![[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
1.1.1.10! root 45: @center last updated 4 December 1990
1.1 root 46: @sp 1
1.1.1.10! root 47: @center for version 1.38
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
517: Gratuitous incompatibilites will burden users. Imagine if each
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
605: Write to or phone your elected representatives to show them how
606: important this issue is.
607:
608: @example
609: Senator So and So Representative So and So
610: United States Senate House of Representatives
611: Washington, DC 20510 Washington, DC 20515
612: @end example
613:
614: You can phone senators and representatives at (202) 225-3121.
615: @end itemize
616:
617: Express your opinion! You can make a difference.
618:
619: @node Options, Installation, Boycott, Top
1.1 root 620: @chapter GNU CC Command Options
621:
622: The GNU C compiler uses a command syntax much like the Unix C compiler.
623: The @code{gcc} program accepts options and file names as operands.
624: Multiple single-letter options may @emph{not} be grouped: @samp{-dr} is
1.1.1.8 root 625: very different from @w{@samp{-d -r}}.
1.1 root 626:
627: When you invoke GNU CC, it normally does preprocessing, compilation,
628: assembly and linking. File names which end in @samp{.c} are taken as C
1.1.1.5 root 629: source to be preprocessed and compiled; file names ending in @samp{.i}
630: are taken as preprocessor output to be compiled; compiler output files
631: plus any input files with names ending in @samp{.s} are assembled; then
632: the resulting object files, plus any other input files, are linked
633: together to produce an executable.
1.1 root 634:
635: Command options allow you to stop this process at an intermediate stage.
636: For example, the @samp{-c} option says not to run the linker. Then the
637: output consists of object files output by the assembler.
638:
1.1.1.5 root 639: Other command options are passed on to one stage of processing. Some
640: options control the preprocessor and others the compiler itself. Yet
641: other options control the assembler and linker; these are not documented
642: here, but you rarely need to use any of them.
1.1 root 643:
644: Here are the options to control the overall compilation process, including
645: those that say whether to link, whether to assemble, and so on.
646:
647: @table @samp
648: @item -o @var{file}
649: Place output in file @var{file}. This applies regardless to whatever
650: sort of output is being produced, whether it be an executable file,
651: an object file, an assembler file or preprocessed C code.
652:
653: If @samp{-o} is not specified, the default is to put an executable file
654: in @file{a.out}, the object file @file{@var{source}.c} in
655: @file{@var{source}.o}, an assembler file in @file{@var{source}.s}, and
656: preprocessed C on standard output.@refill
657:
658: @item -c
659: Compile or assemble the source files, but do not link. Produce object
660: files with names made by replacing @samp{.c} or @samp{.s} with
661: @samp{.o} at the end of the input file names. Do nothing at all for
662: object files specified as input.
663:
664: @item -S
665: Compile into assembler code but do not assemble. The assembler output
666: file name is made by replacing @samp{.c} with @samp{.s} at the end of
667: the input file name. Do nothing at all for assembler source files or
668: object files specified as input.
669:
670: @item -E
671: Run only the C preprocessor. Preprocess all the C source files
672: specified and output the results to standard output.
673:
674: @item -v
675: Compiler driver program prints the commands it executes as it runs
676: the preprocessor, compiler proper, assembler and linker. Some of
677: these are directed to print their own version numbers.
678:
1.1.1.5 root 679: @item -pipe
680: Use pipes rather than temporary files for communication between the
681: various stages of compilation. This fails to work on some systems
682: where the assembler is unable to read from a pipe; but the GNU
683: assembler has no trouble.
684:
1.1 root 685: @item -B@var{prefix}
686: Compiler driver program tries @var{prefix} as a prefix for each
687: program it tries to run. These programs are @file{cpp}, @file{cc1},
688: @file{as} and @file{ld}.
689:
690: For each subprogram to be run, the compiler driver first tries the
691: @samp{-B} prefix, if any. If that name is not found, or if @samp{-B}
692: was not specified, the driver tries two standard prefixes, which are
693: @file{/usr/lib/gcc-} and @file{/usr/local/lib/gcc-}. If neither of
694: those results in a file name that is found, the unmodified program
695: name is searched for using the directories specified in your
696: @samp{PATH} environment variable.
697:
698: The run-time support file @file{gnulib} is also searched for using
699: the @samp{-B} prefix, if needed. If it is not found there, the two
700: standard prefixes above are tried, and that is all. The file is left
701: out of the link if it is not found by those means. Most of the time,
702: on most machines, you can do without it.
1.1.1.5 root 703:
704: You can get a similar result from the environment variable;
705: @code{GCC_EXEC_PREFIX} if it is defined, its value is used as a prefix
706: in the same way. If both the @samp{-B} option and the
707: @code{GCC_EXEC_PREFIX} variable are present, the @samp{-B} option is
708: used first and the environment variable value second.
1.1.1.8 root 709:
710: @item -b@var{prefix}
711: The argument @var{prefix} is used as a second prefix for the compiler
712: executables and libraries. This prefix is optional: the compiler tries
713: each file first with it, then without it. This prefix follows the
714: prefix specified with @samp{-B} or the default prefixes.
715:
716: Thus, @samp{-bvax- -Bcc/} in the presence of environment variable
717: @code{GCC_EXEC_PREFIX} with definition @file{/u/foo/} causes GNU CC to
718: try the following file names for the preprocessor executable:
719:
720: @example
721: cc/vax-cpp
722: cc/cpp
723: /u/foo/vax-cpp
724: /u/foo/cpp
725: /usr/local/lib/gcc-vax-cpp
726: /usr/local/lib/gcc-cpp
727: /usr/lib/gcc-vax-cpp
728: /usr/lib/gcc-cpp
729: @end example
1.1 root 730: @end table
731:
732: These options control the details of C compilation itself.
733:
734: @table @samp
735: @item -ansi
736: Support all ANSI standard C programs.
737:
738: This turns off certain features of GNU C that are incompatible with
739: ANSI C, such as the @code{asm}, @code{inline} and @code{typeof}
740: keywords, and predefined macros such as @code{unix} and @code{vax}
741: that identify the type of system you are using. It also enables the
742: undesirable and rarely used ANSI trigraph feature.
743:
1.1.1.8 root 744: The alternate keywords @code{__asm__}, @code{__inline__} and
745: @code{__typeof__} continue to work despite @samp{-ansi}. You would not
1.1.1.7 root 746: want to use them in an ANSI C program, of course, but it useful to put
747: them in header files that might be included in compilations done with
1.1.1.8 root 748: @samp{-ansi}. Alternate predefined macros such as @code{__unix__} and
749: @code{__vax__} are also available, with or without @samp{-ansi}.
1.1.1.7 root 750:
1.1 root 751: The @samp{-ansi} option does not cause non-ANSI programs to be
752: rejected gratuitously. For that, @samp{-pedantic} is required in
753: addition to @samp{-ansi}.
754:
755: The macro @code{__STRICT_ANSI__} is predefined when the @samp{-ansi}
756: option is used. Some header files may notice this macro and refrain
757: from declaring certain functions or defining certain macros that the
1.1.1.7 root 758: ANSI standard doesn't call for; this is to avoid interfering with any
759: programs that might use these names for other things.
1.1 root 760:
761: @item -traditional
762: Attempt to support some aspects of traditional C compilers.
763: Specifically:
764:
765: @itemize @bullet
766: @item
767: All @code{extern} declarations take effect globally even if they
768: are written inside of a function definition. This includes implicit
769: declarations of functions.
770:
771: @item
772: The keywords @code{typeof}, @code{inline}, @code{signed}, @code{const}
1.1.1.10! root 773: and @code{volatile} are not recognized. (You can still use the alternative
! 774: keywords such as @code{__typeof__}, @code{__inline__}, and so on.)
1.1 root 775:
776: @item
777: Comparisons between pointers and integers are always allowed.
778:
779: @item
780: Integer types @code{unsigned short} and @code{unsigned char} promote
781: to @code{unsigned int}.
782:
783: @item
784: Out-of-range floating point literals are not an error.
785:
786: @item
1.1.1.8 root 787: String ``constants'' are not necessarily constant; they are stored in
788: writable space, and identical looking constants are allocated
789: separately.
790:
791: @item
1.1.1.2 root 792: All automatic variables not declared @code{register} are preserved by
793: @code{longjmp}. Ordinarily, GNU C follows ANSI C: automatic variables
794: not declared @code{volatile} may be clobbered.
795:
796: @item
1.1 root 797: In the preprocessor, comments convert to nothing at all, rather than
798: to a space. This allows traditional token concatenation.
799:
800: @item
801: In the preprocessor, macro arguments are recognized within string
802: constants in a macro definition (and their values are stringified,
803: though without additional quote marks, when they appear in such a
804: context). The preprocessor always considers a string constant to end
805: at a newline.
806:
807: @item
808: The predefined macro @code{__STDC__} is not defined when you use
809: @samp{-traditional}, but @code{__GNUC__} is (since the GNU extensions
810: which @code{__GNUC__} indicates are not affected by
811: @samp{-traditional}). If you need to write header files that work
812: differently depending on whether @samp{-traditional} is in use, by
813: testing both of these predefined macros you can distinguish four
814: situations: GNU C, traditional GNU C, other ANSI C compilers, and
815: other old C compilers.
816: @end itemize
817:
818: @item -O
819: Optimize. Optimizing compilation takes somewhat more time, and a lot
820: more memory for a large function.
821:
822: Without @samp{-O}, the compiler's goal is to reduce the cost of
823: compilation and to make debugging produce the expected results.
824: Statements are independent: if you stop the program with a breakpoint
825: between statements, you can then assign a new value to any variable or
826: change the program counter to any other statement in the function and
827: get exactly the results you would expect from the source code.
828:
829: Without @samp{-O}, only variables declared @code{register} are
830: allocated in registers. The resulting compiled code is a little worse
831: than produced by PCC without @samp{-O}.
832:
833: With @samp{-O}, the compiler tries to reduce code size and execution
834: time.
835:
836: Some of the @samp{-f} options described below turn specific kinds of
837: optimization on or off.
838:
839: @item -g
840: Produce debugging information in the operating system's native format
841: (for DBX or SDB). GDB also can work with this debugging information.
842:
843: Unlike most other C compilers, GNU CC allows you to use @samp{-g} with
844: @samp{-O}. The shortcuts taken by optimized code may occasionally
845: produce surprising results: some variables you declared may not exist
846: at all; flow of control may briefly move where you did not expect it;
847: some statements may not be executed because they compute constant
848: results or their values were already at hand; some statements may
849: execute in different places because they were moved out of loops.
850: Nevertheless it proves possible to debug optimized output. This makes
851: it reasonable to use the optimizer for programs that might have bugs.
852:
853: @item -gg
1.1.1.8 root 854: Produce debugging information in the old GDB format. This is obsolete.
1.1 root 855:
856: @item -w
857: Inhibit all warning messages.
858:
859: @item -W
860: Print extra warning messages for these events:
861:
862: @itemize @bullet
863: @item
864: An automatic variable is used without first being initialized.
865:
866: These warnings are possible only in optimizing compilation,
867: because they require data flow information that is computed only
1.1.1.6 root 868: when optimizing. If you don't specify @samp{-O}, you simply won't
869: get these warnings.
870:
871: These warnings occur only for variables that are candidates for
872: register allocation. Therefore, they do not occur for a variable that
873: is declared @code{volatile}, or whose address is taken, or whose size
874: is other than 1, 2, 4 or 8 bytes. Also, they do not occur for
875: structures, unions or arrays, even when they are in registers.
876:
877: Note that there may be no warning about a variable that is used only
878: to compute a value that itself is never used, because such
879: computations may be deleted by data flow analysis before the warnings
880: are printed.
1.1 root 881:
882: These warnings are made optional because GNU CC is not smart
883: enough to see all the reasons why the code might be correct
884: despite appearing to have an error. Here is one example of how
885: this can happen:
886:
887: @example
888: @{
889: int x;
890: switch (y)
891: @{
892: case 1: x = 1;
893: break;
894: case 2: x = 4;
895: break;
896: case 3: x = 5;
897: @}
898: foo (x);
899: @}
900: @end example
901:
902: @noindent
903: If the value of @code{y} is always 1, 2 or 3, then @code{x} is
904: always initialized, but GNU CC doesn't know this. Here is
905: another common case:
906:
907: @example
908: @{
909: int save_y;
910: if (change_y) save_y = y, y = new_y;
911: @dots{}
912: if (change_y) y = save_y;
913: @}
914: @end example
915:
916: @noindent
917: This has no bug because @code{save_y} is used only if it is set.
918:
1.1.1.5 root 919: Some spurious warnings can be avoided if you declare as
920: @code{volatile} all the functions you use that never return.
921: @xref{Function Attributes}.
922:
1.1 root 923: @item
924: A nonvolatile automatic variable might be changed by a call to
925: @code{longjmp}. These warnings as well are possible only in
926: optimizing compilation.
927:
928: The compiler sees only the calls to @code{setjmp}. It cannot know
929: where @code{longjmp} will be called; in fact, a signal handler could
930: call it at any point in the code. As a result, you may get a warning
931: even when there is in fact no problem because @code{longjmp} cannot
932: in fact be called at the place which would cause a problem.
933:
934: @item
935: A function can return either with or without a value. (Falling
936: off the end of the function body is considered returning without
1.1.1.6 root 937: a value.) For example, this function would evoke such a
1.1 root 938: warning:
939:
940: @example
941: foo (a)
942: @{
943: if (a > 0)
944: return a;
945: @}
946: @end example
947:
948: Spurious warnings can occur because GNU CC does not realize that
949: certain functions (including @code{abort} and @code{longjmp})
950: will never return.
1.1.1.4 root 951:
952: @item
953: An expression-statement contains no side effects.
1.1 root 954: @end itemize
955:
956: In the future, other useful warnings may also be enabled by this
957: option.
958:
959: @item -Wimplicit
960: Warn whenever a function is implicitly declared.
961:
962: @item -Wreturn-type
963: Warn whenever a function is defined with a return-type that defaults
964: to @code{int}. Also warn about any @code{return} statement with no
965: return-value in a function whose return-type is not @code{void}.
966:
967: @item -Wunused
1.1.1.5 root 968: Warn whenever a local variable is unused aside from its declaration,
1.1.1.8 root 969: whenever a function is declared static but never defined, and whenever
970: a statement computes a result that is explicitly not used.
1.1.1.7 root 971:
972: @item -Wswitch
973: Warn whenever a @code{switch} statement has an index of enumeral type
974: and lacks a @code{case} for one or more of the named codes of that
975: enumeration. (The presence of a @code{default} label prevents this
976: warning.) @code{case} labels outside the enumeration range also
977: provoke warnings when this option is used.
