![[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:
1.1.1.10 root 9: Copyright (C) 1988, 1989, 1990 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
1.1.1.9 root 24: sections entitled ``GNU General Public License'' and ``Protect Your
25: Freedom---Fight `Look And Feel'@w{}'' are included exactly as in the
26: original, and provided that the entire resulting derived work is
27: distributed under the terms of a permission notice identical to this
28: one.
1.1 root 29:
30: Permission is granted to copy and distribute translations of this manual
31: into another language, under the above conditions for modified versions,
1.1.1.9 root 32: except that the sections entitled ``GNU General Public License'' and
33: ``Protect Your Freedom---Fight `Look And Feel'@w{}'' and this permission
34: notice may be included in translations approved by the Free Software
35: Foundation instead of in the original English.
1.1 root 36: @end ifinfo
37:
38: @setchapternewpage odd
39:
40: @titlepage
41: @center @titlefont{Using and Porting GNU CC}
42: @sp 2
43: @center Richard M. Stallman
44: @sp 3
1.1.1.11! root 45: @center last updated 15 January 1991
1.1 root 46: @sp 1
1.1.1.11! root 47: @center for version 1.39
1.1 root 48: @page
49: @vskip 0pt plus 1filll
1.1.1.10 root 50: Copyright @copyright{} 1988, 1989, 1990 Free Software Foundation, Inc.
1.1 root 51:
52: Permission is granted to make and distribute verbatim copies of
53: this manual provided the copyright notice and this permission notice
54: are preserved on all copies.
55:
56: Permission is granted to copy and distribute modified versions of this
57: manual under the conditions for verbatim copying, provided also that the
1.1.1.9 root 58: sections entitled ``GNU General Public License'' and ``Protect Your
59: Freedom---Fight `Look And Feel'@w{}'' are included exactly as in the
60: original, and provided that the entire resulting derived work is
61: distributed under the terms of a permission notice identical to this
62: one.
1.1 root 63:
64: Permission is granted to copy and distribute translations of this manual
65: into another language, under the above conditions for modified versions,
1.1.1.9 root 66: except that the sections entitled ``GNU General Public License'' and
67: ``Protect Your Freedom---Fight `Look And Feel'@w{}'' and this permission
68: notice may be included in translations approved by the Free Software
69: Foundation instead of in the original English.
1.1 root 70: @end titlepage
71: @page
72:
73: @ifinfo
74: @node Top, Copying,, (DIR)
75: @ichapter Introduction
76:
77: This manual documents how to run, install and port the GNU C compiler, as
78: well as its new features and incompatibilities, and how to report bugs.
79:
80: @end ifinfo
81: @menu
1.1.1.6 root 82: * Copying:: GNU General Public License says
1.1 root 83: how you can copy and share GNU CC.
84: * Contributors:: People who have contributed to GNU CC.
1.1.1.9 root 85: * Boycott:: Protect your freedom---fight ``look and feel''.
1.1 root 86: * Options:: Command options supported by @samp{gcc}.
87: * Installation:: How to configure, compile and install GNU CC.
88: * Trouble:: If you have trouble installing GNU CC.
1.1.1.10 root 89: * Service:: How to find suppliers of services for GNU CC users.
1.1 root 90: * Incompatibilities:: Incompatibilities of GNU CC.
91: * Extensions:: GNU extensions to the C language.
92: * Bugs:: How to report bugs (if you want to get them fixed).
93: * Portability:: Goals of GNU CC's portability features.
94: * Interface:: Function-call interface of GNU CC output.
95: * Passes:: Order of passes, what they do, and what each file is for.
96: * RTL:: The intermediate representation that most passes work on.
97: * Machine Desc:: How to write machine description instruction patterns.
98: * Machine Macros:: How to write the machine description C macros.
1.1.1.8 root 99: * Config:: Writing the @file{xm-@var{machine}.h} file.
1.1 root 100: @end menu
101:
102: @node Copying, Contributors, Top, Top
1.1.1.6 root 103: @unnumbered GNU GENERAL PUBLIC LICENSE
104: @center Version 1, February 1989
1.1 root 105:
1.1.1.6 root 106: @display
107: Copyright @copyright{} 1989 Free Software Foundation, Inc.
108: 675 Mass Ave, Cambridge, MA 02139, USA
109:
110: Everyone is permitted to copy and distribute verbatim copies
111: of this license document, but changing it is not allowed.
112: @end display
113:
114: @unnumberedsec Preamble
115:
116: The license agreements of most software companies try to keep users
117: at the mercy of those companies. By contrast, our General Public
118: License is intended to guarantee your freedom to share and change free
119: software---to make sure the software is free for all its users. The
120: General Public License applies to the Free Software Foundation's
121: software and to any other program whose authors commit to using it.
122: You can use it for your programs, too.
123:
124: When we speak of free software, we are referring to freedom, not
125: price. Specifically, the General Public License is designed to make
126: sure that you have the freedom to give away or sell copies of free
127: software, that you receive source code or can get it if you want it,
128: that you can change the software or use pieces of it in new free
129: programs; and that you know you can do these things.
130:
131: To protect your rights, we need to make restrictions that forbid
132: anyone to deny you these rights or to ask you to surrender the rights.
133: These restrictions translate to certain responsibilities for you if you
134: distribute copies of the software, or if you modify it.
135:
136: For example, if you distribute copies of a such a program, whether
137: gratis or for a fee, you must give the recipients all the rights that
138: you have. You must make sure that they, too, receive or can get the
1.1 root 139: source code. And you must tell them their rights.
140:
1.1.1.6 root 141: We protect your rights with two steps: (1) copyright the software, and
142: (2) offer you this license which gives you legal permission to copy,
143: distribute and/or modify the software.
144:
145: Also, for each author's protection and ours, we want to make certain
146: that everyone understands that there is no warranty for this free
147: software. If the software is modified by someone else and passed on, we
148: want its recipients to know that what they have is not the original, so
149: that any problems introduced by others will not reflect on the original
150: authors' reputations.
1.1 root 151:
1.1.1.6 root 152: The precise terms and conditions for copying, distribution and
153: modification follow.
1.1 root 154:
1.1.1.6 root 155: @iftex
156: @unnumberedsec TERMS AND CONDITIONS
157: @end iftex
158: @ifinfo
159: @center TERMS AND CONDITIONS
160: @end ifinfo
1.1 root 161:
1.1.1.6 root 162: @enumerate
1.1 root 163: @item
1.1.1.6 root 164: This License Agreement applies to any program or other work which
165: contains a notice placed by the copyright holder saying it may be
166: distributed under the terms of this General Public License. The
167: ``Program'', below, refers to any such program or work, and a ``work based
168: on the Program'' means either the Program or any work containing the
169: Program or a portion of it, either verbatim or with modifications. Each
170: licensee is addressed as ``you''.
171:
172: @item
173: You may copy and distribute verbatim copies of the Program's source
174: code as you receive it, in any medium, provided that you conspicuously and
175: appropriately publish on each copy an appropriate copyright notice and
176: disclaimer of warranty; keep intact all the notices that refer to this
177: General Public License and to the absence of any warranty; and give any
178: other recipients of the Program a copy of this General Public License
179: along with the Program. You may charge a fee for the physical act of
180: transferring a copy.
181:
182: @item
183: You may modify your copy or copies of the Program or any portion of
184: it, and copy and distribute such modifications under the terms of Paragraph
185: 1 above, provided that you also do the following:
1.1 root 186:
187: @itemize @bullet
188: @item
1.1.1.6 root 189: cause the modified files to carry prominent notices stating that
190: you changed the files and the date of any change; and
1.1 root 191:
192: @item
193: cause the whole of any work that you distribute or publish, that
1.1.1.6 root 194: in whole or in part contains the Program or any part thereof, either
195: with or without modifications, to be licensed at no charge to all
196: third parties under the terms of this General Public License (except
197: that you may choose to grant warranty protection to some or all
198: third parties, at your option).
199:
200: @item
201: If the modified program normally reads commands interactively when
202: run, you must cause it, when started running for such interactive use
203: in the simplest and most usual way, to print or display an
204: announcement including an appropriate copyright notice and a notice
205: that there is no warranty (or else, saying that you provide a
206: warranty) and that users may redistribute the program under these
207: conditions, and telling the user how to view a copy of this General
208: Public License.
209:
210: @item
211: You may charge a fee for the physical act of transferring a
212: copy, and you may at your option offer warranty protection in
213: exchange for a fee.
1.1 root 214: @end itemize
215:
1.1.1.6 root 216: Mere aggregation of another independent work with the Program (or its
1.1 root 217: derivative) on a volume of a storage or distribution medium does not bring
1.1.1.6 root 218: the other work under the scope of these terms.
1.1 root 219:
220: @item
1.1.1.6 root 221: You may copy and distribute the Program (or a portion or derivative of
222: it, under Paragraph 2) in object code or executable form under the terms of
223: Paragraphs 1 and 2 above provided that you also do one of the following:
1.1 root 224:
225: @itemize @bullet
226: @item
227: accompany it with the complete corresponding machine-readable
228: source code, which must be distributed under the terms of
229: Paragraphs 1 and 2 above; or,
230:
231: @item
232: accompany it with a written offer, valid for at least three
1.1.1.6 root 233: years, to give any third party free (except for a nominal charge
234: for the cost of distribution) a complete machine-readable copy of the
1.1 root 235: corresponding source code, to be distributed under the terms of
236: Paragraphs 1 and 2 above; or,
237:
238: @item
239: accompany it with the information you received as to where the
240: corresponding source code may be obtained. (This alternative is
241: allowed only for noncommercial distribution and only if you
242: received the program in object code or executable form alone.)
243: @end itemize
244:
1.1.1.6 root 245: Source code for a work means the preferred form of the work for making
246: modifications to it. For an executable file, complete source code means
247: all the source code for all modules it contains; but, as a special
248: exception, it need not include source code for modules which are standard
249: libraries that accompany the operating system on which the executable
250: file runs, or for standard header files or definitions files that
251: accompany that operating system.
252:
253: @item
254: You may not copy, modify, sublicense, distribute or transfer the
255: Program except as expressly provided under this General Public License.
256: Any attempt otherwise to copy, modify, sublicense, distribute or transfer
257: the Program is void, and will automatically terminate your rights to use
258: the Program under this License. However, parties who have received
259: copies, or rights to use copies, from you under this General Public
260: License will not have their licenses terminated so long as such parties
261: remain in full compliance.
262:
263: @item
264: By copying, distributing or modifying the Program (or any work based
265: on the Program) you indicate your acceptance of this license to do so,
266: and all its terms and conditions.
267:
268: @item
269: Each time you redistribute the Program (or any work based on the
270: Program), the recipient automatically receives a license from the original
271: licensor to copy, distribute or modify the Program subject to these
272: terms and conditions. You may not impose any further restrictions on the
273: recipients' exercise of the rights granted herein.
274:
275: @item
276: The Free Software Foundation may publish revised and/or new versions
277: of the General Public License from time to time. Such new versions will
278: be similar in spirit to the present version, but may differ in detail to
279: address new problems or concerns.
280:
281: Each version is given a distinguishing version number. If the Program
282: specifies a version number of the license which applies to it and ``any
283: later version'', you have the option of following the terms and conditions
284: either of that version or of any later version published by the Free
285: Software Foundation. If the Program does not specify a version number of
286: the license, you may choose any version ever published by the Free Software
287: Foundation.
288:
289: @item
290: If you wish to incorporate parts of the Program into other free
291: programs whose distribution conditions are different, write to the author
292: to ask for permission. For software which is copyrighted by the Free
293: Software Foundation, write to the Free Software Foundation; we sometimes
294: make exceptions for this. Our decision will be guided by the two goals
295: of preserving the free status of all derivatives of our free software and
296: of promoting the sharing and reuse of software generally.
297:
298: @iftex
299: @heading NO WARRANTY
300: @end iftex
301: @ifinfo
302: @center NO WARRANTY
303: @end ifinfo
304:
305: @item
306: BECAUSE THE PROGRAM IS LICENSED FREE OF CHARGE, THERE IS NO WARRANTY
307: FOR THE PROGRAM, TO THE EXTENT PERMITTED BY APPLICABLE LAW. EXCEPT WHEN
308: OTHERWISE STATED IN WRITING THE COPYRIGHT HOLDERS AND/OR OTHER PARTIES
309: PROVIDE THE PROGRAM ``AS IS'' WITHOUT WARRANTY OF ANY KIND, EITHER EXPRESSED
310: OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
311: MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. THE ENTIRE RISK AS
312: TO THE QUALITY AND PERFORMANCE OF THE PROGRAM IS WITH YOU. SHOULD THE
313: PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF ALL NECESSARY SERVICING,
314: REPAIR OR CORRECTION.
315:
316: @item
317: IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING WILL
318: ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MAY MODIFY AND/OR
319: REDISTRIBUTE THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES,
320: INCLUDING ANY GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES
321: ARISING OUT OF THE USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT
322: LIMITED TO LOSS OF DATA OR DATA BEING RENDERED INACCURATE OR LOSSES
323: SUSTAINED BY YOU OR THIRD PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE
324: WITH ANY OTHER PROGRAMS), EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN
325: ADVISED OF THE POSSIBILITY OF SUCH DAMAGES.
1.1 root 326: @end enumerate
327:
1.1.1.6 root 328: @iftex
329: @heading END OF TERMS AND CONDITIONS
330: @end iftex
331: @ifinfo
332: @center END OF TERMS AND CONDITIONS
333: @end ifinfo
334:
335: @page
336: @unnumberedsec Appendix: How to Apply These Terms to Your New Programs
337:
338: If you develop a new program, and you want it to be of the greatest
339: possible use to humanity, the best way to achieve this is to make it
340: free software which everyone can redistribute and change under these
341: terms.
342:
343: To do so, attach the following notices to the program. It is safest to
344: attach them to the start of each source file to most effectively convey
345: the exclusion of warranty; and each file should have at least the
346: ``copyright'' line and a pointer to where the full notice is found.
347:
348: @smallexample
349: @var{one line to give the program's name and a brief idea of what it does.}
350: Copyright (C) 19@var{yy} @var{name of author}
351:
352: This program is free software; you can redistribute it and/or modify
353: it under the terms of the GNU General Public License as published by
354: the Free Software Foundation; either version 1, or (at your option)
355: any later version.
356:
357: This program is distributed in the hope that it will be useful,
358: but WITHOUT ANY WARRANTY; without even the implied warranty of
359: MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
360: GNU General Public License for more details.
361:
362: You should have received a copy of the GNU General Public License
363: along with this program; if not, write to the Free Software
364: Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
365: @end smallexample
366:
367: Also add information on how to contact you by electronic and paper mail.
368:
369: If the program is interactive, make it output a short notice like this
370: when it starts in an interactive mode:
371:
372: @smallexample
373: Gnomovision version 69, Copyright (C) 19@var{yy} @var{name of author}
374: Gnomovision comes with ABSOLUTELY NO WARRANTY; for details type `show w'.
375: This is free software, and you are welcome to redistribute it
376: under certain conditions; type `show c' for details.
377: @end smallexample
378:
379: The hypothetical commands `show w' and `show c' should show the
380: appropriate parts of the General Public License. Of course, the
381: commands you use may be called something other than `show w' and `show
382: c'; they could even be mouse-clicks or menu items---whatever suits your
383: program.
384:
385: You should also get your employer (if you work as a programmer) or your
386: school, if any, to sign a ``copyright disclaimer'' for the program, if
387: necessary. Here a sample; alter the names:
388:
389: @example
390: Yoyodyne, Inc., hereby disclaims all copyright interest in the
391: program `Gnomovision' (a program to direct compilers to make passes
392: at assemblers) written by James Hacker.
393:
394: @var{signature of Ty Coon}, 1 April 1989
395: Ty Coon, President of Vice
396: @end example
397:
398: That's all there is to it!
1.1 root 399:
1.1.1.9 root 400: @node Contributors, Boycott, Copying, Top
1.1 root 401: @unnumbered Contributors to GNU CC
402:
403: In addition to Richard Stallman, several people have written parts
404: of GNU CC.
405:
406: @itemize @bullet
407: @item
408: The idea of using RTL and some of the optimization ideas came from the
409: U. of Arizona Portable Optimizer, written by Jack Davidson and
410: Christopher Fraser. See ``Register Allocation and Exhaustive Peephole
411: Optimization'', Software Practice and Experience 14 (9), Sept. 1984,
412: 857-866.
413:
414: @item
415: Paul Rubin wrote most of the preprocessor.
416:
417: @item
1.1.1.6 root 418: Leonard Tower wrote parts of the parser, RTL generator, and RTL
1.1 root 419: definitions, and of the Vax machine description.
420:
421: @item
422: Ted Lemon wrote parts of the RTL reader and printer.
423:
424: @item
425: Jim Wilson implemented loop strength reduction and some other
426: loop optimizations.
427:
428: @item
429: Nobuyuki Hikichi of Software Research Associates, Tokyo, contributed
1.1.1.8 root 430: the support for the Sony NEWS machine.
1.1 root 431:
432: @item
433: Charles LaBrec contributed the support for the Integrated Solutions
434: 68020 system.
435:
436: @item
437: Michael Tiemann of MCC wrote most of the description of the National
438: Semiconductor 32000 series cpu. He also wrote the code for inline
439: function integration and for the SPARC cpu and Motorola 88000 cpu
440: and part of the Sun FPA support.
441:
442: @item
443: Jan Stein of the Chalmers Computer Society provided support for
444: Genix, as well as part of the 32000 machine description.
445:
446: @item
447: Randy Smith finished the Sun FPA support.
448:
449: @item
450: Robert Brown implemented the support for Encore 32000 systems.
451:
452: @item
453: David Kashtan of SRI adapted GNU CC to the Vomit-Making System.
454:
455: @item
456: Alex Crain provided changes for the 3b1.
457:
458: @item
459: Greg Satz and Chris Hanson assisted in making GNU CC work on HP-UX for
460: the 9000 series 300.
461:
462: @item
463: William Schelter did most of the work on the Intel 80386 support.
1.1.1.5 root 464:
465: @item
466: Christopher Smith did the port for Convex machines.
467:
468: @item
469: Paul Petersen wrote the machine description for the Alliant FX/8.
1.1.1.7 root 470:
471: @item
1.1.1.8 root 472: Alain Lichnewsky ported GNU CC to the Mips cpu.
473:
474: @item
475: Devon Bowen, Dale Wiles and Kevin Zachmann ported GNU CC to the Tahoe.
476:
477: @item
478: Jonathan Stone wrote the machine description for the Pyramid computer.
1.1 root 479: @end itemize
480:
1.1.1.9 root 481: @node Boycott, Options, Contributors, Top
482: @chapter Protect Your Freedom---Fight ``Look And Feel''
483:
1.1.1.10 root 484: @quotation
485: @i{This section is a political message from the League for Programming
486: Freedom to the users of GNU CC. It is included here as an expression
487: of support for the League on the part of the Free Software Foundation
488: and Richard Stallman.}
489: @end quotation
490:
1.1.1.9 root 491: Ashton-Tate, Apple, Lotus and Xerox are trying to create a new form of
492: legal monopoly: a copyright on a class of user interfaces. These
493: monopolies would cause serious problems for users and developers of
494: computer software and systems.
495:
1.1.1.10 root 496: Until a few years ago, the law seemed clear: no one could restrict
1.1.1.9 root 497: others from using a user interface; programmers were free to implement
498: any interface they chose. Imitating interfaces, sometimes with changes,
499: was standard practice in the computer field. The interfaces we know
500: evolved gradually in this way; for example, the Macintosh user interface
501: drew ideas from the Xerox interface, which in turn drew on work done at
1.1.1.10 root 502: Stanford and SRI. 1-2-3 imitated VisiCalc, and dBase imitated a
503: database program from JPL.
1.1.1.9 root 504:
505: Most computer companies, and nearly all computer users, were happy with
506: this state of affairs. The companies that are suing say it does not
507: offer ``enough incentive'' to develop their products, but they must have
508: considered it ``enough'' when they made their decision to do so. It
509: seems they are not satisfied with the opportunity to continue to compete
510: in the marketplace---not even with a head start.
511:
512: If Xerox, Lotus, Apple and Ashton-Tate are permitted to make law through
513: the courts, the precedent will hobble the software industry:
514:
515: @itemize @bullet
516: @item
1.1.1.11! root 517: Gratuitous incompatibilities will burden users. Imagine if each
1.1.1.9 root 518: car manufacturer had to arrange the pedals in a different order.
519:
520: @item
521: Software will become and remain more expensive. Users will be
522: ``locked in'' to proprietary interfaces, for which there is no real
523: competition.
524:
525: @item
526: Large companies have an unfair advantage wherever lawsuits become
527: commonplace. Since they can easily afford to sue, they can intimidate
528: small companies with threats even when they don't really have a case.
529:
530: @item
531: User interface improvements will come slower, since incremental
532: evolution through creative imitation will no longer be permitted.
533:
534: @item
535: Even Apple, etc., will find it harder to make improvements if
536: they can no longer adapt the good ideas that others introduce, for
537: fear of weakening their own legal positions. Some users suggest that
538: this stagnation may already have started.
1.1.1.10 root 539:
540: @item
541: If you use GNU software, you might find it of some concern that user
542: interface copyright will make it hard for the Free Software Foundation
543: to develop programs compatible with the interfaces that you already
544: know.
1.1.1.9 root 545: @end itemize
546:
1.1.1.10 root 547: To protect our freedom from lawsuits like these, a group of programmers
548: and users have formed a new grass-roots political organization, the
549: League for Programming Freedom.
550:
551: The purpose of the League is to oppose new monopolistic practices such
552: as user-interface copyright and software patents; it calls for a return
553: to the legal policies of the recent past, in which these practices were
554: not allowed. The League is not concerned with free software as an
555: issue, and not affiliated with the Free Software Foundation.
556:
557: The League's membership rolls include John McCarthy, inventor of Lisp,
558: Marvin Minsky, founder of the Artificial Intelligence lab, Guy L.
559: Steele, Jr., author of well-known books on Lisp and C, as well as
560: Richard Stallman, the developer of GNU CC. Please join and add your
561: name to the list. Membership dues in the League are $42 per year for
562: programmers, managers and professionals; $10.50 for students; $21 for
563: others.
564:
565: The League needs both activist members and members who only pay their
566: dues.
567:
568: To join, or for more information, phone (617) 492-0023 or write to:
569:
570: @example
571: League for Programming Freedom
572: 1 Kendall Square #143
573: P.O. Box 9171
574: Cambridge, MA 02139 league@@prep.ai.mit.edu
575: @end example
576:
577: Here are some suggestions from the League for how you can protect your
578: freedom to write programs:
1.1.1.9 root 579:
580: @itemize @bullet
581: @item
582: Don't buy from Xerox, Lotus, Apple or Ashton-Tate. Buy from their
583: competitors or from the defendants they are suing.
584:
585: @item
586: Don't develop software to work with the systems made by these companies.
587:
588: @item
589: Port your existing software to competing systems, so that you encourage
590: users to switch.
591:
592: @item
593: Write letters to company presidents to let them know their conduct
594: is unacceptable.
595:
596: @item
597: Tell your friends and colleagues about this issue and how it threatens
598: to ruin the computer industry.
599:
600: @item
601: Above all, don't work for the look-and-feel plaintiffs, and don't
602: accept contracts from them.
603:
604: @item
1.1.1.11! root 605: Write to Congress to explain the importance of this issue.
1.1.1.9 root 606:
607: @example
1.1.1.11! root 608: House Subcommittee on Intellectual Property
! 609: 2137 Rayburn Bldg
! 610: Washington, DC 20515
1.1.1.9 root 611:
1.1.1.11! root 612: Senate Subcommittee on Patents, Trademarks and Copyrights
! 613: United States Senate
! 614: Washington, DC 20510
! 615: @end example
1.1.1.9 root 616: @end itemize
617:
618: Express your opinion! You can make a difference.
619:
620: @node Options, Installation, Boycott, Top
1.1 root 621: @chapter GNU CC Command Options
622:
623: The GNU C compiler uses a command syntax much like the Unix C compiler.
624: The @code{gcc} program accepts options and file names as operands.
625: Multiple single-letter options may @emph{not} be grouped: @samp{-dr} is
1.1.1.8 root 626: very different from @w{@samp{-d -r}}.
1.1 root 627:
628: When you invoke GNU CC, it normally does preprocessing, compilation,
629: assembly and linking. File names which end in @samp{.c} are taken as C
1.1.1.5 root 630: source to be preprocessed and compiled; file names ending in @samp{.i}
631: are taken as preprocessor output to be compiled; compiler output files
632: plus any input files with names ending in @samp{.s} are assembled; then
633: the resulting object files, plus any other input files, are linked
634: together to produce an executable.
1.1 root 635:
636: Command options allow you to stop this process at an intermediate stage.
637: For example, the @samp{-c} option says not to run the linker. Then the
638: output consists of object files output by the assembler.
639:
1.1.1.5 root 640: Other command options are passed on to one stage of processing. Some
641: options control the preprocessor and others the compiler itself. Yet
642: other options control the assembler and linker; these are not documented
643: here, but you rarely need to use any of them.
1.1 root 644:
645: Here are the options to control the overall compilation process, including
646: those that say whether to link, whether to assemble, and so on.
647:
648: @table @samp
649: @item -o @var{file}
650: Place output in file @var{file}. This applies regardless to whatever
651: sort of output is being produced, whether it be an executable file,
652: an object file, an assembler file or preprocessed C code.
653:
654: If @samp{-o} is not specified, the default is to put an executable file
655: in @file{a.out}, the object file @file{@var{source}.c} in
656: @file{@var{source}.o}, an assembler file in @file{@var{source}.s}, and
657: preprocessed C on standard output.@refill
658:
659: @item -c
660: Compile or assemble the source files, but do not link. Produce object
661: files with names made by replacing @samp{.c} or @samp{.s} with
662: @samp{.o} at the end of the input file names. Do nothing at all for
663: object files specified as input.
664:
665: @item -S
666: Compile into assembler code but do not assemble. The assembler output
667: file name is made by replacing @samp{.c} with @samp{.s} at the end of
668: the input file name. Do nothing at all for assembler source files or
669: object files specified as input.
670:
671: @item -E
672: Run only the C preprocessor. Preprocess all the C source files
673: specified and output the results to standard output.
674:
675: @item -v
676: Compiler driver program prints the commands it executes as it runs
677: the preprocessor, compiler proper, assembler and linker. Some of
678: these are directed to print their own version numbers.
679:
1.1.1.5 root 680: @item -pipe
681: Use pipes rather than temporary files for communication between the
682: various stages of compilation. This fails to work on some systems
683: where the assembler is unable to read from a pipe; but the GNU
684: assembler has no trouble.
685:
1.1 root 686: @item -B@var{prefix}
687: Compiler driver program tries @var{prefix} as a prefix for each
688: program it tries to run. These programs are @file{cpp}, @file{cc1},
689: @file{as} and @file{ld}.
690:
691: For each subprogram to be run, the compiler driver first tries the
692: @samp{-B} prefix, if any. If that name is not found, or if @samp{-B}
693: was not specified, the driver tries two standard prefixes, which are
694: @file{/usr/lib/gcc-} and @file{/usr/local/lib/gcc-}. If neither of
695: those results in a file name that is found, the unmodified program
696: name is searched for using the directories specified in your
697: @samp{PATH} environment variable.
698:
699: The run-time support file @file{gnulib} is also searched for using
700: the @samp{-B} prefix, if needed. If it is not found there, the two
701: standard prefixes above are tried, and that is all. The file is left
702: out of the link if it is not found by those means. Most of the time,
703: on most machines, you can do without it.
1.1.1.5 root 704:
705: You can get a similar result from the environment variable;
706: @code{GCC_EXEC_PREFIX} if it is defined, its value is used as a prefix
707: in the same way. If both the @samp{-B} option and the
708: @code{GCC_EXEC_PREFIX} variable are present, the @samp{-B} option is
709: used first and the environment variable value second.
1.1.1.8 root 710:
711: @item -b@var{prefix}
712: The argument @var{prefix} is used as a second prefix for the compiler
713: executables and libraries. This prefix is optional: the compiler tries
714: each file first with it, then without it. This prefix follows the
715: prefix specified with @samp{-B} or the default prefixes.
716:
717: Thus, @samp{-bvax- -Bcc/} in the presence of environment variable
718: @code{GCC_EXEC_PREFIX} with definition @file{/u/foo/} causes GNU CC to
719: try the following file names for the preprocessor executable:
720:
721: @example
722: cc/vax-cpp
723: cc/cpp
724: /u/foo/vax-cpp
725: /u/foo/cpp
726: /usr/local/lib/gcc-vax-cpp
727: /usr/local/lib/gcc-cpp
728: /usr/lib/gcc-vax-cpp
729: /usr/lib/gcc-cpp
730: @end example
1.1 root 731: @end table
732:
733: These options control the details of C compilation itself.
734:
735: @table @samp
736: @item -ansi
737: Support all ANSI standard C programs.
738:
739: This turns off certain features of GNU C that are incompatible with
740: ANSI C, such as the @code{asm}, @code{inline} and @code{typeof}
741: keywords, and predefined macros such as @code{unix} and @code{vax}
742: that identify the type of system you are using. It also enables the
743: undesirable and rarely used ANSI trigraph feature.
744:
1.1.1.8 root 745: The alternate keywords @code{__asm__}, @code{__inline__} and
746: @code{__typeof__} continue to work despite @samp{-ansi}. You would not
1.1.1.7 root 747: want to use them in an ANSI C program, of course, but it useful to put
748: them in header files that might be included in compilations done with
1.1.1.8 root 749: @samp{-ansi}. Alternate predefined macros such as @code{__unix__} and
750: @code{__vax__} are also available, with or without @samp{-ansi}.
1.1.1.7 root 751:
1.1 root 752: The @samp{-ansi} option does not cause non-ANSI programs to be
753: rejected gratuitously. For that, @samp{-pedantic} is required in
754: addition to @samp{-ansi}.
755:
756: The macro @code{__STRICT_ANSI__} is predefined when the @samp{-ansi}
757: option is used. Some header files may notice this macro and refrain
758: from declaring certain functions or defining certain macros that the
1.1.1.7 root 759: ANSI standard doesn't call for; this is to avoid interfering with any
760: programs that might use these names for other things.
1.1 root 761:
762: @item -traditional
763: Attempt to support some aspects of traditional C compilers.
764: Specifically:
765:
766: @itemize @bullet
767: @item
768: All @code{extern} declarations take effect globally even if they
769: are written inside of a function definition. This includes implicit
770: declarations of functions.
771:
772: @item
773: The keywords @code{typeof}, @code{inline}, @code{signed}, @code{const}
1.1.1.10 root 774: and @code{volatile} are not recognized. (You can still use the alternative
775: keywords such as @code{__typeof__}, @code{__inline__}, and so on.)
1.1 root 776:
777: @item
778: Comparisons between pointers and integers are always allowed.
779:
780: @item
781: Integer types @code{unsigned short} and @code{unsigned char} promote
782: to @code{unsigned int}.
783:
784: @item
785: Out-of-range floating point literals are not an error.
786:
787: @item
1.1.1.8 root 788: String ``constants'' are not necessarily constant; they are stored in
789: writable space, and identical looking constants are allocated
790: separately.
791:
792: @item
1.1.1.2 root 793: All automatic variables not declared @code{register} are preserved by
794: @code{longjmp}. Ordinarily, GNU C follows ANSI C: automatic variables
795: not declared @code{volatile} may be clobbered.
796:
797: @item
1.1 root 798: In the preprocessor, comments convert to nothing at all, rather than
799: to a space. This allows traditional token concatenation.
800:
801: @item
802: In the preprocessor, macro arguments are recognized within string
803: constants in a macro definition (and their values are stringified,
804: though without additional quote marks, when they appear in such a
805: context). The preprocessor always considers a string constant to end
806: at a newline.
807:
808: @item
809: The predefined macro @code{__STDC__} is not defined when you use
810: @samp{-traditional}, but @code{__GNUC__} is (since the GNU extensions
811: which @code{__GNUC__} indicates are not affected by
812: @samp{-traditional}). If you need to write header files that work
813: differently depending on whether @samp{-traditional} is in use, by
814: testing both of these predefined macros you can distinguish four
815: situations: GNU C, traditional GNU C, other ANSI C compilers, and
816: other old C compilers.
817: @end itemize
818:
819: @item -O
820: Optimize. Optimizing compilation takes somewhat more time, and a lot
821: more memory for a large function.
822:
823: Without @samp{-O}, the compiler's goal is to reduce the cost of
824: compilation and to make debugging produce the expected results.
825: Statements are independent: if you stop the program with a breakpoint
826: between statements, you can then assign a new value to any variable or
827: change the program counter to any other statement in the function and
828: get exactly the results you would expect from the source code.
829:
830: Without @samp{-O}, only variables declared @code{register} are
831: allocated in registers. The resulting compiled code is a little worse
832: than produced by PCC without @samp{-O}.
833:
834: With @samp{-O}, the compiler tries to reduce code size and execution
835: time.
836:
837: Some of the @samp{-f} options described below turn specific kinds of
838: optimization on or off.
839:
840: @item -g
841: Produce debugging information in the operating system's native format
842: (for DBX or SDB). GDB also can work with this debugging information.
843:
844: Unlike most other C compilers, GNU CC allows you to use @samp{-g} with
845: @samp{-O}. The shortcuts taken by optimized code may occasionally
846: produce surprising results: some variables you declared may not exist
847: at all; flow of control may briefly move where you did not expect it;
848: some statements may not be executed because they compute constant
849: results or their values were already at hand; some statements may
850: execute in different places because they were moved out of loops.
851: Nevertheless it proves possible to debug optimized output. This makes
852: it reasonable to use the optimizer for programs that might have bugs.
853:
854: @item -gg
1.1.1.8 root 855: Produce debugging information in the old GDB format. This is obsolete.
1.1 root 856:
857: @item -w
858: Inhibit all warning messages.
859:
860: @item -W
861: Print extra warning messages for these events:
862:
863: @itemize @bullet
864: @item
865: An automatic variable is used without first being initialized.
866:
867: These warnings are possible only in optimizing compilation,
868: because they require data flow information that is computed only
1.1.1.6 root 869: when optimizing. If you don't specify @samp{-O}, you simply won't
870: get these warnings.
871:
872: These warnings occur only for variables that are candidates for
873: register allocation. Therefore, they do not occur for a variable that
874: is declared @code{volatile}, or whose address is taken, or whose size
875: is other than 1, 2, 4 or 8 bytes. Also, they do not occur for
876: structures, unions or arrays, even when they are in registers.
877:
878: Note that there may be no warning about a variable that is used only
879: to compute a value that itself is never used, because such
880: computations may be deleted by data flow analysis before the warnings
881: are printed.
1.1 root 882:
883: These warnings are made optional because GNU CC is not smart
884: enough to see all the reasons why the code might be correct
885: despite appearing to have an error. Here is one example of how
886: this can happen:
887:
888: @example
889: @{
890: int x;
891: switch (y)
892: @{
893: case 1: x = 1;
894: break;
895: case 2: x = 4;
896: break;
897: case 3: x = 5;
898: @}
899: foo (x);
900: @}
901: @end example
902:
903: @noindent
904: If the value of @code{y} is always 1, 2 or 3, then @code{x} is
905: always initialized, but GNU CC doesn't know this. Here is
906: another common case:
907:
908: @example
909: @{
910: int save_y;
911: if (change_y) save_y = y, y = new_y;
912: @dots{}
913: if (change_y) y = save_y;
914: @}
915: @end example
916:
917: @noindent
918: This has no bug because @code{save_y} is used only if it is set.
919:
1.1.1.5 root 920: Some spurious warnings can be avoided if you declare as
921: @code{volatile} all the functions you use that never return.
922: @xref{Function Attributes}.
923:
1.1 root 924: @item
925: A nonvolatile automatic variable might be changed by a call to
926: @code{longjmp}. These warnings as well are possible only in
927: optimizing compilation.
928:
929: The compiler sees only the calls to @code{setjmp}. It cannot know
930: where @code{longjmp} will be called; in fact, a signal handler could
931: call it at any point in the code. As a result, you may get a warning
932: even when there is in fact no problem because @code{longjmp} cannot
933: in fact be called at the place which would cause a problem.
934:
935: @item
936: A function can return either with or without a value. (Falling
937: off the end of the function body is considered returning without
1.1.1.6 root 938: a value.) For example, this function would evoke such a
1.1 root 939: warning:
940:
941: @example
942: foo (a)
943: @{
944: if (a > 0)
945: return a;
946: @}
947: @end example
948:
949: Spurious warnings can occur because GNU CC does not realize that
950: certain functions (including @code{abort} and @code{longjmp})
951: will never return.
1.1.1.4 root 952:
953: @item
954: An expression-statement contains no side effects.
1.1 root 955: @end itemize
956:
957: In the future, other useful warnings may also be enabled by this
958: option.
959:
960: @item -Wimplicit
961: Warn whenever a function is implicitly declared.
962:
963: @item -Wreturn-type
964: Warn whenever a function is defined with a return-type that defaults
965: to @code{int}. Also warn about any @code{return} statement with no
966: return-value in a function whose return-type is not @code{void}.
967:
968: @item -Wunused
1.1.1.5 root 969: Warn whenever a local variable is unused aside from its declaration,
1.1.1.8 root 970: whenever a function is declared static but never defined, and whenever
971: a statement computes a result that is explicitly not used.
1.1.1.7 root 972:
973: @item -Wswitch
974: Warn whenever a @code{switch} statement has an index of enumeral type
975: and lacks a @code{case} for one or more of the named codes of that
976: enumeration. (The presence of a @code{default} label prevents this
977: warning.) @code{case} labels outside the enumeration range also
978: provoke warnings when this option is used.
979:
1.1 root 980: @item -Wcomment
981: Warn whenever a comment-start sequence @samp{/*} appears in a comment.
982:
1.1.1.7 root 983: @item -Wtrigraphs
984: Warn if any trigraphs are encountered (assuming they are enabled).
985:
1.1 root 986: @item -Wall
1.1.1.8 root 987: All of the above @samp{-W} options combined. These are all the
988: options which pertain to usage that we recommend avoiding and that we
989: believe is easy to avoid, even in conjunction with macros.
990:
991: The other @samp{-W@dots{}} options below are not implied by @samp{-Wall}
992: because certain kinds of useful macros are almost impossible to write
993: without causing those warnings.
994:
995: @item -Wshadow
996: Warn whenever a local variable shadows another local variable.
997:
998: @item -Wid-clash-@var{len}
999: Warn whenever two distinct identifiers match in the first @var{len}
1000: characters. This may help you prepare a program that will compile
1001: with certain obsolete, brain-damaged compilers.
1002:
1003: @item -Wpointer-arith
1004: Warn about anything that depends on the ``size of'' a function type or
1005: of @code{void}. GNU C assigns these types a size of 1, for
1006: convenience in calculations with @code{void *} pointers and pointers
1007: to functions.
1.1 root 1008:
1.1.1.6 root 1009: @item -Wcast-qual
1010: Warn whenever a pointer is cast so as to remove a type qualifier from
1011: the target type. For example, warn if a @code{const char *} is cast
1012: to an ordinary @code{char *}.
1013:
1.1 root 1014: @item -Wwrite-strings
1015: Give string constants the type @code{const char[@var{length}]} so that
1016: copying the address of one into a non-@code{const} @code{char *}
1017: pointer will get a warning. These warnings will help you find at
1018: compile time code that can try to write into a string constant, but
1019: only if you have been very careful about using @code{const} in
1020: declarations and prototypes. Otherwise, it will just be a nuisance;
1021: this is why we did not make @samp{-Wall} request these warnings.
1022:
1023: @item -p
1024: Generate extra code to write profile information suitable for the
1025: analysis program @code{prof}.
1026:
1027: @item -pg
1028: Generate extra code to write profile information suitable for the
1029: analysis program @code{gprof}.
1030:
1.1.1.6 root 1031: @item -a
1.1.1.8 root 1032: Generate extra code to write profile information for basic blocks, which
1033: will record the number of times each basic block is executed. This data
1034: could be analyzed by a program like @code{tcov}. Note, however, that
1035: the format of the data is not what @code{tcov} expects. Eventually GNU
1.1.1.6 root 1036: @code{gprof} should be extended to process this data.
1037:
1.1 root 1038: @item -l@var{library}
1039: Search a standard list of directories for a library named
1040: @var{library}, which is actually a file named
1041: @file{lib@var{library}.a}. The linker uses this file as if it
1042: had been specified precisely by name.
1043:
1044: The directories searched include several standard system directories
1045: plus any that you specify with @samp{-L}.
1046:
1047: Normally the files found this way are library files---archive files
1048: whose members are object files. The linker handles an archive file by
1049: scanning through it for members which define symbols that have so far
1050: been referenced but not defined. But if the file that is found is an
1051: ordinary object file, it is linked in the usual fashion. The only
1052: difference between using an @samp{-l} option and specifying a file name
1053: is that @samp{-l} searches several directories.
1054:
1055: @item -L@var{dir}
1056: Add directory @var{dir} to the list of directories to be searched
1057: for @samp{-l}.
1058:
1059: @item -nostdlib
1.1.1.9 root 1060: Don't use the standard system libraries and startup files when linking.
1061: Only the files you specify will be passed to the linker.
1.1 root 1062:
1063: @item -m@var{machinespec}
1064: Machine-dependent option specifying something about the type of target
1065: machine. These options are defined by the macro
1066: @code{TARGET_SWITCHES} in the machine description. The default for
1067: the options is also defined by that macro, which enables you to change
1068: the defaults.@refill
1069:
1070: These are the @samp{-m} options defined in the 68000 machine
1071: description:
1072:
1073: @table @samp
1074: @item -m68020
1075: @itemx -mc68020
1076: Generate output for a 68020 (rather than a 68000). This is the
1077: default if you use the unmodified sources.
1078:
1079: @item -m68000
1080: @item -mc68000
1081: Generate output for a 68000 (rather than a 68020).
1082:
1083: @item -m68881
1084: Generate output containing 68881 instructions for floating point.
1085: This is the default if you use the unmodified sources.
1086:
1087: @item -mfpa
1088: Generate output containing Sun FPA instructions for floating point.
1089:
1090: @item -msoft-float
1091: Generate output containing library calls for floating point.
1092:
1093: @item -mshort
1094: Consider type @code{int} to be 16 bits wide, like @code{short int}.
1095:
1096: @item -mnobitfield
1097: Do not use the bit-field instructions. @samp{-m68000} implies
1098: @samp{-mnobitfield}.
1099:
1100: @item -mbitfield
1101: Do use the bit-field instructions. @samp{-m68020} implies
1102: @samp{-mbitfield}. This is the default if you use the unmodified
1103: sources.
1104:
1105: @item -mrtd
1106: Use a different function-calling convention, in which functions
1107: that take a fixed number of arguments return with the @code{rtd}
1108: instruction, which pops their arguments while returning. This
1109: saves one instruction in the caller since there is no need to pop
1110: the arguments there.
1111:
1112: This calling convention is incompatible with the one normally
1113: used on Unix, so you cannot use it if you need to call libraries
1114: compiled with the Unix compiler.
1115:
1116: Also, you must provide function prototypes for all functions that
1117: take variable numbers of arguments (including @code{printf});
1118: otherwise incorrect code will be generated for calls to those
1119: functions.
1120:
1121: In addition, seriously incorrect code will result if you call a
1122: function with too many arguments. (Normally, extra arguments are
1123: harmlessly ignored.)
1124:
1125: The @code{rtd} instruction is supported by the 68010 and 68020
1126: processors, but not by the 68000.
1127: @end table
1128:
1129: These @samp{-m} options are defined in the Vax machine description:
1130:
1131: @table @samp
1132: @item -munix
1133: Do not output certain jump instructions (@code{aobleq} and so on)
1134: that the Unix assembler for the Vax cannot handle across long
1135: ranges.
1136:
1137: @item -mgnu
1138: Do output those jump instructions, on the assumption that you
1139: will assemble with the GNU assembler.
1140:
1141: @item -mg
1142: Output code for g-format floating point numbers instead of d-format.
1143: @end table
1144:
1.1.1.5 root 1145: These @samp{-m} switches are supported on the Sparc:
1146:
1147: @table @samp
1148: @item -mfpu
1149: Generate output containing floating point instructions. This is the
1150: default if you use the unmodified sources.
1151:
1.1.1.9 root 1152: @ignore
1.1.1.5 root 1153: @item -msoft-float
1154: Generate output containing library calls for floating point.
1155:
1.1.1.9 root 1156: @end ignore
1.1.1.5 root 1157: @item -mno-epilogue
1.1.1.6 root 1158: Generate separate return instructions for @code{return} statements.
1159: This has both advantages and disadvantages; I don't recall what they
1160: are.
1.1.1.5 root 1161: @end table
1162:
1163: These @samp{-m} options are defined in the Convex machine description:
1164:
1165: @table @samp
1166: @item -mc1
1167: Generate output for a C1. This is the default when the compiler is
1168: configured for a C1.
1169:
1170: @item -mc2
1171: Generate output for a C2. This is the default when the compiler is
1172: configured for a C2.
1173:
1174: @item -margcount
1175: Generate code which puts an argument count in the word preceding each
1176: argument list. Some nonportable Convex and Vax programs need this
1177: word. (Debuggers don't; this info is in the symbol table.)
1178:
1179: @item -mnoargcount
1180: Omit the argument count word. This is the default if you use the
1181: unmodified sources.
1182: @end table
1183:
1.1 root 1184: @item -f@var{flag}
1.1.1.4 root 1185: Specify machine-independent flags. Most flags have both positive and
1186: negative forms; the negative form of @samp{-ffoo} would be
1187: @samp{-fno-foo}. In the table below, only one of the forms is
1188: listed---the one which is not the default. You can figure out the
1189: other form by either removing @samp{no-} or adding it.
1.1 root 1190:
1191: @table @samp
1.1.1.6 root 1192: @item -fpcc-struct-return
1193: Use the same convention for returning @code{struct} and @code{union}
1194: values that is used by the usual C compiler on your system. This
1195: convention is less efficient for small structures, and on many
1196: machines it fails to be reentrant; but it has the advantage of
1197: allowing intercallability between GCC-compiled code and PCC-compiled
1198: code.
1199:
1.1 root 1200: @item -ffloat-store
1201: Do not store floating-point variables in registers. This
1202: prevents undesirable excess precision on machines such as the
1203: 68000 where the floating registers (of the 68881) keep more
1204: precision than a @code{double} is supposed to have.
1205:
1206: For most programs, the excess precision does only good, but a few
1207: programs rely on the precise definition of IEEE floating point.
1208: Use @samp{-ffloat-store} for such programs.
1209:
1210: @item -fno-asm
1211: Do not recognize @code{asm}, @code{inline} or @code{typeof} as a
1.1.1.7 root 1212: keyword. These words may then be used as identifiers. You can
1.1.1.8 root 1213: use @code{__asm__}, @code{__inline__} and @code{__typeof__} instead.
1.1 root 1214:
1215: @item -fno-defer-pop
1216: Always pop the arguments to each function call as soon as that
1217: function returns. Normally the compiler (when optimizing) lets
1218: arguments accumulate on the stack for several function calls and
1219: pops them all at once.
1220:
1221: @item -fstrength-reduce
1222: Perform the optimizations of loop strength reduction and
1223: elimination of iteration variables.
1224:
1225: @item -fcombine-regs
1226: Allow the combine pass to combine an instruction that copies one
1227: register into another. This might or might not produce better
1228: code when used in addition to @samp{-O}. I am interested in
1229: hearing about the difference this makes.
1230:
1231: @item -fforce-mem
1232: Force memory operands to be copied into registers before doing
1233: arithmetic on them. This may produce better code by making all
1234: memory references potential common subexpressions. When they are
1235: not common subexpressions, instruction combination should
1236: eliminate the separate register-load. I am interested in hearing
1237: about the difference this makes.
1238:
1239: @item -fforce-addr
1240: Force memory address constants to be copied into registers before
1241: doing arithmetic on them. This may produce better code just as
1242: @samp{-fforce-mem} may. I am interested in hearing about the
1243: difference this makes.
1244:
1245: @item -fomit-frame-pointer
1246: Don't keep the frame pointer in a register for functions that
1247: don't need one. This avoids the instructions to save, set up and
1248: restore frame pointers; it also makes an extra register available
1249: in many functions. @strong{It also makes debugging impossible.}
1250:
1251: On some machines, such as the Vax, this flag has no effect,
1252: because the standard calling sequence automatically handles the
1253: frame pointer and nothing is saved by pretending it doesn't
1254: exist. The machine-description macro
1255: @code{FRAME_POINTER_REQUIRED} controls whether a target machine
1256: supports this flag. @xref{Registers}.@refill
1257:
1258: @item -finline-functions
1259: Integrate all simple functions into their callers. The compiler
1260: heuristically decides which functions are simple enough to be
1261: worth integrating in this way.
1262:
1263: If all calls to a given function are integrated, and the function
1264: is declared @code{static}, then the function is normally not
1265: output as assembler code in its own right.
1266:
1.1.1.6 root 1267: @item -fcaller-saves
1268: Enable values to be allocated in registers that will be clobbered by
1269: function calls, by emitting extra instructions to save and restore the
1270: registers around such calls. Such allocation is done only when it
1271: seems to result in better code than would otherwise be produced.
1272:
1273: This option is enabled by default on certain machines, usually those
1274: which have no call-preserved registers to use instead.
1275:
1.1 root 1276: @item -fkeep-inline-functions
1277: Even if all calls to a given function are integrated, and the
1278: function is declared @code{static}, nevertheless output a
1279: separate run-time callable version of the function.
1280:
1281: @item -fwritable-strings
1.1.1.8 root 1282: Store string constants in the writable data segment and don't uniquize
1283: them. This is for compatibility with old programs which assume they can
1284: write into string constants. @samp{-traditional} also has this effect.
1285:
1286: Writing into string constants is a very bad idea; ``constants'' should
1287: be constant.
1.1 root 1288:
1.1.1.4 root 1289: @item -fcond-mismatch
1290: Allow conditional expressions with mismatched types in the second and
1291: third arguments. The value of such an expression is void.
1292:
1.1 root 1293: @item -fno-function-cse
1294: Do not put function addresses in registers; make each instruction
1295: that calls a constant function contain the function's address
1296: explicitly.
1297:
1298: This option results in less efficient code, but some strange
1299: hacks that alter the assembler output may be confused by the
1300: optimizations performed when this option is not used.
1301:
1302: @item -fvolatile
1303: Consider all memory references through pointers to be volatile.
1304:
1.1.1.4 root 1305: @item -fshared-data
1306: Requests that the data and non-@code{const} variables of this
1307: compilation be shared data rather than private data. The distinction
1308: makes sense only on certain operating systems, where shared data is
1309: shared between processes running the same program, while private data
1310: exists in one copy per process.
1311:
1.1 root 1312: @item -funsigned-char
1.1.1.4 root 1313: Let the type @code{char} be the unsigned, like @code{unsigned char}.
1.1 root 1314:
1315: Each kind of machine has a default for what @code{char} should
1316: be. It is either like @code{unsigned char} by default or like
1317: @code{signed char} by default. (Actually, at present, the
1318: default is always signed.)
1319:
1320: The type @code{char} is always a distinct type from either
1321: @code{signed char} or @code{unsigned char}, even though its
1322: behavior is always just like one of those two.
1323:
1.1.1.4 root 1324: Note that this is equivalent to @samp{-fno-signed-char}, which is the
1325: negative form of @samp{-fsigned-char}.
1326:
1.1 root 1327: @item -fsigned-char
1328: Let the type @code{char} be signed, like @code{signed char}.
1329:
1.1.1.4 root 1330: Note that this is equivalent to @samp{-fno-unsigned-char}, which is
1331: the negative form of @samp{-funsigned-char}.
1332:
1.1.1.8 root 1333: @item -fdelayed-branch
1334: If supported for the target machine, attempt to reorder instructions
1335: to exploit instruction slots available after delayed branch
1336: instructions.
1337:
1.1 root 1338: @item -ffixed-@var{reg}
1339: Treat the register named @var{reg} as a fixed register; generated
1340: code should never refer to it (except perhaps as a stack pointer,
1341: frame pointer or in some other fixed role).
1342:
1343: @var{reg} must be the name of a register. The register names
1344: accepted are machine-specific and are defined in the
1345: @code{REGISTER_NAMES} macro in the machine description macro
1346: file.
1347:
1.1.1.4 root 1348: This flag does not have a negative form, because it specifies a
1349: three-way choice.
1350:
1.1 root 1351: @item -fcall-used-@var{reg}
1352: Treat the register named @var{reg} as an allocatable register
1353: that is clobbered by function calls. It may be allocated for
1354: temporaries or variables that do not live across a call.
1355: Functions compiled this way will not save and restore the
1356: register @var{reg}.
1357:
1358: Use of this flag for a register that has a fixed pervasive role
1359: in the machine's execution model, such as the stack pointer or
1360: frame pointer, will produce disastrous results.
1361:
1.1.1.4 root 1362: This flag does not have a negative form, because it specifies a
1363: three-way choice.
1364:
1.1 root 1365: @item -fcall-saved-@var{reg}
1366: Treat the register named @var{reg} as an allocatable register
1367: saved by functions. It may be allocated even for temporaries or
1368: variables that live across a call. Functions compiled this way
1369: will save and restore the register @var{reg} if they use it.
1370:
1371: Use of this flag for a register that has a fixed pervasive role
1372: in the machine's execution model, such as the stack pointer or
1373: frame pointer, will produce disastrous results.
1374:
1375: A different sort of disaster will result from the use of this
1376: flag for a register in which function values may be returned.
1.1.1.4 root 1377:
1378: This flag does not have a negative form, because it specifies a
1379: three-way choice.
1.1 root 1380: @end table
1381:
1382: @item -d@var{letters}
1383: Says to make debugging dumps at times specified by @var{letters}.
1384: Here are the possible letters:
1385:
1386: @table @samp
1387: @item r
1388: Dump after RTL generation.
1389: @item j
1390: Dump after first jump optimization.
1391: @item s
1392: Dump after CSE (including the jump optimization that sometimes
1393: follows CSE).
1394: @item L
1395: Dump after loop optimization.
1396: @item f
1397: Dump after flow analysis.
1398: @item c
1399: Dump after instruction combination.
1400: @item l
1401: Dump after local register allocation.
1402: @item g
1403: Dump after global register allocation.
1.1.1.8 root 1404: @item d
1405: Dump after delayed branch scheduling.
1406: @item J
1407: Dump after last jump optimization.
1.1 root 1408: @item m
1409: Print statistics on memory usage, at the end of the run.
1410: @end table
1411:
1412: @item -pedantic
1413: Issue all the warnings demanded by strict ANSI standard C; reject
1414: all programs that use forbidden extensions.
1415:
1416: Valid ANSI standard C programs should compile properly with or without
1417: this option (though a rare few will require @samp{-ansi}). However,
1418: without this option, certain GNU extensions and traditional C features
1419: are supported as well. With this option, they are rejected. There is
1420: no reason to @i{use} this option; it exists only to satisfy pedants.
1.1.1.5 root 1421:
1.1.1.8 root 1422: @samp{-pedantic} does not cause warning messages for use of the
1423: alternate keywords whose names begin and end with @samp{__}.
1424: @xref{Alternate Keywords}.
1425:
1.1.1.5 root 1426: @item -static
1427: On Suns running version 4, this prevents linking with the shared
1428: libraries. (@samp{-g} has the same effect.)
1.1 root 1429: @end table
1430:
1431: These options control the C preprocessor, which is run on each C source
1432: file before actual compilation. If you use the @samp{-E} option, nothing
1433: is done except C preprocessing. Some of these options make sense only
1434: together with @samp{-E} because they request preprocessor output that is
1435: not suitable for actual compilation.
1436:
1437: @table @samp
1438: @item -C
1439: Tell the preprocessor not to discard comments. Used with the
1440: @samp{-E} option.
1441:
1442: @item -I@var{dir}
1443: Search directory @var{dir} for include files.
1444:
1445: @item -I-
1446: Any directories specified with @samp{-I} options before the @samp{-I-}
1447: option are searched only for the case of @samp{#include "@var{file}"};
1448: they are not searched for @samp{#include <@var{file}>}.
1449:
1450: If additional directories are specified with @samp{-I} options after
1451: the @samp{-I-}, these directories are searched for all @samp{#include}
1452: directives. (Ordinarily @emph{all} @samp{-I} directories are used
1453: this way.)
1454:
1455: In addition, the @samp{-I-} option inhibits the use of the current
1.1.1.8 root 1456: directory (where the current input file came from) as the first search
1457: directory for @samp{#include "@var{file}"}. There is no way to override
1458: this effect of @samp{-I-}. With @samp{-I.} you can specify searching
1459: the directory which was current when the compiler was invoked. That is
1460: not exactly the same as what the preprocessor does by default, but it is
1461: often satisfactory.
1462:
1463: @samp{-I-} does not inhibit the use of the standard system directories
1464: for header files. Thus, @samp{-I-} and @samp{-nostdinc} are
1465: independent.
