC on a Unix system.
1284: @menu
1285: * VMS Install:: See below for installation on VMS.
1286: @end menu
1287: @iftex
1288: (See below for VMS.)
1289: @end iftex
1290:
1291: @enumerate
1292: @item
1293: Edit @file{Makefile}. If you are using HPUX, or any form of system V,
1294: you must make a few changes described in comments at the beginning of
1.1.1.4 root 1295: the file. Genix requires changes also.
1.1 root 1296:
1297: @item
1298: On a Sequent system, go to the Berkeley universe.
1299:
1300: @item
1.1.1.2 root 1301: Choose configuration files. The easy way to do this is to run the
1302: command file @file{config.gcc} with a single argument, which is the
1.1.1.4 root 1303: name of the machine (and operating system, in some cases).
1304:
1305: Here is a list of the possible arguments:
1306:
1307: @table @samp
1308: @item vax
1309: Vaxes running BSD.
1310: @item vms
1311: Vaxes running VMS.
1312: @item vax-sysv
1313: Vaxes running system V.
1314: @item i386-sysv
1315: Intel 386 PCs running system V.
1.1.1.5 root 1316: @item i386-sysv-gas
1317: Intel 386 PCs running system V, using the GNU assembler and GNU
1318: linker.
1.1.1.6 ! root 1319: @item sequent-i386
1.1.1.4 root 1320: Sequent with Intel 386 processors.
1321: @item sun2
1322: Sun 2 running system version 2 or 3.
1323: @item sun3
1.1.1.5 root 1324: Sun 3 running system version 2 or 3, with 68881.
1325: @item sun3-nfp
1326: Sun 3 running system version 2 or 3, without 68881.
1327: @item sun3-fpa
1328: Sun 3 running system version 2 or 3, with 68881 and fpa.
1.1.1.4 root 1329: @item sun4
1330: Sun 4 running system version 2 or 3.
1331: @item sun2-os4
1332: Sun 2 running system version 4.
1333: @item sun3-os4
1.1.1.5 root 1334: Sun 3 running system version 4, with 68881.
1335: @item sun3-nfp-os4
1336: Sun 3 running system version 4, without 68881.
1337: @item sun3-fpa-os4
1338: Sun 3 running system version 4, with 68881 and fpa.
1.1.1.4 root 1339: @item sun4-os4
1340: Sun 4 running system version 4.
1341: @item sun386
1342: Sun 386 (``roadrunner'').
1.1.1.5 root 1343: @item alliant
1344: Alliant FX/8 computer. Currently, there are bugs in the support for
1345: floating point. Also note that Alliant's version of dbx does not
1346: manage to work with the output from GNU CC.
1347: @item convex-c1
1348: Convex C1 computer.
1349: @item convex-c2
1350: Convex C2 computer.
1.1.1.4 root 1351: @item hp9k320
1352: HP 9000 series 300 using HPUX assembler.
1.1.1.6 ! root 1353: @item hp9k320g
1.1.1.4 root 1354: HP 9000 series 300 using GNU assembler, linker and debugger.
1355: This requires the HP-adapt package which is or will soon be
1.1.1.5 root 1356: available along with the linker.
1.1.1.4 root 1357: @item isi68
1358: ISI 68000 or 68020 system.
1359: @item news800
1360: Sony NEWS 68020 system.
1.1.1.6 ! root 1361: @item next
! 1362: NeXT system.
1.1.1.4 root 1363: @item 3b1
1364: AT&T 3b1, a.k.a. 7300 PC.
1365: @item sequent-ns32k
1366: Sequent containing ns32000 processors.
1367: @item encore
1368: Encore ns32000 system.
1369: @item genix
1370: National Semiconductor ns32000 system.
1371: @item 88000
1372: Motorola 88000 processor. This port is not finished.
1373: @end table
1.1.1.2 root 1374:
1.1.1.4 root 1375: Here we spell out what files need to be set up:
1.1 root 1376:
1377: @itemize @bullet
1378: @item
1379: Make a symbolic link named @file{config.h} to the top-level
1380: config file for the machine you are using (@pxref{Config}). This
1381: file is responsible for defining information about the host
1382: machine. It includes @file{tm.h}.
1383:
1.1.1.3 root 1384: The file's name should be @file{xm-@var{machine}.h}, with these
1.1 root 1385: exceptions:
1386:
1387: @table @file
1.1.1.3 root 1388: @item xm-vms.h
1.1 root 1389: for vaxen running VMS.
1.1.1.3 root 1390: @item xm-vaxv.h
1.1 root 1391: for vaxen running system V.
1.1.1.3 root 1392: @item xm-i386v.h
1.1 root 1393: for Intel 80386's running system V.
1.1.1.3 root 1394: @item xm-sunos4.h
1.1 root 1395: for Suns (model 2, 3 or 4) running @emph{operating system} version 4.
1.1.1.3 root 1396: (Use @file{xm-m68k.h} or @file{xm-sparc.h} for version 3.)
1397: @item xm-sun386i.h
1398: for Sun roadrunner running any version of the operating system.
1399: @item xm-hp9k320.h
1.1 root 1400: for the HP 9000 series 300.
1.1.1.4 root 1401: @item xm-genix.h
1.1 root 1402: for the ns32000 running Genix
1403: @end table
1404:
1405: If your system does not support symbolic links, you might want to
1406: set up @file{config.h} to contain a @samp{#include} command which
1407: refers to the appropriate file.
1408:
1409: @item
1410: Make a symbolic link named @file{tm.h} to the machine-description
1411: macro file for your machine (its name should be
1412: @file{tm-@var{machine}.h}).
1413:
1414: If your system is a 68000, don't use the file @file{tm-m68k.h}
1415: directly. Instead, use one of these files:
1416:
1417: @table @file
1418: @item tm-sun3.h
1.1.1.5 root 1419: for Sun 3 machines with 68881.
1420: @item tm-sun3-fpa.h
1421: for Sun 3 machines with floating point accelerator.
1422: @item tm-sun3-nfp.h
1423: for Sun 3 machines with no hardware floating point.
1.1 root 1424: @item tm-sun2.h
1425: for Sun 2 machines.
1426: @item tm-3b1.h
1427: for AT&T 3b1 (aka 7300 Unix PC).
1428: @item tm-isi68.h
1.1.1.3 root 1429: for Integrated Solutions systems. This file assumes you
1430: use the GNU assembler.
1.1 root 1431: @item tm-news800.h
1432: for SONY News systems.
1433: @item tm-hp9k320.h
1434: for HPUX systems, if you are using GNU CC with the system's
1435: assembler and linker.
1436: @item tm-hp9k320g.h
1437: for HPUX systems, if you are using the GNU assembler, linker and
1438: other utilities. Not all of the pieces of GNU software needed
1439: for this mode of operation are as yet in distribution; full
1440: instructions will appear here in the future.@refill
1441: @end table
1442:
1443: For the vax, use @file{tm-vax.h} on BSD Unix, @file{tm-vaxv.h} on
1444: system V, or @file{tm-vms.h} on VMS.@refill
1445:
1446: For the Motorola 88000, use @file{tm-m88k.h}. The support for the
1447: 88000 has a few unfinished spots because there was no way to run the
1.1.1.2 root 1448: output. Bugs are suspected in handling of branch-tables and in the
1449: function prologue and epilogue.
1.1 root 1450:
1451: For the 80386, don't use @file{tm-i386.h} directly. Use
1452: @file{tm-i386v.h} if the target machine is running system V,
1.1.1.5 root 1453: @file{tm-i386gas.h} if it is running system V but you are using the
1454: GNU assembler and linker, @file{tm-seq386.h} for a Sequent 386 system,
1455: or @file{tm-compaq.h} for a Compaq, or @file{tm-sun386i.h} for a Sun
1456: 386 system.
1.1 root 1457:
1458: For the 32000, use @file{tm-sequent.h} if you are using a Sequent
1459: machine, or @file{tm-encore.h} for an Encore machine, or
1.1.1.4 root 1460: @file{tm-genix.h} if you are using Genix version 3; otherwise, perhaps
1.1 root 1461: @file{tm-ns32k.h} will work for you.
1462:
1463: Note that Genix has bugs in @code{alloca} and @code{malloc}; you must
1464: get the compiled versions of these from GNU Emacs and edit GNU CC's
1465: @file{Makefile} to use them.
1466:
1467: Note that Encore systems are supported only under BSD.
1468:
1.1.1.6 ! root 1469: For Sparc (Sun 4) machines, use @file{tm-sparc.h} with operating system
! 1470: version 4, and @file{tm-sun4os3.h} with system version 3.
! 1471:
1.1 root 1472: @item
1473: Make a symbolic link named @file{md} to the machine description
1.1.1.2 root 1474: pattern file. Its name should be @file{@var{machine}.md}, but
1475: @var{machine} is often not the same as the name used in the
1476: @file{tm.h} file because the @file{md} files are more general.
1.1 root 1477:
1478: @item
1479: Make a symbolic link named @file{aux-output.c} to the output
1480: subroutine file for your machine (its name should be
1481: @file{output-@var{machine}.c}).
1482: @end itemize
1483:
1484: @item
1485: Make sure the Bison parser generator is installed. (This is
1486: unnecessary if the Bison output files @file{c-parse.tab.c} and
1487: @file{cexp.c} are more recent than @file{c-parse.y} and @file{cexp.y}
1488: and you do not plan to change the @samp{.y} files.)
1489:
1490: Bison versions older that Sept 8, 1988 will produce incorrect output
1491: for @file{c-parse.tab.c}.
1492:
1493: @item
1494: If you are using a Sun, make sure the environment variable
1495: @code{FLOAT_OPTION} is not set. If this option were set to
1496: @code{f68881} when @file{gnulib} is compiled, the resulting code would
1497: demand to be linked with a special startup file and will not link
1498: properly without special pains.
1499:
1500: @item
1501: Build the compiler. Just type @samp{make} in the compiler directory.
1502:
1.1.1.2 root 1503: Ignore any warnings you may see about ``statement not reached'' in the
1504: @file{insn-emit.c}; they are normal. Any other compilation errors may
1505: represent bugs in the port to your machine or operating system, and
1506: should be investigated and reported (@pxref{Bugs}).
1507:
1.1 root 1508: @item
1.1.1.5 root 1509: If you are using COFF-encapsulation, you must convert @file{gnulib} to
1510: a GNU-format library at this point. See the file @file{README-ENCAP}
1511: in the directory containing the GNU binary file utilities, for
1512: directions.
1513:
1514: @item
1.1 root 1515: Move the first-stage object files and executables into a subdirectory
1516: with this command:
1517:
1518: @example
1519: make stage1
1520: @end example
1521:
1522: The files are moved into a subdirectory named @file{stage1}.
1523: Once installation is complete, you may wish to delete these files
1524: with @code{rm -r stage1}.
1525:
1526: @item
1527: Recompile the compiler with itself, with this command:
1528:
1529: @example
1530: make CC=stage1/gcc CFLAGS="-g -O -Bstage1/"
1531: @end example
1532:
1533: On a 68000 or 68020 system lacking floating point hardware,
1534: unless you have selected a @file{tm.h} file that expects by default
1535: that there is no such hardware, do this instead:
1536:
1537: @example
1538: make CC=stage1/gcc CFLAGS="-g -O -Bstage1/ -msoft-float"
1539: @end example
1540:
1541: @item
1542: If you wish to test the compiler by compiling it with itself one more
1543: time, do this:
1544:
1545: @example
1546: make stage2
1547: make CC=stage2/gcc CFLAGS="-g -O -Bstage2/"
1548: foreach file (*.o)
1549: cmp $file stage2/$file
1550: end
1551: @end example
1552:
1553: This will notify you if any of these stage 3 object files differs from
1554: those of stage 2. Any difference, no matter how innocuous, indicates
1555: that the stage 2 compiler has compiled GNU CC incorrectly, and is
1556: therefore a potentially serious bug which you should investigate and
1557: report (@pxref{Bugs}).
1558:
1559: Aside from the @samp{-B} option, the options should be the same as
1560: when you made stage 2.
1561:
1562: @item
1563: Install the compiler driver, the compiler's passes and run-time support.
1564: You can use the following command:
1565:
1566: @example
1567: make install
1568: @end example
1569:
1570: @noindent
1571: This copies the files @file{cc1}, @file{cpp} and @file{gnulib} to
1572: files @file{gcc-cc1}, @file{gcc-cpp} and @file{gcc-gnulib} in
1573: directory @file{/usr/local/lib}, which is where the compiler driver
1574: program looks for them. It also copies the driver program @file{gcc}
1.1.1.6 ! root 1575: into the directory @file{/usr/local/bin}, so that it appears in typical
1.1 root 1576: execution search paths.@refill
1577:
1578: @strong{Warning: there is a bug in @code{alloca} in the Sun library.
1579: To avoid this bug, install the binaries of GNU CC that were compiled
1580: by GNU CC. They use @code{alloca} as a built-in function and never
1581: the one in the library.}
1582:
1583: @strong{Warning: the GNU CPP may not work for @file{ioctl.h},
1584: @file{ttychars.h} and other system header files unless the
1585: @samp{-traditional} option is used.} The bug is in the header files:
1586: at least on some machines, they rely on behavior that is incompatible
1587: with ANSI C. This behavior consists of substituting for macro
1588: argument names when they appear inside of character constants. The
1589: @samp{-traditional} option tells GNU CC to behave the way these
1590: headers expect.
1591:
1592: Because of this problem, you might prefer to configure GNU CC to use
1593: the system's own C preprocessor. To do so, make the file
1594: @file{/usr/local/lib/gcc-cpp} a link to @file{/lib/cpp}.
1595:
1596: Alternatively, on Sun systems and 4.3BSD at least, you can correct the
1597: include files by running the shell script @file{fixincludes}. This
1598: installs modified, corrected copies of the files @file{ioctl.h},
1599: @file{ttychars.h} and many others, in a special directory where only
1.1.1.2 root 1600: GNU CC will normally look for them. This script will work on various
1.1.1.6 ! root 1601: systems because it chooses the files by searching all the system
1.1.1.2 root 1602: headers for the problem cases that we know about.
1.1 root 1603: @end enumerate
1604:
1605: If you cannot install the compiler's passes and run-time support in
1606: @file{/usr/local/lib}, you can alternatively use the @samp{-B} option to
1607: specify a prefix by which they may be found. The compiler concatenates
1608: the prefix with the names @file{cpp}, @file{cc1} and @file{gnulib}.
1609: Thus, you can put the files in a directory @file{/usr/foo/gcc} and
1610: specify @samp{-B/usr/foo/gcc/} when you run GNU CC.
1611:
1612: Also, you can specify an alternative default directory for these files
1613: by setting the Make variable @code{libdir} when you make GNU CC.
1614:
1615: @node VMS Install,, Installation, Installation
1616: @section Installing GNU CC on VMS
1617:
1.1.1.4 root 1618: The VMS version of GNU CC is distributed in a backup saveset containing
1619: both source code and precompiled binaries.
1620:
1621: Sometimes the binaries will be from an older version that the sources,
1622: because we don't always have time to update them. In this case, you
1623: should use the binaries you get to recompile the sources. If you must
1.1 root 1624: recompile, here is how:
1625:
1626: @enumerate
1627: @item
1.1.1.3 root 1628: Copy the file @file{tm-vms.h} to @file{tm.h}, @file{xm-vms.h} to
1.1 root 1629: @file{config.h}, @file{vax.md} to @file{md.} and @file{output-vax.c}
1630: to @file{aux-output.c}.@refill
1631:
1632: @item
1633: Type @samp{@@make} to do recompile everything.
1.1.1.5 root 1634:
1635: If you are compiling with a version of GCC older than 1.33, specify
1636: @samp{/DEFINE=("inline=")} as an option in all the compilations. This
1637: requires editing all the @code{gcc} commands in @file{make-cc1.com}.
1638: (The older versions had problems supporting @code{inline}.) Once you
1639: have a working 1.33 or newer GCC, you can change this file back.
1.1 root 1640: @end enumerate
1641:
1642: To install the @samp{GCC} command so you can use the compiler easily, in
1643: the same manner as you use the VMS C compiler, you must install the VMS CLD
1644: file for GNU CC as follows:
1645:
1646: @enumerate
1647: @item
1648: Define the VMS logical names @samp{GNU_CC} and @samp{GNU_CC_INCLUDE}
1649: to point to the directories where the GNU CC executables
1650: (@samp{gcc-cpp}, @samp{gcc-cc1}, etc.) and the C include files are
1651: kept. This should be done with the commands:@refill
1652:
1653: @example
1654: $ assign /super /system disk:[gcc] gnu_cc
1655: $ assign /super /system disk:[gcc.include] gnu_cc_include
1656: @end example
1657:
1658: @noindent
1659: with the appropriate disk and directory names. These commands can be
1660: placed in your system startup file so they will be executed whenever
1661: the machine is rebooted.
1662:
1663: @item
1664: Install the @samp{GCC} command with the command line:
1665:
1666: @example
1667: $ set command /table=sys$library:dcltables gnu_cc:gcc
1668: @end example
1669:
1670: @noindent
1671: Now you can invoke the compiler with a command like @samp{gcc /verbose
1672: file.c}, which is equivalent to the command @samp{gcc -v -c file.c} in
1673: Unix.
1674: @end enumerate
1675:
1.1.1.5 root 1676: There is a known problem on VMS: @code{const} global variables don't
1677: work compatibly with the VMS C compiler; we don't know a way to get
1678: them to the linker properly.
1679:
1.1 root 1680: @node Trouble, Incompatibilities, Installation, Top
1681: @chapter Known Causes of Trouble with GNU CC.
1682:
1683: Here are some of the things that have caused trouble for people installing
1684: or using GNU CC.
1685:
1686: @itemize @bullet
1687: @item
1688: On certain systems, defining certain environment variables such as
1689: @samp{CC} can interfere with the functioning of @code{make}.
1690:
1691: @item
1692: Cross compilation can run into trouble for certain machines because
1693: some target machines' assemblers require floating point numbers to be
1694: written as @emph{integer} constants in certain contexts.
1695:
1696: The compiler writes these integer constants by examining the floating
1697: point value as an integer and printing that integer, because this is
1698: simple to write and independent of the details of the floating point
1699: representation. But this does not work if the compiler is running on
1700: a different machine with an incompatible floating point format, or
1701: even a different byte-ordering.
1702:
1.1.1.5 root 1703: In addition, correct constant folding of floating point values
1704: requires representing them in the target machine's format.
1705: (The C standard does not quite require this, but in practice
1706: it is the only way to win.)
1707:
1708: It is now possible to overcome these problems by defining macros such
1709: as @code{REAL_VALUE_TYPE}. But doing so is a substantial amount of
1710: work for each target machine. @xref{Cross-compilation}.
1.1 root 1711:
1712: @item
1713: DBX rejects some files produced by GNU CC, though it accepts similar
1714: constructs in output from PCC. Until someone can supply a coherent
1715: description of what is valid DBX input and what is not, there is
1716: nothing I can do about these problems. You are on your own.
1.1.1.2 root 1717:
1718: @item
1719: Users often think it is a bug when GNU CC reports an error for code
1720: like this:
1721:
1722: @example
1723: int foo (short);
1724:
1725: int foo (x)
1726: short x;
1727: @{@dots{}@}
1728: @end example
1729:
1.1.1.4 root 1730: The error message is correct: this code really is erroneous, because the
1731: old-style non-prototype definition passes subword integers in their
1732: promoted types. In other words, the argument is really an @code{int},
1733: not a @code{short}. The correct prototype is this:
1.1.1.2 root 1734:
1735: @example
1736: int foo (int);
1737: @end example
1738:
1739: @item
1740: Users often think it is a bug when GNU CC reports an error for code
1741: like this:
1742:
1743: @example
1744: int foo (struct mumble *);
1745:
1746: struct mumble @{ @dots{} @};
1747:
1748: int foo (struct mumble *x)
1749: @{ @dots{} @}
1750: @end example
1751:
1752: This code really is erroneous, because the scope of @code{struct
1753: mumble} the prototype is limited to the argument list containing it.
1754: It does not refer to the @code{struct mumble} defined with file scope
1755: immediately below---they are two unrelated types with similar names in
1756: different scopes.
1757:
1758: But in the definition of @code{foo}, the file-scope type is used
1759: because that is available to be inherited. Thus, the definition and
1760: the prototype do not match, and you get an error.
1761:
1762: This behavior may seem silly, but it's what the ANSI standard
1763: specifies. It is easy enough for you to make your code work by moving
1764: the definition of @code{struct mumble} above the prototype. I don't
1765: think it's worth being incompatible for.
1.1 root 1766: @end itemize
1767:
1768: @node Incompatibilities, Extensions, Trouble, Top
1769: @chapter Incompatibilities of GNU CC
1770:
1771: There are several noteworthy incompatibilities between GNU C and most
1772: existing (non-ANSI) versions of C.
1773:
1774: Ultimately our intention is that the @samp{-traditional} option will
1775: eliminate most of these incompatibilities by telling GNU C to behave
1776: like the other C compilers.
1777:
1778: @itemize @bullet
1779: @item
1780: GNU CC normally makes string constants read-only. If several
1781: identical-looking string constants are used, GNU CC stores only one
1782: copy of the string.
1783:
1784: One consequence is that you cannot call @code{mktemp} with a string
1785: constant argument. The function @code{mktemp} always alters the
1786: string its argument points to.
1787:
1788: Another consequence is that @code{sscanf} does not work on some
1789: systems when passed a string constant as its format control string.
1790: This is because @code{sscanf} incorrectly tries to write into the
1.1.1.4 root 1791: string constant. Likewise @code{fscanf} and @code{scanf}.
1.1 root 1792:
1793: The best solution to these problems is to change the program to use
1794: @code{char}-array variables with initialization strings for these
1795: purposes instead of string constants. But if this is not possible,
1796: you can use the @samp{-fwritable-strings} flag, which directs GNU CC
1797: to handle string constants the same way most C compilers do.
1798:
1799: @item
1800: GNU CC does not substitute macro arguments when they appear inside of
1801: string constants. For example, the following macro in GNU CC
1802:
1803: @example
1804: #define foo(a) "a"
1805: @end example
1806:
1807: @noindent
1808: will produce output @samp{"a"} regardless of what the argument @var{a} is.
1809:
1810: The @samp{-traditional} option directs GNU CC to handle such cases
1811: (among others) in the old-fashioned (non-ANSI) fashion.
1812:
1813: @item
1814: When you use @code{setjmp} and @code{longjmp}, the only automatic
1815: variables guaranteed to remain valid are those declared
1816: @code{volatile}. This is a consequence of automatic register
1817: allocation. Consider this function:
1818:
1819: @example
1820: jmp_buf j;
1821:
1822: foo ()
1823: @{
1824: int a, b;
1825:
1826: a = fun1 ();
1827: if (setjmp (j))
1828: return a;
1829:
1830: a = fun2 ();
1831: /* @r{@code{longjmp (j)} may be occur in @code{fun3}.} */
1832: return a + fun3 ();
1833: @}
1834: @end example
1835:
1836: Here @code{a} may or may not be restored to its first value when the
1837: @code{longjmp} occurs. If @code{a} is allocated in a register, then
1838: its first value is restored; otherwise, it keeps the last value stored
1839: in it.
1840:
1841: If you use the @samp{-W} option with the @samp{-O} option, you will
1842: get a warning when GNU CC thinks such a problem might be possible.
1843:
1.1.1.2 root 1844: The @samp{-traditional} option directs GNU C to put variables in
1845: the stack by default, rather than in registers, in functions that
1846: call @code{setjmp}. This results in the behavior found in
1847: traditional C compilers.
1848:
1.1 root 1849: @item
1850: Declarations of external variables and functions within a block apply
1851: only to the block containing the declaration. In other words, they
1852: have the same scope as any other declaration in the same place.
1853:
1854: In some other C compilers, a @code{extern} declaration affects all the
1855: rest of the file even if it happens within a block.
1856:
1857: The @samp{-traditional} option directs GNU C to treat all @code{extern}
1858: declarations as global, like traditional compilers.
1859:
1860: @item
1861: In traditional C, you can combine @code{long}, etc., with a typedef name,
1862: as shown here:
1863:
1864: @example
1865: typedef int foo;
1866: typedef long foo bar;
1867: @end example
1868:
1869: In ANSI C, this is not allowed: @code{long} and other type modifiers
1870: require an explicit @code{int}. Because this criterion is expressed
1871: by Bison grammar rules rather than C code, the @samp{-traditional}
1872: flag cannot alter it.
1873:
1874: @item
1875: PCC allows typedef names to be used as function parameters. The
1876: difficulty described immediately above applies here too.
1877:
1878: @item
1879: PCC allows whitespace in the middle of compound assignment operators
1880: such as @samp{+=}. GNU CC, following the ANSI standard, does not
1881: allow this. The difficulty described immediately above applies here
1882: too.
1883:
1884: @item
1885: GNU CC will flag unterminated character constants inside of preprocessor
1886: conditionals that fail. Some programs have English comments enclosed in
1887: conditionals that are guaranteed to fail; if these comments contain
1888: apostrophes, GNU CC will probably report an error. For example,
1889: this code would produce an error:
1890:
1891: @example
1892: #if 0
1893: You can't expect this to work.
1894: #endif
1895: @end example
1896:
1897: The best solution to such a problem is to put the text into an actual
1898: C comment delimited by @samp{/*@dots{}*/}. However,
1899: @samp{-traditional} suppresses these error messages.
1900:
1901: @item
1902: When compiling functions that return @code{float}, PCC converts it to
1903: a double. GNU CC actually returns a @code{float}. If you are concerned
1904: with PCC compatibility, you should declare your functions to return
1905: @code{double}; you might as well say what you mean.
1906:
1907: @item
1908: When compiling functions that return structures or unions, GNU CC
1.1.1.6 ! root 1909: output code normally uses a method different from that used on most
! 1910: versions of Unix. As a result, code compiled with GNU CC cannot call
! 1911: a structure-returning function compiled with PCC, and vice versa.
1.1 root 1912:
1.1.1.6 ! root 1913: The method used by GNU CC is as follows: a structure or union which is 1,
1.1 root 1914: 2, 4 or 8 bytes long is returned like a scalar. A structure or union
1915: with any other size is stored into an address supplied by the caller
1916: in a special, fixed register.
1917:
1918: PCC usually handles all sizes of structures and unions by returning
1919: the address of a block of static storage containing the value. This
1.1.1.6 ! root 1920: method is not used in GNU CC because it is slower and nonreentrant.
1.1.1.5 root 1921:
1.1.1.6 ! root 1922: You can tell GNU CC to use the PCC convention with the option
! 1923: @samp{-fpcc-struct-return}.
1.1 root 1924: @end itemize
1925:
1926: @node Extensions, Bugs, Incompatibilities, Top
1927: @chapter GNU Extensions to the C Language
1928:
1929: GNU C provides several language features not found in ANSI standard C.
1930: (The @samp{-pedantic} option directs GNU CC to print a warning message if
1931: any of these features is used.) To test for the availability of these
1932: features in conditional compilation, check for a predefined macro
1933: @code{__GNUC__}, which is always defined under GNU CC.
1934:
1935: @menu
1936: * Statement Exprs:: Putting statements and declarations inside expressions.
1937: * Naming Types:: Giving a name to the type of some expression.
1938: * Typeof:: @code{typeof}: referring to the type of an expression.
1939: * Lvalues:: Using @samp{?:}, @samp{,} and casts in lvalues.
1940: * Conditionals:: Omitting the middle operand of a @samp{?:} expression.
1941: * Zero-Length:: Zero-length arrays.
1942: * Variable-Length:: Arrays whose length is computed at run time.
1943: * Subscripting:: Any array can be subscripted, even if not an lvalue.
1944: * Pointer Arith:: Arithmetic on @code{void}-pointers and function pointers.
1.1.1.5 root 1945: * Initializers:: Non-constant initializers.
1.1 root 1946: * Constructors:: Constructor expressions give structures, unions
1947: or arrays as values.
1.1.1.5 root 1948: * Function Attributes:: Declaring that functions have no side effects,
1949: or that they can never return.
1.1 root 1950: * Dollar Signs:: Dollar sign is allowed in identifiers.
1951: * Alignment:: Inquiring about the alignment of a type or variable.
1952: * Inline:: Defining inline functions (as fast as macros).
1953: * Extended Asm:: Assembler instructions with C expressions as operands.
1954: (With them you can define ``built-in'' functions.)
1955: * Asm Labels:: Specifying the assembler name to use for a C symbol.
1.1.1.5 root 1956: * Global Reg Vars:: Defining global variables which reside in registers.
1.1 root 1957: @end menu
1958:
1959: @node Statement Exprs, Naming Types, Extensions, Extensions
1960: @section Statements and Declarations inside of Expressions
1961:
1962: A compound statement in parentheses may appear inside an expression in GNU
1963: C. This allows you to declare variables within an expression. For
1964: example:
1965:
1966: @example
1967: (@{ int y = foo (); int z;
1968: if (y > 0) z = y;
1969: else z = - y;
1970: z; @})
1971: @end example
1972:
1973: @noindent
1974: is a valid (though slightly more complex than necessary) expression
1975: for the absolute value of @code{foo ()}.
1976:
1977: This feature is especially useful in making macro definitions ``safe'' (so
1978: that they evaluate each operand exactly once). For example, the
1979: ``maximum'' function is commonly defined as a macro in standard C as
1980: follows:
1981:
1982: @example
1983: #define max(a,b) ((a) > (b) ? (a) : (b))
1984: @end example
1985:
1986: @noindent
1987: But this definition computes either @var{a} or @var{b} twice, with bad
1988: results if the operand has side effects. In GNU C, if you know the
1989: type of the operands (here let's assume @code{int}), you can define
1990: the macro safely as follows:
1991:
1992: @example
1993: #define maxint(a,b) \
1994: (@{int _a = (a), _b = (b); _a > _b ? _a : _b; @})
1995: @end example
1996:
1997: Embedded statements are not allowed in constant expressions, such as
1998: the value of an enumeration constant, the width of a bit field, or
1999: the initial value of a static variable.
2000:
2001: If you don't know the type of the operand, you can still do this, but you
2002: must use @code{typeof} (@pxref{Typeof}) or type naming (@pxref{Naming
2003: Types}).
2004:
2005: @node Naming Types, Typeof, Statement Exprs, Extensions
2006: @section Naming an Expression's Type
2007:
2008: You can give a name to the type of an expression using a @code{typedef}
2009: declaration with an initializer. Here is how to define @var{name} as a
2010: type name for the type of @var{exp}:
2011:
2012: @example
2013: typedef @var{name} = @var{exp};
2014: @end example
2015:
2016: This is useful in conjunction with the statements-within-expressions
2017: feature. Here is how the two together can be used to define a safe
2018: ``maximum'' macro that operates on any arithmetic type:
2019:
2020: @example
2021: #define max(a,b) \
2022: (@{typedef _ta = (a), _tb = (b); \
2023: _ta _a = (a); _tb _b = (b); \
2024: _a > _b ? _a : _b; @})
2025: @end example
2026:
2027: The reason for using names that start with underscores for the local
2028: variables is to avoid conflicts with variable names that occur within the
2029: expressions that are substituted for @code{a} and @code{b}. Eventually we
2030: hope to design a new form of declaration syntax that allows you to declare
2031: variables whose scopes start only after their initializers; this will be a
2032: more reliable way to prevent such conflicts.
2033:
2034: @node Typeof, Lvalues, Naming Types, Extensions
2035: @section Referring to a Type with @code{typeof}
2036:
2037: Another way to refer to the type of an expression is with @code{typeof}.
2038: The syntax of using of this keyword looks like @code{sizeof}, but the
2039: construct acts semantically like a type name defined with @code{typedef}.
2040:
2041: There are two ways of writing the argument to @code{typeof}: with an
2042: expression or with a type. Here is an example with an expression:
2043:
2044: @example
2045: typeof (x[0](1))
2046: @end example
2047:
2048: @noindent
2049: This assumes that @code{x} is an array of functions; the type described
2050: is that of the values of the functions.
2051:
2052: Here is an example with a typename as the argument:
2053:
2054: @example
2055: typeof (int *)
2056: @end example
2057:
2058: @noindent
2059: Here the type described is that of pointers to @code{int}.
2060:
2061: A @code{typeof}-construct can be used anywhere a typedef name could be
2062: used. For example, you can use it in a declaration, in a cast, or inside
2063: of @code{sizeof} or @code{typeof}.
2064:
2065: @itemize @bullet
2066: @item
2067: This declares @code{y} with the type of what @code{x} points to.
2068:
2069: @example
2070: typeof (*x) y;
2071: @end example
2072:
2073: @item
2074: This declares @code{y} as an array of such values.
2075:
2076: @example
2077: typeof (*x) y[4];
2078: @end example
2079:
2080: @item
2081: This declares @code{y} as an array of pointers to characters:
2082:
2083: @example
2084: typeof (typeof (char *)[4]) y;
2085: @end example
2086:
2087: @noindent
2088: It is equivalent to the following traditional C declaration:
2089:
2090: @example
2091: char *y[4];
2092: @end example
2093:
2094: To see the meaning of the declaration using @code{typeof}, and why it
2095: might be a useful way to write, let's rewrite it with these macros:
2096:
2097: @example
2098: #define pointer(T) typeof(T *)
2099: #define array(T, N) typeof(T [N])
2100: @end example
2101:
2102: @noindent
2103: Now the declaration can be rewritten this way:
2104:
2105: @example
2106: array (pointer (char), 4) y;
2107: @end example
2108:
2109: @noindent
2110: Thus, @samp{array (pointer (char), 4)} is the type of arrays of 4
2111: pointers to @code{char}.
2112: @end itemize
2113:
2114: @node Lvalues, Conditionals, Typeof, Extensions
2115: @section Generalized Lvalues
2116:
2117: Compound expressions, conditional expressions and casts are allowed as
2118: lvalues provided their operands are lvalues. This means that you can take
2119: their addresses or store values into them.
2120:
2121: For example, a compound expression can be assigned, provided the last
2122: expression in the sequence is an lvalue. These two expressions are
2123: equivalent:
2124:
2125: @example
2126: (a, b) += 5
2127: a, (b += 5)
2128: @end example
2129:
2130: Similarly, the address of the compound expression can be taken. These two
2131: expressions are equivalent:
2132:
2133: @example
2134: &(a, b)
2135: a, &b
2136: @end example
2137:
2138: A conditional expression is a valid lvalue if its type is not void and the
2139: true and false branches are both valid lvalues. For example, these two
2140: expressions are equivalent:
2141:
2142: @example
2143: (a ? b : c) = 5
2144: (a ? b = 5 : (c = 5))
2145: @end example
2146:
2147: A cast is a valid lvalue if its operand is valid. Taking the address of
2148: the cast is the same as taking the address without a cast, except for the
2149: type of the result. For example, these two expressions are equivalent (but
2150: the second may be valid when the type of @samp{a} does not permit a cast to
2151: @samp{int *}).
2152:
2153: @example
2154: &(int *)a
2155: (int **)&a
2156: @end example
2157:
2158: A simple assignment whose left-hand side is a cast works by converting the
2159: right-hand side first to the specified type, then to the type of the inner
2160: left-hand side expression. After this is stored, the value is converter
2161: back to the specified type to become the value of the assignment. Thus, if
2162: @samp{a} has type @samp{char *}, the following two expressions are
2163: equivalent:
2164:
2165: @example
2166: (int)a = 5
2167: (int)(a = (char *)5)
2168: @end example
2169:
2170: An assignment-with-arithmetic operation such as @samp{+=} applied to a cast
2171: performs the arithmetic using the type resulting from the cast, and then
2172: continues as in the previous case. Therefore, these two expressions are
2173: equivalent:
2174:
2175: @example
2176: (int)a += 5
2177: (int)(a = (char *) ((int)a + 5))
2178: @end example
2179:
2180: @node Conditionals, Zero-Length, Lvalues, Extensions
2181: @section Conditional Expressions with Omitted Middle-Operands
2182:
2183: The middle operand in a conditional expression may be omitted. Then
2184: if the first operand is nonzero, its value is the value of the conditional
2185: expression.
2186:
2187: Therefore, the expression
2188:
2189: @example
2190: x ? : y
2191: @end example
2192:
2193: @noindent
2194: has the value of @code{x} if that is nonzero; otherwise, the value of
2195: @code{y}.
2196:
2197: This example is perfectly equivalent to
2198:
2199: @example
2200: x ? x : y
2201: @end example
2202:
2203: @noindent
2204: In this simple case, the ability to omit the middle operand is not
2205: especially useful. When it becomes useful is when the first operand does,
2206: or may (if it is a macro argument), contain a side effect. Then repeating
2207: the operand in the middle would perform the side effect twice. Omitting
2208: the middle operand uses the value already computed without the undesirable
2209: effects of recomputing it.
2210:
2211: @node Zero-Length, Variable-Length, Conditionals, Extensions
2212: @section Arrays of Length Zero
2213:
2214: Zero-length arrays are allowed in GNU C. They are very useful as the last
2215: element of a structure which is really a header for a variable-length
2216: object:
2217:
2218: @example
2219: struct line @{
2220: int length;
2221: char contents[0];
2222: @};
2223:
2224: @{
2225: struct line *thisline
2226: = (struct line *) malloc (sizeof (struct line) + this_length);
2227: thisline->length = this_length;
2228: @}
2229: @end example
2230:
2231: In standard C, you would have to give @code{contents} a length of 1, which
2232: means either you waste space or complicate the argument to @code{malloc}.
2233:
2234: @node Variable-Length, Subscripting, Zero-Length, Extensions
2235: @section Arrays of Variable Length
2236:
2237: Variable-length automatic arrays are allowed in GNU C. These arrays are
2238: declared like any other automatic arrays, but with a length that is not a
2239: constant expression. The storage is allocated at that time and
2240: deallocated when the brace-level is exited. For example:
2241:
2242: @example
2243: FILE *concat_fopen (char *s1, char *s2, char *mode)
2244: @{
2245: char str[strlen (s1) + strlen (s2) + 1];
2246: strcpy (str, s1);
2247: strcat (str, s2);
2248: return fopen (str, mode);
2249: @}
2250: @end example
2251:
2252: You can also define structure types containing variable-length arrays, and
2253: use them even for arguments or function values, as shown here:
2254:
2255: @example
2256: int foo;
2257:
2258: struct entry
2259: @{
2260: char data[foo];
2261: @};
2262:
2263: struct entry
2264: tester (struct entry arg)
2265: @{
2266: struct entry new;
2267: int i;
2268: for (i = 0; i < foo; i++)
2269: new.data[i] = arg.data[i] + 1;
2270: return new;
2271: @}
2272: @end example
2273:
2274: @noindent
2275: (Eventually there will be a way to say that the size of the array is
2276: another member of the same structure.)
2277:
2278: The length of an array is computed on entry to the brace-level where the
2279: array is declared and is remembered for the scope of the array in case you
2280: access it with @code{sizeof}.
2281:
2282: Jumping or breaking out of the scope of the array name will also deallocate
2283: the storage. Jumping into the scope is not allowed; you will get an error
2284: message for it.
2285:
2286: You can use the function @code{alloca} to get an effect much like
2287: variable-length arrays. The function @code{alloca} is available in
2288: many other C implementations (but not in all). On the other hand,
2289: variable-length arrays are more elegant.
2290:
2291: There are other differences between these two methods. Space allocated
2292: with @code{alloca} exists until the containing @emph{function} returns.
2293: The space for a variable-length array is deallocated as soon as the array
2294: name's scope ends. (If you use both variable-length arrays and
2295: @code{alloca} in the same function, deallocation of a variable-length array
2296: will also deallocate anything more recently allocated with @code{alloca}.)
2297:
2298: @node Subscripting, Pointer Arith, Variable-Length, Extensions
2299: @section Non-Lvalue Arrays May Have Subscripts
2300:
2301: Subscripting is allowed on arrays that are not lvalues, even though the
2302: unary @samp{&} operator is not. For example, this is valid in GNU C though
2303: not valid in other C dialects:
2304:
2305: @example
2306: struct foo @{int a[4];@};
2307:
2308: struct foo f();
2309:
2310: bar (int index)
2311: @{
2312: return f().a[index];
2313: @}
2314: @end example
2315:
2316: @node Pointer Arith, Initializers, Subscripting, Extensions
2317: @section Arithmetic on @code{void}-Pointers and Function Pointers
2318:
2319: In GNU C, addition and subtraction operations are supported on pointers to
2320: @code{void} and on pointers to functions. This is done by treating the
2321: size of a @code{void} or of a function as 1.
2322:
2323: A consequence of this is that @code{sizeof} is also allowed on @code{void}
2324: and on function types, and returns 1.
2325:
2326: @node Initializers, Constructors, Pointer Arith, Extensions
2327: @section Non-Constant Initializers
2328:
2329: The elements of an aggregate initializer are not required to be constant
2330: expressions in GNU C. Here is an example of an initializer with run-time
2331: varying elements:
2332:
2333: @example
2334: foo (float f, float g)
2335: @{
2336: float beat_freqs[2] = @{ f-g, f+g @};
2337: @dots{}
2338: @}
2339: @end example
2340:
1.1.1.5 root 2341: @node Constructors, Function Attributes, Initializers, Extensions
1.1 root 2342: @section Constructor Expressions
2343:
2344: GNU C supports constructor expressions. A constructor looks like a cast
2345: containing an initializer. Its value is an object of the type specified in
2346: the cast, containing the elements specified in the initializer. The type
2347: must be a structure, union or array type.