978:
1.1 root 979: @item -Wcomment
980: Warn whenever a comment-start sequence @samp{/*} appears in a comment.
981:
1.1.1.7 root 982: @item -Wtrigraphs
983: Warn if any trigraphs are encountered (assuming they are enabled).
984:
1.1 root 985: @item -Wall
1.1.1.8 root 986: All of the above @samp{-W} options combined. These are all the
987: options which pertain to usage that we recommend avoiding and that we
988: believe is easy to avoid, even in conjunction with macros.
989:
990: The other @samp{-W@dots{}} options below are not implied by @samp{-Wall}
991: because certain kinds of useful macros are almost impossible to write
992: without causing those warnings.
993:
994: @item -Wshadow
995: Warn whenever a local variable shadows another local variable.
996:
997: @item -Wid-clash-@var{len}
998: Warn whenever two distinct identifiers match in the first @var{len}
999: characters. This may help you prepare a program that will compile
1000: with certain obsolete, brain-damaged compilers.
1001:
1002: @item -Wpointer-arith
1003: Warn about anything that depends on the ``size of'' a function type or
1004: of @code{void}. GNU C assigns these types a size of 1, for
1005: convenience in calculations with @code{void *} pointers and pointers
1006: to functions.
1.1 root 1007:
1.1.1.6 root 1008: @item -Wcast-qual
1009: Warn whenever a pointer is cast so as to remove a type qualifier from
1010: the target type. For example, warn if a @code{const char *} is cast
1011: to an ordinary @code{char *}.
1012:
1.1 root 1013: @item -Wwrite-strings
1014: Give string constants the type @code{const char[@var{length}]} so that
1015: copying the address of one into a non-@code{const} @code{char *}
1016: pointer will get a warning. These warnings will help you find at
1017: compile time code that can try to write into a string constant, but
1018: only if you have been very careful about using @code{const} in
1019: declarations and prototypes. Otherwise, it will just be a nuisance;
1020: this is why we did not make @samp{-Wall} request these warnings.
1021:
1022: @item -p
1023: Generate extra code to write profile information suitable for the
1024: analysis program @code{prof}.
1025:
1026: @item -pg
1027: Generate extra code to write profile information suitable for the
1028: analysis program @code{gprof}.
1029:
1.1.1.6 root 1030: @item -a
1.1.1.8 root 1031: Generate extra code to write profile information for basic blocks, which
1032: will record the number of times each basic block is executed. This data
1033: could be analyzed by a program like @code{tcov}. Note, however, that
1034: the format of the data is not what @code{tcov} expects. Eventually GNU
1.1.1.6 root 1035: @code{gprof} should be extended to process this data.
1036:
1.1 root 1037: @item -l@var{library}
1038: Search a standard list of directories for a library named
1039: @var{library}, which is actually a file named
1040: @file{lib@var{library}.a}. The linker uses this file as if it
1041: had been specified precisely by name.
1042:
1043: The directories searched include several standard system directories
1044: plus any that you specify with @samp{-L}.
1045:
1046: Normally the files found this way are library files---archive files
1047: whose members are object files. The linker handles an archive file by
1048: scanning through it for members which define symbols that have so far
1049: been referenced but not defined. But if the file that is found is an
1050: ordinary object file, it is linked in the usual fashion. The only
1051: difference between using an @samp{-l} option and specifying a file name
1052: is that @samp{-l} searches several directories.
1053:
1054: @item -L@var{dir}
1055: Add directory @var{dir} to the list of directories to be searched
1056: for @samp{-l}.
1057:
1058: @item -nostdlib
1.1.1.9 root 1059: Don't use the standard system libraries and startup files when linking.
1060: Only the files you specify will be passed to the linker.
1.1 root 1061:
1062: @item -m@var{machinespec}
1063: Machine-dependent option specifying something about the type of target
1064: machine. These options are defined by the macro
1065: @code{TARGET_SWITCHES} in the machine description. The default for
1066: the options is also defined by that macro, which enables you to change
1067: the defaults.@refill
1068:
1069: These are the @samp{-m} options defined in the 68000 machine
1070: description:
1071:
1072: @table @samp
1073: @item -m68020
1074: @itemx -mc68020
1075: Generate output for a 68020 (rather than a 68000). This is the
1076: default if you use the unmodified sources.
1077:
1078: @item -m68000
1079: @item -mc68000
1080: Generate output for a 68000 (rather than a 68020).
1081:
1082: @item -m68881
1083: Generate output containing 68881 instructions for floating point.
1084: This is the default if you use the unmodified sources.
1085:
1086: @item -mfpa
1087: Generate output containing Sun FPA instructions for floating point.
1088:
1089: @item -msoft-float
1090: Generate output containing library calls for floating point.
1091:
1092: @item -mshort
1093: Consider type @code{int} to be 16 bits wide, like @code{short int}.
1094:
1095: @item -mnobitfield
1096: Do not use the bit-field instructions. @samp{-m68000} implies
1097: @samp{-mnobitfield}.
1098:
1099: @item -mbitfield
1100: Do use the bit-field instructions. @samp{-m68020} implies
1101: @samp{-mbitfield}. This is the default if you use the unmodified
1102: sources.
1103:
1104: @item -mrtd
1105: Use a different function-calling convention, in which functions
1106: that take a fixed number of arguments return with the @code{rtd}
1107: instruction, which pops their arguments while returning. This
1108: saves one instruction in the caller since there is no need to pop
1109: the arguments there.
1110:
1111: This calling convention is incompatible with the one normally
1112: used on Unix, so you cannot use it if you need to call libraries
1113: compiled with the Unix compiler.
1114:
1115: Also, you must provide function prototypes for all functions that
1116: take variable numbers of arguments (including @code{printf});
1117: otherwise incorrect code will be generated for calls to those
1118: functions.
1119:
1120: In addition, seriously incorrect code will result if you call a
1121: function with too many arguments. (Normally, extra arguments are
1122: harmlessly ignored.)
1123:
1124: The @code{rtd} instruction is supported by the 68010 and 68020
1125: processors, but not by the 68000.
1126: @end table
1127:
1128: These @samp{-m} options are defined in the Vax machine description:
1129:
1130: @table @samp
1131: @item -munix
1132: Do not output certain jump instructions (@code{aobleq} and so on)
1133: that the Unix assembler for the Vax cannot handle across long
1134: ranges.
1135:
1136: @item -mgnu
1137: Do output those jump instructions, on the assumption that you
1138: will assemble with the GNU assembler.
1139:
1140: @item -mg
1141: Output code for g-format floating point numbers instead of d-format.
1142: @end table
1143:
1.1.1.5 root 1144: These @samp{-m} switches are supported on the Sparc:
1145:
1146: @table @samp
1147: @item -mfpu
1148: Generate output containing floating point instructions. This is the
1149: default if you use the unmodified sources.
1150:
1.1.1.9 root 1151: @ignore
1.1.1.5 root 1152: @item -msoft-float
1153: Generate output containing library calls for floating point.
1154:
1.1.1.9 root 1155: @end ignore
1.1.1.5 root 1156: @item -mno-epilogue
1.1.1.6 root 1157: Generate separate return instructions for @code{return} statements.
1158: This has both advantages and disadvantages; I don't recall what they
1159: are.
1.1.1.5 root 1160: @end table
1161:
1162: These @samp{-m} options are defined in the Convex machine description:
1163:
1164: @table @samp
1165: @item -mc1
1166: Generate output for a C1. This is the default when the compiler is
1167: configured for a C1.
1168:
1169: @item -mc2
1170: Generate output for a C2. This is the default when the compiler is
1171: configured for a C2.
1172:
1173: @item -margcount
1174: Generate code which puts an argument count in the word preceding each
1175: argument list. Some nonportable Convex and Vax programs need this
1176: word. (Debuggers don't; this info is in the symbol table.)
1177:
1178: @item -mnoargcount
1179: Omit the argument count word. This is the default if you use the
1180: unmodified sources.
1181: @end table
1182:
1.1 root 1183: @item -f@var{flag}
1.1.1.4 root 1184: Specify machine-independent flags. Most flags have both positive and
1185: negative forms; the negative form of @samp{-ffoo} would be
1186: @samp{-fno-foo}. In the table below, only one of the forms is
1187: listed---the one which is not the default. You can figure out the
1188: other form by either removing @samp{no-} or adding it.
1.1 root 1189:
1190: @table @samp
1.1.1.6 root 1191: @item -fpcc-struct-return
1192: Use the same convention for returning @code{struct} and @code{union}
1193: values that is used by the usual C compiler on your system. This
1194: convention is less efficient for small structures, and on many
1195: machines it fails to be reentrant; but it has the advantage of
1196: allowing intercallability between GCC-compiled code and PCC-compiled
1197: code.
1198:
1.1 root 1199: @item -ffloat-store
1200: Do not store floating-point variables in registers. This
1201: prevents undesirable excess precision on machines such as the
1202: 68000 where the floating registers (of the 68881) keep more
1203: precision than a @code{double} is supposed to have.
1204:
1205: For most programs, the excess precision does only good, but a few
1206: programs rely on the precise definition of IEEE floating point.
1207: Use @samp{-ffloat-store} for such programs.
1208:
1209: @item -fno-asm
1210: Do not recognize @code{asm}, @code{inline} or @code{typeof} as a
1.1.1.7 root 1211: keyword. These words may then be used as identifiers. You can
1.1.1.8 root 1212: use @code{__asm__}, @code{__inline__} and @code{__typeof__} instead.
1.1 root 1213:
1214: @item -fno-defer-pop
1215: Always pop the arguments to each function call as soon as that
1216: function returns. Normally the compiler (when optimizing) lets
1217: arguments accumulate on the stack for several function calls and
1218: pops them all at once.
1219:
1220: @item -fstrength-reduce
1221: Perform the optimizations of loop strength reduction and
1222: elimination of iteration variables.
1223:
1224: @item -fcombine-regs
1225: Allow the combine pass to combine an instruction that copies one
1226: register into another. This might or might not produce better
1227: code when used in addition to @samp{-O}. I am interested in
1228: hearing about the difference this makes.
1229:
1230: @item -fforce-mem
1231: Force memory operands to be copied into registers before doing
1232: arithmetic on them. This may produce better code by making all
1233: memory references potential common subexpressions. When they are
1234: not common subexpressions, instruction combination should
1235: eliminate the separate register-load. I am interested in hearing
1236: about the difference this makes.
1237:
1238: @item -fforce-addr
1239: Force memory address constants to be copied into registers before
1240: doing arithmetic on them. This may produce better code just as
1241: @samp{-fforce-mem} may. I am interested in hearing about the
1242: difference this makes.
1243:
1244: @item -fomit-frame-pointer
1245: Don't keep the frame pointer in a register for functions that
1246: don't need one. This avoids the instructions to save, set up and
1247: restore frame pointers; it also makes an extra register available
1248: in many functions. @strong{It also makes debugging impossible.}
1249:
1250: On some machines, such as the Vax, this flag has no effect,
1251: because the standard calling sequence automatically handles the
1252: frame pointer and nothing is saved by pretending it doesn't
1253: exist. The machine-description macro
1254: @code{FRAME_POINTER_REQUIRED} controls whether a target machine
1255: supports this flag. @xref{Registers}.@refill
1256:
1257: @item -finline-functions
1258: Integrate all simple functions into their callers. The compiler
1259: heuristically decides which functions are simple enough to be
1260: worth integrating in this way.
1261:
1262: If all calls to a given function are integrated, and the function
1263: is declared @code{static}, then the function is normally not
1264: output as assembler code in its own right.
1265:
1.1.1.6 root 1266: @item -fcaller-saves
1267: Enable values to be allocated in registers that will be clobbered by
1268: function calls, by emitting extra instructions to save and restore the
1269: registers around such calls. Such allocation is done only when it
1270: seems to result in better code than would otherwise be produced.
1271:
1272: This option is enabled by default on certain machines, usually those
1273: which have no call-preserved registers to use instead.
1274:
1.1 root 1275: @item -fkeep-inline-functions
1276: Even if all calls to a given function are integrated, and the
1277: function is declared @code{static}, nevertheless output a
1278: separate run-time callable version of the function.
1279:
1280: @item -fwritable-strings
1.1.1.8 root 1281: Store string constants in the writable data segment and don't uniquize
1282: them. This is for compatibility with old programs which assume they can
1283: write into string constants. @samp{-traditional} also has this effect.
1284:
1285: Writing into string constants is a very bad idea; ``constants'' should
1286: be constant.
1.1 root 1287:
1.1.1.4 root 1288: @item -fcond-mismatch
1289: Allow conditional expressions with mismatched types in the second and
1290: third arguments. The value of such an expression is void.
1291:
1.1 root 1292: @item -fno-function-cse
1293: Do not put function addresses in registers; make each instruction
1294: that calls a constant function contain the function's address
1295: explicitly.
1296:
1297: This option results in less efficient code, but some strange
1298: hacks that alter the assembler output may be confused by the
1299: optimizations performed when this option is not used.
1300:
1301: @item -fvolatile
1302: Consider all memory references through pointers to be volatile.
1303:
1.1.1.4 root 1304: @item -fshared-data
1305: Requests that the data and non-@code{const} variables of this
1306: compilation be shared data rather than private data. The distinction
1307: makes sense only on certain operating systems, where shared data is
1308: shared between processes running the same program, while private data
1309: exists in one copy per process.
1310:
1.1 root 1311: @item -funsigned-char
1.1.1.4 root 1312: Let the type @code{char} be the unsigned, like @code{unsigned char}.
1.1 root 1313:
1314: Each kind of machine has a default for what @code{char} should
1315: be. It is either like @code{unsigned char} by default or like
1316: @code{signed char} by default. (Actually, at present, the
1317: default is always signed.)
1318:
1319: The type @code{char} is always a distinct type from either
1320: @code{signed char} or @code{unsigned char}, even though its
1321: behavior is always just like one of those two.
1322:
1.1.1.4 root 1323: Note that this is equivalent to @samp{-fno-signed-char}, which is the
1324: negative form of @samp{-fsigned-char}.
1325:
1.1 root 1326: @item -fsigned-char
1327: Let the type @code{char} be signed, like @code{signed char}.
1328:
1.1.1.4 root 1329: Note that this is equivalent to @samp{-fno-unsigned-char}, which is
1330: the negative form of @samp{-funsigned-char}.