1466:
1467: @item -i @var{file}
1468: Process @var{file} as input, discarding the resulting output, before
1469: processing the regular input file. Because the output generated from
1470: @var{file} is discarded, the only effect of @samp{-i @var{file}} is to
1471: make the macros defined in @var{file} available for use in the main
1472: input.
1.1 root 1473:
1474: @item -nostdinc
1475: Do not search the standard system directories for header files. Only
1476: the directories you have specified with @samp{-I} options (and the
1477: current directory, if appropriate) are searched.
1478:
1479: Between @samp{-nostdinc} and @samp{-I-}, you can eliminate all
1480: directories from the search path except those you specify.
1481:
1482: @item -M
1483: Tell the preprocessor to output a rule suitable for @code{make}
1.1.1.10 root 1484: describing the dependencies of each object file. For each source
1.1 root 1485: file, the preprocessor outputs one @code{make}-rule whose target is
1486: the object file name for that source file and whose dependencies are
1487: all the files @samp{#include}d in it. This rule may be a single line
1488: or may be continued with @samp{\}-newline if it is long.
1489:
1490: @samp{-M} implies @samp{-E}.
1491:
1492: @item -MM
1493: Like @samp{-M} but the output mentions only the user-header files
1494: included with @samp{#include "@var{file}"}. System header files
1495: included with @samp{#include <@var{file}>} are omitted.
1496:
1497: @samp{-MM} implies @samp{-E}.
1498:
1499: @item -D@var{macro}
1.1.1.8 root 1500: Define macro @var{macro} with the string @samp{1} as its definition.
1.1 root 1501:
1502: @item -D@var{macro}=@var{defn}
1503: Define macro @var{macro} as @var{defn}.
1504:
1505: @item -U@var{macro}
1506: Undefine macro @var{macro}.
1507:
1.1.1.7 root 1508: @item -trigraphs
1.1 root 1509: Support ANSI C trigraphs. You don't want to know about this
1510: brain-damage. The @samp{-ansi} option also has this effect.
1511: @end table
1512:
1513: @node Installation, Trouble, Options, Top
1514: @chapter Installing GNU CC
1515:
1516: Here is the procedure for installing GNU CC on a Unix system.
1.1.1.8 root 1517:
1.1 root 1518: @menu
1.1.1.8 root 1519: * Other Dir:: Compiling in a separate directory (not where the source is).
1520: * Sun Install:: See below for installation on the Sun.
1521: * 3B1 Install:: See below for installation on the 3B1.
1.1.1.11! root 1522: * SCO Install:: See below for installation on SCO System V 3.2. (Or ESIX.)
1.1 root 1523: * VMS Install:: See below for installation on VMS.
1.1.1.9 root 1524: * HPUX Install:: See below for installation on HPUX.
1.1.1.10 root 1525: * MIPS Install:: See below for installation on MIPS.
1.1.1.11! root 1526: * Tower Install:: See below for installation on an NCR Tower.
1.1 root 1527: @end menu
1528: @iftex
1.1.1.9 root 1529: See below for VMS systems, and modified procedures needed on Sun
1530: systems, 3b1 machines and HPUX. The following section says how to
1531: compile in a separate directory on Unix; here we assume you compile in
1532: the same directory that contains the source files.
1.1 root 1533: @end iftex
1534:
1535: @enumerate
1536: @item
1537: Edit @file{Makefile}. If you are using HPUX, or any form of system V,
1538: you must make a few changes described in comments at the beginning of
1.1.1.9 root 1539: the file. Genix requires changes also, and so does the Pyramid.
1.1 root 1540:
1541: @item
1542: On a Sequent system, go to the Berkeley universe.
1543:
1544: @item
1.1.1.2 root 1545: Choose configuration files. The easy way to do this is to run the
1.1.1.8 root 1546: command file @file{config.gcc} with a single argument, which specifies
1547: the type of machine (and in some cases which operating system).
1.1.1.4 root 1548:
1549: Here is a list of the possible arguments:
1550:
1551: @table @samp
1552: @item vax
1553: Vaxes running BSD.
1554: @item vms
1555: Vaxes running VMS.
1556: @item vax-sysv
1557: Vaxes running system V.
1558: @item i386-sysv
1559: Intel 386 PCs running system V.
1.1.1.5 root 1560: @item i386-sysv-gas
1561: Intel 386 PCs running system V, using the GNU assembler and GNU
1562: linker.
1.1.1.6 root 1563: @item sequent-i386
1.1.1.4 root 1564: Sequent with Intel 386 processors.
1.1.1.8 root 1565: @item i386-aix
1566: Intel 386 PCs or PS/2s running AIX.
1.1.1.4 root 1567: @item sun2
1568: Sun 2 running system version 2 or 3.
1569: @item sun3
1.1.1.5 root 1570: Sun 3 running system version 2 or 3, with 68881.
1.1.1.7 root 1571: Note there we do not provide a configuration file to use an FPA
1.1.1.8 root 1572: by default, because programs that establish signal handlers for
1.1.1.7 root 1573: floating point traps inherently cannot work with the FPA.
1.1.1.5 root 1574: @item sun3-nfp
1575: Sun 3 running system version 2 or 3, without 68881.
1.1.1.4 root 1576: @item sun4
1.1.1.8 root 1577: Sun 4 running system version 2 or 3. @xref{Incompatibilities},
1578: for calling convention incompatibilities on the Sun 4 (sparc).
1.1.1.4 root 1579: @item sun2-os4
1580: Sun 2 running system version 4.
1581: @item sun3-os4
1.1.1.5 root 1582: Sun 3 running system version 4, with 68881.
1583: @item sun3-nfp-os4
1584: Sun 3 running system version 4, without 68881.
1.1.1.4 root 1585: @item sun4-os4
1.1.1.8 root 1586: Sun 4 running system version 4. @xref{Incompatibilities},
1587: for calling convention incompatibilities on the Sun 4 (sparc).
1.1.1.4 root 1588: @item sun386
1589: Sun 386 (``roadrunner'').
1.1.1.5 root 1590: @item alliant
1.1.1.8 root 1591: Alliant FX/8 computer. Note that the standard installed C compiler in
1592: Concentrix 5.0 has a bug which prevent it from compiling GNU CC
1593: correctly. You can patch the compiler bug as follows:
1594:
1595: @example
1596: cp /bin/pcc ./pcc
1.1.1.9 root 1597: adb -w ./pcc - << EOF
1.1.1.8 root 1598: 15f6?w 6610
1599: EOF
1600: @end example
1601:
1602: Then you must use the @samp{-ip12} option when compiling GNU CC
1603: with the patched compiler, as shown here:
1604:
1605: @example
1606: make CC="./pcc -ip12" CFLAGS=-w
1607: @end example
1608:
1609: Note also that Alliant's version of DBX does not manage to work with the
1610: output from GNU CC.
1611: @item tahoe
1612: The tahoe computer (running BSD, and using DBX).
1613: @item decstation
1614: The DEC 3100 Mips machine (``pmax''). Note that GNU CC cannot generate
1615: debugging information in the unusual format used on the Mips.
1616: @item mips-sysv
1617: The Mips computer, RS series, with the System V environment as default.
1618: Note that GNU CC cannot generate debugging information in the unusual
1619: format used on the Mips.
1620: @item mips-bsd43
1621: The Mips computer, RS series, with the BSD 4.3 environment as default.
1622: Note that GNU CC cannot generate debugging information in the unusual
1623: format used on the Mips.
1.1.1.7 root 1624: @item mips
1.1.1.8 root 1625: The Mips computer, M series. Note that GNU CC cannot generate debugging
1626: information in the unusual format used on the Mips.
1627: @item iris
1.1.1.10 root 1628: Another variant of the Mips computer, the Silicon Graphics Iris 4D.
1629: Note that GNU CC cannot generate debugging information in the unusual
1630: format used on the Mips.
1.1.1.5 root 1631: @item convex-c1
1632: Convex C1 computer.
1633: @item convex-c2
1634: Convex C2 computer.
1.1.1.8 root 1635: @item pyramid
1636: Pyramid computer.
1.1.1.4 root 1637: @item hp9k320
1.1.1.7 root 1638: HP 9000 series 300 using HPUX assembler. Note there is no
1639: support in GNU CC for HP's debugger; thus, @samp{-g} is not
1640: available in this configuration.
1.1.1.8 root 1641: @item hp9k320-gas
1.1.1.4 root 1642: HP 9000 series 300 using GNU assembler, linker and debugger.
1.1.1.7 root 1643: This requires the HP-adapt package, which is available along with
1644: the GNU linker as part of the ``binutils'' distribution.
1645: This is on the GNU CC distribution tape.
1.1.1.8 root 1646: @item hp9k320-old
1647: HP 9000 series 300 using HPUX assembler, in operating system versions
1648: older than 6.5. Note there is no support in GNU CC for HP's debugger;
1649: thus, @samp{-g} is not available in this configuration.
1650: @item hp9k320-bsd
1651: HP 9000 series 300 running BSD.
1.1.1.4 root 1652: @item isi68
1.1.1.8 root 1653: ISI 68000 or 68020 system with a 68881.
1654: @item isi68-nfp
1655: ISI 68000 or 68020 system without a 68881.
1.1.1.4 root 1656: @item news800
1657: Sony NEWS 68020 system.
1.1.1.6 root 1658: @item next
1659: NeXT system.
1.1.1.10 root 1660: @item tower
1661: NCR Tower 32 system.
1.1.1.7 root 1662: @item altos
1663: Altos 3068. Note that you must use the GNU assembler, linker and
1664: debugger, with COFF-encapsulation. Also, you must fix a kernel
1665: bug. Details in the file @file{ALTOS-README}.
1.1.1.4 root 1666: @item 3b1
1.1.1.8 root 1667: AT&T 3b1, a.k.a. 7300 PC. Note that special procedures are needed
1668: to compile GNU CC with this machine's standard C compiler, due to
1669: bugs in that compiler. @xref{3b1 Install}. You can bootstrap it
1670: more easily with previous versions of GNU CC if you have them.
1.1.1.9 root 1671: @item 3b1-gas
1672: AT&T 3b1 using the GNU assembler.
1.1.1.4 root 1673: @item sequent-ns32k
1674: Sequent containing ns32000 processors.
1675: @item encore
1676: Encore ns32000 system.
1677: @item genix
1678: National Semiconductor ns32000 system.
1679: @item 88000
1680: Motorola 88000 processor. This port is not finished.
1681: @end table
1.1.1.2 root 1682:
1.1.1.4 root 1683: Here we spell out what files need to be set up:
1.1 root 1684:
1685: @itemize @bullet
1686: @item
1687: Make a symbolic link named @file{config.h} to the top-level
1688: config file for the machine you are using (@pxref{Config}). This
1689: file is responsible for defining information about the host
1690: machine. It includes @file{tm.h}.
1691:
1.1.1.7 root 1692: The file is located in the subdirectory @file{config}. Its name
1693: should be @file{xm-@var{machine}.h}, with these exceptions:
1.1 root 1694:
1695: @table @file
1.1.1.3 root 1696: @item xm-vms.h
1.1 root 1697: for vaxen running VMS.
1.1.1.3 root 1698: @item xm-vaxv.h
1.1 root 1699: for vaxen running system V.
1.1.1.3 root 1700: @item xm-i386v.h
1.1 root 1701: for Intel 80386's running system V.
1.1.1.3 root 1702: @item xm-sun386i.h
1703: for Sun roadrunner running any version of the operating system.
1704: @item xm-hp9k320.h
1.1 root 1705: for the HP 9000 series 300.
1.1.1.4 root 1706: @item xm-genix.h
1.1 root 1707: for the ns32000 running Genix
1708: @end table
1709:
1710: If your system does not support symbolic links, you might want to
1711: set up @file{config.h} to contain a @samp{#include} command which
1712: refers to the appropriate file.
1713:
1714: @item
1715: Make a symbolic link named @file{tm.h} to the machine-description
1.1.1.7 root 1716: macro file for your machine. It should be in the subdirectory
1717: @file{config} and its name should be @file{tm-@var{machine}.h}.
1.1 root 1718:
1719: If your system is a 68000, don't use the file @file{tm-m68k.h}
1720: directly. Instead, use one of these files:
1721:
1722: @table @file
1723: @item tm-sun3.h
1.1.1.5 root 1724: for Sun 3 machines with 68881.
1725: @item tm-sun3-nfp.h
1726: for Sun 3 machines with no hardware floating point.
1.1.1.8 root 1727: @item tm-sun3os3.h
1728: for Sun 3 machines with 68881, running Sunos version 3.
1729: @item tm-sun3os3nf.h
1730: for Sun 3 machines with no hardware floating point, running Sunos
1731: version 3.
1.1 root 1732: @item tm-sun2.h
1733: for Sun 2 machines.
1734: @item tm-3b1.h
1735: for AT&T 3b1 (aka 7300 Unix PC).
1736: @item tm-isi68.h
1.1.1.3 root 1737: for Integrated Solutions systems. This file assumes you
1738: use the GNU assembler.
1.1.1.8 root 1739: @item tm-isi68-nfp.h
1740: for Integrated Solutions systems without a 68881. This file assumes you
1741: use the GNU assembler.
1.1 root 1742: @item tm-news800.h
1.1.1.8 root 1743: for Sony NEWS systems.
1.1 root 1744: @item tm-hp9k320.h
1745: for HPUX systems, if you are using GNU CC with the system's
1746: assembler and linker.
1747: @item tm-hp9k320g.h
1748: for HPUX systems, if you are using the GNU assembler, linker and
1749: other utilities. Not all of the pieces of GNU software needed
1750: for this mode of operation are as yet in distribution; full
1751: instructions will appear here in the future.@refill
1.1.1.10 root 1752: @item tm-tower-as.h
1753: for NCR Tower 32 systems, using the standard system assembler.
1.1 root 1754: @end table
1755:
1756: For the vax, use @file{tm-vax.h} on BSD Unix, @file{tm-vaxv.h} on
1757: system V, or @file{tm-vms.h} on VMS.@refill
1758:
1759: For the Motorola 88000, use @file{tm-m88k.h}. The support for the
1.1.1.9 root 1760: 88000 does not currently work; it requires extensive changes which
1761: we hope to reconcile in version 2.
1.1 root 1762:
1763: For the 80386, don't use @file{tm-i386.h} directly. Use
1764: @file{tm-i386v.h} if the target machine is running system V,
1.1.1.5 root 1765: @file{tm-i386gas.h} if it is running system V but you are using the
1766: GNU assembler and linker, @file{tm-seq386.h} for a Sequent 386 system,
1767: or @file{tm-compaq.h} for a Compaq, or @file{tm-sun386i.h} for a Sun
1768: 386 system.
1.1 root 1769:
1.1.1.8 root 1770: For the Mips computer, there are five choices: @file{tm-mips.h} for the
1771: M series, @file{tm-mips-bsd.h} for the RS series with BSD,
1772: @file{tm-mips-sysv.h} for the RS series with System V, @file{tm-iris.h}
1773: for the Iris version of the machine, and @file{tm-decstatn.h} for the
1774: Decstation.
1775:
1.1 root 1776: For the 32000, use @file{tm-sequent.h} if you are using a Sequent
1777: machine, or @file{tm-encore.h} for an Encore machine, or
1.1.1.4 root 1778: @file{tm-genix.h} if you are using Genix version 3; otherwise, perhaps
1.1 root 1779: @file{tm-ns32k.h} will work for you.
1780:
1781: Note that Genix has bugs in @code{alloca} and @code{malloc}; you must
1782: get the compiled versions of these from GNU Emacs and edit GNU CC's
1783: @file{Makefile} to use them.
1784:
1785: Note that Encore systems are supported only under BSD.
1786:
1.1.1.6 root 1787: For Sparc (Sun 4) machines, use @file{tm-sparc.h} with operating system
1788: version 4, and @file{tm-sun4os3.h} with system version 3.
1789:
1.1.1.10 root 1790: For Convex systems before version 8.1, use @file{tm-conv1os7.h} or
1791: @file{tm-conv2os7.h}. For versions 8.1 and greater, use @file{tm-convex1.h}
1792: or @file{tm-convex2.h}. You should also bootstrap GCC with @code{pcc}
1793: rather than @code{cc}; one way to do this is with the following commands.
1794:
1795: @example
1796: ln -s /bin/pcc ./cc
1797: set path = (. $path)
1798: @end example
1799:
1.1 root 1800: @item
1801: Make a symbolic link named @file{md} to the machine description
1.1.1.7 root 1802: pattern file. It should be in the @file{config} subdirectory and its
1803: name should be @file{@var{machine}.md}; but @var{machine} is often not
1804: the same as the name used in the @file{tm.h} file because the
1805: @file{md} files are more general.
1.1 root 1806:
1807: @item
1808: Make a symbolic link named @file{aux-output.c} to the output
1.1.1.7 root 1809: subroutine file for your machine. It should be in the @file{config}
1810: subdirectory and its name should be @file{out-@var{machine}.c}.
1.1 root 1811: @end itemize
1812:
1813: @item
1814: Make sure the Bison parser generator is installed. (This is
1815: unnecessary if the Bison output files @file{c-parse.tab.c} and
1816: @file{cexp.c} are more recent than @file{c-parse.y} and @file{cexp.y}
1817: and you do not plan to change the @samp{.y} files.)
1818:
1.1.1.9 root 1819: Bison versions older than Sept 8, 1988 will produce incorrect output
1.1 root 1820: for @file{c-parse.tab.c}.
1821:
1822: @item
1.1.1.10 root 1823: If you have a previous version of GCC installed, then chances are
1824: you can compile the new version with that. Do the following:
1825:
1826: @example
1827: make CC="gcc -O"
1828: @end example
1829:
1830: @noindent
1831: Since this produces an optimized executable right away, there is no need
1832: to bootstrap the result with itself except to test it. Therefore, you can
1833: skip directly to the @samp{make install} step below.
1834:
1835: @item
1.1 root 1836: Build the compiler. Just type @samp{make} in the compiler directory.
1837:
1.1.1.2 root 1838: Ignore any warnings you may see about ``statement not reached'' in the
1839: @file{insn-emit.c}; they are normal. Any other compilation errors may
1840: represent bugs in the port to your machine or operating system, and
1841: should be investigated and reported (@pxref{Bugs}).
1842:
1.1.1.9 root 1843: Some commercial compilers fail to compile GNU CC because they have bugs
1844: or limitations. For example, the Microsoft compiler is said to run out
1845: of macro space. Some Ultrix compilers run out of expression space; then
1846: you need to break up the statement where the problem happens.
1.1.1.7 root 1847:
1848: @item
1.1.1.5 root 1849: If you are using COFF-encapsulation, you must convert @file{gnulib} to
1850: a GNU-format library at this point. See the file @file{README-ENCAP}
1851: in the directory containing the GNU binary file utilities, for
1852: directions.
1853:
1854: @item
1.1 root 1855: Move the first-stage object files and executables into a subdirectory
1856: with this command:
1857:
1858: @example
1859: make stage1
1860: @end example
1861:
1862: The files are moved into a subdirectory named @file{stage1}.
1863: Once installation is complete, you may wish to delete these files
1864: with @code{rm -r stage1}.
1865:
1866: @item
1867: Recompile the compiler with itself, with this command:
1868:
1869: @example
1870: make CC=stage1/gcc CFLAGS="-g -O -Bstage1/"
1871: @end example
1872:
1.1.1.10 root 1873: This is called making the stage 2 compiler.
1874:
1.1 root 1875: On a 68000 or 68020 system lacking floating point hardware,
1876: unless you have selected a @file{tm.h} file that expects by default
1877: that there is no such hardware, do this instead:
1878:
1879: @example
1880: make CC=stage1/gcc CFLAGS="-g -O -Bstage1/ -msoft-float"
1881: @end example
1882:
1883: @item
1884: If you wish to test the compiler by compiling it with itself one more
1.1.1.7 root 1885: time, do this (in C shell):
1.1 root 1886:
1887: @example
1888: make stage2
1889: make CC=stage2/gcc CFLAGS="-g -O -Bstage2/"
1890: foreach file (*.o)
1891: cmp $file stage2/$file
1892: end
1893: @end example
1894:
1.1.1.7 root 1895: @noindent
1.1.1.10 root 1896: This is called making the stage 3 compiler. Aside from the @samp{-B}
1897: option, the options should be the same as when you made the stage 2
1898: compiler.
1.1 root 1899:
1.1.1.7 root 1900: The @code{foreach} command (written in C shell) will notify you if any of
1901: these stage 3 object files differs from those of stage 2. On BSD systems,
1902: any difference, no matter how innocuous, indicates that the stage 2
1903: compiler has compiled GNU CC incorrectly, and is therefore a potentially
1904: serious bug which you should investigate and report (@pxref{Bugs}).
1905:
1906: On systems that use COFF object files, bytes 5 to 8 will always be
1907: different, since it is a timestamp. On these systems, you can do the
1908: comparison as follows (in Bourne shell):
1909:
1910: @example
1911: for file in *.o; do
1912: echo $file
1.1.1.10 root 1913: tail +10c $file > foo1
1914: tail +10c stage2/$file > foo2
1.1.1.7 root 1915: cmp foo1 foo2
1916: done
1917: @end example
1918:
1.1.1.10 root 1919: On MIPS machines, you should use the shell script @file{ecoff-cmp}
1920: to compare two object files.
1921:
1.1 root 1922: @item
1923: Install the compiler driver, the compiler's passes and run-time support.
1924: You can use the following command:
1925:
1926: @example
1927: make install
1928: @end example
1929:
1930: @noindent
1931: This copies the files @file{cc1}, @file{cpp} and @file{gnulib} to
1932: files @file{gcc-cc1}, @file{gcc-cpp} and @file{gcc-gnulib} in
1933: directory @file{/usr/local/lib}, which is where the compiler driver
1934: program looks for them. It also copies the driver program @file{gcc}
1.1.1.6 root 1935: into the directory @file{/usr/local/bin}, so that it appears in typical
1.1 root 1936: execution search paths.@refill
1937:
1938: @strong{Warning: there is a bug in @code{alloca} in the Sun library.
1939: To avoid this bug, install the binaries of GNU CC that were compiled
1940: by GNU CC. They use @code{alloca} as a built-in function and never
1941: the one in the library.}
1942:
1943: @strong{Warning: the GNU CPP may not work for @file{ioctl.h},
1944: @file{ttychars.h} and other system header files unless the
1945: @samp{-traditional} option is used.} The bug is in the header files:
1946: at least on some machines, they rely on behavior that is incompatible
1947: with ANSI C. This behavior consists of substituting for macro
1948: argument names when they appear inside of character constants. The
1949: @samp{-traditional} option tells GNU CC to behave the way these
1950: headers expect.
1951:
1952: Because of this problem, you might prefer to configure GNU CC to use
1953: the system's own C preprocessor. To do so, make the file
1954: @file{/usr/local/lib/gcc-cpp} a link to @file{/lib/cpp}.
1955:
1956: Alternatively, on Sun systems and 4.3BSD at least, you can correct the
1957: include files by running the shell script @file{fixincludes}. This
1958: installs modified, corrected copies of the files @file{ioctl.h},
1959: @file{ttychars.h} and many others, in a special directory where only
1.1.1.2 root 1960: GNU CC will normally look for them. This script will work on various
1.1.1.6 root 1961: systems because it chooses the files by searching all the system
1.1.1.2 root 1962: headers for the problem cases that we know about.
1.1.1.10 root 1963:
1964: Use the following command to do this:
1965:
1966: @example
1967: make includes
1968: @end example
1969:
1970: @noindent
1971: If you selected a different directory for GNU CC installation when you
1972: installed it, by specifying the Make variable @code{prefix} or
1973: @code{libdir}, specify it the same way in this command.
1974:
1975: Note that some systems are starting to come with ANSI C system header
1976: files. On these systems, don't run @file{fixincludes}; it may not work,
1977: and is certainly not necessary.
1.1 root 1978: @end enumerate
1979:
1980: If you cannot install the compiler's passes and run-time support in
1981: @file{/usr/local/lib}, you can alternatively use the @samp{-B} option to
1982: specify a prefix by which they may be found. The compiler concatenates
1983: the prefix with the names @file{cpp}, @file{cc1} and @file{gnulib}.
1984: Thus, you can put the files in a directory @file{/usr/foo/gcc} and
1985: specify @samp{-B/usr/foo/gcc/} when you run GNU CC.
1986:
1987: Also, you can specify an alternative default directory for these files
1988: by setting the Make variable @code{libdir} when you make GNU CC.
1989:
1.1.1.8 root 1990: @node Other Dir, Sun Install, Installation, Installation
1991: @section Compilation in a Separate Directory
1.1 root 1992:
1.1.1.8 root 1993: If you wish to build the object files and executables in a directory
1994: other than the one containing the source files, here is what you must
1995: do differently:
1996:
1997: @enumerate
1998: @item
1999: Go to that directory before running @file{config.gcc}:
2000:
2001: @example
2002: mkdir gcc-sun3
2003: cd gcc-sun3
2004: @end example
1.1.1.4 root 2005:
1.1.1.8 root 2006: On systems that do not support symbolic links, this directory must be
2007: on the same file system as the source code directory.
2008:
2009: @item
2010: Specify where to find @file{config.gcc} when you run it:
2011:
2012: @example
2013: ../gcc-1.36/config.gcc @dots{}
2014: @end example
2015:
2016: @item
2017: Specify where to find the sources, as an argument to @file{config.gcc}:
2018:
2019: @example
2020: ../gcc-1.36/config.gcc -srcdir=../gcc-1.36 sun3
2021: @end example
2022:
2023: The @samp{-srcdir=@var{dir}} option is not needed when the source
2024: directory is the parent of the current directory, because
2025: @file{config.gcc} detects that case automatically.
2026: @end enumerate
2027:
2028: Now, you can run @code{make} in that directory. You need not repeat the
2029: configuration steps shown above, when ordinary source files change. You
2030: must, however, run @code{config.gcc} again when the configuration files
2031: change, if your system does not support symbolic links.
2032:
2033: @node Sun Install, 3b1 Install, Other Dir, Installation
2034: @section Installing GNU CC on the Sun
2035:
2036: Make sure the environment variable @code{FLOAT_OPTION} is not set when
2037: you compile @file{gnulib}. If this option were set to @code{f68881}
2038: when @file{gnulib} is compiled, the resulting code would demand to be
2039: linked with a special startup file and would not link properly without
2040: special pains.
2041:
2042: There is a bug in @code{alloca} in certain versions of the Sun library.
2043: To avoid this bug, install the binaries of GNU CC that were compiled by
2044: GNU CC. They use @code{alloca} as a built-in function and never the one
2045: in the library.
2046:
1.1.1.11! root 2047: Some versions of the Sun compiler crash when compiling GNU CC, with a
! 2048: segmentation fault in cpp. This can sometimes be due to the bulk of
! 2049: data in the environment variables. You may be able to avoid it by using
! 2050: the following command to compile GNU CC with Sun CC:
1.1.1.8 root 2051:
2052: @example
2053: make CC="TERMCAP=x OBJS=x LIBFUNCS=x STAGESTUFF=x cc"
2054: @end example
2055:
1.1.1.11! root 2056: Another problem that often happens on Suns is that you get a crash when
! 2057: building stage 2, when @code{genflags} is run.
! 2058:
! 2059: One reason for such as crash is if you configured GNU CC for the wrong
! 2060: version of SunOS. Starting with version 1.38, configurations @code{sun3}
! 2061: and @code{sun4} are for SunOS 4, so this problem should no longer happen.
! 2062:
! 2063: Another cause of the same symptom is having installed the GNU linker
! 2064: with an earlier version of SunOS. The version that worked before
! 2065: stopped working due to a change in the format of executables in SunOS
! 2066: 4.1. Many sites have installed the GNU linker as
! 2067: @file{/usr/local/lib/gcc-ld}, often as part of installing GNU C++. So
! 2068: if you get such crashes and you have used the proper configuration, try
! 2069: deleting @file{/usr/local/lib/gcc-ld}.
! 2070:
! 2071: The current version of the GNU linker, found in the current binutils
! 2072: release, does work with SunOS 4.1.
! 2073:
1.1.1.10 root 2074: @node 3b1 Install, SCO Install, Sun Install, Installation
1.1.1.8 root 2075: @section Installing GNU CC on the 3b1
2076:
2077: Installing GNU CC on the 3b1 is difficult if you do not already have
2078: GNU CC running, due to bugs in the installed C compiler. However,
2079: the following procedure might work. We are unable to test it.
1.1 root 2080:
2081: @enumerate
2082: @item
1.1.1.8 root 2083: Comment out the @samp{#include "config.h"} line on line 37 of
2084: @file{cccp.c} and do @samp{make cpp}. This makes a preliminary version
2085: of GNU cpp.
1.1 root 2086:
2087: @item
1.1.1.8 root 2088: Save the old @file{/lib/cpp} and copy the preliminary GNU cpp to that
2089: file name.
1.1.1.5 root 2090:
1.1.1.8 root 2091: @item
2092: Undo your change in @file{cccp.c}, or reinstall the original version,
2093: and do @samp{make cpp} again.
2094:
2095: @item
2096: Copy this final version of GNU cpp into @file{/lib/cpp}.
2097:
2098: @item
1.1.1.9 root 2099: Replace every occurrence of @code{obstack_free} in @file{tree.c}
1.1.1.8 root 2100: with @code{_obstack_free}.
2101:
2102: @item
2103: Run @code{make} to get the first-stage GNU CC.
2104:
2105: @item
2106: Reinstall the original version of @file{/lib/cpp}.
2107:
2108: @item
2109: Now you can compile GNU CC with itself and install it in the normal
2110: fashion.
1.1 root 2111: @end enumerate
2112:
1.1.1.9 root 2113: If you have installed an earlier version of GCC, you can compile the
2114: newer version with that. However, you will run into trouble compiling
2115: @file{gnulib}, since that is normally compiled with CC. To solve the
2116: problem, uncomment this line in @file{Makefile}:
2117:
2118: @example
2119: CCLIBFLAGS = -B/usr/local/lib/gcc- -tp -Wp,-traditional
2120: @end example
2121:
1.1.1.10 root 2122: @node SCO Install, VMS Install, 3B1 Install, Installation
2123: @section Installing GNU CC on SCO System V 3.2
2124: @cindex Installation on SCO systems
2125:
2126: The compiler that comes with this system does not work properly with
2127: @samp{-O}. Therefore, you should redefine the Make variable
2128: @code{CCLIBFLAGS} not to use @samp{-O}.
2129:
1.1.1.11! root 2130: You should also edit @file{Makefile} to enable the lines that set
! 2131: @code{CLIB} to @code{-lPW}, and the ones specifically labeled as being
! 2132: for SCO, that set @code{RANLIB}, and that set @code{CC} and @code{OLDCC}
! 2133: to @code{rcc}.
1.1.1.10 root 2134:
1.1.1.11! root 2135: Also, edit the definition of @code{USER_H} to remove the file @file{limits.h}.
1.1.1.10 root 2136:
1.1.1.11! root 2137: Then you can run @samp{config.gcc i386-sco} and finish building GNU CC
! 2138: normally.
1.1.1.10 root 2139:
1.1.1.11! root 2140: The same recipe should work on ESIX, but use @samp{config.gcc i386-esix}
! 2141: instead.
1.1.1.10 root 2142:
2143: @node VMS Install, HPUX Install, SCO Install, Installation
1.1.1.8 root 2144: @section Installing GNU CC on VMS
2145:
2146: The VMS version of GNU CC is distributed in a backup saveset containing
2147: both source code and precompiled binaries.
2148:
2149: To install the @file{gcc} command so you can use the compiler easily, in
1.1 root 2150: the same manner as you use the VMS C compiler, you must install the VMS CLD
2151: file for GNU CC as follows:
2152:
2153: @enumerate
2154: @item
2155: Define the VMS logical names @samp{GNU_CC} and @samp{GNU_CC_INCLUDE}
2156: to point to the directories where the GNU CC executables
1.1.1.8 root 2157: (@file{gcc-cpp}, @file{gcc-cc1}, etc.) and the C include files are
1.1 root 2158: kept. This should be done with the commands:@refill
2159:
2160: @example
1.1.1.8 root 2161: $ assign /super /system disk:[gcc.] gnu_cc
2162: $ assign /super /system disk:[gcc.include.] gnu_cc_include
1.1 root 2163: @end example
2164:
2165: @noindent
2166: with the appropriate disk and directory names. These commands can be
2167: placed in your system startup file so they will be executed whenever
1.1.1.8 root 2168: the machine is rebooted. You may, if you choose, do this via the
2169: @file{GCC_INSTALL.COM} script in the @file{[GCC]} directory.
1.1 root 2170:
2171: @item
1.1.1.8 root 2172: Install the @file{GCC} command with the command line:
1.1 root 2173:
2174: @example
1.1.1.8 root 2175: $ set command /table=sys$library:dcltables gnu_cc:[000000]gcc
1.1 root 2176: @end example
2177:
1.1.1.7 root 2178: @item
2179: To install the help file, do the following:
2180:
2181: @example
2182: $ lib/help sys$library:helplib.hlb gcc.hlp
2183: @end example
2184:
1.1 root 2185: @noindent
2186: Now you can invoke the compiler with a command like @samp{gcc /verbose
2187: file.c}, which is equivalent to the command @samp{gcc -v -c file.c} in
2188: Unix.
2189: @end enumerate
2190:
1.1.1.8 root 2191: We try to put corresponding binaries and sources on the VMS distribution
2192: tape. But sometimes the binaries will be from an older version that the
2193: sources, because we don't always have time to update them. (Use the
2194: @samp{/verbose} option to determine the version number of the binaries and
2195: compare it with the source file @file{version.c} to tell whether this is
2196: so.) In this case, you should use the binaries you get to recompile the
2197: sources. If you must recompile, here is how:
2198:
2199: @enumerate
2200: @item
2201: Copy the file @file{tm-vms.h} to @file{tm.h}, @file{xm-vms.h} to
2202: @file{config.h}, @file{vax.md} to @file{md.} and @file{out-vax.c}
2203: to @file{aux-output.c}. The files to be copied are found in the
2204: subdirectory named @file{config}; they should be copied to the
2205: main directory of GNU CC.@refill
2206:
2207: @item
2208: Setup the logical names and command tables as defined above. In
2209: addition, define the vms logical name @samp{GNU_BISON} to point at the
2210: to the directories where the Bison executable is kept. This should be
2211: done with the command:@refill
2212:
2213: @example
2214: $ assign /super /system disk:[bison.] gnu_bison
2215: @end example
2216:
2217: You may, if you choose, use the @file{INSTALL_BISON.COM} script in the
2218: @file{[BISON]} directory.
2219:
2220: @item
2221: Install the @samp{BISON} command with the command line:@refill
2222:
2223: @example
2224: $ set command /table=sys$library:dcltables gnu_bison:[000000]bison
2225: @end example
2226:
2227: @item
2228: Type @samp{@@make} to do recompile everything.
2229:
2230: If you are compiling with a version of GNU CC older than 1.33, specify
2231: @samp{/DEFINE=("inline=")} as an option in all the compilations. This
2232: requires editing all the @code{gcc} commands in @file{make-cc1.com}.
2233: (The older versions had problems supporting @code{inline}.) Once you
2234: have a working 1.33 or newer GNU CC, you can change this file back.
2235: @end enumerate
2236:
1.1.1.10 root 2237: With this version of GNU CC, @code{const} global variables now work
2238: properly. Unless, however, the @code{const} modifier is also specified
2239: in every external declaration of the variable in all of the source files
2240: that use that variable, the linker will issue warnings about conflicting
2241: attributes for the variable, since the linker does not know if the
2242: variable should be read-only. The program will still work, but the
2243: variable will be placed in writable storage.
2244:
2245: Under previous versions of GNU CC, the generated code would occasionally
2246: give strange results when linked to the sharable @file{VAXCRTL} library.
2247: Now this should work.
2248:
2249: Even with this version, however, GNU CC itself should not be linked to
2250: the sharable @file{VAXCRTL}, unless you force the linker to use the
2251: @code{qsort} routine from @file{gcclib.olb}. The @file{qsort} routine
2252: supplied with @file{VAXCRTL} has a bug which causes a compiler crash.
2253: The executable that is generated by @file{make-cc1.com} uses the
2254: non-shared version of @file{VAXCRTL} (and thus the @file{qsort} routine
2255: from @file{gcclib.olb}).
2256:
2257: Note that GNU CC on VMS now generates debugging information to describe
2258: the programs symbols to the VMS debugger. However, you need version 1.37
2259: or later of GAS in order to output them properly in the object file.
1.1.1.7 root 2260:
1.1.1.10 root 2261: @node HPUX Install, MIPS Install, VMS Install, Installation
1.1.1.9 root 2262: @section Installing GNU CC on HPUX
2263:
2264: To install GNU CC on HPUX, you must start by editing the file
2265: @file{Makefile}. Search for the string @samp{HPUX} to find comments
2266: saying what to change. You need to change some variable definitions and
2267: (if you are using GAS) some lines in the rule for the target
2268: @samp{gnulib}.
2269:
1.1.1.10 root 2270: To avoid errors when linking programs with @samp{-g}, create an empty
2271: library named @file{libg.a}. An easy way to do this is:
2272:
2273: @example
2274: ar rc /usr/local/lib/libg.a
2275: @end example
2276:
1.1.1.9 root 2277: To compile with the HPUX C compiler, you must specify get the file
2278: @file{alloca.c} from GNU Emacs. Then, when you run @code{make}, use
2279: this argument:
2280:
2281: @example
2282: make ALLOCA=alloca.o
2283: @end example
2284:
2285: When recompiling GNU CC with itself, do not define @code{ALLOCA}.
2286: Instead, an @samp{-I} option needs to be added to @code{CFLAGS} as
2287: follows:
2288:
2289: @example
2290: make CC=stage1/gcc CFLAGS="-g -O -Bstage1/ -I../binutils/hp-include"
2291: @end example
2292:
1.1.1.11! root 2293: @node MIPS Install, Tower Install, HPUX Install, Installation
1.1.1.10 root 2294: @section Installing GNU CC on MIPS
2295:
2296: To avoid errors when linking programs with @samp{-g}, create an empty
2297: library named @file{libg.a}. An easy way to do this is:
2298:
2299: @example
2300: ar rc /usr/local/lib/libg.a
2301: @end example
2302:
1.1.1.11! root 2303: @node Tower Install,, MIPS Install, Installation
! 2304: @section Installing GNU CC on an NCR Tower
! 2305:
! 2306: On an NCR Tower model 4x0 or 6x0, you may have trouble because the
! 2307: default maximum virtual address size of a process is just 1 Mb. Most
! 2308: often you will find this problem while compiling GNU CC with itself.
! 2309:
! 2310: The only way to solve the problem is to reconfigure the kernel.
! 2311: Add a line such as this to the configuration file:
! 2312:
! 2313: @example
! 2314: MAXUMEM = 4096
! 2315: @end example
! 2316:
! 2317: @noindent
! 2318: and then relink the kernel and reboot the machine.
! 2319:
! 2320:
1.1.1.10 root 2321: @node Trouble, Service, Installation, Top
2322: @chapter Known Causes of Trouble with GNU CC
1.1 root 2323:
2324: Here are some of the things that have caused trouble for people installing
2325: or using GNU CC.
2326:
2327: @itemize @bullet
2328: @item
2329: On certain systems, defining certain environment variables such as
1.1.1.8 root 2330: @code{CC} can interfere with the functioning of @code{make}.
1.1 root 2331:
2332: @item
2333: Cross compilation can run into trouble for certain machines because
2334: some target machines' assemblers require floating point numbers to be
2335: written as @emph{integer} constants in certain contexts.
2336:
2337: The compiler writes these integer constants by examining the floating
2338: point value as an integer and printing that integer, because this is
2339: simple to write and independent of the details of the floating point
2340: representation. But this does not work if the compiler is running on
2341: a different machine with an incompatible floating point format, or
2342: even a different byte-ordering.
2343:
1.1.1.5 root 2344: In addition, correct constant folding of floating point values
2345: requires representing them in the target machine's format.
2346: (The C standard does not quite require this, but in practice
2347: it is the only way to win.)
2348:
2349: It is now possible to overcome these problems by defining macros such
2350: as @code{REAL_VALUE_TYPE}. But doing so is a substantial amount of
2351: work for each target machine. @xref{Cross-compilation}.
1.1 root 2352:
2353: @item
2354: DBX rejects some files produced by GNU CC, though it accepts similar
2355: constructs in output from PCC. Until someone can supply a coherent
2356: description of what is valid DBX input and what is not, there is
2357: nothing I can do about these problems. You are on your own.
1.1.1.2 root 2358:
2359: @item
2360: Users often think it is a bug when GNU CC reports an error for code
2361: like this:
2362:
2363: @example
2364: int foo (short);
2365:
2366: int foo (x)
2367: short x;
2368: @{@dots{}@}
2369: @end example
2370:
1.1.1.4 root 2371: The error message is correct: this code really is erroneous, because the
2372: old-style non-prototype definition passes subword integers in their
2373: promoted types. In other words, the argument is really an @code{int},
2374: not a @code{short}. The correct prototype is this:
1.1.1.2 root 2375:
2376: @example
2377: int foo (int);
2378: @end example
2379:
2380: @item
2381: Users often think it is a bug when GNU CC reports an error for code
2382: like this:
2383:
2384: @example
2385: int foo (struct mumble *);
2386:
2387: struct mumble @{ @dots{} @};
2388:
2389: int foo (struct mumble *x)
2390: @{ @dots{} @}
2391: @end example
2392:
2393: This code really is erroneous, because the scope of @code{struct
2394: mumble} the prototype is limited to the argument list containing it.
2395: It does not refer to the @code{struct mumble} defined with file scope
2396: immediately below---they are two unrelated types with similar names in
2397: different scopes.
2398:
2399: But in the definition of @code{foo}, the file-scope type is used
2400: because that is available to be inherited. Thus, the definition and
2401: the prototype do not match, and you get an error.
2402:
2403: This behavior may seem silly, but it's what the ANSI standard
2404: specifies. It is easy enough for you to make your code work by moving
2405: the definition of @code{struct mumble} above the prototype. I don't
2406: think it's worth being incompatible for.
1.1 root 2407: @end itemize
2408:
1.1.1.10 root 2409: @node Service, Incompatibilities, Trouble, Top
2410: @chapter How To Get Help with GNU CC
2411:
2412: If you need help installing, using or changing GNU CC, there are two
2413: ways to find it:
2414:
2415: @itemize @bullet
2416: @item
2417: Send a message to a suitable network mailing list. First try
2418: @code{bug-gcc@@prep.ai.mit.edu}, and if that brings no response, try
2419: @code{info-gcc@@prep.ai.mit.edu}.
2420:
2421: @item
2422: Look in the service directory for someone who might help you for a fee.
2423: The service directory is found in the file named @file{SERVICE} in the
2424: GNU CC distribution.
2425: @end itemize
2426:
2427: @node Incompatibilities, Extensions, Service, Top
1.1 root 2428: @chapter Incompatibilities of GNU CC
2429:
2430: There are several noteworthy incompatibilities between GNU C and most
1.1.1.9 root 2431: existing (non-ANSI) versions of C. The @samp{-traditional} option
2432: eliminates most of these incompatibilities, @emph{but not all}, by
2433: telling GNU C to behave like older C compilers.
1.1 root 2434:
2435: @itemize @bullet
2436: @item
2437: GNU CC normally makes string constants read-only. If several
2438: identical-looking string constants are used, GNU CC stores only one
2439: copy of the string.
2440:
2441: One consequence is that you cannot call @code{mktemp} with a string
2442: constant argument. The function @code{mktemp} always alters the
2443: string its argument points to.
2444:
2445: Another consequence is that @code{sscanf} does not work on some
2446: systems when passed a string constant as its format control string.
2447: This is because @code{sscanf} incorrectly tries to write into the
1.1.1.4 root 2448: string constant. Likewise @code{fscanf} and @code{scanf}.
1.1 root 2449:
2450: The best solution to these problems is to change the program to use
2451: @code{char}-array variables with initialization strings for these
2452: purposes instead of string constants. But if this is not possible,
2453: you can use the @samp{-fwritable-strings} flag, which directs GNU CC
2454: to handle string constants the same way most C compilers do.
1.1.1.8 root 2455: @samp{-traditional} also has this effect, among others.
1.1 root 2456:
2457: @item
2458: GNU CC does not substitute macro arguments when they appear inside of
2459: string constants. For example, the following macro in GNU CC
2460:
2461: @example
2462: #define foo(a) "a"
2463: @end example
2464:
2465: @noindent
1.1.1.8 root 2466: will produce output @code{"a"} regardless of what the argument @var{a} is.
1.1 root 2467:
2468: The @samp{-traditional} option directs GNU CC to handle such cases
2469: (among others) in the old-fashioned (non-ANSI) fashion.
2470:
2471: @item
2472: When you use @code{setjmp} and @code{longjmp}, the only automatic
2473: variables guaranteed to remain valid are those declared
2474: @code{volatile}. This is a consequence of automatic register
2475: allocation. Consider this function:
2476:
2477: @example
2478: jmp_buf j;
2479:
2480: foo ()
2481: @{
2482: int a, b;
2483:
2484: a = fun1 ();
2485: if (setjmp (j))
2486: return a;
2487:
2488: a = fun2 ();
2489: /* @r{@code{longjmp (j)} may be occur in @code{fun3}.} */
2490: return a + fun3 ();
2491: @}
2492: @end example
2493:
2494: Here @code{a} may or may not be restored to its first value when the
2495: @code{longjmp} occurs. If @code{a} is allocated in a register, then
2496: its first value is restored; otherwise, it keeps the last value stored
2497: in it.
2498:
2499: If you use the @samp{-W} option with the @samp{-O} option, you will
2500: get a warning when GNU CC thinks such a problem might be possible.
2501:
1.1.1.2 root 2502: The @samp{-traditional} option directs GNU C to put variables in
2503: the stack by default, rather than in registers, in functions that
2504: call @code{setjmp}. This results in the behavior found in
2505: traditional C compilers.
2506:
1.1 root 2507: @item
2508: Declarations of external variables and functions within a block apply
2509: only to the block containing the declaration. In other words, they
2510: have the same scope as any other declaration in the same place.
2511:
2512: In some other C compilers, a @code{extern} declaration affects all the
2513: rest of the file even if it happens within a block.
2514:
2515: The @samp{-traditional} option directs GNU C to treat all @code{extern}
2516: declarations as global, like traditional compilers.
2517:
2518: @item
2519: In traditional C, you can combine @code{long}, etc., with a typedef name,
2520: as shown here:
2521:
2522: @example
2523: typedef int foo;
2524: typedef long foo bar;
2525: @end example
2526:
2527: In ANSI C, this is not allowed: @code{long} and other type modifiers
2528: require an explicit @code{int}. Because this criterion is expressed
2529: by Bison grammar rules rather than C code, the @samp{-traditional}
2530: flag cannot alter it.
2531:
2532: @item
2533: PCC allows typedef names to be used as function parameters. The
2534: difficulty described immediately above applies here too.
2535:
2536: @item
2537: PCC allows whitespace in the middle of compound assignment operators
2538: such as @samp{+=}. GNU CC, following the ANSI standard, does not
2539: allow this. The difficulty described immediately above applies here
2540: too.
2541:
2542: @item
2543: GNU CC will flag unterminated character constants inside of preprocessor
2544: conditionals that fail. Some programs have English comments enclosed in
2545: conditionals that are guaranteed to fail; if these comments contain
2546: apostrophes, GNU CC will probably report an error. For example,
2547: this code would produce an error:
2548:
2549: @example
2550: #if 0
2551: You can't expect this to work.
2552: #endif
2553: @end example
2554:
2555: The best solution to such a problem is to put the text into an actual
2556: C comment delimited by @samp{/*@dots{}*/}. However,
2557: @samp{-traditional} suppresses these error messages.
2558:
2559: @item
2560: When compiling functions that return @code{float}, PCC converts it to
2561: a double. GNU CC actually returns a @code{float}. If you are concerned
2562: with PCC compatibility, you should declare your functions to return
2563: @code{double}; you might as well say what you mean.
2564:
2565: @item
2566: When compiling functions that return structures or unions, GNU CC
1.1.1.6 root 2567: output code normally uses a method different from that used on most
2568: versions of Unix. As a result, code compiled with GNU CC cannot call
2569: a structure-returning function compiled with PCC, and vice versa.
1.1 root 2570:
1.1.1.6 root 2571: The method used by GNU CC is as follows: a structure or union which is 1,
1.1 root 2572: 2, 4 or 8 bytes long is returned like a scalar. A structure or union
2573: with any other size is stored into an address supplied by the caller
2574: in a special, fixed register.
2575:
2576: PCC usually handles all sizes of structures and unions by returning
2577: the address of a block of static storage containing the value. This
1.1.1.6 root 2578: method is not used in GNU CC because it is slower and nonreentrant.
1.1.1.5 root 2579:
1.1.1.6 root 2580: You can tell GNU CC to use the PCC convention with the option
2581: @samp{-fpcc-struct-return}.
1.1.1.8 root 2582:
2583: @item
2584: On the Sparc, GNU CC uses an incompatible calling convention for
2585: structures. It passes them by including their contents in the argument
2586: list, whereas the standard compiler passes them effectively by
2587: reference.
2588:
2589: This really ought to be fixed, but such calling conventions are not
2590: yet supported in GNU CC, so it isn't straightforward to fix it.
2591:
2592: The convention for structure returning is also incompatible, and
2593: @samp{-fpcc-struct-return} does not help.
1.1.1.10 root 2594:
2595: @item
2596: On Ultrix, the Fortran compiler expects registers 2 through 5 to be saved
2597: by function calls. We have not been able to tell whether the C compiler
2598: agrees with the Fortran compiler. Currently, GNU CC treats these registers
2599: as temporaries on the Vax, which is compatible with BSD Unix.
2600:
2601: If we learn for certain that Ultrix has departed from the traditional
2602: BSD calling convention, we will change GNU CC for Ultrix to fit. In the
2603: mean time, you can use these options to produce code compatible with the
2604: Fortran compiler:
2605:
2606: @example
2607: -fcall-saved-r2 -fcall-saved-r3 -fcall-saved-r4 -fcall-saved-r5
2608: @end example
1.1 root 2609: @end itemize
2610:
2611: @node Extensions, Bugs, Incompatibilities, Top
2612: @chapter GNU Extensions to the C Language
2613:
2614: GNU C provides several language features not found in ANSI standard C.
2615: (The @samp{-pedantic} option directs GNU CC to print a warning message if
2616: any of these features is used.) To test for the availability of these
2617: features in conditional compilation, check for a predefined macro
2618: @code{__GNUC__}, which is always defined under GNU CC.
2619:
2620: @menu
2621: * Statement Exprs:: Putting statements and declarations inside expressions.
2622: * Naming Types:: Giving a name to the type of some expression.
1.1.1.9 root 2623: * Typeof:: @code{typeof}: referring to the type of an expression.
2624: * Lvalues:: Using @samp{?:}, @samp{,} and casts in lvalues.
2625: * Conditionals:: Omitting the middle operand of a @samp{?:} expression.
2626: * Zero-Length:: Zero-length arrays.
2627: * Variable-Length:: Arrays whose length is computed at run time.
2628: * Subscripting:: Any array can be subscripted, even if not an lvalue.
2629: * Pointer Arith:: Arithmetic on @code{void}-pointers and function pointers.
2630: * Initializers:: Non-constant initializers.
2631: * Constructors:: Constructor expressions give structures, unions
2632: or arrays as values.
1.1.1.5 root 2633: * Function Attributes:: Declaring that functions have no side effects,
1.1.1.9 root 2634: or that they can never return.
1.1 root 2635: * Dollar Signs:: Dollar sign is allowed in identifiers.
2636: * Alignment:: Inquiring about the alignment of a type or variable.
2637: * Inline:: Defining inline functions (as fast as macros).
1.1.1.9 root 2638: * Extended Asm:: Assembler instructions with C expressions as operands.
2639: (With them you can define ``built-in'' functions.)
2640: * Asm Labels:: Specifying the assembler name to use for a C symbol.
1.1.1.8 root 2641: * Explicit Reg Vars:: Defining variables residing in specified registers.
2642: * Alternate Keywords:: @code{__const__}, @code{__asm__}, etc., for header files.
1.1 root 2643: @end menu
2644:
2645: @node Statement Exprs, Naming Types, Extensions, Extensions
2646: @section Statements and Declarations inside of Expressions
2647:
2648: A compound statement in parentheses may appear inside an expression in GNU
2649: C. This allows you to declare variables within an expression. For
2650: example:
2651:
2652: @example
2653: (@{ int y = foo (); int z;
2654: if (y > 0) z = y;
2655: else z = - y;
2656: z; @})
2657: @end example
2658:
2659: @noindent
2660: is a valid (though slightly more complex than necessary) expression
2661: for the absolute value of @code{foo ()}.
2662:
2663: This feature is especially useful in making macro definitions ``safe'' (so
2664: that they evaluate each operand exactly once). For example, the
2665: ``maximum'' function is commonly defined as a macro in standard C as
2666: follows:
2667:
2668: @example
2669: #define max(a,b) ((a) > (b) ? (a) : (b))
2670: @end example
2671:
2672: @noindent
2673: But this definition computes either @var{a} or @var{b} twice, with bad
2674: results if the operand has side effects. In GNU C, if you know the
2675: type of the operands (here let's assume @code{int}), you can define
2676: the macro safely as follows:
2677:
2678: @example
2679: #define maxint(a,b) \
2680: (@{int _a = (a), _b = (b); _a > _b ? _a : _b; @})
2681: @end example
2682:
2683: Embedded statements are not allowed in constant expressions, such as
2684: the value of an enumeration constant, the width of a bit field, or
2685: the initial value of a static variable.
2686:
2687: If you don't know the type of the operand, you can still do this, but you
2688: must use @code{typeof} (@pxref{Typeof}) or type naming (@pxref{Naming
2689: Types}).
2690:
2691: @node Naming Types, Typeof, Statement Exprs, Extensions
2692: @section Naming an Expression's Type
2693:
2694: You can give a name to the type of an expression using a @code{typedef}
2695: declaration with an initializer. Here is how to define @var{name} as a
2696: type name for the type of @var{exp}:
2697:
2698: @example
2699: typedef @var{name} = @var{exp};
2700: @end example
2701:
2702: This is useful in conjunction with the statements-within-expressions
2703: feature. Here is how the two together can be used to define a safe
2704: ``maximum'' macro that operates on any arithmetic type:
2705:
2706: @example
2707: #define max(a,b) \
2708: (@{typedef _ta = (a), _tb = (b); \
2709: _ta _a = (a); _tb _b = (b); \
2710: _a > _b ? _a : _b; @})
2711: @end example
2712:
2713: The reason for using names that start with underscores for the local
2714: variables is to avoid conflicts with variable names that occur within the
2715: expressions that are substituted for @code{a} and @code{b}. Eventually we
2716: hope to design a new form of declaration syntax that allows you to declare
2717: variables whose scopes start only after their initializers; this will be a
2718: more reliable way to prevent such conflicts.
2719:
2720: @node Typeof, Lvalues, Naming Types, Extensions
2721: @section Referring to a Type with @code{typeof}
2722:
2723: Another way to refer to the type of an expression is with @code{typeof}.
2724: The syntax of using of this keyword looks like @code{sizeof}, but the
2725: construct acts semantically like a type name defined with @code{typedef}.
2726:
2727: There are two ways of writing the argument to @code{typeof}: with an
2728: expression or with a type. Here is an example with an expression:
2729:
2730: @example
2731: typeof (x[0](1))
2732: @end example
2733:
2734: @noindent
2735: This assumes that @code{x} is an array of functions; the type described
2736: is that of the values of the functions.
2737:
2738: Here is an example with a typename as the argument:
2739:
2740: @example
2741: typeof (int *)
2742: @end example
2743:
2744: @noindent
2745: Here the type described is that of pointers to @code{int}.
2746:
1.1.1.7 root 2747: If you are writing a header file that must work when included in ANSI C
1.1.1.8 root 2748: programs, write @code{__typeof__} instead of @code{typeof}.
1.1.1.7 root 2749: @xref{Alternate Keywords}.
2750:
1.1 root 2751: A @code{typeof}-construct can be used anywhere a typedef name could be
2752: used. For example, you can use it in a declaration, in a cast, or inside
2753: of @code{sizeof} or @code{typeof}.
2754:
2755: @itemize @bullet
2756: @item
2757: This declares @code{y} with the type of what @code{x} points to.
2758:
2759: @example
2760: typeof (*x) y;
2761: @end example
2762:
2763: @item
2764: This declares @code{y} as an array of such values.
2765:
2766: @example
2767: typeof (*x) y[4];
2768: @end example
2769:
2770: @item
2771: This declares @code{y} as an array of pointers to characters:
2772:
2773: @example
2774: typeof (typeof (char *)[4]) y;
2775: @end example
2776:
2777: @noindent
2778: It is equivalent to the following traditional C declaration:
2779:
2780: @example
2781: char *y[4];
2782: @end example
2783:
2784: To see the meaning of the declaration using @code{typeof}, and why it
2785: might be a useful way to write, let's rewrite it with these macros:
2786:
2787: @example
2788: #define pointer(T) typeof(T *)
2789: #define array(T, N) typeof(T [N])
2790: @end example
2791:
2792: @noindent
2793: Now the declaration can be rewritten this way:
2794:
2795: @example
2796: array (pointer (char), 4) y;
2797: @end example
2798:
2799: @noindent
1.1.1.8 root 2800: Thus, @code{array (pointer (char), 4)} is the type of arrays of 4
1.1 root 2801: pointers to @code{char}.