2348:
2349: Assume that @code{struct foo} and @code{structure} are declared as shown:
2350:
2351: @example
2352: struct foo @{int a; char b[2];@} structure;
2353: @end example
2354:
2355: @noindent
2356: Here is an example of constructing a @samp{struct foo} with a constructor:
2357:
2358: @example
2359: structure = ((struct foo) @{x + y, 'a', 0@});
2360: @end example
2361:
2362: @noindent
2363: This is equivalent to writing the following:
2364:
2365: @example
2366: @{
2367: struct foo temp = @{x + y, 'a', 0@};
2368: structure = temp;
2369: @}
2370: @end example
2371:
2372: You can also construct an array. If all the elements of the constructor
2373: are (made up of) simple constant expressions, suitable for use in
2374: initializers, then the constructor is an lvalue and can be coerced to a
2375: pointer to its first element, as shown here:
2376:
2377: @example
2378: char **foo = (char *[]) @{ "x", "y", "z" @};
2379: @end example
2380:
2381: Array constructors whose elements are not simple constants are not very
2382: useful, because the constructor is not an lvalue. There are only two valid
2383: ways to use it: to subscript it, or initialize an array variable with it.
2384: The former is probably slower than a @code{switch} statement, while the
2385: latter does the same thing an ordinary C initializer would do.
2386:
2387: @example
2388: output = ((int[]) @{ 2, x, 28 @}) [input];
2389: @end example
2390:
1.1.1.5 root 2391: @node Function Attributes, Dollar Signs, Constructors
2392: @section Declaring Attributes of Functions
2393:
2394: In GNU C, you declare certain things about functions called in your program
2395: which help the compiler optimize function calls.
2396:
2397: A few functions, such as @code{abort} and @code{exit}, cannot return.
2398: These functions should be declared @code{volatile}. For example,
2399:
2400: @example
2401: extern volatile void abort ();
2402: @end example
2403:
2404: @noindent
2405: tells the compiler that it can assume that @code{abort} will not return.
2406: This makes slightly better code, but more importantly it helps avoid
2407: spurious warnings of uninitialized variables.
2408:
2409: Many functions do not examine any values except their arguments, and
2410: have no effects except the return value. Such a function can be subject
2411: to common subexpression elimination and loop optimization just as an
2412: arithmetic operator would be. These functions should be declared
2413: @code{const}. For example,
2414:
2415: @example
2416: extern const void square ();
2417: @end example
2418:
2419: @noindent
2420: says that the hypothetical function @code{square} is safe to call
2421: fewer times than the program says.
2422:
2423: Note that a function that has pointer arguments and examines the data
2424: pointed to must @emph{not} be declared @code{const}. Likewise, a
2425: function that calls a non-@code{const} function must not be
2426: @code{const}.
2427:
2428: Some people object to this feature, claiming that ANSI C's @code{#pragma}
2429: should be used instead. There are two reasons I did not do this.
2430:
2431: @enumerate
2432: @item
2433: It is impossible to generate @code{#pragma} commands from a macro.
2434:
2435: @item
2436: The @code{#pragma} command is just as likely as these keywords to mean
2437: something else in another compiler.
2438: @end enumerate
2439:
2440: These two reasons apply to @emph{any} application whatever: as far as
2441: I can see, @code{#pragma} is never useful.
2442:
2443: @node Dollar Signs, Alignment, Function Attributes, Extensions
1.1 root 2444: @section Dollar Signs in Identifier Names
2445:
2446: In GNU C, you may use dollar signs in identifier names. This is because
2447: many traditional C implementations allow such identifiers.
2448:
2449: @node Alignment, Inline, Dollar Signs, Extensions
2450: @section Inquiring about the Alignment of a Type or Variable
2451:
2452: The keyword @code{__alignof} allows you to inquire about how an object
2453: is aligned, or the minimum alignment usually required by a type. Its
2454: syntax is just like @code{sizeof}.
2455:
2456: For example, if the target machine requires a @code{double} value to be
2457: aligned on an 8-byte boundary, then @code{__alignof (double)} is 8. This
2458: is true on many RISC machines. On more traditional machine designs,
2459: @code{__alignof (double)} is 4 or even 2.
2460:
2461: Some machines never actually require alignment; they allow reference to any
2462: data type even at an odd addresses. For these machines, @code{__alignof}
2463: reports the @emph{recommended} alignment of a type.
2464:
2465: When the operand of @code{__alignof} is an lvalue rather than a type, the
2466: value is the largest alignment that the lvalue is known to have. It may
2467: have this alignment as a result of its data type, or because it is part of
2468: a structure and inherits alignment from that structure. For example, after
2469: this declaration:
2470:
2471: @example
2472: struct foo @{ int x; char y; @} foo1;
2473: @end example
2474:
2475: @noindent
2476: the value of @code{__alignof (foo1.y)} is probably 2 or 4, the same as
2477: @code{__alignof (int)}, even though the data type of @code{foo1.y} does not
2478: itself demand any alignment.@refill
2479:
2480: @node Inline, Extended Asm, Alignment, Extensions
2481: @section An Inline Function is As Fast As a Macro
2482:
2483: By declaring a function @code{inline}, you can direct GNU CC to integrate
2484: that function's code into the code for its callers. This makes execution
2485: faster by eliminating the function-call overhead; in addition, if any of
2486: the actual argument values are constant, their known values may permit
2487: simplifications at compile time so that not all of the inline function's
2488: code needs to be included.
2489:
2490: To declare a function inline, use the @code{inline} keyword in its
2491: declaration, like this:
2492:
2493: @example
2494: inline int
2495: inc (int *a)
2496: @{
2497: (*a)++;
2498: @}
2499: @end example
2500:
2501: You can also make all ``simple enough'' functions inline with the
2502: option @samp{-finline-functions}. Note that certain usages in a
2503: function definition can make it unsuitable for inline substitution.
2504:
2505: When a function is both inline and @code{static}, if all calls to the
2506: function are integrated into the caller, then the function's own assembler
2507: code is never referenced. In this case, GNU CC does not actually output
2508: assembler code for the function, unless you specify the option
2509: @samp{-fkeep-inline-functions}. Some calls cannot be integrated for
2510: various reasons (in particular, calls that precede the function's
2511: definition cannot be integrated, and neither can recursive calls within the
2512: definition). If there is a nonintegrated call, then the function is
2513: compiled to assembler code as usual.
2514:
2515: When an inline function is not @code{static}, then the compiler must assume
2516: that there may be calls from other source files; since a global symbol can
2517: be defined only once in any program, the function must not be defined in
2518: the other source files, so the calls therein cannot be integrated.
2519: Therefore, a non-@code{static} inline function is always compiled on its
2520: own in the usual fashion.
2521:
2522: @node Extended Asm, Asm Labels, Inline, Extensions
2523: @section Assembler Instructions with C Expression Operands
2524:
2525: In an assembler instruction using @code{asm}, you can now specify the
2526: operands of the instruction using C expressions. This means no more
2527: guessing which registers or memory locations will contain the data you want
2528: to use.
2529:
2530: You must specify an assembler instruction template much like what appears
2531: in a machine description, plus an operand constraint string for each
2532: operand.
2533:
2534: For example, here is how to use the 68881's @code{fsinx} instruction:
2535:
2536: @example
2537: asm ("fsinx %1,%0" : "=f" (result) : "f" (angle));
2538: @end example
2539:
2540: @noindent
2541: Here @code{angle} is the C expression for the input operand while
2542: @code{result} is that of the output operand. Each has @samp{"f"} as its
2543: operand constraint, saying that a floating-point register is required. The
1.1.1.5 root 2544: @samp{=} in @samp{=f} indicates that the operand is an output; all output
1.1.1.4 root 2545: operands' constraints must use @samp{=}. The constraints use the same
2546: language used in the machine description (@pxref{Constraints}).
1.1 root 2547:
2548: Each operand is described by an operand-constraint string followed by the C
2549: expression in parentheses. A colon separates the assembler template from
2550: the first output operand, and another separates the last output operand
2551: from the first input, if any. Commas separate output operands and separate
1.1.1.4 root 2552: inputs. The total number of operands is limited to the maximum number of
1.1 root 2553: operands in any instruction pattern in the machine description.
2554:
1.1.1.4 root 2555: If there are no output operands, and there are input operands, then there
2556: must be two consecutive colons surrounding the place where the output
2557: operands would go.
2558:
1.1 root 2559: Output operand expressions must be lvalues; the compiler can check this.
2560: The input operands need not be lvalues. The compiler cannot check whether
2561: the operands have data types that are reasonable for the instruction being
2562: executed. It does not parse the assembler instruction template and does
2563: not know what it means, or whether it is valid assembler input. The
2564: extended @code{asm} feature is most often used for machine instructions
2565: that the compiler itself does not know exist.
2566:
2567: The output operands must be write-only; GNU CC will assume that the values
2568: in these operands before the instruction are dead and need not be
2569: generated. For an operand that is read-write, or in which not all bits are
2570: written and the other bits contain useful information, you must logically
2571: split its function into two separate operands, one input operand and one
2572: write-only output operand. The connection between them is expressed by
2573: constraints which say they need to be in the same location when the
2574: instruction executes. You can use the same C expression for both operands,
2575: or different expressions. For example, here we write the (fictitious)
2576: @samp{combine} instruction with @code{bar} as its read-only source operand
2577: and @code{foo} as its read-write destination:
2578:
2579: @example
2580: asm ("combine %2,%0" : "=r" (foo) : "0" (foo), "g" (bar));
2581: @end example
2582:
2583: @noindent
2584: The constraint @samp{"0"} for operand 1 says that it must occupy the same
1.1.1.5 root 2585: location as operand 0. A digit in constraint is allowed only in an input
2586: operand, and it must refer to an output operand.
1.1 root 2587:
2588: Only a digit in the constraint can guarantee that one operand will be in
2589: the same place as another. The mere fact that @code{foo} is the value of
2590: both operands is not enough to guarantee that they will be in the same
2591: place in the generated assembler code. The following would not work:
2592:
2593: @example
2594: asm ("combine %2,%0" : "=r" (foo) : "r" (foo), "g" (bar));
2595: @end example
2596:
2597: Various optimizations or reloading could cause operands 0 and 1 to be in
2598: different registers; GNU CC knows no reason not to do so. For example, the
2599: compiler might find a copy of the value of @code{foo} in one register and
2600: use it for operand 1, but generate the output operand 0 in a different
2601: register (copying it afterward to @code{foo}'s own address). Of course,
2602: since the register for operand 1 is not even mentioned in the assembler
2603: code, the result will not work, but GNU CC can't tell that.
2604:
2605: Unless an output operand has the @samp{&} constraint modifier, GNU CC may
2606: allocate it in the same register as an unrelated input operand, on the
2607: assumption that the inputs are consumed before the outputs are produced.
2608: This assumption may be false if the assembler code actually consists of
2609: more than one instruction. In such a case, use @samp{&} for each output
2610: operand that may not overlap an input. @xref{Modifiers}.
2611:
1.1.1.4 root 2612: Some instructions clobber specific hard registers. To describe this, write
2613: a third colon after the input operands, followed by the names of the
2614: clobbered hard registers (given as strings). Here is a realistic example
2615: for the vax:
1.1 root 2616:
2617: @example
2618: asm volatile ("movc3 %0,%1,%2"
2619: : /* no outputs */
2620: : "g" (from), "g" (to), "g" (count)
2621: : "r0", "r1", "r2", "r3", "r4", "r5");
2622: @end example
2623:
1.1.1.4 root 2624: You can put multiple assembler instructions together in a single @code{asm}
2625: template, separated with semicolons. The input operands are guaranteed not
2626: to use any of the clobbered registers, and neither will the output
2627: operands' addresses, so you can read and write the clobbered registers as
2628: many times as you like. Here is an example of multiple instructions in a
2629: template; it assumes that the subroutine @code{_foo} accepts arguments in
2630: registers 9 and 10:
2631:
2632: @example
2633: asm ("movl %0,r9;movl %1,r10;call _foo"
2634: : /* no outputs */
2635: : "g" (from), "g" (to)
2636: : "r9", "r10");
2637: @end example
2638:
1.1 root 2639: Usually the most convenient way to use these @code{asm} instructions is to
2640: encapsulate them in macros that look like functions. For example,
2641:
2642: @example
2643: #define sin(x) \
2644: (@{ double __value, __arg = (x); \
2645: asm ("fsinx %1,%0": "=f" (__value): "f" (__arg)); \
2646: __value; @})
2647: @end example
2648:
2649: @noindent
2650: Here the variable @code{__arg} is used to make sure that the instruction
2651: operates on a proper @code{double} value, and to accept only those
2652: arguments @code{x} which can convert automatically to a @code{double}.
2653:
2654: Another way to make sure the instruction operates on the correct data type
2655: is to use a cast in the @code{asm}. This is different from using a
2656: variable @code{__arg} in that it converts more different types. For
2657: example, if the desired type were @code{int}, casting the argument to
2658: @code{int} would accept a pointer with no complaint, while assigning the
2659: argument to an @code{int} variable named @code{__arg} would warn about
2660: using a pointer unless the caller explicitly casts it.
2661:
1.1.1.4 root 2662: If an @code{asm} has output operands, GNU CC assumes for optimization
2663: purposes that the instruction has no side effects except to change the
2664: output operands. This does not mean that instructions with a side effect
2665: cannot be used, but you must be careful, because the compiler may eliminate
2666: them if the output operands aren't used, or move them out of loops, or
2667: replace two with one if they constitute a common subexpression. Also, if
2668: your instruction does have a side effect on a variable that otherwise
2669: appears not to change, the old value of the variable may be reused later if
2670: it happens to be found in a register.
1.1 root 2671:
2672: You can prevent an @code{asm} instruction from being deleted, moved or
2673: combined by writing the keyword @code{volatile} after the @code{asm}. For
2674: example:
2675:
2676: @example
2677: #define set_priority(x) \
2678: asm volatile ("set_priority %0": /* no outputs */ : "g" (x))
2679: @end example
2680:
1.1.1.4 root 2681: If there are no output operands, the instruction will not be deleted or
2682: moved.
2683:
1.1 root 2684: It is a natural idea to look for a way to give access to the condition
2685: code left by the assembler instruction. However, when we attempted to
2686: implement this, we found no way to make it work reliably. The problem
2687: is that output operands might need reloading, which would result in
2688: additional following ``store'' instructions. On most machines, these
2689: instructions would alter the condition code before there was time to
2690: test it. This problem doesn't arise for ordinary ``test'' and
2691: ``compare'' instructions because they don't have any output operands.
2692:
1.1.1.5 root 2693: @node Asm Labels, Global Reg Vars, Extended Asm, Extensions
1.1 root 2694: @section Controlling Names Used in Assembler Code
2695:
2696: You can specify the name to be used in the assembler code for a C function
2697: or variable by writing the @code{asm} keyword after the declarator as
2698: follows:
2699:
2700: @example
2701: int foo asm ("myfoo") = 2;
2702: @end example
2703:
2704: @noindent
2705: This specifies that the name to be used for the variable @code{foo} in
2706: the assembler code should be @samp{myfoo} rather than the usual
2707: @samp{_foo}.
2708:
2709: On systems where an underscore is normally prepended to the name of a C
2710: function or variable, this feature allows you to define names for the
2711: linker that do not start with an underscore.
2712:
2713: You cannot use @code{asm} in this way in a function @emph{definition}; but
2714: you can get the same effect by writing a declaration for the function
2715: before its definition and putting @code{asm} there, like this:
2716:
2717: @example
2718: extern func () asm ("FUNC");
2719:
2720: func (x, y)
2721: int x, y;
2722: @dots{}
2723: @end example
2724:
2725: It is up to you to make sure that the assembler names you choose do not
2726: conflict with any other assembler symbols. Also, you must not use a
2727: register name; that would produce completely invalid assembler code. GNU
2728: CC does not as yet have the ability to store static variables in registers.
2729: Perhaps that will be added.
2730:
1.1.1.5 root 2731: @node Global Reg Vars,, Asm Labels, Extensions
2732: @section Global Variables in Registers
2733:
2734: A few programs, such as programming language interpreters, may have a
2735: couple of global variables that are accessed so often that it is worth
2736: while to reserve registers throughout the program just for them.
2737:
2738: You can define a global register variable in GNU C like this:
2739:
2740: @example
2741: register int *foo asm ("a5");
2742: @end example
2743:
2744: @noindent
2745: Here @code{a5} is the name of the register which should be used. Choose a
2746: register which is normally saved and restored by function calls on your
2747: machine, so that library routines will not clobber it.
2748:
2749: Naturally the register name is cpu-dependent, so you would need to
2750: conditionalize your program according to cpu type. The register
2751: @code{a5} would be a good choice on a 68000 for a variable of pointer
2752: type. On machines with register windows, be sure to choose a ``global''
2753: register that is not affected by the function call mechanism.
2754:
2755: In addition, operating systems on one type of cpu may differ in how they
2756: name the registers; then you would need additional conditionals. For
2757: example, some 68000 operating systems call this register @code{%a5}.
2758:
2759: Eventually there may be a way of asking the compiler to choose a register
2760: automatically, but first we need to figure out how it should choose and
1.1.1.6 ! root 2761: how to enable you to guide the choice. No solution is evident.
1.1.1.5 root 2762:
2763: Defining a global register variable in a certain register reserves that
2764: register entirely for this use, at least within the current compilation.
2765: The register will not be allocated for any other purpose in the functions
2766: in the current compilation. The register will not be saved and restored by
2767: these functions. Stores into this register are never deleted even if they
2768: would appear to be dead, but references may be deleted or moved or
2769: simplified.
2770:
2771: It is not safe to access the global register variables from signal
2772: handlers, or from more than one thread of control, because the system
2773: library routines may temporarily use the register for other things (unless
2774: you recompile them specially for the task at hand).
2775:
2776: It is not safe for one function that uses a global register variable to
2777: call another such function @code{foo} by way of a third function
2778: @code{lose} that was compiled without knowledge of this variable (i.e. in a
2779: different source file in which the variable wasn't declared). This is
2780: because @code{lose} might save the register and put some other value there.
2781: For example, you can't expect a global register variable to be available in
2782: the comparison-function that you pass to @code{qsort}, since @code{qsort}
2783: might have put something else in that register. (If you are prepared to
2784: recompile @code{qsort} with the same global register variable, you can
2785: solve this problem.)
2786:
2787: If you want to recompile @code{qsort} or other source files which do not
2788: actually use your global register variable, so that they will not use that
2789: register for any other purpose, then it suffices to specify the compiler
2790: option @samp{-ffixed-@var{reg}}. You need not actually add a global
2791: register declaration to their source code.
2792:
2793: A function which can alter the value of a global register variable cannot
2794: safely be called from a function compiled without this variable, because it
2795: could clobber the value the caller expects to find there on return.
2796: Therefore, the function which is the entry point into the part of the
2797: program that uses the global register variable must explicitly save and
2798: restore the value which belongs to its caller.
2799:
2800: On most machines, @code{longjmp} will restore to each global register
2801: variable the value it had at the time of the @code{setjmp}. On some
2802: machines, however, @code{longjmp} will not change the value of global
2803: register variables. To be portable, the function that called @code{setjmp}
2804: should make other arrangements to save the values of the global register
2805: variables, and to restore them if a @code{longjmp}. This way, the the same
2806: thing will happen regardless of what @code{longjmp} does.
2807:
2808: All global register variable declarations must precede all function
2809: definitions. If such a declaration could appear after function
2810: definitions, the declaration would be too late to prevent the register from
2811: being used for other purposes in the preceding functions.
2812:
1.1.1.6 ! root 2813: Global register variables may not have initial values, because an
! 2814: executable file has no means to supply initial contents for a register.
! 2815:
1.1 root 2816: @node Bugs, Portability, Extensions, Top
2817: @chapter Reporting Bugs
2818:
2819: Your bug reports play an essential role in making GNU CC reliable.
2820:
2821: Reporting a bug may help you by bringing a solution to your problem, or it
2822: may not. But in any case the important function of a bug report is to help
2823: the entire community by making the next version of GNU CC work better. Bug
2824: reports are your contribution to the maintenance of GNU CC.
2825:
2826: In order for a bug report to serve its purpose, you must include the
2827: information that makes for fixing the bug.
2828:
2829: @menu
2830: * Criteria: Bug Criteria. Have you really found a bug?
2831: * Reporting: Bug Reporting. How to report a bug effectively.
2832: @end menu
2833:
2834: @node Bug Criteria, Bug Reporting, Bugs, Bugs
2835: @section Have You Found a Bug?
2836:
2837: If you are not sure whether you have found a bug, here are some guidelines:
2838:
2839: @itemize @bullet
2840: @item
2841: If the compiler gets a fatal signal, for any input whatever, that is a
2842: compiler bug. Reliable compilers never crash.
2843:
2844: @item
2845: If the compiler produces invalid assembly code, for any input whatever
2846: (except an @code{asm} statement), that is a compiler bug, unless the
2847: compiler reports errors (not just warnings) which would ordinarily
2848: prevent the assembler from being run.
2849:
2850: @item
2851: If the compiler produces valid assembly code that does not correctly
2852: execute the input source code, that is a compiler bug.
2853:
2854: However, you must double-check to make sure, because you may have run
2855: into an incompatibility between GNU C and traditional C
2856: (@pxref{Incompatibilities}). These incompatibilities might be considered
2857: bugs, but they are inescapable consequences of valuable features.
2858:
2859: Or you may have a program whose behavior is undefined, which happened
2860: by chance to give the desired results with another C compiler.
2861:
2862: For example, in many nonoptimizing compilers, you can write @samp{x;}
2863: at the end of a function instead of @samp{return x;}, with the same
2864: results. But the value of the function is undefined if @samp{return}
2865: is omitted; it is not a bug when GNU CC produces different results.
2866:
2867: Problems often result from expressions with two increment operators,
2868: as in @samp{f (*p++, *p++)}. Your previous compiler might have
2869: interpreted that expression the way you intended; GNU CC might
2870: interpret it another way; neither compiler is wrong.
2871:
2872: After you have localized the error to a single source line, it should
2873: be easy to check for these things. If your program is correct and
2874: well defined, you have found a compiler bug.
2875:
2876: @item
2877: If the compiler produces an error message for valid input, that is a
2878: compiler bug.
2879:
2880: Note that the following is not valid input, and the error message for
2881: it is not a bug:
2882:
2883: @example
2884: int foo (char);
2885:
2886: int
2887: foo (x)
2888: char x;
2889: @{ @dots{} @}
2890: @end example
2891:
2892: @noindent
2893: The prototype says to pass a @code{char}, while the definition says to
2894: pass an @code{int} and treat the value as a @code{char}. This is what
2895: the ANSI standard says, and it makes sense.
2896:
2897: @item
2898: If the compiler does not produce an error message for invalid input,
2899: that is a compiler bug. However, you should note that your idea of
2900: ``invalid input'' might be my idea of ``an extension'' or ``support
2901: for traditional practice''.
2902:
2903: @item
2904: If you are an experienced user of C compilers, your suggestions
2905: for improvement of GNU CC are welcome in any case.
2906: @end itemize
2907:
2908: @node Bug Reporting,, Bug Criteria, Bugs
2909: @section How to Report Bugs
2910:
2911: Send bug reports for GNU C to one of these addresses:
2912:
2913: @example
2914: bug-gcc@@prep.ai.mit.edu
2915: @{ucbvax|mit-eddie|uunet@}!prep.ai.mit.edu!bug-gcc
2916: @end example
2917:
2918: As a last resort, snail them to:
2919:
2920: @example
2921: GNU Compiler Bugs
2922: 545 Tech Sq
2923: Cambridge, MA 02139
2924: @end example
2925:
2926: The fundamental principle of reporting bugs usefully is this:
2927: @strong{report all the facts}. If you are not sure whether to mention a
2928: fact or leave it out, mention it!
2929:
2930: Often people omit facts because they think they know what causes the
2931: problem and they conclude that some details don't matter. Thus, you might
2932: assume that the name of the variable you use in an example does not matter.
2933: Well, probably it doesn't, but one cannot be sure. Perhaps the bug is a
2934: stray memory reference which happens to fetch from the location where that
2935: name is stored in memory; perhaps, if the name were different, the contents
2936: of that location would fool the compiler into doing the right thing despite
2937: the bug. Play it safe and give an exact example.
2938:
2939: If you want to enable me to fix the bug, you should include all these
2940: things:
2941:
2942: @itemize @bullet
2943: @item
2944: The version of GNU CC. You can get this by running it with the
2945: @samp{-v} option.
2946:
2947: Without this, I won't know whether there is any point in looking for
2948: the bug in the current version of GNU CC.
2949:
2950: @item
2951: A complete input file that will reproduce the bug. If the bug is in
2952: the C preprocessor, send me a source file and any header files that it
2953: requires. If the bug is in the compiler proper (@file{cc1}), run your
2954: source file through the C preprocessor by doing @samp{gcc -E
2955: @var{sourcefile} > @var{outfile}}, then include the contents of
2956: @var{outfile} in the bug report. (Any @samp{-I}, @samp{-D} or
2957: @samp{-U} options that you used in actual compilation should also be
2958: used when doing this.)
2959:
2960: A single statement is not enough of an example. In order to compile
2961: it, it must be embedded in a function definition; and the bug might
2962: depend on the details of how this is done.
2963:
2964: Without a real example I can compile, all I can do about your bug
2965: report is wish you luck. It would be futile to try to guess how to
2966: provoke the bug. For example, bugs in register allocation and
2967: reloading frequently depend on every little detail of the function
2968: they happen in.
2969:
2970: @item
2971: The command arguments you gave GNU CC to compile that example and
2972: observe the bug. For example, did you use @samp{-O}? To guarantee
2973: you won't omit something important, list them all.
2974:
2975: If I were to try to guess the arguments, I would probably guess wrong
2976: and then I would not encounter the bug.
2977:
2978: @item
2979: The names of the files that you used for @file{tm.h} and @file{md}
2980: when you installed the compiler.
2981:
2982: @item
2983: The type of machine you are using, and the operating system name and
2984: version number.
2985:
2986: @item
2987: A description of what behavior you observe that you believe is
2988: incorrect. For example, ``It gets a fatal signal,'' or, ``There is an
2989: incorrect assembler instruction in the output.''
2990:
2991: Of course, if the bug is that the compiler gets a fatal signal, then I
2992: will certainly notice it. But if the bug is incorrect output, I might
2993: not notice unless it is glaringly wrong. I won't study all the
2994: assembler code from a 50-line C program just on the off chance that it
2995: might be wrong.
2996:
2997: Even if the problem you experience is a fatal signal, you should still
2998: say so explicitly. Suppose something strange is going on, such as,
2999: your copy of the compiler is out of synch, or you have encountered a
3000: bug in the C library on your system. (This has happened!) Your copy
3001: might crash and mine would not. If you @i{told} me to expect a crash,
3002: then when mine fails to crash, I would know that the bug was not
3003: happening for me. If you had not told me to expect a crash, then I
3004: would not be able to draw any conclusion from my observations.
3005:
3006: In cases where GNU CC generates incorrect code, if you send me a small
3007: complete sample program I will find the error myself by running the
3008: program under a debugger. If you send me a large example or a part of
3009: a larger program, I cannot do this; you must debug the compiled
3010: program and narrow the problem down to one source line. Tell me which
3011: source line it is, and what you believe is incorrect about the code
3012: generated for that line.
3013:
3014: @item
3015: If you send me examples of output from GNU CC, please use @samp{-g}
3016: when you make them. The debugging information includes source line
3017: numbers which are essential for correlating the output with the input.
3018:
3019: @item
3020: If you wish to suggest changes to the GNU CC source, send me context
3021: diffs. If you even discuss something in the GNU CC source, refer to
3022: it by context, not by line number.
3023:
3024: The line numbers in my development sources don't match those in your
3025: sources. Your line numbers would convey no useful information to me.
3026:
3027: @item
3028: Additional information from a debugger might enable me to find
3029: a problem on a machine which I do not have available myself.
3030: However, you need to think when you collect this information if
3031: you want it to have any chance of being useful.
3032:
3033: For example, many people send just a backtrace, but that is never
3034: useful by itself. A simple backtrace with arguments conveys little
3035: about GNU CC because the compiler is largely data-driven; the same
3036: functions are called over and over for different RTL insns, doing
3037: different things depending on the details of the insn.
3038:
3039: Most of the arguments listed in the backtrace are useless because they
3040: are pointers to RTL list structure. The numeric values of the
3041: pointers, which the debugger prints in the backtrace, have no
3042: significance whatever; all that matters is the contents of the objects
3043: they point to (and most of the contents are other such pointers).
3044:
3045: In addition, most compiler passes consist of one or more loops that
3046: scan the RTL insn sequence. The most vital piece of information about
3047: such a loop--which insn it has reached--is usually in a local variable,
3048: not in an argument.
3049:
3050: What you need to provide in addition to a backtrace are the values of
3051: the local variables for several stack frames up. When a local
3052: variable or an argument is an RTX, first print its value and then use
3053: the GDB command @code{pr} to print the RTL expression that it points
3054: to. (If GDB doesn't run on your machine, use your debugger to call
3055: the function @code{debug_rtx} with the RTX as an argument.) In
3056: general, whenever a variable is a pointer, its value is no use
3057: without the data it points to.
3058:
3059: In addition, include a debugging dump from just before the pass
3060: in which the crash happens. Most bugs involve a series of insns,
3061: not just one.
3062: @end itemize
3063:
3064: Here are some things that are not necessary:
3065:
3066: @itemize @bullet
3067: @item
3068: A description of the envelope of the bug.
3069:
3070: Often people who encounter a bug spend a lot of time investigating
3071: which changes to the input file will make the bug go away and which
3072: changes will not affect it.
3073:
3074: This is often time consuming and not very useful, because the way I
3075: will find the bug is by running a single example under the debugger
3076: with breakpoints, not by pure deduction from a series of examples.
3077:
3078: Of course, if you can find a simpler example to report @emph{instead}
3079: of the original one, that is a convenience for me. Errors in the
3080: output will be easier to spot, running under the debugger will take
3081: less time, etc. Most GNU CC bugs involve just one function, so the
3082: most straightforward way to simplify an example is to delete all the
3083: function definitions except the one where the bug occurs. Those
3084: earlier in the file may be replaced by external declarations if the
3085: crucial function depends on them.
3086:
3087: However, simplification is not vital; if you don't want to do this,
3088: report the bug anyway.
3089:
3090: @item
3091: A patch for the bug.
3092:
3093: A patch for the bug does help me if it is a good one. But don't omit
3094: the necessary information, such as the test case, because I might see
3095: problems with your patch and decide to fix the problem another way.
3096:
3097: Sometimes with a program as complicated as GNU CC it is very hard to
3098: construct an example that will make the program follow a certain path
3099: through the code. If you don't send me the example, I won't be able
3100: to construct one, so I won't be able to verify that the bug is fixed.
3101:
3102: @item
3103: A guess about what the bug is or what it depends on.
3104:
3105: Such guesses are usually wrong. Even I can't guess right about such
3106: things without using the debugger to find the facts.
3107: @end itemize
3108:
3109: @node Portability, Interface, Bugs, Top
3110: @chapter GNU CC and Portability
3111:
3112: The main goal of GNU CC was to make a good, fast compiler for machines in
3113: the class that the GNU system aims to run on: 32-bit machines that address
3114: 8-bit bytes and have several general registers. Elegance, theoretical
3115: power and simplicity are only secondary.
3116:
3117: GNU CC gets most of the information about the target machine from a machine
3118: description which gives an algebraic formula for each of the machine's
3119: instructions. This is a very clean way to describe the target. But when
3120: the compiler needs information that is difficult to express in this
3121: fashion, I have not hesitated to define an ad-hoc parameter to the machine
3122: description. The purpose of portability is to reduce the total work needed
3123: on the compiler; it was not of interest for its own sake.
3124:
3125: GNU CC does not contain machine dependent code, but it does contain code
3126: that depends on machine parameters such as endianness (whether the most
3127: significant byte has the highest or lowest address of the bytes in a word)
3128: and the availability of autoincrement addressing. In the RTL-generation
3129: pass, it is often necessary to have multiple strategies for generating code
3130: for a particular kind of syntax tree, strategies that are usable for different
3131: combinations of parameters. Often I have not tried to address all possible
3132: cases, but only the common ones or only the ones that I have encountered.
3133: As a result, a new target may require additional strategies. You will know
3134: if this happens because the compiler will call @code{abort}. Fortunately,
3135: the new strategies can be added in a machine-independent fashion, and will
3136: affect only the target machines that need them.
3137:
3138: @node Interface, Passes, Portability, Top
3139: @chapter Interfacing to GNU CC Output
3140:
3141: GNU CC is normally configured to use the same function calling convention
3142: normally in use on the target system. This is done with the
3143: machine-description macros described (@pxref{Machine Macros}).
3144:
3145: However, returning of structure and union values is done differently on
3146: some target machines. As a result, functions compiled with PCC
3147: returning such types cannot be called from code compiled with GNU CC,
3148: and vice versa. This does not cause trouble often because few Unix
3149: library routines return structures or unions.
3150:
3151: GNU CC code returns structures and unions that are 1, 2, 4 or 8 bytes
3152: long in the same registers used for @code{int} or @code{double} return
3153: values. (GNU CC typically allocates variables of such types in
3154: registers also.) Structures and unions of other sizes are returned by
3155: storing them into an address passed by the caller (usually in a
3156: register). The machine-description macros @code{STRUCT_VALUE} and
3157: @code{STRUCT_INCOMING_VALUE} tell GNU CC where to pass this address.
3158:
3159: By contrast, PCC on most target machines returns structures and unions
3160: of any size by copying the data into an area of static storage, and then
3161: returning the address of that storage as if it were a pointer value.
3162: The caller must copy the data from that memory area to the place where
3163: the value is wanted. This is slower than the method used by GNU CC, and
3164: fails to be reentrant.
3165:
3166: On some target machines, such as RISC machines and the 80386, the
3167: standard system convention is to pass to the subroutine the address of
3168: where to return the value. On these machines, GNU CC has been
3169: configured to be compatible with the standard compiler, when this method
3170: is used. It may not be compatible for structures of 1, 2, 4 or 8 bytes.
3171:
3172: GNU CC uses the system's standard convention for passing arguments. On
3173: some machines, the first few arguments are passed in registers; in
3174: others, all are passed on the stack. It would be possible to use
3175: registers for argument passing on any machine, and this would probably
3176: result in a significant speedup. But the result would be complete
3177: incompatibility with code that follows the standard convention. So this
3178: change is practical only if you are switching to GNU CC as the sole C
3179: compiler for the system. We may implement register argument passing on
3180: certain machines once we have a complete GNU system so that we can
3181: compile the libraries with GNU CC.
3182:
3183: If you use @code{longjmp}, beware of automatic variables. ANSI C says that
3184: automatic variables that are not declared @code{volatile} have undefined
3185: values after a @code{longjmp}. And this is all GNU CC promises to do,
3186: because it is very difficult to restore register variables correctly, and
3187: one of GNU CC's features is that it can put variables in registers without
3188: your asking it to.
3189:
3190: If you want a variable to be unaltered by @code{longjmp}, and you don't
3191: want to write @code{volatile} because old C compilers don't accept it,
3192: just take the address of the variable. If a variable's address is ever
3193: taken, even if just to compute it and ignore it, then the variable cannot
3194: go in a register:
3195:
3196: @example
3197: @{
3198: int careful;
3199: &careful;
3200: @dots{}
3201: @}
3202: @end example
3203:
3204: Code compiled with GNU CC may call certain library routines. Most of
3205: them handle arithmetic for which there are no instructions. This
3206: includes multiply and divide on some machines, and floating point
3207: operations on any machine for which floating point support is disabled
3208: with @samp{-msoft-float}. Some standard parts of the C library, such as
3209: @code{bcopy} or @code{memcpy}, are also called automatically. The usual
3210: function call interface is used for calling the library routines.
3211:
3212: These library routines should be defined in the library @file{gnulib},
3213: which GNU CC automatically searches whenever it links a program. On
3214: machines that have multiply and divide instructions, if hardware
3215: floating point is in use, normally @file{gnulib} is not needed, but it
3216: is searched just in case.
3217:
3218: Each arithmetic function is defined in @file{gnulib.c} to use the
3219: corresponding C arithmetic operator. As long as the file is compiled
3220: with another C compiler, which supports all the C arithmetic operators,
3221: this file will work portably. However, @file{gnulib.c} does not work if
3222: compiled with GNU CC, because each arithmetic function would compile
3223: into a call to itself!
3224:
3225: @node Passes, RTL, Interface, Top
3226: @chapter Passes and Files of the Compiler
3227:
3228: The overall control structure of the compiler is in @file{toplev.c}. This
3229: file is responsible for initialization, decoding arguments, opening and
3230: closing files, and sequencing the passes.
3231:
3232: The parsing pass is invoked only once, to parse the entire input. The RTL
3233: intermediate code for a function is generated as the function is parsed, a
3234: statement at a time. Each statement is read in as a syntax tree and then
3235: converted to RTL; then the storage for the tree for the statement is
3236: reclaimed. Storage for types (and the expressions for their sizes),
3237: declarations, and a representation of the binding contours and how they nest,
3238: remains until the function is finished being compiled; these are all needed
3239: to output the debugging information.
3240:
3241: Each time the parsing pass reads a complete function definition or
3242: top-level declaration, it calls the function
3243: @code{rest_of_compilation} or @code{rest_of_decl_compilation} in
3244: @file{toplev.c}, which are responsible for all further processing
3245: necessary, ending with output of the assembler language. All other
3246: compiler passes run, in sequence, within @code{rest_of_compilation}.
3247: When that function returns from compiling a function definition, the
3248: storage used for that function definition's compilation is entirely
3249: freed, unless it is an inline function (@pxref{Inline}).
3250:
3251: Here is a list of all the passes of the compiler and their source files.
3252: Also included is a description of where debugging dumps can be requested
3253: with @samp{-d} options.
3254:
3255: @itemize @bullet
3256: @item
3257: Parsing. This pass reads the entire text of a function definition,
3258: constructing partial syntax trees. This and RTL generation are no longer
3259: truly separate passes (formerly they were), but it is easier to think
3260: of them as separate.
3261:
3262: The tree representation does not entirely follow C syntax, because it is
3263: intended to support other languages as well.
3264:
3265: C data type analysis is also done in this pass, and every tree node
3266: that represents an expression has a data type attached. Variables are
3267: represented as declaration nodes.
3268:
3269: Constant folding and associative-law simplifications are also done
3270: during this pass.
3271:
3272: The source files for parsing are @file{c-parse.y}, @file{c-decl.c},
3273: @file{c-typeck.c}, @file{c-convert.c}, @file{stor-layout.c},
3274: @file{fold-const.c}, and @file{tree.c}. The last three files are
3275: intended to be language-independent. There are also header files
3276: @file{c-parse.h}, @file{c-tree.h}, @file{tree.h} and @file{tree.def}.
3277: The last two define the format of the tree representation.@refill
3278:
3279: @item
3280: RTL generation. This is the conversion of syntax tree into RTL code.
3281: It is actually done statement-by-statement during parsing, but for
3282: most purposes it can be thought of as a separate pass.
3283:
3284: This is where the bulk of target-parameter-dependent code is found,
3285: since often it is necessary for strategies to apply only when certain
3286: standard kinds of instructions are available. The purpose of named
3287: instruction patterns is to provide this information to the RTL
3288: generation pass.
3289:
3290: Optimization is done in this pass for @code{if}-conditions that are
3291: comparisons, boolean operations or conditional expressions. Tail
3292: recursion is detected at this time also. Decisions are made about how
3293: best to arrange loops and how to output @code{switch} statements.
3294:
3295: The source files for RTL generation are @file{stmt.c}, @file{expr.c},
3296: @file{explow.c}, @file{expmed.c}, @file{optabs.c} and @file{emit-rtl.c}.
3297: Also, the file @file{insn-emit.c}, generated from the machine description
3298: by the program @code{genemit}, is used in this pass. The header files
3299: @file{expr.h} is used for communication within this pass.@refill
3300:
3301: The header files @file{insn-flags.h} and @file{insn-codes.h},
3302: generated from the machine description by the programs @code{genflags}
3303: and @code{gencodes}, tell this pass which standard names are available
3304: for use and which patterns correspond to them.@refill
3305:
3306: Aside from debugging information output, none of the following passes
3307: refers to the tree structure representation of the function (only
3308: part of which is saved).
3309:
3310: The decision of whether the function can and should be expanded inline
3311: in its subsequent callers is made at the end of rtl generation. The
3312: function must meet certain criteria, currently related to the size of
3313: the function and the types and number of parameters it has. Note that
3314: this function may contain loops, recursive calls to itself
3315: (tail-recursive functions can be inlined!), gotos, in short, all
3316: constructs supported by GNU CC.
3317:
3318: The option @samp{-dr} causes a debugging dump of the RTL code after
3319: this pass. This dump file's name is made by appending @samp{.rtl} to
3320: the input file name.
3321:
3322: @item
3323: Jump optimization. This pass simplifies jumps to the following
3324: instruction, jumps across jumps, and jumps to jumps. It deletes
3325: unreferenced labels and unreachable code, except that unreachable code
3326: that contains a loop is not recognized as unreachable in this pass.
3327: (Such loops are deleted later in the basic block analysis.)
3328:
3329: Jump optimization is performed two or three times. The first time is
3330: immediately following RTL generation. The second time is after CSE,
3331: but only if CSE says repeated jump optimization is needed. The
3332: last time is right before the final pass. That time, cross-jumping
3333: and deletion of no-op move instructions are done together with the
3334: optimizations described above.
3335:
3336: The source file of this pass is @file{jump.c}.
3337:
3338: The option @samp{-dj} causes a debugging dump of the RTL code after
3339: this pass is run for the first time. This dump file's name is made by
3340: appending @samp{.jump} to the input file name.
3341:
3342: @item
3343: Register scan. This pass finds the first and last use of each
3344: register, as a guide for common subexpression elimination. Its source
3345: is in @file{regclass.c}.