1331:
1.1.1.8 root 1332: @item -fdelayed-branch
1333: If supported for the target machine, attempt to reorder instructions
1334: to exploit instruction slots available after delayed branch
1335: instructions.
1336:
1.1 root 1337: @item -ffixed-@var{reg}
1338: Treat the register named @var{reg} as a fixed register; generated
1339: code should never refer to it (except perhaps as a stack pointer,
1340: frame pointer or in some other fixed role).
1341:
1342: @var{reg} must be the name of a register. The register names
1343: accepted are machine-specific and are defined in the
1344: @code{REGISTER_NAMES} macro in the machine description macro
1345: file.
1346:
1.1.1.4 root 1347: This flag does not have a negative form, because it specifies a
1348: three-way choice.
1349:
1.1 root 1350: @item -fcall-used-@var{reg}
1351: Treat the register named @var{reg} as an allocatable register
1352: that is clobbered by function calls. It may be allocated for
1353: temporaries or variables that do not live across a call.
1354: Functions compiled this way will not save and restore the
1355: register @var{reg}.
1356:
1357: Use of this flag for a register that has a fixed pervasive role
1358: in the machine's execution model, such as the stack pointer or
1359: frame pointer, will produce disastrous results.
1360:
1.1.1.4 root 1361: This flag does not have a negative form, because it specifies a
1362: three-way choice.
1363:
1.1 root 1364: @item -fcall-saved-@var{reg}
1365: Treat the register named @var{reg} as an allocatable register
1366: saved by functions. It may be allocated even for temporaries or
1367: variables that live across a call. Functions compiled this way
1368: will save and restore the register @var{reg} if they use it.
1369:
1370: Use of this flag for a register that has a fixed pervasive role
1371: in the machine's execution model, such as the stack pointer or
1372: frame pointer, will produce disastrous results.
1373:
1374: A different sort of disaster will result from the use of this
1375: flag for a register in which function values may be returned.
1.1.1.4 root 1376:
1377: This flag does not have a negative form, because it specifies a
1378: three-way choice.
1.1 root 1379: @end table
1380:
1381: @item -d@var{letters}
1382: Says to make debugging dumps at times specified by @var{letters}.
1383: Here are the possible letters:
1384:
1385: @table @samp
1386: @item r
1387: Dump after RTL generation.
1388: @item j
1389: Dump after first jump optimization.
1390: @item s
1391: Dump after CSE (including the jump optimization that sometimes
1392: follows CSE).
1393: @item L
1394: Dump after loop optimization.
1395: @item f
1396: Dump after flow analysis.
1397: @item c
1398: Dump after instruction combination.
1399: @item l
1400: Dump after local register allocation.
1401: @item g
1402: Dump after global register allocation.
1.1.1.8 root 1403: @item d
1404: Dump after delayed branch scheduling.
1405: @item J
1406: Dump after last jump optimization.
1.1 root 1407: @item m
1408: Print statistics on memory usage, at the end of the run.
1409: @end table
1410:
1411: @item -pedantic
1412: Issue all the warnings demanded by strict ANSI standard C; reject
1413: all programs that use forbidden extensions.
1414:
1415: Valid ANSI standard C programs should compile properly with or without
1416: this option (though a rare few will require @samp{-ansi}). However,
1417: without this option, certain GNU extensions and traditional C features
1418: are supported as well. With this option, they are rejected. There is
1419: no reason to @i{use} this option; it exists only to satisfy pedants.
1.1.1.5 root 1420:
1.1.1.8 root 1421: @samp{-pedantic} does not cause warning messages for use of the
1422: alternate keywords whose names begin and end with @samp{__}.
1423: @xref{Alternate Keywords}.
1424:
1.1.1.5 root 1425: @item -static
1426: On Suns running version 4, this prevents linking with the shared
1427: libraries. (@samp{-g} has the same effect.)
1.1 root 1428: @end table
1429:
1430: These options control the C preprocessor, which is run on each C source
1431: file before actual compilation. If you use the @samp{-E} option, nothing
1432: is done except C preprocessing. Some of these options make sense only
1433: together with @samp{-E} because they request preprocessor output that is
1434: not suitable for actual compilation.
1435:
1436: @table @samp
1437: @item -C
1438: Tell the preprocessor not to discard comments. Used with the
1439: @samp{-E} option.
1440:
1441: @item -I@var{dir}
1442: Search directory @var{dir} for include files.
1443:
1444: @item -I-
1445: Any directories specified with @samp{-I} options before the @samp{-I-}
1446: option are searched only for the case of @samp{#include "@var{file}"};
1447: they are not searched for @samp{#include <@var{file}>}.
1448:
1449: If additional directories are specified with @samp{-I} options after
1450: the @samp{-I-}, these directories are searched for all @samp{#include}
1451: directives. (Ordinarily @emph{all} @samp{-I} directories are used
1452: this way.)
1453:
1454: In addition, the @samp{-I-} option inhibits the use of the current
1.1.1.8 root 1455: directory (where the current input file came from) as the first search
1456: directory for @samp{#include "@var{file}"}. There is no way to override
1457: this effect of @samp{-I-}. With @samp{-I.} you can specify searching
1458: the directory which was current when the compiler was invoked. That is
1459: not exactly the same as what the preprocessor does by default, but it is
1460: often satisfactory.
1461:
1462: @samp{-I-} does not inhibit the use of the standard system directories
1463: for header files. Thus, @samp{-I-} and @samp{-nostdinc} are
1464: independent.
1465:
1466: @item -i @var{file}
1467: Process @var{file} as input, discarding the resulting output, before
1468: processing the regular input file. Because the output generated from
1469: @var{file} is discarded, the only effect of @samp{-i @var{file}} is to
1470: make the macros defined in @var{file} available for use in the main
1471: input.
1.1 root 1472:
1473: @item -nostdinc
1474: Do not search the standard system directories for header files. Only
1475: the directories you have specified with @samp{-I} options (and the
1476: current directory, if appropriate) are searched.
1477:
1478: Between @samp{-nostdinc} and @samp{-I-}, you can eliminate all
1479: directories from the search path except those you specify.
1480:
1481: @item -M
1482: Tell the preprocessor to output a rule suitable for @code{make}
1.1.1.10! root 1483: describing the dependencies of each object file. For each source
1.1 root 1484: file, the preprocessor outputs one @code{make}-rule whose target is
1485: the object file name for that source file and whose dependencies are
1486: all the files @samp{#include}d in it. This rule may be a single line
1487: or may be continued with @samp{\}-newline if it is long.
1488:
1489: @samp{-M} implies @samp{-E}.
1490:
1491: @item -MM
1492: Like @samp{-M} but the output mentions only the user-header files
1493: included with @samp{#include "@var{file}"}. System header files
1494: included with @samp{#include <@var{file}>} are omitted.
1495:
1496: @samp{-MM} implies @samp{-E}.
1497:
1498: @item -D@var{macro}
1.1.1.8 root 1499: Define macro @var{macro} with the string @samp{1} as its definition.
1.1 root 1500:
1501: @item -D@var{macro}=@var{defn}
1502: Define macro @var{macro} as @var{defn}.
1503:
1504: @item -U@var{macro}
1505: Undefine macro @var{macro}.
1506:
1.1.1.7 root 1507: @item -trigraphs
1.1 root 1508: Support ANSI C trigraphs. You don't want to know about this
1509: brain-damage. The @samp{-ansi} option also has this effect.
1510: @end table
1511:
1512: @node Installation, Trouble, Options, Top
1513: @chapter Installing GNU CC
1514:
1515: Here is the procedure for installing GNU CC on a Unix system.
1.1.1.8 root 1516:
1.1 root 1517: @menu
1.1.1.8 root 1518: * Other Dir:: Compiling in a separate directory (not where the source is).
1519: * Sun Install:: See below for installation on the Sun.
1520: * 3B1 Install:: See below for installation on the 3B1.
1.1.1.10! root 1521: * SCO Install:: See below for installation on SCO System V 3.2.
1.1 root 1522: * VMS Install:: See below for installation on VMS.
1.1.1.9 root 1523: * HPUX Install:: See below for installation on HPUX.
1.1.1.10! root 1524: * MIPS Install:: See below for installation on MIPS.
1.1 root 1525: @end menu
1526: @iftex
1.1.1.9 root 1527: See below for VMS systems, and modified procedures needed on Sun
1528: systems, 3b1 machines and HPUX. The following section says how to
1529: compile in a separate directory on Unix; here we assume you compile in
1530: the same directory that contains the source files.
1.1 root 1531: @end iftex
1532:
1533: @enumerate
1534: @item
1535: Edit @file{Makefile}. If you are using HPUX, or any form of system V,
1536: you must make a few changes described in comments at the beginning of
1.1.1.9 root 1537: the file. Genix requires changes also, and so does the Pyramid.
1.1 root 1538:
1539: @item
1540: On a Sequent system, go to the Berkeley universe.
1541:
1542: @item
1.1.1.2 root 1543: Choose configuration files. The easy way to do this is to run the
1.1.1.8 root 1544: command file @file{config.gcc} with a single argument, which specifies
1545: the type of machine (and in some cases which operating system).
1.1.1.4 root 1546:
1547: Here is a list of the possible arguments:
1548:
1549: @table @samp
1550: @item vax
1551: Vaxes running BSD.
1552: @item vms
1553: Vaxes running VMS.
1554: @item vax-sysv
1555: Vaxes running system V.
1556: @item i386-sysv
1557: Intel 386 PCs running system V.
1.1.1.5 root 1558: @item i386-sysv-gas
1559: Intel 386 PCs running system V, using the GNU assembler and GNU
1560: linker.
1.1.1.6 root 1561: @item sequent-i386
1.1.1.4 root 1562: Sequent with Intel 386 processors.
1.1.1.8 root 1563: @item i386-aix
1564: Intel 386 PCs or PS/2s running AIX.
1.1.1.4 root 1565: @item sun2
1566: Sun 2 running system version 2 or 3.
1567: @item sun3
1.1.1.5 root 1568: Sun 3 running system version 2 or 3, with 68881.
1.1.1.7 root 1569: Note there we do not provide a configuration file to use an FPA
1.1.1.8 root 1570: by default, because programs that establish signal handlers for
1.1.1.7 root 1571: floating point traps inherently cannot work with the FPA.
1.1.1.5 root 1572: @item sun3-nfp
1573: Sun 3 running system version 2 or 3, without 68881.
1.1.1.4 root 1574: @item sun4
1.1.1.8 root 1575: Sun 4 running system version 2 or 3. @xref{Incompatibilities},
1576: for calling convention incompatibilities on the Sun 4 (sparc).
1.1.1.4 root 1577: @item sun2-os4
1578: Sun 2 running system version 4.
1579: @item sun3-os4
1.1.1.5 root 1580: Sun 3 running system version 4, with 68881.
1581: @item sun3-nfp-os4
1582: Sun 3 running system version 4, without 68881.
1.1.1.4 root 1583: @item sun4-os4
1.1.1.8 root 1584: Sun 4 running system version 4. @xref{Incompatibilities},
1585: for calling convention incompatibilities on the Sun 4 (sparc).
1.1.1.4 root 1586: @item sun386
1587: Sun 386 (``roadrunner'').
1.1.1.5 root 1588: @item alliant
1.1.1.8 root 1589: Alliant FX/8 computer. Note that the standard installed C compiler in
1590: Concentrix 5.0 has a bug which prevent it from compiling GNU CC
1591: correctly. You can patch the compiler bug as follows:
1592:
1593: @example
1594: cp /bin/pcc ./pcc
1.1.1.9 root 1595: adb -w ./pcc - << EOF
1.1.1.8 root 1596: 15f6?w 6610
1597: EOF
1598: @end example
1599:
1600: Then you must use the @samp{-ip12} option when compiling GNU CC
1601: with the patched compiler, as shown here:
1602:
1603: @example
1604: make CC="./pcc -ip12" CFLAGS=-w
1605: @end example
1606:
1607: Note also that Alliant's version of DBX does not manage to work with the
1608: output from GNU CC.
1609: @item tahoe
1610: The tahoe computer (running BSD, and using DBX).
1611: @item decstation
1612: The DEC 3100 Mips machine (``pmax''). Note that GNU CC cannot generate
1613: debugging information in the unusual format used on the Mips.
1614: @item mips-sysv
1615: The Mips computer, RS series, with the System V environment as default.
1616: Note that GNU CC cannot generate debugging information in the unusual
1617: format used on the Mips.
1618: @item mips-bsd43
1619: The Mips computer, RS series, with the BSD 4.3 environment as default.
1620: Note that GNU CC cannot generate debugging information in the unusual
1621: format used on the Mips.
1.1.1.7 root 1622: @item mips
1.1.1.8 root 1623: The Mips computer, M series. Note that GNU CC cannot generate debugging
1624: information in the unusual format used on the Mips.
1625: @item iris
1.1.1.10! root 1626: Another variant of the Mips computer, the Silicon Graphics Iris 4D.
! 1627: Note that GNU CC cannot generate debugging information in the unusual
! 1628: format used on the Mips.
1.1.1.5 root 1629: @item convex-c1
1630: Convex C1 computer.
1631: @item convex-c2
1632: Convex C2 computer.
1.1.1.8 root 1633: @item pyramid
1634: Pyramid computer.
1.1.1.4 root 1635: @item hp9k320
1.1.1.7 root 1636: HP 9000 series 300 using HPUX assembler. Note there is no
1637: support in GNU CC for HP's debugger; thus, @samp{-g} is not
1638: available in this configuration.
1.1.1.8 root 1639: @item hp9k320-gas
1.1.1.4 root 1640: HP 9000 series 300 using GNU assembler, linker and debugger.
1.1.1.7 root 1641: This requires the HP-adapt package, which is available along with
1642: the GNU linker as part of the ``binutils'' distribution.
1643: This is on the GNU CC distribution tape.
1.1.1.8 root 1644: @item hp9k320-old
1645: HP 9000 series 300 using HPUX assembler, in operating system versions
1646: older than 6.5. Note there is no support in GNU CC for HP's debugger;
1647: thus, @samp{-g} is not available in this configuration.
1648: @item hp9k320-bsd
1649: HP 9000 series 300 running BSD.
1.1.1.4 root 1650: @item isi68
1.1.1.8 root 1651: ISI 68000 or 68020 system with a 68881.
1652: @item isi68-nfp
1653: ISI 68000 or 68020 system without a 68881.
1.1.1.4 root 1654: @item news800
1655: Sony NEWS 68020 system.
1.1.1.6 root 1656: @item next
1657: NeXT system.
1.1.1.10! root 1658: @item tower
! 1659: NCR Tower 32 system.