2802: @end itemize
2803:
2804: @node Lvalues, Conditionals, Typeof, Extensions
2805: @section Generalized Lvalues
2806:
2807: Compound expressions, conditional expressions and casts are allowed as
2808: lvalues provided their operands are lvalues. This means that you can take
2809: their addresses or store values into them.
2810:
2811: For example, a compound expression can be assigned, provided the last
2812: expression in the sequence is an lvalue. These two expressions are
2813: equivalent:
2814:
2815: @example
2816: (a, b) += 5
2817: a, (b += 5)
2818: @end example
2819:
2820: Similarly, the address of the compound expression can be taken. These two
2821: expressions are equivalent:
2822:
2823: @example
2824: &(a, b)
2825: a, &b
2826: @end example
2827:
2828: A conditional expression is a valid lvalue if its type is not void and the
2829: true and false branches are both valid lvalues. For example, these two
2830: expressions are equivalent:
2831:
2832: @example
2833: (a ? b : c) = 5
2834: (a ? b = 5 : (c = 5))
2835: @end example
2836:
2837: A cast is a valid lvalue if its operand is valid. Taking the address of
2838: the cast is the same as taking the address without a cast, except for the
2839: type of the result. For example, these two expressions are equivalent (but
1.1.1.8 root 2840: the second may be valid when the type of @code{a} does not permit a cast to
2841: @code{int *}).
1.1 root 2842:
2843: @example
2844: &(int *)a
2845: (int **)&a
2846: @end example
2847:
2848: A simple assignment whose left-hand side is a cast works by converting the
2849: right-hand side first to the specified type, then to the type of the inner
2850: left-hand side expression. After this is stored, the value is converter
2851: back to the specified type to become the value of the assignment. Thus, if
1.1.1.8 root 2852: @code{a} has type @code{char *}, the following two expressions are
1.1 root 2853: equivalent:
2854:
2855: @example
2856: (int)a = 5
2857: (int)(a = (char *)5)
2858: @end example
2859:
2860: An assignment-with-arithmetic operation such as @samp{+=} applied to a cast
2861: performs the arithmetic using the type resulting from the cast, and then
2862: continues as in the previous case. Therefore, these two expressions are
2863: equivalent:
2864:
2865: @example
2866: (int)a += 5
2867: (int)(a = (char *) ((int)a + 5))
2868: @end example
2869:
2870: @node Conditionals, Zero-Length, Lvalues, Extensions
2871: @section Conditional Expressions with Omitted Middle-Operands
2872:
2873: The middle operand in a conditional expression may be omitted. Then
2874: if the first operand is nonzero, its value is the value of the conditional
2875: expression.
2876:
2877: Therefore, the expression
2878:
2879: @example
2880: x ? : y
2881: @end example
2882:
2883: @noindent
2884: has the value of @code{x} if that is nonzero; otherwise, the value of
2885: @code{y}.
2886:
2887: This example is perfectly equivalent to
2888:
2889: @example
2890: x ? x : y
2891: @end example
2892:
2893: @noindent
2894: In this simple case, the ability to omit the middle operand is not
2895: especially useful. When it becomes useful is when the first operand does,
2896: or may (if it is a macro argument), contain a side effect. Then repeating
2897: the operand in the middle would perform the side effect twice. Omitting
2898: the middle operand uses the value already computed without the undesirable
2899: effects of recomputing it.
2900:
2901: @node Zero-Length, Variable-Length, Conditionals, Extensions
2902: @section Arrays of Length Zero
2903:
2904: Zero-length arrays are allowed in GNU C. They are very useful as the last
2905: element of a structure which is really a header for a variable-length
2906: object:
2907:
2908: @example
2909: struct line @{
2910: int length;
2911: char contents[0];
2912: @};
2913:
2914: @{
2915: struct line *thisline
2916: = (struct line *) malloc (sizeof (struct line) + this_length);
2917: thisline->length = this_length;
2918: @}
2919: @end example
2920:
2921: In standard C, you would have to give @code{contents} a length of 1, which
2922: means either you waste space or complicate the argument to @code{malloc}.
2923:
2924: @node Variable-Length, Subscripting, Zero-Length, Extensions
2925: @section Arrays of Variable Length
2926:
2927: Variable-length automatic arrays are allowed in GNU C. These arrays are
2928: declared like any other automatic arrays, but with a length that is not a
2929: constant expression. The storage is allocated at that time and
2930: deallocated when the brace-level is exited. For example:
2931:
2932: @example
2933: FILE *concat_fopen (char *s1, char *s2, char *mode)
2934: @{
2935: char str[strlen (s1) + strlen (s2) + 1];
2936: strcpy (str, s1);
2937: strcat (str, s2);
2938: return fopen (str, mode);
2939: @}
2940: @end example
2941:
1.1.1.7 root 2942: You can also use variable-length arrays as arguments to functions:
1.1 root 2943:
2944: @example
2945: struct entry
1.1.1.7 root 2946: tester (int len, char data[len])
1.1 root 2947: @{
1.1.1.7 root 2948: @dots{}
1.1 root 2949: @}
2950: @end example
2951:
2952: The length of an array is computed on entry to the brace-level where the
2953: array is declared and is remembered for the scope of the array in case you
2954: access it with @code{sizeof}.
2955:
2956: Jumping or breaking out of the scope of the array name will also deallocate
2957: the storage. Jumping into the scope is not allowed; you will get an error
2958: message for it.
2959:
2960: You can use the function @code{alloca} to get an effect much like
2961: variable-length arrays. The function @code{alloca} is available in
2962: many other C implementations (but not in all). On the other hand,
2963: variable-length arrays are more elegant.
2964:
2965: There are other differences between these two methods. Space allocated
2966: with @code{alloca} exists until the containing @emph{function} returns.
2967: The space for a variable-length array is deallocated as soon as the array
2968: name's scope ends. (If you use both variable-length arrays and
2969: @code{alloca} in the same function, deallocation of a variable-length array
2970: will also deallocate anything more recently allocated with @code{alloca}.)
2971:
2972: @node Subscripting, Pointer Arith, Variable-Length, Extensions
2973: @section Non-Lvalue Arrays May Have Subscripts
2974:
2975: Subscripting is allowed on arrays that are not lvalues, even though the
2976: unary @samp{&} operator is not. For example, this is valid in GNU C though
2977: not valid in other C dialects:
2978:
2979: @example
2980: struct foo @{int a[4];@};
2981:
2982: struct foo f();
2983:
2984: bar (int index)
2985: @{
2986: return f().a[index];
2987: @}
2988: @end example
2989:
2990: @node Pointer Arith, Initializers, Subscripting, Extensions
2991: @section Arithmetic on @code{void}-Pointers and Function Pointers
2992:
2993: In GNU C, addition and subtraction operations are supported on pointers to
2994: @code{void} and on pointers to functions. This is done by treating the
2995: size of a @code{void} or of a function as 1.
2996:
2997: A consequence of this is that @code{sizeof} is also allowed on @code{void}
2998: and on function types, and returns 1.
2999:
1.1.1.8 root 3000: The option @samp{-Wpointer-arith} requests a warning if these extensions
3001: are used.
3002:
1.1 root 3003: @node Initializers, Constructors, Pointer Arith, Extensions
3004: @section Non-Constant Initializers
3005:
1.1.1.8 root 3006: The elements of an aggregate initializer for an automatic variable are
3007: not required to be constant expressions in GNU C. Here is an example of
3008: an initializer with run-time varying elements:
1.1 root 3009:
3010: @example
3011: foo (float f, float g)
3012: @{
3013: float beat_freqs[2] = @{ f-g, f+g @};
3014: @dots{}
3015: @}
3016: @end example
3017:
1.1.1.5 root 3018: @node Constructors, Function Attributes, Initializers, Extensions
1.1 root 3019: @section Constructor Expressions
3020:
3021: GNU C supports constructor expressions. A constructor looks like a cast
3022: containing an initializer. Its value is an object of the type specified in
3023: the cast, containing the elements specified in the initializer. The type
3024: must be a structure, union or array type.
3025:
3026: Assume that @code{struct foo} and @code{structure} are declared as shown:
3027:
3028: @example
3029: struct foo @{int a; char b[2];@} structure;
3030: @end example
3031:
3032: @noindent
1.1.1.8 root 3033: Here is an example of constructing a @code{struct foo} with a constructor:
1.1 root 3034:
3035: @example
3036: structure = ((struct foo) @{x + y, 'a', 0@});
3037: @end example
3038:
3039: @noindent
3040: This is equivalent to writing the following:
3041:
3042: @example
3043: @{
3044: struct foo temp = @{x + y, 'a', 0@};
3045: structure = temp;
3046: @}
3047: @end example
3048:
3049: You can also construct an array. If all the elements of the constructor
3050: are (made up of) simple constant expressions, suitable for use in
3051: initializers, then the constructor is an lvalue and can be coerced to a
3052: pointer to its first element, as shown here:
3053:
3054: @example
3055: char **foo = (char *[]) @{ "x", "y", "z" @};
3056: @end example
3057:
3058: Array constructors whose elements are not simple constants are not very
3059: useful, because the constructor is not an lvalue. There are only two valid
3060: ways to use it: to subscript it, or initialize an array variable with it.
3061: The former is probably slower than a @code{switch} statement, while the
3062: latter does the same thing an ordinary C initializer would do.
3063:
3064: @example
3065: output = ((int[]) @{ 2, x, 28 @}) [input];
3066: @end example
3067:
1.1.1.8 root 3068: @node Function Attributes, Dollar Signs, Constructors, Extensions
1.1.1.5 root 3069: @section Declaring Attributes of Functions
3070:
3071: In GNU C, you declare certain things about functions called in your program
3072: which help the compiler optimize function calls.
3073:
3074: A few functions, such as @code{abort} and @code{exit}, cannot return.
3075: These functions should be declared @code{volatile}. For example,
3076:
3077: @example
3078: extern volatile void abort ();
3079: @end example
3080:
3081: @noindent
3082: tells the compiler that it can assume that @code{abort} will not return.
3083: This makes slightly better code, but more importantly it helps avoid
3084: spurious warnings of uninitialized variables.
3085:
3086: Many functions do not examine any values except their arguments, and
3087: have no effects except the return value. Such a function can be subject
3088: to common subexpression elimination and loop optimization just as an
3089: arithmetic operator would be. These functions should be declared
3090: @code{const}. For example,
3091:
3092: @example
3093: extern const void square ();
3094: @end example
3095:
3096: @noindent
3097: says that the hypothetical function @code{square} is safe to call
3098: fewer times than the program says.
3099:
3100: Note that a function that has pointer arguments and examines the data
3101: pointed to must @emph{not} be declared @code{const}. Likewise, a
1.1.1.10 root 3102: function that calls a non-@code{const} function usually must not be
1.1.1.5 root 3103: @code{const}.
3104:
3105: Some people object to this feature, claiming that ANSI C's @code{#pragma}
3106: should be used instead. There are two reasons I did not do this.
3107:
3108: @enumerate
3109: @item
3110: It is impossible to generate @code{#pragma} commands from a macro.
3111:
3112: @item
3113: The @code{#pragma} command is just as likely as these keywords to mean
3114: something else in another compiler.
3115: @end enumerate
3116:
3117: These two reasons apply to @emph{any} application whatever: as far as
3118: I can see, @code{#pragma} is never useful.
3119:
3120: @node Dollar Signs, Alignment, Function Attributes, Extensions
1.1 root 3121: @section Dollar Signs in Identifier Names
3122:
3123: In GNU C, you may use dollar signs in identifier names. This is because
3124: many traditional C implementations allow such identifiers.
3125:
1.1.1.9 root 3126: Dollar signs are allowed if you specify @samp{-traditional}; they are
3127: not allowed if you specify @samp{-ansi}. Whether they are allowed by
3128: default depends on the target machine; usually, they are not.
3129:
1.1 root 3130: @node Alignment, Inline, Dollar Signs, Extensions
3131: @section Inquiring about the Alignment of a Type or Variable
3132:
1.1.1.8 root 3133: The keyword @code{__alignof__} allows you to inquire about how an object
1.1 root 3134: is aligned, or the minimum alignment usually required by a type. Its
3135: syntax is just like @code{sizeof}.
3136:
3137: For example, if the target machine requires a @code{double} value to be
1.1.1.8 root 3138: aligned on an 8-byte boundary, then @code{__alignof__ (double)} is 8.
3139: This is true on many RISC machines. On more traditional machine
3140: designs, @code{__alignof__ (double)} is 4 or even 2.
1.1 root 3141:
3142: Some machines never actually require alignment; they allow reference to any
1.1.1.8 root 3143: data type even at an odd addresses. For these machines, @code{__alignof__}
1.1 root 3144: reports the @emph{recommended} alignment of a type.
3145:
1.1.1.8 root 3146: When the operand of @code{__alignof__} is an lvalue rather than a type, the
1.1 root 3147: value is the largest alignment that the lvalue is known to have. It may
3148: have this alignment as a result of its data type, or because it is part of
3149: a structure and inherits alignment from that structure. For example, after
3150: this declaration:
3151:
3152: @example
3153: struct foo @{ int x; char y; @} foo1;
3154: @end example
3155:
3156: @noindent
1.1.1.8 root 3157: the value of @code{__alignof__ (foo1.y)} is probably 2 or 4, the same as
3158: @code{__alignof__ (int)}, even though the data type of @code{foo1.y}
3159: does not itself demand any alignment.@refill
1.1 root 3160:
3161: @node Inline, Extended Asm, Alignment, Extensions
3162: @section An Inline Function is As Fast As a Macro
3163:
3164: By declaring a function @code{inline}, you can direct GNU CC to integrate
3165: that function's code into the code for its callers. This makes execution
3166: faster by eliminating the function-call overhead; in addition, if any of
3167: the actual argument values are constant, their known values may permit
3168: simplifications at compile time so that not all of the inline function's
3169: code needs to be included.
3170:
3171: To declare a function inline, use the @code{inline} keyword in its
3172: declaration, like this:
3173:
3174: @example
3175: inline int
3176: inc (int *a)
3177: @{
3178: (*a)++;
3179: @}
3180: @end example
3181:
1.1.1.7 root 3182: (If you are writing a header file to be included in ANSI C programs, write
1.1.1.8 root 3183: @code{__inline__} instead of @code{inline}. @xref{Alternate Keywords}.)
1.1.1.7 root 3184:
3185: You can also make all ``simple enough'' functions inline with the option
3186: @samp{-finline-functions}. Note that certain usages in a function
3187: definition can make it unsuitable for inline substitution.
1.1 root 3188:
3189: When a function is both inline and @code{static}, if all calls to the
1.1.1.8 root 3190: function are integrated into the caller, and the function's address is
3191: never used, then the function's own assembler code is never referenced.
3192: In this case, GNU CC does not actually output assembler code for the
3193: function, unless you specify the option @samp{-fkeep-inline-functions}.
3194: Some calls cannot be integrated for various reasons (in particular,
3195: calls that precede the function's definition cannot be integrated, and
3196: neither can recursive calls within the definition). If there is a
3197: nonintegrated call, then the function is compiled to assembler code as
3198: usual. The function must also be compiled as usual if the program
3199: refers to its address, because that can't be inlined.
1.1 root 3200:
3201: When an inline function is not @code{static}, then the compiler must assume
3202: that there may be calls from other source files; since a global symbol can
3203: be defined only once in any program, the function must not be defined in
3204: the other source files, so the calls therein cannot be integrated.
3205: Therefore, a non-@code{static} inline function is always compiled on its
3206: own in the usual fashion.
3207:
1.1.1.8 root 3208: If you specify both @code{inline} and @code{extern} in the function
3209: definition, then the definition is used only for inlining. In no case
3210: is the function compiled on its own, not even if you refer to its
3211: address explicitly. Such an address becomes an external reference, as
3212: if you had only declared the function, and had not defined it.
3213:
3214: This combination of @code{inline} and @code{extern} has almost the
3215: effect of a macro. The way to use it is to put a function definition in
3216: a header file with these keywords, and put another copy of the
3217: definition (lacking @code{inline} and @code{extern}) in a library file.
3218: The definition in the header file will cause most calls to the function
3219: to be inlined. If any uses of the function remain, they will refer to
3220: the single copy in the library.
3221:
1.1 root 3222: @node Extended Asm, Asm Labels, Inline, Extensions
3223: @section Assembler Instructions with C Expression Operands
3224:
3225: In an assembler instruction using @code{asm}, you can now specify the
3226: operands of the instruction using C expressions. This means no more
3227: guessing which registers or memory locations will contain the data you want
3228: to use.
3229:
3230: You must specify an assembler instruction template much like what appears
3231: in a machine description, plus an operand constraint string for each
3232: operand.
3233:
3234: For example, here is how to use the 68881's @code{fsinx} instruction:
3235:
3236: @example
3237: asm ("fsinx %1,%0" : "=f" (result) : "f" (angle));
3238: @end example
3239:
3240: @noindent
3241: Here @code{angle} is the C expression for the input operand while
3242: @code{result} is that of the output operand. Each has @samp{"f"} as its
3243: operand constraint, saying that a floating-point register is required. The
1.1.1.5 root 3244: @samp{=} in @samp{=f} indicates that the operand is an output; all output
1.1.1.4 root 3245: operands' constraints must use @samp{=}. The constraints use the same
3246: language used in the machine description (@pxref{Constraints}).
1.1 root 3247:
3248: Each operand is described by an operand-constraint string followed by the C
3249: expression in parentheses. A colon separates the assembler template from
3250: the first output operand, and another separates the last output operand
3251: from the first input, if any. Commas separate output operands and separate
1.1.1.4 root 3252: inputs. The total number of operands is limited to the maximum number of
1.1 root 3253: operands in any instruction pattern in the machine description.
3254:
1.1.1.4 root 3255: If there are no output operands, and there are input operands, then there
3256: must be two consecutive colons surrounding the place where the output
3257: operands would go.
3258:
1.1 root 3259: Output operand expressions must be lvalues; the compiler can check this.
3260: The input operands need not be lvalues. The compiler cannot check whether
3261: the operands have data types that are reasonable for the instruction being
3262: executed. It does not parse the assembler instruction template and does
3263: not know what it means, or whether it is valid assembler input. The
3264: extended @code{asm} feature is most often used for machine instructions
3265: that the compiler itself does not know exist.
3266:
3267: The output operands must be write-only; GNU CC will assume that the values
3268: in these operands before the instruction are dead and need not be
1.1.1.8 root 3269: generated. Extended asm does not support input-output or read-write
3270: operands. For this reason, the constraint character @samp{+}, which
3271: indicates such an operand, may not be used.
3272:
3273: When the assembler instruction has a read-write operand, or an operand
3274: in which only some of the bits are to be changed, you must logically
1.1 root 3275: split its function into two separate operands, one input operand and one
3276: write-only output operand. The connection between them is expressed by
3277: constraints which say they need to be in the same location when the
1.1.1.8 root 3278: instruction executes. You can use the same C expression for both
3279: operands, or different expressions. For example, here we write the
3280: (fictitious) @samp{combine} instruction with @code{bar} as its read-only
3281: source operand and @code{foo} as its read-write destination:
1.1 root 3282:
3283: @example
3284: asm ("combine %2,%0" : "=r" (foo) : "0" (foo), "g" (bar));
3285: @end example
3286:
3287: @noindent
3288: The constraint @samp{"0"} for operand 1 says that it must occupy the same
1.1.1.5 root 3289: location as operand 0. A digit in constraint is allowed only in an input
3290: operand, and it must refer to an output operand.
1.1 root 3291:
3292: Only a digit in the constraint can guarantee that one operand will be in
3293: the same place as another. The mere fact that @code{foo} is the value of
3294: both operands is not enough to guarantee that they will be in the same
3295: place in the generated assembler code. The following would not work:
3296:
3297: @example
3298: asm ("combine %2,%0" : "=r" (foo) : "r" (foo), "g" (bar));
3299: @end example
3300:
3301: Various optimizations or reloading could cause operands 0 and 1 to be in
3302: different registers; GNU CC knows no reason not to do so. For example, the
3303: compiler might find a copy of the value of @code{foo} in one register and
3304: use it for operand 1, but generate the output operand 0 in a different
3305: register (copying it afterward to @code{foo}'s own address). Of course,
3306: since the register for operand 1 is not even mentioned in the assembler
3307: code, the result will not work, but GNU CC can't tell that.
3308:
3309: Unless an output operand has the @samp{&} constraint modifier, GNU CC may
3310: allocate it in the same register as an unrelated input operand, on the
3311: assumption that the inputs are consumed before the outputs are produced.
3312: This assumption may be false if the assembler code actually consists of
3313: more than one instruction. In such a case, use @samp{&} for each output
3314: operand that may not overlap an input. @xref{Modifiers}.
3315:
1.1.1.4 root 3316: Some instructions clobber specific hard registers. To describe this, write
3317: a third colon after the input operands, followed by the names of the
3318: clobbered hard registers (given as strings). Here is a realistic example
3319: for the vax:
1.1 root 3320:
3321: @example
3322: asm volatile ("movc3 %0,%1,%2"
3323: : /* no outputs */
3324: : "g" (from), "g" (to), "g" (count)
3325: : "r0", "r1", "r2", "r3", "r4", "r5");
3326: @end example
3327:
1.1.1.4 root 3328: You can put multiple assembler instructions together in a single @code{asm}
1.1.1.7 root 3329: template, separated either with newlines (written as @samp{\n}) or with
3330: semicolons if the assembler allows such semicolons. The GNU assembler
3331: allows semicolons and all Unix assemblers seem to do so. The input
3332: operands are guaranteed not to use any of the clobbered registers, and
3333: neither will the output operands' addresses, so you can read and write the
3334: clobbered registers as many times as you like. Here is an example of
3335: multiple instructions in a template; it assumes that the subroutine
3336: @code{_foo} accepts arguments in registers 9 and 10:
1.1.1.4 root 3337:
3338: @example
3339: asm ("movl %0,r9;movl %1,r10;call _foo"
3340: : /* no outputs */
3341: : "g" (from), "g" (to)
3342: : "r9", "r10");
3343: @end example
3344:
1.1.1.7 root 3345: If you want to test the condition code produced by an assembler instruction,
3346: you must include a branch and a label in the @code{asm} construct, as follows:
3347:
3348: @example
3349: asm ("clr %0;frob %1;beq 0f;mov #1,%0;0:"
3350: : "g" (result)
3351: : "g" (input));
3352: @end example
3353:
3354: @noindent
3355: This assumes your assembler supports local labels, as the GNU assembler
3356: and most Unix assemblers do.
3357:
1.1 root 3358: Usually the most convenient way to use these @code{asm} instructions is to
3359: encapsulate them in macros that look like functions. For example,
3360:
3361: @example
3362: #define sin(x) \
3363: (@{ double __value, __arg = (x); \
3364: asm ("fsinx %1,%0": "=f" (__value): "f" (__arg)); \
3365: __value; @})
3366: @end example
3367:
3368: @noindent
3369: Here the variable @code{__arg} is used to make sure that the instruction
3370: operates on a proper @code{double} value, and to accept only those
3371: arguments @code{x} which can convert automatically to a @code{double}.
3372:
3373: Another way to make sure the instruction operates on the correct data type
3374: is to use a cast in the @code{asm}. This is different from using a
3375: variable @code{__arg} in that it converts more different types. For
3376: example, if the desired type were @code{int}, casting the argument to
3377: @code{int} would accept a pointer with no complaint, while assigning the
3378: argument to an @code{int} variable named @code{__arg} would warn about
3379: using a pointer unless the caller explicitly casts it.
3380:
1.1.1.4 root 3381: If an @code{asm} has output operands, GNU CC assumes for optimization
3382: purposes that the instruction has no side effects except to change the
3383: output operands. This does not mean that instructions with a side effect
3384: cannot be used, but you must be careful, because the compiler may eliminate
3385: them if the output operands aren't used, or move them out of loops, or
3386: replace two with one if they constitute a common subexpression. Also, if
3387: your instruction does have a side effect on a variable that otherwise
3388: appears not to change, the old value of the variable may be reused later if
3389: it happens to be found in a register.
1.1 root 3390:
3391: You can prevent an @code{asm} instruction from being deleted, moved or
3392: combined by writing the keyword @code{volatile} after the @code{asm}. For
3393: example:
3394:
3395: @example
3396: #define set_priority(x) \
3397: asm volatile ("set_priority %0": /* no outputs */ : "g" (x))
3398: @end example
3399:
1.1.1.7 root 3400: @noindent
3401: (However, an instruction without output operands will not be deleted
3402: or moved, regardless, unless it is unreachable.)
1.1.1.4 root 3403:
1.1 root 3404: It is a natural idea to look for a way to give access to the condition
3405: code left by the assembler instruction. However, when we attempted to
3406: implement this, we found no way to make it work reliably. The problem
3407: is that output operands might need reloading, which would result in
3408: additional following ``store'' instructions. On most machines, these
3409: instructions would alter the condition code before there was time to
3410: test it. This problem doesn't arise for ordinary ``test'' and
3411: ``compare'' instructions because they don't have any output operands.
3412:
1.1.1.7 root 3413: If you are writing a header file that should be includable in ANSI C
1.1.1.8 root 3414: programs, write @code{__asm__} instead of @code{asm}. @xref{Alternate
1.1.1.7 root 3415: Keywords}.
3416:
1.1.1.8 root 3417: @node Asm Labels, Explicit Reg Vars, Extended Asm, Extensions
1.1 root 3418: @section Controlling Names Used in Assembler Code
3419:
1.1.1.8 root 3420: You can specify the name to be used in the assembler code for a C
3421: function or variable by writing the @code{asm} (or @code{__asm__})
3422: keyword after the declarator as follows:
1.1 root 3423:
3424: @example
3425: int foo asm ("myfoo") = 2;
3426: @end example
3427:
3428: @noindent
3429: This specifies that the name to be used for the variable @code{foo} in
3430: the assembler code should be @samp{myfoo} rather than the usual
3431: @samp{_foo}.
3432:
3433: On systems where an underscore is normally prepended to the name of a C
3434: function or variable, this feature allows you to define names for the
3435: linker that do not start with an underscore.
3436:
3437: You cannot use @code{asm} in this way in a function @emph{definition}; but
3438: you can get the same effect by writing a declaration for the function
3439: before its definition and putting @code{asm} there, like this:
3440:
3441: @example
3442: extern func () asm ("FUNC");
3443:
3444: func (x, y)
3445: int x, y;
3446: @dots{}
3447: @end example
3448:
3449: It is up to you to make sure that the assembler names you choose do not
3450: conflict with any other assembler symbols. Also, you must not use a
3451: register name; that would produce completely invalid assembler code. GNU
3452: CC does not as yet have the ability to store static variables in registers.
3453: Perhaps that will be added.
3454:
1.1.1.8 root 3455: @node Explicit Reg Vars, Alternate Keywords, Asm Labels, Extensions
3456: @section Variables in Specified Registers
3457:
3458: GNU C allows you to put a few global variables into specified hardware
3459: registers. You can also specify the register in which an ordinary
3460: register variable should be allocated.
3461:
3462: @itemize @bullet
3463: @item
3464: Global register variables reserve registers throughout the program.
3465: This may be useful in programs such as programming language
3466: interpreters which have a couple of global variables that are accessed
3467: very often.
3468:
3469: @item
3470: Local register variables in specific registers do not reserve the
3471: registers. The compiler's data flow analysis is capable of
3472: determining where the specified registers contain live values, and
3473: where they are available for other uses. These local variables are
3474: sometimes convenient for use with the extended @code{asm} feature
3475: (@pxref{Extended Asm}).
3476: @end itemize
3477:
3478: @menu
3479: * Global Reg Vars::
3480: * Local Reg Vars::
3481: @end menu
1.1.1.5 root 3482:
1.1.1.8 root 3483: @node Global Reg Vars, Local Reg Vars, Explicit Reg Vars, Explicit Reg Vars
3484: @subsection Defining Global Register Variables
1.1.1.5 root 3485:
3486: You can define a global register variable in GNU C like this:
3487:
3488: @example
3489: register int *foo asm ("a5");
3490: @end example
3491:
3492: @noindent
3493: Here @code{a5} is the name of the register which should be used. Choose a
3494: register which is normally saved and restored by function calls on your
3495: machine, so that library routines will not clobber it.
3496:
3497: Naturally the register name is cpu-dependent, so you would need to
3498: conditionalize your program according to cpu type. The register
3499: @code{a5} would be a good choice on a 68000 for a variable of pointer
3500: type. On machines with register windows, be sure to choose a ``global''
1.1.1.8 root 3501: register that is not affected magically by the function call mechanism.
1.1.1.5 root 3502:
3503: In addition, operating systems on one type of cpu may differ in how they
3504: name the registers; then you would need additional conditionals. For
3505: example, some 68000 operating systems call this register @code{%a5}.
3506:
3507: Eventually there may be a way of asking the compiler to choose a register
3508: automatically, but first we need to figure out how it should choose and
1.1.1.6 root 3509: how to enable you to guide the choice. No solution is evident.
1.1.1.5 root 3510:
3511: Defining a global register variable in a certain register reserves that
3512: register entirely for this use, at least within the current compilation.
3513: The register will not be allocated for any other purpose in the functions
3514: in the current compilation. The register will not be saved and restored by
3515: these functions. Stores into this register are never deleted even if they
3516: would appear to be dead, but references may be deleted or moved or
3517: simplified.
3518:
3519: It is not safe to access the global register variables from signal
3520: handlers, or from more than one thread of control, because the system
3521: library routines may temporarily use the register for other things (unless
3522: you recompile them specially for the task at hand).
3523:
3524: It is not safe for one function that uses a global register variable to
3525: call another such function @code{foo} by way of a third function
3526: @code{lose} that was compiled without knowledge of this variable (i.e. in a
3527: different source file in which the variable wasn't declared). This is
3528: because @code{lose} might save the register and put some other value there.
3529: For example, you can't expect a global register variable to be available in
3530: the comparison-function that you pass to @code{qsort}, since @code{qsort}
3531: might have put something else in that register. (If you are prepared to
3532: recompile @code{qsort} with the same global register variable, you can
3533: solve this problem.)
3534:
3535: If you want to recompile @code{qsort} or other source files which do not
3536: actually use your global register variable, so that they will not use that
3537: register for any other purpose, then it suffices to specify the compiler
3538: option @samp{-ffixed-@var{reg}}. You need not actually add a global
3539: register declaration to their source code.
3540:
3541: A function which can alter the value of a global register variable cannot
3542: safely be called from a function compiled without this variable, because it
3543: could clobber the value the caller expects to find there on return.
3544: Therefore, the function which is the entry point into the part of the
3545: program that uses the global register variable must explicitly save and
3546: restore the value which belongs to its caller.
3547:
3548: On most machines, @code{longjmp} will restore to each global register
3549: variable the value it had at the time of the @code{setjmp}. On some
3550: machines, however, @code{longjmp} will not change the value of global
3551: register variables. To be portable, the function that called @code{setjmp}
3552: should make other arrangements to save the values of the global register
1.1.1.10 root 3553: variables, and to restore them in a @code{longjmp}. This way, the same
1.1.1.5 root 3554: thing will happen regardless of what @code{longjmp} does.
3555:
3556: All global register variable declarations must precede all function
3557: definitions. If such a declaration could appear after function
3558: definitions, the declaration would be too late to prevent the register from
3559: being used for other purposes in the preceding functions.
3560:
1.1.1.6 root 3561: Global register variables may not have initial values, because an
3562: executable file has no means to supply initial contents for a register.
3563:
1.1.1.9 root 3564: @node Local Reg Vars,, Global Reg Vars, Explicit Reg Vars
1.1.1.8 root 3565: @subsection Specifying Registers for Local Variables
3566:
3567: You can define a local register variable with a specified register
3568: like this:
3569:
3570: @example
3571: register int *foo asm ("a5");
3572: @end example
3573:
3574: @noindent
3575: Here @code{a5} is the name of the register which should be used. Note
3576: that this is the same syntax used for defining global register
3577: variables, but for a local variable it would appear within a function.
3578:
3579: Naturally the register name is cpu-dependent, but this is not a
3580: problem, since specific registers are most often useful with explicit
3581: assembler instructions (@pxref{Extended Asm}). Both of these things
3582: generally require that you conditionalize your program according to
3583: cpu type.
3584:
3585: In addition, operating systems on one type of cpu may differ in how they
3586: name the registers; then you would need additional conditionals. For
3587: example, some 68000 operating systems call this register @code{%a5}.
3588:
3589: Eventually there may be a way of asking the compiler to choose a register
3590: automatically, but first we need to figure out how it should choose and
3591: how to enable you to guide the choice. No solution is evident.
3592:
3593: Defining such a register variable does not reserve the register; it
1.1.1.10 root 3594: remains available for other uses in places where flow control determines
3595: the variable's value is not live. However, these registers are made
3596: unavailable for use in the reload pass. I would not be surprised if
3597: excessive use of this feature leaves the compiler too few available
1.1.1.8 root 3598: registers to compile certain functions.
3599:
3600: @node Alternate Keywords,, Explicit Reg Vars, Extensions
1.1.1.7 root 3601: @section Alternate Keywords
3602:
3603: The option @samp{-traditional} disables certain keywords; @samp{-ansi}
3604: disables certain others. This causes trouble when you want to use GNU C
3605: extensions, or ANSI C features, in a general-purpose header file that
3606: should be usable by all programs, including ANSI C programs and traditional
3607: ones. The keywords @code{asm}, @code{typeof} and @code{inline} cannot be
3608: used since they won't work in a program compiled with @samp{-ansi}, while
3609: the keywords @code{const}, @code{volatile}, @code{signed}, @code{typeof}
3610: and @code{inline} won't work in a program compiled with
3611: @samp{-traditional}.@refill
3612:
1.1.1.8 root 3613: The way to solve these problems is to put @samp{__} at the beginning and
3614: end of each problematical keyword. For example, use @code{__asm__}
3615: instead of @code{asm}, @code{__const__} instead of @code{const}, and
3616: @code{__inline__} instead of @code{inline}.
1.1.1.7 root 3617:
3618: Other C compilers won't accept these alternative keywords; if you want to
3619: compile with another compiler, you can define the alternate keywords as
3620: macros to replace them with the customary keywords. It looks like this:
3621:
3622: @example
3623: #ifndef __GNUC__
1.1.1.8 root 3624: #define __asm__ asm
1.1.1.7 root 3625: #endif
3626: @end example
3627:
1.1 root 3628: @node Bugs, Portability, Extensions, Top
3629: @chapter Reporting Bugs
3630:
3631: Your bug reports play an essential role in making GNU CC reliable.
3632:
1.1.1.11! root 3633: When you encounter a problem, the first thing to do is to see if it is
! 3634: already known. @xref{Trouble}. Also look in @ref{Incompatibilities}.
! 3635: If it isn't known, then you should report the problem.
! 3636:
! 3637: Reporting a bug may help you by bringing a solution to your problem, or
! 3638: it may not. (If it does not, look in the service directory; see
! 3639: @ref{Service}.) In any case, the principal function of a bug report
! 3640: is to help the entire community by making the next version of GNU CC
! 3641: work better. Bug reports are your contribution to the maintenance of
! 3642: GNU CC.
1.1 root 3643:
3644: In order for a bug report to serve its purpose, you must include the
3645: information that makes for fixing the bug.
3646:
3647: @menu
3648: * Criteria: Bug Criteria. Have you really found a bug?
3649: * Reporting: Bug Reporting. How to report a bug effectively.
3650: @end menu
3651:
3652: @node Bug Criteria, Bug Reporting, Bugs, Bugs
3653: @section Have You Found a Bug?
3654:
3655: If you are not sure whether you have found a bug, here are some guidelines:
3656:
3657: @itemize @bullet
3658: @item
3659: If the compiler gets a fatal signal, for any input whatever, that is a
3660: compiler bug. Reliable compilers never crash.
3661:
3662: @item
3663: If the compiler produces invalid assembly code, for any input whatever
3664: (except an @code{asm} statement), that is a compiler bug, unless the
3665: compiler reports errors (not just warnings) which would ordinarily
3666: prevent the assembler from being run.
3667:
3668: @item
3669: If the compiler produces valid assembly code that does not correctly
3670: execute the input source code, that is a compiler bug.
3671:
3672: However, you must double-check to make sure, because you may have run
3673: into an incompatibility between GNU C and traditional C
3674: (@pxref{Incompatibilities}). These incompatibilities might be considered
3675: bugs, but they are inescapable consequences of valuable features.
3676:
3677: Or you may have a program whose behavior is undefined, which happened
3678: by chance to give the desired results with another C compiler.
3679:
3680: For example, in many nonoptimizing compilers, you can write @samp{x;}
3681: at the end of a function instead of @samp{return x;}, with the same
1.1.1.8 root 3682: results. But the value of the function is undefined if @code{return}
1.1 root 3683: is omitted; it is not a bug when GNU CC produces different results.
3684:
3685: Problems often result from expressions with two increment operators,
1.1.1.8 root 3686: as in @code{f (*p++, *p++)}. Your previous compiler might have
1.1 root 3687: interpreted that expression the way you intended; GNU CC might
1.1.1.8 root 3688: interpret it another way. Neither compiler is wrong. The bug is
3689: in your code.
1.1 root 3690:
3691: After you have localized the error to a single source line, it should
3692: be easy to check for these things. If your program is correct and
3693: well defined, you have found a compiler bug.
3694:
3695: @item
3696: If the compiler produces an error message for valid input, that is a
3697: compiler bug.
3698:
3699: Note that the following is not valid input, and the error message for
3700: it is not a bug:
3701:
3702: @example
3703: int foo (char);
3704:
3705: int
3706: foo (x)
3707: char x;
3708: @{ @dots{} @}
3709: @end example
3710:
3711: @noindent
3712: The prototype says to pass a @code{char}, while the definition says to
3713: pass an @code{int} and treat the value as a @code{char}. This is what
3714: the ANSI standard says, and it makes sense.
3715:
3716: @item
3717: If the compiler does not produce an error message for invalid input,
3718: that is a compiler bug. However, you should note that your idea of
3719: ``invalid input'' might be my idea of ``an extension'' or ``support
3720: for traditional practice''.
3721:
3722: @item
3723: If you are an experienced user of C compilers, your suggestions
3724: for improvement of GNU CC are welcome in any case.
3725: @end itemize
3726:
3727: @node Bug Reporting,, Bug Criteria, Bugs
3728: @section How to Report Bugs
3729:
3730: Send bug reports for GNU C to one of these addresses:
3731:
3732: @example
3733: bug-gcc@@prep.ai.mit.edu
3734: @{ucbvax|mit-eddie|uunet@}!prep.ai.mit.edu!bug-gcc
3735: @end example
3736:
1.1.1.8 root 3737: @strong{Do not send bug reports to @samp{info-gcc}, or to the newsgroup
3738: @samp{gnu.gcc}.} Most users of GNU CC do not want to receive bug
3739: reports. Those that do, have asked to be on @samp{bug-gcc}.
3740:
3741: The mailing list @samp{bug-gcc} has a newsgroup which serves as a
3742: repeater. The mailing list and the newsgroup carry exactly the same
3743: messages. Often people think of posting bug reports to the newsgroup
3744: instead of mailing them. This appears to work, but it has one problem
3745: which can be crucial: a newsgroup posting does not contain a mail path
3746: back to the sender. Thus, if I need to ask for more information, I
3747: may be unable to reach you. For this reason, it is better to send bug
3748: reports to the mailing list.
3749:
3750: As a last resort, send bug reports on paper to:
1.1 root 3751:
3752: @example
3753: GNU Compiler Bugs
3754: 545 Tech Sq
3755: Cambridge, MA 02139
3756: @end example
3757:
3758: The fundamental principle of reporting bugs usefully is this:
1.1.1.8 root 3759: @strong{report all the facts}. If you are not sure whether to state a
3760: fact or leave it out, state it!
1.1 root 3761:
3762: Often people omit facts because they think they know what causes the
3763: problem and they conclude that some details don't matter. Thus, you might
3764: assume that the name of the variable you use in an example does not matter.
3765: Well, probably it doesn't, but one cannot be sure. Perhaps the bug is a
3766: stray memory reference which happens to fetch from the location where that
3767: name is stored in memory; perhaps, if the name were different, the contents
3768: of that location would fool the compiler into doing the right thing despite
1.1.1.8 root 3769: the bug. Play it safe and give a specific, complete example. That is the
3770: easiest thing for you to do, and the most helpful.
1.1 root 3771:
1.1.1.8 root 3772: Keep in mind that the purpose of a bug report is to enable me to fix
3773: the bug if it is not known. It isn't very important what happens if
3774: the bug is already known. Therefore, always write your bug reports on
3775: the assumption that the bug is not known.
3776:
3777: Sometimes people give a few sketchy facts and ask, ``Does this ring a
3778: bell?'' Those bug reports are useless, and I urge everyone to
3779: @emph{refuse to respond to them} except to chide the sender to report
3780: bugs properly.
3781:
3782: To enable me to fix the bug, you should include all these things:
1.1 root 3783:
3784: @itemize @bullet
3785: @item
3786: The version of GNU CC. You can get this by running it with the
3787: @samp{-v} option.
3788:
3789: Without this, I won't know whether there is any point in looking for
3790: the bug in the current version of GNU CC.
3791:
3792: @item
3793: A complete input file that will reproduce the bug. If the bug is in
3794: the C preprocessor, send me a source file and any header files that it
3795: requires. If the bug is in the compiler proper (@file{cc1}), run your
3796: source file through the C preprocessor by doing @samp{gcc -E
3797: @var{sourcefile} > @var{outfile}}, then include the contents of
3798: @var{outfile} in the bug report. (Any @samp{-I}, @samp{-D} or
3799: @samp{-U} options that you used in actual compilation should also be
3800: used when doing this.)
3801:
3802: A single statement is not enough of an example. In order to compile
3803: it, it must be embedded in a function definition; and the bug might
3804: depend on the details of how this is done.
3805:
3806: Without a real example I can compile, all I can do about your bug
3807: report is wish you luck. It would be futile to try to guess how to
3808: provoke the bug. For example, bugs in register allocation and
3809: reloading frequently depend on every little detail of the function
3810: they happen in.
3811:
3812: @item
3813: The command arguments you gave GNU CC to compile that example and
3814: observe the bug. For example, did you use @samp{-O}? To guarantee
3815: you won't omit something important, list them all.
3816:
3817: If I were to try to guess the arguments, I would probably guess wrong
3818: and then I would not encounter the bug.
3819:
3820: @item
3821: The names of the files that you used for @file{tm.h} and @file{md}
3822: when you installed the compiler.
3823:
3824: @item
3825: The type of machine you are using, and the operating system name and
3826: version number.
3827:
3828: @item
3829: A description of what behavior you observe that you believe is
3830: incorrect. For example, ``It gets a fatal signal,'' or, ``There is an
3831: incorrect assembler instruction in the output.''
3832:
3833: Of course, if the bug is that the compiler gets a fatal signal, then I
3834: will certainly notice it. But if the bug is incorrect output, I might
3835: not notice unless it is glaringly wrong. I won't study all the
3836: assembler code from a 50-line C program just on the off chance that it
3837: might be wrong.
3838:
3839: Even if the problem you experience is a fatal signal, you should still
3840: say so explicitly. Suppose something strange is going on, such as,
3841: your copy of the compiler is out of synch, or you have encountered a
3842: bug in the C library on your system. (This has happened!) Your copy
3843: might crash and mine would not. If you @i{told} me to expect a crash,
3844: then when mine fails to crash, I would know that the bug was not
3845: happening for me. If you had not told me to expect a crash, then I
3846: would not be able to draw any conclusion from my observations.
3847:
1.1.1.8 root 3848: Often the observed symptom is incorrect output when your program is run.
3849: Sad to say, this is not enough information for me unless the program is
3850: short and simple. If you send me a large program, I don't have time to
3851: figure out how it would work if compiled correctly, much less which line
3852: of it was compiled wrong. So you will have to do that. Tell me which
3853: source line it is, and what incorrect result happens when that line is
3854: executed. A person who understands the test program can find this as
3855: easily as a bug in the program itself.
1.1 root 3856:
3857: @item
3858: If you send me examples of output from GNU CC, please use @samp{-g}
3859: when you make them. The debugging information includes source line
3860: numbers which are essential for correlating the output with the input.
3861:
3862: @item
3863: If you wish to suggest changes to the GNU CC source, send me context
3864: diffs. If you even discuss something in the GNU CC source, refer to
3865: it by context, not by line number.
3866:
3867: The line numbers in my development sources don't match those in your
3868: sources. Your line numbers would convey no useful information to me.
3869:
3870: @item
3871: Additional information from a debugger might enable me to find
3872: a problem on a machine which I do not have available myself.
3873: However, you need to think when you collect this information if
3874: you want it to have any chance of being useful.
3875:
3876: For example, many people send just a backtrace, but that is never
3877: useful by itself. A simple backtrace with arguments conveys little
3878: about GNU CC because the compiler is largely data-driven; the same
3879: functions are called over and over for different RTL insns, doing
3880: different things depending on the details of the insn.
3881:
3882: Most of the arguments listed in the backtrace are useless because they
3883: are pointers to RTL list structure. The numeric values of the
3884: pointers, which the debugger prints in the backtrace, have no
3885: significance whatever; all that matters is the contents of the objects
3886: they point to (and most of the contents are other such pointers).
3887:
3888: In addition, most compiler passes consist of one or more loops that
3889: scan the RTL insn sequence. The most vital piece of information about
1.1.1.8 root 3890: such a loop---which insn it has reached---is usually in a local variable,
1.1 root 3891: not in an argument.
3892:
3893: What you need to provide in addition to a backtrace are the values of
3894: the local variables for several stack frames up. When a local
3895: variable or an argument is an RTX, first print its value and then use
3896: the GDB command @code{pr} to print the RTL expression that it points
3897: to. (If GDB doesn't run on your machine, use your debugger to call
3898: the function @code{debug_rtx} with the RTX as an argument.) In
3899: general, whenever a variable is a pointer, its value is no use
3900: without the data it points to.
3901:
3902: In addition, include a debugging dump from just before the pass
3903: in which the crash happens. Most bugs involve a series of insns,
3904: not just one.
3905: @end itemize
3906:
3907: Here are some things that are not necessary:
3908:
3909: @itemize @bullet
3910: @item
3911: A description of the envelope of the bug.
3912:
3913: Often people who encounter a bug spend a lot of time investigating
3914: which changes to the input file will make the bug go away and which
3915: changes will not affect it.
3916:
3917: This is often time consuming and not very useful, because the way I
3918: will find the bug is by running a single example under the debugger
3919: with breakpoints, not by pure deduction from a series of examples.
1.1.1.8 root 3920: I recommend that you save your time for something else.
1.1 root 3921:
3922: Of course, if you can find a simpler example to report @emph{instead}
3923: of the original one, that is a convenience for me. Errors in the
3924: output will be easier to spot, running under the debugger will take
3925: less time, etc. Most GNU CC bugs involve just one function, so the
3926: most straightforward way to simplify an example is to delete all the
3927: function definitions except the one where the bug occurs. Those
3928: earlier in the file may be replaced by external declarations if the
1.1.1.8 root 3929: crucial function depends on them. (Exception: inline functions may
3930: affect compilation of functions defined later in the file.)
1.1 root 3931:
3932: However, simplification is not vital; if you don't want to do this,
1.1.1.8 root 3933: report the bug anyway and send me the entire test case you used.
1.1 root 3934:
3935: @item
3936: A patch for the bug.
3937:
3938: A patch for the bug does help me if it is a good one. But don't omit
1.1.1.8 root 3939: the necessary information, such as the test case, on the assumption that
3940: a patch is all I need. I might see problems with your patch and decide
3941: to fix the problem another way, or I might not understand it at all.
1.1 root 3942:
3943: Sometimes with a program as complicated as GNU CC it is very hard to
3944: construct an example that will make the program follow a certain path
3945: through the code. If you don't send me the example, I won't be able
3946: to construct one, so I won't be able to verify that the bug is fixed.
3947:
1.1.1.8 root 3948: And if I can't understand what bug you are trying to fix, or why your
3949: patch should be an improvement, I won't install it. A test case will
3950: help me to understand.
3951:
1.1 root 3952: @item
3953: A guess about what the bug is or what it depends on.
3954:
3955: Such guesses are usually wrong. Even I can't guess right about such
1.1.1.8 root 3956: things without first using the debugger to find the facts.
1.1 root 3957: @end itemize
3958:
3959: @node Portability, Interface, Bugs, Top
3960: @chapter GNU CC and Portability
3961:
3962: The main goal of GNU CC was to make a good, fast compiler for machines in
3963: the class that the GNU system aims to run on: 32-bit machines that address
3964: 8-bit bytes and have several general registers. Elegance, theoretical
3965: power and simplicity are only secondary.
3966:
3967: GNU CC gets most of the information about the target machine from a machine
3968: description which gives an algebraic formula for each of the machine's
3969: instructions. This is a very clean way to describe the target. But when
3970: the compiler needs information that is difficult to express in this
3971: fashion, I have not hesitated to define an ad-hoc parameter to the machine
3972: description. The purpose of portability is to reduce the total work needed
3973: on the compiler; it was not of interest for its own sake.
3974:
3975: GNU CC does not contain machine dependent code, but it does contain code
3976: that depends on machine parameters such as endianness (whether the most
3977: significant byte has the highest or lowest address of the bytes in a word)
3978: and the availability of autoincrement addressing. In the RTL-generation
3979: pass, it is often necessary to have multiple strategies for generating code
3980: for a particular kind of syntax tree, strategies that are usable for different
3981: combinations of parameters. Often I have not tried to address all possible
3982: cases, but only the common ones or only the ones that I have encountered.
3983: As a result, a new target may require additional strategies. You will know
3984: if this happens because the compiler will call @code{abort}. Fortunately,
3985: the new strategies can be added in a machine-independent fashion, and will
3986: affect only the target machines that need them.
3987:
3988: @node Interface, Passes, Portability, Top
3989: @chapter Interfacing to GNU CC Output
3990:
3991: GNU CC is normally configured to use the same function calling convention
3992: normally in use on the target system. This is done with the
3993: machine-description macros described (@pxref{Machine Macros}).
3994:
3995: However, returning of structure and union values is done differently on
3996: some target machines. As a result, functions compiled with PCC
3997: returning such types cannot be called from code compiled with GNU CC,
3998: and vice versa. This does not cause trouble often because few Unix
3999: library routines return structures or unions.
4000:
4001: GNU CC code returns structures and unions that are 1, 2, 4 or 8 bytes
4002: long in the same registers used for @code{int} or @code{double} return
4003: values. (GNU CC typically allocates variables of such types in
4004: registers also.) Structures and unions of other sizes are returned by
4005: storing them into an address passed by the caller (usually in a
4006: register). The machine-description macros @code{STRUCT_VALUE} and
4007: @code{STRUCT_INCOMING_VALUE} tell GNU CC where to pass this address.
4008:
4009: By contrast, PCC on most target machines returns structures and unions
4010: of any size by copying the data into an area of static storage, and then
4011: returning the address of that storage as if it were a pointer value.
4012: The caller must copy the data from that memory area to the place where
4013: the value is wanted. This is slower than the method used by GNU CC, and
4014: fails to be reentrant.
4015:
4016: On some target machines, such as RISC machines and the 80386, the
4017: standard system convention is to pass to the subroutine the address of
4018: where to return the value. On these machines, GNU CC has been
4019: configured to be compatible with the standard compiler, when this method
4020: is used. It may not be compatible for structures of 1, 2, 4 or 8 bytes.
4021:
4022: GNU CC uses the system's standard convention for passing arguments. On
4023: some machines, the first few arguments are passed in registers; in
4024: others, all are passed on the stack. It would be possible to use
4025: registers for argument passing on any machine, and this would probably
4026: result in a significant speedup. But the result would be complete
4027: incompatibility with code that follows the standard convention. So this
4028: change is practical only if you are switching to GNU CC as the sole C
4029: compiler for the system. We may implement register argument passing on
4030: certain machines once we have a complete GNU system so that we can
4031: compile the libraries with GNU CC.
4032:
4033: If you use @code{longjmp}, beware of automatic variables. ANSI C says that
4034: automatic variables that are not declared @code{volatile} have undefined
4035: values after a @code{longjmp}. And this is all GNU CC promises to do,
4036: because it is very difficult to restore register variables correctly, and
4037: one of GNU CC's features is that it can put variables in registers without
4038: your asking it to.
4039:
4040: If you want a variable to be unaltered by @code{longjmp}, and you don't
4041: want to write @code{volatile} because old C compilers don't accept it,
4042: just take the address of the variable. If a variable's address is ever
4043: taken, even if just to compute it and ignore it, then the variable cannot
4044: go in a register:
4045:
4046: @example
4047: @{
4048: int careful;
4049: &careful;
4050: @dots{}
4051: @}
4052: @end example
4053:
4054: Code compiled with GNU CC may call certain library routines. Most of
4055: them handle arithmetic for which there are no instructions. This
4056: includes multiply and divide on some machines, and floating point
4057: operations on any machine for which floating point support is disabled
4058: with @samp{-msoft-float}. Some standard parts of the C library, such as
4059: @code{bcopy} or @code{memcpy}, are also called automatically. The usual
4060: function call interface is used for calling the library routines.
4061:
4062: These library routines should be defined in the library @file{gnulib},
4063: which GNU CC automatically searches whenever it links a program. On
4064: machines that have multiply and divide instructions, if hardware
4065: floating point is in use, normally @file{gnulib} is not needed, but it
4066: is searched just in case.
4067:
4068: Each arithmetic function is defined in @file{gnulib.c} to use the
4069: corresponding C arithmetic operator. As long as the file is compiled
4070: with another C compiler, which supports all the C arithmetic operators,
4071: this file will work portably. However, @file{gnulib.c} does not work if
4072: compiled with GNU CC, because each arithmetic function would compile
4073: into a call to itself!
4074:
4075: @node Passes, RTL, Interface, Top
4076: @chapter Passes and Files of the Compiler
4077:
4078: The overall control structure of the compiler is in @file{toplev.c}. This
4079: file is responsible for initialization, decoding arguments, opening and
4080: closing files, and sequencing the passes.
4081:
4082: The parsing pass is invoked only once, to parse the entire input. The RTL
4083: intermediate code for a function is generated as the function is parsed, a
4084: statement at a time. Each statement is read in as a syntax tree and then
4085: converted to RTL; then the storage for the tree for the statement is
4086: reclaimed. Storage for types (and the expressions for their sizes),
4087: declarations, and a representation of the binding contours and how they nest,
4088: remains until the function is finished being compiled; these are all needed
4089: to output the debugging information.
4090:
4091: Each time the parsing pass reads a complete function definition or
4092: top-level declaration, it calls the function
4093: @code{rest_of_compilation} or @code{rest_of_decl_compilation} in
4094: @file{toplev.c}, which are responsible for all further processing
4095: necessary, ending with output of the assembler language. All other
4096: compiler passes run, in sequence, within @code{rest_of_compilation}.
4097: When that function returns from compiling a function definition, the
4098: storage used for that function definition's compilation is entirely
4099: freed, unless it is an inline function (@pxref{Inline}).
4100:
4101: Here is a list of all the passes of the compiler and their source files.
4102: Also included is a description of where debugging dumps can be requested
4103: with @samp{-d} options.
4104:
4105: @itemize @bullet
4106: @item
4107: Parsing. This pass reads the entire text of a function definition,
4108: constructing partial syntax trees. This and RTL generation are no longer
4109: truly separate passes (formerly they were), but it is easier to think
4110: of them as separate.
4111:
4112: The tree representation does not entirely follow C syntax, because it is
4113: intended to support other languages as well.
4114:
4115: C data type analysis is also done in this pass, and every tree node
4116: that represents an expression has a data type attached. Variables are
4117: represented as declaration nodes.
4118:
4119: Constant folding and associative-law simplifications are also done
4120: during this pass.
4121:
4122: The source files for parsing are @file{c-parse.y}, @file{c-decl.c},
4123: @file{c-typeck.c}, @file{c-convert.c}, @file{stor-layout.c},
4124: @file{fold-const.c}, and @file{tree.c}. The last three files are
4125: intended to be language-independent. There are also header files
4126: @file{c-parse.h}, @file{c-tree.h}, @file{tree.h} and @file{tree.def}.
4127: The last two define the format of the tree representation.@refill
4128:
4129: @item
4130: RTL generation. This is the conversion of syntax tree into RTL code.
4131: It is actually done statement-by-statement during parsing, but for
4132: most purposes it can be thought of as a separate pass.
4133:
4134: This is where the bulk of target-parameter-dependent code is found,
4135: since often it is necessary for strategies to apply only when certain
4136: standard kinds of instructions are available. The purpose of named
4137: instruction patterns is to provide this information to the RTL
4138: generation pass.
4139:
4140: Optimization is done in this pass for @code{if}-conditions that are
4141: comparisons, boolean operations or conditional expressions. Tail
4142: recursion is detected at this time also. Decisions are made about how
4143: best to arrange loops and how to output @code{switch} statements.