3346:
3347: @item
3348: Common subexpression elimination. This pass also does constant
3349: propagation. Its source file is @file{cse.c}. If constant
3350: propagation causes conditional jumps to become unconditional or to
3351: become no-ops, jump optimization is run again when CSE is finished.
3352:
3353: The option @samp{-ds} causes a debugging dump of the RTL code after
3354: this pass. This dump file's name is made by appending @samp{.cse} to
3355: the input file name.
3356:
3357: @item
3358: Loop optimization. This pass moves constant expressions out of loops.
3359: Its source file is @file{loop.c}.
3360:
3361: The option @samp{-dL} causes a debugging dump of the RTL code after
3362: this pass. This dump file's name is made by appending @samp{.loop} to
3363: the input file name.
3364:
3365: @item
3366: Stupid register allocation is performed at this point in a
3367: nonoptimizing compilation. It does a little data flow analysis as
3368: well. When stupid register allocation is in use, the next pass
3369: executed is the reloading pass; the others in between are skipped.
3370: The source file is @file{stupid.c}.
3371:
3372: @item
3373: Data flow analysis (@file{flow.c}). This pass divides the program
3374: into basic blocks (and in the process deletes unreachable loops); then
3375: it computes which pseudo-registers are live at each point in the
3376: program, and makes the first instruction that uses a value point at
3377: the instruction that computed the value.
3378:
3379: This pass also deletes computations whose results are never used, and
3380: combines memory references with add or subtract instructions to make
3381: autoincrement or autodecrement addressing.
3382:
3383: The option @samp{-df} causes a debugging dump of the RTL code after
3384: this pass. This dump file's name is made by appending @samp{.flow} to
3385: the input file name. If stupid register allocation is in use, this
3386: dump file reflects the full results of such allocation.
3387:
3388: @item
3389: Instruction combination (@file{combine.c}). This pass attempts to
3390: combine groups of two or three instructions that are related by data
3391: flow into single instructions. It combines the RTL expressions for
3392: the instructions by substitution, simplifies the result using algebra,
3393: and then attempts to match the result against the machine description.
3394:
3395: The option @samp{-dc} causes a debugging dump of the RTL code after
3396: this pass. This dump file's name is made by appending @samp{.combine}
3397: to the input file name.
3398:
3399: @item
3400: Register class preferencing. The RTL code is scanned to find out
3401: which register class is best for each pseudo register. The source
3402: file is @file{regclass.c}.
3403:
3404: @item
3405: Local register allocation (@file{local-alloc.c}). This pass allocates
3406: hard registers to pseudo registers that are used only within one basic
3407: block. Because the basic block is linear, it can use fast and
3408: powerful techniques to do a very good job.
3409:
3410: The option @samp{-dl} causes a debugging dump of the RTL code after
3411: this pass. This dump file's name is made by appending @samp{.lreg} to
3412: the input file name.
3413:
3414: @item
3415: Global register allocation (@file{global-alloc.c}). This pass
3416: allocates hard registers for the remaining pseudo registers (those
3417: whose life spans are not contained in one basic block).
3418:
3419: @item
3420: Reloading. This pass renumbers pseudo registers with the hardware
3421: registers numbers they were allocated. Pseudo registers that did not
3422: get hard registers are replaced with stack slots. Then it finds
3423: instructions that are invalid because a value has failed to end up in
3424: a register, or has ended up in a register of the wrong kind. It fixes
3425: up these instructions by reloading the problematical values
3426: temporarily into registers. Additional instructions are generated to
3427: do the copying.
3428:
3429: Source files are @file{reload.c} and @file{reload1.c}, plus the header
3430: @file{reload.h} used for communication between them.
3431:
3432: The option @samp{-dg} causes a debugging dump of the RTL code after
3433: this pass. This dump file's name is made by appending @samp{.greg} to
3434: the input file name.
3435:
3436: @item
3437: Jump optimization is repeated, this time including cross-jumping
1.1.1.5 root 3438: and deletion of no-op move instructions.
1.1 root 3439:
3440: The option @samp{-dJ} causes a debugging dump of the RTL code after
3441: this pass. This dump file's name is made by appending @samp{.jump2}
3442: to the input file name.
3443:
3444: @item
3445: Final. This pass outputs the assembler code for the function. It is
3446: also responsible for identifying spurious test and compare
1.1.1.5 root 3447: instructions. Machine-specific peephole optimizations are performed
3448: at the same time. The function entry and exit sequences are generated
1.1 root 3449: directly as assembler code in this pass; they never exist as RTL.
3450:
3451: The source files are @file{final.c} plus @file{insn-output.c}; the
3452: latter is generated automatically from the machine description by the
3453: tool @file{genoutput}. The header file @file{conditions.h} is used
3454: for communication between these files.
3455:
3456: @item
3457: Debugging information output. This is run after final because it must
3458: output the stack slot offsets for pseudo registers that did not get
3459: hard registers. Source files are @file{dbxout.c} for DBX symbol table
3460: format and @file{symout.c} for GDB's own symbol table format.
3461: @end itemize
3462:
3463: Some additional files are used by all or many passes:
3464:
3465: @itemize @bullet
3466: @item
3467: Every pass uses @file{machmode.def}, which defines the machine modes.
3468:
3469: @item
3470: All the passes that work with RTL use the header files @file{rtl.h}
3471: and @file{rtl.def}, and subroutines in file @file{rtl.c}. The tools
3472: @code{gen*} also use these files to read and work with the machine
3473: description RTL.
3474:
3475: @item
3476: Several passes refer to the header file @file{insn-config.h} which
3477: contains a few parameters (C macro definitions) generated
3478: automatically from the machine description RTL by the tool
3479: @code{genconfig}.
3480:
3481: @item
3482: Several passes use the instruction recognizer, which consists of
3483: @file{recog.c} and @file{recog.h}, plus the files @file{insn-recog.c}
3484: and @file{insn-extract.c} that are generated automatically from the
3485: machine description by the tools @file{genrecog} and
3486: @file{genextract}.@refill
3487:
3488: @item
3489: Several passes use the header files @file{regs.h} which defines the
3490: information recorded about pseudo register usage, and @file{basic-block.h}
3491: which defines the information recorded about basic blocks.
3492:
3493: @item
3494: @file{hard-reg-set.h} defines the type @code{HARD_REG_SET}, a bit-vector
3495: with a bit for each hard register, and some macros to manipulate it.
3496: This type is just @code{int} if the machine has few enough hard registers;
3497: otherwise it is an array of @code{int} and some of the macros expand
3498: into loops.
3499: @end itemize
3500:
3501: @node RTL, Machine Desc, Passes, Top
3502: @chapter RTL Representation
3503:
3504: Most of the work of the compiler is done on an intermediate representation
3505: called register transfer language. In this language, the instructions to be
3506: output are described, pretty much one by one, in an algebraic form that
3507: describes what the instruction does.
3508:
3509: RTL is inspired by Lisp lists. It has both an internal form, made up of
3510: structures that point at other structures, and a textual form that is used
3511: in the machine description and in printed debugging dumps. The textual
3512: form uses nested parentheses to indicate the pointers in the internal form.
3513:
3514: @menu
3515: * RTL Objects:: Expressions vs vectors vs strings vs integers.
3516: * Accessors:: Macros to access expression operands or vector elts.
3517: * Flags:: Other flags in an RTL expression.
3518: * Machine Modes:: Describing the size and format of a datum.
3519: * Constants:: Expressions with constant values.
3520: * Regs and Memory:: Expressions representing register contents or memory.
3521: * Arithmetic:: Expressions representing arithmetic on other expressions.
3522: * Comparisons:: Expressions representing comparison of expressions.
3523: * Bit Fields:: Expressions representing bit-fields in memory or reg.
3524: * Conversions:: Extending, truncating, floating or fixing.
3525: * RTL Declarations:: Declaring volatility, constancy, etc.
3526: * Side Effects:: Expressions for storing in registers, etc.
3527: * Incdec:: Embedded side-effects for autoincrement addressing.
3528: * Assembler:: Representing @code{asm} with operands.
3529: * Insns:: Expression types for entire insns.
3530: * Calls:: RTL representation of function call insns.
3531: * Sharing:: Some expressions are unique; others *must* be copied.
3532: @end menu
3533:
3534: @node RTL Objects, Accessors, RTL, RTL
3535: @section RTL Object Types
3536:
3537: RTL uses four kinds of objects: expressions, integers, strings and vectors.
3538: Expressions are the most important ones. An RTL expression (``RTX'', for
3539: short) is a C structure, but it is usually referred to with a pointer; a
3540: type that is given the typedef name @code{rtx}.
3541:
3542: An integer is simply an @code{int}, and a string is a @code{char *}.
3543: Within RTL code, strings appear only inside @samp{symbol_ref} expressions,
3544: but they appear in other contexts in the RTL expressions that make up
3545: machine descriptions. Their written form uses decimal digits.
3546:
3547: A string is a sequence of characters. In core it is represented as a
3548: @code{char *} in usual C fashion, and it is written in C syntax as well.
3549: However, strings in RTL may never be null. If you write an empty string in
3550: a machine description, it is represented in core as a null pointer rather
3551: than as a pointer to a null character. In certain contexts, these null
3552: pointers instead of strings are valid.
3553:
3554: A vector contains an arbitrary, specified number of pointers to
3555: expressions. The number of elements in the vector is explicitly present in
3556: the vector. The written form of a vector consists of square brackets
3557: (@samp{[@dots{}]}) surrounding the elements, in sequence and with
3558: whitespace separating them. Vectors of length zero are not created; null
3559: pointers are used instead.
3560:
3561: Expressions are classified by @dfn{expression codes} (also called RTX
3562: codes). The expression code is a name defined in @file{rtl.def}, which is
3563: also (in upper case) a C enumeration constant. The possible expression
3564: codes and their meanings are machine-independent. The code of an RTX can
3565: be extracted with the macro @code{GET_CODE (@var{x})} and altered with
3566: @code{PUT_CODE (@var{x}, @var{newcode})}.
3567:
3568: The expression code determines how many operands the expression contains,
3569: and what kinds of objects they are. In RTL, unlike Lisp, you cannot tell
3570: by looking at an operand what kind of object it is. Instead, you must know
3571: from its context---from the expression code of the containing expression.
3572: For example, in an expression of code @samp{subreg}, the first operand is
3573: to be regarded as an expression and the second operand as an integer. In
3574: an expression of code @samp{plus}, there are two operands, both of which
3575: are to be regarded as expressions. In a @samp{symbol_ref} expression,
3576: there is one operand, which is to be regarded as a string.
3577:
3578: Expressions are written as parentheses containing the name of the
3579: expression type, its flags and machine mode if any, and then the operands
3580: of the expression (separated by spaces).
3581:
3582: Expression code names in the @samp{md} file are written in lower case,
3583: but when they appear in C code they are written in upper case. In this
3584: manual, they are shown as follows: @samp{const_int}.
3585:
3586: In a few contexts a null pointer is valid where an expression is normally
1.1.1.4 root 3587: wanted. The written form of this is @code{(nil)}.
1.1 root 3588:
3589: @node Accessors, Flags, RTL Objects, RTL
3590: @section Access to Operands
3591:
3592: For each expression type @file{rtl.def} specifies the number of contained
3593: objects and their kinds, with four possibilities: @samp{e} for expression
3594: (actually a pointer to an expression), @samp{i} for integer, @samp{s} for
3595: string, and @samp{E} for vector of expressions. The sequence of letters
3596: for an expression code is called its @dfn{format}. Thus, the format of
3597: @samp{subreg} is @samp{ei}.@refill
3598:
3599: Two other format characters are used occasionally: @samp{u} and @samp{0}.
3600: @samp{u} is equivalent to @samp{e} except that it is printed differently in
3601: debugging dumps, and @samp{0} means a slot whose contents do not fit any
3602: normal category. @samp{0} slots are not printed at all in dumps, and are
3603: often used in special ways by small parts of the compiler.@refill
3604:
3605: There are macros to get the number of operands and the format of an
3606: expression code:
3607:
3608: @table @code
3609: @item GET_RTX_LENGTH (@var{code})
3610: Number of operands of an RTX of code @var{code}.
3611:
3612: @item GET_RTX_FORMAT (@var{code})
3613: The format of an RTX of code @var{code}, as a C string.
3614: @end table
3615:
3616: Operands of expressions are accessed using the macros @code{XEXP},
3617: @code{XINT} and @code{XSTR}. Each of these macros takes two arguments: an
3618: expression-pointer (RTX) and an operand number (counting from zero).
3619: Thus,@refill
3620:
3621: @example
3622: XEXP (@var{x}, 2)
3623: @end example
3624:
3625: @noindent
3626: accesses operand 2 of expression @var{x}, as an expression.
3627:
3628: @example
3629: XINT (@var{x}, 2)
3630: @end example
3631:
3632: @noindent
3633: accesses the same operand as an integer. @code{XSTR}, used in the same
3634: fashion, would access it as a string.
3635:
3636: Any operand can be accessed as an integer, as an expression or as a string.
3637: You must choose the correct method of access for the kind of value actually
3638: stored in the operand. You would do this based on the expression code of
3639: the containing expression. That is also how you would know how many
3640: operands there are.
3641:
3642: For example, if @var{x} is a @samp{subreg} expression, you know that it has
3643: two operands which can be correctly accessed as @code{XEXP (@var{x}, 0)}
3644: and @code{XINT (@var{x}, 1)}. If you did @code{XINT (@var{x}, 0)}, you
3645: would get the address of the expression operand but cast as an integer;
3646: that might occasionally be useful, but it would be cleaner to write
3647: @code{(int) XEXP (@var{x}, 0)}. @code{XEXP (@var{x}, 1)} would also
3648: compile without error, and would return the second, integer operand cast as
3649: an expression pointer, which would probably result in a crash when
3650: accessed. Nothing stops you from writing @code{XEXP (@var{x}, 28)} either,
3651: but this will access memory past the end of the expression with
3652: unpredictable results.@refill
3653:
3654: Access to operands which are vectors is more complicated. You can use the
3655: macro @code{XVEC} to get the vector-pointer itself, or the macros
3656: @code{XVECEXP} and @code{XVECLEN} to access the elements and length of a
3657: vector.
3658:
3659: @table @code
3660: @item XVEC (@var{exp}, @var{idx})
3661: Access the vector-pointer which is operand number @var{idx} in @var{exp}.
3662:
3663: @item XVECLEN (@var{exp}, @var{idx})
3664: Access the length (number of elements) in the vector which is
3665: in operand number @var{idx} in @var{exp}. This value is an @code{int}.
3666:
3667: @item XVECEXP (@var{exp}, @var{idx}, @var{eltnum})
3668: Access element number @var{eltnum} in the vector which is
3669: in operand number @var{idx} in @var{exp}. This value is an RTX.
3670:
3671: It is up to you to make sure that @var{eltnum} is not negative
3672: and is less than @code{XVECLEN (@var{exp}, @var{idx})}.
3673: @end table
3674:
3675: All the macros defined in this section expand into lvalues and therefore
3676: can be used to assign the operands, lengths and vector elements as well as
3677: to access them.
3678:
3679: @node Flags, Machine Modes, Accessors, RTL
3680: @section Flags in an RTL Expression
3681:
3682: RTL expressions contain several flags (one-bit bit-fields) that are used
3683: in certain types of expression. Most often they are accessed with the
3684: following macros:
3685:
3686: @table @code
3687: @item MEM_VOLATILE_P (@var{x})
3688: In @samp{mem} expressions, nonzero for volatile memory references.
3689: Stored in the @code{volatil} field and printed as @samp{/v}.
3690:
3691: @item MEM_IN_STRUCT_P (@var{x})
3692: In @samp{mem} expressions, nonzero for reference to an entire
3693: structure, union or array, or to a component of one. Zero for
3694: references to a scalar variable or through a pointer to a scalar.
3695: Stored in the @code{in_struct} field and printed as @samp{/s}.
3696:
3697: @item REG_USER_VAR_P (@var{x})
3698: In a @samp{reg}, nonzero if it corresponds to a variable present in
3699: the user's source code. Zero for temporaries generated internally by
3700: the compiler. Stored in the @code{volatil} field and printed as
3701: @samp{/v}.
3702:
3703: @item REG_FUNCTION_VALUE_P (@var{x})
3704: Nonzero in a @samp{reg} if it is the place in which this function's
3705: value is going to be returned. (This happens only in a hard
3706: register.) Stored in the @code{integrated} field and printed as
3707: @samp{/i}.
3708:
3709: The same hard register may be used also for collecting the values of
3710: functions called by this one, but @code{REG_FUNCTION_VALUE_P} is zero
3711: in this kind of use.
3712:
3713: @item RTX_UNCHANGING_P (@var{x})
3714: Nonzero in a @samp{reg} or @samp{mem} if the value is not changed
3715: explicitly by the current function. (If it is a memory reference then
3716: it may be changed by other functions or by aliasing.) Stored in the
3717: @code{unchanging} field and printed as @samp{/u}.
3718:
3719: @item RTX_INTEGRATED_P (@var{insn})
3720: Nonzero in an insn if it resulted from an in-line function call.
3721: Stored in the @code{integrated} field and printed as @samp{/i}. This
3722: may be deleted; nothing currently depends on it.
3723:
3724: @item INSN_DELETED_P (@var{insn})
3725: In an insn, nonzero if the insn has been deleted. Stored in the
3726: @code{volatil} field and printed as @samp{/v}.
3727:
3728: @item CONSTANT_POOL_ADDRESS_P (@var{x})
3729: Nonzero in a @samp{symbol_ref} if it refers to part of the current
3730: function's ``constants pool''. These are addresses close to the
3731: beginning of the function, and GNU CC assumes they can be addressed
3732: directly (perhaps with the help of base registers). Stored in the
3733: @code{unchanging} field and printed as @samp{/u}.
3734: @end table
3735:
3736: These are the fields which the above macros refer to:
3737:
3738: @table @code
3739: @item used
3740: This flag is used only momentarily, at the end of RTL generation for a
3741: function, to count the number of times an expression appears in insns.
3742: Expressions that appear more than once are copied, according to the
3743: rules for shared structure (@pxref{Sharing}).
3744:
3745: @item volatil
3746: This flag is used in @samp{mem} and @samp{reg} expressions and in insns.
3747: In RTL dump files, it is printed as @samp{/v}.
3748:
3749: In a @samp{mem} expression, it is 1 if the memory reference is volatile.
3750: Volatile memory references may not be deleted, reordered or combined.
3751:
3752: In a @samp{reg} expression, it is 1 if the value is a user-level variable.
3753: 0 indicates an internal compiler temporary.
3754:
3755: In an insn, 1 means the insn has been deleted.
3756:
3757: @item in_struct
3758: This flag is used in @samp{mem} expressions. It is 1 if the memory
3759: datum referred to is all or part of a structure or array; 0 if it is (or
3760: might be) a scalar variable. A reference through a C pointer has 0
3761: because the pointer might point to a scalar variable.
3762:
3763: This information allows the compiler to determine something about possible
3764: cases of aliasing.
3765:
3766: In an RTL dump, this flag is represented as @samp{/s}.
3767:
3768: @item unchanging
3769: This flag is used in @samp{reg} and @samp{mem} expressions. 1 means
3770: that the value of the expression never changes (at least within the
3771: current function).
3772:
3773: In an RTL dump, this flag is represented as @samp{/u}.
3774:
3775: @item integrated
3776: In some kinds of expressions, including insns, this flag means the
3777: rtl was produced by procedure integration.
3778:
3779: In a @samp{reg} expression, this flag indicates the register
3780: containing the value to be returned by the current function. On
3781: machines that pass parameters in registers, the same register number
3782: may be used for parameters as well, but this flag is not set on such
3783: uses.
3784: @end table
3785:
3786: @node Machine Modes, Constants, Flags, RTL
3787: @section Machine Modes
3788:
3789: A machine mode describes a size of data object and the representation used
3790: for it. In the C code, machine modes are represented by an enumeration
3791: type, @code{enum machine_mode}, defined in @file{machmode.def}. Each RTL
3792: expression has room for a machine mode and so do certain kinds of tree
3793: expressions (declarations and types, to be precise).
3794:
3795: In debugging dumps and machine descriptions, the machine mode of an RTL
3796: expression is written after the expression code with a colon to separate
3797: them. The letters @samp{mode} which appear at the end of each machine mode
3798: name are omitted. For example, @code{(reg:SI 38)} is a @samp{reg}
3799: expression with machine mode @code{SImode}. If the mode is
3800: @code{VOIDmode}, it is not written at all.
3801:
3802: Here is a table of machine modes.
3803:
3804: @table @code
3805: @item QImode
3806: ``Quarter-Integer'' mode represents a single byte treated as an integer.
3807:
3808: @item HImode
3809: ``Half-Integer'' mode represents a two-byte integer.
3810:
3811: @item SImode
3812: ``Single Integer'' mode represents a four-byte integer.
3813:
3814: @item DImode
3815: ``Double Integer'' mode represents an eight-byte integer.
3816:
3817: @item TImode
3818: ``Tetra Integer'' (?) mode represents a sixteen-byte integer.
3819:
3820: @item SFmode
3821: ``Single Floating'' mode represents a single-precision (four byte) floating
3822: point number.
3823:
3824: @item DFmode
3825: ``Double Floating'' mode represents a double-precision (eight byte) floating
3826: point number.
3827:
3828: @item TFmode
3829: ``Tetra Floating'' mode represents a quadruple-precision (sixteen byte)
3830: floating point number.
3831:
3832: @item BLKmode
3833: ``Block'' mode represents values that are aggregates to which none of
3834: the other modes apply. In RTL, only memory references can have this mode,
3835: and only if they appear in string-move or vector instructions. On machines
3836: which have no such instructions, @code{BLKmode} will not appear in RTL.
3837:
3838: @item VOIDmode
3839: Void mode means the absence of a mode or an unspecified mode.
3840: For example, RTL expressions of code @samp{const_int} have mode
3841: @code{VOIDmode} because they can be taken to have whatever mode the context
3842: requires. In debugging dumps of RTL, @code{VOIDmode} is expressed by
3843: the absence of any mode.
3844:
3845: @item EPmode
3846: ``Entry Pointer'' mode is intended to be used for function variables in
3847: Pascal and other block structured languages. Such values contain
3848: both a function address and a static chain pointer for access to
3849: automatic variables of outer levels. This mode is only partially
3850: implemented since C does not use it.
3851:
3852: @item CSImode@r{, @dots{}}
3853: ``Complex Single Integer'' mode stands for a complex number represented
3854: as a pair of @code{SImode} integers. Any of the integer and floating modes
3855: may have @samp{C} prefixed to its name to obtain a complex number mode.
3856: For example, there are @code{CQImode}, @code{CSFmode}, and @code{CDFmode}.
3857: Since C does not support complex numbers, these machine modes are only
3858: partially implemented.
3859:
3860: @item BImode
3861: This is the machine mode of a bit-field in a structure. It is used
3862: only in the syntax tree, never in RTL, and in the syntax tree it appears
3863: only in declaration nodes. In C, it appears only in @code{FIELD_DECL}
3864: nodes for structure fields defined with a bit size.
3865: @end table
3866:
3867: The machine description defines @code{Pmode} as a C macro which expands
3868: into the machine mode used for addresses. Normally this is @code{SImode}.
3869:
3870: The only modes which a machine description @i{must} support are
3871: @code{QImode}, @code{SImode}, @code{SFmode} and @code{DFmode}. The
3872: compiler will attempt to use @code{DImode} for two-word structures and
3873: unions, but it would not be hard to program it to avoid this. Likewise,
3874: you can arrange for the C type @code{short int} to avoid using
3875: @code{HImode}. In the long term it would be desirable to make the set of
3876: available machine modes machine-dependent and eliminate all assumptions
3877: about specific machine modes or their uses from the machine-independent
3878: code of the compiler.
3879:
1.1.1.4 root 3880: To help begin this process, the machine modes are divided into mode
3881: classes. These are represented by the enumeration type @code{enum
3882: mode_class} defined in @file{rtl.h}. The possible mode classes are:
3883:
3884: @table @code
3885: @item MODE_INT
3886: Integer modes. By default these are @code{QImode}, @code{HImode},
3887: @code{SImode}, @code{DImode}, @code{TImode}, and also @code{BImode}.
3888:
3889: @item MODE_FLOAT
3890: Floating-point modes. By default these are @code{QFmode},
3891: @code{HFmode}, @code{SFmode}, @code{DFmode} and @code{TFmode}, but the
3892: MC68881 also defines @code{XFmode} to be an 80-bit extended-precision
3893: floating-point mode.
3894:
3895: @item MODE_COMPLEX_INT
3896: Complex integer modes. By default these are @code{CQImode},
3897: @code{CHImode}, @code{CSImode}, @code{CDImode} and @code{CTImode}.
3898:
3899: @item MODE_COMPLEX_FLOAT
3900: Complex floating-point modes. By default these are @code{CQFmode},
3901: @code{CHFmode}, @code{CSFmode}, @code{CDFmode} and @code{CTFmode},
3902:
3903: @item MODE_FUNCTION
3904: Algol or Pascal function variables including a static chain.
3905: (These are not currently implemented).
3906:
3907: @item MODE_RANDOM
3908: This is a catchall mode class for modes which don't fit into the above
3909: classes. Currently @code{VOIDmode}, @code{BLKmode} and @code{EPmode}
3910: are in @code{MODE_RANDOM}.
3911: @end table
3912:
1.1 root 3913: Here are some C macros that relate to machine modes:
3914:
3915: @table @code
3916: @item GET_MODE (@var{x})
3917: Returns the machine mode of the RTX @var{x}.
3918:
3919: @item PUT_MODE (@var{x}, @var{newmode})
3920: Alters the machine mode of the RTX @var{x} to be @var{newmode}.
3921:
1.1.1.4 root 3922: @item NUM_MACHINE_MODES
3923: Stands for the number of machine modes available on the target
3924: machine. This is one greater than the largest numeric value of any
3925: machine mode.
3926:
3927: @item GET_MODE_NAME (@var{m})
3928: Returns the name of mode @var{m} as a string.
3929:
3930: @item GET_MODE_CLASS (@var{m})
3931: Returns the mode class of mode @var{m}.
3932:
1.1 root 3933: @item GET_MODE_SIZE (@var{m})
3934: Returns the size in bytes of a datum of mode @var{m}.
3935:
3936: @item GET_MODE_BITSIZE (@var{m})
3937: Returns the size in bits of a datum of mode @var{m}.
3938:
3939: @item GET_MODE_UNIT_SIZE (@var{m})
3940: Returns the size in bits of the subunits of a datum of mode @var{m}.
3941: This is the same as @code{GET_MODE_SIZE} except in the case of
3942: complex modes and @code{EPmode}. For them, the unit size is the
3943: size of the real or imaginary part, or the size of the function
3944: pointer or the context pointer.
3945: @end table
3946:
3947: @node Constants, Regs and Memory, Machine Modes, RTL
3948: @section Constant Expression Types
3949:
3950: The simplest RTL expressions are those that represent constant values.
3951:
3952: @table @code
3953: @item (const_int @var{i})
3954: This type of expression represents the integer value @var{i}. @var{i}
3955: is customarily accessed with the macro @code{INTVAL} as in
3956: @code{INTVAL (@var{exp})}, which is equivalent to @code{XINT (@var{exp}, 0)}.
3957:
3958: There is only one expression object for the integer value zero;
3959: it is the value of the variable @code{const0_rtx}. Likewise, the
3960: only expression for integer value one is found in @code{const1_rtx}.
3961: Any attempt to create an expression of code @samp{const_int} and
3962: value zero or one will return @code{const0_rtx} or @code{const1_rtx}
3963: as appropriate.
3964:
3965: @item (const_double:@var{m} @var{i0} @var{i1})
1.1.1.6 ! root 3966: Represents a 64-bit constant of mode @var{m}. All floating point
1.1 root 3967: constants are represented in this way, and so are 64-bit @code{DImode}
3968: integer constants.
3969:
3970: The two integers @var{i0} and @var{i1} together contain the bits of
3971: the value. If the constant is floating point (either single or double
3972: precision), then they represent a @code{double}. To convert them to a
3973: @code{double}, do
3974:
3975: @example
3976: union @{ double d; int i[2];@} u;
3977: u.i[0] = XINT (x, 0);
3978: u.i[1] = XINT (x, 1);
3979: @end example
3980:
3981: @noindent
3982: and then refer to @code{u.d}.
3983:
3984: The global variables @code{dconst0_rtx} and @code{fconst0_rtx} hold
3985: @samp{const_double} expressions with value 0, in modes @code{DFmode} and
3986: @code{SFmode}, respectively.
3987:
3988: @item (symbol_ref @var{symbol})
3989: Represents the value of an assembler label for data. @var{symbol} is
3990: a string that describes the name of the assembler label. If it starts
3991: with a @samp{*}, the label is the rest of @var{symbol} not including
3992: the @samp{*}. Otherwise, the label is @var{symbol}, prefixed with
3993: @samp{_}.
3994:
3995: @item (label_ref @var{label})
3996: Represents the value of an assembler label for code. It contains one
3997: operand, an expression, which must be a @samp{code_label} that appears
3998: in the instruction sequence to identify the place where the label
3999: should go.
4000:
4001: The reason for using a distinct expression type for code label
4002: references is so that jump optimization can distinguish them.
4003:
4004: @item (const @var{exp})
4005: Represents a constant that is the result of an assembly-time
4006: arithmetic computation. The operand, @var{exp}, is an expression that
4007: contains only constants (@samp{const_int}, @samp{symbol_ref} and
4008: @samp{label_ref} expressions) combined with @samp{plus} and
4009: @samp{minus}. However, not all combinations are valid, since the
4010: assembler cannot do arbitrary arithmetic on relocatable symbols.
4011: @end table
4012:
4013: @node Regs and Memory, Arithmetic, Constants, RTL
4014: @section Registers and Memory
4015:
4016: Here are the RTL expression types for describing access to machine
4017: registers and to main memory.
4018:
4019: @table @code
4020: @item (reg:@var{m} @var{n})
4021: For small values of the integer @var{n} (less than
4022: @code{FIRST_PSEUDO_REGISTER}), this stands for a reference to machine
4023: register number @var{n}: a @dfn{hard register}. For larger values of
4024: @var{n}, it stands for a temporary value or @dfn{pseudo register}.
4025: The compiler's strategy is to generate code assuming an unlimited
4026: number of such pseudo registers, and later convert them into hard
4027: registers or into memory references.
4028:
4029: The symbol @code{FIRST_PSEUDO_REGISTER} is defined by the machine
4030: description, since the number of hard registers on the machine is an
4031: invariant characteristic of the machine. Note, however, that not
4032: all of the machine registers must be general registers. All the
4033: machine registers that can be used for storage of data are given
4034: hard register numbers, even those that can be used only in certain
4035: instructions or can hold only certain types of data.
4036:
4037: Each pseudo register number used in a function's RTL code is
4038: represented by a unique @samp{reg} expression.
4039:
4040: @var{m} is the machine mode of the reference. It is necessary because
4041: machines can generally refer to each register in more than one mode.
4042: For example, a register may contain a full word but there may be
4043: instructions to refer to it as a half word or as a single byte, as
4044: well as instructions to refer to it as a floating point number of
4045: various precisions.
4046:
4047: Even for a register that the machine can access in only one mode,
4048: the mode must always be specified.
4049:
4050: A hard register may be accessed in various modes throughout one
4051: function, but each pseudo register is given a natural mode
4052: and is accessed only in that mode. When it is necessary to describe
4053: an access to a pseudo register using a nonnatural mode, a @samp{subreg}
4054: expression is used.
4055:
4056: A @samp{reg} expression with a machine mode that specifies more than
4057: one word of data may actually stand for several consecutive registers.
4058: If in addition the register number specifies a hardware register, then
4059: it actually represents several consecutive hardware registers starting
4060: with the specified one.
4061:
1.1.1.5 root 4062: Such multi-word hardware register @samp{reg} expressions must not be live
1.1 root 4063: across the boundary of a basic block. The lifetime analysis pass does not
4064: know how to record properly that several consecutive registers are
4065: actually live there, and therefore register allocation would be confused.
4066: The CSE pass must go out of its way to make sure the situation does
4067: not arise.
4068:
4069: @item (subreg:@var{m} @var{reg} @var{wordnum})
4070: @samp{subreg} expressions are used to refer to a register in a machine
4071: mode other than its natural one, or to refer to one register of
4072: a multi-word @samp{reg} that actually refers to several registers.
4073:
4074: Each pseudo-register has a natural mode. If it is necessary to
4075: operate on it in a different mode---for example, to perform a fullword
4076: move instruction on a pseudo-register that contains a single byte---
4077: the pseudo-register must be enclosed in a @samp{subreg}. In such
4078: a case, @var{wordnum} is zero.
4079:
4080: The other use of @samp{subreg} is to extract the individual registers
4081: of a multi-register value. Machine modes such as @code{DImode} and
4082: @code{EPmode} indicate values longer than a word, values which usually
4083: require two consecutive registers. To access one of the registers,
4084: use a @samp{subreg} with mode @code{SImode} and a @var{wordnum} that
4085: says which register.
4086:
4087: The compilation parameter @code{WORDS_BIG_ENDIAN}, if defined, says
4088: that word number zero is the most significant part; otherwise, it is
4089: the least significant part.
4090:
4091: Between the combiner pass and the reload pass, it is possible to have
4092: a @samp{subreg} which contains a @samp{mem} instead of a @samp{reg} as
4093: its first operand. The reload pass eliminates these cases by
4094: reloading the @samp{mem} into a suitable register.
4095:
4096: Note that it is not valid to access a @code{DFmode} value in @code{SFmode}
4097: using a @samp{subreg}. On some machines the most significant part of a
4098: @code{DFmode} value does not have the same format as a single-precision
4099: floating value.
4100:
4101: @item (cc0)
4102: This refers to the machine's condition code register. It has no
4103: operands and may not have a machine mode. It may be validly used in
4104: only two contexts: as the destination of an assignment (in test and
4105: compare instructions) and in comparison operators comparing against
4106: zero (@samp{const_int} with value zero; that is to say,
4107: @code{const0_rtx}).
4108:
4109: There is only one expression object of code @samp{cc0}; it is the
4110: value of the variable @code{cc0_rtx}. Any attempt to create an
4111: expression of code @samp{cc0} will return @code{cc0_rtx}.
4112:
4113: One special thing about the condition code register is that
4114: instructions can set it implicitly. On many machines, nearly all
4115: instructions set the condition code based on the value that they
4116: compute or store. It is not necessary to record these actions
4117: explicitly in the RTL because the machine description includes a
4118: prescription for recognizing the instructions that do so (by means of
4119: the macro @code{NOTICE_UPDATE_CC}). Only instructions whose sole
4120: purpose is to set the condition code, and instructions that use the
4121: condition code, need mention @code{(cc0)}.
4122:
4123: @item (pc)
4124: This represents the machine's program counter. It has no operands and
4125: may not have a machine mode. @code{(pc)} may be validly used only in
4126: certain specific contexts in jump instructions.
4127:
4128: There is only one expression object of code @samp{pc}; it is the value
4129: of the variable @code{pc_rtx}. Any attempt to create an expression of
4130: code @samp{pc} will return @code{pc_rtx}.
4131:
4132: All instructions that do not jump alter the program counter implicitly
4133: by incrementing it, but there is no need to mention this in the RTL.
4134:
4135: @item (mem:@var{m} @var{addr})
4136: This RTX represents a reference to main memory at an address
4137: represented by the expression @var{addr}. @var{m} specifies how large
4138: a unit of memory is accessed.
4139: @end table
4140:
4141: @node Arithmetic, Comparisons, Regs and Memory, RTL
4142: @section RTL Expressions for Arithmetic
4143:
4144: @table @code
4145: @item (plus:@var{m} @var{x} @var{y})
4146: Represents the sum of the values represented by @var{x} and @var{y}
4147: carried out in machine mode @var{m}. This is valid only if
4148: @var{x} and @var{y} both are valid for mode @var{m}.
4149:
4150: @item (minus:@var{m} @var{x} @var{y})
4151: Like @samp{plus} but represents subtraction.
4152:
1.1.1.6 ! root 4153: @item (compare @var{x} @var{y})
1.1 root 4154: Represents the result of subtracting @var{y} from @var{x}
4155: for purposes of comparison. The absence of a machine mode
1.1.1.6 ! root 4156: in the @samp{compare} expression indicates that the result is
1.1 root 4157: computed without overflow, as if with infinite precision.
4158:
4159: Of course, machines can't really subtract with infinite precision.
4160: However, they can pretend to do so when only the sign of the
4161: result will be used, which is the case when the result is stored
4162: in @code{(cc0)}. And that is the only way this kind of expression
4163: may validly be used: as a value to be stored in the condition codes.
4164:
4165: @item (neg:@var{m} @var{x})
4166: Represents the negation (subtraction from zero) of the value
4167: represented by @var{x}, carried out in mode @var{m}. @var{x} must be
4168: valid for mode @var{m}.
4169:
4170: @item (mult:@var{m} @var{x} @var{y})
4171: Represents the signed product of the values represented by @var{x} and
4172: @var{y} carried out in machine mode @var{m}. If
4173: @var{x} and @var{y} are both valid for mode @var{m}, this is ordinary
4174: size-preserving multiplication. Alternatively, both @var{x} and @var{y}
4175: may be valid for a different, narrower mode. This represents the
4176: kind of multiplication that generates a product wider than the operands.
4177: Widening multiplication and same-size multiplication are completely
4178: distinct and supported by different machine instructions; machines may
4179: support one but not the other.@refill
4180:
1.1.1.5 root 4181: @samp{mult} may be used for floating point multiplication as well.
1.1 root 4182: Then @var{m} is a floating point machine mode.
4183:
4184: @item (umult:@var{m} @var{x} @var{y})
4185: Like @samp{mult} but represents unsigned multiplication. It may be
4186: used in both same-size and widening forms, like @samp{mult}.
4187: @samp{umult} is used only for fixed-point multiplication.
4188:
4189: @item (div:@var{m} @var{x} @var{y})
4190: Represents the quotient in signed division of @var{x} by @var{y},
4191: carried out in machine mode @var{m}. If @var{m} is a floating-point
4192: mode, it represents the exact quotient; otherwise, the integerized
4193: quotient. If @var{x} and @var{y} are both valid for mode @var{m},
4194: this is ordinary size-preserving division. Some machines have
4195: division instructions in which the operands and quotient widths are
4196: not all the same; such instructions are represented by @samp{div}
4197: expressions in which the machine modes are not all the same.
4198:
4199: @item (udiv:@var{m} @var{x} @var{y})
4200: Like @samp{div} but represents unsigned division.
4201:
4202: @item (mod:@var{m} @var{x} @var{y})
4203: @itemx (umod:@var{m} @var{x} @var{y})
4204: Like @samp{div} and @samp{udiv} but represent the remainder instead of
4205: the quotient.
4206:
4207: @item (not:@var{m} @var{x})
4208: Represents the bitwise complement of the value represented by @var{x},
4209: carried out in mode @var{m}, which must be a fixed-point machine mode.
4210: @var{x} must be valid for mode @var{m}, which must be a fixed-point mode.
4211:
4212: @item (and:@var{m} @var{x} @var{y})
4213: Represents the bitwise logical-and of the values represented by
4214: @var{x} and @var{y}, carried out in machine mode @var{m}. This is
4215: valid only if @var{x} and @var{y} both are valid for mode @var{m},
4216: which must be a fixed-point mode.
4217:
4218: @item (ior:@var{m} @var{x} @var{y})
4219: Represents the bitwise inclusive-or of the values represented by
4220: @var{x} and @var{y}, carried out in machine mode @var{m}. This is
4221: valid only if @var{x} and @var{y} both are valid for mode @var{m},
4222: which must be a fixed-point mode.
4223:
4224: @item (xor:@var{m} @var{x} @var{y})
4225: Represents the bitwise exclusive-or of the values represented by
4226: @var{x} and @var{y}, carried out in machine mode @var{m}. This is
4227: valid only if @var{x} and @var{y} both are valid for mode @var{m},
4228: which must be a fixed-point mode.
4229:
4230: @item (lshift:@var{m} @var{x} @var{c})
4231: Represents the result of logically shifting @var{x} left by @var{c}
4232: places. @var{x} must be valid for the mode @var{m}, a fixed-point
4233: machine mode. @var{c} must be valid for a fixed-point mode;
4234: which mode is determined by the mode called for in the machine
4235: description entry for the left-shift instruction. For example,
4236: on the Vax, the mode of @var{c} is @code{QImode} regardless of @var{m}.
4237:
4238: On some machines, negative values of @var{c} may be meaningful; this
4239: is why logical left shift and arithmetic left shift are distinguished.
4240: For example, Vaxes have no right-shift instructions, and right shifts
4241: are represented as left-shift instructions whose counts happen
4242: to be negative constants or else computed (in a previous instruction)
4243: by negation.
4244:
4245: @item (ashift:@var{m} @var{x} @var{c})
4246: Like @samp{lshift} but for arithmetic left shift.
4247:
4248: @item (lshiftrt:@var{m} @var{x} @var{c})
4249: @itemx (ashiftrt:@var{m} @var{x} @var{c})
4250: Like @samp{lshift} and @samp{ashift} but for right shift.
4251:
4252: @item (rotate:@var{m} @var{x} @var{c})
4253: @itemx (rotatert:@var{m} @var{x} @var{c})
4254: Similar but represent left and right rotate.
4255:
4256: @item (abs:@var{m} @var{x})
4257: Represents the absolute value of @var{x}, computed in mode @var{m}.
4258: @var{x} must be valid for @var{m}.