1.1.1.7 root 1660: @item altos
1661: Altos 3068. Note that you must use the GNU assembler, linker and
1662: debugger, with COFF-encapsulation. Also, you must fix a kernel
1663: bug. Details in the file @file{ALTOS-README}.
1.1.1.4 root 1664: @item 3b1
1.1.1.8 root 1665: AT&T 3b1, a.k.a. 7300 PC. Note that special procedures are needed
1666: to compile GNU CC with this machine's standard C compiler, due to
1667: bugs in that compiler. @xref{3b1 Install}. You can bootstrap it
1668: more easily with previous versions of GNU CC if you have them.
1.1.1.9 root 1669: @item 3b1-gas
1670: AT&T 3b1 using the GNU assembler.
1.1.1.4 root 1671: @item sequent-ns32k
1672: Sequent containing ns32000 processors.
1673: @item encore
1674: Encore ns32000 system.
1675: @item genix
1676: National Semiconductor ns32000 system.
1677: @item 88000
1678: Motorola 88000 processor. This port is not finished.
1679: @end table
1.1.1.2 root 1680:
1.1.1.4 root 1681: Here we spell out what files need to be set up:
1.1 root 1682:
1683: @itemize @bullet
1684: @item
1685: Make a symbolic link named @file{config.h} to the top-level
1686: config file for the machine you are using (@pxref{Config}). This
1687: file is responsible for defining information about the host
1688: machine. It includes @file{tm.h}.
1689:
1.1.1.7 root 1690: The file is located in the subdirectory @file{config}. Its name
1691: should be @file{xm-@var{machine}.h}, with these exceptions:
1.1 root 1692:
1693: @table @file
1.1.1.3 root 1694: @item xm-vms.h
1.1 root 1695: for vaxen running VMS.
1.1.1.3 root 1696: @item xm-vaxv.h
1.1 root 1697: for vaxen running system V.
1.1.1.3 root 1698: @item xm-i386v.h
1.1 root 1699: for Intel 80386's running system V.
1.1.1.3 root 1700: @item xm-sun386i.h
1701: for Sun roadrunner running any version of the operating system.
1702: @item xm-hp9k320.h
1.1 root 1703: for the HP 9000 series 300.
1.1.1.4 root 1704: @item xm-genix.h
1.1 root 1705: for the ns32000 running Genix
1706: @end table
1707:
1708: If your system does not support symbolic links, you might want to
1709: set up @file{config.h} to contain a @samp{#include} command which
1710: refers to the appropriate file.
1711:
1712: @item
1713: Make a symbolic link named @file{tm.h} to the machine-description
1.1.1.7 root 1714: macro file for your machine. It should be in the subdirectory
1715: @file{config} and its name should be @file{tm-@var{machine}.h}.
1.1 root 1716:
1717: If your system is a 68000, don't use the file @file{tm-m68k.h}
1718: directly. Instead, use one of these files:
1719:
1720: @table @file
1721: @item tm-sun3.h
1.1.1.5 root 1722: for Sun 3 machines with 68881.
1723: @item tm-sun3-nfp.h
1724: for Sun 3 machines with no hardware floating point.
1.1.1.8 root 1725: @item tm-sun3os3.h
1726: for Sun 3 machines with 68881, running Sunos version 3.
1727: @item tm-sun3os3nf.h
1728: for Sun 3 machines with no hardware floating point, running Sunos
1729: version 3.
1.1 root 1730: @item tm-sun2.h
1731: for Sun 2 machines.
1732: @item tm-3b1.h
1733: for AT&T 3b1 (aka 7300 Unix PC).
1734: @item tm-isi68.h
1.1.1.3 root 1735: for Integrated Solutions systems. This file assumes you
1736: use the GNU assembler.
1.1.1.8 root 1737: @item tm-isi68-nfp.h
1738: for Integrated Solutions systems without a 68881. This file assumes you
1739: use the GNU assembler.
1.1 root 1740: @item tm-news800.h
1.1.1.8 root 1741: for Sony NEWS systems.
1.1 root 1742: @item tm-hp9k320.h
1743: for HPUX systems, if you are using GNU CC with the system's
1744: assembler and linker.
1745: @item tm-hp9k320g.h
1746: for HPUX systems, if you are using the GNU assembler, linker and
1747: other utilities. Not all of the pieces of GNU software needed
1748: for this mode of operation are as yet in distribution; full
1749: instructions will appear here in the future.@refill
1.1.1.10! root 1750: @item tm-tower-as.h
! 1751: for NCR Tower 32 systems, using the standard system assembler.
1.1 root 1752: @end table
1753:
1754: For the vax, use @file{tm-vax.h} on BSD Unix, @file{tm-vaxv.h} on
1755: system V, or @file{tm-vms.h} on VMS.@refill
1756:
1757: For the Motorola 88000, use @file{tm-m88k.h}. The support for the
1.1.1.9 root 1758: 88000 does not currently work; it requires extensive changes which
1759: we hope to reconcile in version 2.
1.1 root 1760:
1761: For the 80386, don't use @file{tm-i386.h} directly. Use
1762: @file{tm-i386v.h} if the target machine is running system V,
1.1.1.5 root 1763: @file{tm-i386gas.h} if it is running system V but you are using the
1764: GNU assembler and linker, @file{tm-seq386.h} for a Sequent 386 system,
1765: or @file{tm-compaq.h} for a Compaq, or @file{tm-sun386i.h} for a Sun
1766: 386 system.
1.1 root 1767:
1.1.1.8 root 1768: For the Mips computer, there are five choices: @file{tm-mips.h} for the
1769: M series, @file{tm-mips-bsd.h} for the RS series with BSD,
1770: @file{tm-mips-sysv.h} for the RS series with System V, @file{tm-iris.h}
1771: for the Iris version of the machine, and @file{tm-decstatn.h} for the
1772: Decstation.
1773:
1.1 root 1774: For the 32000, use @file{tm-sequent.h} if you are using a Sequent
1775: machine, or @file{tm-encore.h} for an Encore machine, or
1.1.1.4 root 1776: @file{tm-genix.h} if you are using Genix version 3; otherwise, perhaps
1.1 root 1777: @file{tm-ns32k.h} will work for you.
1778:
1779: Note that Genix has bugs in @code{alloca} and @code{malloc}; you must
1780: get the compiled versions of these from GNU Emacs and edit GNU CC's
1781: @file{Makefile} to use them.
1782:
1783: Note that Encore systems are supported only under BSD.
1784:
1.1.1.6 root 1785: For Sparc (Sun 4) machines, use @file{tm-sparc.h} with operating system
1786: version 4, and @file{tm-sun4os3.h} with system version 3.
1787:
1.1.1.10! root 1788: For Convex systems before version 8.1, use @file{tm-conv1os7.h} or
! 1789: @file{tm-conv2os7.h}. For versions 8.1 and greater, use @file{tm-convex1.h}
! 1790: or @file{tm-convex2.h}. You should also bootstrap GCC with @code{pcc}
! 1791: rather than @code{cc}; one way to do this is with the following commands.
! 1792:
! 1793: @example
! 1794: ln -s /bin/pcc ./cc
! 1795: set path = (. $path)
! 1796: @end example
! 1797:
1.1 root 1798: @item
1799: Make a symbolic link named @file{md} to the machine description
1.1.1.7 root 1800: pattern file. It should be in the @file{config} subdirectory and its
1801: name should be @file{@var{machine}.md}; but @var{machine} is often not
1802: the same as the name used in the @file{tm.h} file because the
1803: @file{md} files are more general.
1.1 root 1804:
1805: @item
1806: Make a symbolic link named @file{aux-output.c} to the output
1.1.1.7 root 1807: subroutine file for your machine. It should be in the @file{config}
1808: subdirectory and its name should be @file{out-@var{machine}.c}.
1.1 root 1809: @end itemize
1810:
1811: @item
1812: Make sure the Bison parser generator is installed. (This is
1813: unnecessary if the Bison output files @file{c-parse.tab.c} and
1814: @file{cexp.c} are more recent than @file{c-parse.y} and @file{cexp.y}
1815: and you do not plan to change the @samp{.y} files.)
1816:
1.1.1.9 root 1817: Bison versions older than Sept 8, 1988 will produce incorrect output
1.1 root 1818: for @file{c-parse.tab.c}.
1819:
1820: @item
1.1.1.10! root 1821: If you have a previous version of GCC installed, then chances are
! 1822: you can compile the new version with that. Do the following:
! 1823:
! 1824: @example
! 1825: make CC="gcc -O"
! 1826: @end example
! 1827:
! 1828: @noindent
! 1829: Since this produces an optimized executable right away, there is no need
! 1830: to bootstrap the result with itself except to test it. Therefore, you can
! 1831: skip directly to the @samp{make install} step below.
! 1832:
! 1833: @item
1.1 root 1834: Build the compiler. Just type @samp{make} in the compiler directory.
1835:
1.1.1.2 root 1836: Ignore any warnings you may see about ``statement not reached'' in the
1837: @file{insn-emit.c}; they are normal. Any other compilation errors may
1838: represent bugs in the port to your machine or operating system, and
1839: should be investigated and reported (@pxref{Bugs}).
1840:
1.1.1.9 root 1841: Some commercial compilers fail to compile GNU CC because they have bugs
1842: or limitations. For example, the Microsoft compiler is said to run out
1843: of macro space. Some Ultrix compilers run out of expression space; then
1844: you need to break up the statement where the problem happens.
1.1.1.7 root 1845:
1846: @item
1.1.1.5 root 1847: If you are using COFF-encapsulation, you must convert @file{gnulib} to
1848: a GNU-format library at this point. See the file @file{README-ENCAP}
1849: in the directory containing the GNU binary file utilities, for
1850: directions.
1851:
1852: @item
1.1 root 1853: Move the first-stage object files and executables into a subdirectory
1854: with this command:
1855:
1856: @example
1857: make stage1
1858: @end example
1859:
1860: The files are moved into a subdirectory named @file{stage1}.
1861: Once installation is complete, you may wish to delete these files
1862: with @code{rm -r stage1}.
1863:
1864: @item
1865: Recompile the compiler with itself, with this command:
1866:
1867: @example
1868: make CC=stage1/gcc CFLAGS="-g -O -Bstage1/"
1869: @end example
1870:
1.1.1.10! root 1871: This is called making the stage 2 compiler.
! 1872:
1.1 root 1873: On a 68000 or 68020 system lacking floating point hardware,
1874: unless you have selected a @file{tm.h} file that expects by default
1875: that there is no such hardware, do this instead:
1876:
1877: @example
1878: make CC=stage1/gcc CFLAGS="-g -O -Bstage1/ -msoft-float"
1879: @end example
1880:
1881: @item
1882: If you wish to test the compiler by compiling it with itself one more
1.1.1.7 root 1883: time, do this (in C shell):
1.1 root 1884:
1885: @example
1886: make stage2
1887: make CC=stage2/gcc CFLAGS="-g -O -Bstage2/"
1888: foreach file (*.o)
1889: cmp $file stage2/$file
1890: end
1891: @end example
1892:
1.1.1.7 root 1893: @noindent
1.1.1.10! root 1894: This is called making the stage 3 compiler. Aside from the @samp{-B}
! 1895: option, the options should be the same as when you made the stage 2
! 1896: compiler.
1.1 root 1897:
1.1.1.7 root 1898: The @code{foreach} command (written in C shell) will notify you if any of
1899: these stage 3 object files differs from those of stage 2. On BSD systems,
1900: any difference, no matter how innocuous, indicates that the stage 2
1901: compiler has compiled GNU CC incorrectly, and is therefore a potentially
1902: serious bug which you should investigate and report (@pxref{Bugs}).
1903:
1904: On systems that use COFF object files, bytes 5 to 8 will always be
1905: different, since it is a timestamp. On these systems, you can do the
1906: comparison as follows (in Bourne shell):
1907:
1908: @example
1909: for file in *.o; do
1910: echo $file
1.1.1.10! root 1911: tail +10c $file > foo1
! 1912: tail +10c stage2/$file > foo2
1.1.1.7 root 1913: cmp foo1 foo2
1914: done
1915: @end example
1916:
1.1.1.10! root 1917: On MIPS machines, you should use the shell script @file{ecoff-cmp}
! 1918: to compare two object files.
! 1919:
1.1 root 1920: @item
1921: Install the compiler driver, the compiler's passes and run-time support.
1922: You can use the following command:
1923:
1924: @example
1925: make install
1926: @end example
1927:
1928: @noindent
1929: This copies the files @file{cc1}, @file{cpp} and @file{gnulib} to
1930: files @file{gcc-cc1}, @file{gcc-cpp} and @file{gcc-gnulib} in
1931: directory @file{/usr/local/lib}, which is where the compiler driver
1932: program looks for them. It also copies the driver program @file{gcc}
1.1.1.6 root 1933: into the directory @file{/usr/local/bin}, so that it appears in typical
1.1 root 1934: execution search paths.@refill
1935:
1936: @strong{Warning: there is a bug in @code{alloca} in the Sun library.
1937: To avoid this bug, install the binaries of GNU CC that were compiled
1938: by GNU CC. They use @code{alloca} as a built-in function and never
1939: the one in the library.}
1940:
1941: @strong{Warning: the GNU CPP may not work for @file{ioctl.h},
1942: @file{ttychars.h} and other system header files unless the
1943: @samp{-traditional} option is used.} The bug is in the header files:
1944: at least on some machines, they rely on behavior that is incompatible
1945: with ANSI C. This behavior consists of substituting for macro
1946: argument names when they appear inside of character constants. The
1947: @samp{-traditional} option tells GNU CC to behave the way these
1948: headers expect.
1949:
1950: Because of this problem, you might prefer to configure GNU CC to use
1951: the system's own C preprocessor. To do so, make the file
1952: @file{/usr/local/lib/gcc-cpp} a link to @file{/lib/cpp}.
1953:
1954: Alternatively, on Sun systems and 4.3BSD at least, you can correct the
1955: include files by running the shell script @file{fixincludes}. This
1956: installs modified, corrected copies of the files @file{ioctl.h},
1957: @file{ttychars.h} and many others, in a special directory where only
1.1.1.2 root 1958: GNU CC will normally look for them. This script will work on various
1.1.1.6 root 1959: systems because it chooses the files by searching all the system
1.1.1.2 root 1960: headers for the problem cases that we know about.
1.1.1.10! root 1961:
! 1962: Use the following command to do this:
! 1963:
! 1964: @example
! 1965: make includes
! 1966: @end example
! 1967:
! 1968: @noindent
! 1969: If you selected a different directory for GNU CC installation when you
! 1970: installed it, by specifying the Make variable @code{prefix} or
! 1971: @code{libdir}, specify it the same way in this command.