4144:
4145: The source files for RTL generation are @file{stmt.c}, @file{expr.c},
4146: @file{explow.c}, @file{expmed.c}, @file{optabs.c} and @file{emit-rtl.c}.
4147: Also, the file @file{insn-emit.c}, generated from the machine description
4148: by the program @code{genemit}, is used in this pass. The header files
4149: @file{expr.h} is used for communication within this pass.@refill
4150:
4151: The header files @file{insn-flags.h} and @file{insn-codes.h},
4152: generated from the machine description by the programs @code{genflags}
4153: and @code{gencodes}, tell this pass which standard names are available
4154: for use and which patterns correspond to them.@refill
4155:
4156: Aside from debugging information output, none of the following passes
4157: refers to the tree structure representation of the function (only
4158: part of which is saved).
4159:
4160: The decision of whether the function can and should be expanded inline
4161: in its subsequent callers is made at the end of rtl generation. The
4162: function must meet certain criteria, currently related to the size of
4163: the function and the types and number of parameters it has. Note that
4164: this function may contain loops, recursive calls to itself
4165: (tail-recursive functions can be inlined!), gotos, in short, all
4166: constructs supported by GNU CC.
4167:
4168: The option @samp{-dr} causes a debugging dump of the RTL code after
4169: this pass. This dump file's name is made by appending @samp{.rtl} to
4170: the input file name.
4171:
4172: @item
4173: Jump optimization. This pass simplifies jumps to the following
4174: instruction, jumps across jumps, and jumps to jumps. It deletes
4175: unreferenced labels and unreachable code, except that unreachable code
4176: that contains a loop is not recognized as unreachable in this pass.
4177: (Such loops are deleted later in the basic block analysis.)
4178:
4179: Jump optimization is performed two or three times. The first time is
4180: immediately following RTL generation. The second time is after CSE,
4181: but only if CSE says repeated jump optimization is needed. The
4182: last time is right before the final pass. That time, cross-jumping
4183: and deletion of no-op move instructions are done together with the
4184: optimizations described above.
4185:
4186: The source file of this pass is @file{jump.c}.
4187:
4188: The option @samp{-dj} causes a debugging dump of the RTL code after
4189: this pass is run for the first time. This dump file's name is made by
4190: appending @samp{.jump} to the input file name.
4191:
4192: @item
4193: Register scan. This pass finds the first and last use of each
4194: register, as a guide for common subexpression elimination. Its source
4195: is in @file{regclass.c}.
4196:
4197: @item
4198: Common subexpression elimination. This pass also does constant
4199: propagation. Its source file is @file{cse.c}. If constant
4200: propagation causes conditional jumps to become unconditional or to
4201: become no-ops, jump optimization is run again when CSE is finished.
4202:
4203: The option @samp{-ds} causes a debugging dump of the RTL code after
4204: this pass. This dump file's name is made by appending @samp{.cse} to
4205: the input file name.
4206:
4207: @item
1.1.1.8 root 4208: Loop optimization. This pass moves constant expressions out of loops,
4209: and optionally does strength-reduction as well. Its source file is
4210: @file{loop.c}.
1.1 root 4211:
4212: The option @samp{-dL} causes a debugging dump of the RTL code after
4213: this pass. This dump file's name is made by appending @samp{.loop} to
4214: the input file name.
4215:
4216: @item
4217: Stupid register allocation is performed at this point in a
4218: nonoptimizing compilation. It does a little data flow analysis as
4219: well. When stupid register allocation is in use, the next pass
4220: executed is the reloading pass; the others in between are skipped.
4221: The source file is @file{stupid.c}.
4222:
4223: @item
4224: Data flow analysis (@file{flow.c}). This pass divides the program
4225: into basic blocks (and in the process deletes unreachable loops); then
4226: it computes which pseudo-registers are live at each point in the
4227: program, and makes the first instruction that uses a value point at
4228: the instruction that computed the value.
4229:
4230: This pass also deletes computations whose results are never used, and
4231: combines memory references with add or subtract instructions to make
4232: autoincrement or autodecrement addressing.
4233:
4234: The option @samp{-df} causes a debugging dump of the RTL code after
4235: this pass. This dump file's name is made by appending @samp{.flow} to
4236: the input file name. If stupid register allocation is in use, this
4237: dump file reflects the full results of such allocation.
4238:
4239: @item
4240: Instruction combination (@file{combine.c}). This pass attempts to
4241: combine groups of two or three instructions that are related by data
4242: flow into single instructions. It combines the RTL expressions for
4243: the instructions by substitution, simplifies the result using algebra,
4244: and then attempts to match the result against the machine description.
4245:
4246: The option @samp{-dc} causes a debugging dump of the RTL code after
4247: this pass. This dump file's name is made by appending @samp{.combine}
4248: to the input file name.
4249:
4250: @item
4251: Register class preferencing. The RTL code is scanned to find out
4252: which register class is best for each pseudo register. The source
4253: file is @file{regclass.c}.
4254:
4255: @item
4256: Local register allocation (@file{local-alloc.c}). This pass allocates
4257: hard registers to pseudo registers that are used only within one basic
4258: block. Because the basic block is linear, it can use fast and
4259: powerful techniques to do a very good job.
4260:
4261: The option @samp{-dl} causes a debugging dump of the RTL code after
4262: this pass. This dump file's name is made by appending @samp{.lreg} to
4263: the input file name.
4264:
4265: @item
4266: Global register allocation (@file{global-alloc.c}). This pass
4267: allocates hard registers for the remaining pseudo registers (those
4268: whose life spans are not contained in one basic block).
4269:
4270: @item
4271: Reloading. This pass renumbers pseudo registers with the hardware
4272: registers numbers they were allocated. Pseudo registers that did not
4273: get hard registers are replaced with stack slots. Then it finds
4274: instructions that are invalid because a value has failed to end up in
4275: a register, or has ended up in a register of the wrong kind. It fixes
4276: up these instructions by reloading the problematical values
4277: temporarily into registers. Additional instructions are generated to
4278: do the copying.
4279:
4280: Source files are @file{reload.c} and @file{reload1.c}, plus the header
4281: @file{reload.h} used for communication between them.
4282:
4283: The option @samp{-dg} causes a debugging dump of the RTL code after
4284: this pass. This dump file's name is made by appending @samp{.greg} to
4285: the input file name.
4286:
4287: @item
4288: Jump optimization is repeated, this time including cross-jumping
1.1.1.5 root 4289: and deletion of no-op move instructions.
1.1 root 4290:
4291: The option @samp{-dJ} causes a debugging dump of the RTL code after
4292: this pass. This dump file's name is made by appending @samp{.jump2}
4293: to the input file name.
4294:
4295: @item
1.1.1.8 root 4296: Delayed branch scheduling may be done at this point. The source file
4297: name is @file{dbranch.c}.
4298:
4299: The option @samp{-dd} causes a debugging dump of the RTL code after
4300: this pass. This dump file's name is made by appending @samp{.dbr}
4301: to the input file name.
4302:
4303: @item
1.1 root 4304: Final. This pass outputs the assembler code for the function. It is
4305: also responsible for identifying spurious test and compare
1.1.1.5 root 4306: instructions. Machine-specific peephole optimizations are performed
4307: at the same time. The function entry and exit sequences are generated
1.1 root 4308: directly as assembler code in this pass; they never exist as RTL.
4309:
4310: The source files are @file{final.c} plus @file{insn-output.c}; the
4311: latter is generated automatically from the machine description by the
4312: tool @file{genoutput}. The header file @file{conditions.h} is used
4313: for communication between these files.
4314:
4315: @item
4316: Debugging information output. This is run after final because it must
4317: output the stack slot offsets for pseudo registers that did not get
4318: hard registers. Source files are @file{dbxout.c} for DBX symbol table
4319: format and @file{symout.c} for GDB's own symbol table format.
4320: @end itemize
4321:
4322: Some additional files are used by all or many passes:
4323:
4324: @itemize @bullet
4325: @item
4326: Every pass uses @file{machmode.def}, which defines the machine modes.
4327:
4328: @item
4329: All the passes that work with RTL use the header files @file{rtl.h}
4330: and @file{rtl.def}, and subroutines in file @file{rtl.c}. The tools
4331: @code{gen*} also use these files to read and work with the machine
4332: description RTL.
4333:
4334: @item
4335: Several passes refer to the header file @file{insn-config.h} which
4336: contains a few parameters (C macro definitions) generated
4337: automatically from the machine description RTL by the tool
4338: @code{genconfig}.
4339:
4340: @item
4341: Several passes use the instruction recognizer, which consists of
4342: @file{recog.c} and @file{recog.h}, plus the files @file{insn-recog.c}
4343: and @file{insn-extract.c} that are generated automatically from the
4344: machine description by the tools @file{genrecog} and
4345: @file{genextract}.@refill
4346:
4347: @item
4348: Several passes use the header files @file{regs.h} which defines the
4349: information recorded about pseudo register usage, and @file{basic-block.h}
4350: which defines the information recorded about basic blocks.
4351:
4352: @item
4353: @file{hard-reg-set.h} defines the type @code{HARD_REG_SET}, a bit-vector
4354: with a bit for each hard register, and some macros to manipulate it.
4355: This type is just @code{int} if the machine has few enough hard registers;
4356: otherwise it is an array of @code{int} and some of the macros expand
4357: into loops.
4358: @end itemize
4359:
4360: @node RTL, Machine Desc, Passes, Top
4361: @chapter RTL Representation
4362:
4363: Most of the work of the compiler is done on an intermediate representation
4364: called register transfer language. In this language, the instructions to be
4365: output are described, pretty much one by one, in an algebraic form that
4366: describes what the instruction does.
4367:
4368: RTL is inspired by Lisp lists. It has both an internal form, made up of
4369: structures that point at other structures, and a textual form that is used
4370: in the machine description and in printed debugging dumps. The textual
4371: form uses nested parentheses to indicate the pointers in the internal form.
4372:
4373: @menu
4374: * RTL Objects:: Expressions vs vectors vs strings vs integers.
4375: * Accessors:: Macros to access expression operands or vector elts.
4376: * Flags:: Other flags in an RTL expression.
4377: * Machine Modes:: Describing the size and format of a datum.
4378: * Constants:: Expressions with constant values.
4379: * Regs and Memory:: Expressions representing register contents or memory.
4380: * Arithmetic:: Expressions representing arithmetic on other expressions.
4381: * Comparisons:: Expressions representing comparison of expressions.
4382: * Bit Fields:: Expressions representing bit-fields in memory or reg.
4383: * Conversions:: Extending, truncating, floating or fixing.
4384: * RTL Declarations:: Declaring volatility, constancy, etc.
4385: * Side Effects:: Expressions for storing in registers, etc.
4386: * Incdec:: Embedded side-effects for autoincrement addressing.
1.1.1.9 root 4387: * Assembler:: Representing @code{asm} with operands.
1.1 root 4388: * Insns:: Expression types for entire insns.
1.1.1.9 root 4389: * Calls:: RTL representation of function call insns.
1.1 root 4390: * Sharing:: Some expressions are unique; others *must* be copied.
4391: @end menu
4392:
4393: @node RTL Objects, Accessors, RTL, RTL
4394: @section RTL Object Types
4395:
4396: RTL uses four kinds of objects: expressions, integers, strings and vectors.
4397: Expressions are the most important ones. An RTL expression (``RTX'', for
4398: short) is a C structure, but it is usually referred to with a pointer; a
4399: type that is given the typedef name @code{rtx}.
4400:
4401: An integer is simply an @code{int}, and a string is a @code{char *}.
1.1.1.8 root 4402: Within RTL code, strings appear only inside @code{symbol_ref} expressions,
1.1 root 4403: but they appear in other contexts in the RTL expressions that make up
4404: machine descriptions. Their written form uses decimal digits.
4405:
4406: A string is a sequence of characters. In core it is represented as a
4407: @code{char *} in usual C fashion, and it is written in C syntax as well.
4408: However, strings in RTL may never be null. If you write an empty string in
4409: a machine description, it is represented in core as a null pointer rather
4410: than as a pointer to a null character. In certain contexts, these null
4411: pointers instead of strings are valid.
4412:
4413: A vector contains an arbitrary, specified number of pointers to
4414: expressions. The number of elements in the vector is explicitly present in
4415: the vector. The written form of a vector consists of square brackets
4416: (@samp{[@dots{}]}) surrounding the elements, in sequence and with
4417: whitespace separating them. Vectors of length zero are not created; null
4418: pointers are used instead.
4419:
4420: Expressions are classified by @dfn{expression codes} (also called RTX
4421: codes). The expression code is a name defined in @file{rtl.def}, which is
4422: also (in upper case) a C enumeration constant. The possible expression
4423: codes and their meanings are machine-independent. The code of an RTX can
4424: be extracted with the macro @code{GET_CODE (@var{x})} and altered with
4425: @code{PUT_CODE (@var{x}, @var{newcode})}.
4426:
4427: The expression code determines how many operands the expression contains,
4428: and what kinds of objects they are. In RTL, unlike Lisp, you cannot tell
4429: by looking at an operand what kind of object it is. Instead, you must know
4430: from its context---from the expression code of the containing expression.
1.1.1.8 root 4431: For example, in an expression of code @code{subreg}, the first operand is
1.1 root 4432: to be regarded as an expression and the second operand as an integer. In
1.1.1.8 root 4433: an expression of code @code{plus}, there are two operands, both of which
4434: are to be regarded as expressions. In a @code{symbol_ref} expression,
1.1 root 4435: there is one operand, which is to be regarded as a string.
4436:
4437: Expressions are written as parentheses containing the name of the
4438: expression type, its flags and machine mode if any, and then the operands
4439: of the expression (separated by spaces).
4440:
4441: Expression code names in the @samp{md} file are written in lower case,
4442: but when they appear in C code they are written in upper case. In this
1.1.1.8 root 4443: manual, they are shown as follows: @code{const_int}.
1.1 root 4444:
4445: In a few contexts a null pointer is valid where an expression is normally
1.1.1.4 root 4446: wanted. The written form of this is @code{(nil)}.
1.1 root 4447:
4448: @node Accessors, Flags, RTL Objects, RTL
4449: @section Access to Operands
4450:
4451: For each expression type @file{rtl.def} specifies the number of contained
4452: objects and their kinds, with four possibilities: @samp{e} for expression
4453: (actually a pointer to an expression), @samp{i} for integer, @samp{s} for
4454: string, and @samp{E} for vector of expressions. The sequence of letters
4455: for an expression code is called its @dfn{format}. Thus, the format of
1.1.1.8 root 4456: @code{subreg} is @samp{ei}.@refill
1.1 root 4457:
4458: Two other format characters are used occasionally: @samp{u} and @samp{0}.
4459: @samp{u} is equivalent to @samp{e} except that it is printed differently in
4460: debugging dumps, and @samp{0} means a slot whose contents do not fit any
4461: normal category. @samp{0} slots are not printed at all in dumps, and are
4462: often used in special ways by small parts of the compiler.@refill
4463:
4464: There are macros to get the number of operands and the format of an
4465: expression code:
4466:
4467: @table @code
4468: @item GET_RTX_LENGTH (@var{code})
4469: Number of operands of an RTX of code @var{code}.
4470:
4471: @item GET_RTX_FORMAT (@var{code})
4472: The format of an RTX of code @var{code}, as a C string.
4473: @end table
4474:
4475: Operands of expressions are accessed using the macros @code{XEXP},
4476: @code{XINT} and @code{XSTR}. Each of these macros takes two arguments: an
4477: expression-pointer (RTX) and an operand number (counting from zero).
4478: Thus,@refill
4479:
4480: @example
4481: XEXP (@var{x}, 2)
4482: @end example
4483:
4484: @noindent
4485: accesses operand 2 of expression @var{x}, as an expression.
4486:
4487: @example
4488: XINT (@var{x}, 2)
4489: @end example
4490:
4491: @noindent
4492: accesses the same operand as an integer. @code{XSTR}, used in the same
4493: fashion, would access it as a string.
4494:
4495: Any operand can be accessed as an integer, as an expression or as a string.
4496: You must choose the correct method of access for the kind of value actually
4497: stored in the operand. You would do this based on the expression code of
4498: the containing expression. That is also how you would know how many
4499: operands there are.
4500:
1.1.1.8 root 4501: For example, if @var{x} is a @code{subreg} expression, you know that it has
1.1 root 4502: two operands which can be correctly accessed as @code{XEXP (@var{x}, 0)}
4503: and @code{XINT (@var{x}, 1)}. If you did @code{XINT (@var{x}, 0)}, you
4504: would get the address of the expression operand but cast as an integer;
4505: that might occasionally be useful, but it would be cleaner to write
4506: @code{(int) XEXP (@var{x}, 0)}. @code{XEXP (@var{x}, 1)} would also
4507: compile without error, and would return the second, integer operand cast as
4508: an expression pointer, which would probably result in a crash when
4509: accessed. Nothing stops you from writing @code{XEXP (@var{x}, 28)} either,
4510: but this will access memory past the end of the expression with
4511: unpredictable results.@refill
4512:
4513: Access to operands which are vectors is more complicated. You can use the
4514: macro @code{XVEC} to get the vector-pointer itself, or the macros
4515: @code{XVECEXP} and @code{XVECLEN} to access the elements and length of a
4516: vector.
4517:
4518: @table @code
4519: @item XVEC (@var{exp}, @var{idx})
4520: Access the vector-pointer which is operand number @var{idx} in @var{exp}.
4521:
4522: @item XVECLEN (@var{exp}, @var{idx})
4523: Access the length (number of elements) in the vector which is
4524: in operand number @var{idx} in @var{exp}. This value is an @code{int}.
4525:
4526: @item XVECEXP (@var{exp}, @var{idx}, @var{eltnum})
4527: Access element number @var{eltnum} in the vector which is
4528: in operand number @var{idx} in @var{exp}. This value is an RTX.
4529:
4530: It is up to you to make sure that @var{eltnum} is not negative
4531: and is less than @code{XVECLEN (@var{exp}, @var{idx})}.
4532: @end table
4533:
4534: All the macros defined in this section expand into lvalues and therefore
4535: can be used to assign the operands, lengths and vector elements as well as
4536: to access them.
4537:
4538: @node Flags, Machine Modes, Accessors, RTL
4539: @section Flags in an RTL Expression
4540:
4541: RTL expressions contain several flags (one-bit bit-fields) that are used
4542: in certain types of expression. Most often they are accessed with the
4543: following macros:
4544:
4545: @table @code
1.1.1.10 root 4546: @item EXTERNAL_SYMBOL_P (@var{x})
4547: In a @code{symbol_ref} expression, nonzero if it corresponds to a variable
4548: declared extern in the users code. Zero for all other variables. Stored in
4549: the @code{volatil} field and printed as @samp{/v}.
4550:
1.1 root 4551: @item MEM_VOLATILE_P (@var{x})
1.1.1.8 root 4552: In @code{mem} expressions, nonzero for volatile memory references.
1.1 root 4553: Stored in the @code{volatil} field and printed as @samp{/v}.
4554:
4555: @item MEM_IN_STRUCT_P (@var{x})
1.1.1.8 root 4556: In @code{mem} expressions, nonzero for reference to an entire
1.1 root 4557: structure, union or array, or to a component of one. Zero for
4558: references to a scalar variable or through a pointer to a scalar.
4559: Stored in the @code{in_struct} field and printed as @samp{/s}.
4560:
4561: @item REG_USER_VAR_P (@var{x})
1.1.1.8 root 4562: In a @code{reg}, nonzero if it corresponds to a variable present in
1.1 root 4563: the user's source code. Zero for temporaries generated internally by
4564: the compiler. Stored in the @code{volatil} field and printed as
4565: @samp{/v}.
4566:
4567: @item REG_FUNCTION_VALUE_P (@var{x})
1.1.1.8 root 4568: Nonzero in a @code{reg} if it is the place in which this function's
1.1 root 4569: value is going to be returned. (This happens only in a hard
4570: register.) Stored in the @code{integrated} field and printed as
4571: @samp{/i}.
4572:
4573: The same hard register may be used also for collecting the values of
4574: functions called by this one, but @code{REG_FUNCTION_VALUE_P} is zero
4575: in this kind of use.
4576:
4577: @item RTX_UNCHANGING_P (@var{x})
1.1.1.8 root 4578: Nonzero in a @code{reg} or @code{mem} if the value is not changed
1.1 root 4579: explicitly by the current function. (If it is a memory reference then
4580: it may be changed by other functions or by aliasing.) Stored in the
4581: @code{unchanging} field and printed as @samp{/u}.
4582:
4583: @item RTX_INTEGRATED_P (@var{insn})
4584: Nonzero in an insn if it resulted from an in-line function call.
4585: Stored in the @code{integrated} field and printed as @samp{/i}. This
4586: may be deleted; nothing currently depends on it.
4587:
4588: @item INSN_DELETED_P (@var{insn})
4589: In an insn, nonzero if the insn has been deleted. Stored in the
4590: @code{volatil} field and printed as @samp{/v}.
4591:
4592: @item CONSTANT_POOL_ADDRESS_P (@var{x})
1.1.1.8 root 4593: Nonzero in a @code{symbol_ref} if it refers to part of the current
1.1 root 4594: function's ``constants pool''. These are addresses close to the
4595: beginning of the function, and GNU CC assumes they can be addressed
4596: directly (perhaps with the help of base registers). Stored in the
4597: @code{unchanging} field and printed as @samp{/u}.
4598: @end table
4599:
4600: These are the fields which the above macros refer to:
4601:
4602: @table @code
4603: @item used
4604: This flag is used only momentarily, at the end of RTL generation for a
4605: function, to count the number of times an expression appears in insns.
4606: Expressions that appear more than once are copied, according to the
4607: rules for shared structure (@pxref{Sharing}).
4608:
4609: @item volatil
1.1.1.10 root 4610: This flag is used in @code{mem},@code{symbol_ref} and @code{reg} expressions
4611: and in insns. In RTL dump files, it is printed as @samp{/v}.
1.1 root 4612:
1.1.1.8 root 4613: In a @code{mem} expression, it is 1 if the memory reference is volatile.
1.1 root 4614: Volatile memory references may not be deleted, reordered or combined.
4615:
1.1.1.8 root 4616: In a @code{reg} expression, it is 1 if the value is a user-level variable.
1.1 root 4617: 0 indicates an internal compiler temporary.
4618:
1.1.1.10 root 4619: In a @code{symbol_ref} expression, it is 1 if the symbol is declared
4620: @code{extern}.
4621:
1.1 root 4622: In an insn, 1 means the insn has been deleted.
4623:
4624: @item in_struct
1.1.1.8 root 4625: This flag is used in @code{mem} expressions. It is 1 if the memory
1.1 root 4626: datum referred to is all or part of a structure or array; 0 if it is (or
4627: might be) a scalar variable. A reference through a C pointer has 0
4628: because the pointer might point to a scalar variable.
4629:
4630: This information allows the compiler to determine something about possible
4631: cases of aliasing.
4632:
4633: In an RTL dump, this flag is represented as @samp{/s}.
4634:
4635: @item unchanging
1.1.1.8 root 4636: This flag is used in @code{reg} and @code{mem} expressions. 1 means
1.1 root 4637: that the value of the expression never changes (at least within the
4638: current function).
4639:
4640: In an RTL dump, this flag is represented as @samp{/u}.
4641:
4642: @item integrated
4643: In some kinds of expressions, including insns, this flag means the
4644: rtl was produced by procedure integration.
4645:
1.1.1.8 root 4646: In a @code{reg} expression, this flag indicates the register
1.1 root 4647: containing the value to be returned by the current function. On
4648: machines that pass parameters in registers, the same register number
4649: may be used for parameters as well, but this flag is not set on such
4650: uses.
4651: @end table
4652:
4653: @node Machine Modes, Constants, Flags, RTL
4654: @section Machine Modes
4655:
4656: A machine mode describes a size of data object and the representation used
4657: for it. In the C code, machine modes are represented by an enumeration
4658: type, @code{enum machine_mode}, defined in @file{machmode.def}. Each RTL
4659: expression has room for a machine mode and so do certain kinds of tree
4660: expressions (declarations and types, to be precise).
4661:
4662: In debugging dumps and machine descriptions, the machine mode of an RTL
4663: expression is written after the expression code with a colon to separate
4664: them. The letters @samp{mode} which appear at the end of each machine mode
1.1.1.8 root 4665: name are omitted. For example, @code{(reg:SI 38)} is a @code{reg}
1.1 root 4666: expression with machine mode @code{SImode}. If the mode is
4667: @code{VOIDmode}, it is not written at all.
4668:
4669: Here is a table of machine modes.
4670:
4671: @table @code
4672: @item QImode
4673: ``Quarter-Integer'' mode represents a single byte treated as an integer.
4674:
4675: @item HImode
4676: ``Half-Integer'' mode represents a two-byte integer.
4677:
1.1.1.7 root 4678: @item PSImode
4679: ``Partial Single Integer'' mode represents an integer which occupies
4680: four bytes but which doesn't really use all four. On some machines,
4681: this is the right mode to use for pointers.
4682:
1.1 root 4683: @item SImode
4684: ``Single Integer'' mode represents a four-byte integer.
4685:
1.1.1.7 root 4686: @item PDImode
4687: ``Partial Double Integer'' mode represents an integer which occupies
4688: eight bytes but which doesn't really use all eight. On some machines,
4689: this is the right mode to use for certain pointers.
4690:
1.1 root 4691: @item DImode
4692: ``Double Integer'' mode represents an eight-byte integer.
4693:
4694: @item TImode
4695: ``Tetra Integer'' (?) mode represents a sixteen-byte integer.
4696:
4697: @item SFmode
4698: ``Single Floating'' mode represents a single-precision (four byte) floating
4699: point number.
4700:
4701: @item DFmode
4702: ``Double Floating'' mode represents a double-precision (eight byte) floating
4703: point number.
4704:
1.1.1.7 root 4705: @item XFmode
4706: ``Extended Floating'' mode represents a triple-precision (twelve byte)
4707: floating point number. This mode is used for IEEE extended floating
4708: point.
4709:
1.1 root 4710: @item TFmode
4711: ``Tetra Floating'' mode represents a quadruple-precision (sixteen byte)
4712: floating point number.
4713:
4714: @item BLKmode
4715: ``Block'' mode represents values that are aggregates to which none of
4716: the other modes apply. In RTL, only memory references can have this mode,
4717: and only if they appear in string-move or vector instructions. On machines
4718: which have no such instructions, @code{BLKmode} will not appear in RTL.
4719:
4720: @item VOIDmode
4721: Void mode means the absence of a mode or an unspecified mode.
1.1.1.8 root 4722: For example, RTL expressions of code @code{const_int} have mode
1.1 root 4723: @code{VOIDmode} because they can be taken to have whatever mode the context
4724: requires. In debugging dumps of RTL, @code{VOIDmode} is expressed by
4725: the absence of any mode.
4726:
4727: @item EPmode
4728: ``Entry Pointer'' mode is intended to be used for function variables in
4729: Pascal and other block structured languages. Such values contain
4730: both a function address and a static chain pointer for access to
4731: automatic variables of outer levels. This mode is only partially
4732: implemented since C does not use it.
4733:
4734: @item CSImode@r{, @dots{}}
4735: ``Complex Single Integer'' mode stands for a complex number represented
4736: as a pair of @code{SImode} integers. Any of the integer and floating modes
4737: may have @samp{C} prefixed to its name to obtain a complex number mode.
4738: For example, there are @code{CQImode}, @code{CSFmode}, and @code{CDFmode}.
4739: Since C does not support complex numbers, these machine modes are only
4740: partially implemented.
4741:
4742: @item BImode
4743: This is the machine mode of a bit-field in a structure. It is used
4744: only in the syntax tree, never in RTL, and in the syntax tree it appears
4745: only in declaration nodes. In C, it appears only in @code{FIELD_DECL}
4746: nodes for structure fields defined with a bit size.
4747: @end table
4748:
4749: The machine description defines @code{Pmode} as a C macro which expands
4750: into the machine mode used for addresses. Normally this is @code{SImode}.
4751:
4752: The only modes which a machine description @i{must} support are
4753: @code{QImode}, @code{SImode}, @code{SFmode} and @code{DFmode}. The
4754: compiler will attempt to use @code{DImode} for two-word structures and
1.1.1.7 root 4755: unions, but this can be prevented by overriding the definition of
4756: @code{MAX_FIXED_MODE_SIZE}. Likewise, you can arrange for the C type
4757: @code{short int} to avoid using @code{HImode}. In the long term it
4758: might be desirable to make the set of available machine modes
4759: machine-dependent and eliminate all assumptions about specific machine
4760: modes or their uses from the machine-independent code of the compiler.
1.1 root 4761:
1.1.1.4 root 4762: To help begin this process, the machine modes are divided into mode
4763: classes. These are represented by the enumeration type @code{enum
4764: mode_class} defined in @file{rtl.h}. The possible mode classes are:
4765:
4766: @table @code
4767: @item MODE_INT
4768: Integer modes. By default these are @code{QImode}, @code{HImode},
4769: @code{SImode}, @code{DImode}, @code{TImode}, and also @code{BImode}.
4770:
4771: @item MODE_FLOAT
4772: Floating-point modes. By default these are @code{QFmode},
4773: @code{HFmode}, @code{SFmode}, @code{DFmode} and @code{TFmode}, but the
4774: MC68881 also defines @code{XFmode} to be an 80-bit extended-precision
4775: floating-point mode.
4776:
4777: @item MODE_COMPLEX_INT
4778: Complex integer modes. By default these are @code{CQImode},
4779: @code{CHImode}, @code{CSImode}, @code{CDImode} and @code{CTImode}.
4780:
4781: @item MODE_COMPLEX_FLOAT
4782: Complex floating-point modes. By default these are @code{CQFmode},
4783: @code{CHFmode}, @code{CSFmode}, @code{CDFmode} and @code{CTFmode},
4784:
4785: @item MODE_FUNCTION
4786: Algol or Pascal function variables including a static chain.
4787: (These are not currently implemented).
4788:
4789: @item MODE_RANDOM
4790: This is a catchall mode class for modes which don't fit into the above
4791: classes. Currently @code{VOIDmode}, @code{BLKmode} and @code{EPmode}
4792: are in @code{MODE_RANDOM}.
4793: @end table
4794:
1.1 root 4795: Here are some C macros that relate to machine modes:
4796:
4797: @table @code
4798: @item GET_MODE (@var{x})
4799: Returns the machine mode of the RTX @var{x}.
4800:
4801: @item PUT_MODE (@var{x}, @var{newmode})
4802: Alters the machine mode of the RTX @var{x} to be @var{newmode}.
4803:
1.1.1.4 root 4804: @item NUM_MACHINE_MODES
4805: Stands for the number of machine modes available on the target
4806: machine. This is one greater than the largest numeric value of any
4807: machine mode.
4808:
4809: @item GET_MODE_NAME (@var{m})
4810: Returns the name of mode @var{m} as a string.
4811:
4812: @item GET_MODE_CLASS (@var{m})
4813: Returns the mode class of mode @var{m}.
4814:
1.1 root 4815: @item GET_MODE_SIZE (@var{m})
4816: Returns the size in bytes of a datum of mode @var{m}.
4817:
4818: @item GET_MODE_BITSIZE (@var{m})
4819: Returns the size in bits of a datum of mode @var{m}.
4820:
4821: @item GET_MODE_UNIT_SIZE (@var{m})
4822: Returns the size in bits of the subunits of a datum of mode @var{m}.
4823: This is the same as @code{GET_MODE_SIZE} except in the case of
4824: complex modes and @code{EPmode}. For them, the unit size is the
4825: size of the real or imaginary part, or the size of the function
4826: pointer or the context pointer.
4827: @end table
4828:
4829: @node Constants, Regs and Memory, Machine Modes, RTL
4830: @section Constant Expression Types
4831:
4832: The simplest RTL expressions are those that represent constant values.
4833:
4834: @table @code
4835: @item (const_int @var{i})
4836: This type of expression represents the integer value @var{i}. @var{i}
4837: is customarily accessed with the macro @code{INTVAL} as in
4838: @code{INTVAL (@var{exp})}, which is equivalent to @code{XINT (@var{exp}, 0)}.
4839:
4840: There is only one expression object for the integer value zero;
4841: it is the value of the variable @code{const0_rtx}. Likewise, the
4842: only expression for integer value one is found in @code{const1_rtx}.
1.1.1.8 root 4843: Any attempt to create an expression of code @code{const_int} and
1.1 root 4844: value zero or one will return @code{const0_rtx} or @code{const1_rtx}
4845: as appropriate.
4846:
4847: @item (const_double:@var{m} @var{i0} @var{i1})
1.1.1.6 root 4848: Represents a 64-bit constant of mode @var{m}. All floating point
1.1 root 4849: constants are represented in this way, and so are 64-bit @code{DImode}
4850: integer constants.
4851:
4852: The two integers @var{i0} and @var{i1} together contain the bits of
4853: the value. If the constant is floating point (either single or double
4854: precision), then they represent a @code{double}. To convert them to a
4855: @code{double}, do
4856:
4857: @example
4858: union @{ double d; int i[2];@} u;
1.1.1.8 root 4859: u.i[0] = CONST_DOUBLE_LOW(x);
4860: u.i[1] = CONST_DOUBLE_HIGH(x);
1.1 root 4861: @end example
4862:
4863: @noindent
4864: and then refer to @code{u.d}.
4865:
4866: The global variables @code{dconst0_rtx} and @code{fconst0_rtx} hold
1.1.1.8 root 4867: @code{const_double} expressions with value 0, in modes @code{DFmode}
1.1.1.7 root 4868: and @code{SFmode}, respectively. The macro @code{CONST0_RTX
1.1.1.8 root 4869: (@var{mode})} refers to a @code{const_double} expression with value 0
1.1.1.7 root 4870: in mode @var{mode}. The mode @var{mode} must be of mode class
4871: @code{MODE_FLOAT}.
1.1 root 4872:
4873: @item (symbol_ref @var{symbol})
4874: Represents the value of an assembler label for data. @var{symbol} is
4875: a string that describes the name of the assembler label. If it starts
4876: with a @samp{*}, the label is the rest of @var{symbol} not including
4877: the @samp{*}. Otherwise, the label is @var{symbol}, prefixed with
4878: @samp{_}.
4879:
4880: @item (label_ref @var{label})
4881: Represents the value of an assembler label for code. It contains one
1.1.1.8 root 4882: operand, an expression, which must be a @code{code_label} that appears
1.1 root 4883: in the instruction sequence to identify the place where the label
4884: should go.
4885:
4886: The reason for using a distinct expression type for code label
4887: references is so that jump optimization can distinguish them.
4888:
4889: @item (const @var{exp})
4890: Represents a constant that is the result of an assembly-time
4891: arithmetic computation. The operand, @var{exp}, is an expression that
1.1.1.8 root 4892: contains only constants (@code{const_int}, @code{symbol_ref} and
4893: @code{label_ref} expressions) combined with @code{plus} and
4894: @code{minus}. However, not all combinations are valid, since the
1.1 root 4895: assembler cannot do arbitrary arithmetic on relocatable symbols.
4896: @end table
4897:
4898: @node Regs and Memory, Arithmetic, Constants, RTL
4899: @section Registers and Memory
4900:
4901: Here are the RTL expression types for describing access to machine
4902: registers and to main memory.
4903:
4904: @table @code
4905: @item (reg:@var{m} @var{n})
4906: For small values of the integer @var{n} (less than
4907: @code{FIRST_PSEUDO_REGISTER}), this stands for a reference to machine
4908: register number @var{n}: a @dfn{hard register}. For larger values of
4909: @var{n}, it stands for a temporary value or @dfn{pseudo register}.
4910: The compiler's strategy is to generate code assuming an unlimited
4911: number of such pseudo registers, and later convert them into hard
4912: registers or into memory references.
4913:
4914: The symbol @code{FIRST_PSEUDO_REGISTER} is defined by the machine
4915: description, since the number of hard registers on the machine is an
4916: invariant characteristic of the machine. Note, however, that not
4917: all of the machine registers must be general registers. All the
4918: machine registers that can be used for storage of data are given
4919: hard register numbers, even those that can be used only in certain
4920: instructions or can hold only certain types of data.
4921:
4922: Each pseudo register number used in a function's RTL code is
1.1.1.8 root 4923: represented by a unique @code{reg} expression.
1.1 root 4924:
4925: @var{m} is the machine mode of the reference. It is necessary because
4926: machines can generally refer to each register in more than one mode.
4927: For example, a register may contain a full word but there may be
4928: instructions to refer to it as a half word or as a single byte, as
4929: well as instructions to refer to it as a floating point number of
4930: various precisions.
4931:
4932: Even for a register that the machine can access in only one mode,
4933: the mode must always be specified.
4934:
4935: A hard register may be accessed in various modes throughout one
4936: function, but each pseudo register is given a natural mode
4937: and is accessed only in that mode. When it is necessary to describe
1.1.1.8 root 4938: an access to a pseudo register using a nonnatural mode, a @code{subreg}
1.1 root 4939: expression is used.
4940:
1.1.1.8 root 4941: A @code{reg} expression with a machine mode that specifies more than
1.1 root 4942: one word of data may actually stand for several consecutive registers.
4943: If in addition the register number specifies a hardware register, then
4944: it actually represents several consecutive hardware registers starting
4945: with the specified one.
4946:
1.1.1.8 root 4947: Such multi-word hardware register @code{reg} expressions must not be live
1.1 root 4948: across the boundary of a basic block. The lifetime analysis pass does not
4949: know how to record properly that several consecutive registers are
4950: actually live there, and therefore register allocation would be confused.
4951: The CSE pass must go out of its way to make sure the situation does
4952: not arise.
4953:
4954: @item (subreg:@var{m} @var{reg} @var{wordnum})
1.1.1.8 root 4955: @code{subreg} expressions are used to refer to a register in a machine
1.1 root 4956: mode other than its natural one, or to refer to one register of
1.1.1.8 root 4957: a multi-word @code{reg} that actually refers to several registers.
1.1 root 4958:
4959: Each pseudo-register has a natural mode. If it is necessary to
4960: operate on it in a different mode---for example, to perform a fullword
1.1.1.8 root 4961: move instruction on a pseudo-register that contains a single
4962: byte---the pseudo-register must be enclosed in a @code{subreg}. In
4963: such a case, @var{wordnum} is zero.
1.1 root 4964:
1.1.1.8 root 4965: The other use of @code{subreg} is to extract the individual registers
1.1 root 4966: of a multi-register value. Machine modes such as @code{DImode} and
4967: @code{EPmode} indicate values longer than a word, values which usually
4968: require two consecutive registers. To access one of the registers,
1.1.1.8 root 4969: use a @code{subreg} with mode @code{SImode} and a @var{wordnum} that
1.1 root 4970: says which register.
4971:
4972: The compilation parameter @code{WORDS_BIG_ENDIAN}, if defined, says
4973: that word number zero is the most significant part; otherwise, it is
4974: the least significant part.
4975:
4976: Between the combiner pass and the reload pass, it is possible to have
1.1.1.8 root 4977: a @code{subreg} which contains a @code{mem} instead of a @code{reg} as
1.1 root 4978: its first operand. The reload pass eliminates these cases by
1.1.1.8 root 4979: reloading the @code{mem} into a suitable register.
1.1 root 4980:
4981: Note that it is not valid to access a @code{DFmode} value in @code{SFmode}
1.1.1.8 root 4982: using a @code{subreg}. On some machines the most significant part of a
1.1 root 4983: @code{DFmode} value does not have the same format as a single-precision
4984: floating value.
4985:
4986: @item (cc0)
4987: This refers to the machine's condition code register. It has no
1.1.1.10 root 4988: operands and may not have a machine mode. There are two ways to use it:
4989:
4990: @itemize @bullet
4991: @item
4992: To stand for a complete set of condition code flags. This is best on
4993: most machines, where each comparison sets the entire series of flags.
4994:
4995: With this technique, @code{(cc0)} may be validly used in only two
4996: contexts: as the destination of an assignment (in test and compare
4997: instructions) and in comparison operators comparing against zero
4998: (@code{const_int} with value zero; that is to say, @code{const0_rtx}).
4999:
5000: @item
5001: To stand for a single flag that is the result of a single condition.
5002: This is useful on machines that have only a single flag bit, and in
5003: which comparison instructions must specify the condition to test.
5004:
5005: With this technique, @code{(cc0)} may be validly used in only two
5006: contexts: as the destination of an assignment (in test and compare
5007: instructions) where the source is a comparison operator, and as the
5008: first operand of @code{if_then_else} (in a conditional branch).
5009: @end itemize
1.1 root 5010:
1.1.1.8 root 5011: There is only one expression object of code @code{cc0}; it is the
1.1 root 5012: value of the variable @code{cc0_rtx}. Any attempt to create an
1.1.1.8 root 5013: expression of code @code{cc0} will return @code{cc0_rtx}.
1.1 root 5014:
5015: One special thing about the condition code register is that
5016: instructions can set it implicitly. On many machines, nearly all
5017: instructions set the condition code based on the value that they
5018: compute or store. It is not necessary to record these actions
5019: explicitly in the RTL because the machine description includes a
5020: prescription for recognizing the instructions that do so (by means of
5021: the macro @code{NOTICE_UPDATE_CC}). Only instructions whose sole
5022: purpose is to set the condition code, and instructions that use the
5023: condition code, need mention @code{(cc0)}.
5024:
1.1.1.10 root 5025: In some cases, better code may result from recognizing combinations or
5026: peepholes that include instructions that set the condition codes, even
5027: in cases where some reloading is inevitable. For examples, search for
5028: @samp{addcc} and @samp{andcc} in @file{sparc.md}.
5029:
1.1 root 5030: @item (pc)
5031: This represents the machine's program counter. It has no operands and
5032: may not have a machine mode. @code{(pc)} may be validly used only in
5033: certain specific contexts in jump instructions.
5034:
1.1.1.8 root 5035: There is only one expression object of code @code{pc}; it is the value
1.1 root 5036: of the variable @code{pc_rtx}. Any attempt to create an expression of
1.1.1.8 root 5037: code @code{pc} will return @code{pc_rtx}.
1.1 root 5038:
5039: All instructions that do not jump alter the program counter implicitly
5040: by incrementing it, but there is no need to mention this in the RTL.
5041:
5042: @item (mem:@var{m} @var{addr})
5043: This RTX represents a reference to main memory at an address
5044: represented by the expression @var{addr}. @var{m} specifies how large
5045: a unit of memory is accessed.
5046: @end table
5047:
5048: @node Arithmetic, Comparisons, Regs and Memory, RTL
5049: @section RTL Expressions for Arithmetic
5050:
5051: @table @code
5052: @item (plus:@var{m} @var{x} @var{y})
5053: Represents the sum of the values represented by @var{x} and @var{y}
5054: carried out in machine mode @var{m}. This is valid only if
5055: @var{x} and @var{y} both are valid for mode @var{m}.
5056:
5057: @item (minus:@var{m} @var{x} @var{y})
1.1.1.8 root 5058: Like @code{plus} but represents subtraction.
1.1 root 5059:
1.1.1.6 root 5060: @item (compare @var{x} @var{y})
1.1 root 5061: Represents the result of subtracting @var{y} from @var{x}
5062: for purposes of comparison. The absence of a machine mode
1.1.1.8 root 5063: in the @code{compare} expression indicates that the result is
1.1 root 5064: computed without overflow, as if with infinite precision.
5065:
5066: Of course, machines can't really subtract with infinite precision.
5067: However, they can pretend to do so when only the sign of the
5068: result will be used, which is the case when the result is stored
5069: in @code{(cc0)}. And that is the only way this kind of expression
5070: may validly be used: as a value to be stored in the condition codes.
5071:
5072: @item (neg:@var{m} @var{x})
5073: Represents the negation (subtraction from zero) of the value
5074: represented by @var{x}, carried out in mode @var{m}. @var{x} must be
5075: valid for mode @var{m}.
5076:
5077: @item (mult:@var{m} @var{x} @var{y})
5078: Represents the signed product of the values represented by @var{x} and
5079: @var{y} carried out in machine mode @var{m}. If
5080: @var{x} and @var{y} are both valid for mode @var{m}, this is ordinary
5081: size-preserving multiplication. Alternatively, both @var{x} and @var{y}
5082: may be valid for a different, narrower mode. This represents the
5083: kind of multiplication that generates a product wider than the operands.
5084: Widening multiplication and same-size multiplication are completely
5085: distinct and supported by different machine instructions; machines may
5086: support one but not the other.@refill
5087:
1.1.1.8 root 5088: @code{mult} may be used for floating point multiplication as well.
1.1 root 5089: Then @var{m} is a floating point machine mode.
5090:
5091: @item (umult:@var{m} @var{x} @var{y})
1.1.1.8 root 5092: Like @code{mult} but represents unsigned multiplication. It may be
5093: used in both same-size and widening forms, like @code{mult}.
5094: @code{umult} is used only for fixed-point multiplication.
1.1 root 5095:
5096: @item (div:@var{m} @var{x} @var{y})
5097: Represents the quotient in signed division of @var{x} by @var{y},
5098: carried out in machine mode @var{m}. If @var{m} is a floating-point
5099: mode, it represents the exact quotient; otherwise, the integerized
5100: quotient. If @var{x} and @var{y} are both valid for mode @var{m},
5101: this is ordinary size-preserving division. Some machines have
5102: division instructions in which the operands and quotient widths are
1.1.1.8 root 5103: not all the same; such instructions are represented by @code{div}
1.1 root 5104: expressions in which the machine modes are not all the same.
5105:
5106: @item (udiv:@var{m} @var{x} @var{y})
1.1.1.8 root 5107: Like @code{div} but represents unsigned division.
1.1 root 5108:
5109: @item (mod:@var{m} @var{x} @var{y})
5110: @itemx (umod:@var{m} @var{x} @var{y})
1.1.1.8 root 5111: Like @code{div} and @code{udiv} but represent the remainder instead of
1.1 root 5112: the quotient.
5113:
5114: @item (not:@var{m} @var{x})
5115: Represents the bitwise complement of the value represented by @var{x},
5116: carried out in mode @var{m}, which must be a fixed-point machine mode.
5117: @var{x} must be valid for mode @var{m}, which must be a fixed-point mode.
5118:
5119: @item (and:@var{m} @var{x} @var{y})
5120: Represents the bitwise logical-and of the values represented by
5121: @var{x} and @var{y}, carried out in machine mode @var{m}. This is
5122: valid only if @var{x} and @var{y} both are valid for mode @var{m},
5123: which must be a fixed-point mode.
5124:
5125: @item (ior:@var{m} @var{x} @var{y})
5126: Represents the bitwise inclusive-or of the values represented by
5127: @var{x} and @var{y}, carried out in machine mode @var{m}. This is
5128: valid only if @var{x} and @var{y} both are valid for mode @var{m},
5129: which must be a fixed-point mode.
5130:
5131: @item (xor:@var{m} @var{x} @var{y})
5132: Represents the bitwise exclusive-or of the values represented by
5133: @var{x} and @var{y}, carried out in machine mode @var{m}. This is
5134: valid only if @var{x} and @var{y} both are valid for mode @var{m},
5135: which must be a fixed-point mode.
5136:
5137: @item (lshift:@var{m} @var{x} @var{c})
5138: Represents the result of logically shifting @var{x} left by @var{c}
5139: places. @var{x} must be valid for the mode @var{m}, a fixed-point
5140: machine mode. @var{c} must be valid for a fixed-point mode;
5141: which mode is determined by the mode called for in the machine
5142: description entry for the left-shift instruction. For example,
5143: on the Vax, the mode of @var{c} is @code{QImode} regardless of @var{m}.
5144:
5145: On some machines, negative values of @var{c} may be meaningful; this
5146: is why logical left shift and arithmetic left shift are distinguished.
5147: For example, Vaxes have no right-shift instructions, and right shifts
5148: are represented as left-shift instructions whose counts happen
5149: to be negative constants or else computed (in a previous instruction)
5150: by negation.
5151:
5152: @item (ashift:@var{m} @var{x} @var{c})
1.1.1.8 root 5153: Like @code{lshift} but for arithmetic left shift.
1.1 root 5154:
5155: @item (lshiftrt:@var{m} @var{x} @var{c})
5156: @itemx (ashiftrt:@var{m} @var{x} @var{c})
1.1.1.8 root 5157: Like @code{lshift} and @code{ashift} but for right shift.
1.1 root 5158:
5159: @item (rotate:@var{m} @var{x} @var{c})
5160: @itemx (rotatert:@var{m} @var{x} @var{c})
5161: Similar but represent left and right rotate.
5162:
5163: @item (abs:@var{m} @var{x})
5164: Represents the absolute value of @var{x}, computed in mode @var{m}.
5165: @var{x} must be valid for @var{m}.
5166:
5167: @item (sqrt:@var{m} @var{x})
5168: Represents the square root of @var{x}, computed in mode @var{m}.
5169: @var{x} must be valid for @var{m}. Most often @var{m} will be
5170: a floating point mode.
5171:
5172: @item (ffs:@var{m} @var{x})
1.1.1.10 root 5173: Represents one plus the index of the least significant 1-bit in
1.1 root 5174: @var{x}, represented as an integer of mode @var{m}. (The value is
5175: zero if @var{x} is zero.) The mode of @var{x} need not be @var{m};
5176: depending on the target machine, various mode combinations may be
5177: valid.
5178: @end table
5179:
5180: @node Comparisons, Bit Fields, Arithmetic, RTL
5181: @section Comparison Operations
5182:
1.1.1.10 root 5183: Comparison operators test a relation on two operands and are considered
5184: to represent a machine-dependent nonzero value (@code{STORE_FLAG_VALUE})
5185: if the relation holds, or zero if it does not. The mode of the
5186: comparison is determined by the operands; they must both be valid for a
5187: common machine mode. A comparison with both operands constant would be
5188: invalid as the machine mode could not be deduced from it, but such a
5189: comparison should never exist in RTL due to constant folding.
1.1 root 5190:
5191: Inequality comparisons come in two flavors, signed and unsigned. Thus,
1.1.1.8 root 5192: there are distinct expression codes @code{gt} and @code{gtu} for signed and
1.1 root 5193: unsigned greater-than. These can produce different results for the same
5194: pair of integer values: for example, 1 is signed greater-than -1 but not
5195: unsigned greater-than, because -1 when regarded as unsigned is actually
5196: @code{0xffffffff} which is greater than 1.
5197:
5198: The signed comparisons are also used for floating point values. Floating
5199: point comparisons are distinguished by the machine modes of the operands.
5200:
5201: The comparison operators may be used to compare the condition codes
5202: @code{(cc0)} against zero, as in @code{(eq (cc0) (const_int 0))}. Such a
5203: construct actually refers to the result of the preceding instruction in
5204: which the condition codes were set. The above example stands for 1 if the
5205: condition codes were set to say ``zero'' or ``equal'', 0 otherwise.
5206: Although the same comparison operators are used for this as may be used in
5207: other contexts on actual data, no confusion can result since the machine
5208: description would never allow both kinds of uses in the same context.
5209:
5210: @table @code
5211: @item (eq @var{x} @var{y})
5212: 1 if the values represented by @var{x} and @var{y} are equal,
5213: otherwise 0.
5214:
5215: @item (ne @var{x} @var{y})
5216: 1 if the values represented by @var{x} and @var{y} are not equal,
5217: otherwise 0.
5218:
5219: @item (gt @var{x} @var{y})
5220: 1 if the @var{x} is greater than @var{y}. If they are fixed-point,
5221: the comparison is done in a signed sense.
5222:
5223: @item (gtu @var{x} @var{y})
1.1.1.8 root 5224: Like @code{gt} but does unsigned comparison, on fixed-point numbers only.
1.1 root 5225:
5226: @item (lt @var{x} @var{y})
5227: @item (ltu @var{x} @var{y})
1.1.1.8 root 5228: Like @code{gt} and @code{gtu} but test for ``less than''.
1.1 root 5229:
5230: @item (ge @var{x} @var{y})
5231: @item (geu @var{x} @var{y})
1.1.1.8 root 5232: Like @code{gt} and @code{gtu} but test for ``greater than or equal''.
1.1 root 5233:
5234: @item (le @var{x} @var{y})
5235: @item (leu @var{x} @var{y})
1.1.1.8 root 5236: Like @code{gt} and @code{gtu} but test for ``less than or equal''.
1.1 root 5237:
5238: @item (if_then_else @var{cond} @var{then} @var{else})
5239: This is not a comparison operation but is listed here because it is
5240: always used in conjunction with a comparison operation. To be
5241: precise, @var{cond} is a comparison expression. This expression
5242: represents a choice, according to @var{cond}, between the value
5243: represented by @var{then} and the one represented by @var{else}.
5244:
1.1.1.8 root 5245: On most machines, @code{if_then_else} expressions are valid only
1.1 root 5246: to express conditional jumps.
5247: @end table
5248:
5249: @node Bit Fields, Conversions, Comparisons, RTL
5250: @section Bit-fields
5251:
5252: Special expression codes exist to represent bit-field instructions.
5253: These types of expressions are lvalues in RTL; they may appear
1.1.1.10 root 5254: on the left side of an assignment, indicating insertion of a value
1.1 root 5255: into the specified bit field.
5256:
5257: @table @code
5258: @item (sign_extract:SI @var{loc} @var{size} @var{pos})
5259: This represents a reference to a sign-extended bit-field contained or
5260: starting in @var{loc} (a memory or register reference). The bit field
5261: is @var{size} bits wide and starts at bit @var{pos}. The compilation
5262: option @code{BITS_BIG_ENDIAN} says which end of the memory unit
5263: @var{pos} counts from.
5264:
5265: Which machine modes are valid for @var{loc} depends on the machine,
5266: but typically @var{loc} should be a single byte when in memory
5267: or a full word in a register.
5268:
5269: @item (zero_extract:SI @var{loc} @var{size} @var{pos})
1.1.1.8 root 5270: Like @code{sign_extract} but refers to an unsigned or zero-extended
1.1 root 5271: bit field. The same sequence of bits are extracted, but they
5272: are filled to an entire word with zeros instead of by sign-extension.
5273: @end table
5274:
5275: @node Conversions, RTL Declarations, Bit Fields, RTL
5276: @section Conversions
5277:
5278: All conversions between machine modes must be represented by
5279: explicit conversion operations. For example, an expression
5280: which is the sum of a byte and a full word cannot be written as
1.1.1.8 root 5281: @code{(plus:SI (reg:QI 34) (reg:SI 80))} because the @code{plus}
1.1 root 5282: operation requires two operands of the same machine mode.
5283: Therefore, the byte-sized operand is enclosed in a conversion
5284: operation, as in
5285:
5286: @example
5287: (plus:SI (sign_extend:SI (reg:QI 34)) (reg:SI 80))
5288: @end example
5289:
5290: The conversion operation is not a mere placeholder, because there
5291: may be more than one way of converting from a given starting mode
5292: to the desired final mode. The conversion operation code says how
5293: to do it.
5294:
5295: @table @code
5296: @item (sign_extend:@var{m} @var{x})
5297: Represents the result of sign-extending the value @var{x}
5298: to machine mode @var{m}. @var{m} must be a fixed-point mode
5299: and @var{x} a fixed-point value of a mode narrower than @var{m}.
5300:
5301: @item (zero_extend:@var{m} @var{x})
5302: Represents the result of zero-extending the value @var{x}
5303: to machine mode @var{m}. @var{m} must be a fixed-point mode
5304: and @var{x} a fixed-point value of a mode narrower than @var{m}.
5305:
5306: @item (float_extend:@var{m} @var{x})
5307: Represents the result of extending the value @var{x}
5308: to machine mode @var{m}. @var{m} must be a floating point mode
5309: and @var{x} a floating point value of a mode narrower than @var{m}.
5310:
5311: @item (truncate:@var{m} @var{x})
5312: Represents the result of truncating the value @var{x}
5313: to machine mode @var{m}. @var{m} must be a fixed-point mode
5314: and @var{x} a fixed-point value of a mode wider than @var{m}.
5315:
5316: @item (float_truncate:@var{m} @var{x})
5317: Represents the result of truncating the value @var{x}
5318: to machine mode @var{m}. @var{m} must be a floating point mode
5319: and @var{x} a floating point value of a mode wider than @var{m}.
5320:
5321: @item (float:@var{m} @var{x})
5322: Represents the result of converting fixed point value @var{x},
5323: regarded as signed, to floating point mode @var{m}.
5324:
5325: @item (unsigned_float:@var{m} @var{x})
5326: Represents the result of converting fixed point value @var{x},
5327: regarded as unsigned, to floating point mode @var{m}.
5328:
5329: @item (fix:@var{m} @var{x})
5330: When @var{m} is a fixed point mode, represents the result of
5331: converting floating point value @var{x} to mode @var{m}, regarded as
5332: signed. How rounding is done is not specified, so this operation may
5333: be used validly in compiling C code only for integer-valued operands.
5334:
5335: @item (unsigned_fix:@var{m} @var{x})
5336: Represents the result of converting floating point value @var{x} to
5337: fixed point mode @var{m}, regarded as unsigned. How rounding is done
5338: is not specified.