4259:
4260: @item (sqrt:@var{m} @var{x})
4261: Represents the square root of @var{x}, computed in mode @var{m}.
4262: @var{x} must be valid for @var{m}. Most often @var{m} will be
4263: a floating point mode.
4264:
4265: @item (ffs:@var{m} @var{x})
4266: Represents the one plus the index of the least significant 1-bit in
4267: @var{x}, represented as an integer of mode @var{m}. (The value is
4268: zero if @var{x} is zero.) The mode of @var{x} need not be @var{m};
4269: depending on the target machine, various mode combinations may be
4270: valid.
4271: @end table
4272:
4273: @node Comparisons, Bit Fields, Arithmetic, RTL
4274: @section Comparison Operations
4275:
4276: Comparison operators test a relation on two operands and are considered to
4277: represent the value 1 if the relation holds, or zero if it does not. The
4278: mode of the comparison is determined by the operands; they must both be
4279: valid for a common machine mode. A comparison with both operands constant
4280: would be invalid as the machine mode could not be deduced from it, but such
4281: a comparison should never exist in RTL due to constant folding.
4282:
4283: Inequality comparisons come in two flavors, signed and unsigned. Thus,
4284: there are distinct expression codes @samp{gt} and @samp{gtu} for signed and
4285: unsigned greater-than. These can produce different results for the same
4286: pair of integer values: for example, 1 is signed greater-than -1 but not
4287: unsigned greater-than, because -1 when regarded as unsigned is actually
4288: @code{0xffffffff} which is greater than 1.
4289:
4290: The signed comparisons are also used for floating point values. Floating
4291: point comparisons are distinguished by the machine modes of the operands.
4292:
4293: The comparison operators may be used to compare the condition codes
4294: @code{(cc0)} against zero, as in @code{(eq (cc0) (const_int 0))}. Such a
4295: construct actually refers to the result of the preceding instruction in
4296: which the condition codes were set. The above example stands for 1 if the
4297: condition codes were set to say ``zero'' or ``equal'', 0 otherwise.
4298: Although the same comparison operators are used for this as may be used in
4299: other contexts on actual data, no confusion can result since the machine
4300: description would never allow both kinds of uses in the same context.
4301:
4302: @table @code
4303: @item (eq @var{x} @var{y})
4304: 1 if the values represented by @var{x} and @var{y} are equal,
4305: otherwise 0.
4306:
4307: @item (ne @var{x} @var{y})
4308: 1 if the values represented by @var{x} and @var{y} are not equal,
4309: otherwise 0.
4310:
4311: @item (gt @var{x} @var{y})
4312: 1 if the @var{x} is greater than @var{y}. If they are fixed-point,
4313: the comparison is done in a signed sense.
4314:
4315: @item (gtu @var{x} @var{y})
4316: Like @samp{gt} but does unsigned comparison, on fixed-point numbers only.
4317:
4318: @item (lt @var{x} @var{y})
4319: @item (ltu @var{x} @var{y})
4320: Like @samp{gt} and @samp{gtu} but test for ``less than''.
4321:
4322: @item (ge @var{x} @var{y})
4323: @item (geu @var{x} @var{y})
4324: Like @samp{gt} and @samp{gtu} but test for ``greater than or equal''.
4325:
4326: @item (le @var{x} @var{y})
4327: @item (leu @var{x} @var{y})
4328: Like @samp{gt} and @samp{gtu} but test for ``less than or equal''.
4329:
4330: @item (if_then_else @var{cond} @var{then} @var{else})
4331: This is not a comparison operation but is listed here because it is
4332: always used in conjunction with a comparison operation. To be
4333: precise, @var{cond} is a comparison expression. This expression
4334: represents a choice, according to @var{cond}, between the value
4335: represented by @var{then} and the one represented by @var{else}.
4336:
4337: On most machines, @samp{if_then_else} expressions are valid only
4338: to express conditional jumps.
4339: @end table
4340:
4341: @node Bit Fields, Conversions, Comparisons, RTL
4342: @section Bit-fields
4343:
4344: Special expression codes exist to represent bit-field instructions.
4345: These types of expressions are lvalues in RTL; they may appear
4346: on the left side of a assignment, indicating insertion of a value
4347: into the specified bit field.
4348:
4349: @table @code
4350: @item (sign_extract:SI @var{loc} @var{size} @var{pos})
4351: This represents a reference to a sign-extended bit-field contained or
4352: starting in @var{loc} (a memory or register reference). The bit field
4353: is @var{size} bits wide and starts at bit @var{pos}. The compilation
4354: option @code{BITS_BIG_ENDIAN} says which end of the memory unit
4355: @var{pos} counts from.
4356:
4357: Which machine modes are valid for @var{loc} depends on the machine,
4358: but typically @var{loc} should be a single byte when in memory
4359: or a full word in a register.
4360:
4361: @item (zero_extract:SI @var{loc} @var{size} @var{pos})
4362: Like @samp{sign_extract} but refers to an unsigned or zero-extended
4363: bit field. The same sequence of bits are extracted, but they
4364: are filled to an entire word with zeros instead of by sign-extension.
4365: @end table
4366:
4367: @node Conversions, RTL Declarations, Bit Fields, RTL
4368: @section Conversions
4369:
4370: All conversions between machine modes must be represented by
4371: explicit conversion operations. For example, an expression
4372: which is the sum of a byte and a full word cannot be written as
4373: @code{(plus:SI (reg:QI 34) (reg:SI 80))} because the @samp{plus}
4374: operation requires two operands of the same machine mode.
4375: Therefore, the byte-sized operand is enclosed in a conversion
4376: operation, as in
4377:
4378: @example
4379: (plus:SI (sign_extend:SI (reg:QI 34)) (reg:SI 80))
4380: @end example
4381:
4382: The conversion operation is not a mere placeholder, because there
4383: may be more than one way of converting from a given starting mode
4384: to the desired final mode. The conversion operation code says how
4385: to do it.
4386:
4387: @table @code
4388: @item (sign_extend:@var{m} @var{x})
4389: Represents the result of sign-extending the value @var{x}
4390: to machine mode @var{m}. @var{m} must be a fixed-point mode
4391: and @var{x} a fixed-point value of a mode narrower than @var{m}.
4392:
4393: @item (zero_extend:@var{m} @var{x})
4394: Represents the result of zero-extending the value @var{x}
4395: to machine mode @var{m}. @var{m} must be a fixed-point mode
4396: and @var{x} a fixed-point value of a mode narrower than @var{m}.
4397:
4398: @item (float_extend:@var{m} @var{x})
4399: Represents the result of extending the value @var{x}
4400: to machine mode @var{m}. @var{m} must be a floating point mode
4401: and @var{x} a floating point value of a mode narrower than @var{m}.
4402:
4403: @item (truncate:@var{m} @var{x})
4404: Represents the result of truncating the value @var{x}
4405: to machine mode @var{m}. @var{m} must be a fixed-point mode
4406: and @var{x} a fixed-point value of a mode wider than @var{m}.
4407:
4408: @item (float_truncate:@var{m} @var{x})
4409: Represents the result of truncating the value @var{x}
4410: to machine mode @var{m}. @var{m} must be a floating point mode
4411: and @var{x} a floating point value of a mode wider than @var{m}.
4412:
4413: @item (float:@var{m} @var{x})
4414: Represents the result of converting fixed point value @var{x},
4415: regarded as signed, to floating point mode @var{m}.
4416:
4417: @item (unsigned_float:@var{m} @var{x})
4418: Represents the result of converting fixed point value @var{x},
4419: regarded as unsigned, to floating point mode @var{m}.
4420:
4421: @item (fix:@var{m} @var{x})
4422: When @var{m} is a fixed point mode, represents the result of
4423: converting floating point value @var{x} to mode @var{m}, regarded as
4424: signed. How rounding is done is not specified, so this operation may
4425: be used validly in compiling C code only for integer-valued operands.
4426:
4427: @item (unsigned_fix:@var{m} @var{x})
4428: Represents the result of converting floating point value @var{x} to
4429: fixed point mode @var{m}, regarded as unsigned. How rounding is done
4430: is not specified.
4431:
4432: @item (fix:@var{m} @var{x})
4433: When @var{m} is a floating point mode, represents the result of
4434: converting floating point value @var{x} (valid for mode @var{m}) to an
4435: integer, still represented in floating point mode @var{m}, by rounding
4436: towards zero.
4437: @end table
4438:
4439: @node RTL Declarations, Side Effects, Conversions, RTL
4440: @section Declarations
4441:
4442: Declaration expression codes do not represent arithmetic operations
4443: but rather state assertions about their operands.
4444:
4445: @table @code
4446: @item (strict_low_part (subreg:@var{m} (reg:@var{n} @var{r}) 0))
4447: This expression code is used in only one context: operand 0 of a
4448: @samp{set} expression. In addition, the operand of this expression
4449: must be a @samp{subreg} expression.
4450:
4451: The presence of @samp{strict_low_part} says that the part of the
4452: register which is meaningful in mode @var{n}, but is not part of
4453: mode @var{m}, is not to be altered. Normally, an assignment to such
4454: a subreg is allowed to have undefined effects on the rest of the
4455: register when @var{m} is less than a word.
4456: @end table
4457:
4458: @node Side Effects, Incdec, RTL Declarations, RTL
4459: @section Side Effect Expressions
4460:
4461: The expression codes described so far represent values, not actions.
4462: But machine instructions never produce values; they are meaningful
4463: only for their side effects on the state of the machine. Special
4464: expression codes are used to represent side effects.
4465:
4466: The body of an instruction is always one of these side effect codes;
4467: the codes described above, which represent values, appear only as
4468: the operands of these.
4469:
4470: @table @code
4471: @item (set @var{lval} @var{x})
4472: Represents the action of storing the value of @var{x} into the place
4473: represented by @var{lval}. @var{lval} must be an expression
4474: representing a place that can be stored in: @samp{reg} (or
4475: @samp{subreg} or @samp{strict_low_part}), @samp{mem}, @samp{pc} or
4476: @samp{cc0}.@refill
4477:
4478: If @var{lval} is a @samp{reg}, @samp{subreg} or @samp{mem}, it has a
4479: machine mode; then @var{x} must be valid for that mode.@refill
4480:
4481: If @var{lval} is a @samp{reg} whose machine mode is less than the full
4482: width of the register, then it means that the part of the register
4483: specified by the machine mode is given the specified value and the
4484: rest of the register receives an undefined value. Likewise, if
4485: @var{lval} is a @samp{subreg} whose machine mode is narrower than
4486: @code{SImode}, the rest of the register can be changed in an undefined way.
4487:
4488: If @var{lval} is a @samp{strict_low_part} of a @samp{subreg}, then the
4489: part of the register specified by the machine mode of the
4490: @samp{subreg} is given the value @var{x} and the rest of the register
4491: is not changed.@refill
4492:
4493: If @var{lval} is @code{(cc0)}, it has no machine mode, and @var{x} may
4494: have any mode. This represents a ``test'' or ``compare'' instruction.@refill
4495:
4496: If @var{lval} is @code{(pc)}, we have a jump instruction, and the
4497: possibilities for @var{x} are very limited. It may be a
4498: @samp{label_ref} expression (unconditional jump). It may be an
4499: @samp{if_then_else} (conditional jump), in which case either the
4500: second or the third operand must be @code{(pc)} (for the case which
4501: does not jump) and the other of the two must be a @samp{label_ref}
4502: (for the case which does jump). @var{x} may also be a @samp{mem} or
4503: @code{(plus:SI (pc) @var{y})}, where @var{y} may be a @samp{reg} or a
4504: @samp{mem}; these unusual patterns are used to represent jumps through
4505: branch tables.@refill
4506:
4507: @item (return)
4508: Represents a return from the current function, on machines where this
4509: can be done with one instruction, such as Vaxes. On machines where a
4510: multi-instruction ``epilogue'' must be executed in order to return
4511: from the function, returning is done by jumping to a label which
4512: precedes the epilogue, and the @samp{return} expression code is never
4513: used.
4514:
4515: @item (call @var{function} @var{nargs})
4516: Represents a function call. @var{function} is a @samp{mem} expression
4517: whose address is the address of the function to be called.
4518: @var{nargs} is an expression which can be used for two purposes: on
4519: some machines it represents the number of bytes of stack argument; on
4520: others, it represents the number of argument registers.
4521:
4522: Each machine has a standard machine mode which @var{function} must
4523: have. The machine description defines macro @code{FUNCTION_MODE} to
4524: expand into the requisite mode name. The purpose of this mode is to
4525: specify what kind of addressing is allowed, on machines where the
4526: allowed kinds of addressing depend on the machine mode being
4527: addressed.
4528:
4529: @item (clobber @var{x})
4530: Represents the storing or possible storing of an unpredictable,
4531: undescribed value into @var{x}, which must be a @samp{reg} or
4532: @samp{mem} expression.
4533:
4534: One place this is used is in string instructions that store standard
4535: values into particular hard registers. It may not be worth the
4536: trouble to describe the values that are stored, but it is essential to
4537: inform the compiler that the registers will be altered, lest it
4538: attempt to keep data in them across the string instruction.
4539:
4540: @var{x} may also be null---a null C pointer, no expression at all.
4541: Such a @code{(clobber (null))} expression means that all memory
4542: locations must be presumed clobbered.
4543:
4544: Note that the machine description classifies certain hard registers as
4545: ``call-clobbered''. All function call instructions are assumed by
4546: default to clobber these registers, so there is no need to use
4547: @samp{clobber} expressions to indicate this fact. Also, each function
1.1.1.6 ! root 4548: call is assumed to have the potential to alter any memory location,
! 4549: unless the function is declared @code{const}.
1.1 root 4550:
1.1.1.4 root 4551: When a @samp{clobber} expression for a register appears inside a
4552: @samp{parallel} with other side effects, GNU CC guarantees that the
4553: register is unoccupied both before and after that insn. Therefore, it
4554: is safe for the assembler code produced by the insn to use the
4555: register as a temporary. You can clobber either a specific hard
4556: register or a pseudo register; in the latter case, GNU CC will
4557: allocate a hard register that is available there for use as a
4558: temporary.
4559:
1.1 root 4560: @item (use @var{x})
4561: Represents the use of the value of @var{x}. It indicates that the
4562: value in @var{x} at this point in the program is needed, even though
4563: it may not be apparent why this is so. Therefore, the compiler will
1.1.1.4 root 4564: not attempt to delete previous instructions whose only effect is to
4565: store a value in @var{x}. @var{x} must be a @samp{reg} expression.
1.1 root 4566:
4567: @item (parallel [@var{x0} @var{x1} @dots{}])
4568: Represents several side effects performed in parallel. The square
4569: brackets stand for a vector; the operand of @samp{parallel} is a
4570: vector of expressions. @var{x0}, @var{x1} and so on are individual
1.1.1.4 root 4571: side effect expressions---expressions of code @samp{set}, @samp{call},
1.1 root 4572: @samp{return}, @samp{clobber} or @samp{use}.@refill
4573:
4574: ``In parallel'' means that first all the values used in the individual
4575: side-effects are computed, and second all the actual side-effects are
4576: performed. For example,
4577:
4578: @example
4579: (parallel [(set (reg:SI 1) (mem:SI (reg:SI 1)))
4580: (set (mem:SI (reg:SI 1)) (reg:SI 1))])
4581: @end example
4582:
4583: @noindent
4584: says unambiguously that the values of hard register 1 and the memory
4585: location addressed by it are interchanged. In both places where
4586: @code{(reg:SI 1)} appears as a memory address it refers to the value
1.1.1.4 root 4587: in register 1 @emph{before} the execution of the insn.
4588:
4589: It follows that it is @emph{incorrect} to use @samp{parallel} and
4590: expect the result of one @samp{set} to be available for the next one.
4591: For example, people sometimes attempt to represent a jump-if-zero
4592: instruction this way:
4593:
4594: @example
4595: (parallel [(set (cc0) (reg:SI 34))
4596: (set (pc) (if_then_else
4597: (eq (cc0) (const_int 0))
4598: (label_ref @dots{})
4599: (pc)))])
4600: @end example
4601:
4602: @noindent
4603: But this is incorrect, because it says that the jump condition depends
4604: on the condition code value @emph{before} this instruction, not on the
4605: new value that is set by this instruction.
1.1 root 4606:
1.1.1.5 root 4607: Peephole optimization, which takes place in together with final assembly
4608: code output, can produce insns whose patterns consist of a @samp{parallel}
1.1 root 4609: whose elements are the operands needed to output the resulting
4610: assembler code--often @samp{reg}, @samp{mem} or constant expressions.
4611: This would not be well-formed RTL at any other stage in compilation,
4612: but it is ok then because no further optimization remains to be done.
1.1.1.4 root 4613: However, the definition of the macro @code{NOTICE_UPDATE_CC} must
4614: deal with such insns if you define any peephole optimizations.
1.1 root 4615:
4616: @item (sequence [@var{insns} @dots{}])
4617: Represents a sequence of insns. Each of the @var{insns} that appears
4618: in the vector is suitable for appearing in the chain of insns, so it
4619: must be an @samp{insn}, @samp{jump_insn}, @samp{call_insn},
4620: @samp{code_label}, @samp{barrier} or @samp{note}.
4621:
4622: A @samp{sequence} RTX never appears in an actual insn. It represents
4623: the sequence of insns that result from a @samp{define_expand}
4624: @emph{before} those insns are passed to @code{emit_insn} to insert
4625: them in the chain of insns. When actually inserted, the individual
4626: sub-insns are separated out and the @samp{sequence} is forgotten.
4627: @end table
4628:
4629: Three expression codes appear in place of a side effect, as the body of an
4630: insn, though strictly speaking they do not describe side effects as such:
4631:
4632: @table @code
4633: @item (asm_input @var{s})
4634: Represents literal assembler code as described by the string @var{s}.
4635:
4636: @item (addr_vec:@var{m} [@var{lr0} @var{lr1} @dots{}])
4637: Represents a table of jump addresses. The vector elements @var{lr0},
4638: etc., are @samp{label_ref} expressions. The mode @var{m} specifies
4639: how much space is given to each address; normally @var{m} would be
4640: @code{Pmode}.
4641:
4642: @item (addr_diff_vec:@var{m} @var{base} [@var{lr0} @var{lr1} @dots{}])
4643: Represents a table of jump addresses expressed as offsets from
4644: @var{base}. The vector elements @var{lr0}, etc., are @samp{label_ref}
4645: expressions and so is @var{base}. The mode @var{m} specifies how much
4646: space is given to each address-difference.@refill
4647: @end table
4648:
4649: @node Incdec, Assembler, Side Effects, RTL
4650: @section Embedded Side-Effects on Addresses
4651:
4652: Four special side-effect expression codes appear as memory addresses.
4653:
4654: @table @code
4655: @item (pre_dec:@var{m} @var{x})
4656: Represents the side effect of decrementing @var{x} by a standard
4657: amount and represents also the value that @var{x} has after being
4658: decremented. @var{x} must be a @samp{reg} or @samp{mem}, but most
4659: machines allow only a @samp{reg}. @var{m} must be the machine mode
4660: for pointers on the machine in use. The amount @var{x} is decremented
4661: by is the length in bytes of the machine mode of the containing memory
4662: reference of which this expression serves as the address. Here is an
4663: example of its use:@refill
4664:
4665: @example
4666: (mem:DF (pre_dec:SI (reg:SI 39)))
4667: @end example
4668:
4669: @noindent
4670: This says to decrement pseudo register 39 by the length of a @code{DFmode}
4671: value and use the result to address a @code{DFmode} value.
4672:
4673: @item (pre_inc:@var{m} @var{x})
4674: Similar, but specifies incrementing @var{x} instead of decrementing it.
4675:
4676: @item (post_dec:@var{m} @var{x})
4677: Represents the same side effect as @samp{pre_decrement} but a different
4678: value. The value represented here is the value @var{x} has @i{before}
4679: being decremented.
4680:
4681: @item (post_inc:@var{m} @var{x})
4682: Similar, but specifies incrementing @var{x} instead of decrementing it.
4683: @end table
4684:
4685: These embedded side effect expressions must be used with care. Instruction
4686: patterns may not use them. Until the @samp{flow} pass of the compiler,
4687: they may occur only to represent pushes onto the stack. The @samp{flow}
4688: pass finds cases where registers are incremented or decremented in one
4689: instruction and used as an address shortly before or after; these cases are
4690: then transformed to use pre- or post-increment or -decrement.
4691:
4692: Explicit popping of the stack could be represented with these embedded
4693: side effect operators, but that would not be safe; the instruction
4694: combination pass could move the popping past pushes, thus changing
4695: the meaning of the code.
4696:
4697: An instruction that can be represented with an embedded side effect
4698: could also be represented using @samp{parallel} containing an additional
4699: @samp{set} to describe how the address register is altered. This is not
4700: done because machines that allow these operations at all typically
4701: allow them wherever a memory address is called for. Describing them as
4702: additional parallel stores would require doubling the number of entries
4703: in the machine description.
4704:
4705: @node Assembler, Insns, IncDec, RTL
4706: @section Assembler Instructions as Expressions
4707:
4708: The RTX code @samp{asm_operands} represents a value produced by a
4709: user-specified assembler instruction. It is used to represent
4710: an @code{asm} statement with arguments. An @code{asm} statement with
4711: a single output operand, like this:
4712:
4713: @example
1.1.1.6 ! root 4714: asm ("foo %1,%2,%0" : "=a" (outputvar) : "g" (x + y), "di" (*z));
1.1 root 4715: @end example
4716:
4717: @noindent
4718: is represented using a single @samp{asm_operands} RTX which represents
4719: the value that is stored in @code{outputvar}:
4720:
4721: @example
4722: (set @var{rtx-for-outputvar}
4723: (asm_operands "foo %1,%2,%0" "a" 0
4724: [@var{rtx-for-addition-result} @var{rtx-for-*z}]
4725: [(asm_input:@var{m1} "g")
4726: (asm_input:@var{m2} "di")]))
4727: @end example
4728:
4729: @noindent
4730: Here the operands of the @samp{asm_operands} RTX are the assembler
4731: template string, the output-operand's constraint, the index-number of the
4732: output operand among the output operands specified, a vector of input
4733: operand RTX's, and a vector of input-operand modes and constraints. The
4734: mode @var{m1} is the mode of the sum @code{x+y}; @var{m2} is that of
4735: @code{*z}.
4736:
4737: When an @code{asm} statement has multiple output values, its insn has
4738: several such @samp{set} RTX's inside of a @samp{parallel}. Each @samp{set}
4739: contains a @samp{asm_operands}; all of these share the same assembler
4740: template and vectors, but each contains the constraint for the respective
4741: output operand. They are also distinguished by the output-operand index
4742: number, which is 0, 1, @dots{} for successive output operands.
4743:
4744: @node Insns, Calls, Assembler, RTL
4745: @section Insns
4746:
4747: The RTL representation of the code for a function is a doubly-linked
4748: chain of objects called @dfn{insns}. Insns are expressions with
4749: special codes that are used for no other purpose. Some insns are
4750: actual instructions; others represent dispatch tables for @code{switch}
4751: statements; others represent labels to jump to or various sorts of
4752: declarative information.
4753:
4754: In addition to its own specific data, each insn must have a unique id-number
4755: that distinguishes it from all other insns in the current function, and
4756: chain pointers to the preceding and following insns. These three fields
4757: occupy the same position in every insn, independent of the expression code
4758: of the insn. They could be accessed with @code{XEXP} and @code{XINT},
4759: but instead three special macros are always used:
4760:
4761: @table @code
4762: @item INSN_UID (@var{i})
4763: Accesses the unique id of insn @var{i}.
4764:
4765: @item PREV_INSN (@var{i})
4766: Accesses the chain pointer to the insn preceding @var{i}.
4767: If @var{i} is the first insn, this is a null pointer.
4768:
4769: @item NEXT_INSN (@var{i})
4770: Accesses the chain pointer to the insn following @var{i}.
4771: If @var{i} is the last insn, this is a null pointer.
4772: @end table
4773:
4774: The @code{NEXT_INSN} and @code{PREV_INSN} pointers must always
1.1.1.6 ! root 4775: correspond: if @var{insn} is not the first insn,
1.1 root 4776:
4777: @example
4778: NEXT_INSN (PREV_INSN (@var{insn})) == @var{insn}
4779: @end example
4780:
4781: @noindent
4782: is always true.
4783:
4784: Every insn has one of the following six expression codes:
4785:
4786: @table @samp
4787: @item insn
4788: The expression code @samp{insn} is used for instructions that do not jump
4789: and do not do function calls. Insns with code @samp{insn} have four
4790: additional fields beyond the three mandatory ones listed above.
4791: These four are described in a table below.
4792:
4793: @item jump_insn
4794: The expression code @samp{jump_insn} is used for instructions that may jump
4795: (or, more generally, may contain @samp{label_ref} expressions).
4796: @samp{jump_insn} insns have the same extra fields as @samp{insn} insns,
4797: accessed in the same way.
4798:
4799: @item call_insn
4800: The expression code @samp{call_insn} is used for instructions that may do
4801: function calls. It is important to distinguish these instructions because
4802: they imply that certain registers and memory locations may be altered
4803: unpredictably.
4804:
4805: @samp{call_insn} insns have the same extra fields as @samp{insn} insns,
4806: accessed in the same way.
4807:
4808: @item code_label
4809: A @samp{code_label} insn represents a label that a jump insn can jump to.
4810: It contains one special field of data in addition to the three standard ones.
4811: It is used to hold the @dfn{label number}, a number that identifies this
4812: label uniquely among all the labels in the compilation (not just in the
4813: current function). Ultimately, the label is represented in the assembler
4814: output as an assembler label @samp{L@var{n}} where @var{n} is the label number.
4815:
4816: @item barrier
4817: Barriers are placed in the instruction stream after unconditional
4818: jump instructions to indicate that the jumps are unconditional.
4819: They contain no information beyond the three standard fields.
4820:
4821: @item note
4822: @samp{note} insns are used to represent additional debugging and
4823: declarative information. They contain two nonstandard fields, an
4824: integer which is accessed with the macro @code{NOTE_LINE_NUMBER} and a
4825: string accessed with @code{NOTE_SOURCE_FILE}.
4826:
4827: If @code{NOTE_LINE_NUMBER} is positive, the note represents the
4828: position of a source line and @code{NOTE_SOURCE_FILE} is the source file name
4829: that the line came from. These notes control generation of line
4830: number data in the assembler output.
4831:
4832: Otherwise, @code{NOTE_LINE_NUMBER} is not really a line number but a
4833: code with one of the following values (and @code{NOTE_SOURCE_FILE}
4834: must contain a null pointer):
4835:
4836: @table @code
4837: @item NOTE_INSN_DELETED
4838: Such a note is completely ignorable. Some passes of the compiler
4839: delete insns by altering them into notes of this kind.
4840:
4841: @item NOTE_INSN_BLOCK_BEG
4842: @itemx NOTE_INSN_BLOCK_END
4843: These types of notes indicate the position of the beginning and end
4844: of a level of scoping of variable names. They control the output
4845: of debugging information.
4846:
4847: @item NOTE_INSN_LOOP_BEG
4848: @itemx NOTE_INSN_LOOP_END
4849: These types of notes indicate the position of the beginning and end
4850: of a @code{while} or @code{for} loop. They enable the loop optimizer
4851: to find loops quickly.
1.1.1.6 ! root 4852: @item NOTE_INSN_FUNCTION_END
! 4853: Appears near the end of the function body, just before the label that
! 4854: @code{return} statements jump to (on machine where a single instruction
! 4855: does not suffice for returning). This note may be deleted by jump
! 4856: optimization.
! 4857: @item NOTE_INSN_SETJMP
! 4858: Appears following each call to @code{setjmp} or a related function.
1.1 root 4859: @end table
4860: @end table
4861:
1.1.1.6 ! root 4862: The machine mode of an insn is normally zero (@code{VOIDmode}), but the
! 4863: reload pass sets it to @code{QImode} if the insn needs reloading.
! 4864:
1.1 root 4865: Here is a table of the extra fields of @samp{insn}, @samp{jump_insn}
4866: and @samp{call_insn} insns:
4867:
4868: @table @code
4869: @item PATTERN (@var{i})
4870: An expression for the side effect performed by this insn.
4871:
1.1.1.6 ! root 4872: @item INSN_CODE (@var{i})
! 4873: An integer that says which pattern in the machine description matches
! 4874: this insn, or -1 if the matching has not yet been attempted.
! 4875:
! 4876: Such matching is never attempted and this field is not used on an insn
! 4877: whose pattern consists of a single @samp{use}, @samp{clobber},
! 4878: @samp{asm}, @samp{addr_vec} or @samp{addr_diff_vec} expression.
1.1 root 4879:
4880: @item LOG_LINKS (@var{i})
4881: A list (chain of @samp{insn_list} expressions) of previous ``related''
4882: insns: insns which store into registers values that are used for the
4883: first time in this insn. (An additional constraint is that neither a
4884: jump nor a label may come between the related insns). This list is
4885: set up by the flow analysis pass; it is a null pointer until then.
4886:
1.1.1.6 ! root 4887: @item REG_NOTES (@var{i})
! 4888: A list (chain of @samp{expr_list} expressions) giving information
! 4889: about the usage of registers in this insn. This list is set up by the
! 4890: flow analysis pass; it is a null pointer until then.
1.1 root 4891: @end table
4892:
4893: The @code{LOG_LINKS} field of an insn is a chain of @samp{insn_list}
4894: expressions. Each of these has two operands: the first is an insn,
4895: and the second is another @samp{insn_list} expression (the next one in
4896: the chain). The last @samp{insn_list} in the chain has a null pointer
4897: as second operand. The significant thing about the chain is which
4898: insns appear in it (as first operands of @samp{insn_list}
4899: expressions). Their order is not significant.
4900:
4901: The @code{REG_NOTES} field of an insn is a similar chain but of
1.1.1.5 root 4902: @samp{expr_list} expressions instead of @samp{insn_list}. There are
4903: several kinds of register notes, which are distinguished by the machine
4904: mode of the @samp{expr_list}, which in a register note is really
4905: understood as being an @code{enum reg_note}. The first operand @var{op}
4906: of the @samp{expr_list} is data whose meaning depends on the kind of
4907: note. Here are the kinds of register note:
1.1 root 4908:
4909: @table @code
4910: @item REG_DEAD
4911: The register @var{op} dies in this insn; that is to say, altering the
4912: value immediately after this insn would not affect the future behavior
4913: of the program.
4914:
4915: @item REG_INC
4916: The register @var{op} is incremented (or decremented; at this level
4917: there is no distinction) by an embedded side effect inside this insn.
4918: This means it appears in a @code{POST_INC}, @code{PRE_INC},
4919: @code{POST_DEC} or @code{PRE_DEC} RTX.
4920:
4921: @item REG_EQUIV
4922: The register that is set by this insn will be equal to @var{op} at run
4923: time, and could validly be replaced in all its occurrences by
4924: @var{op}. (``Validly'' here refers to the data flow of the program;
4925: simple replacement may make some insns invalid.)
4926:
4927: The value which the insn explicitly copies into the register may look
4928: different from @var{op}, but they will be equal at run time.
4929:
4930: For example, when a constant is loaded into a register that is never
4931: assigned any other value, this kind of note is used.
4932:
4933: When a parameter is copied into a pseudo-register at entry to a function,
4934: a note of this kind records that the register is equivalent to the stack
4935: slot where the parameter was passed. Although in this case the register
4936: may be set by other insns, it is still valid to replace the register
4937: by the stack slot throughout the function.
4938:
4939: @item REG_EQUAL
4940: The register that is set by this insn will be equal to @var{op} at run
4941: time at the end of this insn (but not necessarily elsewhere in the
4942: function).
4943:
4944: The RTX @var{op} is typically an arithmetic expression. For example,
4945: when a sequence of insns such as a library call is used to perform an
4946: arithmetic operation, this kind of note is attached to the insn that
4947: produces or copies the final value. It tells the CSE pass how to
4948: think of that value.
4949:
4950: @item REG_RETVAL
4951: This insn copies the value of a library call, and @var{op} is the
4952: first insn that was generated to set up the arguments for the library
4953: call.
4954:
4955: Flow analysis uses this note to delete all of a library call whose
4956: result is dead.
4957:
4958: @item REG_WAS_0
4959: The register @var{op} contained zero before this insn. You can rely
4960: on this note if it is present; its absence implies nothing.
4961:
4962: @item REG_LIBCALL
4963: This is the inverse of @code{REG_RETVAL}: it is placed on the first
4964: insn of a library call, and it points to the last one.
4965:
4966: Loop optimization uses this note to move an entire library call out
4967: of a loop when its value is constant.
4968:
4969: @item REG_NONNEG
4970: The register @var{op} is known to have nonnegative value when this
4971: insn is reached.
4972: @end table
4973:
4974: (The only difference between the expression codes @samp{insn_list} and
4975: @samp{expr_list} is that the first operand of an @samp{insn_list} is
4976: assumed to be an insn and is printed in debugging dumps as the insn's
4977: unique id; the first operand of an @samp{expr_list} is printed in the
4978: ordinary way as an expression.)
4979:
4980: @node Calls, Sharing, Insns, RTL
4981: @section RTL Representation of Function-Call Insns
4982:
4983: Insns that call subroutines have the RTL expression code @samp{call_insn}.
4984: These insns must satisfy special rules, and their bodies must use a special
4985: RTL expression code, @samp{call}.
4986:
4987: A @samp{call} expression has two operands, as follows:
4988:
4989: @example
1.1.1.6 ! root 4990: (call (mem:@var{fm} @var{addr}) @var{nbytes})
1.1 root 4991: @end example
4992:
4993: @noindent
4994: Here @var{nbytes} is an operand that represents the number of bytes of
4995: argument data being passed to the subroutine, @var{fm} is a machine mode
4996: (which must equal as the definition of the @code{FUNCTION_MODE} macro in
4997: the machine description) and @var{addr} represents the address of the
4998: subroutine.
4999:
5000: For a subroutine that returns no value, the @samp{call} RTX as shown above
5001: is the entire body of the insn.
5002:
5003: For a subroutine that returns a value whose mode is not @code{BLKmode},
5004: the value is returned in a hard register. If this register's number is
5005: @var{r}, then the body of the call insn looks like this:
5006:
5007: @example
5008: (set (reg:@var{m} @var{r})
5009: (call @var{nbytes} (mem:@var{fm} @var{addr})))
5010: @end example
5011:
5012: @noindent
5013: This RTL expression makes it clear (to the optimizer passes) that the
5014: appropriate register receives a useful value in this insn.
5015:
5016: Immediately after RTL generation, if the value of the subroutine is
5017: actually used, this call insn is always followed closely by an insn which
5018: refers to the register @var{r}. This remains true through all the
5019: optimizer passes until cross jumping occurs.
5020:
5021: The following insn has one of two forms. Either it copies the value into a
5022: pseudo-register, like this:
5023:
5024: @example
5025: (set (reg:@var{m} @var{p}) (reg:@var{m} @var{r}))
5026: @end example
5027:
5028: @noindent
5029: or (in the case where the calling function will simply return whatever
5030: value the call produced, and no operation is needed to do this):
5031:
5032: @example
5033: (use (reg:@var{m} @var{r}))
5034: @end example
5035:
5036: @noindent
5037: Between the call insn and this following insn there may intervene only a
5038: stack-adjustment insn (and perhaps some @samp{note} insns).
5039:
5040: When a subroutine returns a @code{BLKmode} value, it is handled by
5041: passing to the subroutine the address of a place to store the value.
5042: So the call insn itself does not ``return'' any value, and it has the
5043: same RTL form as a call that returns nothing.
5044:
5045: @node Sharing,, Calls, RTL
5046: @section Structure Sharing Assumptions
5047:
5048: The compiler assumes that certain kinds of RTL expressions are unique;
5049: there do not exist two distinct objects representing the same value.
5050: In other cases, it makes an opposite assumption: that no RTL expression
5051: object of a certain kind appears in more than one place in the
5052: containing structure.
5053:
5054: These assumptions refer to a single function; except for the RTL
5055: objects that describe global variables and external functions,
5056: no RTL objects are common to two functions.
5057:
5058: @itemize @bullet
5059: @item
5060: Each pseudo-register has only a single @samp{reg} object to represent it,
5061: and therefore only a single machine mode.
5062:
5063: @item
5064: For any symbolic label, there is only one @samp{symbol_ref} object
5065: referring to it.
5066:
5067: @item
5068: There is only one @samp{const_int} expression with value zero,
5069: and only one with value one.
5070:
5071: @item
5072: There is only one @samp{pc} expression.
5073:
5074: @item
5075: There is only one @samp{cc0} expression.
5076:
5077: @item
5078: There is only one @samp{const_double} expression with mode
5079: @code{SFmode} and value zero, and only one with mode @code{DFmode} and
5080: value zero.
5081:
5082: @item
5083: No @samp{label_ref} appears in more than one place in the RTL
5084: structure; in other words, it is safe to do a tree-walk of all the
5085: insns in the function and assume that each time a @samp{label_ref} is
5086: seen it is distinct from all others that are seen.
5087:
5088: @item
5089: Only one @samp{mem} object is normally created for each static
5090: variable or stack slot, so these objects are frequently shared in all
5091: the places they appear. However, separate but equal objects for these
5092: variables are occasionally made.
5093:
5094: @item
1.1.1.5 root 5095: When a single @code{asm} statement has multiple output operands,
5096: a distinct @code{asm_operands} RTX is made for each output operand.
5097: However, these all share the vector which contains the sequence of
5098: input operands. Because this sharing is used later on to test whether
5099: two @code{asm_operands} RTX's come from the same statement, the sharing
5100: must be guaranteed to be preserved.
5101:
5102: @item
1.1 root 5103: No RTL object appears in more than one place in the RTL structure
5104: except as described above. Many passes of the compiler rely on this
5105: by assuming that they can modify RTL objects in place without unwanted
5106: side-effects on other insns.
5107:
5108: @item
5109: During initial RTL generation, shared structure is freely introduced.
5110: After all the RTL for a function has been generated, all shared
5111: structure is copied by @code{unshare_all_rtl} in @file{emit-rtl.c},
5112: after which the above rules are guaranteed to be followed.
5113:
5114: @item
5115: During the combiner pass, shared structure with an insn can exist
5116: temporarily. However, the shared structure is copied before the
5117: combiner is finished with the insn. This is done by
5118: @code{copy_substitutions} in @samp{combine.c}.
5119: @end itemize
5120:
5121: @node Machine Desc, Machine Macros, RTL, Top
5122: @chapter Machine Descriptions
5123:
5124: A machine description has two parts: a file of instruction patterns
5125: (@file{.md} file) and a C header file of macro definitions.
5126:
5127: The @file{.md} file for a target machine contains a pattern for each
5128: instruction that the target machine supports (or at least each instruction
5129: that is worth telling the compiler about). It may also contain comments.
5130: A semicolon causes the rest of the line to be a comment, unless the semicolon
5131: is inside a quoted string.
5132:
5133: See the next chapter for information on the C header file.
5134:
5135: @menu
5136: * Patterns:: How to write instruction patterns.
5137: * Example:: An explained example of a @samp{define_insn} pattern.
5138: * RTL Template:: The RTL template defines what insns match a pattern.
5139: * Output Template:: The output template says how to make assembler code
5140: from such an insn.
5141: * Output Statement:: For more generality, write C code to output
5142: the assembler code.
5143: * Constraints:: When not all operands are general operands.
5144: * Standard Names:: Names mark patterns to use for code generation.
5145: * Pattern Ordering:: When the order of patterns makes a difference.
5146: * Dependent Patterns:: Having one pattern may make you need another.
5147: * Jump Patterns:: Special considerations for patterns for jump insns.
5148: * Peephole Definitions::Defining machine-specific peephole optimizations.
5149: * Expander Definitions::Generating a sequence of several RTL insns
5150: for a standard operation.
5151: @end menu
5152:
5153: @node Patterns, Example, Machine Desc, Machine Desc
5154: @section Everything about Instruction Patterns
5155:
5156: Each instruction pattern contains an incomplete RTL expression, with pieces
5157: to be filled in later, operand constraints that restrict how the pieces can
5158: be filled in, and an output pattern or C code to generate the assembler
5159: output, all wrapped up in a @samp{define_insn} expression.
5160:
5161: A @samp{define_insn} is an RTL expression containing four or five operands:
5162:
5163: @enumerate
5164: @item
5165: An optional name. The presence of a name indicate that this instruction
5166: pattern can perform a certain standard job for the RTL-generation
5167: pass of the compiler. This pass knows certain names and will use
5168: the instruction patterns with those names, if the names are defined
5169: in the machine description.
5170:
5171: The absence of a name is indicated by writing an empty string
5172: where the name should go. Nameless instruction patterns are never
5173: used for generating RTL code, but they may permit several simpler insns
5174: to be combined later on.
5175:
5176: Names that are not thus known and used in RTL-generation have no
5177: effect; they are equivalent to no name at all.
5178:
5179: @item
5180: The @dfn{RTL template} (@pxref{RTL Template}) is a vector of
5181: incomplete RTL expressions which show what the instruction should look
5182: like. It is incomplete because it may contain @samp{match_operand}
5183: and @samp{match_dup} expressions that stand for operands of the
5184: instruction.
5185:
5186: If the vector has only one element, that element is what the
5187: instruction should look like. If the vector has multiple elements,
5188: then the instruction looks like a @samp{parallel} expression
5189: containing that many elements as described.
5190:
5191: @item
5192: A condition. This is a string which contains a C expression that is
5193: the final test to decide whether an insn body matches this pattern.
5194:
5195: For a named pattern, the condition (if present) may not depend on
5196: the data in the insn being matched, but only the target-machine-type
5197: flags. The compiler needs to test these conditions during
5198: initialization in order to learn exactly which named instructions are
5199: available in a particular run.