! 1972:
! 1973: Note that some systems are starting to come with ANSI C system header
! 1974: files. On these systems, don't run @file{fixincludes}; it may not work,
! 1975: and is certainly not necessary.
1.1 root 1976: @end enumerate
1977:
1978: If you cannot install the compiler's passes and run-time support in
1979: @file{/usr/local/lib}, you can alternatively use the @samp{-B} option to
1980: specify a prefix by which they may be found. The compiler concatenates
1981: the prefix with the names @file{cpp}, @file{cc1} and @file{gnulib}.
1982: Thus, you can put the files in a directory @file{/usr/foo/gcc} and
1983: specify @samp{-B/usr/foo/gcc/} when you run GNU CC.
1984:
1985: Also, you can specify an alternative default directory for these files
1986: by setting the Make variable @code{libdir} when you make GNU CC.
1987:
1.1.1.8 root 1988: @node Other Dir, Sun Install, Installation, Installation
1989: @section Compilation in a Separate Directory
1.1 root 1990:
1.1.1.8 root 1991: If you wish to build the object files and executables in a directory
1992: other than the one containing the source files, here is what you must
1993: do differently:
1994:
1995: @enumerate
1996: @item
1997: Go to that directory before running @file{config.gcc}:
1998:
1999: @example
2000: mkdir gcc-sun3
2001: cd gcc-sun3
2002: @end example
1.1.1.4 root 2003:
1.1.1.8 root 2004: On systems that do not support symbolic links, this directory must be
2005: on the same file system as the source code directory.
2006:
2007: @item
2008: Specify where to find @file{config.gcc} when you run it:
2009:
2010: @example
2011: ../gcc-1.36/config.gcc @dots{}
2012: @end example
2013:
2014: @item
2015: Specify where to find the sources, as an argument to @file{config.gcc}:
2016:
2017: @example
2018: ../gcc-1.36/config.gcc -srcdir=../gcc-1.36 sun3
2019: @end example
2020:
2021: The @samp{-srcdir=@var{dir}} option is not needed when the source
2022: directory is the parent of the current directory, because
2023: @file{config.gcc} detects that case automatically.
2024: @end enumerate
2025:
2026: Now, you can run @code{make} in that directory. You need not repeat the
2027: configuration steps shown above, when ordinary source files change. You
2028: must, however, run @code{config.gcc} again when the configuration files
2029: change, if your system does not support symbolic links.
2030:
2031: @node Sun Install, 3b1 Install, Other Dir, Installation
2032: @section Installing GNU CC on the Sun
2033:
2034: Make sure the environment variable @code{FLOAT_OPTION} is not set when
2035: you compile @file{gnulib}. If this option were set to @code{f68881}
2036: when @file{gnulib} is compiled, the resulting code would demand to be
2037: linked with a special startup file and would not link properly without
2038: special pains.
2039:
2040: There is a bug in @code{alloca} in certain versions of the Sun library.
2041: To avoid this bug, install the binaries of GNU CC that were compiled by
2042: GNU CC. They use @code{alloca} as a built-in function and never the one
2043: in the library.
2044:
2045: Some versions of the Sun compiler crash when compiling GNU CC.
2046: The problem is a segmentation fault in cpp.
2047:
2048: This problem seems to be due to the bulk of data in the environment
2049: variables. You may be able to avoid it by using the following
2050: command to compile GNU CC with Sun CC:
2051:
2052: @example
2053: make CC="TERMCAP=x OBJS=x LIBFUNCS=x STAGESTUFF=x cc"
2054: @end example
2055:
1.1.1.10! root 2056: @node 3b1 Install, SCO Install, Sun Install, Installation
1.1.1.8 root 2057: @section Installing GNU CC on the 3b1
2058:
2059: Installing GNU CC on the 3b1 is difficult if you do not already have
2060: GNU CC running, due to bugs in the installed C compiler. However,
2061: the following procedure might work. We are unable to test it.
1.1 root 2062:
2063: @enumerate
2064: @item
1.1.1.8 root 2065: Comment out the @samp{#include "config.h"} line on line 37 of
2066: @file{cccp.c} and do @samp{make cpp}. This makes a preliminary version
2067: of GNU cpp.
1.1 root 2068:
2069: @item
1.1.1.8 root 2070: Save the old @file{/lib/cpp} and copy the preliminary GNU cpp to that
2071: file name.
1.1.1.5 root 2072:
1.1.1.8 root 2073: @item
2074: Undo your change in @file{cccp.c}, or reinstall the original version,
2075: and do @samp{make cpp} again.
2076:
2077: @item
2078: Copy this final version of GNU cpp into @file{/lib/cpp}.
2079:
2080: @item
1.1.1.9 root 2081: Replace every occurrence of @code{obstack_free} in @file{tree.c}
1.1.1.8 root 2082: with @code{_obstack_free}.
2083:
2084: @item
2085: Run @code{make} to get the first-stage GNU CC.
2086:
2087: @item
2088: Reinstall the original version of @file{/lib/cpp}.
2089:
2090: @item
2091: Now you can compile GNU CC with itself and install it in the normal
2092: fashion.
1.1 root 2093: @end enumerate
2094:
1.1.1.9 root 2095: If you have installed an earlier version of GCC, you can compile the
2096: newer version with that. However, you will run into trouble compiling
2097: @file{gnulib}, since that is normally compiled with CC. To solve the
2098: problem, uncomment this line in @file{Makefile}:
2099:
2100: @example
2101: CCLIBFLAGS = -B/usr/local/lib/gcc- -tp -Wp,-traditional
2102: @end example
2103:
1.1.1.10! root 2104: @node SCO Install, VMS Install, 3B1 Install, Installation
! 2105: @section Installing GNU CC on SCO System V 3.2
! 2106: @cindex Installation on SCO systems
! 2107:
! 2108: The compiler that comes with this system does not work properly with
! 2109: @samp{-O}. Therefore, you should redefine the Make variable
! 2110: @code{CCLIBFLAGS} not to use @samp{-O}.
! 2111:
! 2112: In addition, the compiler produces incorrect output when compiling parts
! 2113: of GNU CC; the resulting executable @file{cc1} does not work properly
! 2114: when it is used with @samp{-O}.
! 2115:
! 2116: Therefore, what you must do after building the first stage
! 2117: is use GNU CC to compile itself without optimization. Here is how:
! 2118:
! 2119: @example
! 2120: make -k cc1 CC="./gcc -B./"
! 2121: @end example
! 2122:
! 2123: You can think of this as ``stage 1.1'' of the installation process.
! 2124: However, using this command has the effect of discarding the faulty
! 2125: stage 1 executable for @file{cc1} and replacing it with stage 1.1. You
! 2126: can then proceed with @samp{make stage1} and the rest of installation.
! 2127:
! 2128: @node VMS Install, HPUX Install, SCO Install, Installation
1.1.1.8 root 2129: @section Installing GNU CC on VMS
2130:
2131: The VMS version of GNU CC is distributed in a backup saveset containing
2132: both source code and precompiled binaries.
2133:
2134: To install the @file{gcc} command so you can use the compiler easily, in
1.1 root 2135: the same manner as you use the VMS C compiler, you must install the VMS CLD
2136: file for GNU CC as follows:
2137:
2138: @enumerate
2139: @item
2140: Define the VMS logical names @samp{GNU_CC} and @samp{GNU_CC_INCLUDE}
2141: to point to the directories where the GNU CC executables
1.1.1.8 root 2142: (@file{gcc-cpp}, @file{gcc-cc1}, etc.) and the C include files are
1.1 root 2143: kept. This should be done with the commands:@refill
2144:
2145: @example
1.1.1.8 root 2146: $ assign /super /system disk:[gcc.] gnu_cc
2147: $ assign /super /system disk:[gcc.include.] gnu_cc_include
1.1 root 2148: @end example
2149:
2150: @noindent
2151: with the appropriate disk and directory names. These commands can be
2152: placed in your system startup file so they will be executed whenever
1.1.1.8 root 2153: the machine is rebooted. You may, if you choose, do this via the
2154: @file{GCC_INSTALL.COM} script in the @file{[GCC]} directory.
1.1 root 2155:
2156: @item
1.1.1.8 root 2157: Install the @file{GCC} command with the command line:
1.1 root 2158:
2159: @example
1.1.1.8 root 2160: $ set command /table=sys$library:dcltables gnu_cc:[000000]gcc
1.1 root 2161: @end example
2162:
1.1.1.7 root 2163: @item
2164: To install the help file, do the following:
2165:
2166: @example
2167: $ lib/help sys$library:helplib.hlb gcc.hlp
2168: @end example
2169:
1.1 root 2170: @noindent
2171: Now you can invoke the compiler with a command like @samp{gcc /verbose
2172: file.c}, which is equivalent to the command @samp{gcc -v -c file.c} in
2173: Unix.
2174: @end enumerate
2175:
1.1.1.8 root 2176: We try to put corresponding binaries and sources on the VMS distribution
2177: tape. But sometimes the binaries will be from an older version that the
2178: sources, because we don't always have time to update them. (Use the
2179: @samp{/verbose} option to determine the version number of the binaries and
2180: compare it with the source file @file{version.c} to tell whether this is
2181: so.) In this case, you should use the binaries you get to recompile the
2182: sources. If you must recompile, here is how:
2183:
2184: @enumerate
2185: @item
2186: Copy the file @file{tm-vms.h} to @file{tm.h}, @file{xm-vms.h} to
2187: @file{config.h}, @file{vax.md} to @file{md.} and @file{out-vax.c}
2188: to @file{aux-output.c}. The files to be copied are found in the
2189: subdirectory named @file{config}; they should be copied to the
2190: main directory of GNU CC.@refill
2191:
2192: @item
2193: Setup the logical names and command tables as defined above. In
2194: addition, define the vms logical name @samp{GNU_BISON} to point at the
2195: to the directories where the Bison executable is kept. This should be
2196: done with the command:@refill
2197:
2198: @example
2199: $ assign /super /system disk:[bison.] gnu_bison
2200: @end example
2201:
2202: You may, if you choose, use the @file{INSTALL_BISON.COM} script in the
2203: @file{[BISON]} directory.
2204:
2205: @item
2206: Install the @samp{BISON} command with the command line:@refill
2207:
2208: @example
2209: $ set command /table=sys$library:dcltables gnu_bison:[000000]bison
2210: @end example
2211:
2212: @item
2213: Type @samp{@@make} to do recompile everything.
2214:
2215: If you are compiling with a version of GNU CC older than 1.33, specify
2216: @samp{/DEFINE=("inline=")} as an option in all the compilations. This
2217: requires editing all the @code{gcc} commands in @file{make-cc1.com}.
2218: (The older versions had problems supporting @code{inline}.) Once you
2219: have a working 1.33 or newer GNU CC, you can change this file back.
2220: @end enumerate
2221:
1.1.1.10! root 2222: With this version of GNU CC, @code{const} global variables now work
! 2223: properly. Unless, however, the @code{const} modifier is also specified
! 2224: in every external declaration of the variable in all of the source files
! 2225: that use that variable, the linker will issue warnings about conflicting
! 2226: attributes for the variable, since the linker does not know if the
! 2227: variable should be read-only. The program will still work, but the
! 2228: variable will be placed in writable storage.
! 2229:
! 2230: Under previous versions of GNU CC, the generated code would occasionally
! 2231: give strange results when linked to the sharable @file{VAXCRTL} library.
! 2232: Now this should work.
! 2233:
! 2234: Even with this version, however, GNU CC itself should not be linked to
! 2235: the sharable @file{VAXCRTL}, unless you force the linker to use the
! 2236: @code{qsort} routine from @file{gcclib.olb}. The @file{qsort} routine
! 2237: supplied with @file{VAXCRTL} has a bug which causes a compiler crash.
! 2238: The executable that is generated by @file{make-cc1.com} uses the
! 2239: non-shared version of @file{VAXCRTL} (and thus the @file{qsort} routine
! 2240: from @file{gcclib.olb}).
! 2241:
! 2242: Note that GNU CC on VMS now generates debugging information to describe
! 2243: the programs symbols to the VMS debugger. However, you need version 1.37
! 2244: or later of GAS in order to output them properly in the object file.
1.1.1.7 root 2245:
1.1.1.10! root 2246: @node HPUX Install, MIPS Install, VMS Install, Installation
1.1.1.9 root 2247: @section Installing GNU CC on HPUX
2248:
2249: To install GNU CC on HPUX, you must start by editing the file
2250: @file{Makefile}. Search for the string @samp{HPUX} to find comments
2251: saying what to change. You need to change some variable definitions and
2252: (if you are using GAS) some lines in the rule for the target
2253: @samp{gnulib}.
2254:
1.1.1.10! root 2255: To avoid errors when linking programs with @samp{-g}, create an empty
! 2256: library named @file{libg.a}. An easy way to do this is:
! 2257:
! 2258: @example
! 2259: ar rc /usr/local/lib/libg.a
! 2260: @end example
! 2261:
1.1.1.9 root 2262: To compile with the HPUX C compiler, you must specify get the file
2263: @file{alloca.c} from GNU Emacs. Then, when you run @code{make}, use
2264: this argument:
2265:
2266: @example
2267: make ALLOCA=alloca.o
2268: @end example
2269:
2270: When recompiling GNU CC with itself, do not define @code{ALLOCA}.
2271: Instead, an @samp{-I} option needs to be added to @code{CFLAGS} as
2272: follows:
2273:
2274: @example
2275: make CC=stage1/gcc CFLAGS="-g -O -Bstage1/ -I../binutils/hp-include"
2276: @end example
2277:
1.1.1.10! root 2278: @node MIPS Install,, HPUX Install, Installation
! 2279: @section Installing GNU CC on MIPS
! 2280:
! 2281: To avoid errors when linking programs with @samp{-g}, create an empty
! 2282: library named @file{libg.a}. An easy way to do this is:
! 2283:
! 2284: @example
! 2285: ar rc /usr/local/lib/libg.a
! 2286: @end example
! 2287:
! 2288: @node Trouble, Service, Installation, Top
! 2289: @chapter Known Causes of Trouble with GNU CC
1.1 root 2290:
2291: Here are some of the things that have caused trouble for people installing
2292: or using GNU CC.
2293:
2294: @itemize @bullet
2295: @item
2296: On certain systems, defining certain environment variables such as
1.1.1.8 root 2297: @code{CC} can interfere with the functioning of @code{make}.