5339:
5340: @item (fix:@var{m} @var{x})
5341: When @var{m} is a floating point mode, represents the result of
5342: converting floating point value @var{x} (valid for mode @var{m}) to an
5343: integer, still represented in floating point mode @var{m}, by rounding
5344: towards zero.
5345: @end table
5346:
5347: @node RTL Declarations, Side Effects, Conversions, RTL
5348: @section Declarations
5349:
5350: Declaration expression codes do not represent arithmetic operations
5351: but rather state assertions about their operands.
5352:
5353: @table @code
5354: @item (strict_low_part (subreg:@var{m} (reg:@var{n} @var{r}) 0))
5355: This expression code is used in only one context: operand 0 of a
1.1.1.8 root 5356: @code{set} expression. In addition, the operand of this expression
5357: must be a @code{subreg} expression.
1.1 root 5358:
1.1.1.8 root 5359: The presence of @code{strict_low_part} says that the part of the
1.1 root 5360: register which is meaningful in mode @var{n}, but is not part of
5361: mode @var{m}, is not to be altered. Normally, an assignment to such
5362: a subreg is allowed to have undefined effects on the rest of the
5363: register when @var{m} is less than a word.
5364: @end table
5365:
5366: @node Side Effects, Incdec, RTL Declarations, RTL
5367: @section Side Effect Expressions
5368:
5369: The expression codes described so far represent values, not actions.
5370: But machine instructions never produce values; they are meaningful
5371: only for their side effects on the state of the machine. Special
5372: expression codes are used to represent side effects.
5373:
5374: The body of an instruction is always one of these side effect codes;
5375: the codes described above, which represent values, appear only as
5376: the operands of these.
5377:
5378: @table @code
5379: @item (set @var{lval} @var{x})
5380: Represents the action of storing the value of @var{x} into the place
5381: represented by @var{lval}. @var{lval} must be an expression
1.1.1.8 root 5382: representing a place that can be stored in: @code{reg} (or
5383: @code{subreg} or @code{strict_low_part}), @code{mem}, @code{pc} or
5384: @code{cc0}.@refill
1.1 root 5385:
1.1.1.8 root 5386: If @var{lval} is a @code{reg}, @code{subreg} or @code{mem}, it has a
1.1 root 5387: machine mode; then @var{x} must be valid for that mode.@refill
5388:
1.1.1.8 root 5389: If @var{lval} is a @code{reg} whose machine mode is less than the full
1.1 root 5390: width of the register, then it means that the part of the register
5391: specified by the machine mode is given the specified value and the
5392: rest of the register receives an undefined value. Likewise, if
1.1.1.8 root 5393: @var{lval} is a @code{subreg} whose machine mode is narrower than
1.1 root 5394: @code{SImode}, the rest of the register can be changed in an undefined way.
5395:
1.1.1.8 root 5396: If @var{lval} is a @code{strict_low_part} of a @code{subreg}, then the
1.1 root 5397: part of the register specified by the machine mode of the
1.1.1.8 root 5398: @code{subreg} is given the value @var{x} and the rest of the register
1.1 root 5399: is not changed.@refill
5400:
5401: If @var{lval} is @code{(cc0)}, it has no machine mode, and @var{x} may
5402: have any mode. This represents a ``test'' or ``compare'' instruction.@refill
5403:
5404: If @var{lval} is @code{(pc)}, we have a jump instruction, and the
5405: possibilities for @var{x} are very limited. It may be a
1.1.1.8 root 5406: @code{label_ref} expression (unconditional jump). It may be an
5407: @code{if_then_else} (conditional jump), in which case either the
1.1 root 5408: second or the third operand must be @code{(pc)} (for the case which
1.1.1.8 root 5409: does not jump) and the other of the two must be a @code{label_ref}
5410: (for the case which does jump). @var{x} may also be a @code{mem} or
5411: @code{(plus:SI (pc) @var{y})}, where @var{y} may be a @code{reg} or a
5412: @code{mem}; these unusual patterns are used to represent jumps through
1.1 root 5413: branch tables.@refill
5414:
5415: @item (return)
5416: Represents a return from the current function, on machines where this
5417: can be done with one instruction, such as Vaxes. On machines where a
5418: multi-instruction ``epilogue'' must be executed in order to return
5419: from the function, returning is done by jumping to a label which
1.1.1.8 root 5420: precedes the epilogue, and the @code{return} expression code is never
1.1 root 5421: used.
5422:
5423: @item (call @var{function} @var{nargs})
1.1.1.8 root 5424: Represents a function call. @var{function} is a @code{mem} expression
1.1 root 5425: whose address is the address of the function to be called.
5426: @var{nargs} is an expression which can be used for two purposes: on
5427: some machines it represents the number of bytes of stack argument; on
5428: others, it represents the number of argument registers.
5429:
5430: Each machine has a standard machine mode which @var{function} must
5431: have. The machine description defines macro @code{FUNCTION_MODE} to
5432: expand into the requisite mode name. The purpose of this mode is to
5433: specify what kind of addressing is allowed, on machines where the
5434: allowed kinds of addressing depend on the machine mode being
5435: addressed.
5436:
5437: @item (clobber @var{x})
5438: Represents the storing or possible storing of an unpredictable,
1.1.1.8 root 5439: undescribed value into @var{x}, which must be a @code{reg} or
5440: @code{mem} expression.
1.1 root 5441:
5442: One place this is used is in string instructions that store standard
5443: values into particular hard registers. It may not be worth the
5444: trouble to describe the values that are stored, but it is essential to
5445: inform the compiler that the registers will be altered, lest it
5446: attempt to keep data in them across the string instruction.
5447:
5448: @var{x} may also be null---a null C pointer, no expression at all.
5449: Such a @code{(clobber (null))} expression means that all memory
5450: locations must be presumed clobbered.
5451:
5452: Note that the machine description classifies certain hard registers as
5453: ``call-clobbered''. All function call instructions are assumed by
5454: default to clobber these registers, so there is no need to use
1.1.1.8 root 5455: @code{clobber} expressions to indicate this fact. Also, each function
1.1.1.6 root 5456: call is assumed to have the potential to alter any memory location,
5457: unless the function is declared @code{const}.
1.1 root 5458:
1.1.1.8 root 5459: When a @code{clobber} expression for a register appears inside a
5460: @code{parallel} with other side effects, GNU CC guarantees that the
1.1.1.4 root 5461: register is unoccupied both before and after that insn. Therefore, it
5462: is safe for the assembler code produced by the insn to use the
5463: register as a temporary. You can clobber either a specific hard
5464: register or a pseudo register; in the latter case, GNU CC will
5465: allocate a hard register that is available there for use as a
5466: temporary.
5467:
1.1.1.8 root 5468: If you clobber a pseudo register in this way, use a pseudo register
5469: which appears nowhere else---generate a new one each time. Otherwise,
5470: you may confuse CSE.
5471:
5472: There is one other known use for clobbering a pseudo register in a
5473: @code{parallel}: when one of the input operands of the insn is also
5474: clobbered by the insn. In this case, using the same pseudo register in
5475: the clobber and elsewhere in the insn produces the expected results.
5476:
1.1 root 5477: @item (use @var{x})
5478: Represents the use of the value of @var{x}. It indicates that the
5479: value in @var{x} at this point in the program is needed, even though
5480: it may not be apparent why this is so. Therefore, the compiler will
1.1.1.4 root 5481: not attempt to delete previous instructions whose only effect is to
1.1.1.8 root 5482: store a value in @var{x}. @var{x} must be a @code{reg} expression.
1.1 root 5483:
5484: @item (parallel [@var{x0} @var{x1} @dots{}])
5485: Represents several side effects performed in parallel. The square
1.1.1.8 root 5486: brackets stand for a vector; the operand of @code{parallel} is a
1.1 root 5487: vector of expressions. @var{x0}, @var{x1} and so on are individual
1.1.1.8 root 5488: side effect expressions---expressions of code @code{set}, @code{call},
5489: @code{return}, @code{clobber} or @code{use}.@refill
1.1 root 5490:
5491: ``In parallel'' means that first all the values used in the individual
5492: side-effects are computed, and second all the actual side-effects are
5493: performed. For example,
5494:
5495: @example
5496: (parallel [(set (reg:SI 1) (mem:SI (reg:SI 1)))
5497: (set (mem:SI (reg:SI 1)) (reg:SI 1))])
5498: @end example
5499:
5500: @noindent
5501: says unambiguously that the values of hard register 1 and the memory
5502: location addressed by it are interchanged. In both places where
5503: @code{(reg:SI 1)} appears as a memory address it refers to the value
1.1.1.4 root 5504: in register 1 @emph{before} the execution of the insn.
5505:
1.1.1.8 root 5506: It follows that it is @emph{incorrect} to use @code{parallel} and
5507: expect the result of one @code{set} to be available for the next one.
1.1.1.4 root 5508: For example, people sometimes attempt to represent a jump-if-zero
5509: instruction this way:
5510:
5511: @example
5512: (parallel [(set (cc0) (reg:SI 34))
1.1.1.9 root 5513: (set (pc) (if_then_else
5514: (eq (cc0) (const_int 0))
5515: (label_ref @dots{})
5516: (pc)))])
1.1.1.4 root 5517: @end example
5518:
5519: @noindent
5520: But this is incorrect, because it says that the jump condition depends
5521: on the condition code value @emph{before} this instruction, not on the
5522: new value that is set by this instruction.
1.1 root 5523:
1.1.1.5 root 5524: Peephole optimization, which takes place in together with final assembly
1.1.1.8 root 5525: code output, can produce insns whose patterns consist of a @code{parallel}
1.1 root 5526: whose elements are the operands needed to output the resulting
1.1.1.8 root 5527: assembler code---often @code{reg}, @code{mem} or constant expressions.
1.1 root 5528: This would not be well-formed RTL at any other stage in compilation,
5529: but it is ok then because no further optimization remains to be done.
1.1.1.4 root 5530: However, the definition of the macro @code{NOTICE_UPDATE_CC} must
5531: deal with such insns if you define any peephole optimizations.
1.1 root 5532:
5533: @item (sequence [@var{insns} @dots{}])
5534: Represents a sequence of insns. Each of the @var{insns} that appears
5535: in the vector is suitable for appearing in the chain of insns, so it
1.1.1.8 root 5536: must be an @code{insn}, @code{jump_insn}, @code{call_insn},
5537: @code{code_label}, @code{barrier} or @code{note}.
1.1 root 5538:
1.1.1.8 root 5539: A @code{sequence} RTX never appears in an actual insn. It represents
5540: the sequence of insns that result from a @code{define_expand}
1.1 root 5541: @emph{before} those insns are passed to @code{emit_insn} to insert
5542: them in the chain of insns. When actually inserted, the individual
1.1.1.8 root 5543: sub-insns are separated out and the @code{sequence} is forgotten.
1.1 root 5544: @end table
5545:
5546: Three expression codes appear in place of a side effect, as the body of an
5547: insn, though strictly speaking they do not describe side effects as such:
5548:
5549: @table @code
5550: @item (asm_input @var{s})
5551: Represents literal assembler code as described by the string @var{s}.
5552:
5553: @item (addr_vec:@var{m} [@var{lr0} @var{lr1} @dots{}])
5554: Represents a table of jump addresses. The vector elements @var{lr0},
1.1.1.8 root 5555: etc., are @code{label_ref} expressions. The mode @var{m} specifies
1.1 root 5556: how much space is given to each address; normally @var{m} would be
5557: @code{Pmode}.
5558:
5559: @item (addr_diff_vec:@var{m} @var{base} [@var{lr0} @var{lr1} @dots{}])
5560: Represents a table of jump addresses expressed as offsets from
1.1.1.8 root 5561: @var{base}. The vector elements @var{lr0}, etc., are @code{label_ref}
1.1 root 5562: expressions and so is @var{base}. The mode @var{m} specifies how much
5563: space is given to each address-difference.@refill
5564: @end table
5565:
5566: @node Incdec, Assembler, Side Effects, RTL
5567: @section Embedded Side-Effects on Addresses
5568:
5569: Four special side-effect expression codes appear as memory addresses.
5570:
5571: @table @code
5572: @item (pre_dec:@var{m} @var{x})
5573: Represents the side effect of decrementing @var{x} by a standard
5574: amount and represents also the value that @var{x} has after being
1.1.1.8 root 5575: decremented. @var{x} must be a @code{reg} or @code{mem}, but most
5576: machines allow only a @code{reg}. @var{m} must be the machine mode
1.1 root 5577: for pointers on the machine in use. The amount @var{x} is decremented
5578: by is the length in bytes of the machine mode of the containing memory
5579: reference of which this expression serves as the address. Here is an
5580: example of its use:@refill
5581:
5582: @example
5583: (mem:DF (pre_dec:SI (reg:SI 39)))
5584: @end example
5585:
5586: @noindent
5587: This says to decrement pseudo register 39 by the length of a @code{DFmode}
5588: value and use the result to address a @code{DFmode} value.
5589:
5590: @item (pre_inc:@var{m} @var{x})
5591: Similar, but specifies incrementing @var{x} instead of decrementing it.
5592:
5593: @item (post_dec:@var{m} @var{x})
1.1.1.8 root 5594: Represents the same side effect as @code{pre_dec} but a different
1.1 root 5595: value. The value represented here is the value @var{x} has @i{before}
5596: being decremented.
5597:
5598: @item (post_inc:@var{m} @var{x})
5599: Similar, but specifies incrementing @var{x} instead of decrementing it.
5600: @end table
5601:
5602: These embedded side effect expressions must be used with care. Instruction
5603: patterns may not use them. Until the @samp{flow} pass of the compiler,
5604: they may occur only to represent pushes onto the stack. The @samp{flow}
5605: pass finds cases where registers are incremented or decremented in one
5606: instruction and used as an address shortly before or after; these cases are
5607: then transformed to use pre- or post-increment or -decrement.
5608:
5609: Explicit popping of the stack could be represented with these embedded
5610: side effect operators, but that would not be safe; the instruction
5611: combination pass could move the popping past pushes, thus changing
5612: the meaning of the code.
5613:
5614: An instruction that can be represented with an embedded side effect
1.1.1.8 root 5615: could also be represented using @code{parallel} containing an additional
5616: @code{set} to describe how the address register is altered. This is not
1.1 root 5617: done because machines that allow these operations at all typically
5618: allow them wherever a memory address is called for. Describing them as
5619: additional parallel stores would require doubling the number of entries
5620: in the machine description.
5621:
5622: @node Assembler, Insns, IncDec, RTL
5623: @section Assembler Instructions as Expressions
5624:
1.1.1.8 root 5625: The RTX code @code{asm_operands} represents a value produced by a
1.1 root 5626: user-specified assembler instruction. It is used to represent
5627: an @code{asm} statement with arguments. An @code{asm} statement with
5628: a single output operand, like this:
5629:
5630: @example
1.1.1.6 root 5631: asm ("foo %1,%2,%0" : "=a" (outputvar) : "g" (x + y), "di" (*z));
1.1 root 5632: @end example
5633:
5634: @noindent
1.1.1.8 root 5635: is represented using a single @code{asm_operands} RTX which represents
1.1 root 5636: the value that is stored in @code{outputvar}:
5637:
5638: @example
5639: (set @var{rtx-for-outputvar}
5640: (asm_operands "foo %1,%2,%0" "a" 0
5641: [@var{rtx-for-addition-result} @var{rtx-for-*z}]
5642: [(asm_input:@var{m1} "g")
5643: (asm_input:@var{m2} "di")]))
5644: @end example
5645:
5646: @noindent
1.1.1.8 root 5647: Here the operands of the @code{asm_operands} RTX are the assembler
1.1 root 5648: template string, the output-operand's constraint, the index-number of the
5649: output operand among the output operands specified, a vector of input
5650: operand RTX's, and a vector of input-operand modes and constraints. The
5651: mode @var{m1} is the mode of the sum @code{x+y}; @var{m2} is that of
5652: @code{*z}.
5653:
5654: When an @code{asm} statement has multiple output values, its insn has
1.1.1.8 root 5655: several such @code{set} RTX's inside of a @code{parallel}. Each @code{set}
5656: contains a @code{asm_operands}; all of these share the same assembler
1.1 root 5657: template and vectors, but each contains the constraint for the respective
5658: output operand. They are also distinguished by the output-operand index
5659: number, which is 0, 1, @dots{} for successive output operands.
5660:
5661: @node Insns, Calls, Assembler, RTL
5662: @section Insns
5663:
5664: The RTL representation of the code for a function is a doubly-linked
5665: chain of objects called @dfn{insns}. Insns are expressions with
5666: special codes that are used for no other purpose. Some insns are
5667: actual instructions; others represent dispatch tables for @code{switch}
5668: statements; others represent labels to jump to or various sorts of
5669: declarative information.
5670:
5671: In addition to its own specific data, each insn must have a unique id-number
5672: that distinguishes it from all other insns in the current function, and
5673: chain pointers to the preceding and following insns. These three fields
5674: occupy the same position in every insn, independent of the expression code
5675: of the insn. They could be accessed with @code{XEXP} and @code{XINT},
5676: but instead three special macros are always used:
5677:
5678: @table @code
5679: @item INSN_UID (@var{i})
5680: Accesses the unique id of insn @var{i}.
5681:
5682: @item PREV_INSN (@var{i})
5683: Accesses the chain pointer to the insn preceding @var{i}.
5684: If @var{i} is the first insn, this is a null pointer.
5685:
5686: @item NEXT_INSN (@var{i})
5687: Accesses the chain pointer to the insn following @var{i}.
5688: If @var{i} is the last insn, this is a null pointer.
5689: @end table
5690:
5691: The @code{NEXT_INSN} and @code{PREV_INSN} pointers must always
1.1.1.6 root 5692: correspond: if @var{insn} is not the first insn,
1.1 root 5693:
5694: @example
5695: NEXT_INSN (PREV_INSN (@var{insn})) == @var{insn}
5696: @end example
5697:
5698: @noindent
5699: is always true.
5700:
5701: Every insn has one of the following six expression codes:
5702:
1.1.1.8 root 5703: @table @code
1.1 root 5704: @item insn
1.1.1.8 root 5705: The expression code @code{insn} is used for instructions that do not jump
5706: and do not do function calls. Insns with code @code{insn} have four
1.1 root 5707: additional fields beyond the three mandatory ones listed above.
5708: These four are described in a table below.
5709:
5710: @item jump_insn
1.1.1.8 root 5711: The expression code @code{jump_insn} is used for instructions that may jump
5712: (or, more generally, may contain @code{label_ref} expressions).
5713: @code{jump_insn} insns have the same extra fields as @code{insn} insns,
1.1.1.10 root 5714: accessed in the same way. If there is an instruction to return from the
5715: current function, it is recorded as a @code{jump_insn}.
1.1 root 5716:
5717: @item call_insn
1.1.1.8 root 5718: The expression code @code{call_insn} is used for instructions that may do
1.1 root 5719: function calls. It is important to distinguish these instructions because
5720: they imply that certain registers and memory locations may be altered
5721: unpredictably.
5722:
1.1.1.8 root 5723: @code{call_insn} insns have the same extra fields as @code{insn} insns,
1.1 root 5724: accessed in the same way.
5725:
5726: @item code_label
1.1.1.8 root 5727: A @code{code_label} insn represents a label that a jump insn can jump to.
1.1 root 5728: It contains one special field of data in addition to the three standard ones.
5729: It is used to hold the @dfn{label number}, a number that identifies this
5730: label uniquely among all the labels in the compilation (not just in the
5731: current function). Ultimately, the label is represented in the assembler
5732: output as an assembler label @samp{L@var{n}} where @var{n} is the label number.
5733:
5734: @item barrier
5735: Barriers are placed in the instruction stream after unconditional
5736: jump instructions to indicate that the jumps are unconditional.
5737: They contain no information beyond the three standard fields.
5738:
5739: @item note
1.1.1.8 root 5740: @code{note} insns are used to represent additional debugging and
1.1 root 5741: declarative information. They contain two nonstandard fields, an
5742: integer which is accessed with the macro @code{NOTE_LINE_NUMBER} and a
5743: string accessed with @code{NOTE_SOURCE_FILE}.
5744:
5745: If @code{NOTE_LINE_NUMBER} is positive, the note represents the
5746: position of a source line and @code{NOTE_SOURCE_FILE} is the source file name
5747: that the line came from. These notes control generation of line
5748: number data in the assembler output.
5749:
5750: Otherwise, @code{NOTE_LINE_NUMBER} is not really a line number but a
5751: code with one of the following values (and @code{NOTE_SOURCE_FILE}
5752: must contain a null pointer):
5753:
5754: @table @code
5755: @item NOTE_INSN_DELETED
5756: Such a note is completely ignorable. Some passes of the compiler
5757: delete insns by altering them into notes of this kind.
5758:
5759: @item NOTE_INSN_BLOCK_BEG
5760: @itemx NOTE_INSN_BLOCK_END
5761: These types of notes indicate the position of the beginning and end
5762: of a level of scoping of variable names. They control the output
5763: of debugging information.
5764:
5765: @item NOTE_INSN_LOOP_BEG
5766: @itemx NOTE_INSN_LOOP_END
5767: These types of notes indicate the position of the beginning and end
5768: of a @code{while} or @code{for} loop. They enable the loop optimizer
5769: to find loops quickly.
1.1.1.6 root 5770: @item NOTE_INSN_FUNCTION_END
5771: Appears near the end of the function body, just before the label that
5772: @code{return} statements jump to (on machine where a single instruction
5773: does not suffice for returning). This note may be deleted by jump
5774: optimization.
5775: @item NOTE_INSN_SETJMP
5776: Appears following each call to @code{setjmp} or a related function.
1.1.1.7 root 5777:
1.1.1.10 root 5778: @item NOTE_INSN_LOOP_CONT
1.1.1.7 root 5779: Appears at the place in a loop that @code{continue} statements jump to.
1.1 root 5780: @end table
1.1.1.7 root 5781:
5782: These codes are printed symbolically when they appear in debugging dumps.
1.1 root 5783: @end table
5784:
1.1.1.6 root 5785: The machine mode of an insn is normally zero (@code{VOIDmode}), but the
5786: reload pass sets it to @code{QImode} if the insn needs reloading.
5787:
1.1.1.8 root 5788: Here is a table of the extra fields of @code{insn}, @code{jump_insn}
5789: and @code{call_insn} insns:
1.1 root 5790:
5791: @table @code
5792: @item PATTERN (@var{i})
5793: An expression for the side effect performed by this insn.
5794:
1.1.1.6 root 5795: @item INSN_CODE (@var{i})
5796: An integer that says which pattern in the machine description matches
5797: this insn, or -1 if the matching has not yet been attempted.
5798:
5799: Such matching is never attempted and this field is not used on an insn
1.1.1.8 root 5800: whose pattern consists of a single @code{use}, @code{clobber},
5801: @code{asm}, @code{addr_vec} or @code{addr_diff_vec} expression.
1.1 root 5802:
5803: @item LOG_LINKS (@var{i})
1.1.1.8 root 5804: A list (chain of @code{insn_list} expressions) of previous ``related''
1.1 root 5805: insns: insns which store into registers values that are used for the
5806: first time in this insn. (An additional constraint is that neither a
5807: jump nor a label may come between the related insns). This list is
5808: set up by the flow analysis pass; it is a null pointer until then.
5809:
1.1.1.6 root 5810: @item REG_NOTES (@var{i})
1.1.1.8 root 5811: A list (chain of @code{expr_list} expressions) giving information
1.1.1.6 root 5812: about the usage of registers in this insn. This list is set up by the
5813: flow analysis pass; it is a null pointer until then.
1.1 root 5814: @end table
5815:
1.1.1.8 root 5816: The @code{LOG_LINKS} field of an insn is a chain of @code{insn_list}
1.1 root 5817: expressions. Each of these has two operands: the first is an insn,
1.1.1.8 root 5818: and the second is another @code{insn_list} expression (the next one in
5819: the chain). The last @code{insn_list} in the chain has a null pointer
1.1 root 5820: as second operand. The significant thing about the chain is which
1.1.1.8 root 5821: insns appear in it (as first operands of @code{insn_list}
1.1 root 5822: expressions). Their order is not significant.
5823:
5824: The @code{REG_NOTES} field of an insn is a similar chain but of
1.1.1.8 root 5825: @code{expr_list} expressions instead of @code{insn_list}. There are
1.1.1.5 root 5826: several kinds of register notes, which are distinguished by the machine
1.1.1.8 root 5827: mode of the @code{expr_list}, which in a register note is really
1.1.1.5 root 5828: understood as being an @code{enum reg_note}. The first operand @var{op}
1.1.1.8 root 5829: of the @code{expr_list} is data whose meaning depends on the kind of
1.1.1.5 root 5830: note. Here are the kinds of register note:
1.1 root 5831:
5832: @table @code
5833: @item REG_DEAD
5834: The register @var{op} dies in this insn; that is to say, altering the
5835: value immediately after this insn would not affect the future behavior
5836: of the program.
5837:
5838: @item REG_INC
5839: The register @var{op} is incremented (or decremented; at this level
5840: there is no distinction) by an embedded side effect inside this insn.
1.1.1.8 root 5841: This means it appears in a @code{post_inc}, @code{pre_inc},
5842: @code{post_dec} or @code{pre_dec} RTX.
1.1 root 5843:
5844: @item REG_EQUIV
5845: The register that is set by this insn will be equal to @var{op} at run
5846: time, and could validly be replaced in all its occurrences by
5847: @var{op}. (``Validly'' here refers to the data flow of the program;
5848: simple replacement may make some insns invalid.)
5849:
5850: The value which the insn explicitly copies into the register may look
5851: different from @var{op}, but they will be equal at run time.
5852:
5853: For example, when a constant is loaded into a register that is never
5854: assigned any other value, this kind of note is used.
5855:
5856: When a parameter is copied into a pseudo-register at entry to a function,
5857: a note of this kind records that the register is equivalent to the stack
5858: slot where the parameter was passed. Although in this case the register
5859: may be set by other insns, it is still valid to replace the register
5860: by the stack slot throughout the function.
5861:
5862: @item REG_EQUAL
5863: The register that is set by this insn will be equal to @var{op} at run
5864: time at the end of this insn (but not necessarily elsewhere in the
5865: function).
5866:
5867: The RTX @var{op} is typically an arithmetic expression. For example,
5868: when a sequence of insns such as a library call is used to perform an
5869: arithmetic operation, this kind of note is attached to the insn that
5870: produces or copies the final value. It tells the CSE pass how to
5871: think of that value.
5872:
5873: @item REG_RETVAL
5874: This insn copies the value of a library call, and @var{op} is the
5875: first insn that was generated to set up the arguments for the library
5876: call.
5877:
5878: Flow analysis uses this note to delete all of a library call whose
5879: result is dead.
5880:
5881: @item REG_WAS_0
5882: The register @var{op} contained zero before this insn. You can rely
5883: on this note if it is present; its absence implies nothing.
5884:
5885: @item REG_LIBCALL
5886: This is the inverse of @code{REG_RETVAL}: it is placed on the first
5887: insn of a library call, and it points to the last one.
5888:
5889: Loop optimization uses this note to move an entire library call out
5890: of a loop when its value is constant.
5891:
5892: @item REG_NONNEG
5893: The register @var{op} is known to have nonnegative value when this
5894: insn is reached.
5895: @end table
5896:
1.1.1.8 root 5897: For convenience, the machine mode in an @code{insn_list} or
5898: @code{expr_list} is printed using these symbolic codes in debugging dumps.
1.1.1.7 root 5899:
1.1.1.8 root 5900: The only difference between the expression codes @code{insn_list} and
5901: @code{expr_list} is that the first operand of an @code{insn_list} is
1.1 root 5902: assumed to be an insn and is printed in debugging dumps as the insn's
1.1.1.8 root 5903: unique id; the first operand of an @code{expr_list} is printed in the
1.1.1.7 root 5904: ordinary way as an expression.
1.1 root 5905:
5906: @node Calls, Sharing, Insns, RTL
5907: @section RTL Representation of Function-Call Insns
5908:
1.1.1.8 root 5909: Insns that call subroutines have the RTL expression code @code{call_insn}.
1.1 root 5910: These insns must satisfy special rules, and their bodies must use a special
1.1.1.8 root 5911: RTL expression code, @code{call}.
1.1 root 5912:
1.1.1.8 root 5913: A @code{call} expression has two operands, as follows:
1.1 root 5914:
5915: @example
1.1.1.6 root 5916: (call (mem:@var{fm} @var{addr}) @var{nbytes})
1.1 root 5917: @end example
5918:
5919: @noindent
5920: Here @var{nbytes} is an operand that represents the number of bytes of
5921: argument data being passed to the subroutine, @var{fm} is a machine mode
5922: (which must equal as the definition of the @code{FUNCTION_MODE} macro in
5923: the machine description) and @var{addr} represents the address of the
5924: subroutine.
5925:
1.1.1.8 root 5926: For a subroutine that returns no value, the @code{call} RTX as shown above
1.1 root 5927: is the entire body of the insn.
5928:
5929: For a subroutine that returns a value whose mode is not @code{BLKmode},
5930: the value is returned in a hard register. If this register's number is
5931: @var{r}, then the body of the call insn looks like this:
5932:
5933: @example
5934: (set (reg:@var{m} @var{r})
1.1.1.9 root 5935: (call (mem:@var{fm} @var{addr}) @var{nbytes}))
1.1 root 5936: @end example
5937:
5938: @noindent
5939: This RTL expression makes it clear (to the optimizer passes) that the
5940: appropriate register receives a useful value in this insn.
5941:
5942: Immediately after RTL generation, if the value of the subroutine is
5943: actually used, this call insn is always followed closely by an insn which
5944: refers to the register @var{r}. This remains true through all the
5945: optimizer passes until cross jumping occurs.
5946:
5947: The following insn has one of two forms. Either it copies the value into a
5948: pseudo-register, like this:
5949:
5950: @example
5951: (set (reg:@var{m} @var{p}) (reg:@var{m} @var{r}))
5952: @end example
5953:
5954: @noindent
5955: or (in the case where the calling function will simply return whatever
5956: value the call produced, and no operation is needed to do this):
5957:
5958: @example
5959: (use (reg:@var{m} @var{r}))
5960: @end example
5961:
5962: @noindent
5963: Between the call insn and this following insn there may intervene only a
1.1.1.8 root 5964: stack-adjustment insn (and perhaps some @code{note} insns).
1.1 root 5965:
5966: When a subroutine returns a @code{BLKmode} value, it is handled by
5967: passing to the subroutine the address of a place to store the value.
5968: So the call insn itself does not ``return'' any value, and it has the
5969: same RTL form as a call that returns nothing.
5970:
5971: @node Sharing,, Calls, RTL
5972: @section Structure Sharing Assumptions
5973:
5974: The compiler assumes that certain kinds of RTL expressions are unique;
5975: there do not exist two distinct objects representing the same value.
5976: In other cases, it makes an opposite assumption: that no RTL expression
5977: object of a certain kind appears in more than one place in the
5978: containing structure.
5979:
5980: These assumptions refer to a single function; except for the RTL
5981: objects that describe global variables and external functions,
5982: no RTL objects are common to two functions.
5983:
5984: @itemize @bullet
5985: @item
1.1.1.8 root 5986: Each pseudo-register has only a single @code{reg} object to represent it,
1.1 root 5987: and therefore only a single machine mode.
5988:
5989: @item
1.1.1.8 root 5990: For any symbolic label, there is only one @code{symbol_ref} object
1.1 root 5991: referring to it.
5992:
5993: @item
1.1.1.8 root 5994: There is only one @code{const_int} expression with value zero,
1.1 root 5995: and only one with value one.
5996:
5997: @item
1.1.1.8 root 5998: There is only one @code{pc} expression.
1.1 root 5999:
6000: @item
1.1.1.8 root 6001: There is only one @code{cc0} expression.
1.1 root 6002:
6003: @item
1.1.1.8 root 6004: There is only one @code{const_double} expression with mode
1.1 root 6005: @code{SFmode} and value zero, and only one with mode @code{DFmode} and
6006: value zero.
6007:
6008: @item
1.1.1.8 root 6009: No @code{label_ref} appears in more than one place in the RTL
1.1 root 6010: structure; in other words, it is safe to do a tree-walk of all the
1.1.1.8 root 6011: insns in the function and assume that each time a @code{label_ref} is
1.1 root 6012: seen it is distinct from all others that are seen.
6013:
6014: @item
1.1.1.8 root 6015: Only one @code{mem} object is normally created for each static
1.1 root 6016: variable or stack slot, so these objects are frequently shared in all
6017: the places they appear. However, separate but equal objects for these
6018: variables are occasionally made.
6019:
6020: @item
1.1.1.5 root 6021: When a single @code{asm} statement has multiple output operands,
6022: a distinct @code{asm_operands} RTX is made for each output operand.
6023: However, these all share the vector which contains the sequence of
6024: input operands. Because this sharing is used later on to test whether
6025: two @code{asm_operands} RTX's come from the same statement, the sharing
6026: must be guaranteed to be preserved.
6027:
6028: @item
1.1 root 6029: No RTL object appears in more than one place in the RTL structure
6030: except as described above. Many passes of the compiler rely on this
6031: by assuming that they can modify RTL objects in place without unwanted
6032: side-effects on other insns.
6033:
6034: @item
6035: During initial RTL generation, shared structure is freely introduced.
6036: After all the RTL for a function has been generated, all shared
6037: structure is copied by @code{unshare_all_rtl} in @file{emit-rtl.c},
6038: after which the above rules are guaranteed to be followed.
6039:
6040: @item
6041: During the combiner pass, shared structure with an insn can exist
6042: temporarily. However, the shared structure is copied before the
6043: combiner is finished with the insn. This is done by
1.1.1.8 root 6044: @code{copy_substitutions} in @file{combine.c}.
1.1 root 6045: @end itemize
6046:
6047: @node Machine Desc, Machine Macros, RTL, Top
6048: @chapter Machine Descriptions
6049:
6050: A machine description has two parts: a file of instruction patterns
6051: (@file{.md} file) and a C header file of macro definitions.
6052:
6053: The @file{.md} file for a target machine contains a pattern for each
6054: instruction that the target machine supports (or at least each instruction
6055: that is worth telling the compiler about). It may also contain comments.
6056: A semicolon causes the rest of the line to be a comment, unless the semicolon
6057: is inside a quoted string.
6058:
6059: See the next chapter for information on the C header file.
6060:
6061: @menu
6062: * Patterns:: How to write instruction patterns.
1.1.1.8 root 6063: * Example:: An explained example of a @code{define_insn} pattern.
1.1 root 6064: * RTL Template:: The RTL template defines what insns match a pattern.
6065: * Output Template:: The output template says how to make assembler code
6066: from such an insn.
6067: * Output Statement:: For more generality, write C code to output
6068: the assembler code.
6069: * Constraints:: When not all operands are general operands.
6070: * Standard Names:: Names mark patterns to use for code generation.
6071: * Pattern Ordering:: When the order of patterns makes a difference.
6072: * Dependent Patterns:: Having one pattern may make you need another.
6073: * Jump Patterns:: Special considerations for patterns for jump insns.
6074: * Peephole Definitions::Defining machine-specific peephole optimizations.
6075: * Expander Definitions::Generating a sequence of several RTL insns
6076: for a standard operation.
6077: @end menu
6078:
6079: @node Patterns, Example, Machine Desc, Machine Desc
6080: @section Everything about Instruction Patterns
6081:
6082: Each instruction pattern contains an incomplete RTL expression, with pieces
6083: to be filled in later, operand constraints that restrict how the pieces can
6084: be filled in, and an output pattern or C code to generate the assembler
1.1.1.8 root 6085: output, all wrapped up in a @code{define_insn} expression.
1.1 root 6086:
1.1.1.8 root 6087: A @code{define_insn} is an RTL expression containing four or five operands:
1.1 root 6088:
6089: @enumerate
6090: @item
6091: An optional name. The presence of a name indicate that this instruction
6092: pattern can perform a certain standard job for the RTL-generation
6093: pass of the compiler. This pass knows certain names and will use
6094: the instruction patterns with those names, if the names are defined
6095: in the machine description.
6096:
6097: The absence of a name is indicated by writing an empty string
6098: where the name should go. Nameless instruction patterns are never
6099: used for generating RTL code, but they may permit several simpler insns
6100: to be combined later on.
6101:
6102: Names that are not thus known and used in RTL-generation have no
6103: effect; they are equivalent to no name at all.
6104:
6105: @item
6106: The @dfn{RTL template} (@pxref{RTL Template}) is a vector of
6107: incomplete RTL expressions which show what the instruction should look
1.1.1.8 root 6108: like. It is incomplete because it may contain @code{match_operand}
6109: and @code{match_dup} expressions that stand for operands of the
1.1 root 6110: instruction.
6111:
1.1.1.10 root 6112: If the vector has only one element, that element is the template for the
6113: instruction pattern. If the vector has multiple elements, then the
6114: instruction pattern is a @code{parallel} expression containing the
6115: elements described.
1.1 root 6116:
6117: @item
6118: A condition. This is a string which contains a C expression that is
6119: the final test to decide whether an insn body matches this pattern.
6120:
6121: For a named pattern, the condition (if present) may not depend on
6122: the data in the insn being matched, but only the target-machine-type
6123: flags. The compiler needs to test these conditions during
6124: initialization in order to learn exactly which named instructions are
6125: available in a particular run.
6126:
6127: For nameless patterns, the condition is applied only when matching an
6128: individual insn, and only after the insn has matched the pattern's
6129: recognition template. The insn's operands may be found in the vector
6130: @code{operands}.
6131:
6132: @item
6133: The @dfn{output template}: a string that says how to output matching
6134: insns as assembler code. @samp{%} in this string specifies where
6135: to substitute the value of an operand. @xref{Output Template}.
6136:
6137: When simple substitution isn't general enough, you can specify a piece
6138: of C code to compute the output. @xref{Output Statement}.
6139:
6140: @item
6141: Optionally, some @dfn{machine-specific information}. The meaning
6142: of this information is defined only by an individual machine description;
6143: typically it might say whether this insn alters the condition codes,
6144: or how many bytes of output it generates.
6145:
6146: This operand is written as a string containing a C initializer
6147: (complete with braces) for the structure type @code{INSN_MACHINE_INFO},
6148: whose definition is up to you (@pxref{Misc}).
6149: @end enumerate
6150:
6151: @node Example, RTL Template, Patterns, Machine Desc
1.1.1.8 root 6152: @section Example of @code{define_insn}
1.1 root 6153:
6154: Here is an actual example of an instruction pattern, for the 68000/68020.
6155:
6156: @example
6157: (define_insn "tstsi"
6158: [(set (cc0)
6159: (match_operand:SI 0 "general_operand" "rm"))]
6160: ""
6161: "*
6162: @{ if (TARGET_68020 || ! ADDRESS_REG_P (operands[0]))
6163: return \"tstl %0\";
6164: return \"cmpl #0,%0\"; @}")
6165: @end example
6166:
6167: This is an instruction that sets the condition codes based on the value of
6168: a general operand. It has no condition, so any insn whose RTL description
6169: has the form shown may be handled according to this pattern. The name
6170: @samp{tstsi} means ``test a @code{SImode} value'' and tells the RTL generation
6171: pass that, when it is necessary to test such a value, an insn to do so
6172: can be constructed using this pattern.
6173:
6174: The output control string is a piece of C code which chooses which
6175: output template to return based on the kind of operand and the specific
6176: type of CPU for which code is being generated.
6177:
6178: @samp{"rm"} is an operand constraint. Its meaning is explained below.
6179:
6180: @node RTL Template, Output Template, Example, Machine Desc
6181: @section RTL Template for Generating and Recognizing Insns
6182:
6183: The RTL template is used to define which insns match the particular pattern
6184: and how to find their operands. For named patterns, the RTL template also
6185: says how to construct an insn from specified operands.
6186:
6187: Construction involves substituting specified operands into a copy of the
6188: template. Matching involves determining the values that serve as the
6189: operands in the insn being matched. Both of these activities are
6190: controlled by special expression types that direct matching and
6191: substitution of the operands.
6192:
6193: @table @code
1.1.1.8 root 6194: @item (match_operand:@var{m} @var{n} @var{pred} @var{constraint})
1.1 root 6195: This expression is a placeholder for operand number @var{n} of
6196: the insn. When constructing an insn, operand number @var{n}
6197: will be substituted at this point. When matching an insn, whatever
6198: appears at this position in the insn will be taken as operand
1.1.1.8 root 6199: number @var{n}; but it must satisfy @var{pred} or this instruction
1.1 root 6200: pattern will not match at all.
6201:
6202: Operand numbers must be chosen consecutively counting from zero in
1.1.1.8 root 6203: each instruction pattern. There may be only one @code{match_operand}
1.1 root 6204: expression in the pattern for each operand number. Usually operands
1.1.1.8 root 6205: are numbered in the order of appearance in @code{match_operand}
1.1 root 6206: expressions.
6207:
1.1.1.8 root 6208: @var{pred} is a string that is the name of a C function that accepts
6209: two arguments, an expression and a machine mode. During matching, the
6210: function will be called with the putative operand as the expression
6211: and @var{m} as the mode argument. If it returns zero, this
6212: instruction pattern fails to match. @var{pred} may be an empty
6213: string; then it means no test is to be done on the operand,
6214: so anything which occurs in this position is valid.
6215:
6216: @var{constraint} controls reloading and the choice of the best register
6217: class to use for a value, as explained later (@pxref{Constraints}).
6218:
6219: People are often unclear on the difference between the constraint and the
6220: predicate. The predicate helps decide whether a given insn matches the
6221: pattern. The constraint plays no role in this decision; instead, it
6222: controls various decisions in the case of an insn which does match.
6223:
6224: Most often, @var{pred} is @code{"general_operand"}. This function checks
6225: that the putative operand is either a constant, a register or a memory
6226: reference, and that it is valid for mode @var{m}.
1.1 root 6227:
1.1.1.8 root 6228: For an operand that must be a register, @var{pred} should be
1.1 root 6229: @code{"register_operand"}. It would be valid to use
6230: @code{"general_operand"}, since the reload pass would copy any
6231: non-register operands through registers, but this would make GNU CC do
6232: extra work, and it would prevent the register allocator from doing the
6233: best possible job.
6234:
1.1.1.8 root 6235: For an operand that must be a constant, either @var{pred} should be
1.1 root 6236: @code{"immediate_operand"}, or the instruction pattern's extra
6237: condition should check for constants, or both. You cannot expect the
6238: constraints to do this work! If the constraints allow only constants,
6239: but the predicate allows something else, the compiler will crash when
6240: that case arises.
6241:
6242: @item (match_dup @var{n})
6243: This expression is also a placeholder for operand number @var{n}.
6244: It is used when the operand needs to appear more than once in the
6245: insn.
6246:
1.1.1.8 root 6247: In construction, @code{match_dup} behaves exactly like
6248: @code{match_operand}: the operand is substituted into the insn being
6249: constructed. But in matching, @code{match_dup} behaves differently.
1.1 root 6250: It assumes that operand number @var{n} has already been determined by
1.1.1.8 root 6251: a @code{match_operand} appearing earlier in the recognition template,
1.1 root 6252: and it matches only an identical-looking expression.
6253:
1.1.1.4 root 6254: @item (match_operator:@var{m} @var{n} "@var{predicate}" [@var{operands}@dots{}])
6255: This pattern is a kind of placeholder for a variable RTL expression
6256: code.
6257:
6258: When constructing an insn, it stands for an RTL expression whose
6259: expression code is taken from that of operand @var{n}, and whose
6260: operands are constructed from the patterns @var{operands}.
6261:
6262: When matching an expression, it matches an expression if the function
6263: @var{predicate} returns nonzero on that expression @emph{and} the
6264: patterns @var{operands} match the operands of the expression.
6265:
6266: Suppose that the function @code{commutative_operator} is defined as
6267: follows, to match any expression whose operator is one of the six
6268: commutative arithmetic operators of RTL and whose mode is @var{mode}:
6269:
6270: @example
6271: int
6272: commutative_operator (x, mode)
6273: rtx x;
6274: enum machine_mode mode;
6275: @{
6276: enum rtx_code code = GET_CODE (x);
6277: if (GET_MODE (x) != mode)
6278: return 0;
6279: return (code == PLUS || code == MULT || code == UMULT
6280: || code == AND || code == IOR || code == XOR);
6281: @}
6282: @end example
6283:
6284: Then the following pattern will match any RTL expression consisting
6285: of a commutative operator applied to two general operands:
6286:
6287: @example
6288: (match_operator:SI 2 "commutative_operator"
6289: [(match_operand:SI 3 "general_operand" "g")
6290: (match_operand:SI 4 "general_operand" "g")])
6291: @end example
6292:
6293: Here the vector @code{[@var{operands}@dots{}]} contains two patterns
6294: because the expressions to be matched all contain two operands.
6295:
6296: When this pattern does match, the two operands of the commutative
6297: operator are recorded as operands 3 and 4 of the insn. (This is done
1.1.1.8 root 6298: by the two instances of @code{match_operand}.) Operand 2 of the insn
1.1.1.4 root 6299: will be the entire commutative expression: use @code{GET_CODE
6300: (operands[2])} to see which commutative operator was used.
6301:
1.1.1.8 root 6302: The machine mode @var{m} of @code{match_operator} works like that of
6303: @code{match_operand}: it is passed as the second argument to the
1.1.1.4 root 6304: predicate function, and that function is solely responsible for
6305: deciding whether the expression to be matched ``has'' that mode.
6306:
6307: When constructing an insn, argument 2 of the gen-function will specify
6308: the operation (i.e. the expression code) for the expression to be
6309: made. It should be an RTL expression, whose expression code is copied
6310: into a new expression whose operands are arguments 3 and 4 of the
6311: gen-function. The subexpressions of argument 2 are not used;
6312: only its expression code matters.
6313:
1.1.1.8 root 6314: There is no way to specify constraints in @code{match_operator}. The
6315: operand of the insn which corresponds to the @code{match_operator}
1.1.1.4 root 6316: never has any constraints because it is never reloaded as a whole.
6317: However, if parts of its @var{operands} are matched by
1.1.1.8 root 6318: @code{match_operand} patterns, those parts may have constraints of
1.1.1.4 root 6319: their own.
6320:
1.1 root 6321: @item (address (match_operand:@var{m} @var{n} "address_operand" ""))
6322: This complex of expressions is a placeholder for an operand number
6323: @var{n} in a ``load address'' instruction: an operand which specifies
6324: a memory location in the usual way, but for which the actual operand
6325: value used is the address of the location, not the contents of the
6326: location.
6327:
1.1.1.8 root 6328: @code{address} expressions never appear in RTL code, only in machine
1.1 root 6329: descriptions. And they are used only in machine descriptions that do
6330: not use the operand constraint feature. When operand constraints are
6331: in use, the letter @samp{p} in the constraint serves this purpose.
6332:
6333: @var{m} is the machine mode of the @emph{memory location being
6334: addressed}, not the machine mode of the address itself. That mode is
6335: always the same on a given target machine (it is @code{Pmode}, which
6336: normally is @code{SImode}), so there is no point in mentioning it;
1.1.1.8 root 6337: thus, no machine mode is written in the @code{address} expression. If
1.1 root 6338: some day support is added for machines in which addresses of different
6339: kinds of objects appear differently or are used differently (such as
6340: the PDP-10), different formats would perhaps need different machine
1.1.1.8 root 6341: modes and these modes might be written in the @code{address}
1.1 root 6342: expression.
6343: @end table
6344:
6345: @node Output Template, Output Statement, RTL Template, Machine Desc
6346: @section Output Templates and Operand Substitution
6347:
1.1.1.6 root 6348: The @dfn{output template} is a string which specifies how to output the
6349: assembler code for an instruction pattern. Most of the template is a
6350: fixed string which is output literally. The character @samp{%} is used
6351: to specify where to substitute an operand; it can also be used to
6352: identify places where different variants of the assembler require
1.1 root 6353: different syntax.
6354:
6355: In the simplest case, a @samp{%} followed by a digit @var{n} says to output
6356: operand @var{n} at that point in the string.
6357:
6358: @samp{%} followed by a letter and a digit says to output an operand in an
6359: alternate fashion. Four letters have standard, built-in meanings described
6360: below. The machine description macro @code{PRINT_OPERAND} can define
6361: additional letters with nonstandard meanings.
6362:
6363: @samp{%c@var{digit}} can be used to substitute an operand that is a
6364: constant value without the syntax that normally indicates an immediate
6365: operand.
6366:
6367: @samp{%n@var{digit}} is like @samp{%c@var{digit}} except that the value of
6368: the constant is negated before printing.
6369:
6370: @samp{%a@var{digit}} can be used to substitute an operand as if it were a
6371: memory reference, with the actual operand treated as the address. This may
6372: be useful when outputting a ``load address'' instruction, because often the
6373: assembler syntax for such an instruction requires you to write the operand
6374: as if it were a memory reference.
6375:
6376: @samp{%l@var{digit}} is used to substitute a @code{label_ref} into a jump
6377: instruction.
6378:
6379: @samp{%} followed by a punctuation character specifies a substitution that
6380: does not use an operand. Only one case is standard: @samp{%%} outputs a
6381: @samp{%} into the assembler code. Other nonstandard cases can be
1.1.1.8 root 6382: defined in the @code{PRINT_OPERAND} macro. You must also define
6383: which punctuation characters are valid with the
6384: @code{PRINT_OPERAND_PUNCT_VALID_P} macro.
1.1 root 6385:
6386: The template may generate multiple assembler instructions. Write the text
6387: for the instructions, with @samp{\;} between them.
6388:
1.1.1.6 root 6389: When the RTL contains two operands which are required by constraint to match
1.1 root 6390: each other, the output template must refer only to the lower-numbered operand.
6391: Matching operands are not always identical, and the rest of the compiler
6392: arranges to put the proper RTL expression for printing into the lower-numbered
6393: operand.
6394:
6395: One use of nonstandard letters or punctuation following @samp{%} is to
6396: distinguish between different assembler languages for the same machine; for
6397: example, Motorola syntax versus MIT syntax for the 68000. Motorola syntax
6398: requires periods in most opcode names, while MIT syntax does not. For
6399: example, the opcode @samp{movel} in MIT syntax is @samp{move.l} in Motorola
6400: syntax. The same file of patterns is used for both kinds of output syntax,
6401: but the character sequence @samp{%.} is used in each place where Motorola
6402: syntax wants a period. The @code{PRINT_OPERAND} macro for Motorola syntax
6403: defines the sequence to output a period; the macro for MIT syntax defines
6404: it to do nothing.
6405:
6406: @node Output Statement, Constraints, Output Template, Machine Desc
6407: @section C Statements for Generating Assembler Output
6408:
6409: Often a single fixed template string cannot produce correct and efficient
6410: assembler code for all the cases that are recognized by a single
6411: instruction pattern. For example, the opcodes may depend on the kinds of
6412: operands; or some unfortunate combinations of operands may require extra
6413: machine instructions.
6414:
6415: If the output control string starts with a @samp{*}, then it is not an
6416: output template but rather a piece of C program that should compute a
6417: template. It should execute a @code{return} statement to return the
6418: template-string you want. Most such templates use C string literals, which
6419: require doublequote characters to delimit them. To include these
6420: doublequote characters in the string, prefix each one with @samp{\}.
6421:
6422: The operands may be found in the array @code{operands}, whose C data type
6423: is @code{rtx []}.
6424:
6425: It is possible to output an assembler instruction and then go on to output
6426: or compute more of them, using the subroutine @code{output_asm_insn}. This
6427: receives two arguments: a template-string and a vector of operands. The
6428: vector may be @code{operands}, or it may be another array of @code{rtx}
6429: that you declare locally and initialize yourself.
6430:
6431: When an insn pattern has multiple alternatives in its constraints, often
1.1.1.5 root 6432: the appearance of the assembler code is determined mostly by which alternative
1.1 root 6433: was matched. When this is so, the C code can test the variable
6434: @code{which_alternative}, which is the ordinal number of the alternative
6435: that was actually satisfied (0 for the first, 1 for the second alternative,
6436: etc.).
6437:
6438: For example, suppose there are two opcodes for storing zero, @samp{clrreg}
6439: for registers and @samp{clrmem} for memory locations. Here is how
6440: a pattern could use @code{which_alternative} to choose between them:
6441:
6442: @example
6443: (define_insn ""
6444: [(set (match_operand:SI 0 "general_operand" "r,m")
6445: (const_int 0))]
6446: ""
6447: "*
6448: return (which_alternative == 0
6449: ? \"clrreg %0\" : \"clrmem %0\");
6450: ")
6451: @end example
6452:
6453: @node Constraints, Standard Names, Output Statement, Machine Desc
6454: @section Operand Constraints
6455:
1.1.1.8 root 6456: Each @code{match_operand} in an instruction pattern can specify a
1.1 root 6457: constraint for the type of operands allowed. Constraints can say whether
6458: an operand may be in a register, and which kinds of register; whether the
6459: operand can be a memory reference, and which kinds of address; whether the
6460: operand may be an immediate constant, and which possible values it may
6461: have. Constraints can also require two operands to match.
6462:
6463: @menu
6464: * Simple Constraints:: Basic use of constraints.
6465: * Multi-Alternative:: When an insn has two alternative constraint-patterns.
6466: * Class Preferences:: Constraints guide which hard register to put things in.
6467: * Modifiers:: More precise control over effects of constraints.
6468: * No Constraints:: Describing a clean machine without constraints.
6469: @end menu
6470:
6471: @node Simple Constraints, Multi-Alternative, Constraints, Constraints
6472: @subsection Simple Constraints
6473:
6474: The simplest kind of constraint is a string full of letters, each of
6475: which describes one kind of operand that is permitted. Here are
6476: the letters that are allowed:
6477:
6478: @table @asis
6479: @item @samp{m}
6480: A memory operand is allowed, with any kind of address that the machine
6481: supports in general.
6482:
6483: @item @samp{o}
6484: A memory operand is allowed, but only if the address is
1.1.1.8 root 6485: @dfn{offsettable}. This means that adding a small integer (actually,
1.1 root 6486: the width in bytes of the operand, as determined by its machine mode)
6487: may be added to the address and the result is also a valid memory
6488: address.
6489:
1.1.1.8 root 6490: For example, an address which is constant is offsettable; so is an
1.1 root 6491: address that is the sum of a register and a constant (as long as a
6492: slightly larger constant is also within the range of address-offsets
6493: supported by the machine); but an autoincrement or autodecrement
1.1.1.8 root 6494: address is not offsettable. More complicated indirect/indexed
6495: addresses may or may not be offsettable depending on the other
1.1 root 6496: addressing modes that the machine supports.
6497:
6498: Note that in an output operand which can be matched by another
6499: operand, the constraint letter @samp{o} is valid only when accompanied
6500: by both @samp{<} (if the target machine has predecrement addressing)
6501: and @samp{>} (if the target machine has preincrement addressing).
6502:
6503: When the constraint letter @samp{o} is used, the reload pass may
1.1.1.8 root 6504: generate instructions which copy a nonoffsettable address into an index
1.1 root 6505: register. The idea is that the register can be used as a replacement
1.1.1.8 root 6506: offsettable address. But this method requires that there be patterns
1.1 root 6507: to copy any kind of address into a register. Auto-increment
6508: and auto-decrement addresses are an exception; there need not be an
6509: instruction that can copy such an address into a register, because
6510: reload handles these cases specially.
6511:
6512: Most older machine designs have ``load address'' instructions which do
6513: just what is needed here. Some RISC machines do not advertise such
6514: instructions, but the possible addresses on these machines are very
6515: limited, so it is easy to fake them.
6516:
6517: @item @samp{<}
6518: A memory operand with autodecrement addressing (either predecrement or
6519: postdecrement) is allowed.
6520:
6521: @item @samp{>}
6522: A memory operand with autoincrement addressing (either preincrement or
6523: postincrement) is allowed.
6524:
6525: @item @samp{r}
6526: A register operand is allowed provided that it is in a general
6527: register.
6528:
6529: @item @samp{d}, @samp{a}, @samp{f}, @dots{}
6530: Other letters can be defined in machine-dependent fashion to stand for
6531: particular classes of registers. @samp{d}, @samp{a} and @samp{f} are
6532: defined on the 68000/68020 to stand for data, address and floating
6533: point registers.
6534:
6535: @item @samp{i}
6536: An immediate integer operand (one with constant value) is allowed.
6537: This includes symbolic constants whose values will be known only at
6538: assembly time.
6539:
6540: @item @samp{n}
6541: An immediate integer operand with a known numeric value is allowed.
6542: Many systems cannot support assembly-time constants for operands less
6543: than a word wide. Constraints for these operands should use @samp{n}
6544: rather than @samp{i}.
6545:
6546: @item @samp{I}, @samp{J}, @samp{K}, @dots{}
6547: Other letters in the range @samp{I} through @samp{M} may be defined in
6548: a machine-dependent fashion to permit immediate integer operands with
6549: explicit integer values in specified ranges. For example, on the
6550: 68000, @samp{I} is defined to stand for the range of values 1 to 8.
6551: This is the range permitted as a shift count in the shift
6552: instructions.
6553:
6554: @item @samp{F}
1.1.1.8 root 6555: An immediate floating operand (expression code @code{const_double}) is
1.1 root 6556: allowed.
6557:
6558: @item @samp{G}, @samp{H}
6559: @samp{G} and @samp{H} may be defined in a machine-dependent fashion to
6560: permit immediate floating operands in particular ranges of values.