5200:
5201: For nameless patterns, the condition is applied only when matching an
5202: individual insn, and only after the insn has matched the pattern's
5203: recognition template. The insn's operands may be found in the vector
5204: @code{operands}.
5205:
5206: @item
5207: The @dfn{output template}: a string that says how to output matching
5208: insns as assembler code. @samp{%} in this string specifies where
5209: to substitute the value of an operand. @xref{Output Template}.
5210:
5211: When simple substitution isn't general enough, you can specify a piece
5212: of C code to compute the output. @xref{Output Statement}.
5213:
5214: @item
5215: Optionally, some @dfn{machine-specific information}. The meaning
5216: of this information is defined only by an individual machine description;
5217: typically it might say whether this insn alters the condition codes,
5218: or how many bytes of output it generates.
5219:
5220: This operand is written as a string containing a C initializer
5221: (complete with braces) for the structure type @code{INSN_MACHINE_INFO},
5222: whose definition is up to you (@pxref{Misc}).
5223: @end enumerate
5224:
5225: @node Example, RTL Template, Patterns, Machine Desc
5226: @section Example of @samp{define_insn}
5227:
5228: Here is an actual example of an instruction pattern, for the 68000/68020.
5229:
5230: @example
5231: (define_insn "tstsi"
5232: [(set (cc0)
5233: (match_operand:SI 0 "general_operand" "rm"))]
5234: ""
5235: "*
5236: @{ if (TARGET_68020 || ! ADDRESS_REG_P (operands[0]))
5237: return \"tstl %0\";
5238: return \"cmpl #0,%0\"; @}")
5239: @end example
5240:
5241: This is an instruction that sets the condition codes based on the value of
5242: a general operand. It has no condition, so any insn whose RTL description
5243: has the form shown may be handled according to this pattern. The name
5244: @samp{tstsi} means ``test a @code{SImode} value'' and tells the RTL generation
5245: pass that, when it is necessary to test such a value, an insn to do so
5246: can be constructed using this pattern.
5247:
5248: The output control string is a piece of C code which chooses which
5249: output template to return based on the kind of operand and the specific
5250: type of CPU for which code is being generated.
5251:
5252: @samp{"rm"} is an operand constraint. Its meaning is explained below.
5253:
5254: @node RTL Template, Output Template, Example, Machine Desc
5255: @section RTL Template for Generating and Recognizing Insns
5256:
5257: The RTL template is used to define which insns match the particular pattern
5258: and how to find their operands. For named patterns, the RTL template also
5259: says how to construct an insn from specified operands.
5260:
5261: Construction involves substituting specified operands into a copy of the
5262: template. Matching involves determining the values that serve as the
5263: operands in the insn being matched. Both of these activities are
5264: controlled by special expression types that direct matching and
5265: substitution of the operands.
5266:
5267: @table @code
5268: @item (match_operand:@var{m} @var{n} @var{testfn} @var{constraint})
5269: This expression is a placeholder for operand number @var{n} of
5270: the insn. When constructing an insn, operand number @var{n}
5271: will be substituted at this point. When matching an insn, whatever
5272: appears at this position in the insn will be taken as operand
5273: number @var{n}; but it must satisfy @var{testfn} or this instruction
5274: pattern will not match at all.
5275:
5276: Operand numbers must be chosen consecutively counting from zero in
5277: each instruction pattern. There may be only one @samp{match_operand}
5278: expression in the pattern for each operand number. Usually operands
5279: are numbered in the order of appearance in @samp{match_operand}
5280: expressions.
5281:
5282: @var{testfn} is a string that is the name of a C function that accepts
5283: two arguments, a machine mode and an expression. During matching,
5284: the function will be called with @var{m} as the mode argument
5285: and the putative operand as the other argument. If it returns zero,
5286: this instruction pattern fails to match. @var{testfn} may be
5287: an empty string; then it means no test is to be done on the operand.
5288:
5289: @var{constraint} is explained later (@pxref{Constraints}).
5290:
5291: Most often, @var{testfn} is @code{"general_operand"}. It checks
5292: that the putative operand is either a constant, a register or a
5293: memory reference, and that it is valid for mode @var{m}.
5294:
5295: For an operand that must be a register, @var{testfn} should be
5296: @code{"register_operand"}. It would be valid to use
5297: @code{"general_operand"}, since the reload pass would copy any
5298: non-register operands through registers, but this would make GNU CC do
5299: extra work, and it would prevent the register allocator from doing the
5300: best possible job.
5301:
5302: For an operand that must be a constant, either @var{testfn} should be
5303: @code{"immediate_operand"}, or the instruction pattern's extra
5304: condition should check for constants, or both. You cannot expect the
5305: constraints to do this work! If the constraints allow only constants,
5306: but the predicate allows something else, the compiler will crash when
5307: that case arises.
5308:
5309: @item (match_dup @var{n})
5310: This expression is also a placeholder for operand number @var{n}.
5311: It is used when the operand needs to appear more than once in the
5312: insn.
5313:
5314: In construction, @samp{match_dup} behaves exactly like
5315: @samp{match_operand}: the operand is substituted into the insn being
5316: constructed. But in matching, @samp{match_dup} behaves differently.
5317: It assumes that operand number @var{n} has already been determined by
5318: a @samp{match_operand} appearing earlier in the recognition template,
5319: and it matches only an identical-looking expression.
5320:
1.1.1.4 root 5321: @item (match_operator:@var{m} @var{n} "@var{predicate}" [@var{operands}@dots{}])
5322: This pattern is a kind of placeholder for a variable RTL expression
5323: code.
5324:
5325: When constructing an insn, it stands for an RTL expression whose
5326: expression code is taken from that of operand @var{n}, and whose
5327: operands are constructed from the patterns @var{operands}.
5328:
5329: When matching an expression, it matches an expression if the function
5330: @var{predicate} returns nonzero on that expression @emph{and} the
5331: patterns @var{operands} match the operands of the expression.
5332:
5333: Suppose that the function @code{commutative_operator} is defined as
5334: follows, to match any expression whose operator is one of the six
5335: commutative arithmetic operators of RTL and whose mode is @var{mode}:
5336:
5337: @example
5338: int
5339: commutative_operator (x, mode)
5340: rtx x;
5341: enum machine_mode mode;
5342: @{
5343: enum rtx_code code = GET_CODE (x);
5344: if (GET_MODE (x) != mode)
5345: return 0;
5346: return (code == PLUS || code == MULT || code == UMULT
5347: || code == AND || code == IOR || code == XOR);
5348: @}
5349: @end example
5350:
5351: Then the following pattern will match any RTL expression consisting
5352: of a commutative operator applied to two general operands:
5353:
5354: @example
5355: (match_operator:SI 2 "commutative_operator"
5356: [(match_operand:SI 3 "general_operand" "g")
5357: (match_operand:SI 4 "general_operand" "g")])
5358: @end example
5359:
5360: Here the vector @code{[@var{operands}@dots{}]} contains two patterns
5361: because the expressions to be matched all contain two operands.
5362:
5363: When this pattern does match, the two operands of the commutative
5364: operator are recorded as operands 3 and 4 of the insn. (This is done
1.1.1.5 root 5365: by the two instances of @samp{match_operand}.) Operand 2 of the insn
1.1.1.4 root 5366: will be the entire commutative expression: use @code{GET_CODE
5367: (operands[2])} to see which commutative operator was used.
5368:
5369: The machine mode @var{m} of @samp{match_operator} works like that of
5370: @samp{match_operand}: it is passed as the second argument to the
5371: predicate function, and that function is solely responsible for
5372: deciding whether the expression to be matched ``has'' that mode.
5373:
5374: When constructing an insn, argument 2 of the gen-function will specify
5375: the operation (i.e. the expression code) for the expression to be
5376: made. It should be an RTL expression, whose expression code is copied
5377: into a new expression whose operands are arguments 3 and 4 of the
5378: gen-function. The subexpressions of argument 2 are not used;
5379: only its expression code matters.
5380:
5381: There is no way to specify constraints in @samp{match_operator}. The
5382: operand of the insn which corresponds to the @samp{match_operator}
5383: never has any constraints because it is never reloaded as a whole.
5384: However, if parts of its @var{operands} are matched by
5385: @samp{match_operand} patterns, those parts may have constraints of
5386: their own.
5387:
1.1 root 5388: @item (address (match_operand:@var{m} @var{n} "address_operand" ""))
5389: This complex of expressions is a placeholder for an operand number
5390: @var{n} in a ``load address'' instruction: an operand which specifies
5391: a memory location in the usual way, but for which the actual operand
5392: value used is the address of the location, not the contents of the
5393: location.
5394:
5395: @samp{address} expressions never appear in RTL code, only in machine
5396: descriptions. And they are used only in machine descriptions that do
5397: not use the operand constraint feature. When operand constraints are
5398: in use, the letter @samp{p} in the constraint serves this purpose.
5399:
5400: @var{m} is the machine mode of the @emph{memory location being
5401: addressed}, not the machine mode of the address itself. That mode is
5402: always the same on a given target machine (it is @code{Pmode}, which
5403: normally is @code{SImode}), so there is no point in mentioning it;
5404: thus, no machine mode is written in the @samp{address} expression. If
5405: some day support is added for machines in which addresses of different
5406: kinds of objects appear differently or are used differently (such as
5407: the PDP-10), different formats would perhaps need different machine
5408: modes and these modes might be written in the @samp{address}
5409: expression.
5410: @end table
5411:
5412: @node Output Template, Output Statement, RTL Template, Machine Desc
5413: @section Output Templates and Operand Substitution
5414:
1.1.1.6 ! root 5415: The @dfn{output template} is a string which specifies how to output the
! 5416: assembler code for an instruction pattern. Most of the template is a
! 5417: fixed string which is output literally. The character @samp{%} is used
! 5418: to specify where to substitute an operand; it can also be used to
! 5419: identify places where different variants of the assembler require
1.1 root 5420: different syntax.
5421:
5422: In the simplest case, a @samp{%} followed by a digit @var{n} says to output
5423: operand @var{n} at that point in the string.
5424:
5425: @samp{%} followed by a letter and a digit says to output an operand in an
5426: alternate fashion. Four letters have standard, built-in meanings described
5427: below. The machine description macro @code{PRINT_OPERAND} can define
5428: additional letters with nonstandard meanings.
5429:
5430: @samp{%c@var{digit}} can be used to substitute an operand that is a
5431: constant value without the syntax that normally indicates an immediate
5432: operand.
5433:
5434: @samp{%n@var{digit}} is like @samp{%c@var{digit}} except that the value of
5435: the constant is negated before printing.
5436:
5437: @samp{%a@var{digit}} can be used to substitute an operand as if it were a
5438: memory reference, with the actual operand treated as the address. This may
5439: be useful when outputting a ``load address'' instruction, because often the
5440: assembler syntax for such an instruction requires you to write the operand
5441: as if it were a memory reference.
5442:
5443: @samp{%l@var{digit}} is used to substitute a @code{label_ref} into a jump
5444: instruction.
5445:
5446: @samp{%} followed by a punctuation character specifies a substitution that
5447: does not use an operand. Only one case is standard: @samp{%%} outputs a
5448: @samp{%} into the assembler code. Other nonstandard cases can be
5449: defined in the @code{PRINT_OPERAND} macro.
5450:
5451: The template may generate multiple assembler instructions. Write the text
5452: for the instructions, with @samp{\;} between them.
5453:
1.1.1.6 ! root 5454: When the RTL contains two operands which are required by constraint to match
1.1 root 5455: each other, the output template must refer only to the lower-numbered operand.
5456: Matching operands are not always identical, and the rest of the compiler
5457: arranges to put the proper RTL expression for printing into the lower-numbered
5458: operand.
5459:
5460: One use of nonstandard letters or punctuation following @samp{%} is to
5461: distinguish between different assembler languages for the same machine; for
5462: example, Motorola syntax versus MIT syntax for the 68000. Motorola syntax
5463: requires periods in most opcode names, while MIT syntax does not. For
5464: example, the opcode @samp{movel} in MIT syntax is @samp{move.l} in Motorola
5465: syntax. The same file of patterns is used for both kinds of output syntax,
5466: but the character sequence @samp{%.} is used in each place where Motorola
5467: syntax wants a period. The @code{PRINT_OPERAND} macro for Motorola syntax
5468: defines the sequence to output a period; the macro for MIT syntax defines
5469: it to do nothing.
5470:
5471: @node Output Statement, Constraints, Output Template, Machine Desc
5472: @section C Statements for Generating Assembler Output
5473:
5474: Often a single fixed template string cannot produce correct and efficient
5475: assembler code for all the cases that are recognized by a single
5476: instruction pattern. For example, the opcodes may depend on the kinds of
5477: operands; or some unfortunate combinations of operands may require extra
5478: machine instructions.
5479:
5480: If the output control string starts with a @samp{*}, then it is not an
5481: output template but rather a piece of C program that should compute a
5482: template. It should execute a @code{return} statement to return the
5483: template-string you want. Most such templates use C string literals, which
5484: require doublequote characters to delimit them. To include these
5485: doublequote characters in the string, prefix each one with @samp{\}.
5486:
5487: The operands may be found in the array @code{operands}, whose C data type
5488: is @code{rtx []}.
5489:
5490: It is possible to output an assembler instruction and then go on to output
5491: or compute more of them, using the subroutine @code{output_asm_insn}. This
5492: receives two arguments: a template-string and a vector of operands. The
5493: vector may be @code{operands}, or it may be another array of @code{rtx}
5494: that you declare locally and initialize yourself.
5495:
5496: When an insn pattern has multiple alternatives in its constraints, often
1.1.1.5 root 5497: the appearance of the assembler code is determined mostly by which alternative
1.1 root 5498: was matched. When this is so, the C code can test the variable
5499: @code{which_alternative}, which is the ordinal number of the alternative
5500: that was actually satisfied (0 for the first, 1 for the second alternative,
5501: etc.).
5502:
5503: For example, suppose there are two opcodes for storing zero, @samp{clrreg}
5504: for registers and @samp{clrmem} for memory locations. Here is how
5505: a pattern could use @code{which_alternative} to choose between them:
5506:
5507: @example
5508: (define_insn ""
5509: [(set (match_operand:SI 0 "general_operand" "r,m")
5510: (const_int 0))]
5511: ""
5512: "*
5513: return (which_alternative == 0
5514: ? \"clrreg %0\" : \"clrmem %0\");
5515: ")
5516: @end example
5517:
5518: @node Constraints, Standard Names, Output Statement, Machine Desc
5519: @section Operand Constraints
5520:
5521: Each @samp{match_operand} in an instruction pattern can specify a
5522: constraint for the type of operands allowed. Constraints can say whether
5523: an operand may be in a register, and which kinds of register; whether the
5524: operand can be a memory reference, and which kinds of address; whether the
5525: operand may be an immediate constant, and which possible values it may
5526: have. Constraints can also require two operands to match.
5527:
5528: @menu
5529: * Simple Constraints:: Basic use of constraints.
5530: * Multi-Alternative:: When an insn has two alternative constraint-patterns.
5531: * Class Preferences:: Constraints guide which hard register to put things in.
5532: * Modifiers:: More precise control over effects of constraints.
5533: * No Constraints:: Describing a clean machine without constraints.
5534: @end menu
5535:
5536: @node Simple Constraints, Multi-Alternative, Constraints, Constraints
5537: @subsection Simple Constraints
5538:
5539: The simplest kind of constraint is a string full of letters, each of
5540: which describes one kind of operand that is permitted. Here are
5541: the letters that are allowed:
5542:
5543: @table @asis
5544: @item @samp{m}
5545: A memory operand is allowed, with any kind of address that the machine
5546: supports in general.
5547:
5548: @item @samp{o}
5549: A memory operand is allowed, but only if the address is
5550: @dfn{offsetable}. This means that adding a small integer (actually,
5551: the width in bytes of the operand, as determined by its machine mode)
5552: may be added to the address and the result is also a valid memory
5553: address.
5554:
5555: For example, an address which is constant is offsetable; so is an
5556: address that is the sum of a register and a constant (as long as a
5557: slightly larger constant is also within the range of address-offsets
5558: supported by the machine); but an autoincrement or autodecrement
5559: address is not offsetable. More complicated indirect/indexed
5560: addresses may or may not be offsetable depending on the other
5561: addressing modes that the machine supports.
5562:
5563: Note that in an output operand which can be matched by another
5564: operand, the constraint letter @samp{o} is valid only when accompanied
5565: by both @samp{<} (if the target machine has predecrement addressing)
5566: and @samp{>} (if the target machine has preincrement addressing).
5567:
5568: When the constraint letter @samp{o} is used, the reload pass may
5569: generate instructions which copy a nonoffsetable address into an index
5570: register. The idea is that the register can be used as a replacement
5571: offsetable address. But this method requires that there be patterns
5572: to copy any kind of address into a register. Auto-increment
5573: and auto-decrement addresses are an exception; there need not be an
5574: instruction that can copy such an address into a register, because
5575: reload handles these cases specially.
5576:
5577: Most older machine designs have ``load address'' instructions which do
5578: just what is needed here. Some RISC machines do not advertise such
5579: instructions, but the possible addresses on these machines are very
5580: limited, so it is easy to fake them.
5581:
5582: @item @samp{<}
5583: A memory operand with autodecrement addressing (either predecrement or
5584: postdecrement) is allowed.
5585:
5586: @item @samp{>}
5587: A memory operand with autoincrement addressing (either preincrement or
5588: postincrement) is allowed.
5589:
5590: @item @samp{r}
5591: A register operand is allowed provided that it is in a general
5592: register.
5593:
5594: @item @samp{d}, @samp{a}, @samp{f}, @dots{}
5595: Other letters can be defined in machine-dependent fashion to stand for
5596: particular classes of registers. @samp{d}, @samp{a} and @samp{f} are
5597: defined on the 68000/68020 to stand for data, address and floating
5598: point registers.
5599:
5600: @item @samp{i}
5601: An immediate integer operand (one with constant value) is allowed.
5602: This includes symbolic constants whose values will be known only at
5603: assembly time.
5604:
5605: @item @samp{n}
5606: An immediate integer operand with a known numeric value is allowed.
5607: Many systems cannot support assembly-time constants for operands less
5608: than a word wide. Constraints for these operands should use @samp{n}
5609: rather than @samp{i}.
5610:
5611: @item @samp{I}, @samp{J}, @samp{K}, @dots{}
5612: Other letters in the range @samp{I} through @samp{M} may be defined in
5613: a machine-dependent fashion to permit immediate integer operands with
5614: explicit integer values in specified ranges. For example, on the
5615: 68000, @samp{I} is defined to stand for the range of values 1 to 8.
5616: This is the range permitted as a shift count in the shift
5617: instructions.
5618:
5619: @item @samp{F}
5620: An immediate floating operand (expression code @samp{const_double}) is
5621: allowed.
5622:
5623: @item @samp{G}, @samp{H}
5624: @samp{G} and @samp{H} may be defined in a machine-dependent fashion to
5625: permit immediate floating operands in particular ranges of values.
5626:
5627: @item @samp{s}
5628: An immediate integer operand whose value is not an explicit integer is
5629: allowed.
5630:
5631: This might appear strange; if an insn allows a constant operand with a
5632: value not known at compile time, it certainly must allow any known
5633: value. So why use @samp{s} instead of @samp{i}? Sometimes it allows
5634: better code to be generated.
5635:
5636: For example, on the 68000 in a fullword instruction it is possible to
5637: use an immediate operand; but if the immediate value is between -32
5638: and 31, better code results from loading the value into a register and
5639: using the register. This is because the load into the register can be
5640: done with a @samp{moveq} instruction. We arrange for this to happen
5641: by defining the letter @samp{K} to mean ``any integer outside the
5642: range -32 to 31'', and then specifying @samp{Ks} in the operand
5643: constraints.
5644:
5645: @item @samp{g}
5646: Any register, memory or immediate integer operand is allowed, except for
5647: registers that are not general registers.
5648:
5649: @item @samp{@var{n}} (a digit)
5650: An operand that matches operand number @var{n} is allowed.
5651: If a digit is used together with letters, the digit should come last.
5652:
5653: This is called a @dfn{matching constraint} and what it really means is
5654: that the assembler has only a single operand that fills two roles
5655: considered separate in the RTL insn. For example, an add insn has two
5656: input operands and one output operand in the RTL, but on most machines
5657: an add instruction really has only two operands, one of them an
5658: input-output operand.
5659:
5660: Matching constraints work only in circumstances like that add insn.
5661: More precisely, the matching constraint must appear in an input-only
5662: operand and the operand that it matches must be an output-only operand
1.1.1.5 root 5663: with a lower number. Thus, operand @var{n} must have @samp{=} in its
5664: constraint.
1.1 root 5665:
5666: For operands to match in a particular case usually means that they
5667: are identical-looking RTL expressions. But in a few special cases
5668: specific kinds of dissimilarity are allowed. For example, @code{*x}
5669: as an input operand will match @code{*x++} as an output operand.
5670: For proper results in such cases, the output template should always
5671: use the output-operand's number when printing the operand.
5672:
5673: @item @samp{p}
5674: An operand that is a valid memory address is allowed. This is
5675: for ``load address'' and ``push address'' instructions.
5676:
5677: If @samp{p} is used in the constraint, the test-function in the
5678: @samp{match_operand} must be @code{address_operand}.
5679: @end table
5680:
5681: In order to have valid assembler code, each operand must satisfy
5682: its constraint. But a failure to do so does not prevent the pattern
5683: from applying to an insn. Instead, it directs the compiler to modify
5684: the code so that the constraint will be satisfied. Usually this is
5685: done by copying an operand into a register.
5686:
5687: Contrast, therefore, the two instruction patterns that follow:
5688:
5689: @example
5690: (define_insn ""
5691: [(set (match_operand:SI 0 "general_operand" "r")
5692: (plus:SI (match_dup 0)
5693: (match_operand:SI 1 "general_operand" "r")))]
5694: ""
5695: "@dots{}")
5696: @end example
5697:
5698: @noindent
5699: which has two operands, one of which must appear in two places, and
5700:
5701: @example
5702: (define_insn ""
5703: [(set (match_operand:SI 0 "general_operand" "r")
5704: (plus:SI (match_operand:SI 1 "general_operand" "0")
5705: (match_operand:SI 2 "general_operand" "r")))]
5706: ""
5707: "@dots{}")
5708: @end example
5709:
5710: @noindent
5711: which has three operands, two of which are required by a constraint to be
5712: identical. If we are considering an insn of the form
5713:
5714: @example
5715: (insn @var{n} @var{prev} @var{next}
5716: (set (reg:SI 3)
5717: (plus:SI (reg:SI 6) (reg:SI 109)))
5718: @dots{})
5719: @end example
5720:
5721: @noindent
5722: the first pattern would not apply at all, because this insn does not
5723: contain two identical subexpressions in the right place. The pattern would
5724: say, ``That does not look like an add instruction; try other patterns.''
5725: The second pattern would say, ``Yes, that's an add instruction, but there
5726: is something wrong with it.'' It would direct the reload pass of the
5727: compiler to generate additional insns to make the constraint true. The
5728: results might look like this:
5729:
5730: @example
5731: (insn @var{n2} @var{prev} @var{n}
5732: (set (reg:SI 3) (reg:SI 6))
5733: @dots{})
5734:
5735: (insn @var{n} @var{n2} @var{next}
5736: (set (reg:SI 3)
5737: (plus:SI (reg:SI 3) (reg:SI 109)))
5738: @dots{})
5739: @end example
5740:
5741: It is up to you to make sure that each operand, in each pattern, has
5742: constraints that can handle any RTL expression that could be present for
5743: that operand. (When multiple alternatives are in use, each pattern must,
5744: for each possible combination of operand expressions, have at least one
5745: alternative which can handle that combination of operands.) The
5746: constraints don't need to @emph{allow} any possible operand---when this is
5747: the case, they do not constrain---but they must at least point the way to
5748: reloading any possible operand so that it will fit.
5749:
5750: @itemize @bullet
5751: @item
5752: If the constraint accepts whatever operands the predicate permits,
5753: there is no problem: reloading is never necessary for this operand.
5754:
5755: For example, an operand whose constraints permit everything except
5756: registers is safe provided its predicate rejects registers.
5757:
5758: An operand whose predicate accepts only constant values is safe
5759: provided its constraints include the letter @samp{i}. If any possible
5760: constant value is accepted, then nothing less than @samp{i} will do;
1.1.1.5 root 5761: if the predicate is more selective, then the constraints may also be
1.1 root 5762: more selective.
5763:
5764: @item
5765: Any operand expression can be reloaded by copying it into a register.
5766: So if an operand's constraints allow some kind of register, it is
5767: certain to be safe. It need not permit all classes of registers; the
5768: compiler knows how to copy a register into another register of the
5769: proper class in order to make an instruction valid.
5770:
5771: @item
5772: A nonoffsetable memory reference can be reloaded by copying the
5773: address into a register. So if the constraint uses the letter
5774: @samp{o}, all memory references are taken care of.
5775:
5776: @item
5777: A constant operand can be reloaded by storing it in memory; it then
5778: becomes an offsetable memory reference. So if the constraint uses the
5779: letters @samp{o} or @samp{m}, constant operands are not a problem.
5780: @end itemize
5781:
5782: If the operand's predicate can recognize registers, but the constraint does
5783: not permit them, it can make the compiler crash. When this operand happens
5784: to be a register, the reload pass will be stymied, because it does not know
5785: how to copy a register temporarily into memory.
5786:
5787: @node Multi-Alternative, Class Preferences, Simple Constraints, Constraints
5788: @subsection Multiple Alternative Constraints
5789:
5790: Sometimes a single instruction has multiple alternative sets of possible
5791: operands. For example, on the 68000, a logical-or instruction can combine
5792: register or an immediate value into memory, or it can combine any kind of
5793: operand into a register; but it cannot combine one memory location into
5794: another.
5795:
5796: These constraints are represented as multiple alternatives. An alternative
5797: can be described by a series of letters for each operand. The overall
5798: constraint for an operand is made from the letters for this operand
5799: from the first alternative, a comma, the letters for this operand from
5800: the second alternative, a comma, and so on until the last alternative.
5801: Here is how it is done for fullword logical-or on the 68000:
5802:
5803: @example
5804: (define_insn "iorsi3"
5805: [(set (match_operand:SI 0 "general_operand" "=%m,d")
5806: (ior:SI (match_operand:SI 1 "general_operand" "0,0")
5807: (match_operand:SI 2 "general_operand" "dKs,dmKs")))]
5808: @dots{})
5809: @end example
5810:
5811: The first alternative has @samp{m} (memory) for operand 0, @samp{0} for
5812: operand 1 (meaning it must match operand 0), and @samp{dKs} for operand 2.
5813: The second alternative has @samp{d} (data register) for operand 0, @samp{0}
5814: for operand 1, and @samp{dmKs} for operand 2. The @samp{=} and @samp{%} in
5815: the constraint for operand 0 are not part of any alternative; their meaning
5816: is explained in the next section.
5817:
5818: If all the operands fit any one alternative, the instruction is valid.
5819: Otherwise, for each alternative, the compiler counts how many instructions
5820: must be added to copy the operands so that that alternative applies.
5821: The alternative requiring the least copying is chosen. If two alternatives
5822: need the same amount of copying, the one that comes first is chosen.
5823: These choices can be altered with the @samp{?} and @samp{!} characters:
5824:
5825: @table @samp
5826: @item ?
5827: Disparage slightly the alternative that the @samp{?} appears in,
5828: as a choice when no alternative applies exactly. The compiler regards
5829: this alternative as one unit more costly for each @samp{?} that appears
5830: in it.
5831:
5832: @item !
5833: Disparage severely the alternative that the @samp{!} appears in.
5834: When operands must be copied into registers, the compiler will
5835: never choose this alternative as the one to strive for.
5836: @end table
5837:
1.1.1.5 root 5838: When an insn pattern has multiple alternatives in its constraints, often
5839: the appearance of the assembler code is determined mostly by which
1.1 root 5840: alternative was matched. When this is so, the C code for writing the
5841: assembler code can use the variable @code{which_alternative}, which is
1.1.1.5 root 5842: the ordinal number of the alternative that was actually satisfied (0 for
5843: the first, 1 for the second alternative, etc.). For example:
1.1 root 5844:
5845: @example
5846: (define_insn ""
5847: [(set (match_operand:SI 0 "general_operand" "r,m")
5848: (const_int 0))]
5849: ""
5850: "*
5851: return (which_alternative == 0
5852: ? \"clrreg %0\" : \"clrmem %0\");
5853: ")
5854: @end example
5855:
5856: @node Class Preferences, Modifiers, Multi-Alternative, Constraints
5857: @subsection Register Class Preferences
5858:
5859: The operand constraints have another function: they enable the compiler
5860: to decide which kind of hardware register a pseudo register is best
5861: allocated to. The compiler examines the constraints that apply to the
5862: insns that use the pseudo register, looking for the machine-dependent
5863: letters such as @samp{d} and @samp{a} that specify classes of registers.
5864: The pseudo register is put in whichever class gets the most ``votes''.
5865: The constraint letters @samp{g} and @samp{r} also vote: they vote in
5866: favor of a general register. The machine description says which registers
5867: are considered general.
5868:
5869: Of course, on some machines all registers are equivalent, and no register
5870: classes are defined. Then none of this complexity is relevant.
5871:
5872: @node Modifiers, No Constraints, Class Preferences, Constraints
5873: @subsection Constraint Modifier Characters
5874:
5875: @table @samp
5876: @item =
5877: Means that this operand is write-only for this instruction: the previous
5878: value is discarded and replaced by output data.
5879:
5880: @item +
5881: Means that this operand is both read and written by the instruction.
5882:
5883: When the compiler fixes up the operands to satisfy the constraints,
5884: it needs to know which operands are inputs to the instruction and
5885: which are outputs from it. @samp{=} identifies an output; @samp{+}
5886: identifies an operand that is both input and output; all other operands
5887: are assumed to be input only.
5888:
5889: @item &
5890: Means (in a particular alternative) that this operand is written
5891: before the instruction is finished using the input operands.
5892: Therefore, this operand may not lie in a register that is used as an
5893: input operand or as part of any memory address.
5894:
5895: @samp{&} applies only to the alternative in which it is written. In
5896: constraints with multiple alternatives, sometimes one alternative
5897: requires @samp{&} while others do not. See, for example, the
5898: @samp{movdf} insn of the 68000.
5899:
5900: @samp{&} does not obviate the need to write @samp{=}.
5901:
5902: @item %
5903: Declares the instruction to be commutative for this operand and the
5904: following operand. This means that the compiler may interchange the
5905: two operands if that is the cheapest way to make all operands fit the
5906: constraints. This is often used in patterns for addition instructions
5907: that really have only two operands: the result must go in one of the
5908: arguments. Here for example, is how the 68000 halfword-add
5909: instruction is defined:
5910:
5911: @example
5912: (define_insn "addhi3"
5913: [(set (match_operand:HI 0 "general_operand" "=m,r")
5914: (plus:HI (match_operand:HI 1 "general_operand" "%0,0")
5915: (match_operand:HI 2 "general_operand" "di,g")))]
5916: @dots{})
5917: @end example
5918:
5919: Note that in previous versions of GNU CC the @samp{%} constraint
5920: modifier always applied to operands 1 and 2 regardless of which
5921: operand it was written in. The usual custom was to write it in
5922: operand 0. Now it must be in operand 1 if the operands to be
5923: exchanged are 1 and 2.
5924:
5925: @item #
5926: Says that all following characters, up to the next comma, are to be
5927: ignored as a constraint. They are significant only for choosing
5928: register preferences.
5929:
5930: @item *
5931: Says that the following character should be ignored when choosing
5932: register preferences. @samp{*} has no effect on the meaning of the
5933: constraint as a constraint.
5934:
5935: Here is an example: the 68000 has an instruction to sign-extend a
5936: halfword in a data register, and can also sign-extend a value by
5937: copying it into an address register. While either kind of register is
5938: acceptable, the constraints on an address-register destination are
5939: less strict, so it is best if register allocation makes an address
5940: register its goal. Therefore, @samp{*} is used so that the @samp{d}
5941: constraint letter (for data register) is ignored when computing
5942: register preferences.
5943:
5944: @example
5945: (define_insn "extendhisi2"
5946: [(set (match_operand:SI 0 "general_operand" "=*d,a")
5947: (sign_extend:SI
5948: (match_operand:HI 1 "general_operand" "0,g")))]
5949: @dots{})
5950: @end example
5951: @end table
5952:
5953: @node No Constraints,, Modifiers, Constraints
5954: @subsection Not Using Constraints
5955:
5956: Some machines are so clean that operand constraints are not required. For
5957: example, on the Vax, an operand valid in one context is valid in any other
5958: context. On such a machine, every operand constraint would be @samp{g},
5959: excepting only operands of ``load address'' instructions which are
5960: written as if they referred to a memory location's contents but actual
5961: refer to its address. They would have constraint @samp{p}.
5962:
5963: For such machines, instead of writing @samp{g} and @samp{p} for all
5964: the constraints, you can choose to write a description with empty constraints.
5965: Then you write @samp{""} for the constraint in every @samp{match_operand}.
5966: Address operands are identified by writing an @samp{address} expression
5967: around the @samp{match_operand}, not by their constraints.
5968:
5969: When the machine description has just empty constraints, certain parts
1.1.1.6 ! root 5970: of compilation are skipped, making the compiler faster. However,
! 5971: few machines actually do not need constraints; all machine descriptions
! 5972: now in existence use constraints.
1.1 root 5973:
5974: @node Standard Names, Pattern Ordering, Constraints, Machine Desc
5975: @section Standard Names for Patterns Used in Generation
5976:
5977: Here is a table of the instruction names that are meaningful in the RTL
5978: generation pass of the compiler. Giving one of these names to an
5979: instruction pattern tells the RTL generation pass that it can use the
5980: pattern in to accomplish a certain task.
5981:
5982: @table @asis
5983: @item @samp{mov@var{m}}
5984: Here @var{m} is a two-letter machine mode name, in lower case. This
5985: instruction pattern moves data with that machine mode from operand 1 to
5986: operand 0. For example, @samp{movsi} moves full-word data.
5987:
5988: If operand 0 is a @samp{subreg} with mode @var{m} of a register whose
5989: natural mode is wider than @var{m}, the effect of this instruction is
5990: to store the specified value in the part of the register that corresponds
5991: to mode @var{m}. The effect on the rest of the register is undefined.
5992:
5993: This class of patterns is special in several ways. First of all, each
5994: of these names @emph{must} be defined, because there is no other way
5995: to copy a datum from one place to another.
5996:
5997: Second, these patterns are not used solely in the RTL generation pass.
5998: Even the reload pass can generate move insns to copy values from stack
5999: slots into temporary registers. When it does so, one of the operands
6000: is a hard register and the other is an operand that can have a reload.
6001:
6002: Therefore, when given such a pair of operands, the pattern must
6003: generate RTL which needs no temporary registers---no registers other
6004: than the operands. For example, if you support the pattern with a
6005: @code{define_expand}, then in such a case you mustn't call
6006: @code{force_reg} or any other such function which might generate new
6007: pseudo registers.
6008:
6009: This requirement exists even for subword modes on a RISC machine where
6010: fetching those modes from memory normally requires several insns and
6011: some temporary registers. Look in @file{spur.md} to see how the
6012: requirement is satisfied.
6013:
6014: The variety of operands that have reloads depends on the rest of the
6015: machine description, but typically on a RISC machine these can only be
6016: pseudo registers that did not get hard registers, while on other
6017: machines explicit memory references will get optional reloads.
6018:
6019: In addition, the constraints must allow any hard register to be moved
6020: to any other hard register (provided that @code{HARD_REGNO_MODE_OK}
6021: permits mode @var{m} in each of the registers).
6022:
6023: @item @samp{movstrict@var{m}}
6024: Like @samp{mov@var{m}} except that if operand 0 is a @samp{subreg}
6025: with mode @var{m} of a register whose natural mode is wider,
6026: the @samp{movstrict@var{m}} instruction is guaranteed not to alter
6027: any of the register except the part which belongs to mode @var{m}.
6028:
6029: @item @samp{add@var{m}3}
6030: Add operand 2 and operand 1, storing the result in operand 0. All operands
6031: must have mode @var{m}. This can be used even on two-address machines, by
6032: means of constraints requiring operands 1 and 0 to be the same location.
6033:
6034: @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}
6035: Similar, for other arithmetic operations.
6036:
6037: There are special considerations for register classes for logical-and
6038: instructions, affecting also the macro @code{PREFERRED_RELOAD_CLASS}.
6039: They apply not only to the patterns with these standard names, but to
6040: any patterns that will match such an instruction. @xref{Register
6041: Classes}.
6042:
6043: @item @samp{mulhisi3}
6044: Multiply operands 1 and 2, which have mode @code{HImode}, and store
6045: a @code{SImode} product in operand 0.
6046:
6047: @item @samp{mulqihi3}, @samp{mulsidi3}
6048: Similar widening-multiplication instructions of other widths.
6049:
6050: @item @samp{umulqihi3}, @samp{umulhisi3}, @samp{umulsidi3}
6051: Similar widening-multiplication instructions that do unsigned
6052: multiplication.
6053:
6054: @item @samp{divmod@var{m}4}
6055: Signed division that produces both a quotient and a remainder.
6056: Operand 1 is divided by operand 2 to produce a quotient stored
6057: in operand 0 and a remainder stored in operand 3.
6058:
6059: @item @samp{udivmod@var{m}4}
6060: Similar, but does unsigned division.
6061:
6062: @item @samp{divmod@var{m}@var{n}4}
6063: Like @samp{divmod@var{m}4} except that only the dividend has mode
6064: @var{m}; the divisor, quotient and remainder have mode @var{n}.
6065: For example, the Vax has a @samp{divmoddisi4} instruction
6066: (but it is omitted from the machine description, because it
6067: is so slow that it is faster to compute remainders by the
6068: circumlocution that the compiler will use if this instruction is
6069: not available).
6070:
6071: @item @samp{ashl@var{m}3}
6072: Arithmetic-shift operand 1 left by a number of bits specified by
6073: operand 2, and store the result in operand 0. Operand 2 has
6074: mode @code{SImode}, not mode @var{m}.
6075:
6076: @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}
6077: Other shift and rotate instructions.
6078:
6079: Logical and arithmetic left shift are the same. Machines that do not
6080: allow negative shift counts often have only one instruction for
6081: shifting left. On such machines, you should define a pattern named
6082: @samp{ashl@var{m}3} and leave @samp{lshl@var{m}3} undefined.
6083:
6084: There are special considerations for register classes for shift
6085: instructions, affecting also the macro @code{PREFERRED_RELOAD_CLASS}.
6086: They apply not only to the patterns with these standard names, but to
6087: any patterns that will match such an instruction. @xref{Register
6088: Classes}.
6089:
6090: @item @samp{neg@var{m}2}
6091: Negate operand 1 and store the result in operand 0.
6092:
6093: @item @samp{abs@var{m}2}
6094: Store the absolute value of operand 1 into operand 0.
6095:
6096: @item @samp{sqrt@var{m}2}
6097: Store the square root of operand 1 into operand 0.
6098:
6099: @item @samp{ffs@var{m}2}
6100: Store into operand 0 one plus the index of the least significant 1-bit
6101: of operand 1. If operand 1 is zero, store zero. @var{m} is the mode
6102: of operand 0; operand 1's mode is specified by the instruction
6103: pattern, and the compiler will convert the operand to that mode before
6104: generating the instruction.
6105:
6106: @item @samp{one_cmpl@var{m}2}
6107: Store the bitwise-complement of operand 1 into operand 0.
6108:
6109: @item @samp{cmp@var{m}}
6110: Compare operand 0 and operand 1, and set the condition codes.
6111: The RTL pattern should look like this:
6112:
6113: @example
1.1.1.6 ! root 6114: (set (cc0) (compare (match_operand:@var{m} 0 @dots{})
! 6115: (match_operand:@var{m} 1 @dots{})))
1.1 root 6116: @end example
6117:
6118: Each such definition in the machine description, for integer mode
6119: @var{m}, must have a corresponding @samp{tst@var{m}} pattern, because
6120: optimization can simplify the compare into a test when operand 1 is
6121: zero.
6122:
6123: @item @samp{tst@var{m}}
6124: Compare operand 0 against zero, and set the condition codes.
6125: The RTL pattern should look like this:
6126:
6127: @example
6128: (set (cc0) (match_operand:@var{m} 0 @dots{}))
6129: @end example
6130:
6131: @item @samp{movstr@var{m}}
6132: Block move instruction. The addresses of the destination and source
6133: strings are the first two operands, and both are in mode @code{Pmode}.
6134: The number of bytes to move is the third operand, in mode @var{m}.
1.1.1.5 root 6135: The fourth operand is the known shared alignment of the source and
6136: destination, in the form of a @code{const_int} rtx.
1.1 root 6137:
6138: @item @samp{cmpstr@var{m}}
6139: Block compare instruction, with operands like @samp{movstr@var{m}}
6140: except that the two memory blocks are compared byte by byte
6141: in lexicographic order. The effect of the instruction is to set
6142: the condition codes.
6143:
6144: @item @samp{float@var{m}@var{n}2}
6145: Convert operand 1 (valid for fixed point mode @var{m}) to floating
6146: point mode @var{n} and store in operand 0 (which has mode @var{n}).
6147:
6148: @item @samp{fix@var{m}@var{n}2}
6149: Convert operand 1 (valid for floating point mode @var{m}) to fixed
6150: point mode @var{n} as a signed number and store in operand 0 (which
6151: has mode @var{n}). This instruction's result is defined only when
6152: the value of operand 1 is an integer.