1.1 root 2298:
2299: @item
2300: Cross compilation can run into trouble for certain machines because
2301: some target machines' assemblers require floating point numbers to be
2302: written as @emph{integer} constants in certain contexts.
2303:
2304: The compiler writes these integer constants by examining the floating
2305: point value as an integer and printing that integer, because this is
2306: simple to write and independent of the details of the floating point
2307: representation. But this does not work if the compiler is running on
2308: a different machine with an incompatible floating point format, or
2309: even a different byte-ordering.
2310:
1.1.1.5 root 2311: In addition, correct constant folding of floating point values
2312: requires representing them in the target machine's format.
2313: (The C standard does not quite require this, but in practice
2314: it is the only way to win.)
2315:
2316: It is now possible to overcome these problems by defining macros such
2317: as @code{REAL_VALUE_TYPE}. But doing so is a substantial amount of
2318: work for each target machine. @xref{Cross-compilation}.
1.1 root 2319:
2320: @item
2321: DBX rejects some files produced by GNU CC, though it accepts similar
2322: constructs in output from PCC. Until someone can supply a coherent
2323: description of what is valid DBX input and what is not, there is
2324: nothing I can do about these problems. You are on your own.
1.1.1.2 root 2325:
2326: @item
2327: Users often think it is a bug when GNU CC reports an error for code
2328: like this:
2329:
2330: @example
2331: int foo (short);
2332:
2333: int foo (x)
2334: short x;
2335: @{@dots{}@}
2336: @end example
2337:
1.1.1.4 root 2338: The error message is correct: this code really is erroneous, because the
2339: old-style non-prototype definition passes subword integers in their
2340: promoted types. In other words, the argument is really an @code{int},
2341: not a @code{short}. The correct prototype is this:
1.1.1.2 root 2342:
2343: @example
2344: int foo (int);
2345: @end example
2346:
2347: @item
2348: Users often think it is a bug when GNU CC reports an error for code
2349: like this:
2350:
2351: @example
2352: int foo (struct mumble *);
2353:
2354: struct mumble @{ @dots{} @};
2355:
2356: int foo (struct mumble *x)
2357: @{ @dots{} @}
2358: @end example
2359:
2360: This code really is erroneous, because the scope of @code{struct
2361: mumble} the prototype is limited to the argument list containing it.
2362: It does not refer to the @code{struct mumble} defined with file scope
2363: immediately below---they are two unrelated types with similar names in
2364: different scopes.
2365:
2366: But in the definition of @code{foo}, the file-scope type is used
2367: because that is available to be inherited. Thus, the definition and
2368: the prototype do not match, and you get an error.
2369:
2370: This behavior may seem silly, but it's what the ANSI standard
2371: specifies. It is easy enough for you to make your code work by moving
2372: the definition of @code{struct mumble} above the prototype. I don't
2373: think it's worth being incompatible for.
1.1 root 2374: @end itemize
2375:
1.1.1.10! root 2376:
! 2377: @node Service, Incompatibilities, Trouble, Top
! 2378: @chapter How To Get Help with GNU CC
! 2379:
! 2380: If you need help installing, using or changing GNU CC, there are two
! 2381: ways to find it:
! 2382:
! 2383: @itemize @bullet
! 2384: @item
! 2385: Send a message to a suitable network mailing list. First try
! 2386: @code{bug-gcc@@prep.ai.mit.edu}, and if that brings no response, try
! 2387: @code{info-gcc@@prep.ai.mit.edu}.
! 2388:
! 2389: @item
! 2390: Look in the service directory for someone who might help you for a fee.
! 2391: The service directory is found in the file named @file{SERVICE} in the
! 2392: GNU CC distribution.
! 2393: @end itemize
! 2394:
! 2395: @node Incompatibilities, Extensions, Service, Top
1.1 root 2396: @chapter Incompatibilities of GNU CC
2397:
2398: There are several noteworthy incompatibilities between GNU C and most
1.1.1.9 root 2399: existing (non-ANSI) versions of C. The @samp{-traditional} option
2400: eliminates most of these incompatibilities, @emph{but not all}, by
2401: telling GNU C to behave like older C compilers.
1.1 root 2402:
2403: @itemize @bullet
2404: @item
2405: GNU CC normally makes string constants read-only. If several
2406: identical-looking string constants are used, GNU CC stores only one
2407: copy of the string.
2408:
2409: One consequence is that you cannot call @code{mktemp} with a string
2410: constant argument. The function @code{mktemp} always alters the
2411: string its argument points to.
2412:
2413: Another consequence is that @code{sscanf} does not work on some
2414: systems when passed a string constant as its format control string.
2415: This is because @code{sscanf} incorrectly tries to write into the
1.1.1.4 root 2416: string constant. Likewise @code{fscanf} and @code{scanf}.
1.1 root 2417:
2418: The best solution to these problems is to change the program to use
2419: @code{char}-array variables with initialization strings for these
2420: purposes instead of string constants. But if this is not possible,
2421: you can use the @samp{-fwritable-strings} flag, which directs GNU CC
2422: to handle string constants the same way most C compilers do.
1.1.1.8 root 2423: @samp{-traditional} also has this effect, among others.
1.1 root 2424:
2425: @item
2426: GNU CC does not substitute macro arguments when they appear inside of
2427: string constants. For example, the following macro in GNU CC
2428:
2429: @example
2430: #define foo(a) "a"
2431: @end example
2432:
2433: @noindent
1.1.1.8 root 2434: will produce output @code{"a"} regardless of what the argument @var{a} is.
1.1 root 2435:
2436: The @samp{-traditional} option directs GNU CC to handle such cases
2437: (among others) in the old-fashioned (non-ANSI) fashion.
2438:
2439: @item
2440: When you use @code{setjmp} and @code{longjmp}, the only automatic
2441: variables guaranteed to remain valid are those declared
2442: @code{volatile}. This is a consequence of automatic register
2443: allocation. Consider this function:
2444:
2445: @example
2446: jmp_buf j;
2447:
2448: foo ()
2449: @{
2450: int a, b;
2451:
2452: a = fun1 ();
2453: if (setjmp (j))
2454: return a;
2455:
2456: a = fun2 ();
2457: /* @r{@code{longjmp (j)} may be occur in @code{fun3}.} */
2458: return a + fun3 ();
2459: @}
2460: @end example
2461:
2462: Here @code{a} may or may not be restored to its first value when the
2463: @code{longjmp} occurs. If @code{a} is allocated in a register, then
2464: its first value is restored; otherwise, it keeps the last value stored
2465: in it.
2466:
2467: If you use the @samp{-W} option with the @samp{-O} option, you will
2468: get a warning when GNU CC thinks such a problem might be possible.
2469:
1.1.1.2 root 2470: The @samp{-traditional} option directs GNU C to put variables in
2471: the stack by default, rather than in registers, in functions that
2472: call @code{setjmp}. This results in the behavior found in
2473: traditional C compilers.
2474:
1.1 root 2475: @item
2476: Declarations of external variables and functions within a block apply
2477: only to the block containing the declaration. In other words, they
2478: have the same scope as any other declaration in the same place.
2479:
2480: In some other C compilers, a @code{extern} declaration affects all the
2481: rest of the file even if it happens within a block.
2482:
2483: The @samp{-traditional} option directs GNU C to treat all @code{extern}
2484: declarations as global, like traditional compilers.
2485:
2486: @item
2487: In traditional C, you can combine @code{long}, etc., with a typedef name,
2488: as shown here:
2489:
2490: @example
2491: typedef int foo;
2492: typedef long foo bar;
2493: @end example
2494:
2495: In ANSI C, this is not allowed: @code{long} and other type modifiers
2496: require an explicit @code{int}. Because this criterion is expressed
2497: by Bison grammar rules rather than C code, the @samp{-traditional}
2498: flag cannot alter it.
2499:
2500: @item
2501: PCC allows typedef names to be used as function parameters. The
2502: difficulty described immediately above applies here too.
2503:
2504: @item
2505: PCC allows whitespace in the middle of compound assignment operators
2506: such as @samp{+=}. GNU CC, following the ANSI standard, does not
2507: allow this. The difficulty described immediately above applies here
2508: too.
2509:
2510: @item
2511: GNU CC will flag unterminated character constants inside of preprocessor
2512: conditionals that fail. Some programs have English comments enclosed in
2513: conditionals that are guaranteed to fail; if these comments contain
2514: apostrophes, GNU CC will probably report an error. For example,
2515: this code would produce an error:
2516:
2517: @example
2518: #if 0
2519: You can't expect this to work.
2520: #endif
2521: @end example
2522:
2523: The best solution to such a problem is to put the text into an actual
2524: C comment delimited by @samp{/*@dots{}*/}. However,
2525: @samp{-traditional} suppresses these error messages.
2526:
2527: @item
2528: When compiling functions that return @code{float}, PCC converts it to
2529: a double. GNU CC actually returns a @code{float}. If you are concerned
2530: with PCC compatibility, you should declare your functions to return
2531: @code{double}; you might as well say what you mean.
2532:
2533: @item
2534: When compiling functions that return structures or unions, GNU CC
1.1.1.6 root 2535: output code normally uses a method different from that used on most
2536: versions of Unix. As a result, code compiled with GNU CC cannot call
2537: a structure-returning function compiled with PCC, and vice versa.
1.1 root 2538:
1.1.1.6 root 2539: The method used by GNU CC is as follows: a structure or union which is 1,
1.1 root 2540: 2, 4 or 8 bytes long is returned like a scalar. A structure or union
2541: with any other size is stored into an address supplied by the caller
2542: in a special, fixed register.
2543:
2544: PCC usually handles all sizes of structures and unions by returning
2545: the address of a block of static storage containing the value. This
1.1.1.6 root 2546: method is not used in GNU CC because it is slower and nonreentrant.
1.1.1.5 root 2547:
1.1.1.6 root 2548: You can tell GNU CC to use the PCC convention with the option
2549: @samp{-fpcc-struct-return}.
1.1.1.8 root 2550:
2551: @item
2552: On the Sparc, GNU CC uses an incompatible calling convention for
2553: structures. It passes them by including their contents in the argument
2554: list, whereas the standard compiler passes them effectively by
2555: reference.
2556:
2557: This really ought to be fixed, but such calling conventions are not
2558: yet supported in GNU CC, so it isn't straightforward to fix it.
2559:
2560: The convention for structure returning is also incompatible, and
2561: @samp{-fpcc-struct-return} does not help.
1.1.1.10! root 2562:
! 2563: @item
! 2564: On Ultrix, the Fortran compiler expects registers 2 through 5 to be saved
! 2565: by function calls. We have not been able to tell whether the C compiler
! 2566: agrees with the Fortran compiler. Currently, GNU CC treats these registers
! 2567: as temporaries on the Vax, which is compatible with BSD Unix.
! 2568:
! 2569: If we learn for certain that Ultrix has departed from the traditional
! 2570: BSD calling convention, we will change GNU CC for Ultrix to fit. In the
! 2571: mean time, you can use these options to produce code compatible with the
! 2572: Fortran compiler:
! 2573:
! 2574: @example
! 2575: -fcall-saved-r2 -fcall-saved-r3 -fcall-saved-r4 -fcall-saved-r5
! 2576: @end example
1.1 root 2577: @end itemize
2578:
2579: @node Extensions, Bugs, Incompatibilities, Top
2580: @chapter GNU Extensions to the C Language
2581:
2582: GNU C provides several language features not found in ANSI standard C.
2583: (The @samp{-pedantic} option directs GNU CC to print a warning message if
2584: any of these features is used.) To test for the availability of these
2585: features in conditional compilation, check for a predefined macro
2586: @code{__GNUC__}, which is always defined under GNU CC.
2587:
2588: @menu
2589: * Statement Exprs:: Putting statements and declarations inside expressions.
2590: * Naming Types:: Giving a name to the type of some expression.
1.1.1.9 root 2591: * Typeof:: @code{typeof}: referring to the type of an expression.
2592: * Lvalues:: Using @samp{?:}, @samp{,} and casts in lvalues.
2593: * Conditionals:: Omitting the middle operand of a @samp{?:} expression.
2594: * Zero-Length:: Zero-length arrays.
2595: * Variable-Length:: Arrays whose length is computed at run time.
2596: * Subscripting:: Any array can be subscripted, even if not an lvalue.
2597: * Pointer Arith:: Arithmetic on @code{void}-pointers and function pointers.
2598: * Initializers:: Non-constant initializers.
2599: * Constructors:: Constructor expressions give structures, unions
2600: or arrays as values.
1.1.1.5 root 2601: * Function Attributes:: Declaring that functions have no side effects,
1.1.1.9 root 2602: or that they can never return.
1.1 root 2603: * Dollar Signs:: Dollar sign is allowed in identifiers.
2604: * Alignment:: Inquiring about the alignment of a type or variable.
2605: * Inline:: Defining inline functions (as fast as macros).
1.1.1.9 root 2606: * Extended Asm:: Assembler instructions with C expressions as operands.
2607: (With them you can define ``built-in'' functions.)
2608: * Asm Labels:: Specifying the assembler name to use for a C symbol.
1.1.1.8 root 2609: * Explicit Reg Vars:: Defining variables residing in specified registers.
2610: * Alternate Keywords:: @code{__const__}, @code{__asm__}, etc., for header files.
1.1 root 2611: @end menu
2612:
2613: @node Statement Exprs, Naming Types, Extensions, Extensions
2614: @section Statements and Declarations inside of Expressions
2615:
2616: A compound statement in parentheses may appear inside an expression in GNU
2617: C. This allows you to declare variables within an expression. For
2618: example:
2619:
2620: @example
2621: (@{ int y = foo (); int z;
2622: if (y > 0) z = y;
2623: else z = - y;
2624: z; @})
2625: @end example
2626:
2627: @noindent
2628: is a valid (though slightly more complex than necessary) expression
2629: for the absolute value of @code{foo ()}.
2630:
2631: This feature is especially useful in making macro definitions ``safe'' (so
2632: that they evaluate each operand exactly once). For example, the
2633: ``maximum'' function is commonly defined as a macro in standard C as
2634: follows:
2635:
2636: @example
2637: #define max(a,b) ((a) > (b) ? (a) : (b))
2638: @end example
2639:
2640: @noindent
2641: But this definition computes either @var{a} or @var{b} twice, with bad
2642: results if the operand has side effects. In GNU C, if you know the
2643: type of the operands (here let's assume @code{int}), you can define
2644: the macro safely as follows:
2645:
2646: @example
2647: #define maxint(a,b) \
2648: (@{int _a = (a), _b = (b); _a > _b ? _a : _b; @})
2649: @end example
2650:
2651: Embedded statements are not allowed in constant expressions, such as
2652: the value of an enumeration constant, the width of a bit field, or
2653: the initial value of a static variable.