6561:
6562: @item @samp{s}
6563: An immediate integer operand whose value is not an explicit integer is
6564: allowed.
6565:
6566: This might appear strange; if an insn allows a constant operand with a
6567: value not known at compile time, it certainly must allow any known
6568: value. So why use @samp{s} instead of @samp{i}? Sometimes it allows
6569: better code to be generated.
6570:
6571: For example, on the 68000 in a fullword instruction it is possible to
1.1.1.9 root 6572: use an immediate operand; but if the immediate value is between -128
6573: and 127, better code results from loading the value into a register and
1.1 root 6574: using the register. This is because the load into the register can be
6575: done with a @samp{moveq} instruction. We arrange for this to happen
6576: by defining the letter @samp{K} to mean ``any integer outside the
1.1.1.9 root 6577: range -128 to 127'', and then specifying @samp{Ks} in the operand
1.1 root 6578: constraints.
6579:
6580: @item @samp{g}
6581: Any register, memory or immediate integer operand is allowed, except for
6582: registers that are not general registers.
6583:
6584: @item @samp{@var{n}} (a digit)
6585: An operand that matches operand number @var{n} is allowed.
6586: If a digit is used together with letters, the digit should come last.
6587:
6588: This is called a @dfn{matching constraint} and what it really means is
6589: that the assembler has only a single operand that fills two roles
6590: considered separate in the RTL insn. For example, an add insn has two
6591: input operands and one output operand in the RTL, but on most machines
6592: an add instruction really has only two operands, one of them an
6593: input-output operand.
6594:
6595: Matching constraints work only in circumstances like that add insn.
6596: More precisely, the matching constraint must appear in an input-only
6597: operand and the operand that it matches must be an output-only operand
1.1.1.5 root 6598: with a lower number. Thus, operand @var{n} must have @samp{=} in its
6599: constraint.
1.1 root 6600:
6601: For operands to match in a particular case usually means that they
6602: are identical-looking RTL expressions. But in a few special cases
6603: specific kinds of dissimilarity are allowed. For example, @code{*x}
6604: as an input operand will match @code{*x++} as an output operand.
6605: For proper results in such cases, the output template should always
6606: use the output-operand's number when printing the operand.
6607:
6608: @item @samp{p}
6609: An operand that is a valid memory address is allowed. This is
6610: for ``load address'' and ``push address'' instructions.
6611:
1.1.1.8 root 6612: @samp{p} in the constraint must be accompanies by @code{address_operand}
6613: as the predicate in the @code{match_operand}.
1.1 root 6614: @end table
6615:
6616: In order to have valid assembler code, each operand must satisfy
6617: its constraint. But a failure to do so does not prevent the pattern
6618: from applying to an insn. Instead, it directs the compiler to modify
6619: the code so that the constraint will be satisfied. Usually this is
6620: done by copying an operand into a register.
6621:
6622: Contrast, therefore, the two instruction patterns that follow:
6623:
6624: @example
6625: (define_insn ""
6626: [(set (match_operand:SI 0 "general_operand" "r")
6627: (plus:SI (match_dup 0)
6628: (match_operand:SI 1 "general_operand" "r")))]
6629: ""
6630: "@dots{}")
6631: @end example
6632:
6633: @noindent
6634: which has two operands, one of which must appear in two places, and
6635:
6636: @example
6637: (define_insn ""
6638: [(set (match_operand:SI 0 "general_operand" "r")
6639: (plus:SI (match_operand:SI 1 "general_operand" "0")
6640: (match_operand:SI 2 "general_operand" "r")))]
6641: ""
6642: "@dots{}")
6643: @end example
6644:
6645: @noindent
6646: which has three operands, two of which are required by a constraint to be
6647: identical. If we are considering an insn of the form
6648:
6649: @example
6650: (insn @var{n} @var{prev} @var{next}
6651: (set (reg:SI 3)
6652: (plus:SI (reg:SI 6) (reg:SI 109)))
6653: @dots{})
6654: @end example
6655:
6656: @noindent
6657: the first pattern would not apply at all, because this insn does not
6658: contain two identical subexpressions in the right place. The pattern would
6659: say, ``That does not look like an add instruction; try other patterns.''
6660: The second pattern would say, ``Yes, that's an add instruction, but there
6661: is something wrong with it.'' It would direct the reload pass of the
6662: compiler to generate additional insns to make the constraint true. The
6663: results might look like this:
6664:
6665: @example
6666: (insn @var{n2} @var{prev} @var{n}
6667: (set (reg:SI 3) (reg:SI 6))
6668: @dots{})
6669:
6670: (insn @var{n} @var{n2} @var{next}
6671: (set (reg:SI 3)
6672: (plus:SI (reg:SI 3) (reg:SI 109)))
6673: @dots{})
6674: @end example
6675:
6676: It is up to you to make sure that each operand, in each pattern, has
6677: constraints that can handle any RTL expression that could be present for
6678: that operand. (When multiple alternatives are in use, each pattern must,
6679: for each possible combination of operand expressions, have at least one
6680: alternative which can handle that combination of operands.) The
6681: constraints don't need to @emph{allow} any possible operand---when this is
6682: the case, they do not constrain---but they must at least point the way to
6683: reloading any possible operand so that it will fit.
6684:
6685: @itemize @bullet
6686: @item
6687: If the constraint accepts whatever operands the predicate permits,
6688: there is no problem: reloading is never necessary for this operand.
6689:
6690: For example, an operand whose constraints permit everything except
6691: registers is safe provided its predicate rejects registers.
6692:
6693: An operand whose predicate accepts only constant values is safe
6694: provided its constraints include the letter @samp{i}. If any possible
6695: constant value is accepted, then nothing less than @samp{i} will do;
1.1.1.5 root 6696: if the predicate is more selective, then the constraints may also be
1.1 root 6697: more selective.
6698:
6699: @item
6700: Any operand expression can be reloaded by copying it into a register.
6701: So if an operand's constraints allow some kind of register, it is
6702: certain to be safe. It need not permit all classes of registers; the
6703: compiler knows how to copy a register into another register of the
6704: proper class in order to make an instruction valid.
6705:
6706: @item
1.1.1.8 root 6707: A nonoffsettable memory reference can be reloaded by copying the
1.1 root 6708: address into a register. So if the constraint uses the letter
6709: @samp{o}, all memory references are taken care of.
6710:
6711: @item
1.1.1.8 root 6712: A constant operand can be reloaded by allocating space in memory to
6713: hold it as preinitialized data. Then the memory reference can be used
6714: in place of the constant. So if the constraint uses the letters
6715: @samp{o} or @samp{m}, constant operands are not a problem.
1.1 root 6716: @end itemize
6717:
6718: If the operand's predicate can recognize registers, but the constraint does
6719: not permit them, it can make the compiler crash. When this operand happens
6720: to be a register, the reload pass will be stymied, because it does not know
6721: how to copy a register temporarily into memory.
6722:
6723: @node Multi-Alternative, Class Preferences, Simple Constraints, Constraints
6724: @subsection Multiple Alternative Constraints
6725:
6726: Sometimes a single instruction has multiple alternative sets of possible
6727: operands. For example, on the 68000, a logical-or instruction can combine
6728: register or an immediate value into memory, or it can combine any kind of
6729: operand into a register; but it cannot combine one memory location into
6730: another.
6731:
6732: These constraints are represented as multiple alternatives. An alternative
6733: can be described by a series of letters for each operand. The overall
6734: constraint for an operand is made from the letters for this operand
6735: from the first alternative, a comma, the letters for this operand from
6736: the second alternative, a comma, and so on until the last alternative.
6737: Here is how it is done for fullword logical-or on the 68000:
6738:
6739: @example
6740: (define_insn "iorsi3"
1.1.1.9 root 6741: [(set (match_operand:SI 0 "general_operand" "=m,d")
6742: (ior:SI (match_operand:SI 1 "general_operand" "%0,0")
1.1 root 6743: (match_operand:SI 2 "general_operand" "dKs,dmKs")))]
6744: @dots{})
6745: @end example
6746:
6747: The first alternative has @samp{m} (memory) for operand 0, @samp{0} for
1.1.1.9 root 6748: operand 1 (meaning it must match operand 0), and @samp{dKs} for operand
6749: 2. The second alternative has @samp{d} (data register) for operand 0,
6750: @samp{0} for operand 1, and @samp{dmKs} for operand 2. The @samp{=} and
6751: @samp{%} in the constraints apply to all the alternatives; their meaning
1.1 root 6752: is explained in the next section.
6753:
6754: If all the operands fit any one alternative, the instruction is valid.
6755: Otherwise, for each alternative, the compiler counts how many instructions
6756: must be added to copy the operands so that that alternative applies.
6757: The alternative requiring the least copying is chosen. If two alternatives
6758: need the same amount of copying, the one that comes first is chosen.
6759: These choices can be altered with the @samp{?} and @samp{!} characters:
6760:
6761: @table @samp
6762: @item ?
6763: Disparage slightly the alternative that the @samp{?} appears in,
6764: as a choice when no alternative applies exactly. The compiler regards
6765: this alternative as one unit more costly for each @samp{?} that appears
6766: in it.
6767:
6768: @item !
6769: Disparage severely the alternative that the @samp{!} appears in.
6770: When operands must be copied into registers, the compiler will
6771: never choose this alternative as the one to strive for.
6772: @end table
6773:
1.1.1.5 root 6774: When an insn pattern has multiple alternatives in its constraints, often
6775: the appearance of the assembler code is determined mostly by which
1.1 root 6776: alternative was matched. When this is so, the C code for writing the
6777: assembler code can use the variable @code{which_alternative}, which is
1.1.1.5 root 6778: the ordinal number of the alternative that was actually satisfied (0 for
6779: the first, 1 for the second alternative, etc.). For example:
1.1 root 6780:
6781: @example
6782: (define_insn ""
6783: [(set (match_operand:SI 0 "general_operand" "r,m")
6784: (const_int 0))]
6785: ""
6786: "*
6787: return (which_alternative == 0
6788: ? \"clrreg %0\" : \"clrmem %0\");
6789: ")
6790: @end example
6791:
6792: @node Class Preferences, Modifiers, Multi-Alternative, Constraints
6793: @subsection Register Class Preferences
6794:
6795: The operand constraints have another function: they enable the compiler
6796: to decide which kind of hardware register a pseudo register is best
6797: allocated to. The compiler examines the constraints that apply to the
6798: insns that use the pseudo register, looking for the machine-dependent
6799: letters such as @samp{d} and @samp{a} that specify classes of registers.
6800: The pseudo register is put in whichever class gets the most ``votes''.
6801: The constraint letters @samp{g} and @samp{r} also vote: they vote in
6802: favor of a general register. The machine description says which registers
6803: are considered general.
6804:
6805: Of course, on some machines all registers are equivalent, and no register
6806: classes are defined. Then none of this complexity is relevant.
6807:
6808: @node Modifiers, No Constraints, Class Preferences, Constraints
6809: @subsection Constraint Modifier Characters
6810:
6811: @table @samp
6812: @item =
6813: Means that this operand is write-only for this instruction: the previous
6814: value is discarded and replaced by output data.
6815:
6816: @item +
6817: Means that this operand is both read and written by the instruction.
6818:
6819: When the compiler fixes up the operands to satisfy the constraints,
6820: it needs to know which operands are inputs to the instruction and
6821: which are outputs from it. @samp{=} identifies an output; @samp{+}
6822: identifies an operand that is both input and output; all other operands
6823: are assumed to be input only.
6824:
6825: @item &
6826: Means (in a particular alternative) that this operand is written
6827: before the instruction is finished using the input operands.
6828: Therefore, this operand may not lie in a register that is used as an
6829: input operand or as part of any memory address.
6830:
6831: @samp{&} applies only to the alternative in which it is written. In
6832: constraints with multiple alternatives, sometimes one alternative
6833: requires @samp{&} while others do not. See, for example, the
6834: @samp{movdf} insn of the 68000.
6835:
6836: @samp{&} does not obviate the need to write @samp{=}.
6837:
6838: @item %
6839: Declares the instruction to be commutative for this operand and the
6840: following operand. This means that the compiler may interchange the
6841: two operands if that is the cheapest way to make all operands fit the
6842: constraints. This is often used in patterns for addition instructions
6843: that really have only two operands: the result must go in one of the
6844: arguments. Here for example, is how the 68000 halfword-add
6845: instruction is defined:
6846:
6847: @example
6848: (define_insn "addhi3"
6849: [(set (match_operand:HI 0 "general_operand" "=m,r")
6850: (plus:HI (match_operand:HI 1 "general_operand" "%0,0")
6851: (match_operand:HI 2 "general_operand" "di,g")))]
6852: @dots{})
6853: @end example
6854:
6855: Note that in previous versions of GNU CC the @samp{%} constraint
6856: modifier always applied to operands 1 and 2 regardless of which
6857: operand it was written in. The usual custom was to write it in
6858: operand 0. Now it must be in operand 1 if the operands to be
6859: exchanged are 1 and 2.
6860:
6861: @item #
6862: Says that all following characters, up to the next comma, are to be
6863: ignored as a constraint. They are significant only for choosing
6864: register preferences.
6865:
6866: @item *
6867: Says that the following character should be ignored when choosing
6868: register preferences. @samp{*} has no effect on the meaning of the
6869: constraint as a constraint.
6870:
6871: Here is an example: the 68000 has an instruction to sign-extend a
6872: halfword in a data register, and can also sign-extend a value by
6873: copying it into an address register. While either kind of register is
6874: acceptable, the constraints on an address-register destination are
6875: less strict, so it is best if register allocation makes an address
6876: register its goal. Therefore, @samp{*} is used so that the @samp{d}
6877: constraint letter (for data register) is ignored when computing
6878: register preferences.
6879:
6880: @example
6881: (define_insn "extendhisi2"
6882: [(set (match_operand:SI 0 "general_operand" "=*d,a")
6883: (sign_extend:SI
6884: (match_operand:HI 1 "general_operand" "0,g")))]
6885: @dots{})
6886: @end example
6887: @end table
6888:
6889: @node No Constraints,, Modifiers, Constraints
6890: @subsection Not Using Constraints
6891:
6892: Some machines are so clean that operand constraints are not required. For
6893: example, on the Vax, an operand valid in one context is valid in any other
6894: context. On such a machine, every operand constraint would be @samp{g},
6895: excepting only operands of ``load address'' instructions which are
6896: written as if they referred to a memory location's contents but actual
6897: refer to its address. They would have constraint @samp{p}.
6898:
6899: For such machines, instead of writing @samp{g} and @samp{p} for all
6900: the constraints, you can choose to write a description with empty constraints.
1.1.1.8 root 6901: Then you write @samp{""} for the constraint in every @code{match_operand}.
6902: Address operands are identified by writing an @code{address} expression
6903: around the @code{match_operand}, not by their constraints.
1.1 root 6904:
6905: When the machine description has just empty constraints, certain parts
1.1.1.6 root 6906: of compilation are skipped, making the compiler faster. However,
6907: few machines actually do not need constraints; all machine descriptions
6908: now in existence use constraints.
1.1 root 6909:
6910: @node Standard Names, Pattern Ordering, Constraints, Machine Desc
6911: @section Standard Names for Patterns Used in Generation
6912:
6913: Here is a table of the instruction names that are meaningful in the RTL
6914: generation pass of the compiler. Giving one of these names to an
6915: instruction pattern tells the RTL generation pass that it can use the
6916: pattern in to accomplish a certain task.
6917:
6918: @table @asis
6919: @item @samp{mov@var{m}}
1.1.1.8 root 6920: Here @var{m} stands for a two-letter machine mode name, in lower case.
6921: This instruction pattern moves data with that machine mode from operand
6922: 1 to operand 0. For example, @samp{movsi} moves full-word data.
1.1 root 6923:
1.1.1.8 root 6924: If operand 0 is a @code{subreg} with mode @var{m} of a register whose
6925: own mode is wider than @var{m}, the effect of this instruction is
1.1 root 6926: to store the specified value in the part of the register that corresponds
6927: to mode @var{m}. The effect on the rest of the register is undefined.
6928:
6929: This class of patterns is special in several ways. First of all, each
6930: of these names @emph{must} be defined, because there is no other way
6931: to copy a datum from one place to another.
6932:
6933: Second, these patterns are not used solely in the RTL generation pass.
6934: Even the reload pass can generate move insns to copy values from stack
1.1.1.8 root 6935: slots into temporary registers. When it does so, one of the operands is
6936: a hard register and the other is an operand that can need to be reloaded
6937: into a register.
6938:
6939: Therefore, when given such a pair of operands, the pattern must generate
6940: RTL which needs no reloading and needs no temporary registers---no
6941: registers other than the operands. For example, if you support the
6942: pattern with a @code{define_expand}, then in such a case the
6943: @code{define_expand} mustn't call @code{force_reg} or any other such
6944: function which might generate new pseudo registers.
1.1 root 6945:
6946: This requirement exists even for subword modes on a RISC machine where
6947: fetching those modes from memory normally requires several insns and
6948: some temporary registers. Look in @file{spur.md} to see how the
1.1.1.8 root 6949: requirement can be satisfied.
1.1 root 6950:
6951: The variety of operands that have reloads depends on the rest of the
6952: machine description, but typically on a RISC machine these can only be
6953: pseudo registers that did not get hard registers, while on other
6954: machines explicit memory references will get optional reloads.
6955:
1.1.1.9 root 6956: The constraints on a @samp{move@var{m}} must allow any hard register to
6957: be moved to any other hard register (provided that
6958: @code{HARD_REGNO_MODE_OK} permits mode @var{m} in both registers).
6959:
6960: It is obligatory to support floating point @samp{move@var{m}}
6961: instructions into and out of any registers that can hold fixed point
6962: values, because unions and structures (which have modes @code{SImode} or
6963: @code{DImode}) can be in those registers and they may have floating
6964: point members.
6965:
6966: There may also be a need to support fixed point @samp{move@var{m}}
6967: instructions in and out of floating point registers. Unfortunately, I
6968: have forgotten why this was so, and I don't know whether it is still
6969: true. If @code{HARD_REGNO_MODE_OK} rejects fixed point values in
6970: floating point registers, then the constraints of the fixed point
6971: @samp{move@var{m}} instructions must be designed to avoid ever trying to
6972: reload into a floating point register.
1.1 root 6973:
6974: @item @samp{movstrict@var{m}}
1.1.1.8 root 6975: Like @samp{mov@var{m}} except that if operand 0 is a @code{subreg}
1.1 root 6976: with mode @var{m} of a register whose natural mode is wider,
6977: the @samp{movstrict@var{m}} instruction is guaranteed not to alter
6978: any of the register except the part which belongs to mode @var{m}.
6979:
6980: @item @samp{add@var{m}3}
6981: Add operand 2 and operand 1, storing the result in operand 0. All operands
6982: must have mode @var{m}. This can be used even on two-address machines, by
6983: means of constraints requiring operands 1 and 0 to be the same location.
6984:
6985: @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}
6986: Similar, for other arithmetic operations.
6987:
6988: There are special considerations for register classes for logical-and
6989: instructions, affecting also the macro @code{PREFERRED_RELOAD_CLASS}.
6990: They apply not only to the patterns with these standard names, but to
6991: any patterns that will match such an instruction. @xref{Register
6992: Classes}.
6993:
6994: @item @samp{mulhisi3}
6995: Multiply operands 1 and 2, which have mode @code{HImode}, and store
6996: a @code{SImode} product in operand 0.
6997:
6998: @item @samp{mulqihi3}, @samp{mulsidi3}
6999: Similar widening-multiplication instructions of other widths.
7000:
7001: @item @samp{umulqihi3}, @samp{umulhisi3}, @samp{umulsidi3}
7002: Similar widening-multiplication instructions that do unsigned
7003: multiplication.
7004:
7005: @item @samp{divmod@var{m}4}
7006: Signed division that produces both a quotient and a remainder.
7007: Operand 1 is divided by operand 2 to produce a quotient stored
7008: in operand 0 and a remainder stored in operand 3.
7009:
7010: @item @samp{udivmod@var{m}4}
7011: Similar, but does unsigned division.
7012:
7013: @item @samp{ashl@var{m}3}
7014: Arithmetic-shift operand 1 left by a number of bits specified by
7015: operand 2, and store the result in operand 0. Operand 2 has
7016: mode @code{SImode}, not mode @var{m}.
7017:
7018: @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}
7019: Other shift and rotate instructions.
7020:
7021: Logical and arithmetic left shift are the same. Machines that do not
7022: allow negative shift counts often have only one instruction for
7023: shifting left. On such machines, you should define a pattern named
7024: @samp{ashl@var{m}3} and leave @samp{lshl@var{m}3} undefined.
7025:
7026: There are special considerations for register classes for shift
7027: instructions, affecting also the macro @code{PREFERRED_RELOAD_CLASS}.
7028: They apply not only to the patterns with these standard names, but to
7029: any patterns that will match such an instruction. @xref{Register
7030: Classes}.
7031:
7032: @item @samp{neg@var{m}2}
7033: Negate operand 1 and store the result in operand 0.
7034:
7035: @item @samp{abs@var{m}2}
7036: Store the absolute value of operand 1 into operand 0.
7037:
7038: @item @samp{sqrt@var{m}2}
7039: Store the square root of operand 1 into operand 0.
7040:
7041: @item @samp{ffs@var{m}2}
7042: Store into operand 0 one plus the index of the least significant 1-bit
7043: of operand 1. If operand 1 is zero, store zero. @var{m} is the mode
7044: of operand 0; operand 1's mode is specified by the instruction
7045: pattern, and the compiler will convert the operand to that mode before
7046: generating the instruction.
7047:
7048: @item @samp{one_cmpl@var{m}2}
7049: Store the bitwise-complement of operand 1 into operand 0.
7050:
7051: @item @samp{cmp@var{m}}
7052: Compare operand 0 and operand 1, and set the condition codes.
7053: The RTL pattern should look like this:
7054:
7055: @example
1.1.1.6 root 7056: (set (cc0) (compare (match_operand:@var{m} 0 @dots{})
7057: (match_operand:@var{m} 1 @dots{})))
1.1 root 7058: @end example
7059:
7060: Each such definition in the machine description, for integer mode
7061: @var{m}, must have a corresponding @samp{tst@var{m}} pattern, because
7062: optimization can simplify the compare into a test when operand 1 is
7063: zero.
7064:
7065: @item @samp{tst@var{m}}
7066: Compare operand 0 against zero, and set the condition codes.
7067: The RTL pattern should look like this:
7068:
7069: @example
7070: (set (cc0) (match_operand:@var{m} 0 @dots{}))
7071: @end example
7072:
7073: @item @samp{movstr@var{m}}
7074: Block move instruction. The addresses of the destination and source
7075: strings are the first two operands, and both are in mode @code{Pmode}.
7076: The number of bytes to move is the third operand, in mode @var{m}.
1.1.1.5 root 7077: The fourth operand is the known shared alignment of the source and
7078: destination, in the form of a @code{const_int} rtx.
1.1 root 7079:
7080: @item @samp{cmpstr@var{m}}
7081: Block compare instruction, with operands like @samp{movstr@var{m}}
7082: except that the two memory blocks are compared byte by byte
7083: in lexicographic order. The effect of the instruction is to set
7084: the condition codes.
7085:
7086: @item @samp{float@var{m}@var{n}2}
1.1.1.9 root 7087: Convert signed integer operand 1 (valid for fixed point mode @var{m}) to
7088: floating point mode @var{n} and store in operand 0 (which has mode
7089: @var{n}).
7090:
7091: @item @samp{floatuns@var{m}@var{n}2}
7092: Convert unsigned integer operand 1 (valid for fixed point mode @var{m})
7093: to floating point mode @var{n} and store in operand 0 (which has mode
7094: @var{n}).
1.1 root 7095:
7096: @item @samp{fix@var{m}@var{n}2}
7097: Convert operand 1 (valid for floating point mode @var{m}) to fixed
7098: point mode @var{n} as a signed number and store in operand 0 (which
7099: has mode @var{n}). This instruction's result is defined only when
7100: the value of operand 1 is an integer.
7101:
7102: @item @samp{fixuns@var{m}@var{n}2}
7103: Convert operand 1 (valid for floating point mode @var{m}) to fixed
7104: point mode @var{n} as an unsigned number and store in operand 0 (which
7105: has mode @var{n}). This instruction's result is defined only when the
7106: value of operand 1 is an integer.
7107:
7108: @item @samp{ftrunc@var{m}2}
7109: Convert operand 1 (valid for floating point mode @var{m}) to an
7110: integer value, still represented in floating point mode @var{m}, and
7111: store it in operand 0 (valid for floating point mode @var{m}).
7112:
7113: @item @samp{fix_trunc@var{m}@var{n}2}
7114: Like @samp{fix@var{m}@var{n}2} but works for any floating point value
7115: of mode @var{m} by converting the value to an integer.
7116:
7117: @item @samp{fixuns_trunc@var{m}@var{n}2}
7118: Like @samp{fixuns@var{m}@var{n}2} but works for any floating point
7119: value of mode @var{m} by converting the value to an integer.
7120:
7121: @item @samp{trunc@var{m}@var{n}}
7122: Truncate operand 1 (valid for mode @var{m}) to mode @var{n} and
7123: store in operand 0 (which has mode @var{n}). Both modes must be fixed
7124: point or both floating point.
7125:
7126: @item @samp{extend@var{m}@var{n}}
7127: Sign-extend operand 1 (valid for mode @var{m}) to mode @var{n} and
7128: store in operand 0 (which has mode @var{n}). Both modes must be fixed
7129: point or both floating point.
7130:
7131: @item @samp{zero_extend@var{m}@var{n}}
7132: Zero-extend operand 1 (valid for mode @var{m}) to mode @var{n} and
7133: store in operand 0 (which has mode @var{n}). Both modes must be fixed
7134: point.
7135:
7136: @item @samp{extv}
7137: Extract a bit-field from operand 1 (a register or memory operand),
7138: where operand 2 specifies the width in bits and operand 3 the starting
1.1.1.10 root 7139: bit, and store it in operand 0. Operand 0 must have @code{SImode}.
1.1 root 7140: Operand 1 may have mode @code{QImode} or @code{SImode}; often
7141: @code{SImode} is allowed only for registers. Operands 2 and 3 must be
7142: valid for @code{SImode}.
7143:
7144: The RTL generation pass generates this instruction only with constants
7145: for operands 2 and 3.
7146:
7147: The bit-field value is sign-extended to a full word integer
7148: before it is stored in operand 0.
7149:
7150: @item @samp{extzv}
7151: Like @samp{extv} except that the bit-field value is zero-extended.
7152:
7153: @item @samp{insv}
7154: Store operand 3 (which must be valid for @code{SImode}) into a
7155: bit-field in operand 0, where operand 1 specifies the width in bits
7156: and operand 2 the starting bit. Operand 0 may have mode @code{QImode}
7157: or @code{SImode}; often @code{SImode} is allowed only for registers.
7158: Operands 1 and 2 must be valid for @code{SImode}.
7159:
7160: The RTL generation pass generates this instruction only with constants
7161: for operands 1 and 2.
7162:
7163: @item @samp{s@var{cond}}
7164: Store zero or nonzero in the operand according to the condition codes.
7165: Value stored is nonzero iff the condition @var{cond} is true.
7166: @var{cond} is the name of a comparison operation expression code, such
1.1.1.8 root 7167: as @code{eq}, @code{lt} or @code{leu}.
1.1 root 7168:
7169: You specify the mode that the operand must have when you write the
7170: @code{match_operand} expression. The compiler automatically sees
7171: which mode you have used and supplies an operand of that mode.
7172:
1.1.1.8 root 7173: The value stored for a true condition must have 1 as its low bit, or
7174: else must be negative. Otherwise the instruction is not suitable and
7175: must be omitted from the machine description. You must tell the
7176: compiler exactly which value is stored by defining the macro
7177: @code{STORE_FLAG_VALUE}.
1.1 root 7178:
7179: @item @samp{b@var{cond}}
1.1.1.8 root 7180: Conditional branch instruction. Operand 0 is a @code{label_ref}
1.1 root 7181: that refers to the label to jump to. Jump if the condition codes
7182: meet condition @var{cond}.
7183:
7184: @item @samp{call}
7185: Subroutine call instruction returning no value. Operand 0 is the
7186: function to call; operand 1 is the number of bytes of arguments pushed
1.1.1.8 root 7187: (in mode @code{SImode}, except it is normally a @code{const_int});
1.1 root 7188: operand 2 is the number of registers used as operands.
7189:
7190: On most machines, operand 2 is not actually stored into the RTL
7191: pattern. It is supplied for the sake of some RISC machines which need
7192: to put this information into the assembler code; they can put it in
7193: the RTL instead of operand 1.
7194:
1.1.1.8 root 7195: Operand 0 should be a @code{mem} RTX whose address is the address of
1.1 root 7196: the function.
7197:
7198: @item @samp{call_value}
7199: Subroutine call instruction returning a value. Operand 0 is the hard
7200: register in which the value is returned. There are three more
7201: operands, the same as the three operands of the @samp{call}
7202: instruction (but with numbers increased by one).
7203:
7204: Subroutines that return @code{BLKmode} objects use the @samp{call}
7205: insn.
7206:
7207: @item @samp{return}
7208: Subroutine return instruction. This instruction pattern name should be
7209: defined only if a single instruction can do all the work of returning
7210: from a function.
7211:
1.1.1.8 root 7212: @item @samp{nop}
7213: No-op instruction. This instruction pattern name should always be defined
7214: to output a no-op in assembler code. @code{(const_int 0)} will do as an
7215: RTL pattern.
7216:
1.1 root 7217: @item @samp{casesi}
7218: Instruction to jump through a dispatch table, including bounds checking.
7219: This instruction takes five operands:
7220:
7221: @enumerate
7222: @item
7223: The index to dispatch on, which has mode @code{SImode}.
7224:
7225: @item
7226: The lower bound for indices in the table, an integer constant.
7227:
7228: @item
1.1.1.6 root 7229: The total range of indices in the table---the largest index
7230: minus the smallest one (both inclusive).
1.1 root 7231:
7232: @item
7233: A label to jump to if the index has a value outside the bounds.
7234: (If the machine-description macro @code{CASE_DROPS_THROUGH} is defined,
7235: then an out-of-bounds index drops through to the code following
7236: the jump table instead of jumping to this label. In that case,
7237: this label is not actually used by the @samp{casesi} instruction,
7238: but it is always provided as an operand.)
7239:
7240: @item
7241: A label that precedes the table itself.
7242: @end enumerate
7243:
1.1.1.8 root 7244: The table is a @code{addr_vec} or @code{addr_diff_vec} inside of a
7245: @code{jump_insn}. The number of elements in the table is one plus the
1.1 root 7246: difference between the upper bound and the lower bound.
7247:
7248: @item @samp{tablejump}
7249: Instruction to jump to a variable address. This is a low-level
7250: capability which can be used to implement a dispatch table when there
7251: is no @samp{casesi} pattern.
7252:
7253: This pattern requires two operands: the address or offset, and a label
7254: which should immediately precede the jump table. If the macro
7255: @code{CASE_VECTOR_PC_RELATIVE} is defined then the first operand is an
1.1.1.10 root 7256: offset that counts from the address of the table; otherwise, it is an
7257: absolute address to jump to.
1.1 root 7258:
7259: The @samp{tablejump} insn is always the last insn before the jump
7260: table it uses. Its assembler code normally has no need to use the
7261: second operand, but you should incorporate it in the RTL pattern so
7262: that the jump optimizer will not delete the table as unreachable code.
7263: @end table
7264:
7265: @node Pattern Ordering, Dependent Patterns, Standard Names, Machine Desc
7266: @section When the Order of Patterns Matters
7267:
7268: Sometimes an insn can match more than one instruction pattern. Then the
7269: pattern that appears first in the machine description is the one used.
7270: Therefore, more specific patterns (patterns that will match fewer things)
7271: and faster instructions (those that will produce better code when they
7272: do match) should usually go first in the description.
7273:
7274: In some cases the effect of ordering the patterns can be used to hide
7275: a pattern when it is not valid. For example, the 68000 has an
7276: instruction for converting a fullword to floating point and another
7277: for converting a byte to floating point. An instruction converting
7278: an integer to floating point could match either one. We put the
7279: pattern to convert the fullword first to make sure that one will
7280: be used rather than the other. (Otherwise a large integer might
7281: be generated as a single-byte immediate quantity, which would not work.)
7282: Instead of using this pattern ordering it would be possible to make the
7283: pattern for convert-a-byte smart enough to deal properly with any
7284: constant value.
7285:
7286: @node Dependent Patterns, Jump Patterns, Pattern Ordering, Machine Desc
7287: @section Interdependence of Patterns
7288:
7289: Every machine description must have a named pattern for each of the
7290: conditional branch names @samp{b@var{cond}}. The recognition template
7291: must always have the form
7292:
7293: @example
7294: (set (pc)
7295: (if_then_else (@var{cond} (cc0) (const_int 0))
7296: (label_ref (match_operand 0 "" ""))
7297: (pc)))
7298: @end example
7299:
7300: @noindent
7301: In addition, every machine description must have an anonymous pattern
7302: for each of the possible reverse-conditional branches. These patterns
7303: look like
7304:
7305: @example
7306: (set (pc)
7307: (if_then_else (@var{cond} (cc0) (const_int 0))
7308: (pc)
7309: (label_ref (match_operand 0 "" ""))))
7310: @end example
7311:
7312: @noindent
7313: They are necessary because jump optimization can turn direct-conditional
7314: branches into reverse-conditional branches.
7315:
7316: The compiler does more with RTL than just create it from patterns
7317: and recognize the patterns: it can perform arithmetic expression codes
7318: when constant values for their operands can be determined. As a result,
7319: sometimes having one pattern can require other patterns. For example, the
7320: Vax has no `and' instruction, but it has `and not' instructions. Here
7321: is the definition of one of them:
7322:
7323: @example
7324: (define_insn "andcbsi2"
7325: [(set (match_operand:SI 0 "general_operand" "")
7326: (and:SI (match_dup 0)
7327: (not:SI (match_operand:SI
7328: 1 "general_operand" ""))))]
7329: ""
7330: "bicl2 %1,%0")
7331: @end example
7332:
7333: @noindent
7334: If operand 1 is an explicit integer constant, an instruction constructed
7335: using that pattern can be simplified into an `and' like this:
7336:
7337: @example
7338: (set (reg:SI 41)
7339: (and:SI (reg:SI 41)
7340: (const_int 0xffff7fff)))
7341: @end example
7342:
7343: @noindent
7344: (where the integer constant is the one's complement of what
7345: appeared in the original instruction).
7346:
7347: To avoid a fatal error, the compiler must have a pattern that recognizes
7348: such an instruction. Here is what is used:
7349:
7350: @example
7351: (define_insn ""
7352: [(set (match_operand:SI 0 "general_operand" "")
7353: (and:SI (match_dup 0)
7354: (match_operand:SI 1 "general_operand" "")))]
7355: "GET_CODE (operands[1]) == CONST_INT"
7356: "*
7357: @{ operands[1]
7358: = gen_rtx (CONST_INT, VOIDmode, ~INTVAL (operands[1]));
7359: return \"bicl2 %1,%0\";
7360: @}")
7361: @end example
7362:
7363: @noindent
7364: Whereas a pattern to match a general `and' instruction is impossible to
7365: support on the Vax, this pattern is possible because it matches only a
7366: constant second argument: a special case that can be output as an `and not'
7367: instruction.
7368:
7369: A ``compare'' instruction whose RTL looks like this:
7370:
7371: @example
1.1.1.6 root 7372: (set (cc0) (compare @var{operand} (const_int 0)))
1.1 root 7373: @end example
7374:
7375: @noindent
7376: may be simplified by optimization into a ``test'' like this:
7377:
7378: @example
7379: (set (cc0) @var{operand})
7380: @end example
7381:
7382: @noindent
7383: So in the machine description, each ``compare'' pattern for an integer
7384: mode must have a corresponding ``test'' pattern that will match the
7385: result of such simplification.
7386:
7387: In some cases machines support instructions identical except for the
7388: machine mode of one or more operands. For example, there may be
7389: ``sign-extend halfword'' and ``sign-extend byte'' instructions whose
7390: patterns are
7391:
7392: @example
7393: (set (match_operand:SI 0 @dots{})
7394: (extend:SI (match_operand:HI 1 @dots{})))
7395:
7396: (set (match_operand:SI 0 @dots{})
7397: (extend:SI (match_operand:QI 1 @dots{})))
7398: @end example
7399:
7400: @noindent
7401: Constant integers do not specify a machine mode, so an instruction to
7402: extend a constant value could match either pattern. The pattern it
7403: actually will match is the one that appears first in the file. For correct
7404: results, this must be the one for the widest possible mode (@code{HImode},
7405: here). If the pattern matches the @code{QImode} instruction, the results
7406: will be incorrect if the constant value does not actually fit that mode.
7407:
7408: Such instructions to extend constants are rarely generated because they are
7409: optimized away, but they do occasionally happen in nonoptimized
7410: compilations.
7411:
7412: When an instruction has the constraint letter @samp{o}, the reload
1.1.1.8 root 7413: pass may generate instructions which copy a nonoffsettable address into
1.1 root 7414: an index register. The idea is that the register can be used as a
1.1.1.8 root 7415: replacement offsettable address. In order for these generated
1.1 root 7416: instructions to work, there must be patterns to copy any kind of valid
7417: address into a register.
7418:
7419: Most older machine designs have ``load address'' instructions which do
7420: just what is needed here. Some RISC machines do not advertise such
7421: instructions, but the possible addresses on these machines are very
7422: limited, so it is easy to fake them.
7423:
7424: Auto-increment and auto-decrement addresses are an exception; there
7425: need not be an instruction that can copy such an address into a
7426: register, because reload handles these cases in a different manner.
7427:
7428: @node Jump Patterns, Peephole Definitions, Dependent Patterns, Machine Desc
7429: @section Defining Jump Instruction Patterns
7430:
7431: GNU CC assumes that the machine has a condition code. A comparison insn
7432: sets the condition code, recording the results of both signed and unsigned
7433: comparison of the given operands. A separate branch insn tests the
7434: condition code and branches or not according its value. The branch insns
7435: come in distinct signed and unsigned flavors. Many common machines, such
7436: as the Vax, the 68000 and the 32000, work this way.
7437:
7438: Some machines have distinct signed and unsigned compare instructions, and
7439: only one set of conditional branch instructions. The easiest way to handle
7440: these machines is to treat them just like the others until the final stage
7441: where assembly code is written. At this time, when outputting code for the
7442: compare instruction, peek ahead at the following branch using
7443: @code{NEXT_INSN (insn)}. (The variable @code{insn} refers to the insn
7444: being output, in the output-writing code in an instruction pattern.) If
7445: the RTL says that is an unsigned branch, output an unsigned compare;
7446: otherwise output a signed compare. When the branch itself is output, you
7447: can treat signed and unsigned branches identically.
7448:
7449: The reason you can do this is that GNU CC always generates a pair of
7450: consecutive RTL insns, one to set the condition code and one to test it,
7451: and keeps the pair inviolate until the end.
7452:
7453: To go with this technique, you must define the machine-description macro
7454: @code{NOTICE_UPDATE_CC} to do @code{CC_STATUS_INIT}; in other words, no
7455: compare instruction is superfluous.
7456:
7457: Some machines have compare-and-branch instructions and no condition code.
7458: A similar technique works for them. When it is time to ``output'' a
7459: compare instruction, record its operands in two static variables. When
7460: outputting the branch-on-condition-code instruction that follows, actually
7461: output a compare-and-branch instruction that uses the remembered operands.
7462:
7463: It also works to define patterns for compare-and-branch instructions.
7464: In optimizing compilation, the pair of compare and branch instructions
1.1.1.5 root 7465: will be combined according to these patterns. But this does not happen
1.1 root 7466: if optimization is not requested. So you must use one of the solutions
7467: above in addition to any special patterns you define.
7468:
7469: @node Peephole Definitions, Expander Definitions, Jump Patterns, Machine Desc
7470: @section Defining Machine-Specific Peephole Optimizers
7471:
7472: In addition to instruction patterns the @file{md} file may contain
7473: definitions of machine-specific peephole optimizations.
7474:
7475: The combiner does not notice certain peephole optimizations when the data
7476: flow in the program does not suggest that it should try them. For example,
7477: sometimes two consecutive insns related in purpose can be combined even
7478: though the second one does not appear to use a register computed in the
7479: first one. A machine-specific peephole optimizer can detect such
7480: opportunities.
7481:
7482: A definition looks like this:
7483:
7484: @example
7485: (define_peephole
7486: [@var{insn-pattern-1}
7487: @var{insn-pattern-2}
7488: @dots{}]
7489: "@var{condition}"
7490: "@var{template}"
7491: "@var{machine-specific info}")
7492: @end example
7493:
7494: @noindent
7495: The last string operand may be omitted if you are not using any
7496: machine-specific information in this machine description. If present,
1.1.1.8 root 7497: it must obey the same rules as in a @code{define_insn}.
1.1 root 7498:
7499: In this skeleton, @var{insn-pattern-1} and so on are patterns to match
1.1.1.5 root 7500: consecutive insns. The optimization applies to a sequence of insns when
7501: @var{insn-pattern-1} matches the first one, @var{insn-pattern-2} matches
7502: the next, and so on.@refill
1.1 root 7503:
1.1.1.8 root 7504: Each of the insns matched by a peephole must also match a
7505: @code{define_insn}. Peepholes are checked only at the last stage just
7506: before code generation, and only optionally. Therefore, any insn which
7507: would match a peephole but no @code{define_insn} will cause a crash in code
7508: generation in an unoptimized compilation, or at various optimization
7509: stages.
1.1 root 7510:
1.1.1.5 root 7511: The operands of the insns are matched with @code{match_operands} and
7512: @code{match_dup}, as usual. What is not usual is that the operand numbers
7513: apply to all the insn patterns in the definition. So, you can check for
7514: identical operands in two insns by using @code{match_operand} in one insn
7515: and @code{match_dup} in the other.
1.1 root 7516:
7517: The operand constraints used in @code{match_operand} patterns do not have
1.1.1.8 root 7518: any direct effect on the applicability of the peephole, but they will
7519: be validated afterward, so make sure your constraints are general enough
7520: to apply whenever the peephole matches. If the peephole matches
7521: but the constraints are not satisfied, the compiler will crash.
7522:
7523: It is safe to omit constraints in all the operands of the peephole; or
7524: you can write constraints which serve as a double-check on the criteria
7525: previously tested.
1.1 root 7526:
1.1.1.5 root 7527: Once a sequence of insns matches the patterns, the @var{condition} is
7528: checked. This is a C expression which makes the final decision whether to
7529: perform the optimization (we do so if the expression is nonzero). If
1.1 root 7530: @var{condition} is omitted (in other words, the string is empty) then the
1.1.1.5 root 7531: optimization is applied to every sequence of insns that matches the
1.1 root 7532: patterns.
7533:
1.1.1.5 root 7534: The defined peephole optimizations are applied after register allocation
7535: is complete. Therefore, the peephole definition can check which
7536: operands have ended up in which kinds of registers, just by looking at
7537: the operands.
1.1 root 7538:
7539: The way to refer to the operands in @var{condition} is to write
7540: @code{operands[@var{i}]} for operand number @var{i} (as matched by
7541: @code{(match_operand @var{i} @dots{})}). Use the variable @code{insn} to
7542: refer to the last of the insns being matched; use @code{PREV_INSN} to find
1.1.1.8 root 7543: the preceding insns (but be careful to skip over any @code{note} insns that
1.1 root 7544: intervene).@refill
7545:
7546: When optimizing computations with intermediate results, you can use
7547: @var{condition} to match only when the intermediate results are not used
7548: elsewhere. Use the C expression @code{dead_or_set_p (@var{insn},
7549: @var{op})}, where @var{insn} is the insn in which you expect the value to
7550: be used for the last time (from the value of @code{insn}, together with use
7551: of @code{PREV_INSN}), and @var{op} is the intermediate value (from
7552: @code{operands[@var{i}]}).@refill
7553:
1.1.1.5 root 7554: Applying the optimization means replacing the sequence of insns with one
7555: new insn. The @var{template} controls ultimate output of assembler code
7556: for this combined insn. It works exactly like the template of a
7557: @code{define_insn}. Operand numbers in this template are the same ones
7558: used in matching the original sequence of insns.
1.1 root 7559:
7560: The result of a defined peephole optimizer does not need to match any of
1.1.1.5 root 7561: the insn patterns in the machine description; it does not even have an
7562: opportunity to match them. The peephole optimizer definition itself serves
7563: as the insn pattern to control how the insn is output.
7564:
7565: Defined peephole optimizers are run as assembler code is being output,
7566: so the insns they produce are never combined or rearranged in any way.
1.1 root 7567:
7568: Here is an example, taken from the 68000 machine description:
7569:
7570: @example
7571: (define_peephole
7572: [(set (reg:SI 15) (plus:SI (reg:SI 15) (const_int 4)))
7573: (set (match_operand:DF 0 "register_operand" "f")
7574: (match_operand:DF 1 "register_operand" "ad"))]
7575: "FP_REG_P (operands[0]) && ! FP_REG_P (operands[1])"
7576: "*
7577: @{
7578: rtx xoperands[2];
7579: xoperands[1] = gen_rtx (REG, SImode, REGNO (operands[1]) + 1);
7580: #ifdef MOTOROLA
7581: output_asm_insn (\"move.l %1,(sp)\", xoperands);
7582: output_asm_insn (\"move.l %1,-(sp)\", operands);
7583: return \"fmove.d (sp)+,%0\";
7584: #else
7585: output_asm_insn (\"movel %1,sp@@\", xoperands);
7586: output_asm_insn (\"movel %1,sp@@-\", operands);
7587: return \"fmoved sp@@+,%0\";
7588: #endif
7589: @}
7590: ")
7591: @end example
7592:
7593: The effect of this optimization is to change
7594:
7595: @example
7596: jbsr _foobar
7597: addql #4,sp
7598: movel d1,sp@@-
7599: movel d0,sp@@-
7600: fmoved sp@@+,fp0
7601: @end example
7602:
7603: @noindent
7604: into
7605:
7606: @example
7607: jbsr _foobar
7608: movel d1,sp@@
7609: movel d0,sp@@-
7610: fmoved sp@@+,fp0
7611: @end example
7612:
1.1.1.5 root 7613: @ignore
7614: If a peephole matches a sequence including one or more jump insns, you must
7615: take account of the flags such as @code{CC_REVERSED} which specify that the
7616: condition codes are represented in an unusual manner. The compiler
7617: automatically alters any ordinary conditional jumps which occur in such
7618: situations, but the compiler cannot alter jumps which have been replaced by
7619: peephole optimizations. So it is up to you to alter the assembler code
7620: that the peephole produces. Supply C code to write the assembler output,
7621: and in this C code check the condition code status flags and change the
7622: assembler code as appropriate.
7623: @end ignore
7624:
1.1.1.8 root 7625: @var{insn-pattern-1} and so on look @emph{almost} like the second
7626: operand of @code{define_insn}. There is one important difference: the
7627: second operand of @code{define_insn} consists of one or more RTX's
7628: enclosed in square brackets. Usually, there is only one: then the same
7629: action can be written as an element of a @code{define_peephole}. But
7630: when there are multiple actions in a @code{define_insn}, they are
7631: implicitly enclosed in a @code{parallel}. Then you must explicitly
7632: write the @code{parallel}, and the square brackets within it, in the
7633: @code{define_peephole}. Thus, if an insn pattern looks like this,
7634:
7635: @example
7636: (define_insn "divmodsi4"
7637: [(set (match_operand:SI 0 "general_operand" "=d")
7638: (div:SI (match_operand:SI 1 "general_operand" "0")
7639: (match_operand:SI 2 "general_operand" "dmsK")))
7640: (set (match_operand:SI 3 "general_operand" "=d")
7641: (mod:SI (match_dup 1) (match_dup 2)))]
7642: "TARGET_68020"
7643: "divsl%.l %2,%3:%0")
7644: @end example
7645:
7646: @noindent
7647: then the way to mention this insn in a peephole is as follows:
7648:
7649: @example
7650: (define_peephole
7651: [@dots{}
7652: (parallel
7653: [(set (match_operand:SI 0 "general_operand" "=d")
7654: (div:SI (match_operand:SI 1 "general_operand" "0")
7655: (match_operand:SI 2 "general_operand" "dmsK")))
7656: (set (match_operand:SI 3 "general_operand" "=d")
7657: (mod:SI (match_dup 1) (match_dup 2)))])
7658: @dots{}]
7659: @dots{})
7660: @end example
7661:
1.1 root 7662: @node Expander Definitions,, Peephole Definitions, Machine Desc
7663: @section Defining RTL Sequences for Code Generation
7664:
7665: On some target machines, some standard pattern names for RTL generation
7666: cannot be handled with single insn, but a sequence of RTL insns can
7667: represent them. For these target machines, you can write a
1.1.1.8 root 7668: @code{define_expand} to specify how to generate the sequence of RTL.
1.1 root 7669:
1.1.1.8 root 7670: A @code{define_expand} is an RTL expression that looks almost like a
7671: @code{define_insn}; but, unlike the latter, a @code{define_expand} is used
1.1 root 7672: only for RTL generation and it can produce more than one RTL insn.
7673:
1.1.1.8 root 7674: A @code{define_expand} RTX has four operands:
1.1 root 7675:
7676: @itemize @bullet
7677: @item
1.1.1.8 root 7678: The name. Each @code{define_expand} must have a name, since the only
1.1 root 7679: use for it is to refer to it by name.
7680:
7681: @item
7682: The RTL template. This is just like the RTL template for a
1.1.1.8 root 7683: @code{define_peephole} in that it is a vector of RTL expressions
1.1 root 7684: each being one insn.
7685:
7686: @item
7687: The condition, a string containing a C expression. This expression is
7688: used to express how the availability of this pattern depends on
7689: subclasses of target machine, selected by command-line options when
7690: GNU CC is run. This is just like the condition of a
1.1.1.8 root 7691: @code{define_insn} that has a standard name.
1.1 root 7692:
7693: @item
7694: The preparation statements, a string containing zero or more C
7695: statements which are to be executed before RTL code is generated from
7696: the RTL template.
7697:
7698: Usually these statements prepare temporary registers for use as
7699: internal operands in the RTL template, but they can also generate RTL
1.1.1.8 root 7700: insns directly by calling routines such as @code{emit_insn}, etc.
1.1 root 7701: Any such insns precede the ones that come from the RTL template.
7702: @end itemize
7703:
1.1.1.8 root 7704: Every RTL insn emitted by a @code{define_expand} must match some
7705: @code{define_insn} in the machine description. Otherwise, the compiler
7706: will crash when trying to generate code for the insn or trying to optimize
7707: it.
7708:
1.1 root 7709: The RTL template, in addition to controlling generation of RTL insns,
7710: also describes the operands that need to be specified when this pattern
7711: is used. In particular, it gives a predicate for each operand.
7712:
7713: A true operand, which need to be specified in order to generate RTL from
1.1.1.8 root 7714: the pattern, should be described with a @code{match_operand} in its first
1.1 root 7715: occurrence in the RTL template. This enters information on the operand's
7716: predicate into the tables that record such things. GNU CC uses the
7717: information to preload the operand into a register if that is required for
7718: valid RTL code. If the operand is referred to more than once, subsequent
1.1.1.8 root 7719: references should use @code{match_dup}.
1.1 root 7720:
7721: The RTL template may also refer to internal ``operands'' which are
7722: temporary registers or labels used only within the sequence made by the
1.1.1.8 root 7723: @code{define_expand}. Internal operands are substituted into the RTL
7724: template with @code{match_dup}, never with @code{match_operand}. The
1.1 root 7725: values of the internal operands are not passed in as arguments by the
7726: compiler when it requests use of this pattern. Instead, they are computed
7727: within the pattern, in the preparation statements. These statements
7728: compute the values and store them into the appropriate elements of
1.1.1.8 root 7729: @code{operands} so that @code{match_dup} can find them.
1.1 root 7730:
7731: There are two special macros defined for use in the preparation statements:
7732: @code{DONE} and @code{FAIL}. Use them with a following semicolon,
7733: as a statement.
7734:
7735: @table @code
7736: @item DONE
7737: Use the @code{DONE} macro to end RTL generation for the pattern. The
7738: only RTL insns resulting from the pattern on this occasion will be
7739: those already emitted by explicit calls to @code{emit_insn} within the
7740: preparation statements; the RTL template will not be generated.
7741:
7742: @item FAIL
7743: Make the pattern fail on this occasion. When a pattern fails, it means
7744: that the pattern was not truly available. The calling routines in the
7745: compiler will try other strategies for code generation using other patterns.
7746:
7747: Failure is currently supported only for binary operations (addition,
7748: multiplication, shifting, etc.).
7749:
7750: Do not emit any insns explicitly with @code{emit_insn} before failing.
7751: @end table
7752:
7753: Here is an example, the definition of left-shift for the SPUR chip:
7754:
7755: @example
7756: (define_expand "ashlsi3"
7757: [(set (match_operand:SI 0 "register_operand" "")
7758: (ashift:SI
7759: (match_operand:SI 1 "register_operand" "")
7760: (match_operand:SI 2 "nonmemory_operand" "")))]
7761: ""
7762: "
7763: @{
7764: if (GET_CODE (operands[2]) != CONST_INT
7765: || (unsigned) INTVAL (operands[2]) > 3)
7766: FAIL;
7767: @}")
7768: @end example
7769:
7770: @noindent
1.1.1.8 root 7771: This example uses @code{define_expand} so that it can generate an RTL insn
1.1 root 7772: for shifting when the shift-count is in the supported range of 0 to 3 but
7773: fail in other cases where machine insns aren't available. When it fails,
7774: the compiler tries another strategy using different patterns (such as, a
7775: library call).
7776:
7777: If the compiler were able to handle nontrivial condition-strings in
1.1.1.8 root 7778: patterns with names, then it would be possible to use a
7779: @code{define_insn} in that case. Here is another case (zero-extension
7780: on the 68000) which makes more use of the power of @code{define_expand}:
1.1 root 7781:
7782: @example
7783: (define_expand "zero_extendhisi2"
7784: [(set (match_operand:SI 0 "general_operand" "")
7785: (const_int 0))
7786: (set (strict_low_part
7787: (subreg:HI
1.1.1.8 root 7788: (match_dup 0)
1.1 root 7789: 0))
7790: (match_operand:HI 1 "general_operand" ""))]
7791: ""
7792: "operands[1] = make_safe_from (operands[1], operands[0]);")
7793: @end example
7794:
7795: @noindent
7796: Here two RTL insns are generated, one to clear the entire output operand
7797: and the other to copy the input operand into its low half. This sequence
7798: is incorrect if the input operand refers to [the old value of] the output
7799: operand, so the preparation statement makes sure this isn't so. The
7800: function @code{make_safe_from} copies the @code{operands[1]} into a
7801: temporary register if it refers to @code{operands[0]}. It does this
7802: by emitting another RTL insn.
7803:
7804: Finally, a third example shows the use of an internal operand.
1.1.1.8 root 7805: Zero-extension on the SPUR chip is done by @code{and}-ing the result
1.1 root 7806: against a halfword mask. But this mask cannot be represented by a
1.1.1.8 root 7807: @code{const_int} because the constant value is too large to be legitimate
1.1 root 7808: on this machine. So it must be copied into a register with
1.1.1.8 root 7809: @code{force_reg} and then the register used in the @code{and}.
1.1 root 7810:
7811: @example
7812: (define_expand "zero_extendhisi2"
7813: [(set (match_operand:SI 0 "register_operand" "")
7814: (and:SI (subreg:SI
7815: (match_operand:HI 1 "register_operand" "")
7816: 0)
7817: (match_dup 2)))]
7818: ""
7819: "operands[2]
7820: = force_reg (SImode, gen_rtx (CONST_INT,
7821: VOIDmode, 65535)); ")
7822: @end example
7823:
1.1.1.8 root 7824: @strong{Note:} If the @code{define_expand} is used to serve a standard
7825: binary or unary arithmetic operation, then the last insn it generates
7826: must not be a @code{code_label}, @code{barrier} or @code{note}. It must
7827: be an @code{insn}, @code{jump_insn} or @code{call_insn}.
7828:
1.1 root 7829: @node Machine Macros, Config, Machine Desc, Top
7830: @chapter Machine Description Macros
7831:
7832: The other half of the machine description is a C header file conventionally
7833: given the name @file{tm-@var{machine}.h}. The file @file{tm.h} should be a
7834: link to it. The header file @file{config.h} includes @file{tm.h} and most
7835: compiler source files include @file{config.h}.
7836:
7837: @menu
1.1.1.9 root 7838: * Run-time Target:: Defining @samp{-m} options like @samp{-m68000} and @samp{-m68020}.
1.1 root 7839: * Storage Layout:: Defining sizes and alignments of data types.
7840: * Registers:: Naming and describing the hardware registers.
7841: * Register Classes:: Defining the classes of hardware registers.
7842: * Stack Layout:: Defining which way the stack grows and by how much.
7843: * Library Names:: Specifying names of subroutines to call automatically.