6153:
6154: @item @samp{fixuns@var{m}@var{n}2}
6155: Convert operand 1 (valid for floating point mode @var{m}) to fixed
6156: point mode @var{n} as an unsigned number and store in operand 0 (which
6157: has mode @var{n}). This instruction's result is defined only when the
6158: value of operand 1 is an integer.
6159:
6160: @item @samp{ftrunc@var{m}2}
6161: Convert operand 1 (valid for floating point mode @var{m}) to an
6162: integer value, still represented in floating point mode @var{m}, and
6163: store it in operand 0 (valid for floating point mode @var{m}).
6164:
6165: @item @samp{fix_trunc@var{m}@var{n}2}
6166: Like @samp{fix@var{m}@var{n}2} but works for any floating point value
6167: of mode @var{m} by converting the value to an integer.
6168:
6169: @item @samp{fixuns_trunc@var{m}@var{n}2}
6170: Like @samp{fixuns@var{m}@var{n}2} but works for any floating point
6171: value of mode @var{m} by converting the value to an integer.
6172:
6173: @item @samp{trunc@var{m}@var{n}}
6174: Truncate operand 1 (valid for mode @var{m}) to mode @var{n} and
6175: store in operand 0 (which has mode @var{n}). Both modes must be fixed
6176: point or both floating point.
6177:
6178: @item @samp{extend@var{m}@var{n}}
6179: Sign-extend operand 1 (valid for mode @var{m}) to mode @var{n} and
6180: store in operand 0 (which has mode @var{n}). Both modes must be fixed
6181: point or both floating point.
6182:
6183: @item @samp{zero_extend@var{m}@var{n}}
6184: Zero-extend operand 1 (valid for mode @var{m}) to mode @var{n} and
6185: store in operand 0 (which has mode @var{n}). Both modes must be fixed
6186: point.
6187:
6188: @item @samp{extv}
6189: Extract a bit-field from operand 1 (a register or memory operand),
6190: where operand 2 specifies the width in bits and operand 3 the starting
6191: bit, and store it in operand 0. Operand 0 must have @code{Simode}.
6192: Operand 1 may have mode @code{QImode} or @code{SImode}; often
6193: @code{SImode} is allowed only for registers. Operands 2 and 3 must be
6194: valid for @code{SImode}.
6195:
6196: The RTL generation pass generates this instruction only with constants
6197: for operands 2 and 3.
6198:
6199: The bit-field value is sign-extended to a full word integer
6200: before it is stored in operand 0.
6201:
6202: @item @samp{extzv}
6203: Like @samp{extv} except that the bit-field value is zero-extended.
6204:
6205: @item @samp{insv}
6206: Store operand 3 (which must be valid for @code{SImode}) into a
6207: bit-field in operand 0, where operand 1 specifies the width in bits
6208: and operand 2 the starting bit. Operand 0 may have mode @code{QImode}
6209: or @code{SImode}; often @code{SImode} is allowed only for registers.
6210: Operands 1 and 2 must be valid for @code{SImode}.
6211:
6212: The RTL generation pass generates this instruction only with constants
6213: for operands 1 and 2.
6214:
6215: @item @samp{s@var{cond}}
6216: Store zero or nonzero in the operand according to the condition codes.
6217: Value stored is nonzero iff the condition @var{cond} is true.
6218: @var{cond} is the name of a comparison operation expression code, such
6219: as @samp{eq}, @samp{lt} or @samp{leu}.
6220:
6221: You specify the mode that the operand must have when you write the
6222: @code{match_operand} expression. The compiler automatically sees
6223: which mode you have used and supplies an operand of that mode.
6224:
6225: The value stored for a true condition must have 1 as its low bit.
6226: Otherwise the instruction is not suitable and must be omitted from the
6227: machine description. You must tell the compiler exactly which value
6228: is stored by defining the macro @code{STORE_FLAG_VALUE}.
6229:
6230: @item @samp{b@var{cond}}
6231: Conditional branch instruction. Operand 0 is a @samp{label_ref}
6232: that refers to the label to jump to. Jump if the condition codes
6233: meet condition @var{cond}.
6234:
6235: @item @samp{call}
6236: Subroutine call instruction returning no value. Operand 0 is the
6237: function to call; operand 1 is the number of bytes of arguments pushed
6238: (in mode @code{SImode}, except it is normally a @samp{const_int});
6239: operand 2 is the number of registers used as operands.
6240:
6241: On most machines, operand 2 is not actually stored into the RTL
6242: pattern. It is supplied for the sake of some RISC machines which need
6243: to put this information into the assembler code; they can put it in
6244: the RTL instead of operand 1.
6245:
6246: Operand 0 should be a @samp{mem} RTX whose address is the address of
6247: the function.
6248:
6249: @item @samp{call_value}
6250: Subroutine call instruction returning a value. Operand 0 is the hard
6251: register in which the value is returned. There are three more
6252: operands, the same as the three operands of the @samp{call}
6253: instruction (but with numbers increased by one).
6254:
6255: Subroutines that return @code{BLKmode} objects use the @samp{call}
6256: insn.
6257:
6258: @item @samp{return}
6259: Subroutine return instruction. This instruction pattern name should be
6260: defined only if a single instruction can do all the work of returning
6261: from a function.
6262:
6263: @item @samp{casesi}
6264: Instruction to jump through a dispatch table, including bounds checking.
6265: This instruction takes five operands:
6266:
6267: @enumerate
6268: @item
6269: The index to dispatch on, which has mode @code{SImode}.
6270:
6271: @item
6272: The lower bound for indices in the table, an integer constant.
6273:
6274: @item
1.1.1.6 ! root 6275: The total range of indices in the table---the largest index
! 6276: minus the smallest one (both inclusive).
1.1 root 6277:
6278: @item
6279: A label to jump to if the index has a value outside the bounds.
6280: (If the machine-description macro @code{CASE_DROPS_THROUGH} is defined,
6281: then an out-of-bounds index drops through to the code following
6282: the jump table instead of jumping to this label. In that case,
6283: this label is not actually used by the @samp{casesi} instruction,
6284: but it is always provided as an operand.)
6285:
6286: @item
6287: A label that precedes the table itself.
6288: @end enumerate
6289:
6290: The table is a @samp{addr_vec} or @samp{addr_diff_vec} inside of a
6291: @samp{jump_insn}. The number of elements in the table is one plus the
6292: difference between the upper bound and the lower bound.
6293:
6294: @item @samp{tablejump}
6295: Instruction to jump to a variable address. This is a low-level
6296: capability which can be used to implement a dispatch table when there
6297: is no @samp{casesi} pattern.
6298:
6299: This pattern requires two operands: the address or offset, and a label
6300: which should immediately precede the jump table. If the macro
6301: @code{CASE_VECTOR_PC_RELATIVE} is defined then the first operand is an
6302: absolute address to jump to; otherwise, it is an offset which counts
6303: from the address of the table.
6304:
6305: The @samp{tablejump} insn is always the last insn before the jump
6306: table it uses. Its assembler code normally has no need to use the
6307: second operand, but you should incorporate it in the RTL pattern so
6308: that the jump optimizer will not delete the table as unreachable code.
6309: @end table
6310:
6311: @node Pattern Ordering, Dependent Patterns, Standard Names, Machine Desc
6312: @section When the Order of Patterns Matters
6313:
6314: Sometimes an insn can match more than one instruction pattern. Then the
6315: pattern that appears first in the machine description is the one used.
6316: Therefore, more specific patterns (patterns that will match fewer things)
6317: and faster instructions (those that will produce better code when they
6318: do match) should usually go first in the description.
6319:
6320: In some cases the effect of ordering the patterns can be used to hide
6321: a pattern when it is not valid. For example, the 68000 has an
6322: instruction for converting a fullword to floating point and another
6323: for converting a byte to floating point. An instruction converting
6324: an integer to floating point could match either one. We put the
6325: pattern to convert the fullword first to make sure that one will
6326: be used rather than the other. (Otherwise a large integer might
6327: be generated as a single-byte immediate quantity, which would not work.)
6328: Instead of using this pattern ordering it would be possible to make the
6329: pattern for convert-a-byte smart enough to deal properly with any
6330: constant value.
6331:
6332: @node Dependent Patterns, Jump Patterns, Pattern Ordering, Machine Desc
6333: @section Interdependence of Patterns
6334:
6335: Every machine description must have a named pattern for each of the
6336: conditional branch names @samp{b@var{cond}}. The recognition template
6337: must always have the form
6338:
6339: @example
6340: (set (pc)
6341: (if_then_else (@var{cond} (cc0) (const_int 0))
6342: (label_ref (match_operand 0 "" ""))
6343: (pc)))
6344: @end example
6345:
6346: @noindent
6347: In addition, every machine description must have an anonymous pattern
6348: for each of the possible reverse-conditional branches. These patterns
6349: look like
6350:
6351: @example
6352: (set (pc)
6353: (if_then_else (@var{cond} (cc0) (const_int 0))
6354: (pc)
6355: (label_ref (match_operand 0 "" ""))))
6356: @end example
6357:
6358: @noindent
6359: They are necessary because jump optimization can turn direct-conditional
6360: branches into reverse-conditional branches.
6361:
6362: The compiler does more with RTL than just create it from patterns
6363: and recognize the patterns: it can perform arithmetic expression codes
6364: when constant values for their operands can be determined. As a result,
6365: sometimes having one pattern can require other patterns. For example, the
6366: Vax has no `and' instruction, but it has `and not' instructions. Here
6367: is the definition of one of them:
6368:
6369: @example
6370: (define_insn "andcbsi2"
6371: [(set (match_operand:SI 0 "general_operand" "")
6372: (and:SI (match_dup 0)
6373: (not:SI (match_operand:SI
6374: 1 "general_operand" ""))))]
6375: ""
6376: "bicl2 %1,%0")
6377: @end example
6378:
6379: @noindent
6380: If operand 1 is an explicit integer constant, an instruction constructed
6381: using that pattern can be simplified into an `and' like this:
6382:
6383: @example
6384: (set (reg:SI 41)
6385: (and:SI (reg:SI 41)
6386: (const_int 0xffff7fff)))
6387: @end example
6388:
6389: @noindent
6390: (where the integer constant is the one's complement of what
6391: appeared in the original instruction).
6392:
6393: To avoid a fatal error, the compiler must have a pattern that recognizes
6394: such an instruction. Here is what is used:
6395:
6396: @example
6397: (define_insn ""
6398: [(set (match_operand:SI 0 "general_operand" "")
6399: (and:SI (match_dup 0)
6400: (match_operand:SI 1 "general_operand" "")))]
6401: "GET_CODE (operands[1]) == CONST_INT"
6402: "*
6403: @{ operands[1]
6404: = gen_rtx (CONST_INT, VOIDmode, ~INTVAL (operands[1]));
6405: return \"bicl2 %1,%0\";
6406: @}")
6407: @end example
6408:
6409: @noindent
6410: Whereas a pattern to match a general `and' instruction is impossible to
6411: support on the Vax, this pattern is possible because it matches only a
6412: constant second argument: a special case that can be output as an `and not'
6413: instruction.
6414:
6415: A ``compare'' instruction whose RTL looks like this:
6416:
6417: @example
1.1.1.6 ! root 6418: (set (cc0) (compare @var{operand} (const_int 0)))
1.1 root 6419: @end example
6420:
6421: @noindent
6422: may be simplified by optimization into a ``test'' like this:
6423:
6424: @example
6425: (set (cc0) @var{operand})
6426: @end example
6427:
6428: @noindent
6429: So in the machine description, each ``compare'' pattern for an integer
6430: mode must have a corresponding ``test'' pattern that will match the
6431: result of such simplification.
6432:
6433: In some cases machines support instructions identical except for the
6434: machine mode of one or more operands. For example, there may be
6435: ``sign-extend halfword'' and ``sign-extend byte'' instructions whose
6436: patterns are
6437:
6438: @example
6439: (set (match_operand:SI 0 @dots{})
6440: (extend:SI (match_operand:HI 1 @dots{})))
6441:
6442: (set (match_operand:SI 0 @dots{})
6443: (extend:SI (match_operand:QI 1 @dots{})))
6444: @end example
6445:
6446: @noindent
6447: Constant integers do not specify a machine mode, so an instruction to
6448: extend a constant value could match either pattern. The pattern it
6449: actually will match is the one that appears first in the file. For correct
6450: results, this must be the one for the widest possible mode (@code{HImode},
6451: here). If the pattern matches the @code{QImode} instruction, the results
6452: will be incorrect if the constant value does not actually fit that mode.
6453:
6454: Such instructions to extend constants are rarely generated because they are
6455: optimized away, but they do occasionally happen in nonoptimized
6456: compilations.
6457:
6458: When an instruction has the constraint letter @samp{o}, the reload
6459: pass may generate instructions which copy a nonoffsetable address into
6460: an index register. The idea is that the register can be used as a
6461: replacement offsetable address. In order for these generated
6462: instructions to work, there must be patterns to copy any kind of valid
6463: address into a register.
6464:
6465: Most older machine designs have ``load address'' instructions which do
6466: just what is needed here. Some RISC machines do not advertise such
6467: instructions, but the possible addresses on these machines are very
6468: limited, so it is easy to fake them.
6469:
6470: Auto-increment and auto-decrement addresses are an exception; there
6471: need not be an instruction that can copy such an address into a
6472: register, because reload handles these cases in a different manner.
6473:
6474: @node Jump Patterns, Peephole Definitions, Dependent Patterns, Machine Desc
6475: @section Defining Jump Instruction Patterns
6476:
6477: GNU CC assumes that the machine has a condition code. A comparison insn
6478: sets the condition code, recording the results of both signed and unsigned
6479: comparison of the given operands. A separate branch insn tests the
6480: condition code and branches or not according its value. The branch insns
6481: come in distinct signed and unsigned flavors. Many common machines, such
6482: as the Vax, the 68000 and the 32000, work this way.
6483:
6484: Some machines have distinct signed and unsigned compare instructions, and
6485: only one set of conditional branch instructions. The easiest way to handle
6486: these machines is to treat them just like the others until the final stage
6487: where assembly code is written. At this time, when outputting code for the
6488: compare instruction, peek ahead at the following branch using
6489: @code{NEXT_INSN (insn)}. (The variable @code{insn} refers to the insn
6490: being output, in the output-writing code in an instruction pattern.) If
6491: the RTL says that is an unsigned branch, output an unsigned compare;
6492: otherwise output a signed compare. When the branch itself is output, you
6493: can treat signed and unsigned branches identically.
6494:
6495: The reason you can do this is that GNU CC always generates a pair of
6496: consecutive RTL insns, one to set the condition code and one to test it,
6497: and keeps the pair inviolate until the end.
6498:
6499: To go with this technique, you must define the machine-description macro
6500: @code{NOTICE_UPDATE_CC} to do @code{CC_STATUS_INIT}; in other words, no
6501: compare instruction is superfluous.
6502:
6503: Some machines have compare-and-branch instructions and no condition code.
6504: A similar technique works for them. When it is time to ``output'' a
6505: compare instruction, record its operands in two static variables. When
6506: outputting the branch-on-condition-code instruction that follows, actually
6507: output a compare-and-branch instruction that uses the remembered operands.
6508:
6509: It also works to define patterns for compare-and-branch instructions.
6510: In optimizing compilation, the pair of compare and branch instructions
1.1.1.5 root 6511: will be combined according to these patterns. But this does not happen
1.1 root 6512: if optimization is not requested. So you must use one of the solutions
6513: above in addition to any special patterns you define.
6514:
6515: @node Peephole Definitions, Expander Definitions, Jump Patterns, Machine Desc
6516: @section Defining Machine-Specific Peephole Optimizers
6517:
6518: In addition to instruction patterns the @file{md} file may contain
6519: definitions of machine-specific peephole optimizations.
6520:
6521: The combiner does not notice certain peephole optimizations when the data
6522: flow in the program does not suggest that it should try them. For example,
6523: sometimes two consecutive insns related in purpose can be combined even
6524: though the second one does not appear to use a register computed in the
6525: first one. A machine-specific peephole optimizer can detect such
6526: opportunities.
6527:
6528: A definition looks like this:
6529:
6530: @example
6531: (define_peephole
6532: [@var{insn-pattern-1}
6533: @var{insn-pattern-2}
6534: @dots{}]
6535: "@var{condition}"
6536: "@var{template}"
6537: "@var{machine-specific info}")
6538: @end example
6539:
6540: @noindent
6541: The last string operand may be omitted if you are not using any
6542: machine-specific information in this machine description. If present,
6543: it must obey the same rules as in a @samp{define_insn}.
6544:
6545: In this skeleton, @var{insn-pattern-1} and so on are patterns to match
1.1.1.5 root 6546: consecutive insns. The optimization applies to a sequence of insns when
6547: @var{insn-pattern-1} matches the first one, @var{insn-pattern-2} matches
6548: the next, and so on.@refill
1.1 root 6549:
6550: @var{insn-pattern-1} and so on look @emph{almost} like the second operand
6551: of @code{define_insn}. There is one important difference: this pattern is
6552: an RTX, not a vector. If the @code{define_insn} pattern would be a vector
6553: of one element, the @var{insn-pattern} should be just that element, no
6554: vector. If the @code{define_insn} pattern would have multiple elements
6555: then the @var{insn-pattern} must place the vector inside an explicit
6556: @code{parallel} RTX.@refill
6557:
1.1.1.5 root 6558: The operands of the insns are matched with @code{match_operands} and
6559: @code{match_dup}, as usual. What is not usual is that the operand numbers
6560: apply to all the insn patterns in the definition. So, you can check for
6561: identical operands in two insns by using @code{match_operand} in one insn
6562: and @code{match_dup} in the other.
1.1 root 6563:
6564: The operand constraints used in @code{match_operand} patterns do not have
6565: any direct effect on the applicability of the optimization, but they will
6566: be validated afterward, so write constraints that are sure to fit whenever
6567: the optimization is applied. It is safe to use @code{"g"} for each
6568: operand.
6569:
1.1.1.5 root 6570: Once a sequence of insns matches the patterns, the @var{condition} is
6571: checked. This is a C expression which makes the final decision whether to
6572: perform the optimization (we do so if the expression is nonzero). If
1.1 root 6573: @var{condition} is omitted (in other words, the string is empty) then the
1.1.1.5 root 6574: optimization is applied to every sequence of insns that matches the
1.1 root 6575: patterns.
6576:
1.1.1.5 root 6577: The defined peephole optimizations are applied after register allocation
6578: is complete. Therefore, the peephole definition can check which
6579: operands have ended up in which kinds of registers, just by looking at
6580: the operands.
1.1 root 6581:
6582: The way to refer to the operands in @var{condition} is to write
6583: @code{operands[@var{i}]} for operand number @var{i} (as matched by
6584: @code{(match_operand @var{i} @dots{})}). Use the variable @code{insn} to
6585: refer to the last of the insns being matched; use @code{PREV_INSN} to find
6586: the preceding insns (but be careful to skip over any @samp{note} insns that
6587: intervene).@refill
6588:
6589: When optimizing computations with intermediate results, you can use
6590: @var{condition} to match only when the intermediate results are not used
6591: elsewhere. Use the C expression @code{dead_or_set_p (@var{insn},
6592: @var{op})}, where @var{insn} is the insn in which you expect the value to
6593: be used for the last time (from the value of @code{insn}, together with use
6594: of @code{PREV_INSN}), and @var{op} is the intermediate value (from
6595: @code{operands[@var{i}]}).@refill
6596:
1.1.1.5 root 6597: Applying the optimization means replacing the sequence of insns with one
6598: new insn. The @var{template} controls ultimate output of assembler code
6599: for this combined insn. It works exactly like the template of a
6600: @code{define_insn}. Operand numbers in this template are the same ones
6601: used in matching the original sequence of insns.
1.1 root 6602:
6603: The result of a defined peephole optimizer does not need to match any of
1.1.1.5 root 6604: the insn patterns in the machine description; it does not even have an
6605: opportunity to match them. The peephole optimizer definition itself serves
6606: as the insn pattern to control how the insn is output.
6607:
6608: Defined peephole optimizers are run as assembler code is being output,
6609: so the insns they produce are never combined or rearranged in any way.
1.1 root 6610:
6611: Here is an example, taken from the 68000 machine description:
6612:
6613: @example
6614: (define_peephole
6615: [(set (reg:SI 15) (plus:SI (reg:SI 15) (const_int 4)))
6616: (set (match_operand:DF 0 "register_operand" "f")
6617: (match_operand:DF 1 "register_operand" "ad"))]
6618: "FP_REG_P (operands[0]) && ! FP_REG_P (operands[1])"
6619: "*
6620: @{
6621: rtx xoperands[2];
6622: xoperands[1] = gen_rtx (REG, SImode, REGNO (operands[1]) + 1);
6623: #ifdef MOTOROLA
6624: output_asm_insn (\"move.l %1,(sp)\", xoperands);
6625: output_asm_insn (\"move.l %1,-(sp)\", operands);
6626: return \"fmove.d (sp)+,%0\";
6627: #else
6628: output_asm_insn (\"movel %1,sp@@\", xoperands);
6629: output_asm_insn (\"movel %1,sp@@-\", operands);
6630: return \"fmoved sp@@+,%0\";
6631: #endif
6632: @}
6633: ")
6634: @end example
6635:
6636: The effect of this optimization is to change
6637:
6638: @example
6639: jbsr _foobar
6640: addql #4,sp
6641: movel d1,sp@@-
6642: movel d0,sp@@-
6643: fmoved sp@@+,fp0
6644: @end example
6645:
6646: @noindent
6647: into
6648:
6649: @example
6650: jbsr _foobar
6651: movel d1,sp@@
6652: movel d0,sp@@-
6653: fmoved sp@@+,fp0
6654: @end example
6655:
1.1.1.5 root 6656: @ignore
6657: If a peephole matches a sequence including one or more jump insns, you must
6658: take account of the flags such as @code{CC_REVERSED} which specify that the
6659: condition codes are represented in an unusual manner. The compiler
6660: automatically alters any ordinary conditional jumps which occur in such
6661: situations, but the compiler cannot alter jumps which have been replaced by
6662: peephole optimizations. So it is up to you to alter the assembler code
6663: that the peephole produces. Supply C code to write the assembler output,
6664: and in this C code check the condition code status flags and change the
6665: assembler code as appropriate.
6666: @end ignore
6667:
1.1 root 6668: @node Expander Definitions,, Peephole Definitions, Machine Desc
6669: @section Defining RTL Sequences for Code Generation
6670:
6671: On some target machines, some standard pattern names for RTL generation
6672: cannot be handled with single insn, but a sequence of RTL insns can
6673: represent them. For these target machines, you can write a
6674: @samp{define_expand} to specify how to generate the sequence of RTL.
6675:
6676: A @samp{define_expand} is an RTL expression that looks almost like a
6677: @samp{define_insn}; but, unlike the latter, a @samp{define_expand} is used
6678: only for RTL generation and it can produce more than one RTL insn.
6679:
6680: A @samp{define_expand} RTX has four operands:
6681:
6682: @itemize @bullet
6683: @item
6684: The name. Each @samp{define_expand} must have a name, since the only
6685: use for it is to refer to it by name.
6686:
6687: @item
6688: The RTL template. This is just like the RTL template for a
6689: @samp{define_peephole} in that it is a vector of RTL expressions
6690: each being one insn.
6691:
6692: @item
6693: The condition, a string containing a C expression. This expression is
6694: used to express how the availability of this pattern depends on
6695: subclasses of target machine, selected by command-line options when
6696: GNU CC is run. This is just like the condition of a
6697: @samp{define_insn} that has a standard name.
6698:
6699: @item
6700: The preparation statements, a string containing zero or more C
6701: statements which are to be executed before RTL code is generated from
6702: the RTL template.
6703:
6704: Usually these statements prepare temporary registers for use as
6705: internal operands in the RTL template, but they can also generate RTL
6706: insns directly by calling routines such as @samp{emit_insn}, etc.
6707: Any such insns precede the ones that come from the RTL template.
6708: @end itemize
6709:
6710: The RTL template, in addition to controlling generation of RTL insns,
6711: also describes the operands that need to be specified when this pattern
6712: is used. In particular, it gives a predicate for each operand.
6713:
6714: A true operand, which need to be specified in order to generate RTL from
6715: the pattern, should be described with a @samp{match_operand} in its first
6716: occurrence in the RTL template. This enters information on the operand's
6717: predicate into the tables that record such things. GNU CC uses the
6718: information to preload the operand into a register if that is required for
6719: valid RTL code. If the operand is referred to more than once, subsequent
6720: references should use @samp{match_dup}.
6721:
6722: The RTL template may also refer to internal ``operands'' which are
6723: temporary registers or labels used only within the sequence made by the
6724: @samp{define_expand}. Internal operands are substituted into the RTL
6725: template with @samp{match_dup}, never with @samp{match_operand}. The
6726: values of the internal operands are not passed in as arguments by the
6727: compiler when it requests use of this pattern. Instead, they are computed
6728: within the pattern, in the preparation statements. These statements
6729: compute the values and store them into the appropriate elements of
6730: @code{operands} so that @samp{match_dup} can find them.
6731:
6732: There are two special macros defined for use in the preparation statements:
6733: @code{DONE} and @code{FAIL}. Use them with a following semicolon,
6734: as a statement.
6735:
6736: @table @code
6737: @item DONE
6738: Use the @code{DONE} macro to end RTL generation for the pattern. The
6739: only RTL insns resulting from the pattern on this occasion will be
6740: those already emitted by explicit calls to @code{emit_insn} within the
6741: preparation statements; the RTL template will not be generated.
6742:
6743: @item FAIL
6744: Make the pattern fail on this occasion. When a pattern fails, it means
6745: that the pattern was not truly available. The calling routines in the
6746: compiler will try other strategies for code generation using other patterns.
6747:
6748: Failure is currently supported only for binary operations (addition,
6749: multiplication, shifting, etc.).
6750:
6751: Do not emit any insns explicitly with @code{emit_insn} before failing.
6752: @end table
6753:
6754: Here is an example, the definition of left-shift for the SPUR chip:
6755:
6756: @example
6757: (define_expand "ashlsi3"
6758: [(set (match_operand:SI 0 "register_operand" "")
6759: (ashift:SI
6760: (match_operand:SI 1 "register_operand" "")
6761: (match_operand:SI 2 "nonmemory_operand" "")))]
6762: ""
6763: "
6764: @{
6765: if (GET_CODE (operands[2]) != CONST_INT
6766: || (unsigned) INTVAL (operands[2]) > 3)
6767: FAIL;
6768: @}")
6769: @end example
6770:
6771: @noindent
6772: This example uses @samp{define_expand} so that it can generate an RTL insn
6773: for shifting when the shift-count is in the supported range of 0 to 3 but
6774: fail in other cases where machine insns aren't available. When it fails,
6775: the compiler tries another strategy using different patterns (such as, a
6776: library call).
6777:
6778: If the compiler were able to handle nontrivial condition-strings in
6779: patterns with names, then there would be possible to use a
6780: @samp{define_insn} in that case. Here is another case (zero-extension on
6781: the 68000) which makes more use of the power of @samp{define_expand}:
6782:
6783: @example
6784: (define_expand "zero_extendhisi2"
6785: [(set (match_operand:SI 0 "general_operand" "")
6786: (const_int 0))
6787: (set (strict_low_part
6788: (subreg:HI
6789: (match_operand:SI 0 "general_operand" "")
6790: 0))
6791: (match_operand:HI 1 "general_operand" ""))]
6792: ""
6793: "operands[1] = make_safe_from (operands[1], operands[0]);")
6794: @end example
6795:
6796: @noindent
6797: Here two RTL insns are generated, one to clear the entire output operand
6798: and the other to copy the input operand into its low half. This sequence
6799: is incorrect if the input operand refers to [the old value of] the output
6800: operand, so the preparation statement makes sure this isn't so. The
6801: function @code{make_safe_from} copies the @code{operands[1]} into a
6802: temporary register if it refers to @code{operands[0]}. It does this
6803: by emitting another RTL insn.
6804:
6805: Finally, a third example shows the use of an internal operand.
6806: Zero-extension on the SPUR chip is done by @samp{and}-ing the result
6807: against a halfword mask. But this mask cannot be represented by a
6808: @samp{const_int} because the constant value is too large to be legitimate
6809: on this machine. So it must be copied into a register with
6810: @code{force_reg} and then the register used in the @samp{and}.
6811:
6812: @example
6813: (define_expand "zero_extendhisi2"
6814: [(set (match_operand:SI 0 "register_operand" "")
6815: (and:SI (subreg:SI
6816: (match_operand:HI 1 "register_operand" "")
6817: 0)
6818: (match_dup 2)))]
6819: ""
6820: "operands[2]
6821: = force_reg (SImode, gen_rtx (CONST_INT,
6822: VOIDmode, 65535)); ")
6823: @end example
6824:
6825: @node Machine Macros, Config, Machine Desc, Top
6826: @chapter Machine Description Macros
6827:
6828: The other half of the machine description is a C header file conventionally
6829: given the name @file{tm-@var{machine}.h}. The file @file{tm.h} should be a
6830: link to it. The header file @file{config.h} includes @file{tm.h} and most
6831: compiler source files include @file{config.h}.
6832:
6833: @menu
6834: * Run-time Target:: Defining -m options like -m68000 and -m68020.
6835: * Storage Layout:: Defining sizes and alignments of data types.
6836: * Registers:: Naming and describing the hardware registers.
6837: * Register Classes:: Defining the classes of hardware registers.
6838: * Stack Layout:: Defining which way the stack grows and by how much.
6839: * Library Names:: Specifying names of subroutines to call automatically.
6840: * Addressing Modes:: Defining addressing modes valid for memory operands.
6841: * Condition Code:: Defining how insns update the condition code.
6842: * Assembler Format:: Defining how to write insns and pseudo-ops to output.
1.1.1.5 root 6843: * Cross-compilation:: Handling floating point for cross-compilers.
1.1 root 6844: * Misc:: Everything else.
6845: @end menu
6846:
6847: @node Run-time Target, Storage Layout, Machine Macros, Machine Macros
6848: @section Run-time Target Specification
6849:
6850: @table @code
6851: @item CPP_PREDEFINES
6852: Define this to be a string constant containing @samp{-D} options to
6853: define the predefined macros that identify this machine and system.
6854: These macros will be predefined unless the @samp{-ansi} option is
6855: specified.
6856:
1.1.1.4 root 6857: In addition, a parallel set of macros are predefined, whose names are
6858: made by appending @samp{__} at the beginning and at the end. These
6859: @samp{__} macros are permitted by the ANSI standard, so they are
6860: predefined regardless of whether @samp{-ansi} is specified.
6861:
6862: For example, on the Sun, one can use the following value:
1.1 root 6863:
6864: @example
6865: "-Dmc68000 -Dsun -Dunix"
6866: @end example
6867:
1.1.1.5 root 6868: The result is to define the macros @samp{__mc68000__}, @samp{__sun__}
1.1.1.4 root 6869: and @samp{__unix__} unconditionally, and the macros @samp{mc68000},
6870: @samp{sun} and @samp{unix} provided @samp{-ansi} is not specified.
6871:
1.1 root 6872: @item CPP_SPEC
6873: A C string constant that tells the GNU CC driver program options to
6874: pass to CPP. It can also specify how to translate options you
6875: give to GNU CC into options for GNU CC to pass to the CPP.
6876:
6877: Do not define this macro if it does not need to do anything.
6878:
6879: @item CC1_SPEC
6880: A C string constant that tells the GNU CC driver program options to
6881: pass to CC1. It can also specify how to translate options you
6882: give to GNU CC into options for GNU CC to pass to the CC1.
6883:
6884: Do not define this macro if it does not need to do anything.
6885:
6886: @item extern int target_flags;
6887: This declaration should be present.
6888:
6889: @item TARGET_@dots{}
6890: This series of macros is to allow compiler command arguments to
6891: enable or disable the use of optional features of the target machine.
6892: For example, one machine description serves both the 68000 and
6893: the 68020; a command argument tells the compiler whether it should
6894: use 68020-only instructions or not. This command argument works
6895: by means of a macro @code{TARGET_68020} that tests a bit in
6896: @code{target_flags}.
6897:
6898: Define a macro @code{TARGET_@var{featurename}} for each such option.
6899: Its definition should test a bit in @code{target_flags}; for example:
6900:
6901: @example
6902: #define TARGET_68020 (target_flags & 1)
6903: @end example
6904:
6905: One place where these macros are used is in the condition-expressions
6906: of instruction patterns. Note how @code{TARGET_68020} appears
6907: frequently in the 68000 machine description file, @file{m68k.md}.
6908: Another place they are used is in the definitions of the other
6909: macros in the @file{tm-@var{machine}.h} file.
6910:
6911: @item TARGET_SWITCHES
6912: This macro defines names of command options to set and clear
6913: bits in @code{target_flags}. Its definition is an initializer
6914: with a subgrouping for each command option.
6915:
6916: Each subgrouping contains a string constant, that defines the option
6917: name, and a number, which contains the bits to set in
6918: @code{target_flags}. A negative number says to clear bits instead;
6919: the negative of the number is which bits to clear. The actual option
6920: name is made by appending @samp{-m} to the specified name.
6921:
6922: One of the subgroupings should have a null string. The number in
6923: this grouping is the default value for @code{target_flags}. Any
6924: target options act starting with that value.
6925:
6926: Here is an example which defines @samp{-m68000} and @samp{-m68020}
6927: with opposite meanings, and picks the latter as the default:
6928:
6929: @example
6930: #define TARGET_SWITCHES \
6931: @{ @{ "68020", 1@}, \
6932: @{ "68000", -1@}, \
6933: @{ "", 1@}@}
6934: @end example
6935:
6936: @item OVERRIDE_OPTIONS
6937: Sometimes certain combinations of command options do not make sense on
6938: a particular target machine. You can define a macro
6939: @code{OVERRIDE_OPTIONS} to take account of this. This macro, if
6940: defined, is executed once just after all the command options have been
6941: parsed.
6942: @end table
6943:
6944: @node Storage Layout, Registers, Run-time Target, Machine Macros
6945: @section Storage Layout
6946:
6947: Note that the definitions of the macros in this table which are sizes or
6948: alignments measured in bits do not need to be constant. They can be C
6949: expressions that refer to static variables, such as the @code{target_flags}.
6950: @xref{Run-time Target}.
6951:
6952: @table @code
6953: @item BITS_BIG_ENDIAN
6954: Define this macro if the most significant bit in a byte has the lowest
6955: number. This means that bit-field instructions count from the most
6956: significant bit. If the machine has no bit-field instructions, this
6957: macro is irrelevant.
6958:
6959: @item BYTES_BIG_ENDIAN
6960: Define this macro if the most significant byte in a word has the
6961: lowest number.
6962:
6963: @item WORDS_BIG_ENDIAN
6964: Define this macro if, in a multiword object, the most significant
6965: word has the lowest number.
6966:
6967: @item BITS_PER_UNIT
6968: Number of bits in an addressable storage unit (byte); normally 8.
6969:
6970: @item BITS_PER_WORD
6971: Number of bits in a word; normally 32.
6972:
6973: @item UNITS_PER_WORD
6974: Number of storage units in a word; normally 4.
6975:
6976: @item POINTER_SIZE
6977: Width of a pointer, in bits.
6978:
6979: @item POINTER_BOUNDARY
6980: Alignment required for pointers stored in memory, in bits.
6981:
6982: @item PARM_BOUNDARY
6983: Alignment required for function parameters on the stack, in bits.
6984:
6985: @item STACK_BOUNDARY
6986: Define this macro if you wish to preserve a certain alignment for
6987: the stack pointer at all times. The definition is a C expression
6988: for the desired alignment (measured in bits).
6989:
6990: @item FUNCTION_BOUNDARY
6991: Alignment required for a function entry point, in bits.
6992:
6993: @item BIGGEST_ALIGNMENT
6994: Biggest alignment that any data type can require on this machine, in bits.
6995:
6996: @item EMPTY_FIELD_BOUNDARY
6997: Alignment in bits to be given to a structure bit field that follows an
6998: empty field such as @code{int : 0;}.
6999:
7000: @item STRUCTURE_SIZE_BOUNDARY
7001: Number of bits which any structure or union's size must be a multiple of.
7002: Each structure or union's size is rounded up to a multiple of this.
7003:
7004: If you do not define this macro, the default is the same as
7005: @code{BITS_PER_UNIT}.
7006:
7007: @item STRICT_ALIGNMENT
7008: Define this if instructions will fail to work if given data not
7009: on the nominal alignment. If instructions will merely go slower
7010: in that case, do not define this macro.
7011:
7012: @item PCC_BITFIELD_TYPE_MATTERS
7013: Define this if you wish to imitate a certain bizarre behavior pattern
7014: of some instances of PCC: a bit field whose declared type is
7015: @code{int} has the same effect on the size and alignment of a
7016: structure as an actual @code{int} would have.
7017:
7018: Just what effect that is in GNU CC depends on other parameters, but on
7019: most machines it would force the structure's alignment and size to a
7020: multiple of 32 or @code{BIGGEST_ALIGNMENT} bits.
7021:
7022: @item CHECK_FLOAT_VALUE (@var{mode}, @var{value})
7023: A C statement to validate the value @var{value} (or type
7024: @code{double}) for mode @var{mode}. This means that you check whether
7025: @var{value} fits within the possible range of values for mode
7026: @var{mode} on this target machine. The mode @var{mode} is always
7027: @code{SFmode} or @code{DFmode}.
7028:
7029: If @var{value} is not valid, you should call @code{error} to print an
7030: error message and then assign some valid value to @var{value}.
7031: Allowing an invalid value to go through the compiler can produce
7032: incorrect assembler code which may even cause Unix assemblers to
7033: crash.
7034:
7035: This macro need not be defined if there is no work for it to do.
7036: @end table
7037:
7038: @node Registers, Register Classes, Storage Layout, Machine Macros
7039: @section Register Usage
7040:
7041: @table @code
7042: @item FIRST_PSEUDO_REGISTER
7043: Number of hardware registers known to the compiler. They receive
7044: numbers 0 through @code{FIRST_PSEUDO_REGISTER-1}; thus, the first
7045: pseudo register's number really is assigned the number
7046: @code{FIRST_PSEUDO_REGISTER}.
7047:
7048: @item FIXED_REGISTERS
7049: An initializer that says which registers are used for fixed purposes
7050: all throughout the compiled code and are therefore not available for
7051: general allocation. These would include the stack pointer, the frame
7052: pointer (except on machines where that can be used as a general
7053: register when no frame pointer is needed), the program counter on
7054: machines where that is considered one of the addressable registers,
7055: and any other numbered register with a standard use.
7056:
7057: This information is expressed as a sequence of numbers, separated by
7058: commas and surrounded by braces. The @var{n}th number is 1 if
7059: register @var{n} is fixed, 0 otherwise.
7060:
7061: The table initialized from this macro, and the table initialized by
7062: the following one, may be overridden at run time either automatically,
7063: by the actions of the macro @code{CONDITIONAL_REGISTER_USAGE}, or by
7064: the user with the command options @samp{-ffixed-@var{reg}},
7065: @samp{-fcall-used-@var{reg}} and @samp{-fcall-saved-@var{reg}}.
7066:
7067: @item CALL_USED_REGISTERS
7068: Like @code{FIXED_REGISTERS} but has 1 for each register that is
7069: clobbered (in general) by function calls as well as for fixed
7070: registers. This macro therefore identifies the registers that are not
7071: available for general allocation of values that must live across
7072: function calls.
7073:
7074: If a register has 0 in @code{CALL_USED_REGISTERS}, the compiler
7075: automatically saves it on function entry and restores it on function
7076: exit, if the register is used within the function.
7077:
1.1.1.6 ! root 7078: @item DEFAULT_CALLER_SAVES
! 7079: Define this macro if the target machine if function calls do not preserve
! 7080: any registers; in other words, if @code{CALL_USED_REGISTERS} has 1
! 7081: for all registers. This macro enables @samp{-fcaller-saves} by default.
! 7082: Eventually that option will be enabled by default on all machines and both
! 7083: the option and this macro will be eliminated.
! 7084:
1.1 root 7085: @item CONDITIONAL_REGISTER_USAGE
7086: Zero or more C statements that may conditionally modify two variables
7087: @code{fixed_regs} and @code{call_used_regs} (both of type @code{char
7088: []}) after they have been initialized from the two preceding macros.
7089:
7090: This is necessary in case the fixed or call-clobbered registers depend
7091: on target flags.
7092:
7093: You need not define this macro if it has no work to do.
7094:
7095: If the usage of an entire class of registers depends on the target
1.1.1.5 root 7096: flags, you may indicate this to GCC by using this macro to modify
1.1 root 7097: @code{fixed_regs} and @code{call_used_regs} to 1 for each of the
1.1.1.5 root 7098: registers in the classes which should not be used by GCC. Also define
1.1 root 7099: the macro @code{REG_CLASS_FROM_LETTER} to return @code{NO_REGS} if it
7100: is called with a letter for a class that shouldn't be used.
7101:
7102: (However, if this class is not included in @code{GENERAL_REGS} and all
7103: of the insn patterns whose constraints permit this class are
7104: controlled by target switches, then GCC will automatically avoid using
7105: these registers when the target switches are opposed to them.)
7106:
7107: @item OVERLAPPING_REGNO_P (@var{regno})
1.1.1.5 root 7108: If defined, this is a C expression whose value is nonzero if hard
7109: register number @var{regno} is an overlapping register. This means a
7110: hard register which overlaps a hard register with a different number.
7111: (Such overlap is undesirable, but occasionally it allows a machine to
7112: be supported which otherwise could not be.) This macro must return
7113: nonzero for @emph{all} the registers which overlap each other. GNU CC
7114: can use an overlapping register only in certain limited ways. It can
7115: be used for allocation within a basic block, and may be spilled for
7116: reloading; that is all.