2654:
2655: If you don't know the type of the operand, you can still do this, but you
2656: must use @code{typeof} (@pxref{Typeof}) or type naming (@pxref{Naming
2657: Types}).
2658:
2659: @node Naming Types, Typeof, Statement Exprs, Extensions
2660: @section Naming an Expression's Type
2661:
2662: You can give a name to the type of an expression using a @code{typedef}
2663: declaration with an initializer. Here is how to define @var{name} as a
2664: type name for the type of @var{exp}:
2665:
2666: @example
2667: typedef @var{name} = @var{exp};
2668: @end example
2669:
2670: This is useful in conjunction with the statements-within-expressions
2671: feature. Here is how the two together can be used to define a safe
2672: ``maximum'' macro that operates on any arithmetic type:
2673:
2674: @example
2675: #define max(a,b) \
2676: (@{typedef _ta = (a), _tb = (b); \
2677: _ta _a = (a); _tb _b = (b); \
2678: _a > _b ? _a : _b; @})
2679: @end example
2680:
2681: The reason for using names that start with underscores for the local
2682: variables is to avoid conflicts with variable names that occur within the
2683: expressions that are substituted for @code{a} and @code{b}. Eventually we
2684: hope to design a new form of declaration syntax that allows you to declare
2685: variables whose scopes start only after their initializers; this will be a
2686: more reliable way to prevent such conflicts.
2687:
2688: @node Typeof, Lvalues, Naming Types, Extensions
2689: @section Referring to a Type with @code{typeof}
2690:
2691: Another way to refer to the type of an expression is with @code{typeof}.
2692: The syntax of using of this keyword looks like @code{sizeof}, but the
2693: construct acts semantically like a type name defined with @code{typedef}.
2694:
2695: There are two ways of writing the argument to @code{typeof}: with an
2696: expression or with a type. Here is an example with an expression:
2697:
2698: @example
2699: typeof (x[0](1))
2700: @end example
2701:
2702: @noindent
2703: This assumes that @code{x} is an array of functions; the type described
2704: is that of the values of the functions.
2705:
2706: Here is an example with a typename as the argument:
2707:
2708: @example
2709: typeof (int *)
2710: @end example
2711:
2712: @noindent
2713: Here the type described is that of pointers to @code{int}.
2714:
1.1.1.7 root 2715: If you are writing a header file that must work when included in ANSI C
1.1.1.8 root 2716: programs, write @code{__typeof__} instead of @code{typeof}.
1.1.1.7 root 2717: @xref{Alternate Keywords}.
2718:
1.1 root 2719: A @code{typeof}-construct can be used anywhere a typedef name could be
2720: used. For example, you can use it in a declaration, in a cast, or inside
2721: of @code{sizeof} or @code{typeof}.
2722:
2723: @itemize @bullet
2724: @item
2725: This declares @code{y} with the type of what @code{x} points to.
2726:
2727: @example
2728: typeof (*x) y;
2729: @end example
2730:
2731: @item
2732: This declares @code{y} as an array of such values.
2733:
2734: @example
2735: typeof (*x) y[4];
2736: @end example
2737:
2738: @item
2739: This declares @code{y} as an array of pointers to characters:
2740:
2741: @example
2742: typeof (typeof (char *)[4]) y;
2743: @end example
2744:
2745: @noindent
2746: It is equivalent to the following traditional C declaration:
2747:
2748: @example
2749: char *y[4];
2750: @end example
2751:
2752: To see the meaning of the declaration using @code{typeof}, and why it
2753: might be a useful way to write, let's rewrite it with these macros:
2754:
2755: @example
2756: #define pointer(T) typeof(T *)
2757: #define array(T, N) typeof(T [N])
2758: @end example
2759:
2760: @noindent
2761: Now the declaration can be rewritten this way:
2762:
2763: @example
2764: array (pointer (char), 4) y;
2765: @end example
2766:
2767: @noindent
1.1.1.8 root 2768: Thus, @code{array (pointer (char), 4)} is the type of arrays of 4
1.1 root 2769: pointers to @code{char}.
2770: @end itemize
2771:
2772: @node Lvalues, Conditionals, Typeof, Extensions
2773: @section Generalized Lvalues
2774:
2775: Compound expressions, conditional expressions and casts are allowed as
2776: lvalues provided their operands are lvalues. This means that you can take
2777: their addresses or store values into them.
2778:
2779: For example, a compound expression can be assigned, provided the last
2780: expression in the sequence is an lvalue. These two expressions are
2781: equivalent:
2782:
2783: @example
2784: (a, b) += 5
2785: a, (b += 5)
2786: @end example
2787:
2788: Similarly, the address of the compound expression can be taken. These two
2789: expressions are equivalent:
2790:
2791: @example
2792: &(a, b)
2793: a, &b
2794: @end example
2795:
2796: A conditional expression is a valid lvalue if its type is not void and the
2797: true and false branches are both valid lvalues. For example, these two
2798: expressions are equivalent:
2799:
2800: @example
2801: (a ? b : c) = 5
2802: (a ? b = 5 : (c = 5))
2803: @end example
2804:
2805: A cast is a valid lvalue if its operand is valid. Taking the address of
2806: the cast is the same as taking the address without a cast, except for the
2807: type of the result. For example, these two expressions are equivalent (but
1.1.1.8 root 2808: the second may be valid when the type of @code{a} does not permit a cast to
2809: @code{int *}).
1.1 root 2810:
2811: @example
2812: &(int *)a
2813: (int **)&a
2814: @end example
2815:
2816: A simple assignment whose left-hand side is a cast works by converting the
2817: right-hand side first to the specified type, then to the type of the inner
2818: left-hand side expression. After this is stored, the value is converter
2819: back to the specified type to become the value of the assignment. Thus, if
1.1.1.8 root 2820: @code{a} has type @code{char *}, the following two expressions are
1.1 root 2821: equivalent:
2822:
2823: @example
2824: (int)a = 5
2825: (int)(a = (char *)5)
2826: @end example
2827:
2828: An assignment-with-arithmetic operation such as @samp{+=} applied to a cast
2829: performs the arithmetic using the type resulting from the cast, and then
2830: continues as in the previous case. Therefore, these two expressions are
2831: equivalent:
2832:
2833: @example
2834: (int)a += 5
2835: (int)(a = (char *) ((int)a + 5))
2836: @end example
2837:
2838: @node Conditionals, Zero-Length, Lvalues, Extensions
2839: @section Conditional Expressions with Omitted Middle-Operands
2840:
2841: The middle operand in a conditional expression may be omitted. Then
2842: if the first operand is nonzero, its value is the value of the conditional
2843: expression.
2844:
2845: Therefore, the expression
2846:
2847: @example
2848: x ? : y
2849: @end example
2850:
2851: @noindent
2852: has the value of @code{x} if that is nonzero; otherwise, the value of
2853: @code{y}.
2854:
2855: This example is perfectly equivalent to
2856:
2857: @example
2858: x ? x : y
2859: @end example
2860:
2861: @noindent
2862: In this simple case, the ability to omit the middle operand is not
2863: especially useful. When it becomes useful is when the first operand does,
2864: or may (if it is a macro argument), contain a side effect. Then repeating
2865: the operand in the middle would perform the side effect twice. Omitting
2866: the middle operand uses the value already computed without the undesirable
2867: effects of recomputing it.
2868:
2869: @node Zero-Length, Variable-Length, Conditionals, Extensions
2870: @section Arrays of Length Zero
2871:
2872: Zero-length arrays are allowed in GNU C. They are very useful as the last
2873: element of a structure which is really a header for a variable-length
2874: object:
2875:
2876: @example
2877: struct line @{
2878: int length;
2879: char contents[0];
2880: @};
2881:
2882: @{
2883: struct line *thisline
2884: = (struct line *) malloc (sizeof (struct line) + this_length);
2885: thisline->length = this_length;
2886: @}
2887: @end example
2888:
2889: In standard C, you would have to give @code{contents} a length of 1, which
2890: means either you waste space or complicate the argument to @code{malloc}.
2891:
2892: @node Variable-Length, Subscripting, Zero-Length, Extensions
2893: @section Arrays of Variable Length
2894:
2895: Variable-length automatic arrays are allowed in GNU C. These arrays are
2896: declared like any other automatic arrays, but with a length that is not a
2897: constant expression. The storage is allocated at that time and
2898: deallocated when the brace-level is exited. For example:
2899:
2900: @example
2901: FILE *concat_fopen (char *s1, char *s2, char *mode)
2902: @{
2903: char str[strlen (s1) + strlen (s2) + 1];
2904: strcpy (str, s1);
2905: strcat (str, s2);
2906: return fopen (str, mode);
2907: @}
2908: @end example
2909:
1.1.1.7 root 2910: You can also use variable-length arrays as arguments to functions:
1.1 root 2911:
2912: @example
2913: struct entry
1.1.1.7 root 2914: tester (int len, char data[len])
1.1 root 2915: @{
1.1.1.7 root 2916: @dots{}
1.1 root 2917: @}
2918: @end example
2919:
2920: The length of an array is computed on entry to the brace-level where the
2921: array is declared and is remembered for the scope of the array in case you
2922: access it with @code{sizeof}.
2923:
2924: Jumping or breaking out of the scope of the array name will also deallocate
2925: the storage. Jumping into the scope is not allowed; you will get an error
2926: message for it.
2927:
2928: You can use the function @code{alloca} to get an effect much like
2929: variable-length arrays. The function @code{alloca} is available in
2930: many other C implementations (but not in all). On the other hand,
2931: variable-length arrays are more elegant.
2932:
2933: There are other differences between these two methods. Space allocated
2934: with @code{alloca} exists until the containing @emph{function} returns.
2935: The space for a variable-length array is deallocated as soon as the array
2936: name's scope ends. (If you use both variable-length arrays and
2937: @code{alloca} in the same function, deallocation of a variable-length array
2938: will also deallocate anything more recently allocated with @code{alloca}.)
2939:
2940: @node Subscripting, Pointer Arith, Variable-Length, Extensions
2941: @section Non-Lvalue Arrays May Have Subscripts
2942:
2943: Subscripting is allowed on arrays that are not lvalues, even though the
2944: unary @samp{&} operator is not. For example, this is valid in GNU C though
2945: not valid in other C dialects:
2946:
2947: @example
2948: struct foo @{int a[4];@};
2949:
2950: struct foo f();
2951:
2952: bar (int index)
2953: @{
2954: return f().a[index];
2955: @}
2956: @end example
2957:
2958: @node Pointer Arith, Initializers, Subscripting, Extensions
2959: @section Arithmetic on @code{void}-Pointers and Function Pointers
2960:
2961: In GNU C, addition and subtraction operations are supported on pointers to
2962: @code{void} and on pointers to functions. This is done by treating the
2963: size of a @code{void} or of a function as 1.
2964:
2965: A consequence of this is that @code{sizeof} is also allowed on @code{void}
2966: and on function types, and returns 1.
2967:
1.1.1.8 root 2968: The option @samp{-Wpointer-arith} requests a warning if these extensions
2969: are used.
2970:
1.1 root 2971: @node Initializers, Constructors, Pointer Arith, Extensions
2972: @section Non-Constant Initializers
2973:
1.1.1.8 root 2974: The elements of an aggregate initializer for an automatic variable are
2975: not required to be constant expressions in GNU C. Here is an example of
2976: an initializer with run-time varying elements:
1.1 root 2977:
2978: @example
2979: foo (float f, float g)
2980: @{
2981: float beat_freqs[2] = @{ f-g, f+g @};
2982: @dots{}
2983: @}
2984: @end example
2985:
1.1.1.5 root 2986: @node Constructors, Function Attributes, Initializers, Extensions
1.1 root 2987: @section Constructor Expressions
2988:
2989: GNU C supports constructor expressions. A constructor looks like a cast
2990: containing an initializer. Its value is an object of the type specified in
2991: the cast, containing the elements specified in the initializer. The type
2992: must be a structure, union or array type.
2993:
2994: Assume that @code{struct foo} and @code{structure} are declared as shown:
2995:
2996: @example
2997: struct foo @{int a; char b[2];@} structure;
2998: @end example
2999:
3000: @noindent
1.1.1.8 root 3001: Here is an example of constructing a @code{struct foo} with a constructor:
1.1 root 3002:
3003: @example
3004: structure = ((struct foo) @{x + y, 'a', 0@});
3005: @end example
3006:
3007: @noindent
3008: This is equivalent to writing the following:
3009:
3010: @example
3011: @{
3012: struct foo temp = @{x + y, 'a', 0@};
3013: structure = temp;
3014: @}
3015: @end example
3016:
3017: You can also construct an array. If all the elements of the constructor
3018: are (made up of) simple constant expressions, suitable for use in
3019: initializers, then the constructor is an lvalue and can be coerced to a
3020: pointer to its first element, as shown here:
3021:
3022: @example
3023: char **foo = (char *[]) @{ "x", "y", "z" @};
3024: @end example
3025:
3026: Array constructors whose elements are not simple constants are not very
3027: useful, because the constructor is not an lvalue. There are only two valid
3028: ways to use it: to subscript it, or initialize an array variable with it.
3029: The former is probably slower than a @code{switch} statement, while the
3030: latter does the same thing an ordinary C initializer would do.
3031:
3032: @example
3033: output = ((int[]) @{ 2, x, 28 @}) [input];
3034: @end example
3035:
1.1.1.8 root 3036: @node Function Attributes, Dollar Signs, Constructors, Extensions
1.1.1.5 root 3037: @section Declaring Attributes of Functions
3038:
3039: In GNU C, you declare certain things about functions called in your program
3040: which help the compiler optimize function calls.
3041:
3042: A few functions, such as @code{abort} and @code{exit}, cannot return.
3043: These functions should be declared @code{volatile}. For example,
3044:
3045: @example
3046: extern volatile void abort ();
3047: @end example
3048:
3049: @noindent
3050: tells the compiler that it can assume that @code{abort} will not return.
3051: This makes slightly better code, but more importantly it helps avoid
3052: spurious warnings of uninitialized variables.
3053:
3054: Many functions do not examine any values except their arguments, and
3055: have no effects except the return value. Such a function can be subject
3056: to common subexpression elimination and loop optimization just as an
3057: arithmetic operator would be. These functions should be declared
3058: @code{const}. For example,
3059:
3060: @example
3061: extern const void square ();
3062: @end example
3063:
3064: @noindent
3065: says that the hypothetical function @code{square} is safe to call
3066: fewer times than the program says.