7844: * Addressing Modes:: Defining addressing modes valid for memory operands.
1.1.1.8 root 7845: * Delayed Branch:: Do branches execute the following instruction?
1.1 root 7846: * Condition Code:: Defining how insns update the condition code.
1.1.1.5 root 7847: * Cross-compilation:: Handling floating point for cross-compilers.
1.1 root 7848: * Misc:: Everything else.
1.1.1.11! root 7849: * Assembler Format:: Defining how to write insns and pseudo-ops to output.
1.1 root 7850: @end menu
7851:
7852: @node Run-time Target, Storage Layout, Machine Macros, Machine Macros
7853: @section Run-time Target Specification
7854:
7855: @table @code
7856: @item CPP_PREDEFINES
7857: Define this to be a string constant containing @samp{-D} options to
7858: define the predefined macros that identify this machine and system.
7859: These macros will be predefined unless the @samp{-ansi} option is
7860: specified.
7861:
1.1.1.4 root 7862: In addition, a parallel set of macros are predefined, whose names are
7863: made by appending @samp{__} at the beginning and at the end. These
7864: @samp{__} macros are permitted by the ANSI standard, so they are
7865: predefined regardless of whether @samp{-ansi} is specified.
7866:
7867: For example, on the Sun, one can use the following value:
1.1 root 7868:
7869: @example
7870: "-Dmc68000 -Dsun -Dunix"
7871: @end example
7872:
1.1.1.8 root 7873: The result is to define the macros @code{__mc68000__}, @code{__sun__}
7874: and @code{__unix__} unconditionally, and the macros @code{mc68000},
7875: @code{sun} and @code{unix} provided @samp{-ansi} is not specified.
1.1.1.4 root 7876:
1.1 root 7877: @item CPP_SPEC
7878: A C string constant that tells the GNU CC driver program options to
7879: pass to CPP. It can also specify how to translate options you
7880: give to GNU CC into options for GNU CC to pass to the CPP.
7881:
7882: Do not define this macro if it does not need to do anything.
7883:
7884: @item CC1_SPEC
7885: A C string constant that tells the GNU CC driver program options to
7886: pass to CC1. It can also specify how to translate options you
7887: give to GNU CC into options for GNU CC to pass to the CC1.
7888:
7889: Do not define this macro if it does not need to do anything.
7890:
7891: @item extern int target_flags;
7892: This declaration should be present.
7893:
7894: @item TARGET_@dots{}
7895: This series of macros is to allow compiler command arguments to
7896: enable or disable the use of optional features of the target machine.
7897: For example, one machine description serves both the 68000 and
7898: the 68020; a command argument tells the compiler whether it should
7899: use 68020-only instructions or not. This command argument works
7900: by means of a macro @code{TARGET_68020} that tests a bit in
7901: @code{target_flags}.
7902:
7903: Define a macro @code{TARGET_@var{featurename}} for each such option.
7904: Its definition should test a bit in @code{target_flags}; for example:
7905:
7906: @example
7907: #define TARGET_68020 (target_flags & 1)
7908: @end example
7909:
7910: One place where these macros are used is in the condition-expressions
7911: of instruction patterns. Note how @code{TARGET_68020} appears
7912: frequently in the 68000 machine description file, @file{m68k.md}.
7913: Another place they are used is in the definitions of the other
7914: macros in the @file{tm-@var{machine}.h} file.
7915:
7916: @item TARGET_SWITCHES
7917: This macro defines names of command options to set and clear
7918: bits in @code{target_flags}. Its definition is an initializer
7919: with a subgrouping for each command option.
7920:
7921: Each subgrouping contains a string constant, that defines the option
7922: name, and a number, which contains the bits to set in
7923: @code{target_flags}. A negative number says to clear bits instead;
7924: the negative of the number is which bits to clear. The actual option
7925: name is made by appending @samp{-m} to the specified name.
7926:
7927: One of the subgroupings should have a null string. The number in
7928: this grouping is the default value for @code{target_flags}. Any
7929: target options act starting with that value.
7930:
7931: Here is an example which defines @samp{-m68000} and @samp{-m68020}
7932: with opposite meanings, and picks the latter as the default:
7933:
7934: @example
7935: #define TARGET_SWITCHES \
7936: @{ @{ "68020", 1@}, \
7937: @{ "68000", -1@}, \
7938: @{ "", 1@}@}
7939: @end example
7940:
7941: @item OVERRIDE_OPTIONS
7942: Sometimes certain combinations of command options do not make sense on
7943: a particular target machine. You can define a macro
7944: @code{OVERRIDE_OPTIONS} to take account of this. This macro, if
7945: defined, is executed once just after all the command options have been
7946: parsed.
7947: @end table
7948:
7949: @node Storage Layout, Registers, Run-time Target, Machine Macros
7950: @section Storage Layout
7951:
7952: Note that the definitions of the macros in this table which are sizes or
7953: alignments measured in bits do not need to be constant. They can be C
7954: expressions that refer to static variables, such as the @code{target_flags}.
7955: @xref{Run-time Target}.
7956:
7957: @table @code
7958: @item BITS_BIG_ENDIAN
7959: Define this macro if the most significant bit in a byte has the lowest
7960: number. This means that bit-field instructions count from the most
7961: significant bit. If the machine has no bit-field instructions, this
7962: macro is irrelevant.
7963:
1.1.1.8 root 7964: This macro does not affect the way structure fields are packed into
7965: bytes or words; that is controlled by @code{BYTES_BIG_ENDIAN}.
7966:
1.1 root 7967: @item BYTES_BIG_ENDIAN
7968: Define this macro if the most significant byte in a word has the
7969: lowest number.
7970:
7971: @item WORDS_BIG_ENDIAN
7972: Define this macro if, in a multiword object, the most significant
7973: word has the lowest number.
7974:
7975: @item BITS_PER_UNIT
7976: Number of bits in an addressable storage unit (byte); normally 8.
7977:
7978: @item BITS_PER_WORD
7979: Number of bits in a word; normally 32.
7980:
7981: @item UNITS_PER_WORD
7982: Number of storage units in a word; normally 4.
7983:
7984: @item POINTER_SIZE
7985: Width of a pointer, in bits.
7986:
7987: @item POINTER_BOUNDARY
7988: Alignment required for pointers stored in memory, in bits.
7989:
7990: @item PARM_BOUNDARY
1.1.1.7 root 7991: Normal alignment required for function parameters on the stack, in
7992: bits. All stack parameters receive least this much alignment
7993: regardless of data type. On most machines, this is the same as the
7994: size of an integer.
7995:
7996: @item MAX_PARM_BOUNDARY
7997: Largest alignment required for any stack parameters, in bits. If the
7998: data type of the parameter calls for more alignment than
7999: @code{PARM_BOUNDARY}, then it is given extra padding up to this limit.
8000:
8001: Don't define this macro if it would be equal to @code{PARM_BOUNDARY};
8002: in other words, if the alignment of a stack parameter should not
8003: depend on its data type (as is the case on most machines).
1.1 root 8004:
8005: @item STACK_BOUNDARY
8006: Define this macro if you wish to preserve a certain alignment for
8007: the stack pointer at all times. The definition is a C expression
8008: for the desired alignment (measured in bits).
8009:
8010: @item FUNCTION_BOUNDARY
8011: Alignment required for a function entry point, in bits.
8012:
8013: @item BIGGEST_ALIGNMENT
8014: Biggest alignment that any data type can require on this machine, in bits.
8015:
1.1.1.8 root 8016: @item CONSTANT_ALIGNMENT (@var{code}, @var{typealign})
8017: A C expression to compute the alignment for a constant. The argument
8018: @var{typealign} is the alignment required for the constant's data type.
8019: @var{code} is the tree code of the constant itself.
8020:
8021: If this macro is not defined, the default is to use @var{typealign}. If
8022: you do define this macro, the value must be a multiple of
8023: @var{typealign}.
8024:
8025: The purpose of defining this macro is usually to cause string constants
8026: to be word aligned so that @file{dhrystone} can be made to run faster.
8027:
1.1 root 8028: @item EMPTY_FIELD_BOUNDARY
8029: Alignment in bits to be given to a structure bit field that follows an
8030: empty field such as @code{int : 0;}.
8031:
8032: @item STRUCTURE_SIZE_BOUNDARY
8033: Number of bits which any structure or union's size must be a multiple of.
8034: Each structure or union's size is rounded up to a multiple of this.
8035:
8036: If you do not define this macro, the default is the same as
8037: @code{BITS_PER_UNIT}.
8038:
8039: @item STRICT_ALIGNMENT
8040: Define this if instructions will fail to work if given data not
8041: on the nominal alignment. If instructions will merely go slower
8042: in that case, do not define this macro.
8043:
8044: @item PCC_BITFIELD_TYPE_MATTERS
8045: Define this if you wish to imitate a certain bizarre behavior pattern
8046: of some instances of PCC: a bit field whose declared type is
8047: @code{int} has the same effect on the size and alignment of a
8048: structure as an actual @code{int} would have.
8049:
1.1.1.10 root 8050: If the macro is defined, then its definition should be a C expression;
8051: a nonzero value for the expression enables PCC-compatible behavior.
8052:
1.1 root 8053: Just what effect that is in GNU CC depends on other parameters, but on
8054: most machines it would force the structure's alignment and size to a
8055: multiple of 32 or @code{BIGGEST_ALIGNMENT} bits.
8056:
1.1.1.7 root 8057: @item MAX_FIXED_MODE_SIZE
8058: An integer expression for the largest integer machine mode that should
8059: actually be used. All integer machine modes of this size or smaller
8060: can be used for structures and unions with the appropriate sizes.
8061:
1.1 root 8062: @item CHECK_FLOAT_VALUE (@var{mode}, @var{value})
8063: A C statement to validate the value @var{value} (or type
8064: @code{double}) for mode @var{mode}. This means that you check whether
8065: @var{value} fits within the possible range of values for mode
8066: @var{mode} on this target machine. The mode @var{mode} is always
8067: @code{SFmode} or @code{DFmode}.
8068:
8069: If @var{value} is not valid, you should call @code{error} to print an
8070: error message and then assign some valid value to @var{value}.
8071: Allowing an invalid value to go through the compiler can produce
8072: incorrect assembler code which may even cause Unix assemblers to
8073: crash.
8074:
8075: This macro need not be defined if there is no work for it to do.
8076: @end table
8077:
8078: @node Registers, Register Classes, Storage Layout, Machine Macros
8079: @section Register Usage
8080:
8081: @table @code
8082: @item FIRST_PSEUDO_REGISTER
8083: Number of hardware registers known to the compiler. They receive
8084: numbers 0 through @code{FIRST_PSEUDO_REGISTER-1}; thus, the first
8085: pseudo register's number really is assigned the number
8086: @code{FIRST_PSEUDO_REGISTER}.
8087:
8088: @item FIXED_REGISTERS
8089: An initializer that says which registers are used for fixed purposes
8090: all throughout the compiled code and are therefore not available for
8091: general allocation. These would include the stack pointer, the frame
8092: pointer (except on machines where that can be used as a general
8093: register when no frame pointer is needed), the program counter on
8094: machines where that is considered one of the addressable registers,
8095: and any other numbered register with a standard use.
8096:
8097: This information is expressed as a sequence of numbers, separated by
8098: commas and surrounded by braces. The @var{n}th number is 1 if
8099: register @var{n} is fixed, 0 otherwise.
8100:
8101: The table initialized from this macro, and the table initialized by
8102: the following one, may be overridden at run time either automatically,
8103: by the actions of the macro @code{CONDITIONAL_REGISTER_USAGE}, or by
8104: the user with the command options @samp{-ffixed-@var{reg}},
8105: @samp{-fcall-used-@var{reg}} and @samp{-fcall-saved-@var{reg}}.
8106:
8107: @item CALL_USED_REGISTERS
8108: Like @code{FIXED_REGISTERS} but has 1 for each register that is
8109: clobbered (in general) by function calls as well as for fixed
8110: registers. This macro therefore identifies the registers that are not
8111: available for general allocation of values that must live across
8112: function calls.
8113:
8114: If a register has 0 in @code{CALL_USED_REGISTERS}, the compiler
8115: automatically saves it on function entry and restores it on function
8116: exit, if the register is used within the function.
8117:
1.1.1.6 root 8118: @item DEFAULT_CALLER_SAVES
1.1.1.8 root 8119: Define this macro if function calls on the target machine do not preserve
1.1.1.6 root 8120: any registers; in other words, if @code{CALL_USED_REGISTERS} has 1
8121: for all registers. This macro enables @samp{-fcaller-saves} by default.
8122: Eventually that option will be enabled by default on all machines and both
8123: the option and this macro will be eliminated.
8124:
1.1 root 8125: @item CONDITIONAL_REGISTER_USAGE
8126: Zero or more C statements that may conditionally modify two variables
8127: @code{fixed_regs} and @code{call_used_regs} (both of type @code{char
8128: []}) after they have been initialized from the two preceding macros.
8129:
8130: This is necessary in case the fixed or call-clobbered registers depend
8131: on target flags.
8132:
8133: You need not define this macro if it has no work to do.
8134:
8135: If the usage of an entire class of registers depends on the target
1.1.1.5 root 8136: flags, you may indicate this to GCC by using this macro to modify
1.1 root 8137: @code{fixed_regs} and @code{call_used_regs} to 1 for each of the
1.1.1.5 root 8138: registers in the classes which should not be used by GCC. Also define
1.1 root 8139: the macro @code{REG_CLASS_FROM_LETTER} to return @code{NO_REGS} if it
8140: is called with a letter for a class that shouldn't be used.
8141:
8142: (However, if this class is not included in @code{GENERAL_REGS} and all
8143: of the insn patterns whose constraints permit this class are
8144: controlled by target switches, then GCC will automatically avoid using
8145: these registers when the target switches are opposed to them.)
8146:
8147: @item OVERLAPPING_REGNO_P (@var{regno})
1.1.1.5 root 8148: If defined, this is a C expression whose value is nonzero if hard
8149: register number @var{regno} is an overlapping register. This means a
8150: hard register which overlaps a hard register with a different number.
8151: (Such overlap is undesirable, but occasionally it allows a machine to
8152: be supported which otherwise could not be.) This macro must return
8153: nonzero for @emph{all} the registers which overlap each other. GNU CC
8154: can use an overlapping register only in certain limited ways. It can
8155: be used for allocation within a basic block, and may be spilled for
8156: reloading; that is all.
1.1 root 8157:
8158: If this macro is not defined, it means that none of the hard registers
8159: overlap each other. This is the usual situation.
8160:
8161: @item INSN_CLOBBERS_REGNO_P (@var{insn}, @var{regno})
8162: If defined, this is a C expression whose value should be nonzero if
8163: the insn @var{insn} has the effect of mysteriously clobbering the
8164: contents of hard register number @var{regno}. By ``mysterious'' we
8165: mean that the insn's RTL expression doesn't describe such an effect.
8166:
8167: If this macro is not defined, it means that no insn clobbers registers
8168: mysteriously. This is the usual situation; all else being equal,
8169: it is best for the RTL expression to show all the activity.
8170:
8171: @item PRESERVE_DEATH_INFO_REGNO_P (@var{regno})
8172: If defined, this is a C expression whose value is nonzero if accurate
8173: @code{REG_DEAD} notes are needed for hard register number @var{regno}
8174: at the time of outputting the assembler code. When this is so, a few
8175: optimizations that take place after register allocation and could
8176: invalidate the death notes are not done when this register is
8177: involved.
8178:
1.1.1.8 root 8179: You would arrange to preserve death info for a register when some of the
8180: code in the machine description which is executed to write the assembler
8181: code looks at the death notes. This is necessary only when the actual
8182: hardware feature which GNU CC thinks of as a register is not actually a
8183: register of the usual sort. (It might, for example, be a hardware
8184: stack.)
1.1 root 8185:
8186: If this macro is not defined, it means that no death notes need to be
8187: preserved. This is the usual situation.
8188:
1.1.1.10 root 8189: @item HARD_REGNO_NREGS (@var{regno}, @var{mode})
1.1 root 8190: A C expression for the number of consecutive hard registers, starting
8191: at register number @var{regno}, required to hold a value of mode
8192: @var{mode}.
8193:
8194: On a machine where all registers are exactly one word, a suitable
8195: definition of this macro is
8196:
8197: @example
8198: #define HARD_REGNO_NREGS(REGNO, MODE) \
8199: ((GET_MODE_SIZE (MODE) + UNITS_PER_WORD - 1) \
8200: / UNITS_PER_WORD))
8201: @end example
8202:
8203: @item HARD_REGNO_MODE_OK (@var{regno}, @var{mode})
8204: A C expression that is nonzero if it is permissible to store a value
8205: of mode @var{mode} in hard register number @var{regno} (or in several
8206: registers starting with that one). For a machine where all registers
8207: are equivalent, a suitable definition is
8208:
8209: @example
8210: #define HARD_REGNO_MODE_OK(REGNO, MODE) 1
8211: @end example
8212:
1.1.1.8 root 8213: It is not necessary for this macro to check for the numbers of fixed
8214: registers, because the allocation mechanism considers them to be always
8215: occupied.
8216:
8217: On some machines, double-precision values must be kept in even/odd
8218: register pairs. The way to implement that is to define this macro
8219: to reject odd register numbers for such modes.
8220:
8221: GNU CC assumes that it can always move values between registers and
8222: (suitably addressed) memory locations. If it is impossible to move a
8223: value of a certain mode between memory and certain registers, then
8224: @code{HARD_REGNO_MODE_OK} must not allow this mode in those registers.
1.1 root 8225:
8226: Many machines have special registers for floating point arithmetic.
8227: Often people assume that floating point machine modes are allowed only
8228: in floating point registers. This is not true. Any registers that
8229: can hold integers can safely @emph{hold} a floating point machine
8230: mode, whether or not floating arithmetic can be done on it in those
8231: registers.
8232:
1.1.1.9 root 8233: On some machines, though, the converse is true: fixed-point machine
8234: modes may not go in floating registers. This is true if the floating
8235: registers normalize any value stored in them, because storing a
8236: non-floating value there would garble it. In this case,
8237: @code{HARD_REGNO_MODE_OK} should reject fixed-point machine modes in
8238: floating registers. But if the floating registers do not automatically
8239: normalize, if you can store any bit pattern in one and retrieve it
8240: unchanged without a trap, then any machine mode may go in a floating
8241: register and this macro should say so.
8242:
8243: The primary significance of special floating registers is rather that
8244: they are the registers acceptable in floating point arithmetic
8245: instructions. However, this is of no concern to
8246: @code{HARD_REGNO_MODE_OK}. You handle it by writing the proper
8247: constraints for those instructions.
8248:
8249: On some machines, the floating registers are especially slow to access,
8250: so that it is better to store a value in a stack frame than in such a
8251: register if floating point arithmetic is not being done. As long as the
8252: floating registers are not in class @code{GENERAL_REGS}, they will not
8253: be used unless some insn's constraint asks for one.
1.1 root 8254:
8255: @item MODES_TIEABLE_P (@var{mode1}, @var{mode2})
8256: A C expression that is nonzero if it is desirable to choose register
8257: allocation so as to avoid move instructions between a value of mode
8258: @var{mode1} and a value of mode @var{mode2}.
8259:
8260: If @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode1})} and
8261: @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode2})} are ever different
8262: for any @var{r}, then @code{MODES_TIEABLE_P (@var{mode1},
8263: @var{mode2})} must be zero.
8264:
8265: @item PC_REGNUM
8266: If the program counter has a register number, define this as that
8267: register number. Otherwise, do not define it.
8268:
8269: @item STACK_POINTER_REGNUM
8270: The register number of the stack pointer register, which must also be
8271: a fixed register according to @code{FIXED_REGISTERS}. On many
8272: machines, the hardware determines which register this is.
8273:
8274: @item FRAME_POINTER_REGNUM
8275: The register number of the frame pointer register, which is used to
8276: access automatic variables in the stack frame. On some machines, the
8277: hardware determines which register this is. On other machines, you
8278: can choose any register you wish for this purpose.
8279:
8280: @item FRAME_POINTER_REQUIRED
1.1.1.9 root 8281: A C expression which is nonzero if a function must have and use a frame
8282: pointer. This expression is evaluated twice: at the beginning of
8283: generating RTL, and in the reload pass. If its value is nonzero at
8284: either time, then the function will have a frame pointer.
8285:
8286: The expression can in principle examine the current function and decide
8287: according to the facts, but on most machines the constant 0 or the
8288: constant 1 suffices. Use 0 when the machine allows code to be generated
8289: with no frame pointer, and doing so saves some time or space. Use 1
8290: when there is no possible advantage to avoiding a frame pointer.
1.1 root 8291:
1.1.1.5 root 8292: In certain cases, the compiler does not know how to produce valid code
8293: without a frame pointer. The compiler recognizes those cases and
8294: automatically gives the function a frame pointer regardless of what
1.1 root 8295: @code{FRAME_POINTER_REQUIRED} says. You don't need to worry about
8296: them.@refill
8297:
8298: In a function that does not require a frame pointer, the frame pointer
8299: register can be allocated for ordinary usage, unless you mark it as a
8300: fixed register. See @code{FIXED_REGISTERS} for more information.
8301:
8302: @item ARG_POINTER_REGNUM
8303: The register number of the arg pointer register, which is used to
8304: access the function's argument list. On some machines, this is the
8305: same as the frame pointer register. On some machines, the hardware
8306: determines which register this is. On other machines, you can choose
8307: any register you wish for this purpose. If this is not the same
8308: register as the frame pointer register, then you must mark it as a
8309: fixed register according to @code{FIXED_REGISTERS}.
8310:
8311: @item STATIC_CHAIN_REGNUM
8312: The register number used for passing a function's static chain
8313: pointer. This is needed for languages such as Pascal and Algol where
8314: functions defined within other functions can access the local
8315: variables of the outer functions; it is not currently used because C
8316: does not provide this feature, but you must define the macro.
8317:
8318: The static chain register need not be a fixed register.
8319:
8320: @item STRUCT_VALUE_REGNUM
8321: When a function's value's mode is @code{BLKmode}, the value is not
8322: returned according to @code{FUNCTION_VALUE}. Instead, the caller
8323: passes the address of a block of memory in which the value should be
8324: stored.
8325:
8326: If this value is passed in a register, then @code{STRUCT_VALUE_REGNUM}
8327: should be the number of that register.
8328:
8329: @item STRUCT_VALUE
8330: If the structure value address is not passed in a register, define
8331: @code{STRUCT_VALUE} as an expression returning an RTX for the place
1.1.1.8 root 8332: where the address is passed. If it returns a @code{mem} RTX, the
1.1 root 8333: address is passed as an ``invisible'' first argument.
8334:
8335: @item STRUCT_VALUE_INCOMING_REGNUM
8336: On some architectures the place where the structure value address
8337: is found by the called function is not the same place that the
8338: caller put it. This can be due to register windows, or it could
8339: be because the function prologue moves it to a different place.
8340:
8341: If the incoming location of the structure value address is in a
8342: register, define this macro as the register number.
8343:
8344: @item STRUCT_VALUE_INCOMING
8345: If the incoming location is not a register, define
8346: @code{STRUCT_VALUE_INCOMING} as an expression for an RTX for where the
8347: called function should find the value. If it should find the value on
1.1.1.8 root 8348: the stack, define this to create a @code{mem} which refers to the
8349: frame pointer. If the value is a @code{mem}, the compiler assumes it
1.1 root 8350: is for an invisible first argument, and leaves space for it when
8351: finding the first real argument.
8352:
8353: @item REG_ALLOC_ORDER
8354: If defined, an initializer for a vector of integers, containing the
8355: numbers of hard registers in the order in which the GNU CC should
8356: prefer to use them (from most preferred to least).
8357:
8358: If this macro is not defined, registers are used lowest numbered first
8359: (all else being equal).
8360:
8361: One use of this macro is on the 360, where the highest numbered
8362: registers must always be saved and the save-multiple-registers
8363: instruction supports only sequences of consecutive registers. This
8364: macro is defined to cause the highest numbered allocatable registers
8365: to be used first.
8366: @end table
8367:
8368: @node Register Classes, Stack Layout, Registers, Machine Macros
8369: @section Register Classes
8370:
8371: On many machines, the numbered registers are not all equivalent.
8372: For example, certain registers may not be allowed for indexed addressing;
8373: certain registers may not be allowed in some instructions. These machine
8374: restrictions are described to the compiler using @dfn{register classes}.
8375:
8376: You define a number of register classes, giving each one a name and saying
8377: which of the registers belong to it. Then you can specify register classes
8378: that are allowed as operands to particular instruction patterns.
8379:
8380: In general, each register will belong to several classes. In fact, one
8381: class must be named @code{ALL_REGS} and contain all the registers. Another
8382: class must be named @code{NO_REGS} and contain no registers. Often the
8383: union of two classes will be another class; however, this is not required.
8384:
8385: One of the classes must be named @code{GENERAL_REGS}. There is nothing
8386: terribly special about the name, but the operand constraint letters
8387: @samp{r} and @samp{g} specify this class. If @code{GENERAL_REGS} is
8388: the same as @code{ALL_REGS}, just define it as a macro which expands
8389: to @code{ALL_REGS}.
8390:
8391: The way classes other than @code{GENERAL_REGS} are specified in operand
8392: constraints is through machine-dependent operand constraint letters.
8393: You can define such letters to correspond to various classes, then use
8394: them in operand constraints.
8395:
8396: You should define a class for the union of two classes whenever some
8397: instruction allows both classes. For example, if an instruction allows
8398: either a floating-point (coprocessor) register or a general register for a
8399: certain operand, you should define a class @code{FLOAT_OR_GENERAL_REGS}
8400: which includes both of them. Otherwise you will get suboptimal code.
8401:
8402: You must also specify certain redundant information about the register
8403: classes: for each class, which classes contain it and which ones are
8404: contained in it; for each pair of classes, the largest class contained
8405: in their union.
8406:
1.1.1.8 root 8407: When a value occupying several consecutive registers is expected in a
8408: certain class, all the registers used must belong to that class.
8409: Therefore, register classes cannot be used to enforce a requirement for
8410: a register pair to start with an even-numbered register. The way to
8411: specify this requirement is with @code{HARD_REGNO_MODE_OK}.
8412:
1.1 root 8413: Register classes used for input-operands of bitwise-and or shift
8414: instructions have a special requirement: each such class must have, for
8415: each fixed-point machine mode, a subclass whose registers can transfer that
8416: mode to or from memory. For example, on some machines, the operations for
8417: single-byte values (@code{QImode}) are limited to certain registers. When
8418: this is so, each register class that is used in a bitwise-and or shift
8419: instruction must have a subclass consisting of registers from which
8420: single-byte values can be loaded or stored. This is so that
8421: @code{PREFERRED_RELOAD_CLASS} can always have a possible value to return.
8422:
8423: @table @code
8424: @item enum reg_class
8425: An enumeral type that must be defined with all the register class names
8426: as enumeral values. @code{NO_REGS} must be first. @code{ALL_REGS}
8427: must be the last register class, followed by one more enumeral value,
8428: @code{LIM_REG_CLASSES}, which is not a register class but rather
8429: tells how many classes there are.
8430:
8431: Each register class has a number, which is the value of casting
8432: the class name to type @code{int}. The number serves as an index
8433: in many of the tables described below.
8434:
8435: @item N_REG_CLASSES
8436: The number of distinct register classes, defined as follows:
8437:
8438: @example
8439: #define N_REG_CLASSES (int) LIM_REG_CLASSES
8440: @end example
8441:
8442: @item REG_CLASS_NAMES
8443: An initializer containing the names of the register classes as C string
8444: constants. These names are used in writing some of the debugging dumps.
8445:
8446: @item REG_CLASS_CONTENTS
8447: An initializer containing the contents of the register classes, as integers
8448: which are bit masks. The @var{n}th integer specifies the contents of class
8449: @var{n}. The way the integer @var{mask} is interpreted is that
8450: register @var{r} is in the class if @code{@var{mask} & (1 << @var{r})} is 1.
8451:
8452: When the machine has more than 32 registers, an integer does not suffice.
8453: Then the integers are replaced by sub-initializers, braced groupings containing
8454: several integers. Each sub-initializer must be suitable as an initializer
8455: for the type @code{HARD_REG_SET} which is defined in @file{hard-reg-set.h}.
8456:
8457: @item REGNO_REG_CLASS (@var{regno})
8458: A C expression whose value is a register class containing hard register
8459: @var{regno}. In general there is more that one such class; choose a class
8460: which is @dfn{minimal}, meaning that no smaller class also contains the
8461: register.
8462:
8463: @item BASE_REG_CLASS
8464: A macro whose definition is the name of the class to which a valid
8465: base register must belong. A base register is one used in an address
8466: which is the register value plus a displacement.
8467:
8468: @item INDEX_REG_CLASS
8469: A macro whose definition is the name of the class to which a valid
8470: index register must belong. An index register is one used in an
8471: address where its value is either multiplied by a scale factor or
8472: added to another register (as well as added to a displacement).
8473:
8474: @item REG_CLASS_FROM_LETTER (@var{char})
8475: A C expression which defines the machine-dependent operand constraint
8476: letters for register classes. If @var{char} is such a letter, the
8477: value should be the register class corresponding to it. Otherwise,
8478: the value should be @code{NO_REGS}.
8479:
8480: @item REGNO_OK_FOR_BASE_P (@var{num})
8481: A C expression which is nonzero if register number @var{num} is
8482: suitable for use as a base register in operand addresses. It may be
8483: either a suitable hard register or a pseudo register that has been
8484: allocated such a hard register.
8485:
8486: @item REGNO_OK_FOR_INDEX_P (@var{num})
8487: A C expression which is nonzero if register number @var{num} is
8488: suitable for use as an index register in operand addresses. It may be
8489: either a suitable hard register or a pseudo register that has been
8490: allocated such a hard register.
8491:
8492: The difference between an index register and a base register is that
8493: the index register may be scaled. If an address involves the sum of
8494: two registers, neither one of them scaled, then either one may be
8495: labeled the ``base'' and the other the ``index''; but whichever
8496: labeling is used must fit the machine's constraints of which registers
8497: may serve in each capacity. The compiler will try both labelings,
8498: looking for one that is valid, and will reload one or both registers
8499: only if neither labeling works.
8500:
8501: @item PREFERRED_RELOAD_CLASS (@var{x}, @var{class})
8502: A C expression that places additional restrictions on the register class
8503: to use when it is necessary to copy value @var{x} into a register in class
8504: @var{class}. The value is a register class; perhaps @var{class}, or perhaps
8505: another, smaller class. On many machines, the definition
8506:
8507: @example
8508: #define PREFERRED_RELOAD_CLASS(X,CLASS) CLASS
8509: @end example
8510:
8511: @noindent
8512: is safe.
8513:
8514: Sometimes returning a more restrictive class makes better code. For
8515: example, on the 68000, when @var{x} is an integer constant that is in range
8516: for a @samp{moveq} instruction, the value of this macro is always
8517: @code{DATA_REGS} as long as @var{class} includes the data registers.
8518: Requiring a data register guarantees that a @samp{moveq} will be used.
8519:
1.1.1.8 root 8520: If @var{x} is a @code{const_double}, by returning @code{NO_REGS}
1.1 root 8521: you can force @var{x} into a memory constant. This is useful on
8522: certain machines where immediate floating values cannot be loaded into
8523: certain kinds of registers.
8524:
8525: In a shift instruction or a bitwise-and instruction, the mode of @var{x},
8526: the value being reloaded, may not be the same as the mode of the
8527: instruction's operand. (They will both be fixed-point modes, however.) In
8528: such a case, @var{class} may not be a safe value to return. @var{class} is
8529: certainly valid for the instruction, but it may not be valid for reloading
8530: @var{x}. This problem can occur on machines such as the 68000 and 80386
8531: where some registers can handle full-word values but cannot handle
8532: single-byte values.
8533:
8534: On such machines, this macro must examine the mode of @var{x} and return a
8535: subclass of @var{class} which can handle loads and stores of that mode. On
8536: the 68000, where address registers cannot handle @code{QImode}, if @var{x}
8537: has @code{QImode} then you must return @code{DATA_REGS}. If @var{class} is
8538: @code{ADDR_REGS}, then there is no correct value to return; but the shift
8539: and bitwise-and instructions don't use @code{ADDR_REGS}, so this fatal case
8540: never arises.
8541:
8542: @item CLASS_MAX_NREGS (@var{class}, @var{mode})
8543: A C expression for the maximum number of consecutive registers
8544: of class @var{class} needed to hold a value of mode @var{mode}.
8545:
8546: This is closely related to the macro @code{HARD_REGNO_NREGS}.
8547: In fact, the value of the macro @code{CLASS_MAX_NREGS (@var{class}, @var{mode})}
8548: should be the maximum value of @code{HARD_REGNO_NREGS (@var{regno}, @var{mode})}
8549: for all @var{regno} values in the class @var{class}.
8550:
8551: This macro helps control the handling of multiple-word values
8552: in the reload pass.
8553: @end table
8554:
8555: Two other special macros describe which constants fit which constraint
8556: letters.
8557:
8558: @table @code
8559: @item CONST_OK_FOR_LETTER_P (@var{value}, @var{c})
8560: A C expression that defines the machine-dependent operand constraint letters
8561: that specify particular ranges of integer values. If @var{c} is one
8562: of those letters, the expression should check that @var{value}, an integer,
8563: is in the appropriate range and return 1 if so, 0 otherwise. If @var{c} is
8564: not one of those letters, the value should be 0 regardless of @var{value}.
8565:
8566: @item CONST_DOUBLE_OK_FOR_LETTER_P (@var{value}, @var{c})
8567: A C expression that defines the machine-dependent operand constraint
8568: letters that specify particular ranges of floating values. If @var{c} is
8569: one of those letters, the expression should check that @var{value}, an RTX
1.1.1.8 root 8570: of code @code{const_double}, is in the appropriate range and return 1 if
1.1 root 8571: so, 0 otherwise. If @var{c} is not one of those letters, the value should
8572: be 0 regardless of @var{value}.
8573: @end table
8574:
8575: @node Stack Layout, Library Names, Register Classes, Machine Macros
8576: @section Describing Stack Layout
8577:
8578: @table @code
8579: @item STACK_GROWS_DOWNWARD
8580: Define this macro if pushing a word onto the stack moves the stack
8581: pointer to a smaller address.
8582:
8583: When we say, ``define this macro if @dots{},'' it means that the
8584: compiler checks this macro only with @code{#ifdef} so the precise
8585: definition used does not matter.
8586:
8587: @item FRAME_GROWS_DOWNWARD
8588: Define this macro if the addresses of local variable slots are at negative
8589: offsets from the frame pointer.
8590:
8591: @item STARTING_FRAME_OFFSET
8592: Offset from the frame pointer to the first local variable slot to be allocated.
8593:
8594: If @code{FRAME_GROWS_DOWNWARD}, the next slot's offset is found by
8595: subtracting the length of the first slot from @code{STARTING_FRAME_OFFSET}.
8596: Otherwise, it is found by adding the length of the first slot to
8597: the value @code{STARTING_FRAME_OFFSET}.
8598:
8599: @item PUSH_ROUNDING (@var{npushed})
8600: A C expression that is the number of bytes actually pushed onto the
8601: stack when an instruction attempts to push @var{npushed} bytes.
8602:
8603: If the target machine does not have a push instruction, do not define
8604: this macro. That directs GNU CC to use an alternate strategy: to
8605: allocate the entire argument block and then store the arguments into
8606: it.
8607:
8608: On some machines, the definition
8609:
8610: @example
8611: #define PUSH_ROUNDING(BYTES) (BYTES)
8612: @end example
8613:
8614: @noindent
8615: will suffice. But on other machines, instructions that appear
8616: to push one byte actually push two bytes in an attempt to maintain
8617: alignment. Then the definition should be
8618:
8619: @example
8620: #define PUSH_ROUNDING(BYTES) (((BYTES) + 1) & ~1)
8621: @end example
8622:
8623: @item FIRST_PARM_OFFSET (@var{fundecl})
8624: Offset from the argument pointer register to the first argument's
8625: address. On some machines it may depend on the data type of the
8626: function. (In the next version of GNU CC, the argument will be
8627: changed to the function data type rather than its declaration.)
8628:
8629: @item FIRST_PARM_CALLER_OFFSET (@var{fundecl})
8630: Define this macro on machines where register parameters have shadow
8631: locations on the stack, at addresses below the nominal parameter.
8632: This matters because certain arguments cannot be passed on the stack.
8633: On these machines, such arguments must be stored into the shadow
8634: locations.
8635:
8636: This macro should expand into a C expression whose value is the offset
8637: of the first parameter's shadow location from the nominal stack
8638: pointer value. (That value is itself computed by adding the value of
8639: @code{STACK_POINTER_OFFSET} to the stack pointer register.)
8640:
1.1.1.9 root 8641: @item REG_PARM_STACK_SPACE
8642: Define this macro if functions should assume that stack space has been
8643: allocated for arguments even when their values are passed in
8644: registers.
8645:
8646: The actual allocation of such space would be done either by
8647: the call instruction or by the function prologue, or by
1.1.1.10 root 8648: defining @code{FIRST_PARM_CALLER_OFFSET}.
1.1.1.9 root 8649:
1.1.1.6 root 8650: @item STACK_ARGS_ADJUST (@var{size})
8651: Define this macro if the machine requires padding on the stack for
8652: certain function calls. This is padding on a per-function-call basis,
8653: not padding for individual arguments.
8654:
1.1.1.7 root 8655: The argument @var{size} will be a C variable of type @code{struct
8656: arg_data} which contains two fields, an integer named @code{constant}
8657: and an RTX named @code{var}. These together represent a size measured
8658: in bytes which is the sum of the integer and the RTX. Most of the
8659: time @code{var} is 0, which means that the size is simply the integer.
8660:
8661: The definition should be a C statement or compound statement
8662: which alters the variable supplied in whatever way you wish.
8663:
8664: Note that the value you leave in the variable @code{size} will
8665: ultimately be rounded up to a multiple of @code{STACK_BOUNDARY} bits.
8666:
8667: This macro is not fully implemented for machines which have push
8668: instructions (i.e., on which @code{PUSH_ROUNDING} is defined).
1.1.1.6 root 8669:
1.1 root 8670: @item RETURN_POPS_ARGS (@var{funtype})
8671: A C expression that should be 1 if a function pops its own arguments
8672: on returning, or 0 if the function pops no arguments and the caller
8673: must therefore pop them all after the function returns.
8674:
8675: @var{funtype} is a C variable whose value is a tree node that
8676: describes the function in question. Normally it is a node of type
8677: @code{FUNCTION_TYPE} that describes the data type of the function.
8678: From this it is possible to obtain the data types of the value and
8679: arguments (if known).
8680:
8681: When a call to a library function is being considered, @var{funtype}
8682: will contain an identifier node for the library function. Thus, if
8683: you need to distinguish among various library functions, you can do so
8684: by their names. Note that ``library function'' in this context means
8685: a function used to perform arithmetic, whose name is known specially
8686: in the compiler and was not mentioned in the C code being compiled.
8687:
8688: On the Vax, all functions always pop their arguments, so the
8689: definition of this macro is 1. On the 68000, using the standard
8690: calling convention, no functions pop their arguments, so the value of
8691: the macro is always 0 in this case. But an alternative calling
8692: convention is available in which functions that take a fixed number of
8693: arguments pop them but other functions (such as @code{printf}) pop
8694: nothing (the caller pops all). When this convention is in use,
8695: @var{funtype} is examined to determine whether a function takes a
8696: fixed number of arguments.
8697:
1.1.1.10 root 8698: When this macro returns nonzero, the macro @code{FRAME_POINTER_REQUIRED}
8699: must also return nonzero for proper operation.
8700:
1.1 root 8701: @item FUNCTION_VALUE (@var{valtype}, @var{func})
8702: A C expression to create an RTX representing the place where a
8703: function returns a value of data type @var{valtype}. @var{valtype} is
8704: a tree node representing a data type. Write @code{TYPE_MODE
8705: (@var{valtype})} to get the machine mode used to represent that type.
8706: On many machines, only the mode is relevant. (Actually, on most
8707: machines, scalar values are returned in the same place regardless of
8708: mode).@refill
8709:
8710: If the precise function being called is known, @var{func} is a tree
8711: node (@code{FUNCTION_DECL}) for it; otherwise, @var{func} is a null
8712: pointer. This makes it possible to use a different value-returning
8713: convention for specific functions when all their calls are
8714: known.@refill
8715:
8716: @item FUNCTION_OUTGOING_VALUE (@var{valtype}, @var{func})
8717: Define this macro if the target machine has ``register windows''
8718: so that the register in which a function returns its value is not
8719: the same as the one in which the caller sees the value.
8720:
8721: For such machines, @code{FUNCTION_VALUE} computes the register in
8722: which the caller will see the value, and
8723: @code{FUNCTION_OUTGOING_VALUE} should be defined in a similar fashion
8724: to tell the function where to put the value.@refill
8725:
8726: If @code{FUNCTION_OUTGOING_VALUE} is not defined,
8727: @code{FUNCTION_VALUE} serves both purposes.@refill
8728:
1.1.1.7 root 8729: @item RETURN_IN_MEMORY (@var{type})
8730: A C expression which can inhibit the returning of certain function
8731: values in registers, based on the type of value. A nonzero value says
8732: to return the function value in memory, just as large structures are
8733: always returned. Here @var{type} will be a C expression of type
8734: @code{tree}, representing the data type of the value.
8735:
8736: Note that values of mode @code{BLKmode} are returned in memory
8737: regardless of this macro. Also, the option @samp{-fpcc-struct-return}
8738: takes effect regardless of this macro. On most systems, it is
8739: possible to leave the macro undefined; this causes a default
8740: definition to be used, whose value is the constant 0.
8741:
1.1 root 8742: @item LIBCALL_VALUE (@var{mode})
8743: A C expression to create an RTX representing the place where a library
8744: function returns a value of mode @var{mode}. If the precise function
8745: being called is known, @var{func} is a tree node
8746: (@code{FUNCTION_DECL}) for it; otherwise, @var{func} is a null
8747: pointer. This makes it possible to use a different value-returning
8748: convention for specific functions when all their calls are
8749: known.@refill
8750:
8751: Note that ``library function'' in this context means a compiler
8752: support routine, used to perform arithmetic, whose name is known
8753: specially by the compiler and was not mentioned in the C code being
8754: compiled.
8755:
8756: @item FUNCTION_VALUE_REGNO_P (@var{regno})
8757: A C expression that is nonzero if @var{regno} is the number of a hard
8758: register in which the values of called function may come back.
8759:
8760: A register whose use for returning values is limited to serving as the
8761: second of a pair (for a value of type @code{double}, say) need not be
8762: recognized by this macro. So for most machines, this definition
8763: suffices:
8764:
8765: @example
8766: #define FUNCTION_VALUE_REGNO_P(N) ((N) == 0)
8767: @end example
8768:
8769: If the machine has register windows, so that the caller and the called
8770: function use different registers for the return value, this macro
8771: should recognize only the caller's register numbers.
8772:
8773: @item FUNCTION_ARG (@var{cum}, @var{mode}, @var{type}, @var{named})
8774: A C expression that controls whether a function argument is passed
8775: in a register, and which register.
8776:
8777: The arguments are @var{cum}, which summarizes all the previous
8778: arguments; @var{mode}, the machine mode of the argument; @var{type},
8779: the data type of the argument as a tree node or 0 if that is not known
8780: (which happens for C support library functions); and @var{named},
8781: which is 1 for an ordinary argument and 0 for nameless arguments that
1.1.1.8 root 8782: correspond to @samp{@dots{}} in the called function's prototype.
1.1 root 8783:
1.1.1.8 root 8784: The value of the expression should either be a @code{reg} RTX for the
1.1 root 8785: hard register in which to pass the argument, or zero to pass the
8786: argument on the stack.
8787:
8788: For the Vax and 68000, where normally all arguments are pushed, zero
8789: suffices as a definition.
8790:
1.1.1.8 root 8791: The usual way to make the ANSI library @file{stdarg.h} work on a machine
8792: where some arguments are usually passed in registers, is to cause
8793: nameless arguments to be passed on the stack instead. This is done
8794: by making @code{FUNCTION_ARG} return 0 whenever @var{named} is 0.
8795:
1.1 root 8796: @item FUNCTION_INCOMING_ARG (@var{cum}, @var{mode}, @var{type}, @var{named})
8797: Define this macro if the target machine has ``register windows'', so
8798: that the register in which a function sees an arguments is not
8799: necessarily the same as the one in which the caller passed the
8800: argument.
8801:
8802: For such machines, @code{FUNCTION_ARG} computes the register in which
8803: the caller passes the value, and @code{FUNCTION_INCOMING_ARG} should
8804: be defined in a similar fashion to tell the function being called
8805: where the arguments will arrive.
8806:
8807: If @code{FUNCTION_INCOMING_ARG} is not defined, @code{FUNCTION_ARG}
8808: serves both purposes.@refill
8809:
8810: @item FUNCTION_ARG_PARTIAL_NREGS (@var{cum}, @var{mode}, @var{type}, @var{named})
8811: A C expression for the number of words, at the beginning of an
8812: argument, must be put in registers. The value must be zero for
8813: arguments that are passed entirely in registers or that are entirely
8814: pushed on the stack.
8815:
8816: On some machines, certain arguments must be passed partially in
8817: registers and partially in memory. On these machines, typically the
8818: first @var{n} words of arguments are passed in registers, and the rest
8819: on the stack. If a multi-word argument (a @code{double} or a
8820: structure) crosses that boundary, its first few words must be passed
8821: in registers and the rest must be pushed. This macro tells the
8822: compiler when this occurs, and how many of the words should go in
8823: registers.
8824:
8825: @code{FUNCTION_ARG} for these arguments should return the first
8826: register to be used by the caller for this argument; likewise
8827: @code{FUNCTION_INCOMING_ARG}, for the called function.
8828:
8829: @item CUMULATIVE_ARGS
8830: A C type for declaring a variable that is used as the first argument
8831: of @code{FUNCTION_ARG} and other related values. For some target
8832: machines, the type @code{int} suffices and can hold the number of
8833: bytes of argument so far.
8834:
8835: @item INIT_CUMULATIVE_ARGS (@var{cum}, @var{fntype})
8836: A C statement (sans semicolon) for initializing the variable @var{cum}
8837: for the state at the beginning of the argument list. The variable has
8838: type @code{CUMULATIVE_ARGS}. The value of @var{fntype} is the tree node
8839: for the data type of the function which will receive the args, or 0
8840: if the args are to a compiler support library function.
8841:
8842: @item FUNCTION_ARG_ADVANCE (@var{cum}, @var{mode}, @var{type}, @var{named})
1.1.1.7 root 8843: A C statement (sans semicolon) to update the summarizer variable
8844: @var{cum} to advance past an argument in the argument list. The
8845: values @var{mode}, @var{type} and @var{named} describe that argument.
8846: Once this is done, the variable @var{cum} is suitable for analyzing
8847: the @emph{following} argument with @code{FUNCTION_ARG}, etc.@refill
1.1 root 8848:
8849: @item FUNCTION_ARG_REGNO_P (@var{regno})
8850: A C expression that is nonzero if @var{regno} is the number of a hard
8851: register in which function arguments are sometimes passed. This does
8852: @emph{not} include implicit arguments such as the static chain and
8853: the structure-value address. On many machines, no registers can be
8854: used for this purpose since all function arguments are pushed on the
8855: stack.
8856:
8857: @item FUNCTION_ARG_PADDING (@var{mode}, @var{size})
8858: If defined, a C expression which determines whether, and in which direction,
8859: to pad out an argument with extra space. The value should be of type
8860: @code{enum direction}: either @code{upward} to pad above the argument,
8861: @code{downward} to pad below, or @code{none} to inhibit padding.
8862:
8863: The argument @var{size} is an RTX which describes the size of the
8864: argument, in bytes. It should be used only if @var{mode} is
8865: @code{BLKmode}. Otherwise, @var{size} is 0.
8866:
8867: This macro does not control the @emph{amount} of padding; that is
8868: always just enough to reach the next multiple of @code{PARM_BOUNDARY}.
8869:
8870: This macro has a default definition which is right for most systems.
8871: For little-endian machines, the default is to pad upward. For
8872: big-endian machines, the default is to pad downward for an argument of
8873: constant size shorter than an @code{int}, and upward otherwise.
8874:
8875: @item FUNCTION_PROLOGUE (@var{file}, @var{size})
8876: A C compound statement that outputs the assembler code for entry to a
8877: function. The prologue is responsible for setting up the stack frame,
8878: initializing the frame pointer register, saving registers that must be
8879: saved, and allocating @var{size} additional bytes of storage for the
8880: local variables. @var{size} is an integer. @var{file} is a stdio
8881: stream to which the assembler code should be output.
8882:
8883: The label for the beginning of the function need not be output by this
8884: macro. That has already been done when the macro is run.
8885:
8886: To determine which registers to save, the macro can refer to the array
8887: @code{regs_ever_live}: element @var{r} is nonzero if hard register
8888: @var{r} is used anywhere within the function. This implies the
8889: function prologue should save register @var{r}, but not if it is one
8890: of the call-used registers.
8891:
8892: On machines where functions may or may not have frame-pointers, the
8893: function entry code must vary accordingly; it must set up the frame
8894: pointer if one is wanted, and not otherwise. To determine whether a
8895: frame pointer is in wanted, the macro can refer to the variable
8896: @code{frame_pointer_needed}. The variable's value will be 1 at run
8897: time in a function that needs a frame pointer.
8898:
1.1.1.10 root 8899: On machines where an argument may be passed partly in registers and
8900: partly in memory, this macro must examine the variable
1.1.1.8 root 8901: @code{current_function_pretend_args_size}, and allocate that many bytes
8902: of uninitialized space on the stack just underneath the first argument
8903: arriving on the stack. (This may not be at the very end of the stack,
8904: if the calling sequence has pushed anything else since pushing the stack
8905: arguments. But usually, on such machines, nothing else has been pushed
8906: yet, because the function prologue itself does all the pushing.)
8907:
1.1 root 8908: @item FUNCTION_PROFILER (@var{file}, @var{labelno})
8909: A C statement or compound statement to output to @var{file} some
8910: assembler code to call the profiling subroutine @code{mcount}.
8911: Before calling, the assembler code must load the address of a
8912: counter variable into a register where @code{mcount} expects to
8913: find the address. The name of this variable is @samp{LP} followed
8914: by the number @var{labelno}, so you would generate the name using
8915: @samp{LP%d} in a @code{fprintf}.
8916:
8917: The details of how the address should be passed to @code{mcount} are
8918: determined by your operating system environment, not by GNU CC. To
8919: figure them out, compile a small program for profiling using the
8920: system's installed C compiler and look at the assembler code that
8921: results.
8922:
1.1.1.6 root 8923: @item FUNCTION_BLOCK_PROFILER (@var{file}, @var{labelno})
8924: A C statement or compound statement to output to @var{file} some
8925: assembler code to initialize basic-block profiling for the current
8926: object module. This code should call the subroutine
8927: @code{__bb_init_func} once per object module, passing it as its sole
8928: argument the address of a block allocated in the object module.
8929:
8930: The name of the block is a local symbol made with this statement:
8931:
8932: @example
8933: ASM_GENERATE_INTERNAL_LABEL (@var{buffer}, "LPBX", 0);
8934: @end example
8935:
8936: Of course, since you are writing the definition of
8937: @code{ASM_GENERATE_INTERNAL_LABEL} as well as that of this macro, you
8938: can take a short cut in the definition of this macro and use the name
8939: that you know will result.
8940:
8941: The first word of this block is a flag which will be nonzero if the
8942: object module has already been initialized. So test this word first,
8943: and do not call @code{__bb_init_func} if the flag is nonzero.
8944:
8945: @item BLOCK_PROFILER (@var{file}, @var{blockno})
8946: A C statement or compound statement to increment the count associated
8947: with the basic block number @var{blockno}. Basic blocks are numbered
8948: separately from zero within each compilation. The count associated
8949: with block number @var{blockno} is at index @var{blockno} in a vector
8950: of words; the name of this array is a local symbol made with this
8951: statement:
8952:
8953: @example
8954: ASM_GENERATE_INTERNAL_LABEL (@var{buffer}, "LPBX", 2);
8955: @end example
8956:
8957: Of course, since you are writing the definition of
8958: @code{ASM_GENERATE_INTERNAL_LABEL} as well as that of this macro, you
8959: can take a short cut in the definition of this macro and use the name
8960: that you know will result.
8961:
1.1.1.10 root 8962: @item EXIT_IGNORE_STACK
1.1 root 8963: Define this macro as a C expression that is nonzero if the return
8964: instruction or the function epilogue ignores the value of the stack
8965: pointer; in other words, if it is safe to delete an instruction to
8966: adjust the stack pointer before a return from the function.
8967:
1.1.1.8 root 8968: Note that this macro's value is relevant only for functions for which
8969: frame pointers are maintained. It is never safe to delete a final
8970: stack adjustment in a function that has no frame pointer, and the
1.1.1.10 root 8971: compiler knows this regardless of @code{EXIT_IGNORE_STACK}.
1.1 root 8972:
8973: @item FUNCTION_EPILOGUE (@var{file}, @var{size})
8974: A C compound statement that outputs the assembler code for exit from a
8975: function. The epilogue is responsible for restoring the saved
8976: registers and stack pointer to their values when the function was
8977: called, and returning control to the caller. This macro takes the
8978: same arguments as the macro @code{FUNCTION_PROLOGUE}, and the
8979: registers to restore are determined from @code{regs_ever_live} and
8980: @code{CALL_USED_REGISTERS} in the same way.
8981:
8982: On some machines, there is a single instruction that does all the work
8983: of returning from the function. On these machines, give that
8984: instruction the name @samp{return} and do not define the macro
8985: @code{FUNCTION_EPILOGUE} at all.
8986:
8987: Do not define a pattern named @samp{return} if you want the
8988: @code{FUNCTION_EPILOGUE} to be used. If you want the target switches
8989: to control whether return instructions or epilogues are used, define a
8990: @samp{return} pattern with a validity condition that tests the target
8991: switches appropriately. If the @samp{return} pattern's validity
8992: condition is false, epilogues will be used.
8993:
8994: On machines where functions may or may not have frame-pointers, the
8995: function exit code must vary accordingly. Sometimes the code for
8996: these two cases is completely different. To determine whether a frame
8997: pointer is in wanted, the macro can refer to the variable
8998: @code{frame_pointer_needed}. The variable's value will be 1 at run
8999: time in a function that needs a frame pointer.
9000:
9001: On some machines, some functions pop their arguments on exit while
9002: others leave that for the caller to do. For example, the 68020 when
9003: given @samp{-mrtd} pops arguments in functions that take a fixed
9004: number of arguments.
9005:
9006: Your definition of the macro @code{RETURN_POPS_ARGS} decides which
9007: functions pop their own arguments. @code{FUNCTION_EPILOGUE} needs to
9008: know what was decided. The variable @code{current_function_pops_args}
9009: is nonzero if the function should pop its own arguments. If so, use
9010: the variable @code{current_function_args_size} as the number of bytes
9011: to pop.
9012:
9013: @item FIX_FRAME_POINTER_ADDRESS (@var{addr}, @var{depth})
9014: A C compound statement to alter a memory address that uses the frame
9015: pointer register so that it uses the stack pointer register instead.
9016: This must be done in the instructions that load parameter values into
9017: registers, when the reload pass determines that a frame pointer is not
9018: necessary for the function. @var{addr} will be a C variable name, and
9019: the updated address should be stored in that variable. @var{depth}
9020: will be the current depth of stack temporaries (number of bytes of
9021: arguments currently pushed). The change in offset between a
9022: frame-pointer-relative address and a stack-pointer-relative address
9023: must include @var{depth}.
9024:
9025: Even if your machine description specifies there will always be a
9026: frame pointer in the frame pointer register, you must still define
9027: @code{FIX_FRAME_POINTER_ADDRESS}, but the definition will never be
9028: executed at run time, so it may be empty.
1.1.1.8 root 9029:
9030: @item LONGJMP_RESTORE_FROM_STACK
9031: Define this macro if the @code{longjmp} function restores registers
9032: from the stack frames, rather than from those saved specifically by
9033: @code{setjmp}. Certain quantities must not be kept in registers
9034: across a call to @code{setjmp} on such machines.
1.1 root 9035: @end table
9036:
9037: @node Library Names, Addressing Modes, Stack Layout, Machine Macros
9038: @section Library Subroutine Names
9039:
9040: @table @code
1.1.1.5 root 9041: @item MULSI3_LIBCALL
9042: A C string constant giving the name of the function to call for
9043: multiplication of one signed full-word by another. If you do not
9044: define this macro, the default name is used, which is @code{__mulsi3},
9045: a function defined in @file{gnulib}.
9046:
9047: @item UMULSI3_LIBCALL
9048: A C string constant giving the name of the function to call for
9049: multiplication of one unsigned full-word by another. If you do not
9050: define this macro, the default name is used, which is
9051: @code{__umulsi3}, a function defined in @file{gnulib}.
9052:
9053: @item DIVSI3_LIBCALL
9054: A C string constant giving the name of the function to call for
9055: division of one signed full-word by another. If you do not define
9056: this macro, the default name is used, which is @code{__divsi3}, a
9057: function defined in @file{gnulib}.