1.1 root 7117:
7118: If this macro is not defined, it means that none of the hard registers
7119: overlap each other. This is the usual situation.
7120:
7121: @item INSN_CLOBBERS_REGNO_P (@var{insn}, @var{regno})
7122: If defined, this is a C expression whose value should be nonzero if
7123: the insn @var{insn} has the effect of mysteriously clobbering the
7124: contents of hard register number @var{regno}. By ``mysterious'' we
7125: mean that the insn's RTL expression doesn't describe such an effect.
7126:
7127: If this macro is not defined, it means that no insn clobbers registers
7128: mysteriously. This is the usual situation; all else being equal,
7129: it is best for the RTL expression to show all the activity.
7130:
7131: @item PRESERVE_DEATH_INFO_REGNO_P (@var{regno})
7132: If defined, this is a C expression whose value is nonzero if accurate
7133: @code{REG_DEAD} notes are needed for hard register number @var{regno}
7134: at the time of outputting the assembler code. When this is so, a few
7135: optimizations that take place after register allocation and could
7136: invalidate the death notes are not done when this register is
7137: involved.
7138:
7139: You would arrange to preserve death info for a register when some
7140: of the code in the machine description which is executed to write
7141: the assembler code looks at the the death notes. This is
7142: necessary only when the actual hardware feature which GNU CC
7143: thinks of as a register is not actually a register of the usual sort.
7144: (It might, for example, be a hardware stack.)
7145:
7146: If this macro is not defined, it means that no death notes need to be
7147: preserved. This is the usual situation.
7148:
7149: @item HARD_REGNO_REGS (@var{regno}, @var{mode})
7150: A C expression for the number of consecutive hard registers, starting
7151: at register number @var{regno}, required to hold a value of mode
7152: @var{mode}.
7153:
7154: On a machine where all registers are exactly one word, a suitable
7155: definition of this macro is
7156:
7157: @example
7158: #define HARD_REGNO_NREGS(REGNO, MODE) \
7159: ((GET_MODE_SIZE (MODE) + UNITS_PER_WORD - 1) \
7160: / UNITS_PER_WORD))
7161: @end example
7162:
7163: @item HARD_REGNO_MODE_OK (@var{regno}, @var{mode})
7164: A C expression that is nonzero if it is permissible to store a value
7165: of mode @var{mode} in hard register number @var{regno} (or in several
7166: registers starting with that one). For a machine where all registers
7167: are equivalent, a suitable definition is
7168:
7169: @example
7170: #define HARD_REGNO_MODE_OK(REGNO, MODE) 1
7171: @end example
7172:
7173: It is not necessary for this macro to check for fixed register numbers
7174: because the allocation mechanism considers them to be always occupied.
7175:
7176: Many machines have special registers for floating point arithmetic.
7177: Often people assume that floating point machine modes are allowed only
7178: in floating point registers. This is not true. Any registers that
7179: can hold integers can safely @emph{hold} a floating point machine
7180: mode, whether or not floating arithmetic can be done on it in those
7181: registers.
7182:
7183: The true significance of special floating registers is rather than
7184: non-floating-point machine modes @emph{may not} go in those registers.
7185: This is true if the floating registers normalize any value stored in
7186: them, because storing a non-floating value there would garble it. If
7187: the floating registers do not automatically normalize, if you can
7188: store any bit pattern in one and retrieve it unchanged without a trap,
7189: then any machine mode may go in a floating register and this macro
7190: should say so.
7191:
7192: Sometimes there are floating registers that are especially slow to
7193: access, so that it is better to store a value in a stack frame than in
7194: such a register if floating point arithmetic is not being done. As long
7195: as the floating registers are not in class @code{GENERAL_REGS}, they
7196: will not be used unless some insn's constraint asks for one.
7197:
7198: It is obligatory to support floating point `move' instructions into
7199: and out of any registers that can hold fixed point values, because
7200: unions and structures (which have modes @samp{SImode} or
7201: @samp{DImode}) can be in those registers and they may have floating
7202: point members.
7203:
7204: There may also be a need to support fixed point `move' instructions in
7205: and out of floating point registers. Unfortunately, I have forgotten
7206: why this was so, and I don't know whether it is still true. If
7207: @code{HARD_REGNO_MODE_OK} rejects fixed point values in floating point
7208: registers, then the constraints of the fixed point `move' instructions
7209: must be designed to avoid ever trying to reload into a floating point
7210: register.
7211:
7212: @item MODES_TIEABLE_P (@var{mode1}, @var{mode2})
7213: A C expression that is nonzero if it is desirable to choose register
7214: allocation so as to avoid move instructions between a value of mode
7215: @var{mode1} and a value of mode @var{mode2}.
7216:
7217: If @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode1})} and
7218: @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode2})} are ever different
7219: for any @var{r}, then @code{MODES_TIEABLE_P (@var{mode1},
7220: @var{mode2})} must be zero.
7221:
7222: @item PC_REGNUM
7223: If the program counter has a register number, define this as that
7224: register number. Otherwise, do not define it.
7225:
7226: @item STACK_POINTER_REGNUM
7227: The register number of the stack pointer register, which must also be
7228: a fixed register according to @code{FIXED_REGISTERS}. On many
7229: machines, the hardware determines which register this is.
7230:
7231: @item FRAME_POINTER_REGNUM
7232: The register number of the frame pointer register, which is used to
7233: access automatic variables in the stack frame. On some machines, the
7234: hardware determines which register this is. On other machines, you
7235: can choose any register you wish for this purpose.
7236:
7237: @item FRAME_POINTER_REQUIRED
7238: A C expression which is nonzero if a function must have and use a
7239: frame pointer. This expression is evaluated in the reload pass, in
7240: the function @code{reload}, and it can in principle examine the
7241: current function and decide according to the facts, but on most
7242: machines the constant 0 or the constant 1 suffices. Use 0 when the
7243: machine allows code to be generated with no frame pointer, and doing
7244: so saves some time or space. Use 1 when there is no possible
7245: advantage to avoiding a frame pointer.
7246:
1.1.1.5 root 7247: In certain cases, the compiler does not know how to produce valid code
7248: without a frame pointer. The compiler recognizes those cases and
7249: automatically gives the function a frame pointer regardless of what
1.1 root 7250: @code{FRAME_POINTER_REQUIRED} says. You don't need to worry about
7251: them.@refill
7252:
7253: In a function that does not require a frame pointer, the frame pointer
7254: register can be allocated for ordinary usage, unless you mark it as a
7255: fixed register. See @code{FIXED_REGISTERS} for more information.
7256:
7257: @item ARG_POINTER_REGNUM
7258: The register number of the arg pointer register, which is used to
7259: access the function's argument list. On some machines, this is the
7260: same as the frame pointer register. On some machines, the hardware
7261: determines which register this is. On other machines, you can choose
7262: any register you wish for this purpose. If this is not the same
7263: register as the frame pointer register, then you must mark it as a
7264: fixed register according to @code{FIXED_REGISTERS}.
7265:
7266: @item STATIC_CHAIN_REGNUM
7267: The register number used for passing a function's static chain
7268: pointer. This is needed for languages such as Pascal and Algol where
7269: functions defined within other functions can access the local
7270: variables of the outer functions; it is not currently used because C
7271: does not provide this feature, but you must define the macro.
7272:
7273: The static chain register need not be a fixed register.
7274:
7275: @item STRUCT_VALUE_REGNUM
7276: When a function's value's mode is @code{BLKmode}, the value is not
7277: returned according to @code{FUNCTION_VALUE}. Instead, the caller
7278: passes the address of a block of memory in which the value should be
7279: stored.
7280:
7281: If this value is passed in a register, then @code{STRUCT_VALUE_REGNUM}
7282: should be the number of that register.
7283:
7284: @item STRUCT_VALUE
7285: If the structure value address is not passed in a register, define
7286: @code{STRUCT_VALUE} as an expression returning an RTX for the place
7287: where the address is passed. If it returns a @samp{mem} RTX, the
7288: address is passed as an ``invisible'' first argument.
7289:
7290: @item STRUCT_VALUE_INCOMING_REGNUM
7291: On some architectures the place where the structure value address
7292: is found by the called function is not the same place that the
7293: caller put it. This can be due to register windows, or it could
7294: be because the function prologue moves it to a different place.
7295:
7296: If the incoming location of the structure value address is in a
7297: register, define this macro as the register number.
7298:
7299: @item STRUCT_VALUE_INCOMING
7300: If the incoming location is not a register, define
7301: @code{STRUCT_VALUE_INCOMING} as an expression for an RTX for where the
7302: called function should find the value. If it should find the value on
7303: the stack, define this to create a @samp{mem} which refers to the
7304: frame pointer. If the value is a @samp{mem}, the compiler assumes it
7305: is for an invisible first argument, and leaves space for it when
7306: finding the first real argument.
7307:
7308: @item REG_ALLOC_ORDER
7309: If defined, an initializer for a vector of integers, containing the
7310: numbers of hard registers in the order in which the GNU CC should
7311: prefer to use them (from most preferred to least).
7312:
7313: If this macro is not defined, registers are used lowest numbered first
7314: (all else being equal).
7315:
7316: One use of this macro is on the 360, where the highest numbered
7317: registers must always be saved and the save-multiple-registers
7318: instruction supports only sequences of consecutive registers. This
7319: macro is defined to cause the highest numbered allocatable registers
7320: to be used first.
7321: @end table
7322:
7323: @node Register Classes, Stack Layout, Registers, Machine Macros
7324: @section Register Classes
7325:
7326: On many machines, the numbered registers are not all equivalent.
7327: For example, certain registers may not be allowed for indexed addressing;
7328: certain registers may not be allowed in some instructions. These machine
7329: restrictions are described to the compiler using @dfn{register classes}.
7330:
7331: You define a number of register classes, giving each one a name and saying
7332: which of the registers belong to it. Then you can specify register classes
7333: that are allowed as operands to particular instruction patterns.
7334:
7335: In general, each register will belong to several classes. In fact, one
7336: class must be named @code{ALL_REGS} and contain all the registers. Another
7337: class must be named @code{NO_REGS} and contain no registers. Often the
7338: union of two classes will be another class; however, this is not required.
7339:
7340: One of the classes must be named @code{GENERAL_REGS}. There is nothing
7341: terribly special about the name, but the operand constraint letters
7342: @samp{r} and @samp{g} specify this class. If @code{GENERAL_REGS} is
7343: the same as @code{ALL_REGS}, just define it as a macro which expands
7344: to @code{ALL_REGS}.
7345:
7346: The way classes other than @code{GENERAL_REGS} are specified in operand
7347: constraints is through machine-dependent operand constraint letters.
7348: You can define such letters to correspond to various classes, then use
7349: them in operand constraints.
7350:
7351: You should define a class for the union of two classes whenever some
7352: instruction allows both classes. For example, if an instruction allows
7353: either a floating-point (coprocessor) register or a general register for a
7354: certain operand, you should define a class @code{FLOAT_OR_GENERAL_REGS}
7355: which includes both of them. Otherwise you will get suboptimal code.
7356:
7357: You must also specify certain redundant information about the register
7358: classes: for each class, which classes contain it and which ones are
7359: contained in it; for each pair of classes, the largest class contained
7360: in their union.
7361:
7362: Register classes used for input-operands of bitwise-and or shift
7363: instructions have a special requirement: each such class must have, for
7364: each fixed-point machine mode, a subclass whose registers can transfer that
7365: mode to or from memory. For example, on some machines, the operations for
7366: single-byte values (@code{QImode}) are limited to certain registers. When
7367: this is so, each register class that is used in a bitwise-and or shift
7368: instruction must have a subclass consisting of registers from which
7369: single-byte values can be loaded or stored. This is so that
7370: @code{PREFERRED_RELOAD_CLASS} can always have a possible value to return.
7371:
7372: @table @code
7373: @item enum reg_class
7374: An enumeral type that must be defined with all the register class names
7375: as enumeral values. @code{NO_REGS} must be first. @code{ALL_REGS}
7376: must be the last register class, followed by one more enumeral value,
7377: @code{LIM_REG_CLASSES}, which is not a register class but rather
7378: tells how many classes there are.
7379:
7380: Each register class has a number, which is the value of casting
7381: the class name to type @code{int}. The number serves as an index
7382: in many of the tables described below.
7383:
7384: @item N_REG_CLASSES
7385: The number of distinct register classes, defined as follows:
7386:
7387: @example
7388: #define N_REG_CLASSES (int) LIM_REG_CLASSES
7389: @end example
7390:
7391: @item REG_CLASS_NAMES
7392: An initializer containing the names of the register classes as C string
7393: constants. These names are used in writing some of the debugging dumps.
7394:
7395: @item REG_CLASS_CONTENTS
7396: An initializer containing the contents of the register classes, as integers
7397: which are bit masks. The @var{n}th integer specifies the contents of class
7398: @var{n}. The way the integer @var{mask} is interpreted is that
7399: register @var{r} is in the class if @code{@var{mask} & (1 << @var{r})} is 1.
7400:
7401: When the machine has more than 32 registers, an integer does not suffice.
7402: Then the integers are replaced by sub-initializers, braced groupings containing
7403: several integers. Each sub-initializer must be suitable as an initializer
7404: for the type @code{HARD_REG_SET} which is defined in @file{hard-reg-set.h}.
7405:
7406: @item REGNO_REG_CLASS (@var{regno})
7407: A C expression whose value is a register class containing hard register
7408: @var{regno}. In general there is more that one such class; choose a class
7409: which is @dfn{minimal}, meaning that no smaller class also contains the
7410: register.
7411:
7412: @item BASE_REG_CLASS
7413: A macro whose definition is the name of the class to which a valid
7414: base register must belong. A base register is one used in an address
7415: which is the register value plus a displacement.
7416:
7417: @item INDEX_REG_CLASS
7418: A macro whose definition is the name of the class to which a valid
7419: index register must belong. An index register is one used in an
7420: address where its value is either multiplied by a scale factor or
7421: added to another register (as well as added to a displacement).
7422:
7423: @item REG_CLASS_FROM_LETTER (@var{char})
7424: A C expression which defines the machine-dependent operand constraint
7425: letters for register classes. If @var{char} is such a letter, the
7426: value should be the register class corresponding to it. Otherwise,
7427: the value should be @code{NO_REGS}.
7428:
7429: @item REGNO_OK_FOR_BASE_P (@var{num})
7430: A C expression which is nonzero if register number @var{num} is
7431: suitable for use as a base register in operand addresses. It may be
7432: either a suitable hard register or a pseudo register that has been
7433: allocated such a hard register.
7434:
7435: @item REGNO_OK_FOR_INDEX_P (@var{num})
7436: A C expression which is nonzero if register number @var{num} is
7437: suitable for use as an index register in operand addresses. It may be
7438: either a suitable hard register or a pseudo register that has been
7439: allocated such a hard register.
7440:
7441: The difference between an index register and a base register is that
7442: the index register may be scaled. If an address involves the sum of
7443: two registers, neither one of them scaled, then either one may be
7444: labeled the ``base'' and the other the ``index''; but whichever
7445: labeling is used must fit the machine's constraints of which registers
7446: may serve in each capacity. The compiler will try both labelings,
7447: looking for one that is valid, and will reload one or both registers
7448: only if neither labeling works.
7449:
7450: @item PREFERRED_RELOAD_CLASS (@var{x}, @var{class})
7451: A C expression that places additional restrictions on the register class
7452: to use when it is necessary to copy value @var{x} into a register in class
7453: @var{class}. The value is a register class; perhaps @var{class}, or perhaps
7454: another, smaller class. On many machines, the definition
7455:
7456: @example
7457: #define PREFERRED_RELOAD_CLASS(X,CLASS) CLASS
7458: @end example
7459:
7460: @noindent
7461: is safe.
7462:
7463: Sometimes returning a more restrictive class makes better code. For
7464: example, on the 68000, when @var{x} is an integer constant that is in range
7465: for a @samp{moveq} instruction, the value of this macro is always
7466: @code{DATA_REGS} as long as @var{class} includes the data registers.
7467: Requiring a data register guarantees that a @samp{moveq} will be used.
7468:
7469: If @var{x} is a @samp{const_double}, by returning @code{NO_REGS}
7470: you can force @var{x} into a memory constant. This is useful on
7471: certain machines where immediate floating values cannot be loaded into
7472: certain kinds of registers.
7473:
7474: In a shift instruction or a bitwise-and instruction, the mode of @var{x},
7475: the value being reloaded, may not be the same as the mode of the
7476: instruction's operand. (They will both be fixed-point modes, however.) In
7477: such a case, @var{class} may not be a safe value to return. @var{class} is
7478: certainly valid for the instruction, but it may not be valid for reloading
7479: @var{x}. This problem can occur on machines such as the 68000 and 80386
7480: where some registers can handle full-word values but cannot handle
7481: single-byte values.
7482:
7483: On such machines, this macro must examine the mode of @var{x} and return a
7484: subclass of @var{class} which can handle loads and stores of that mode. On
7485: the 68000, where address registers cannot handle @code{QImode}, if @var{x}
7486: has @code{QImode} then you must return @code{DATA_REGS}. If @var{class} is
7487: @code{ADDR_REGS}, then there is no correct value to return; but the shift
7488: and bitwise-and instructions don't use @code{ADDR_REGS}, so this fatal case
7489: never arises.
7490:
7491: @item CLASS_MAX_NREGS (@var{class}, @var{mode})
7492: A C expression for the maximum number of consecutive registers
7493: of class @var{class} needed to hold a value of mode @var{mode}.
7494:
7495: This is closely related to the macro @code{HARD_REGNO_NREGS}.
7496: In fact, the value of the macro @code{CLASS_MAX_NREGS (@var{class}, @var{mode})}
7497: should be the maximum value of @code{HARD_REGNO_NREGS (@var{regno}, @var{mode})}
7498: for all @var{regno} values in the class @var{class}.
7499:
7500: This macro helps control the handling of multiple-word values
7501: in the reload pass.
7502: @end table
7503:
7504: Two other special macros describe which constants fit which constraint
7505: letters.
7506:
7507: @table @code
7508: @item CONST_OK_FOR_LETTER_P (@var{value}, @var{c})
7509: A C expression that defines the machine-dependent operand constraint letters
7510: that specify particular ranges of integer values. If @var{c} is one
7511: of those letters, the expression should check that @var{value}, an integer,
7512: is in the appropriate range and return 1 if so, 0 otherwise. If @var{c} is
7513: not one of those letters, the value should be 0 regardless of @var{value}.
7514:
7515: @item CONST_DOUBLE_OK_FOR_LETTER_P (@var{value}, @var{c})
7516: A C expression that defines the machine-dependent operand constraint
7517: letters that specify particular ranges of floating values. If @var{c} is
7518: one of those letters, the expression should check that @var{value}, an RTX
7519: of code @samp{const_double}, is in the appropriate range and return 1 if
7520: so, 0 otherwise. If @var{c} is not one of those letters, the value should
7521: be 0 regardless of @var{value}.
7522: @end table
7523:
7524: @node Stack Layout, Library Names, Register Classes, Machine Macros
7525: @section Describing Stack Layout
7526:
7527: @table @code
7528: @item STACK_GROWS_DOWNWARD
7529: Define this macro if pushing a word onto the stack moves the stack
7530: pointer to a smaller address.
7531:
7532: When we say, ``define this macro if @dots{},'' it means that the
7533: compiler checks this macro only with @code{#ifdef} so the precise
7534: definition used does not matter.
7535:
7536: @item FRAME_GROWS_DOWNWARD
7537: Define this macro if the addresses of local variable slots are at negative
7538: offsets from the frame pointer.
7539:
7540: @item STARTING_FRAME_OFFSET
7541: Offset from the frame pointer to the first local variable slot to be allocated.
7542:
7543: If @code{FRAME_GROWS_DOWNWARD}, the next slot's offset is found by
7544: subtracting the length of the first slot from @code{STARTING_FRAME_OFFSET}.
7545: Otherwise, it is found by adding the length of the first slot to
7546: the value @code{STARTING_FRAME_OFFSET}.
7547:
7548: @item PUSH_ROUNDING (@var{npushed})
7549: A C expression that is the number of bytes actually pushed onto the
7550: stack when an instruction attempts to push @var{npushed} bytes.
7551:
7552: If the target machine does not have a push instruction, do not define
7553: this macro. That directs GNU CC to use an alternate strategy: to
7554: allocate the entire argument block and then store the arguments into
7555: it.
7556:
7557: On some machines, the definition
7558:
7559: @example
7560: #define PUSH_ROUNDING(BYTES) (BYTES)
7561: @end example
7562:
7563: @noindent
7564: will suffice. But on other machines, instructions that appear
7565: to push one byte actually push two bytes in an attempt to maintain
7566: alignment. Then the definition should be
7567:
7568: @example
7569: #define PUSH_ROUNDING(BYTES) (((BYTES) + 1) & ~1)
7570: @end example
7571:
7572: @item FIRST_PARM_OFFSET (@var{fundecl})
7573: Offset from the argument pointer register to the first argument's
7574: address. On some machines it may depend on the data type of the
7575: function. (In the next version of GNU CC, the argument will be
7576: changed to the function data type rather than its declaration.)
7577:
7578: @item FIRST_PARM_CALLER_OFFSET (@var{fundecl})
7579: Define this macro on machines where register parameters have shadow
7580: locations on the stack, at addresses below the nominal parameter.
7581: This matters because certain arguments cannot be passed on the stack.
7582: On these machines, such arguments must be stored into the shadow
7583: locations.
7584:
7585: This macro should expand into a C expression whose value is the offset
7586: of the first parameter's shadow location from the nominal stack
7587: pointer value. (That value is itself computed by adding the value of
7588: @code{STACK_POINTER_OFFSET} to the stack pointer register.)
7589:
1.1.1.6 ! root 7590: @item STACK_ARGS_ADJUST (@var{size})
! 7591: Define this macro if the machine requires padding on the stack for
! 7592: certain function calls. This is padding on a per-function-call basis,
! 7593: not padding for individual arguments.
! 7594:
! 7595: The definition should be a C expression for the total size to be
! 7596: pushed on the stack in a call whose arguments (including their
! 7597: individual padding) occupy @var{size} bytes.
! 7598:
1.1 root 7599: @item RETURN_POPS_ARGS (@var{funtype})
7600: A C expression that should be 1 if a function pops its own arguments
7601: on returning, or 0 if the function pops no arguments and the caller
7602: must therefore pop them all after the function returns.
7603:
7604: @var{funtype} is a C variable whose value is a tree node that
7605: describes the function in question. Normally it is a node of type
7606: @code{FUNCTION_TYPE} that describes the data type of the function.
7607: From this it is possible to obtain the data types of the value and
7608: arguments (if known).
7609:
7610: When a call to a library function is being considered, @var{funtype}
7611: will contain an identifier node for the library function. Thus, if
7612: you need to distinguish among various library functions, you can do so
7613: by their names. Note that ``library function'' in this context means
7614: a function used to perform arithmetic, whose name is known specially
7615: in the compiler and was not mentioned in the C code being compiled.
7616:
7617: On the Vax, all functions always pop their arguments, so the
7618: definition of this macro is 1. On the 68000, using the standard
7619: calling convention, no functions pop their arguments, so the value of
7620: the macro is always 0 in this case. But an alternative calling
7621: convention is available in which functions that take a fixed number of
7622: arguments pop them but other functions (such as @code{printf}) pop
7623: nothing (the caller pops all). When this convention is in use,
7624: @var{funtype} is examined to determine whether a function takes a
7625: fixed number of arguments.
7626:
7627: @item FUNCTION_VALUE (@var{valtype}, @var{func})
7628: A C expression to create an RTX representing the place where a
7629: function returns a value of data type @var{valtype}. @var{valtype} is
7630: a tree node representing a data type. Write @code{TYPE_MODE
7631: (@var{valtype})} to get the machine mode used to represent that type.
7632: On many machines, only the mode is relevant. (Actually, on most
7633: machines, scalar values are returned in the same place regardless of
7634: mode).@refill
7635:
7636: If the precise function being called is known, @var{func} is a tree
7637: node (@code{FUNCTION_DECL}) for it; otherwise, @var{func} is a null
7638: pointer. This makes it possible to use a different value-returning
7639: convention for specific functions when all their calls are
7640: known.@refill
7641:
7642: @item FUNCTION_OUTGOING_VALUE (@var{valtype}, @var{func})
7643: Define this macro if the target machine has ``register windows''
7644: so that the register in which a function returns its value is not
7645: the same as the one in which the caller sees the value.
7646:
7647: For such machines, @code{FUNCTION_VALUE} computes the register in
7648: which the caller will see the value, and
7649: @code{FUNCTION_OUTGOING_VALUE} should be defined in a similar fashion
7650: to tell the function where to put the value.@refill
7651:
7652: If @code{FUNCTION_OUTGOING_VALUE} is not defined,
7653: @code{FUNCTION_VALUE} serves both purposes.@refill
7654:
7655: @item LIBCALL_VALUE (@var{mode})
7656: A C expression to create an RTX representing the place where a library
7657: function returns a value of mode @var{mode}. If the precise function
7658: being called is known, @var{func} is a tree node
7659: (@code{FUNCTION_DECL}) for it; otherwise, @var{func} is a null
7660: pointer. This makes it possible to use a different value-returning
7661: convention for specific functions when all their calls are
7662: known.@refill
7663:
7664: Note that ``library function'' in this context means a compiler
7665: support routine, used to perform arithmetic, whose name is known
7666: specially by the compiler and was not mentioned in the C code being
7667: compiled.
7668:
7669: @item FUNCTION_VALUE_REGNO_P (@var{regno})
7670: A C expression that is nonzero if @var{regno} is the number of a hard
7671: register in which the values of called function may come back.
7672:
7673: A register whose use for returning values is limited to serving as the
7674: second of a pair (for a value of type @code{double}, say) need not be
7675: recognized by this macro. So for most machines, this definition
7676: suffices:
7677:
7678: @example
7679: #define FUNCTION_VALUE_REGNO_P(N) ((N) == 0)
7680: @end example
7681:
7682: If the machine has register windows, so that the caller and the called
7683: function use different registers for the return value, this macro
7684: should recognize only the caller's register numbers.
7685:
7686: @item FUNCTION_ARG (@var{cum}, @var{mode}, @var{type}, @var{named})
7687: A C expression that controls whether a function argument is passed
7688: in a register, and which register.
7689:
7690: The arguments are @var{cum}, which summarizes all the previous
7691: arguments; @var{mode}, the machine mode of the argument; @var{type},
7692: the data type of the argument as a tree node or 0 if that is not known
7693: (which happens for C support library functions); and @var{named},
7694: which is 1 for an ordinary argument and 0 for nameless arguments that
7695: correspond to @samp{...} in the called function's prototype.
7696:
7697: The value of the expression should either be a @samp{reg} RTX for the
7698: hard register in which to pass the argument, or zero to pass the
7699: argument on the stack.
7700:
7701: For the Vax and 68000, where normally all arguments are pushed, zero
7702: suffices as a definition.
7703:
7704: @item FUNCTION_INCOMING_ARG (@var{cum}, @var{mode}, @var{type}, @var{named})
7705: Define this macro if the target machine has ``register windows'', so
7706: that the register in which a function sees an arguments is not
7707: necessarily the same as the one in which the caller passed the
7708: argument.
7709:
7710: For such machines, @code{FUNCTION_ARG} computes the register in which
7711: the caller passes the value, and @code{FUNCTION_INCOMING_ARG} should
7712: be defined in a similar fashion to tell the function being called
7713: where the arguments will arrive.
7714:
7715: If @code{FUNCTION_INCOMING_ARG} is not defined, @code{FUNCTION_ARG}
7716: serves both purposes.@refill
7717:
7718: @item FUNCTION_ARG_PARTIAL_NREGS (@var{cum}, @var{mode}, @var{type}, @var{named})
7719: A C expression for the number of words, at the beginning of an
7720: argument, must be put in registers. The value must be zero for
7721: arguments that are passed entirely in registers or that are entirely
7722: pushed on the stack.
7723:
7724: On some machines, certain arguments must be passed partially in
7725: registers and partially in memory. On these machines, typically the
7726: first @var{n} words of arguments are passed in registers, and the rest
7727: on the stack. If a multi-word argument (a @code{double} or a
7728: structure) crosses that boundary, its first few words must be passed
7729: in registers and the rest must be pushed. This macro tells the
7730: compiler when this occurs, and how many of the words should go in
7731: registers.
7732:
7733: @code{FUNCTION_ARG} for these arguments should return the first
7734: register to be used by the caller for this argument; likewise
7735: @code{FUNCTION_INCOMING_ARG}, for the called function.
7736:
7737: @item CUMULATIVE_ARGS
7738: A C type for declaring a variable that is used as the first argument
7739: of @code{FUNCTION_ARG} and other related values. For some target
7740: machines, the type @code{int} suffices and can hold the number of
7741: bytes of argument so far.
7742:
7743: @item INIT_CUMULATIVE_ARGS (@var{cum}, @var{fntype})
7744: A C statement (sans semicolon) for initializing the variable @var{cum}
7745: for the state at the beginning of the argument list. The variable has
7746: type @code{CUMULATIVE_ARGS}. The value of @var{fntype} is the tree node
7747: for the data type of the function which will receive the args, or 0
7748: if the args are to a compiler support library function.
7749:
7750: @item FUNCTION_ARG_ADVANCE (@var{cum}, @var{mode}, @var{type}, @var{named})
7751: Update the summarizer variable @var{cum} to advance past an argument
7752: in the argument list. The values @var{mode}, @var{type} and
7753: @var{named} describe that argument. Once this is done, the variable
7754: @var{cum} is suitable for analyzing the @emph{following} argument
7755: with @code{FUNCTION_ARG}, etc.@refill
7756:
7757: @item FUNCTION_ARG_REGNO_P (@var{regno})
7758: A C expression that is nonzero if @var{regno} is the number of a hard
7759: register in which function arguments are sometimes passed. This does
7760: @emph{not} include implicit arguments such as the static chain and
7761: the structure-value address. On many machines, no registers can be
7762: used for this purpose since all function arguments are pushed on the
7763: stack.
7764:
7765: @item FUNCTION_ARG_PADDING (@var{mode}, @var{size})
7766: If defined, a C expression which determines whether, and in which direction,
7767: to pad out an argument with extra space. The value should be of type
7768: @code{enum direction}: either @code{upward} to pad above the argument,
7769: @code{downward} to pad below, or @code{none} to inhibit padding.
7770:
7771: The argument @var{size} is an RTX which describes the size of the
7772: argument, in bytes. It should be used only if @var{mode} is
7773: @code{BLKmode}. Otherwise, @var{size} is 0.
7774:
7775: This macro does not control the @emph{amount} of padding; that is
7776: always just enough to reach the next multiple of @code{PARM_BOUNDARY}.
7777:
7778: This macro has a default definition which is right for most systems.
7779: For little-endian machines, the default is to pad upward. For
7780: big-endian machines, the default is to pad downward for an argument of
7781: constant size shorter than an @code{int}, and upward otherwise.
7782:
7783: @item FUNCTION_PROLOGUE (@var{file}, @var{size})
7784: A C compound statement that outputs the assembler code for entry to a
7785: function. The prologue is responsible for setting up the stack frame,
7786: initializing the frame pointer register, saving registers that must be
7787: saved, and allocating @var{size} additional bytes of storage for the
7788: local variables. @var{size} is an integer. @var{file} is a stdio
7789: stream to which the assembler code should be output.
7790:
7791: The label for the beginning of the function need not be output by this
7792: macro. That has already been done when the macro is run.
7793:
7794: To determine which registers to save, the macro can refer to the array
7795: @code{regs_ever_live}: element @var{r} is nonzero if hard register
7796: @var{r} is used anywhere within the function. This implies the
7797: function prologue should save register @var{r}, but not if it is one
7798: of the call-used registers.
7799:
7800: On machines where functions may or may not have frame-pointers, the
7801: function entry code must vary accordingly; it must set up the frame
7802: pointer if one is wanted, and not otherwise. To determine whether a
7803: frame pointer is in wanted, the macro can refer to the variable
7804: @code{frame_pointer_needed}. The variable's value will be 1 at run
7805: time in a function that needs a frame pointer.
7806:
7807: @item FUNCTION_PROFILER (@var{file}, @var{labelno})
7808: A C statement or compound statement to output to @var{file} some
7809: assembler code to call the profiling subroutine @code{mcount}.
7810: Before calling, the assembler code must load the address of a
7811: counter variable into a register where @code{mcount} expects to
7812: find the address. The name of this variable is @samp{LP} followed
7813: by the number @var{labelno}, so you would generate the name using
7814: @samp{LP%d} in a @code{fprintf}.
7815:
7816: The details of how the address should be passed to @code{mcount} are
7817: determined by your operating system environment, not by GNU CC. To
7818: figure them out, compile a small program for profiling using the
7819: system's installed C compiler and look at the assembler code that
7820: results.
7821:
1.1.1.6 ! root 7822: @item FUNCTION_BLOCK_PROFILER (@var{file}, @var{labelno})
! 7823: A C statement or compound statement to output to @var{file} some
! 7824: assembler code to initialize basic-block profiling for the current
! 7825: object module. This code should call the subroutine
! 7826: @code{__bb_init_func} once per object module, passing it as its sole
! 7827: argument the address of a block allocated in the object module.
! 7828:
! 7829: The name of the block is a local symbol made with this statement:
! 7830:
! 7831: @example
! 7832: ASM_GENERATE_INTERNAL_LABEL (@var{buffer}, "LPBX", 0);
! 7833: @end example
! 7834:
! 7835: Of course, since you are writing the definition of
! 7836: @code{ASM_GENERATE_INTERNAL_LABEL} as well as that of this macro, you
! 7837: can take a short cut in the definition of this macro and use the name
! 7838: that you know will result.
! 7839:
! 7840: The first word of this block is a flag which will be nonzero if the
! 7841: object module has already been initialized. So test this word first,
! 7842: and do not call @code{__bb_init_func} if the flag is nonzero.
! 7843:
! 7844: @item BLOCK_PROFILER (@var{file}, @var{blockno})
! 7845: A C statement or compound statement to increment the count associated
! 7846: with the basic block number @var{blockno}. Basic blocks are numbered
! 7847: separately from zero within each compilation. The count associated
! 7848: with block number @var{blockno} is at index @var{blockno} in a vector
! 7849: of words; the name of this array is a local symbol made with this
! 7850: statement:
! 7851:
! 7852: @example
! 7853: ASM_GENERATE_INTERNAL_LABEL (@var{buffer}, "LPBX", 2);
! 7854: @end example
! 7855:
! 7856: Of course, since you are writing the definition of
! 7857: @code{ASM_GENERATE_INTERNAL_LABEL} as well as that of this macro, you
! 7858: can take a short cut in the definition of this macro and use the name
! 7859: that you know will result.
! 7860:
1.1 root 7861: @item EXIT_IGNORES_STACK
7862: Define this macro as a C expression that is nonzero if the return
7863: instruction or the function epilogue ignores the value of the stack
7864: pointer; in other words, if it is safe to delete an instruction to
7865: adjust the stack pointer before a return from the function.
7866:
7867: Note that this macro's value is relevant only for for which frame
7868: pointers are maintained. It is never possible to delete a final stack
7869: adjustment in a function that has no frame pointer, and the compiler
7870: knows this regardless of @code{EXIT_IGNORES_STACK}.
7871:
7872: @item FUNCTION_EPILOGUE (@var{file}, @var{size})
7873: A C compound statement that outputs the assembler code for exit from a
7874: function. The epilogue is responsible for restoring the saved
7875: registers and stack pointer to their values when the function was
7876: called, and returning control to the caller. This macro takes the
7877: same arguments as the macro @code{FUNCTION_PROLOGUE}, and the
7878: registers to restore are determined from @code{regs_ever_live} and
7879: @code{CALL_USED_REGISTERS} in the same way.
7880:
7881: On some machines, there is a single instruction that does all the work
7882: of returning from the function. On these machines, give that
7883: instruction the name @samp{return} and do not define the macro
7884: @code{FUNCTION_EPILOGUE} at all.
7885:
7886: Do not define a pattern named @samp{return} if you want the
7887: @code{FUNCTION_EPILOGUE} to be used. If you want the target switches
7888: to control whether return instructions or epilogues are used, define a
7889: @samp{return} pattern with a validity condition that tests the target
7890: switches appropriately. If the @samp{return} pattern's validity
7891: condition is false, epilogues will be used.
7892:
7893: On machines where functions may or may not have frame-pointers, the
7894: function exit code must vary accordingly. Sometimes the code for
7895: these two cases is completely different. To determine whether a frame
7896: pointer is in wanted, the macro can refer to the variable
7897: @code{frame_pointer_needed}. The variable's value will be 1 at run
7898: time in a function that needs a frame pointer.
7899:
7900: On some machines, some functions pop their arguments on exit while
7901: others leave that for the caller to do. For example, the 68020 when
7902: given @samp{-mrtd} pops arguments in functions that take a fixed
7903: number of arguments.
7904:
7905: Your definition of the macro @code{RETURN_POPS_ARGS} decides which
7906: functions pop their own arguments. @code{FUNCTION_EPILOGUE} needs to
7907: know what was decided. The variable @code{current_function_pops_args}
7908: is nonzero if the function should pop its own arguments. If so, use
7909: the variable @code{current_function_args_size} as the number of bytes
7910: to pop.
7911:
7912: @item FIX_FRAME_POINTER_ADDRESS (@var{addr}, @var{depth})
7913: A C compound statement to alter a memory address that uses the frame
7914: pointer register so that it uses the stack pointer register instead.
7915: This must be done in the instructions that load parameter values into
7916: registers, when the reload pass determines that a frame pointer is not
7917: necessary for the function. @var{addr} will be a C variable name, and
7918: the updated address should be stored in that variable. @var{depth}
7919: will be the current depth of stack temporaries (number of bytes of
7920: arguments currently pushed). The change in offset between a
7921: frame-pointer-relative address and a stack-pointer-relative address
7922: must include @var{depth}.
7923:
7924: Even if your machine description specifies there will always be a
7925: frame pointer in the frame pointer register, you must still define
7926: @code{FIX_FRAME_POINTER_ADDRESS}, but the definition will never be
7927: executed at run time, so it may be empty.
7928: @end table
7929:
7930: @node Library Names, Addressing Modes, Stack Layout, Machine Macros
7931: @section Library Subroutine Names
7932:
7933: @table @code
1.1.1.5 root 7934: @item MULSI3_LIBCALL
7935: A C string constant giving the name of the function to call for
7936: multiplication of one signed full-word by another. If you do not
7937: define this macro, the default name is used, which is @code{__mulsi3},
7938: a function defined in @file{gnulib}.
7939:
7940: @item UMULSI3_LIBCALL
7941: A C string constant giving the name of the function to call for
7942: multiplication of one unsigned full-word by another. If you do not
7943: define this macro, the default name is used, which is
7944: @code{__umulsi3}, a function defined in @file{gnulib}.
7945:
7946: @item DIVSI3_LIBCALL
7947: A C string constant giving the name of the function to call for
7948: division of one signed full-word by another. If you do not define
7949: this macro, the default name is used, which is @code{__divsi3}, a
7950: function defined in @file{gnulib}.
7951:
1.1 root 7952: @item UDIVSI3_LIBCALL
7953: A C string constant giving the name of the function to call for
1.1.1.5 root 7954: division of one unsigned full-word by another. If you do not define
7955: this macro, the default name is used, which is @code{__udivsi3}, a
7956: function defined in @file{gnulib}.
7957:
7958: @item MODSI3_LIBCALL
7959: A C string constant giving the name of the function to call for the
7960: remainder in division of one signed full-word by another. If you do
7961: not define this macro, the default name is used, which is
7962: @code{__modsi3}, a function defined in @file{gnulib}.
1.1 root 7963:
7964: @item UMODSI3_LIBCALL
7965: A C string constant giving the name of the function to call for the
1.1.1.5 root 7966: remainder in division of one unsigned full-word by another. If you do
7967: not define this macro, the default name is used, which is
7968: @code{__umodsi3}, a function defined in @file{gnulib}.
1.1 root 7969:
7970: @item TARGET_MEM_FUNCTIONS
7971: Define this macro if GNU CC should generate calls to the System V
7972: (and ANSI C) library functions @code{memcpy} and @code{memset}
7973: rather than the BSD functions @code{bcopy} and @code{bzero}.
7974: @end table
7975:
1.1.1.5 root 7976: @node Addressing Modes, Cross-compilation, Library Names, Machine Macros
1.1 root 7977: @section Addressing Modes
7978:
7979: @table @code
7980: @item HAVE_POST_INCREMENT
7981: Define this macro if the machine supports post-increment addressing.
7982:
7983: @item HAVE_PRE_INCREMENT
7984: @itemx HAVE_POST_DECREMENT
7985: @itemx HAVE_PRE_DECREMENT
7986: Similar for other kinds of addressing.
7987:
7988: @item CONSTANT_ADDRESS_P (@var{x})
7989: A C expression that is 1 if the RTX @var{x} is a constant whose value
7990: is an integer. This includes integers whose values are not explicitly
7991: known, such as @samp{symbol_ref} and @samp{label_ref} expressions and
7992: @samp{const} arithmetic expressions.
7993:
7994: On most machines, this can be defined as @code{CONSTANT_P (@var{x})},
7995: but a few machines are more restrictive in which constant addresses
7996: are supported.
7997:
7998: @item MAX_REGS_PER_ADDRESS
7999: A number, the maximum number of registers that can appear in a valid
8000: memory address.
8001:
8002: @item GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{label})
8003: A C compound statement with a conditional @code{goto @var{label};}
8004: executed if @var{x} (an RTX) is a legitimate memory address on the
8005: target machine for a memory operand of mode @var{mode}.