9058:
1.1 root 9059: @item UDIVSI3_LIBCALL
9060: A C string constant giving the name of the function to call for
1.1.1.5 root 9061: division of one unsigned full-word by another. If you do not define
9062: this macro, the default name is used, which is @code{__udivsi3}, a
9063: function defined in @file{gnulib}.
9064:
9065: @item MODSI3_LIBCALL
9066: A C string constant giving the name of the function to call for the
9067: remainder in division of one signed full-word by another. If you do
9068: not define this macro, the default name is used, which is
9069: @code{__modsi3}, a function defined in @file{gnulib}.
1.1 root 9070:
9071: @item UMODSI3_LIBCALL
9072: A C string constant giving the name of the function to call for the
1.1.1.5 root 9073: remainder in division of one unsigned full-word by another. If you do
9074: not define this macro, the default name is used, which is
9075: @code{__umodsi3}, a function defined in @file{gnulib}.
1.1 root 9076:
9077: @item TARGET_MEM_FUNCTIONS
9078: Define this macro if GNU CC should generate calls to the System V
9079: (and ANSI C) library functions @code{memcpy} and @code{memset}
9080: rather than the BSD functions @code{bcopy} and @code{bzero}.
9081: @end table
9082:
1.1.1.8 root 9083: @node Addressing Modes, Delayed Branch, Library Names, Machine Macros
1.1 root 9084: @section Addressing Modes
9085:
9086: @table @code
9087: @item HAVE_POST_INCREMENT
9088: Define this macro if the machine supports post-increment addressing.
9089:
9090: @item HAVE_PRE_INCREMENT
9091: @itemx HAVE_POST_DECREMENT
9092: @itemx HAVE_PRE_DECREMENT
9093: Similar for other kinds of addressing.
9094:
9095: @item CONSTANT_ADDRESS_P (@var{x})
9096: A C expression that is 1 if the RTX @var{x} is a constant whose value
9097: is an integer. This includes integers whose values are not explicitly
1.1.1.8 root 9098: known, such as @code{symbol_ref} and @code{label_ref} expressions and
9099: @code{const} arithmetic expressions.
1.1 root 9100:
9101: On most machines, this can be defined as @code{CONSTANT_P (@var{x})},
9102: but a few machines are more restrictive in which constant addresses
9103: are supported.
9104:
9105: @item MAX_REGS_PER_ADDRESS
9106: A number, the maximum number of registers that can appear in a valid
1.1.1.10 root 9107: memory address. Note that it is up to you to specify a value equal to
9108: the maximum number that @code{go_if_legitimate_address} would ever
9109: accept.
1.1 root 9110:
9111: @item GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{label})
9112: A C compound statement with a conditional @code{goto @var{label};}
9113: executed if @var{x} (an RTX) is a legitimate memory address on the
9114: target machine for a memory operand of mode @var{mode}.
9115:
9116: It usually pays to define several simpler macros to serve as
9117: subroutines for this one. Otherwise it may be too complicated to
9118: understand.
9119:
9120: This macro must exist in two variants: a strict variant and a
9121: non-strict one. The strict variant is used in the reload pass. It
9122: must be defined so that any pseudo-register that has not been
9123: allocated a hard register is considered a memory reference. In
9124: contexts where some kind of register is required, a pseudo-register
9125: with no hard register must be rejected.
9126:
9127: The non-strict variant is used in other passes. It must be defined to
9128: accept all pseudo-registers in every context where some kind of
9129: register is required.
9130:
9131: Compiler source files that want to use the strict variant of this
9132: macro define the macro @code{REG_OK_STRICT}. You should use an
9133: @code{#ifdef REG_OK_STRICT} conditional to define the strict variant
9134: in that case and the non-strict variant otherwise.
9135:
9136: Typically among the subroutines used to define
9137: @code{GO_IF_LEGITIMATE_ADDRESS} are subroutines to check for
9138: acceptable registers for various purposes (one for base registers, one
9139: for index registers, and so on). Then only these subroutine macros
9140: need have two variants; the higher levels of macros may be the same
9141: whether strict or not.@refill
9142:
1.1.1.8 root 9143: Normally, constant addresses which are the sum of a @code{symbol_ref}
9144: and an integer are stored inside a @code{const} RTX to mark them as
9145: constant. Therefore, there is no need to recognize such sums as
9146: legitimate addresses.
9147:
9148: Usually @code{PRINT_OPERAND_ADDRESS} is not prepared to handle constant
9149: sums that are not marked with @code{const}. It assumes that a naked
9150: @code{plus} indicates indexing. If so, then you @emph{must} reject such
9151: naked constant sums as illegitimate addresses, so that none of them will
9152: be given to @code{PRINT_OPERAND_ADDRESS}.@refill
9153:
1.1 root 9154: @item REG_OK_FOR_BASE_P (@var{x})
1.1.1.5 root 9155: A C expression that is nonzero if @var{x} (assumed to be a @code{reg}
1.1 root 9156: RTX) is valid for use as a base register. For hard registers, it
9157: should always accept those which the hardware permits and reject the
9158: others. Whether the macro accepts or rejects pseudo registers must be
9159: controlled by @code{REG_OK_STRICT} as described above. This usually
9160: requires two variant definitions, of which @code{REG_OK_STRICT}
9161: controls the one actually used.
9162:
9163: @item REG_OK_FOR_INDEX_P (@var{x})
1.1.1.5 root 9164: A C expression that is nonzero if @var{x} (assumed to be a @code{reg}
1.1 root 9165: RTX) is valid for use as an index register.
9166:
9167: The difference between an index register and a base register is that
9168: the index register may be scaled. If an address involves the sum of
9169: two registers, neither one of them scaled, then either one may be
9170: labeled the ``base'' and the other the ``index''; but whichever
9171: labeling is used must fit the machine's constraints of which registers
9172: may serve in each capacity. The compiler will try both labelings,
9173: looking for one that is valid, and will reload one or both registers
9174: only if neither labeling works.
9175:
9176: @item LEGITIMIZE_ADDRESS (@var{x}, @var{oldx}, @var{mode}, @var{win})
9177: A C compound statement that attempts to replace @var{x} with a valid
9178: memory address for an operand of mode @var{mode}. @var{win} will be a
9179: C statement label elsewhere in the code; the macro definition may use
9180:
9181: @example
9182: GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{win});
9183: @end example
9184:
9185: @noindent
9186: to avoid further processing if the address has become legitimate.
9187:
9188: @var{x} will always be the result of a call to @code{break_out_memory_refs},
9189: and @var{oldx} will be the operand that was given to that function to produce
9190: @var{x}.
9191:
9192: The code generated by this macro should not alter the substructure of
9193: @var{x}. If it transforms @var{x} into a more legitimate form, it
9194: should assign @var{x} (which will always be a C variable) a new value.
9195:
9196: It is not necessary for this macro to come up with a legitimate
9197: address. The compiler has standard ways of doing so in all cases. In
9198: fact, it is safe for this macro to do nothing. But often a
9199: machine-dependent strategy can generate better code.
9200:
9201: @item GO_IF_MODE_DEPENDENT_ADDRESS (@var{addr}, @var{label})
9202: A C statement or compound statement with a conditional @code{goto
9203: @var{label};} executed if memory address @var{x} (an RTX) can have
9204: different meanings depending on the machine mode of the memory
9205: reference it is used for.
9206:
9207: Autoincrement and autodecrement addresses typically have mode-dependent
9208: effects because the amount of the increment or decrement is the size
9209: of the operand being addressed. Some machines have other mode-dependent
9210: addresses. Many RISC machines have no mode-dependent addresses.
9211:
9212: You may assume that @var{addr} is a valid address for the machine.
9213:
9214: @item LEGITIMATE_CONSTANT_P (@var{x})
9215: A C expression that is nonzero if @var{x} is a legitimate constant for
9216: an immediate operand on the target machine. You can assume that
1.1.1.8 root 9217: either @var{x} is a @code{const_double} or it satisfies
1.1 root 9218: @code{CONSTANT_P}, so you need not check these things. In fact,
9219: @samp{1} is a suitable definition for this macro on machines where any
1.1.1.8 root 9220: @code{const_double} is valid and anything @code{CONSTANT_P} is valid.@refill
9221: @end table
9222:
9223: @node Delayed Branch, Condition Code, Addressing Modes, Machine Macros
9224: @section Parameters for Delayed Branch Optimization
9225:
9226: @table @code
9227: @item HAVE_DELAYED_BRANCH
9228: Define this macro if the target machine has delayed branches, that is,
9229: a branch does not take effect immediately, and the actual branch
9230: instruction may be followed by one or more instructions that will be
9231: issued before the PC is actually changed.
9232:
9233: If defined, this allows a special scheduling pass to be run after the
9234: second jump optimization to attempt to reorder instructions to exploit
9235: this. Defining this macro also requires the definition of certain
9236: other macros described below.
9237:
9238: @item DBR_SLOTS_AFTER (@var{insn})
9239: This macro must be defined if @code{HAVE_DELAYED_BRANCH} is defined.
9240: Its definition should be a C expression returning the number of
9241: available delay slots following the instruction(s) output by the
9242: pattern for @var{insn}. The definition of ``slot'' is
9243: machine-dependent, and may denote instructions, bytes, or whatever.
9244:
9245: @item DBR_INSN_SLOTS (@var{insn})
9246: This macro must be defined if @code{HAVE_DELAYED_BRANCH} is defined.
9247: It should be a C expression returning the number of slots (typically
9248: the number of machine instructions) consumed by @var{insn}.
9249:
9250: You may assume that @var{insn} is truly an insn, not a note, label,
9251: barrier, dispatch table, @code{use}, or @code{clobber}.
9252:
9253: @item DBR_INSN_ELIGIBLE_P (@var{insn}, @var{dinsn})
9254: A C expression whose value is non-zero if it is legitimate to put
9255: @var{insn} in the delay slot following @var{dinsn}.
9256:
9257: You do not need to take account of data flow considerations in the
9258: definition of this macro, because the delayed branch optimizer always
9259: does that. This macro is needed only when certain insns may not be
9260: placed in certain delay slots for reasons not evident from the RTL
9261: expressions themselves. If there are no such problems, you don't need
9262: to define this macro.
9263:
9264: You may assume that @var{insn} is truly an insn, not a note, label,
9265: barrier, dispatch table, @code{use}, or @code{clobber}. You may
9266: assume that @var{dinsn} is a jump insn with a delay slot.
9267:
9268: @item DBR_OUTPUT_SEQEND(@var{file})
9269: A C statement, to be executed after all slot-filler instructions have
9270: been output. If necessary, call @code{dbr_sequence_length} to
9271: determine the number of slots filled in a sequence (zero if not
9272: currently outputting a sequence), to decide how many no-ops to output,
9273: or whatever.
9274:
9275: Don't define this macro if it has nothing to do, but it is helpful in
9276: reading assembly output if the extent of the delay sequence is made
9277: explicit (e.g. with white space).
9278:
9279: Note that output routines for instructions with delay slots must be
9280: prepared to deal with not being output as part of a sequence (i.e.
9281: when the scheduling pass is not run, or when no slot fillers could be
9282: found.) The variable @code{final_sequence} is null when not
9283: processing a sequence, otherwise it contains the @code{sequence} rtx
9284: being output.
1.1 root 9285: @end table
9286:
1.1.1.9 root 9287: @node Condition Code, Cross-compilation, Delayed Branch, Machine Macros
1.1.1.8 root 9288: @section Condition Code Information
9289:
9290: The file @file{conditions.h} defines a variable @code{cc_status} to
9291: describe how the condition code was computed (in case the interpretation of
9292: the condition code depends on the instruction that it was set by). This
9293: variable contains the RTL expressions on which the condition code is
9294: currently based, and several standard flags.
9295:
9296: Sometimes additional machine-specific flags must be defined in the machine
9297: description header file. It can also add additional machine-specific
9298: information by defining @code{CC_STATUS_MDEP}.
9299:
9300: @table @code
9301: @item CC_STATUS_MDEP
9302: C code for a data type which is used for declaring the @code{mdep}
9303: component of @code{cc_status}. It defaults to @code{int}.
9304:
9305: @item CC_STATUS_MDEP_INIT
1.1.1.9 root 9306: A C expression to initialize the @code{mdep} field to ``empty''.
9307: The default definition does nothing, since most machines don't use
9308: the field anyway. If you want to use the field, you should probably
9309: define this macro to initialize it.
1.1.1.8 root 9310:
9311: @item NOTICE_UPDATE_CC (@var{exp}, @var{insn})
9312: A C compound statement to set the components of @code{cc_status}
9313: appropriately for an insn @var{insn} whose body is @var{exp}. It is
9314: this macro's responsibility to recognize insns that set the condition
9315: code as a byproduct of other activity as well as those that explicitly
9316: set @code{(cc0)}.
9317:
9318: If there are insn that do not set the condition code but do alter
9319: other machine registers, this macro must check to see whether they
9320: invalidate the expressions that the condition code is recorded as
9321: reflecting. For example, on the 68000, insns that store in address
9322: registers do not set the condition code, which means that usually
9323: @code{NOTICE_UPDATE_CC} can leave @code{cc_status} unaltered for such
9324: insns. But suppose that the previous insn set the condition code
9325: based on location @samp{a4@@(102)} and the current insn stores a new
9326: value in @samp{a4}. Although the condition code is not changed by
9327: this, it will no longer be true that it reflects the contents of
9328: @samp{a4@@(102)}. Therefore, @code{NOTICE_UPDATE_CC} must alter
9329: @code{cc_status} in this case to say that nothing is known about the
9330: condition code value.
9331:
9332: The definition of @code{NOTICE_UPDATE_CC} must be prepared to deal
9333: with the results of peephole optimization: insns whose patterns are
9334: @code{parallel} RTXs containing various @code{reg}, @code{mem} or
9335: constants which are just the operands. The RTL structure of these
9336: insns is not sufficient to indicate what the insns actually do. What
9337: @code{NOTICE_UPDATE_CC} should do when it sees one is just to run
9338: @code{CC_STATUS_INIT}.
9339: @end table
9340:
9341: @node Cross-compilation, Misc, Condition Code, Machine Macros
1.1.1.5 root 9342: @section Cross Compilation and Floating-Point Format
9343:
1.1.1.9 root 9344: While all modern machines use 2's complement representation for integers,
1.1.1.5 root 9345: there are a variety of representations for floating point numbers. This
9346: means that in a cross-compiler the representation of floating point numbers
9347: in the compiled program may be different from that used in the machine
9348: doing the compilation.
9349:
9350: Because different representation systems may offer different amounts of
9351: range and precision, the cross compiler cannot safely use the host
9352: machine's floating point arithmetic. Therefore, floating point constants
9353: must be represented in the target machine's format. This means that the
9354: cross compiler cannot use @code{atof} to parse a floating point constant;
9355: it must have its own special routine to use instead. Also, constant
9356: folding must emulate the target machine's arithmetic (or must not be done
9357: at all).
9358:
9359: The macros in the following table should be defined only if you are cross
9360: compiling between different floating point formats.
9361:
9362: Otherwise, don't define them. Then default definitions will be set up which
9363: use @code{double} as the data type, @code{==} to test for equality, etc.
9364:
9365: You don't need to worry about how many times you use an operand of any
9366: of these macros. The compiler never uses operands which have side effects.
9367:
9368: @table @code
9369: @item REAL_VALUE_TYPE
9370: A macro for the C data type to be used to hold a floating point value
9371: in the target machine's format. Typically this would be a
9372: @code{struct} containing an array of @code{int}.
9373:
9374: @item REAL_VALUES_EQUAL (@var{x}, @var{y})
9375: A macro for a C expression which compares for equality the two values,
9376: @var{x} and @var{y}, both of type @code{REAL_VALUE_TYPE}.
9377:
9378: @item REAL_VALUES_LESS (@var{x}, @var{y})
9379: A macro for a C expression which tests whether @var{x} is less than
9380: @var{y}, both values being of type @code{REAL_VALUE_TYPE} and
9381: interpreted as floating point numbers in the target machine's
9382: representation.
9383:
9384: @item REAL_VALUE_LDEXP (@var{x}, @var{scale})
9385: A macro for a C expression which performs the standard library
9386: function @code{ldexp}, but using the target machine's floating point
9387: representation. Both @var{x} and the value of the expression have
9388: type @code{REAL_VALUE_TYPE}. The second argument, @var{scale}, is an
9389: integer.
9390:
9391: @item REAL_VALUE_ATOF (@var{string})
9392: A macro for a C expression which converts @var{string}, an expression
9393: of type @code{char *}, into a floating point number in the target
9394: machine's representation. The value has type @code{REAL_VALUE_TYPE}.
9395: @end table
9396:
9397: Define the following additional macros if you want to make floating
9398: point constant folding work while cross compiling. If you don't
9399: define them, cross compilation is still possible, but constant folding
9400: will not happen for floating point values.
9401:
9402: @table @code
9403: @item REAL_ARITHMETIC (@var{output}, @var{code}, @var{x}, @var{y})
9404: A macro for a C statement which calculates an arithmetic operation of
9405: the two floating point values @var{x} and @var{y}, both of type
9406: @code{REAL_VALUE_TYPE} in the target machine's representation, to
9407: produce a result of the same type and representation which is stored
9408: in @var{output} (which will be a variable).
9409:
9410: The operation to be performed is specified by @var{code}, a tree code
9411: which will always be one of the following: @code{PLUS_EXPR},
9412: @code{MINUS_EXPR}, @code{MULT_EXPR}, @code{RDIV_EXPR},
9413: @code{MAX_EXPR}, @code{MIN_EXPR}.@refill
9414:
9415: The expansion of this macro is responsible for checking for overflow.
9416: If overflow happens, the macro expansion should execute the statement
9417: @code{return 0;}, which indicates the inability to perform the
9418: arithmetic operation requested.
9419:
9420: @item REAL_VALUE_NEGATE (@var{x})
9421: A macro for a C expression which returns the negative of the floating
9422: point value @var{x}. Both @var{x} and the value of the expression
9423: have type @code{REAL_VALUE_TYPE} and are in the target machine's
9424: floating point representation.
9425:
9426: There is no way for this macro to report overflow, since overflow
9427: can't happen in the negation operation.
9428:
9429: @item REAL_VALUE_TO_INT (@var{low}, @var{high}, @var{x})
9430: A macro for a C expression which converts a floating point value
9431: @var{x} into a double-precision integer which is then stored into
9432: @var{low} and @var{high}, two variables of type @var{int}.
9433:
9434: @item REAL_VALUE_FROM_INT (@var{x}, @var{low}, @var{high})
9435: A macro for a C expression which converts a double-precision integer
9436: found in @var{low} and @var{high}, two variables of type @var{int},
9437: into a floating point value which is then stored into @var{x}.
9438: @end table
9439:
1.1.1.8 root 9440: @node Misc, Assembler Format, Cross-compilation, Machine Macros
1.1 root 9441: @section Miscellaneous Parameters
9442:
9443: @table @code
9444: @item CASE_VECTOR_MODE
9445: An alias for a machine mode name. This is the machine mode that
9446: elements of a jump-table should have.
9447:
9448: @item CASE_VECTOR_PC_RELATIVE
9449: Define this macro if jump-tables should contain relative addresses.
9450:
9451: @item CASE_DROPS_THROUGH
9452: Define this if control falls through a @code{case} insn when the index
9453: value is out of range. This means the specified default-label is
9454: actually ignored by the @code{case} insn proper.
9455:
9456: @item IMPLICIT_FIX_EXPR
9457: An alias for a tree code that should be used by default for conversion
9458: of floating point values to fixed point. Normally,
9459: @code{FIX_ROUND_EXPR} is used.@refill
9460:
9461: @item FIXUNS_TRUNC_LIKE_FIX_TRUNC
9462: Define this macro if the same instructions that convert a floating
9463: point number to a signed fixed point number also convert validly to an
9464: unsigned one.
9465:
9466: @item EASY_DIV_EXPR
9467: An alias for a tree code that is the easiest kind of division to
9468: compile code for in the general case. It may be
9469: @code{TRUNC_DIV_EXPR}, @code{FLOOR_DIV_EXPR}, @code{CEIL_DIV_EXPR} or
9470: @code{ROUND_DIV_EXPR}. These four division operators differ in how
9471: they round the result to an integer. @code{EASY_DIV_EXPR} is used
9472: when it is permissible to use any of those kinds of division and the
9473: choice should be made on the basis of efficiency.@refill
9474:
9475: @item DEFAULT_SIGNED_CHAR
9476: An expression whose value is 1 or 0, according to whether the type
9477: @code{char} should be signed or unsigned by default. The user can
9478: always override this default with the options @samp{-fsigned-char}
9479: and @samp{-funsigned-char}.
9480:
9481: @item SCCS_DIRECTIVE
9482: Define this if the preprocessor should ignore @code{#sccs} directives
9483: and print no error message.
9484:
1.1.1.7 root 9485: @item HAVE_VPRINTF
9486: Define this if the library function @code{vprintf} is available on your
9487: system.
1.1 root 9488:
9489: @item MOVE_MAX
9490: The maximum number of bytes that a single instruction can move quickly
9491: from memory to memory.
9492:
9493: @item INT_TYPE_SIZE
9494: A C expression for the size in bits of the type @code{int} on the
1.1.1.8 root 9495: target machine. If you don't define this, the default is one word.
9496:
9497: @item SHORT_TYPE_SIZE
9498: A C expression for the size in bits of the type @code{short} on the
9499: target machine. If you don't define this, the default is half a word.
9500: (If this would be less than one storage unit, it is rounded up to one
9501: unit.)
9502:
9503: @item LONG_TYPE_SIZE
9504: A C expression for the size in bits of the type @code{long} on the
9505: target machine. If you don't define this, the default is one word.
9506:
9507: @item LONG_LONG_TYPE_SIZE
9508: A C expression for the size in bits of the type @code{long long} on the
9509: target machine. If you don't define this, the default is two
9510: words.
9511:
9512: @item CHAR_TYPE_SIZE
9513: A C expression for the size in bits of the type @code{char} on the
9514: target machine. If you don't define this, the default is one quarter
9515: of a word. (If this would be less than one storage unit, it is rounded up
9516: to one unit.)
9517:
9518: @item FLOAT_TYPE_SIZE
9519: A C expression for the size in bits of the type @code{float} on the
9520: target machine. If you don't define this, the default is one word.
9521:
9522: @item DOUBLE_TYPE_SIZE
9523: A C expression for the size in bits of the type @code{double} on the
9524: target machine. If you don't define this, the default is two
9525: words.
9526:
9527: @item LONG_DOUBLE_TYPE_SIZE
9528: A C expression for the size in bits of the type @code{long double} on
9529: the target machine. If you don't define this, the default is two
9530: words.
1.1 root 9531:
9532: @item SLOW_BYTE_ACCESS
9533: Define this macro as a C expression which is nonzero if accessing less
9534: than a word of memory (i.e. a @code{char} or a @code{short}) is slow
9535: (requires more than one instruction).
9536:
9537: @item SLOW_ZERO_EXTEND
9538: Define this macro if zero-extension (of a @code{char} or @code{short}
9539: to an @code{int}) can be done faster if the destination is a register
9540: that is known to be zero.
9541:
9542: If you define this macro, you must have instruction patterns that
9543: recognize RTL structures like this:
9544:
9545: @example
9546: (set (strict-low-part (subreg:QI (reg:SI @dots{}) 0)) @dots{})
9547: @end example
9548:
9549: @noindent
9550: and likewise for @code{HImode}.
9551:
9552: @item SHIFT_COUNT_TRUNCATED
9553: Define this macro if shift instructions ignore all but the lowest few
9554: bits of the shift count. It implies that a sign-extend or zero-extend
9555: instruction for the shift count can be omitted.
9556:
9557: @item TRULY_NOOP_TRUNCATION (@var{outprec}, @var{inprec})
9558: A C expression which is nonzero if on this machine it is safe to
9559: ``convert'' an integer of @var{inprec} bits to one of @var{outprec}
9560: bits (where @var{outprec} is smaller than @var{inprec}) by merely
9561: operating on it as if it had only @var{outprec} bits.
9562:
9563: On many machines, this expression can be 1.
9564:
9565: @item NO_FUNCTION_CSE
9566: Define this macro if it is as good or better to call a constant
9567: function address than to call an address kept in a register.
9568:
9569: @item PROMOTE_PROTOTYPES
9570: Define this macro if an argument declared as @code{char} or
9571: @code{short} in a prototype should actually be passed as an
9572: @code{int}. In addition to avoiding errors in certain cases of
9573: mismatch, it also makes for better code on certain machines.
9574:
9575: @item STORE_FLAG_VALUE
9576: A C expression for the value stored by a store-flag instruction
9577: (@code{s@var{cond}}) when the condition is true. This is usually 1 or
1.1.1.9 root 9578: -1; it is required to be an odd number or a negative number.
1.1 root 9579:
9580: Do not define @code{STORE_FLAG_VALUE} if the machine has no store-flag
9581: instructions.
9582:
9583: @item Pmode
9584: An alias for the machine mode for pointers. Normally the definition
9585: can be
9586:
9587: @example
9588: #define Pmode SImode
9589: @end example
9590:
9591: @item FUNCTION_MODE
9592: An alias for the machine mode used for memory references to functions
1.1.1.8 root 9593: being called, in @code{call} RTL expressions. On most machines this
1.1 root 9594: should be @code{QImode}.
9595:
9596: @item INSN_MACHINE_INFO
9597: This macro should expand into a C structure type to use for the
9598: machine-dependent info field specified with the optional last argument
1.1.1.8 root 9599: in @code{define_insn} and @code{define_peephole} patterns. For example,
9600: it might expand into @code{struct machine_info}; then it would be up
1.1 root 9601: to you to define this structure in the @file{tm.h} file.
9602:
9603: You do not need to define this macro if you do not write the optional
9604: last argument in any of the patterns in the machine description.
9605:
1.1.1.8 root 9606: @item DEFAULT_MACHINE_INFO
9607: This macro should expand into a C initializer to use to initialize
9608: the machine-dependent info for one insn pattern. It is used for patterns
9609: that do not specify the machine-dependent info.
9610:
9611: If you do not define this macro, zero is used.
9612:
1.1 root 9613: @item CONST_COSTS (@var{x}, @var{code})
9614: A part of a C @code{switch} statement that describes the relative
9615: costs of constant RTL expressions. It must contain @code{case} labels
1.1.1.8 root 9616: for expression codes @code{const_int}, @code{const}, @code{symbol_ref}, @code{label_ref}
9617: and @code{const_double}. Each case must ultimately reach a
1.1 root 9618: @code{return} statement to return the relative cost of the use of that
9619: kind of constant value in an expression. The cost may depend on the
9620: precise value of the constant, which is available for examination in
9621: @var{x}.
9622:
9623: @var{code} is the expression code---redundant, since it can be
9624: obtained with @code{GET_CODE (@var{x})}.
9625:
9626: @item DOLLARS_IN_IDENTIFIERS
9627: Define this to be nonzero if the character @samp{$} should be allowed
9628: by default in identifier names.
9629: @end table
9630:
1.1.1.8 root 9631: @node Assembler Format,, Misc, Machine Macros
1.1 root 9632: @section Output of Assembler Code
9633:
9634: @table @code
9635: @item ASM_SPEC
9636: A C string constant that tells the GNU CC driver program options to
9637: pass to the assembler. It can also specify how to translate options
9638: you give to GNU CC into options for GNU CC to pass to the assembler.
9639: See the file @file{tm-sun3.h} for an example of this.
9640:
9641: Do not define this macro if it does not need to do anything.
9642:
9643: @item LINK_SPEC
9644: A C string constant that tells the GNU CC driver program options to
9645: pass to the linker. It can also specify how to translate options you
9646: give to GNU CC into options for GNU CC to pass to the linker.
9647:
9648: Do not define this macro if it does not need to do anything.
9649:
9650: @item LIB_SPEC
9651: Another C string constant used much like @code{LINK_SPEC}. The difference
9652: between the two is that @code{LIBS_SPEC} is used at the end of the
9653: command given to the linker.
9654:
9655: If this macro is not defined, a default is provided that
9656: loads the standard C library from the usual place. See @file{gcc.c}.
9657:
9658: @item STARTFILE_SPEC
9659: Another C string constant used much like @code{LINK_SPEC}. The
9660: difference between the two is that @code{STARTFILE_SPEC} is used at
9661: the very beginning of the command given to the linker.
9662:
9663: If this macro is not defined, a default is provided that loads the
9664: standard C startup file from the usual place. See @file{gcc.c}.
9665:
1.1.1.7 root 9666: @item STANDARD_EXEC_PREFIX
9667: Define this macro as a C string constant if you wish to override the
9668: standard choice of @file{/usr/local/lib/gcc-} as the default prefix to
9669: try when searching for the executable files of the compiler.
9670:
9671: The prefix specified by the @samp{-B} option, if any, is tried before
9672: the default prefix. After the default prefix, if the executable is
9673: not found that way, @file{/usr/lib/gcc-} is tried next; then the
9674: directories in your search path for shell commands are searched.
9675:
1.1.1.4 root 9676: @item STANDARD_STARTFILE_PREFIX
9677: Define this macro as a C string constant if you wish to override the
1.1.1.7 root 9678: standard choice of @file{/usr/local/lib/} as the default prefix to try
9679: when searching for startup files such as @file{crt0.o}.
9680:
9681: In this search, all the prefixes tried for executable files are tried
9682: first. Then comes the default startfile prefix specified by this
9683: macro, followed by the prefixes @file{/lib/} and @file{/usr/lib/} as
9684: last resorts.
1.1.1.4 root 9685:
1.1 root 9686: @item ASM_FILE_START (@var{stream})
9687: A C expression which outputs to the stdio stream @var{stream}
9688: some appropriate text to go at the start of an assembler file.
9689:
9690: Normally this macro is defined to output a line containing
9691: @samp{#NO_APP}, which is a comment that has no effect on most
9692: assemblers but tells the GNU assembler that it can save time by not
9693: checking for certain assembler constructs.
9694:
9695: On systems that use SDB, it is necessary to output certain commands;
9696: see @file{tm-attasm.h}.
9697:
1.1.1.8 root 9698: @item ASM_FILE_END (@var{stream})
9699: A C expression which outputs to the stdio stream @var{stream}
9700: some appropriate text to go at the end of an assembler file.
9701:
9702: If this macro is not defined, the default is to output nothing
9703: special at the end of the file. Most systems don't require any
9704: definition.
9705:
9706: On systems that use SDB, it is necessary to output certain commands;
9707: see @file{tm-attasm.h}.
9708:
9709: @item ASM_IDENTIFY_GCC (@var{file})
9710: A C statement to output assembler commands which will identify
9711: the object file as having been compiled with GNU CC (or another
9712: GNU compiler).
9713:
9714: If you don't define this macro, the string @samp{gcc_compiled.:}
9715: is output. This string is calculated to define a symbol which,
9716: on BSD systems, will never be defined for any other reason.
9717: GDB checks for the presence of this symbol when reading the
9718: symbol table of an executable.
9719:
9720: On non-BSD systems, you must arrange communication with GDB in
9721: some other fashion. If GDB is not used on your system, you can
9722: define this macro with an empty body.
9723:
1.1 root 9724: @item ASM_APP_ON
9725: A C string constant for text to be output before each @code{asm}
9726: statement or group of consecutive ones. Normally this is
9727: @code{"#APP"}, which is a comment that has no effect on most
9728: assemblers but tells the GNU assembler that it must check the lines
9729: that follow for all valid assembler constructs.
9730:
9731: @item ASM_APP_OFF
9732: A C string constant for text to be output after each @code{asm}
9733: statement or group of consecutive ones. Normally this is
9734: @code{"#NO_APP"}, which tells the GNU assembler to resume making the
9735: time-saving assumptions that are valid for ordinary compiler output.
9736:
9737: @item TEXT_SECTION_ASM_OP
9738: A C string constant for the assembler operation that should precede
9739: instructions and read-only data. Normally @code{".text"} is right.
9740:
9741: @item DATA_SECTION_ASM_OP
9742: A C string constant for the assembler operation to identify the
9743: following data as writable initialized data. Normally @code{".data"}
9744: is right.
9745:
1.1.1.8 root 9746: @item EXTRA_SECTIONS
9747: A list of names for sections other than the standard two, which are
9748: @code{in_text} and @code{in_data}. You need not define this macro
9749: on a system with no other sections (that GCC needs to use).
9750:
9751: @item EXTRA_SECTION_FUNCTIONS
9752: One or more functions to be defined in @file{varasm.c}. These
9753: functions should do jobs analogous to those of @code{text_section} and
9754: @code{data_section}, for your additional sections. Do not define this
9755: macro if you do not define @code{EXTRA_SECTIONS}.
9756:
9757: @item SELECT_SECTION (@var{exp})
9758: A C statement or statements to switch to the appropriate section for
9759: output of @var{exp}. You can assume that @var{exp} is either a
9760: @code{VAR_DECL} node or a constant of some sort. Select the section
9761: by calling @code{text_section} or one of the alternatives for other
9762: sections.
9763:
9764: Do not define this macro if you use only the standard two sections
9765: and put all read-only variables and constants in the text section.
9766:
9767: @item SELECT_RTX_SECTION (@var{mode}, @var{rtx})
9768: A C statement or statements to switch to the appropriate section for
9769: output of @var{rtx} in mode @var{mode}. You can assume that @var{rtx}
9770: is some kind of constant in RTL. The argument @var{mode} is redundant
9771: except in the case of a @code{const_int} rtx. Select the section by
9772: calling @code{text_section} or one of the alternatives for other
9773: sections.
9774:
9775: Do not define this macro if you use only the standard two sections and
9776: put all constants in the text section.
9777:
1.1 root 9778: @item REGISTER_NAMES
9779: A C initializer containing the assembler's names for the machine
9780: registers, each one as a C string constant. This is what translates
9781: register numbers in the compiler into assembler language.
9782:
9783: @item DBX_REGISTER_NUMBER (@var{regno})
9784: A C expression that returns the DBX register number for the compiler
9785: register number @var{regno}. In simple cases, the value of this
9786: expression may be @var{regno} itself. But sometimes there are some
9787: registers that the compiler knows about and DBX does not, or vice
9788: versa. In such cases, some register may need to have one number in
9789: the compiler and another for DBX.
9790:
9791: @item DBX_DEBUGGING_INFO
9792: Define this macro if GNU CC should produce debugging output for DBX
9793: in response to the @samp{-g} option.
9794:
9795: @item SDB_DEBUGGING_INFO
9796: Define this macro if GNU CC should produce debugging output for SDB
9797: in response to the @samp{-g} option.
9798:
9799: @item PUT_SDB_@var{op}
9800: Define these macros to override the assembler syntax for the special
9801: SDB assembler directives. See @file{sdbout.c} for a list of these
9802: macros and their arguments. If the standard syntax is used, you need
9803: not define them yourself.
9804:
9805: @item SDB_GENERATE_FAKE
9806: Define this macro to override the usual method of constructing a dummy
9807: name for anonymous structure and union types. See @file{sdbout.c} for
1.1.1.9 root 9808: more information.
1.1 root 9809:
9810: @item DBX_NO_XREFS
9811: Define this macro if DBX on your system does not support the construct
9812: @samp{xs@var{tagname}}. On some systems, this construct is used to
9813: describe a forward reference to a structure named @var{tagname}.
9814: On other systems, this construct is not supported at all.
9815:
9816: @item DBX_CONTIN_LENGTH
9817: A symbol name in DBX-format debugging information is normally
9818: continued (split into two separate @code{.stabs} directives) when it
9819: exceeds a certain length (by default, 80 characters). On some
9820: operating systems, DBX requires this splitting; on others, splitting
9821: must not be done. You can inhibit splitting by defining this macro
9822: with the value zero. You can override the default splitting-length by
9823: defining this macro as an expression for the length you desire.
9824:
9825: @item DBX_CONTIN_CHAR
9826: Normally continuation is indicated by adding a @samp{\} character to
9827: the end of a @code{.stabs} string when a continuation follows. To use
9828: a different character instead, define this macro as a character
9829: constant for the character you want to use. Do not define this macro
9830: if backslash is correct for your system.
9831:
1.1.1.8 root 9832: @item DBX_STATIC_STAB_DATA_SECTION
9833: Define this macro if it is necessary to go to the data section before
9834: outputting the @samp{.stabs} pseudo-op for a non-global static
9835: variable.
9836:
1.1 root 9837: @item ASM_OUTPUT_LABEL (@var{stream}, @var{name})
9838: A C statement (sans semicolon) to output to the stdio stream
1.1.1.8 root 9839: @var{stream} the assembler definition of a label named @var{name}.
9840: Use the expression @code{assemble_name (@var{stream}, @var{name})} to
9841: output the name itself; before and after that, output the additional
1.1 root 9842: assembler syntax for defining the name, and a newline.
9843:
9844: @item ASM_DECLARE_FUNCTION_NAME (@var{stream}, @var{name}, @var{decl})
9845: A C statement (sans semicolon) to output to the stdio stream
9846: @var{stream} any text necessary for declaring the name @var{name} of a
9847: function which is being defined. This macro is responsible for
9848: outputting the label definition (perhaps using
9849: @code{ASM_OUTPUT_LABEL}). The argument @var{decl} is the
9850: @code{FUNCTION_DECL} tree node representing the function.
9851:
9852: If this macro is not defined, then the function name is defined in the
9853: usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}).
9854:
9855: @item ASM_GLOBALIZE_LABEL (@var{stream}, @var{name})
9856: A C statement (sans semicolon) to output to the stdio stream
9857: @var{stream} some commands that will make the label @var{name} global;
9858: that is, available for reference from other files. Use the expression
9859: @code{assemble_name (@var{stream}, @var{name})} to output the name
9860: itself; before and after that, output the additional assembler syntax
9861: for making that name global, and a newline.
9862:
1.1.1.8 root 9863: @item ASM_OUTPUT_EXTERNAL (@var{stream}, @var{decl}, @var{name})
1.1 root 9864: A C statement (sans semicolon) to output to the stdio stream
9865: @var{stream} any text necessary for declaring the name of an external
9866: symbol named @var{name} which is referenced in this compilation but
9867: not defined. The value of @var{decl} is the tree node for the
9868: declaration.
9869:
9870: This macro need not be defined if it does not need to output anything.
9871: The GNU assembler and most Unix assemblers don't require anything.
9872:
9873: @item ASM_OUTPUT_LABELREF (@var{stream}, @var{name})
1.1.1.8 root 9874: A C statement to output to the stdio stream @var{stream} a reference
9875: in assembler syntax to a label named @var{name}. The character
9876: @samp{_} should be added to the front of the name, if that is
9877: customary on your operating system, as it is in most Berkeley Unix
9878: systems. This macro is used in @code{assemble_name}.
1.1 root 9879:
9880: @item ASM_GENERATE_INTERNAL_LABEL (@var{string}, @var{prefix}, @var{num})
1.1.1.8 root 9881: A C statement to store into the string @var{string} a label whose name
9882: is made from the string @var{prefix} and the number @var{num}.
1.1 root 9883:
9884: This string, when output subsequently by @code{ASM_OUTPUT_LABELREF},
9885: should produce the same output that @code{ASM_OUTPUT_INTERNAL_LABEL}
9886: would produce with the same @var{prefix} and @var{num}.
9887:
9888: @item ASM_OUTPUT_INTERNAL_LABEL (@var{stream}, @var{prefix}, @var{num})
9889: A C statement to output to the stdio stream @var{stream} a label whose
9890: name is made from the string @var{prefix} and the number @var{num}.
9891: These labels are used for internal purposes, and there is no reason
9892: for them to appear in the symbol table of the object file. On many
9893: systems, the letter @samp{L} at the beginning of a label has this
9894: effect. The usual definition of this macro is as follows:
9895:
9896: @example
9897: fprintf (@var{stream}, "L%s%d:\n", @var{prefix}, @var{num})
9898: @end example
9899:
9900: @item ASM_OUTPUT_CASE_LABEL (@var{stream}, @var{prefix}, @var{num}, @var{table})
9901: Define this if the label before a jump-table needs to be output
9902: specially. The first three arguments are the same as for
9903: @code{ASM_OUTPUT_INTERNAL_LABEL}; the fourth argument is the
1.1.1.8 root 9904: jump-table which follows (a @code{jump_insn} containing an
9905: @code{addr_vec} or @code{addr_diff_vec}).
1.1 root 9906:
9907: This feature is used on system V to output a @code{swbeg} statement
9908: for the table.
9909:
9910: If this macro is not defined, these labels are output with
9911: @code{ASM_OUTPUT_INTERNAL_LABEL}.
9912:
9913: @item ASM_OUTPUT_CASE_END (@var{stream}, @var{num}, @var{table})
1.1.1.8 root 9914: Define this if something special must be output at the end of a
9915: jump-table. The definition should be a C statement to be executed
9916: after the assembler code for the table is written. It should write
9917: the appropriate code to stdio stream @var{stream}. The argument
9918: @var{table} is the jump-table insn, and @var{num} is the label-number
9919: of the preceding label.
1.1 root 9920:
9921: If this macro is not defined, nothing special is output at the end of
9922: the jump-table.
9923:
1.1.1.4 root 9924: @item ASM_OUTPUT_ALIGN_CODE (@var{file})
9925: A C expression to output text to align the location counter in the way
9926: that is desirable at a point in the code that is reached only by
9927: jumping.
9928:
9929: This macro need not be defined if you don't want any special alignment
9930: to be done at such a time. Most machine descriptions do not currently
9931: define the macro.
9932:
1.1 root 9933: @item ASM_FORMAT_PRIVATE_NAME (@var{outvar}, @var{name}, @var{number})
9934: A C expression to assign to @var{outvar} (which is a variable of type
9935: @code{char *}) a newly allocated string made from the string
9936: @var{name} and the number @var{number}, with some suitable punctuation
9937: added. Use @code{alloca} to get space for the string.
9938:
9939: This string will be used as the argument to @code{ASM_OUTPUT_LABELREF}
9940: to produce an assembler label for an internal static variable whose
9941: name is @var{name}. Therefore, the string must be such as to result
9942: in valid assembler code. The argument @var{number} is different each
9943: time this macro is executed; it prevents conflicts between
9944: similarly-named internal static variables in different scopes.
9945:
9946: Ideally this string should not be a valid C identifier, to prevent any
9947: conflict with the user's own symbols. Most assemblers allow periods
9948: or percent signs in assembler symbols; putting at least one of these
9949: between the name and the number will suffice.
9950:
9951: @item ASM_OUTPUT_REG_PUSH (@var{stream}, @var{regno})
9952: A C expression to output to @var{stream} some assembler code
9953: which will push hard register number @var{regno} onto the stack.
9954: The code need not be optimal, since this macro is used only when
9955: profiling.
9956:
9957: @item ASM_OUTPUT_REG_POP (@var{stream}, @var{regno})
9958: A C expression to output to @var{stream} some assembler code
9959: which will pop hard register number @var{regno} off of the stack.
9960: The code need not be optimal, since this macro is used only when
9961: profiling.
9962:
9963: @item ASM_OUTPUT_ADDR_DIFF_ELT (@var{stream}, @var{value}, @var{rel})
9964: This macro should be provided on machines where the addresses
9965: in a dispatch table are relative to the table's own address.
9966:
9967: The definition should be a C statement to output to the stdio stream
9968: @var{stream} an assembler pseudo-instruction to generate a difference
9969: between two labels. @var{value} and @var{rel} are the numbers of two
9970: internal labels. The definitions of these labels are output using
9971: @code{ASM_OUTPUT_INTERNAL_LABEL}, and they must be printed in the same
9972: way here. For example,
9973:
9974: @example
9975: fprintf (@var{stream}, "\t.word L%d-L%d\n",
9976: @var{value}, @var{rel})
9977: @end example
9978:
9979: @item ASM_OUTPUT_ADDR_VEC_ELT (@var{stream}, @var{value})
9980: This macro should be provided on machines where the addresses
9981: in a dispatch table are absolute.
9982:
9983: The definition should be a C statement to output to the stdio stream
9984: @var{stream} an assembler pseudo-instruction to generate a reference to
9985: a label. @var{value} is the number of an internal label whose
9986: definition is output using @code{ASM_OUTPUT_INTERNAL_LABEL}.
9987: For example,
9988:
9989: @example
9990: fprintf (@var{stream}, "\t.word L%d\n", @var{value})
9991: @end example
9992:
9993: @item ASM_OUTPUT_DOUBLE (@var{stream}, @var{value})
9994: A C statement to output to the stdio stream @var{stream} an assembler
9995: instruction to assemble a @code{double} constant whose value is
9996: @var{value}. @var{value} will be a C expression of type
9997: @code{double}.
9998:
9999: @item ASM_OUTPUT_FLOAT (@var{stream}, @var{value})
10000: A C statement to output to the stdio stream @var{stream} an assembler
10001: instruction to assemble a @code{float} constant whose value is
10002: @var{value}. @var{value} will be a C expression of type @code{float}.
10003:
10004: @item ASM_OUTPUT_INT (@var{stream}, @var{exp})
10005: @itemx ASM_OUTPUT_SHORT (@var{stream}, @var{exp})
10006: @itemx ASM_OUTPUT_CHAR (@var{stream}, @var{exp})
10007: A C statement to output to the stdio stream @var{stream} an assembler
10008: instruction to assemble a @code{int}, @code{short} or @code{char}
1.1.1.10 root 10009: constant whose value is @var{value}. The argument @var{exp} will be an
10010: RTL expression which represents a constant value. Use
10011: @samp{output_addr_const (@var{stream}, @var{exp})} to output this value
10012: as an assembler expression.@refill
1.1 root 10013:
1.1.1.8 root 10014: @item ASM_OUTPUT_DOUBLE_INT (@var{stream}, @var{exp})
10015: A C statement to output to the stdio stream @var{stream} an assembler
10016: instruction to assemble a @code{long long} constant whose value is
10017: @var{exp}. The argument @var{exp} will be an RTL expression which
10018: represents a constant value. It may be a @code{const_double} RTX,
10019: or it may be an ordinary single-precision constant. In the latter
10020: case, you should zero-extend it.
10021:
1.1 root 10022: @item ASM_OUTPUT_BYTE (@var{stream}, @var{value})
10023: A C statement to output to the stdio stream @var{stream} an assembler
10024: instruction to assemble a single byte containing the number @var{value}.
10025:
10026: @item ASM_OUTPUT_ASCII (@var{stream}, @var{ptr}, @var{len})
10027: A C statement to output to the stdio stream @var{stream} an assembler
10028: instruction to assemble a string constant containing the @var{len}
10029: bytes at @var{ptr}. @var{ptr} will be a C expression of type
10030: @code{char *} and @var{len} a C expression of type @code{int}.
10031:
10032: If the assembler has a @code{.ascii} pseudo-op as found in the
10033: Berkeley Unix assembler, do not define the macro
10034: @code{ASM_OUTPUT_ASCII}.
10035:
10036: @item ASM_OUTPUT_SKIP (@var{stream}, @var{nbytes})
10037: A C statement to output to the stdio stream @var{stream} an assembler
10038: instruction to advance the location counter by @var{nbytes} bytes.
10039: @var{nbytes} will be a C expression of type @code{int}.
10040:
10041: @item ASM_OUTPUT_ALIGN (@var{stream}, @var{power})
10042: A C statement to output to the stdio stream @var{stream} an assembler
10043: instruction to advance the location counter to a multiple of 2 to the
10044: @var{power} bytes. @var{power} will be a C expression of type @code{int}.
10045:
1.1.1.7 root 10046: @item ASM_OUTPUT_COMMON (@var{stream}, @var{name}, @var{size}, @var{rounded})
1.1 root 10047: A C statement (sans semicolon) to output to the stdio stream
1.1.1.7 root 10048: @var{stream} the assembler definition of a common-label named
10049: @var{name} whose size is @var{size} bytes. The variable @var{rounded}
10050: is the size rounded up to whatever alignment the caller wants.
10051:
10052: Use the expression @code{assemble_name (@var{stream}, @var{name})} to
10053: output the name itself; before and after that, output the additional
10054: assembler syntax for defining the name, and a newline.
1.1 root 10055:
10056: This macro controls how the assembler definitions of uninitialized
10057: global variables are output.
10058:
1.1.1.7 root 10059: @item ASM_OUTPUT_LOCAL (@var{stream}, @var{name}, @var{size}, @var{rounded})
1.1 root 10060: A C statement (sans semicolon) to output to the stdio stream
10061: @var{stream} the assembler definition of a local-common-label named
1.1.1.7 root 10062: @var{name} whose size is @var{size} bytes. The variable @var{rounded}
10063: is the size rounded up to whatever alignment the caller wants.
10064:
10065: Use the expression @code{assemble_name (@var{stream}, @var{name})} to
10066: output the name itself; before and after that, output the additional
10067: assembler syntax for defining the name, and a newline.
1.1 root 10068:
10069: This macro controls how the assembler definitions of uninitialized
10070: static variables are output.
10071:
1.1.1.8 root 10072: @item ASM_OUTPUT_SOURCE_FILENAME (@var{stream}, @var{name})
10073: A C statment to output DBX or SDB debugging information which indicates
10074: that filename @var{name} is the current source file to the stdio stream
10075: @var{stream}.
10076:
10077: This macro need not be defined if the standard form of debugging
10078: information for the debugger in use is appropriate.
10079:
1.1 root 10080: @item ASM_OUTPUT_SOURCE_LINE (@var{stream}, @var{line})
10081: A C statment to output DBX or SDB debugging information before code
10082: for line number @var{line} of the current source file to the
10083: stdio stream @var{stream}.
10084:
10085: This macro need not be defined if the standard form of debugging
10086: information for the debugger in use is appropriate.
10087:
10088: @item ASM_OUTPUT_IDENT (@var{stream}, @var{string})
10089: A C statement to output something to the assembler file to handle a
10090: @samp{#ident} directive containing the text @var{string}. If this
1.1.1.7 root 10091: macro is not defined, nothing is output for a @samp{#ident} directive.
1.1 root 10092:
10093: @item TARGET_BELL
10094: A C constant expression for the integer value for escape sequence
10095: @samp{\a}.
10096:
10097: @item TARGET_BS
10098: @itemx TARGET_TAB
10099: @itemx TARGET_NEWLINE
10100: C constant expressions for the integer values for escape sequences
10101: @samp{\b}, @samp{\t} and @samp{\n}.
10102:
10103: @item TARGET_VT
10104: @itemx TARGET_FF
10105: @itemx TARGET_CR
10106: C constant expressions for the integer values for escape sequences
10107: @samp{\v}, @samp{\f} and @samp{\r}.
10108:
10109: @item ASM_OUTPUT_OPCODE (@var{stream}, @var{ptr})
10110: Define this macro if you are using an unusual assembler that
10111: requires different names for the machine instructions.
10112:
10113: The definition is a C statement or statements which output an
10114: assembler instruction opcode to the stdio stream @var{stream}. The
10115: macro-operand @var{ptr} is a variable of type @code{char *} which
10116: points to the opcode name in its ``internal'' form---the form that is
10117: written in the machine description. The definition should output the
10118: opcode name to @var{stream}, performing any translation you desire, and
10119: increment the variable @var{ptr} to point at the end of the opcode
10120: so that it will not be output twice.
10121:
10122: In fact, your macro definition may process less than the entire opcode
10123: name, or more than the opcode name; but if you want to process text
10124: that includes @samp{%}-sequences to substitute operands, you must take
10125: care of the substitution yourself. Just be sure to increment
10126: @var{ptr} over whatever text should not be output normally.
10127:
1.1.1.8 root 10128: If you need to look at the operand values, they can be found as the
10129: elements of @code{recog_operand}.
10130:
1.1 root 10131: If the macro definition does nothing, the instruction is output
10132: in the usual way.
10133:
10134: @item FINAL_PRESCAN_INSN (@var{insn}, @var{opvec}, @var{noperands})
10135: If defined, a C statement to be executed just prior to the output of
10136: assembler code for @var{insn}, to modify the extracted operands so
10137: they will be output differently.
10138:
10139: Here the argument @var{opvec} is the vector containing the operands
10140: extracted from @var{insn}, and @var{noperands} is the number of
10141: elements of the vector which contain meaningful data for this insn.
10142: The contents of this vector are what will be used to convert the insn
10143: template into assembler code, so you can change the assembler output
10144: by changing the contents of the vector.
10145:
10146: This macro is useful when various assembler syntaxes share a single
10147: file of instruction patterns; by defining this macro differently, you
10148: can cause a large class of instructions to be output differently (such
10149: as with rearranged operands). Naturally, variations in assembler
10150: syntax affecting individual insn patterns ought to be handled by
10151: writing conditional output routines in those patterns.
10152:
10153: If this macro is not defined, it is equivalent to a null statement.
10154:
10155: @item PRINT_OPERAND (@var{stream}, @var{x}, @var{code})
10156: A C compound statement to output to stdio stream @var{stream} the
10157: assembler syntax for an instruction operand @var{x}. @var{x} is an
10158: RTL expression.
10159:
10160: @var{code} is a value that can be used to specify one of several ways
10161: of printing the operand. It is used when identical operands must be
10162: printed differently depending on the context. @var{code} comes from
10163: the @samp{%} specification that was used to request printing of the
10164: operand. If the specification was just @samp{%@var{digit}} then
10165: @var{code} is 0; if the specification was @samp{%@var{ltr}
10166: @var{digit}} then @var{code} is the ASCII code for @var{ltr}.
10167:
10168: If @var{x} is a register, this macro should print the register's name.
10169: The names can be found in an array @code{reg_names} whose type is
10170: @code{char *[]}. @code{reg_names} is initialized from
10171: @code{REGISTER_NAMES}.
10172:
10173: When the machine description has a specification @samp{%@var{punct}}
10174: (a @samp{%} followed by a punctuation character), this macro is called
10175: with a null pointer for @var{x} and the punctuation character for
10176: @var{code}.
10177:
1.1.1.8 root 10178: @item PRINT_OPERAND_PUNCT_VALID_P (@var{code})
10179: A C expression which evaluates to true if @var{code} is a valid
10180: punctuation character for use in the @code{PRINT_OPERAND} macro. If
10181: @code{PRINT_OPERAND_PUNCT_VALID_P} is not defined, it means that no
10182: punctuation characters (except for the standard one, @samp{%}) are used
10183: in this way.
10184:
1.1 root 10185: @item PRINT_OPERAND_ADDRESS (@var{stream}, @var{x})
10186: A C compound statement to output to stdio stream @var{stream} the
10187: assembler syntax for an instruction operand that is a memory reference
10188: whose address is @var{x}. @var{x} is an RTL expression.
10189:
10190: @item ASM_OPEN_PAREN
10191: @itemx ASM_CLOSE_PAREN
10192: These macros are defined as C string constant, describing the syntax
10193: in the assembler for grouping arithmetic expressions. The following
10194: definitions are correct for most assemblers:
10195:
10196: @example
10197: #define ASM_OPEN_PAREN "("
10198: #define ASM_CLOSE_PAREN ")"
10199: @end example
10200: @end table
10201:
10202: @node Config,, Machine Macros, Top
10203: @chapter The Configuration File
10204:
1.1.1.3 root 10205: The configuration file @file{xm-@var{machine}.h} contains macro definitions
10206: that describe the machine and system on which the compiler is running.
10207: Most of the values in it are actually the same on all machines that GNU CC
10208: runs on, so large parts of all configuration files are identical. But
1.1 root 10209: there are some macros that vary:
10210:
10211: @table @code
10212: @item FAILURE_EXIT_CODE
10213: A C expression for the status code to be returned when the compiler
10214: exits after serious errors.
10215:
10216: @item SUCCESS_EXIT_CODE
10217: A C expression for the status code to be returned when the compiler
10218: exits without serious errors.
1.1.1.10 root 10219:
10220: @item USE_C_ALLOCA
10221: Define this macro to indicate that the compiler is running with the
10222: @code{alloca} implemented in C. This version of @code{alloca} can be
10223: found in the file @file{alloca.c}; to use it, you must also alter the
10224: @file{Makefile} variable @code{ALLOCA}.
10225:
10226: This macro, unlike most, describes the machine that the compiler is
10227: running on, rather than the one the compiler is compiling for.
10228: Therefore, it should be set in the @file{xm-@var{machine}.h} file
10229: rather than in the @file{tm-@var{machine}.h} file.
10230:
10231: If you do define this macro, you should probably do it as follows:
10232:
10233: @example
10234: #ifndef __GNUC__
10235: #define USE_C_ALLOCA
10236: #else
10237: #define alloca __builtin_alloca
10238: #endif
10239: @end example
10240:
10241: @noindent
10242: so that when the compiler is compiled with GNU CC it uses the more
10243: efficient built-in @code{alloca} function.
1.1 root 10244: @end table
10245:
1.1.1.3 root 10246: In addition, configuration files for system V define @code{bcopy},
10247: @code{bzero} and @code{bcmp} as aliases. Some files define @code{alloca}
10248: as a macro when compiled with GNU CC, in order to take advantage of the
10249: benefit of GNU CC's built-in @code{alloca}.
10250:
1.1 root 10251: @contents
10252: @bye
This archive runs on limited infrastructure. Preserving old code on modern bandwidth. Automated agents are requested to crawl responsibly.