8006:
8007: It usually pays to define several simpler macros to serve as
8008: subroutines for this one. Otherwise it may be too complicated to
8009: understand.
8010:
8011: This macro must exist in two variants: a strict variant and a
8012: non-strict one. The strict variant is used in the reload pass. It
8013: must be defined so that any pseudo-register that has not been
8014: allocated a hard register is considered a memory reference. In
8015: contexts where some kind of register is required, a pseudo-register
8016: with no hard register must be rejected.
8017:
8018: The non-strict variant is used in other passes. It must be defined to
8019: accept all pseudo-registers in every context where some kind of
8020: register is required.
8021:
8022: Compiler source files that want to use the strict variant of this
8023: macro define the macro @code{REG_OK_STRICT}. You should use an
8024: @code{#ifdef REG_OK_STRICT} conditional to define the strict variant
8025: in that case and the non-strict variant otherwise.
8026:
8027: Typically among the subroutines used to define
8028: @code{GO_IF_LEGITIMATE_ADDRESS} are subroutines to check for
8029: acceptable registers for various purposes (one for base registers, one
8030: for index registers, and so on). Then only these subroutine macros
8031: need have two variants; the higher levels of macros may be the same
8032: whether strict or not.@refill
8033:
8034: @item REG_OK_FOR_BASE_P (@var{x})
1.1.1.5 root 8035: A C expression that is nonzero if @var{x} (assumed to be a @code{reg}
1.1 root 8036: RTX) is valid for use as a base register. For hard registers, it
8037: should always accept those which the hardware permits and reject the
8038: others. Whether the macro accepts or rejects pseudo registers must be
8039: controlled by @code{REG_OK_STRICT} as described above. This usually
8040: requires two variant definitions, of which @code{REG_OK_STRICT}
8041: controls the one actually used.
8042:
8043: @item REG_OK_FOR_INDEX_P (@var{x})
1.1.1.5 root 8044: A C expression that is nonzero if @var{x} (assumed to be a @code{reg}
1.1 root 8045: RTX) is valid for use as an index register.
8046:
8047: The difference between an index register and a base register is that
8048: the index register may be scaled. If an address involves the sum of
8049: two registers, neither one of them scaled, then either one may be
8050: labeled the ``base'' and the other the ``index''; but whichever
8051: labeling is used must fit the machine's constraints of which registers
8052: may serve in each capacity. The compiler will try both labelings,
8053: looking for one that is valid, and will reload one or both registers
8054: only if neither labeling works.
8055:
8056: @item LEGITIMIZE_ADDRESS (@var{x}, @var{oldx}, @var{mode}, @var{win})
8057: A C compound statement that attempts to replace @var{x} with a valid
8058: memory address for an operand of mode @var{mode}. @var{win} will be a
8059: C statement label elsewhere in the code; the macro definition may use
8060:
8061: @example
8062: GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{win});
8063: @end example
8064:
8065: @noindent
8066: to avoid further processing if the address has become legitimate.
8067:
8068: @var{x} will always be the result of a call to @code{break_out_memory_refs},
8069: and @var{oldx} will be the operand that was given to that function to produce
8070: @var{x}.
8071:
8072: The code generated by this macro should not alter the substructure of
8073: @var{x}. If it transforms @var{x} into a more legitimate form, it
8074: should assign @var{x} (which will always be a C variable) a new value.
8075:
8076: It is not necessary for this macro to come up with a legitimate
8077: address. The compiler has standard ways of doing so in all cases. In
8078: fact, it is safe for this macro to do nothing. But often a
8079: machine-dependent strategy can generate better code.
8080:
8081: @item GO_IF_MODE_DEPENDENT_ADDRESS (@var{addr}, @var{label})
8082: A C statement or compound statement with a conditional @code{goto
8083: @var{label};} executed if memory address @var{x} (an RTX) can have
8084: different meanings depending on the machine mode of the memory
8085: reference it is used for.
8086:
8087: Autoincrement and autodecrement addresses typically have mode-dependent
8088: effects because the amount of the increment or decrement is the size
8089: of the operand being addressed. Some machines have other mode-dependent
8090: addresses. Many RISC machines have no mode-dependent addresses.
8091:
8092: You may assume that @var{addr} is a valid address for the machine.
8093:
8094: @item LEGITIMATE_CONSTANT_P (@var{x})
8095: A C expression that is nonzero if @var{x} is a legitimate constant for
8096: an immediate operand on the target machine. You can assume that
8097: either @var{x} is a @samp{const_double} or it satisfies
8098: @code{CONSTANT_P}, so you need not check these things. In fact,
8099: @samp{1} is a suitable definition for this macro on machines where any
8100: @samp{const_double} is valid and anything @code{CONSTANT_P} is valid.@refill
8101: @end table
8102:
1.1.1.5 root 8103: @node Cross-compilation, Misc, Addressing Modes, Machine Macros
8104: @section Cross Compilation and Floating-Point Format
8105:
8106: While all modern machines use 2's compliment representation for integers,
8107: there are a variety of representations for floating point numbers. This
8108: means that in a cross-compiler the representation of floating point numbers
8109: in the compiled program may be different from that used in the machine
8110: doing the compilation.
8111:
8112: Because different representation systems may offer different amounts of
8113: range and precision, the cross compiler cannot safely use the host
8114: machine's floating point arithmetic. Therefore, floating point constants
8115: must be represented in the target machine's format. This means that the
8116: cross compiler cannot use @code{atof} to parse a floating point constant;
8117: it must have its own special routine to use instead. Also, constant
8118: folding must emulate the target machine's arithmetic (or must not be done
8119: at all).
8120:
8121: The macros in the following table should be defined only if you are cross
8122: compiling between different floating point formats.
8123:
8124: Otherwise, don't define them. Then default definitions will be set up which
8125: use @code{double} as the data type, @code{==} to test for equality, etc.
8126:
8127: You don't need to worry about how many times you use an operand of any
8128: of these macros. The compiler never uses operands which have side effects.
8129:
8130: @table @code
8131: @item REAL_VALUE_TYPE
8132: A macro for the C data type to be used to hold a floating point value
8133: in the target machine's format. Typically this would be a
8134: @code{struct} containing an array of @code{int}.
8135:
8136: @item REAL_VALUES_EQUAL (@var{x}, @var{y})
8137: A macro for a C expression which compares for equality the two values,
8138: @var{x} and @var{y}, both of type @code{REAL_VALUE_TYPE}.
8139:
8140: @item REAL_VALUES_LESS (@var{x}, @var{y})
8141: A macro for a C expression which tests whether @var{x} is less than
8142: @var{y}, both values being of type @code{REAL_VALUE_TYPE} and
8143: interpreted as floating point numbers in the target machine's
8144: representation.
8145:
8146: @item REAL_VALUE_LDEXP (@var{x}, @var{scale})
8147: A macro for a C expression which performs the standard library
8148: function @code{ldexp}, but using the target machine's floating point
8149: representation. Both @var{x} and the value of the expression have
8150: type @code{REAL_VALUE_TYPE}. The second argument, @var{scale}, is an
8151: integer.
8152:
8153: @item REAL_VALUE_ATOF (@var{string})
8154: A macro for a C expression which converts @var{string}, an expression
8155: of type @code{char *}, into a floating point number in the target
8156: machine's representation. The value has type @code{REAL_VALUE_TYPE}.
8157: @end table
8158:
8159: Define the following additional macros if you want to make floating
8160: point constant folding work while cross compiling. If you don't
8161: define them, cross compilation is still possible, but constant folding
8162: will not happen for floating point values.
8163:
8164: @table @code
8165: @item REAL_ARITHMETIC (@var{output}, @var{code}, @var{x}, @var{y})
8166: A macro for a C statement which calculates an arithmetic operation of
8167: the two floating point values @var{x} and @var{y}, both of type
8168: @code{REAL_VALUE_TYPE} in the target machine's representation, to
8169: produce a result of the same type and representation which is stored
8170: in @var{output} (which will be a variable).
8171:
8172: The operation to be performed is specified by @var{code}, a tree code
8173: which will always be one of the following: @code{PLUS_EXPR},
8174: @code{MINUS_EXPR}, @code{MULT_EXPR}, @code{RDIV_EXPR},
8175: @code{MAX_EXPR}, @code{MIN_EXPR}.@refill
8176:
8177: The expansion of this macro is responsible for checking for overflow.
8178: If overflow happens, the macro expansion should execute the statement
8179: @code{return 0;}, which indicates the inability to perform the
8180: arithmetic operation requested.
8181:
8182: @item REAL_VALUE_NEGATE (@var{x})
8183: A macro for a C expression which returns the negative of the floating
8184: point value @var{x}. Both @var{x} and the value of the expression
8185: have type @code{REAL_VALUE_TYPE} and are in the target machine's
8186: floating point representation.
8187:
8188: There is no way for this macro to report overflow, since overflow
8189: can't happen in the negation operation.
8190:
8191: @item REAL_VALUE_TO_INT (@var{low}, @var{high}, @var{x})
8192: A macro for a C expression which converts a floating point value
8193: @var{x} into a double-precision integer which is then stored into
8194: @var{low} and @var{high}, two variables of type @var{int}.
8195:
8196: @item REAL_VALUE_FROM_INT (@var{x}, @var{low}, @var{high})
8197: A macro for a C expression which converts a double-precision integer
8198: found in @var{low} and @var{high}, two variables of type @var{int},
8199: into a floating point value which is then stored into @var{x}.
8200: @end table
8201:
8202: @node Misc, Condition Code, Cross-compilation, Machine Macros
1.1 root 8203: @section Miscellaneous Parameters
8204:
8205: @table @code
8206: @item CASE_VECTOR_MODE
8207: An alias for a machine mode name. This is the machine mode that
8208: elements of a jump-table should have.
8209:
8210: @item CASE_VECTOR_PC_RELATIVE
8211: Define this macro if jump-tables should contain relative addresses.
8212:
8213: @item CASE_DROPS_THROUGH
8214: Define this if control falls through a @code{case} insn when the index
8215: value is out of range. This means the specified default-label is
8216: actually ignored by the @code{case} insn proper.
8217:
8218: @item IMPLICIT_FIX_EXPR
8219: An alias for a tree code that should be used by default for conversion
8220: of floating point values to fixed point. Normally,
8221: @code{FIX_ROUND_EXPR} is used.@refill
8222:
8223: @item FIXUNS_TRUNC_LIKE_FIX_TRUNC
8224: Define this macro if the same instructions that convert a floating
8225: point number to a signed fixed point number also convert validly to an
8226: unsigned one.
8227:
8228: @item EASY_DIV_EXPR
8229: An alias for a tree code that is the easiest kind of division to
8230: compile code for in the general case. It may be
8231: @code{TRUNC_DIV_EXPR}, @code{FLOOR_DIV_EXPR}, @code{CEIL_DIV_EXPR} or
8232: @code{ROUND_DIV_EXPR}. These four division operators differ in how
8233: they round the result to an integer. @code{EASY_DIV_EXPR} is used
8234: when it is permissible to use any of those kinds of division and the
8235: choice should be made on the basis of efficiency.@refill
8236:
8237: @item DEFAULT_SIGNED_CHAR
8238: An expression whose value is 1 or 0, according to whether the type
8239: @code{char} should be signed or unsigned by default. The user can
8240: always override this default with the options @samp{-fsigned-char}
8241: and @samp{-funsigned-char}.
8242:
8243: @item SCCS_DIRECTIVE
8244: Define this if the preprocessor should ignore @code{#sccs} directives
8245: and print no error message.
8246:
8247: @item IDENT_DIRECTIVE
8248: Define this if the preprocessor should ignore @code{#ident} directives
8249: and print no error message.
8250:
8251: @item MOVE_MAX
8252: The maximum number of bytes that a single instruction can move quickly
8253: from memory to memory.
8254:
8255: @item INT_TYPE_SIZE
8256: A C expression for the size in bits of the type @code{int} on the
8257: target machine.
8258:
8259: @item SLOW_BYTE_ACCESS
8260: Define this macro as a C expression which is nonzero if accessing less
8261: than a word of memory (i.e. a @code{char} or a @code{short}) is slow
8262: (requires more than one instruction).
8263:
8264: @item SLOW_ZERO_EXTEND
8265: Define this macro if zero-extension (of a @code{char} or @code{short}
8266: to an @code{int}) can be done faster if the destination is a register
8267: that is known to be zero.
8268:
8269: If you define this macro, you must have instruction patterns that
8270: recognize RTL structures like this:
8271:
8272: @example
8273: (set (strict-low-part (subreg:QI (reg:SI @dots{}) 0)) @dots{})
8274: @end example
8275:
8276: @noindent
8277: and likewise for @code{HImode}.
8278:
8279: @item SHIFT_COUNT_TRUNCATED
8280: Define this macro if shift instructions ignore all but the lowest few
8281: bits of the shift count. It implies that a sign-extend or zero-extend
8282: instruction for the shift count can be omitted.
8283:
8284: @item TRULY_NOOP_TRUNCATION (@var{outprec}, @var{inprec})
8285: A C expression which is nonzero if on this machine it is safe to
8286: ``convert'' an integer of @var{inprec} bits to one of @var{outprec}
8287: bits (where @var{outprec} is smaller than @var{inprec}) by merely
8288: operating on it as if it had only @var{outprec} bits.
8289:
8290: On many machines, this expression can be 1.
8291:
8292: @item NO_FUNCTION_CSE
8293: Define this macro if it is as good or better to call a constant
8294: function address than to call an address kept in a register.
8295:
8296: @item PROMOTE_PROTOTYPES
8297: Define this macro if an argument declared as @code{char} or
8298: @code{short} in a prototype should actually be passed as an
8299: @code{int}. In addition to avoiding errors in certain cases of
8300: mismatch, it also makes for better code on certain machines.
8301:
8302: @item STORE_FLAG_VALUE
8303: A C expression for the value stored by a store-flag instruction
8304: (@code{s@var{cond}}) when the condition is true. This is usually 1 or
8305: -1; it is required to be an odd number.
8306:
8307: Do not define @code{STORE_FLAG_VALUE} if the machine has no store-flag
8308: instructions.
8309:
8310: @item Pmode
8311: An alias for the machine mode for pointers. Normally the definition
8312: can be
8313:
8314: @example
8315: #define Pmode SImode
8316: @end example
8317:
8318: @item FUNCTION_MODE
8319: An alias for the machine mode used for memory references to functions
8320: being called, in @samp{call} RTL expressions. On most machines this
8321: should be @code{QImode}.
8322:
8323: @item INSN_MACHINE_INFO
8324: This macro should expand into a C structure type to use for the
8325: machine-dependent info field specified with the optional last argument
8326: in @samp{define_insn} and @samp{define_peephole} patterns. For example,
8327: it might expand into @samp{struct machine_info}; then it would be up
8328: to you to define this structure in the @file{tm.h} file.
8329:
8330: You do not need to define this macro if you do not write the optional
8331: last argument in any of the patterns in the machine description.
8332:
8333: @item CONST_COSTS (@var{x}, @var{code})
8334: A part of a C @code{switch} statement that describes the relative
8335: costs of constant RTL expressions. It must contain @code{case} labels
8336: for expression codes @samp{const_int}, @samp{const}, @samp{symbol_ref}, @samp{label_ref}
8337: and @samp{const_double}. Each case must ultimately reach a
8338: @code{return} statement to return the relative cost of the use of that
8339: kind of constant value in an expression. The cost may depend on the
8340: precise value of the constant, which is available for examination in
8341: @var{x}.
8342:
8343: @var{code} is the expression code---redundant, since it can be
8344: obtained with @code{GET_CODE (@var{x})}.
8345:
8346: @item DOLLARS_IN_IDENTIFIERS
8347: Define this to be nonzero if the character @samp{$} should be allowed
8348: by default in identifier names.
8349: @end table
8350:
8351: @node Condition Code, Assembler Format, Misc, Machine Macros
8352: @section Condition Code Information
8353:
8354: The file @file{conditions.h} defines a variable @code{cc_status} to
8355: describe how the condition code was computed (in case the interpretation of
8356: the condition code depends on the instruction that it was set by). This
8357: variable contains the RTL expressions on which the condition code is
8358: currently based, and several standard flags.
8359:
8360: Sometimes additional machine-specific flags must be defined in the machine
8361: description header file. It can also add additional machine-specific
8362: information by defining @code{CC_STATUS_MDEP}.
8363:
8364: @table @code
8365: @item CC_STATUS_MDEP
8366: C code for a data type which is used for declaring the @code{mdep}
8367: component of @code{cc_status}. It defaults to @code{int}.
8368:
8369: @item CC_STATUS_MDEP_INIT
8370: A C expression for the initial value of the @code{mdep} field. It
8371: defaults to 0.
8372:
8373: @item NOTICE_UPDATE_CC (@var{exp}, @var{insn})
8374: A C compound statement to set the components of @code{cc_status}
8375: appropriately for an insn @var{insn} whose body is @var{exp}. It is
8376: this macro's responsibility to recognize insns that set the condition
8377: code as a byproduct of other activity as well as those that explicitly
8378: set @code{(cc0)}.
8379:
8380: If there are insn that do not set the condition code but do alter
8381: other machine registers, this macro must check to see whether they
8382: invalidate the expressions that the condition code is recorded as
8383: reflecting. For example, on the 68000, insns that store in address
8384: registers do not set the condition code, which means that usually
8385: @code{NOTICE_UPDATE_CC} can leave @code{cc_status} unaltered for such
8386: insns. But suppose that the previous insn set the condition code
8387: based on location @samp{a4@@(102)} and the current insn stores a new
8388: value in @samp{a4}. Although the condition code is not changed by
8389: this, it will no longer be true that it reflects the contents of
8390: @samp{a4@@(102)}. Therefore, @code{NOTICE_UPDATE_CC} must alter
8391: @code{cc_status} in this case to say that nothing is known about the
8392: condition code value.
8393:
8394: The definition of @code{NOTICE_UPDATE_CC} must be prepared to deal
8395: with the results of peephole optimization: insns whose patterns are
8396: @samp{parallel} RTXs containing various @samp{reg}, @samp{mem} or
8397: constants which are just the operands. The RTL structure of these
8398: insns is not sufficient to indicate what the insns actually do. What
8399: @code{NOTICE_UPDATE_CC} should do when it sees one is just to run
8400: @code{CC_STATUS_INIT}.
8401: @end table
8402:
8403: @node Assembler Format,, Condition Code, Machine Macros
8404: @section Output of Assembler Code
8405:
8406: @table @code
8407: @item ASM_SPEC
8408: A C string constant that tells the GNU CC driver program options to
8409: pass to the assembler. It can also specify how to translate options
8410: you give to GNU CC into options for GNU CC to pass to the assembler.
8411: See the file @file{tm-sun3.h} for an example of this.
8412:
8413: Do not define this macro if it does not need to do anything.
8414:
8415: @item LINK_SPEC
8416: A C string constant that tells the GNU CC driver program options to
8417: pass to the linker. It can also specify how to translate options you
8418: give to GNU CC into options for GNU CC to pass to the linker.
8419:
8420: Do not define this macro if it does not need to do anything.
8421:
8422: @item LIB_SPEC
8423: Another C string constant used much like @code{LINK_SPEC}. The difference
8424: between the two is that @code{LIBS_SPEC} is used at the end of the
8425: command given to the linker.
8426:
8427: If this macro is not defined, a default is provided that
8428: loads the standard C library from the usual place. See @file{gcc.c}.
8429:
8430: @item STARTFILE_SPEC
8431: Another C string constant used much like @code{LINK_SPEC}. The
8432: difference between the two is that @code{STARTFILE_SPEC} is used at
8433: the very beginning of the command given to the linker.
8434:
8435: If this macro is not defined, a default is provided that loads the
8436: standard C startup file from the usual place. See @file{gcc.c}.
8437:
1.1.1.4 root 8438: @item STANDARD_STARTFILE_PREFIX
8439: Define this macro as a C string constant if you wish to override the
8440: standard choice of @file{/lib/} as the default prefix for where to
8441: find the startup files such as @file{crt0.o}.
8442:
1.1 root 8443: @item ASM_FILE_START (@var{stream})
8444: A C expression which outputs to the stdio stream @var{stream}
8445: some appropriate text to go at the start of an assembler file.
8446:
8447: Normally this macro is defined to output a line containing
8448: @samp{#NO_APP}, which is a comment that has no effect on most
8449: assemblers but tells the GNU assembler that it can save time by not
8450: checking for certain assembler constructs.
8451:
8452: On systems that use SDB, it is necessary to output certain commands;
8453: see @file{tm-attasm.h}.
8454:
8455: @item ASM_APP_ON
8456: A C string constant for text to be output before each @code{asm}
8457: statement or group of consecutive ones. Normally this is
8458: @code{"#APP"}, which is a comment that has no effect on most
8459: assemblers but tells the GNU assembler that it must check the lines
8460: that follow for all valid assembler constructs.
8461:
8462: @item ASM_APP_OFF
8463: A C string constant for text to be output after each @code{asm}
8464: statement or group of consecutive ones. Normally this is
8465: @code{"#NO_APP"}, which tells the GNU assembler to resume making the
8466: time-saving assumptions that are valid for ordinary compiler output.
8467:
8468: @item TEXT_SECTION_ASM_OP
8469: A C string constant for the assembler operation that should precede
8470: instructions and read-only data. Normally @code{".text"} is right.
8471:
8472: @item DATA_SECTION_ASM_OP
8473: A C string constant for the assembler operation to identify the
8474: following data as writable initialized data. Normally @code{".data"}
8475: is right.
8476:
8477: @item REGISTER_NAMES
8478: A C initializer containing the assembler's names for the machine
8479: registers, each one as a C string constant. This is what translates
8480: register numbers in the compiler into assembler language.
8481:
8482: @item DBX_REGISTER_NUMBER (@var{regno})
8483: A C expression that returns the DBX register number for the compiler
8484: register number @var{regno}. In simple cases, the value of this
8485: expression may be @var{regno} itself. But sometimes there are some
8486: registers that the compiler knows about and DBX does not, or vice
8487: versa. In such cases, some register may need to have one number in
8488: the compiler and another for DBX.
8489:
8490: @item DBX_DEBUGGING_INFO
8491: Define this macro if GNU CC should produce debugging output for DBX
8492: in response to the @samp{-g} option.
8493:
8494: @item SDB_DEBUGGING_INFO
8495: Define this macro if GNU CC should produce debugging output for SDB
8496: in response to the @samp{-g} option.
8497:
8498: @item PUT_SDB_@var{op}
8499: Define these macros to override the assembler syntax for the special
8500: SDB assembler directives. See @file{sdbout.c} for a list of these
8501: macros and their arguments. If the standard syntax is used, you need
8502: not define them yourself.
8503:
8504: @item SDB_GENERATE_FAKE
8505: Define this macro to override the usual method of constructing a dummy
8506: name for anonymous structure and union types. See @file{sdbout.c} for
8507: more infomation.
8508:
8509: @item DBX_NO_XREFS
8510: Define this macro if DBX on your system does not support the construct
8511: @samp{xs@var{tagname}}. On some systems, this construct is used to
8512: describe a forward reference to a structure named @var{tagname}.
8513: On other systems, this construct is not supported at all.
8514:
8515: @item DBX_CONTIN_LENGTH
8516: A symbol name in DBX-format debugging information is normally
8517: continued (split into two separate @code{.stabs} directives) when it
8518: exceeds a certain length (by default, 80 characters). On some
8519: operating systems, DBX requires this splitting; on others, splitting
8520: must not be done. You can inhibit splitting by defining this macro
8521: with the value zero. You can override the default splitting-length by
8522: defining this macro as an expression for the length you desire.
8523:
8524: @item DBX_CONTIN_CHAR
8525: Normally continuation is indicated by adding a @samp{\} character to
8526: the end of a @code{.stabs} string when a continuation follows. To use
8527: a different character instead, define this macro as a character
8528: constant for the character you want to use. Do not define this macro
8529: if backslash is correct for your system.
8530:
8531: @item ASM_OUTPUT_LABEL (@var{stream}, @var{name})
8532: A C statement (sans semicolon) to output to the stdio stream
8533: @var{stream} the assembler definition of a label named @var{name}. Use
8534: the expression @code{assemble_name (@var{stream}, @var{name})} to output
8535: the name itself; before and after that, output the additional
8536: assembler syntax for defining the name, and a newline.
8537:
8538: @item ASM_DECLARE_FUNCTION_NAME (@var{stream}, @var{name}, @var{decl})
8539: A C statement (sans semicolon) to output to the stdio stream
8540: @var{stream} any text necessary for declaring the name @var{name} of a
8541: function which is being defined. This macro is responsible for
8542: outputting the label definition (perhaps using
8543: @code{ASM_OUTPUT_LABEL}). The argument @var{decl} is the
8544: @code{FUNCTION_DECL} tree node representing the function.
8545:
8546: If this macro is not defined, then the function name is defined in the
8547: usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}).
8548:
8549: @item ASM_GLOBALIZE_LABEL (@var{stream}, @var{name})
8550: A C statement (sans semicolon) to output to the stdio stream
8551: @var{stream} some commands that will make the label @var{name} global;
8552: that is, available for reference from other files. Use the expression
8553: @code{assemble_name (@var{stream}, @var{name})} to output the name
8554: itself; before and after that, output the additional assembler syntax
8555: for making that name global, and a newline.
8556:
8557: @item ASM_OUTPUT_EXTERNAL (@var{stream}, @var{name}, @var{decl})
8558: A C statement (sans semicolon) to output to the stdio stream
8559: @var{stream} any text necessary for declaring the name of an external
8560: symbol named @var{name} which is referenced in this compilation but
8561: not defined. The value of @var{decl} is the tree node for the
8562: declaration.
8563:
8564: This macro need not be defined if it does not need to output anything.
8565: The GNU assembler and most Unix assemblers don't require anything.
8566:
8567: @item ASM_OUTPUT_LABELREF (@var{stream}, @var{name})
8568: A C statement to output to the stdio stream @var{stream} a reference in
8569: assembler syntax to a label named @var{name}. The character @samp{_}
8570: should be added to the front of the name, if that is customary on your
8571: operating system, as it is in most Berkeley Unix systems. This macro
8572: is used in @code{assemble_name}.
8573:
8574: @item ASM_GENERATE_INTERNAL_LABEL (@var{string}, @var{prefix}, @var{num})
8575: A C statement to store into the string @var{string} a label whose
8576: name is made from the string @var{prefix} and the number @var{num}.
8577:
8578: This string, when output subsequently by @code{ASM_OUTPUT_LABELREF},
8579: should produce the same output that @code{ASM_OUTPUT_INTERNAL_LABEL}
8580: would produce with the same @var{prefix} and @var{num}.
8581:
8582: @item ASM_OUTPUT_INTERNAL_LABEL (@var{stream}, @var{prefix}, @var{num})
8583: A C statement to output to the stdio stream @var{stream} a label whose
8584: name is made from the string @var{prefix} and the number @var{num}.
8585: These labels are used for internal purposes, and there is no reason
8586: for them to appear in the symbol table of the object file. On many
8587: systems, the letter @samp{L} at the beginning of a label has this
8588: effect. The usual definition of this macro is as follows:
8589:
8590: @example
8591: fprintf (@var{stream}, "L%s%d:\n", @var{prefix}, @var{num})
8592: @end example
8593:
8594: @item ASM_OUTPUT_CASE_LABEL (@var{stream}, @var{prefix}, @var{num}, @var{table})
8595: Define this if the label before a jump-table needs to be output
8596: specially. The first three arguments are the same as for
8597: @code{ASM_OUTPUT_INTERNAL_LABEL}; the fourth argument is the
8598: jump-table which follows (a @samp{jump_insn} containing an
8599: @samp{addr_vec} or @samp{addr_diff_vec}).
8600:
8601: This feature is used on system V to output a @code{swbeg} statement
8602: for the table.
8603:
8604: If this macro is not defined, these labels are output with
8605: @code{ASM_OUTPUT_INTERNAL_LABEL}.
8606:
8607: @item ASM_OUTPUT_CASE_END (@var{stream}, @var{num}, @var{table})
8608: Define this if something special must be output at the end of a jump-table.
8609: The definition should be a C statement to be executed after the assembler
8610: code for the table is written. It should write the appropriate code to
8611: stdio stream @var{stream}. The argument @var{table} is the jump-table
8612: insn, and @var{num} is the label-number of the preceding label.
8613:
8614: If this macro is not defined, nothing special is output at the end of
8615: the jump-table.
8616:
1.1.1.4 root 8617: @item ASM_OUTPUT_ALIGN_CODE (@var{file})
8618: A C expression to output text to align the location counter in the way
8619: that is desirable at a point in the code that is reached only by
8620: jumping.
8621:
8622: This macro need not be defined if you don't want any special alignment
8623: to be done at such a time. Most machine descriptions do not currently
8624: define the macro.
8625:
1.1 root 8626: @item ASM_FORMAT_PRIVATE_NAME (@var{outvar}, @var{name}, @var{number})
8627: A C expression to assign to @var{outvar} (which is a variable of type
8628: @code{char *}) a newly allocated string made from the string
8629: @var{name} and the number @var{number}, with some suitable punctuation
8630: added. Use @code{alloca} to get space for the string.
8631:
8632: This string will be used as the argument to @code{ASM_OUTPUT_LABELREF}
8633: to produce an assembler label for an internal static variable whose
8634: name is @var{name}. Therefore, the string must be such as to result
8635: in valid assembler code. The argument @var{number} is different each
8636: time this macro is executed; it prevents conflicts between
8637: similarly-named internal static variables in different scopes.
8638:
8639: Ideally this string should not be a valid C identifier, to prevent any
8640: conflict with the user's own symbols. Most assemblers allow periods
8641: or percent signs in assembler symbols; putting at least one of these
8642: between the name and the number will suffice.
8643:
8644: @item ASM_OUTPUT_REG_PUSH (@var{stream}, @var{regno})
8645: A C expression to output to @var{stream} some assembler code
8646: which will push hard register number @var{regno} onto the stack.
8647: The code need not be optimal, since this macro is used only when
8648: profiling.
8649:
8650: @item ASM_OUTPUT_REG_POP (@var{stream}, @var{regno})
8651: A C expression to output to @var{stream} some assembler code
8652: which will pop hard register number @var{regno} off of the stack.
8653: The code need not be optimal, since this macro is used only when
8654: profiling.
8655:
8656: @item ASM_OUTPUT_ADDR_DIFF_ELT (@var{stream}, @var{value}, @var{rel})
8657: This macro should be provided on machines where the addresses
8658: in a dispatch table are relative to the table's own address.
8659:
8660: The definition should be a C statement to output to the stdio stream
8661: @var{stream} an assembler pseudo-instruction to generate a difference
8662: between two labels. @var{value} and @var{rel} are the numbers of two
8663: internal labels. The definitions of these labels are output using
8664: @code{ASM_OUTPUT_INTERNAL_LABEL}, and they must be printed in the same
8665: way here. For example,
8666:
8667: @example
8668: fprintf (@var{stream}, "\t.word L%d-L%d\n",
8669: @var{value}, @var{rel})
8670: @end example
8671:
8672: @item ASM_OUTPUT_ADDR_VEC_ELT (@var{stream}, @var{value})
8673: This macro should be provided on machines where the addresses
8674: in a dispatch table are absolute.
8675:
8676: The definition should be a C statement to output to the stdio stream
8677: @var{stream} an assembler pseudo-instruction to generate a reference to
8678: a label. @var{value} is the number of an internal label whose
8679: definition is output using @code{ASM_OUTPUT_INTERNAL_LABEL}.
8680: For example,
8681:
8682: @example
8683: fprintf (@var{stream}, "\t.word L%d\n", @var{value})
8684: @end example
8685:
8686: @item ASM_OUTPUT_DOUBLE (@var{stream}, @var{value})
8687: A C statement to output to the stdio stream @var{stream} an assembler
8688: instruction to assemble a @code{double} constant whose value is
8689: @var{value}. @var{value} will be a C expression of type
8690: @code{double}.
8691:
8692: @item ASM_OUTPUT_FLOAT (@var{stream}, @var{value})
8693: A C statement to output to the stdio stream @var{stream} an assembler
8694: instruction to assemble a @code{float} constant whose value is
8695: @var{value}. @var{value} will be a C expression of type @code{float}.
8696:
8697: @item ASM_OUTPUT_INT (@var{stream}, @var{exp})
8698: @itemx ASM_OUTPUT_SHORT (@var{stream}, @var{exp})
8699: @itemx ASM_OUTPUT_CHAR (@var{stream}, @var{exp})
8700: A C statement to output to the stdio stream @var{stream} an assembler
8701: instruction to assemble a @code{int}, @code{short} or @code{char}
8702: constant whose value is @var{value}. The argument @var{exp} will be
8703: an RTL expression which represents a constant value. Use
8704: @samp{output_addr_const (@var{exp})} to output this value as an
8705: assembler expression.@refill
8706:
8707: @item ASM_OUTPUT_BYTE (@var{stream}, @var{value})
8708: A C statement to output to the stdio stream @var{stream} an assembler
8709: instruction to assemble a single byte containing the number @var{value}.
8710:
8711: @item ASM_OUTPUT_ASCII (@var{stream}, @var{ptr}, @var{len})
8712: A C statement to output to the stdio stream @var{stream} an assembler
8713: instruction to assemble a string constant containing the @var{len}
8714: bytes at @var{ptr}. @var{ptr} will be a C expression of type
8715: @code{char *} and @var{len} a C expression of type @code{int}.
8716:
8717: If the assembler has a @code{.ascii} pseudo-op as found in the
8718: Berkeley Unix assembler, do not define the macro
8719: @code{ASM_OUTPUT_ASCII}.
8720:
8721: @item ASM_OUTPUT_SKIP (@var{stream}, @var{nbytes})
8722: A C statement to output to the stdio stream @var{stream} an assembler
8723: instruction to advance the location counter by @var{nbytes} bytes.
8724: @var{nbytes} will be a C expression of type @code{int}.
8725:
8726: @item ASM_OUTPUT_ALIGN (@var{stream}, @var{power})
8727: A C statement to output to the stdio stream @var{stream} an assembler
8728: instruction to advance the location counter to a multiple of 2 to the
8729: @var{power} bytes. @var{power} will be a C expression of type @code{int}.
8730:
8731: @item ASM_OUTPUT_COMMON (@var{stream}, @var{name}, @var{size})
8732: A C statement (sans semicolon) to output to the stdio stream
8733: @var{stream} the assembler definition of a common-label named @var{name}
8734: whose size is @var{size} bytes. Use the expression
8735: @code{assemble_name (@var{stream}, @var{name})} to output the name
8736: itself; before and after that, output the additional assembler syntax
8737: for defining the name, and a newline.
8738:
8739: This macro controls how the assembler definitions of uninitialized
8740: global variables are output.
8741:
8742: @item ASM_OUTPUT_LOCAL (@var{stream}, @var{name}, @var{size})
8743: A C statement (sans semicolon) to output to the stdio stream
8744: @var{stream} the assembler definition of a local-common-label named
8745: @var{name} whose size is @var{size} bytes. Use the expression
8746: @code{assemble_name (@var{stream}, @var{name})} to output the name
8747: itself; before and after that, output the additional assembler syntax
8748: for defining the name, and a newline.
8749:
8750: This macro controls how the assembler definitions of uninitialized
8751: static variables are output.
8752:
8753: @item ASM_OUTPUT_SOURCE_LINE (@var{stream}, @var{line})
8754: A C statment to output DBX or SDB debugging information before code
8755: for line number @var{line} of the current source file to the
8756: stdio stream @var{stream}.
8757:
8758: This macro need not be defined if the standard form of debugging
8759: information for the debugger in use is appropriate.
8760:
8761: @item ASM_OUTPUT_IDENT (@var{stream}, @var{string})
8762: A C statement to output something to the assembler file to handle a
8763: @samp{#ident} directive containing the text @var{string}. If this
8764: macro is not defined, the assembler code @samp{.ident "@var{string}"}
8765: will be output by default.
8766:
8767: This macro is significant only if @code{IDENT_DIRECTIVE} is defined.
8768:
8769: @item TARGET_BELL
8770: A C constant expression for the integer value for escape sequence
8771: @samp{\a}.
8772:
8773: @item TARGET_BS
8774: @itemx TARGET_TAB
8775: @itemx TARGET_NEWLINE
8776: C constant expressions for the integer values for escape sequences
8777: @samp{\b}, @samp{\t} and @samp{\n}.
8778:
8779: @item TARGET_VT
8780: @itemx TARGET_FF
8781: @itemx TARGET_CR
8782: C constant expressions for the integer values for escape sequences
8783: @samp{\v}, @samp{\f} and @samp{\r}.
8784:
8785: @item ASM_OUTPUT_OPCODE (@var{stream}, @var{ptr})
8786: Define this macro if you are using an unusual assembler that
8787: requires different names for the machine instructions.
8788:
8789: The definition is a C statement or statements which output an
8790: assembler instruction opcode to the stdio stream @var{stream}. The
8791: macro-operand @var{ptr} is a variable of type @code{char *} which
8792: points to the opcode name in its ``internal'' form---the form that is
8793: written in the machine description. The definition should output the
8794: opcode name to @var{stream}, performing any translation you desire, and
8795: increment the variable @var{ptr} to point at the end of the opcode
8796: so that it will not be output twice.
8797:
8798: In fact, your macro definition may process less than the entire opcode
8799: name, or more than the opcode name; but if you want to process text
8800: that includes @samp{%}-sequences to substitute operands, you must take
8801: care of the substitution yourself. Just be sure to increment
8802: @var{ptr} over whatever text should not be output normally.
8803:
8804: If the macro definition does nothing, the instruction is output
8805: in the usual way.
8806:
8807: @item FINAL_PRESCAN_INSN (@var{insn}, @var{opvec}, @var{noperands})
8808: If defined, a C statement to be executed just prior to the output of
8809: assembler code for @var{insn}, to modify the extracted operands so
8810: they will be output differently.
8811:
8812: Here the argument @var{opvec} is the vector containing the operands
8813: extracted from @var{insn}, and @var{noperands} is the number of
8814: elements of the vector which contain meaningful data for this insn.
8815: The contents of this vector are what will be used to convert the insn
8816: template into assembler code, so you can change the assembler output
8817: by changing the contents of the vector.
8818:
8819: This macro is useful when various assembler syntaxes share a single
8820: file of instruction patterns; by defining this macro differently, you
8821: can cause a large class of instructions to be output differently (such
8822: as with rearranged operands). Naturally, variations in assembler
8823: syntax affecting individual insn patterns ought to be handled by
8824: writing conditional output routines in those patterns.
8825:
8826: If this macro is not defined, it is equivalent to a null statement.
8827:
8828: @item PRINT_OPERAND (@var{stream}, @var{x}, @var{code})
8829: A C compound statement to output to stdio stream @var{stream} the
8830: assembler syntax for an instruction operand @var{x}. @var{x} is an
8831: RTL expression.
8832:
8833: @var{code} is a value that can be used to specify one of several ways
8834: of printing the operand. It is used when identical operands must be
8835: printed differently depending on the context. @var{code} comes from
8836: the @samp{%} specification that was used to request printing of the
8837: operand. If the specification was just @samp{%@var{digit}} then
8838: @var{code} is 0; if the specification was @samp{%@var{ltr}
8839: @var{digit}} then @var{code} is the ASCII code for @var{ltr}.
8840:
8841: If @var{x} is a register, this macro should print the register's name.
8842: The names can be found in an array @code{reg_names} whose type is
8843: @code{char *[]}. @code{reg_names} is initialized from
8844: @code{REGISTER_NAMES}.
8845:
8846: When the machine description has a specification @samp{%@var{punct}}
8847: (a @samp{%} followed by a punctuation character), this macro is called
8848: with a null pointer for @var{x} and the punctuation character for
8849: @var{code}.
8850:
8851: @item PRINT_OPERAND_ADDRESS (@var{stream}, @var{x})
8852: A C compound statement to output to stdio stream @var{stream} the
8853: assembler syntax for an instruction operand that is a memory reference
8854: whose address is @var{x}. @var{x} is an RTL expression.
8855:
8856: @item ASM_OPEN_PAREN
8857: @itemx ASM_CLOSE_PAREN
8858: These macros are defined as C string constant, describing the syntax
8859: in the assembler for grouping arithmetic expressions. The following
8860: definitions are correct for most assemblers:
8861:
8862: @example
8863: #define ASM_OPEN_PAREN "("
8864: #define ASM_CLOSE_PAREN ")"
8865: @end example
8866: @end table
8867:
8868: @node Config,, Machine Macros, Top
8869: @chapter The Configuration File
8870:
1.1.1.3 root 8871: The configuration file @file{xm-@var{machine}.h} contains macro definitions
8872: that describe the machine and system on which the compiler is running.
8873: Most of the values in it are actually the same on all machines that GNU CC
8874: runs on, so large parts of all configuration files are identical. But
1.1 root 8875: there are some macros that vary:
8876:
8877: @table @code
8878: @item FAILURE_EXIT_CODE
8879: A C expression for the status code to be returned when the compiler
8880: exits after serious errors.
8881:
8882: @item SUCCESS_EXIT_CODE
8883: A C expression for the status code to be returned when the compiler
8884: exits without serious errors.
8885: @end table
8886:
1.1.1.3 root 8887: In addition, configuration files for system V define @code{bcopy},
8888: @code{bzero} and @code{bcmp} as aliases. Some files define @code{alloca}
8889: as a macro when compiled with GNU CC, in order to take advantage of the
8890: benefit of GNU CC's built-in @code{alloca}.
8891:
1.1 root 8892: @contents
8893: @bye
This archive runs on limited infrastructure. Preserving old code on modern bandwidth. Automated agents are requested to crawl responsibly.