3067:
3068: Note that a function that has pointer arguments and examines the data
3069: pointed to must @emph{not} be declared @code{const}. Likewise, a
1.1.1.10! root 3070: function that calls a non-@code{const} function usually must not be
1.1.1.5 root 3071: @code{const}.
3072:
3073: Some people object to this feature, claiming that ANSI C's @code{#pragma}
3074: should be used instead. There are two reasons I did not do this.
3075:
3076: @enumerate
3077: @item
3078: It is impossible to generate @code{#pragma} commands from a macro.
3079:
3080: @item
3081: The @code{#pragma} command is just as likely as these keywords to mean
3082: something else in another compiler.
3083: @end enumerate
3084:
3085: These two reasons apply to @emph{any} application whatever: as far as
3086: I can see, @code{#pragma} is never useful.
3087:
3088: @node Dollar Signs, Alignment, Function Attributes, Extensions
1.1 root 3089: @section Dollar Signs in Identifier Names
3090:
3091: In GNU C, you may use dollar signs in identifier names. This is because
3092: many traditional C implementations allow such identifiers.
3093:
1.1.1.9 root 3094: Dollar signs are allowed if you specify @samp{-traditional}; they are
3095: not allowed if you specify @samp{-ansi}. Whether they are allowed by
3096: default depends on the target machine; usually, they are not.
3097:
1.1 root 3098: @node Alignment, Inline, Dollar Signs, Extensions
3099: @section Inquiring about the Alignment of a Type or Variable
3100:
1.1.1.8 root 3101: The keyword @code{__alignof__} allows you to inquire about how an object
1.1 root 3102: is aligned, or the minimum alignment usually required by a type. Its
3103: syntax is just like @code{sizeof}.
3104:
3105: For example, if the target machine requires a @code{double} value to be
1.1.1.8 root 3106: aligned on an 8-byte boundary, then @code{__alignof__ (double)} is 8.
3107: This is true on many RISC machines. On more traditional machine
3108: designs, @code{__alignof__ (double)} is 4 or even 2.
1.1 root 3109:
3110: Some machines never actually require alignment; they allow reference to any
1.1.1.8 root 3111: data type even at an odd addresses. For these machines, @code{__alignof__}
1.1 root 3112: reports the @emph{recommended} alignment of a type.
3113:
1.1.1.8 root 3114: When the operand of @code{__alignof__} is an lvalue rather than a type, the
1.1 root 3115: value is the largest alignment that the lvalue is known to have. It may
3116: have this alignment as a result of its data type, or because it is part of
3117: a structure and inherits alignment from that structure. For example, after
3118: this declaration:
3119:
3120: @example
3121: struct foo @{ int x; char y; @} foo1;
3122: @end example
3123:
3124: @noindent
1.1.1.8 root 3125: the value of @code{__alignof__ (foo1.y)} is probably 2 or 4, the same as
3126: @code{__alignof__ (int)}, even though the data type of @code{foo1.y}
3127: does not itself demand any alignment.@refill
1.1 root 3128:
3129: @node Inline, Extended Asm, Alignment, Extensions
3130: @section An Inline Function is As Fast As a Macro
3131:
3132: By declaring a function @code{inline}, you can direct GNU CC to integrate
3133: that function's code into the code for its callers. This makes execution
3134: faster by eliminating the function-call overhead; in addition, if any of
3135: the actual argument values are constant, their known values may permit
3136: simplifications at compile time so that not all of the inline function's
3137: code needs to be included.
3138:
3139: To declare a function inline, use the @code{inline} keyword in its
3140: declaration, like this:
3141:
3142: @example
3143: inline int
3144: inc (int *a)
3145: @{
3146: (*a)++;
3147: @}
3148: @end example
3149:
1.1.1.7 root 3150: (If you are writing a header file to be included in ANSI C programs, write
1.1.1.8 root 3151: @code{__inline__} instead of @code{inline}. @xref{Alternate Keywords}.)
1.1.1.7 root 3152:
3153: You can also make all ``simple enough'' functions inline with the option
3154: @samp{-finline-functions}. Note that certain usages in a function
3155: definition can make it unsuitable for inline substitution.
1.1 root 3156:
3157: When a function is both inline and @code{static}, if all calls to the
1.1.1.8 root 3158: function are integrated into the caller, and the function's address is
3159: never used, then the function's own assembler code is never referenced.
3160: In this case, GNU CC does not actually output assembler code for the
3161: function, unless you specify the option @samp{-fkeep-inline-functions}.
3162: Some calls cannot be integrated for various reasons (in particular,
3163: calls that precede the function's definition cannot be integrated, and
3164: neither can recursive calls within the definition). If there is a
3165: nonintegrated call, then the function is compiled to assembler code as
3166: usual. The function must also be compiled as usual if the program
3167: refers to its address, because that can't be inlined.
1.1 root 3168:
3169: When an inline function is not @code{static}, then the compiler must assume
3170: that there may be calls from other source files; since a global symbol can
3171: be defined only once in any program, the function must not be defined in
3172: the other source files, so the calls therein cannot be integrated.
3173: Therefore, a non-@code{static} inline function is always compiled on its
3174: own in the usual fashion.
3175:
1.1.1.8 root 3176: If you specify both @code{inline} and @code{extern} in the function
3177: definition, then the definition is used only for inlining. In no case
3178: is the function compiled on its own, not even if you refer to its
3179: address explicitly. Such an address becomes an external reference, as
3180: if you had only declared the function, and had not defined it.
3181:
3182: This combination of @code{inline} and @code{extern} has almost the
3183: effect of a macro. The way to use it is to put a function definition in
3184: a header file with these keywords, and put another copy of the
3185: definition (lacking @code{inline} and @code{extern}) in a library file.
3186: The definition in the header file will cause most calls to the function
3187: to be inlined. If any uses of the function remain, they will refer to
3188: the single copy in the library.
3189:
1.1 root 3190: @node Extended Asm, Asm Labels, Inline, Extensions
3191: @section Assembler Instructions with C Expression Operands
3192:
3193: In an assembler instruction using @code{asm}, you can now specify the
3194: operands of the instruction using C expressions. This means no more
3195: guessing which registers or memory locations will contain the data you want
3196: to use.
3197:
3198: You must specify an assembler instruction template much like what appears
3199: in a machine description, plus an operand constraint string for each
3200: operand.
3201:
3202: For example, here is how to use the 68881's @code{fsinx} instruction:
3203:
3204: @example
3205: asm ("fsinx %1,%0" : "=f" (result) : "f" (angle));
3206: @end example
3207:
3208: @noindent
3209: Here @code{angle} is the C expression for the input operand while
3210: @code{result} is that of the output operand. Each has @samp{"f"} as its
3211: operand constraint, saying that a floating-point register is required. The
1.1.1.5 root 3212: @samp{=} in @samp{=f} indicates that the operand is an output; all output
1.1.1.4 root 3213: operands' constraints must use @samp{=}. The constraints use the same
3214: language used in the machine description (@pxref{Constraints}).
1.1 root 3215:
3216: Each operand is described by an operand-constraint string followed by the C
3217: expression in parentheses. A colon separates the assembler template from
3218: the first output operand, and another separates the last output operand
3219: from the first input, if any. Commas separate output operands and separate
1.1.1.4 root 3220: inputs. The total number of operands is limited to the maximum number of
1.1 root 3221: operands in any instruction pattern in the machine description.
3222:
1.1.1.4 root 3223: If there are no output operands, and there are input operands, then there
3224: must be two consecutive colons surrounding the place where the output
3225: operands would go.
3226:
1.1 root 3227: Output operand expressions must be lvalues; the compiler can check this.
3228: The input operands need not be lvalues. The compiler cannot check whether
3229: the operands have data types that are reasonable for the instruction being
3230: executed. It does not parse the assembler instruction template and does
3231: not know what it means, or whether it is valid assembler input. The
3232: extended @code{asm} feature is most often used for machine instructions
3233: that the compiler itself does not know exist.
3234:
3235: The output operands must be write-only; GNU CC will assume that the values
3236: in these operands before the instruction are dead and need not be
1.1.1.8 root 3237: generated. Extended asm does not support input-output or read-write
3238: operands. For this reason, the constraint character @samp{+}, which
3239: indicates such an operand, may not be used.
3240:
3241: When the assembler instruction has a read-write operand, or an operand
3242: in which only some of the bits are to be changed, you must logically
1.1 root 3243: split its function into two separate operands, one input operand and one
3244: write-only output operand. The connection between them is expressed by
3245: constraints which say they need to be in the same location when the
1.1.1.8 root 3246: instruction executes. You can use the same C expression for both
3247: operands, or different expressions. For example, here we write the
3248: (fictitious) @samp{combine} instruction with @code{bar} as its read-only
3249: source operand and @code{foo} as its read-write destination:
1.1 root 3250:
3251: @example
3252: asm ("combine %2,%0" : "=r" (foo) : "0" (foo), "g" (bar));
3253: @end example
3254:
3255: @noindent
3256: The constraint @samp{"0"} for operand 1 says that it must occupy the same
1.1.1.5 root 3257: location as operand 0. A digit in constraint is allowed only in an input
3258: operand, and it must refer to an output operand.
1.1 root 3259:
3260: Only a digit in the constraint can guarantee that one operand will be in
3261: the same place as another. The mere fact that @code{foo} is the value of
3262: both operands is not enough to guarantee that they will be in the same
3263: place in the generated assembler code. The following would not work:
3264:
3265: @example
3266: asm ("combine %2,%0" : "=r" (foo) : "r" (foo), "g" (bar));
3267: @end example
3268:
3269: Various optimizations or reloading could cause operands 0 and 1 to be in
3270: different registers; GNU CC knows no reason not to do so. For example, the
3271: compiler might find a copy of the value of @code{foo} in one register and
3272: use it for operand 1, but generate the output operand 0 in a different
3273: register (copying it afterward to @code{foo}'s own address). Of course,
3274: since the register for operand 1 is not even mentioned in the assembler
3275: code, the result will not work, but GNU CC can't tell that.
3276:
3277: Unless an output operand has the @samp{&} constraint modifier, GNU CC may
3278: allocate it in the same register as an unrelated input operand, on the
3279: assumption that the inputs are consumed before the outputs are produced.
3280: This assumption may be false if the assembler code actually consists of
3281: more than one instruction. In such a case, use @samp{&} for each output
3282: operand that may not overlap an input. @xref{Modifiers}.
3283:
1.1.1.4 root 3284: Some instructions clobber specific hard registers. To describe this, write
3285: a third colon after the input operands, followed by the names of the
3286: clobbered hard registers (given as strings). Here is a realistic example
3287: for the vax:
1.1 root 3288:
3289: @example
3290: asm volatile ("movc3 %0,%1,%2"
3291: : /* no outputs */
3292: : "g" (from), "g" (to), "g" (count)
3293: : "r0", "r1", "r2", "r3", "r4", "r5");
3294: @end example
3295:
1.1.1.4 root 3296: You can put multiple assembler instructions together in a single @code{asm}
1.1.1.7 root 3297: template, separated either with newlines (written as @samp{\n}) or with
3298: semicolons if the assembler allows such semicolons. The GNU assembler
3299: allows semicolons and all Unix assemblers seem to do so. The input
3300: operands are guaranteed not to use any of the clobbered registers, and
3301: neither will the output operands' addresses, so you can read and write the
3302: clobbered registers as many times as you like. Here is an example of
3303: multiple instructions in a template; it assumes that the subroutine
3304: @code{_foo} accepts arguments in registers 9 and 10:
1.1.1.4 root 3305:
3306: @example
3307: asm ("movl %0,r9;movl %1,r10;call _foo"
3308: : /* no outputs */
3309: : "g" (from), "g" (to)
3310: : "r9", "r10");
3311: @end example
3312:
1.1.1.7 root 3313: If you want to test the condition code produced by an assembler instruction,
3314: you must include a branch and a label in the @code{asm} construct, as follows:
3315:
3316: @example
3317: asm ("clr %0;frob %1;beq 0f;mov #1,%0;0:"
3318: : "g" (result)
3319: : "g" (input));
3320: @end example
3321:
3322: @noindent
3323: This assumes your assembler supports local labels, as the GNU assembler
3324: and most Unix assemblers do.
3325:
1.1 root 3326: Usually the most convenient way to use these @code{asm} instructions is to
3327: encapsulate them in macros that look like functions. For example,
3328:
3329: @example
3330: #define sin(x) \
3331: (@{ double __value, __arg = (x); \
3332: asm ("fsinx %1,%0": "=f" (__value): "f" (__arg)); \
3333: __value; @})
3334: @end example
3335:
3336: @noindent
3337: Here the variable @code{__arg} is used to make sure that the instruction
3338: operates on a proper @code{double} value, and to accept only those
3339: arguments @code{x} which can convert automatically to a @code{double}.
3340:
3341: Another way to make sure the instruction operates on the correct data type
3342: is to use a cast in the @code{asm}. This is different from using a
3343: variable @code{__arg} in that it converts more different types. For
3344: example, if the desired type were @code{int}, casting the argument to
3345: @code{int} would accept a pointer with no complaint, while assigning the
3346: argument to an @code{int} variable named @code{__arg} would warn about
3347: using a pointer unless the caller explicitly casts it.
3348:
1.1.1.4 root 3349: If an @code{asm} has output operands, GNU CC assumes for optimization
3350: purposes that the instruction has no side effects except to change the
3351: output operands. This does not mean that instructions with a side effect
3352: cannot be used, but you must be careful, because the compiler may eliminate
3353: them if the output operands aren't used, or move them out of loops, or
3354: replace two with one if they constitute a common subexpression. Also, if
3355: your instruction does have a side effect on a variable that otherwise
3356: appears not to change, the old value of the variable may be reused later if
3357: it happens to be found in a register.
1.1 root 3358:
3359: You can prevent an @code{asm} instruction from being deleted, moved or
3360: combined by writing the keyword @code{volatile} after the @code{asm}. For
3361: example:
3362:
3363: @example
3364: #define set_priority(x) \
3365: asm volatile ("set_priority %0": /* no outputs */ : "g" (x))
3366: @end example
3367:
1.1.1.7 root 3368: @noindent
3369: (However, an instruction without output operands will not be deleted
3370: or moved, regardless, unless it is unreachable.)
1.1.1.4 root 3371:
1.1 root 3372: It is a natural idea to look for a way to give access to the condition
3373: code left by the assembler instruction. However, when we attempted to
3374: implement this, we found no way to make it work reliably. The problem
3375: is that output operands might need reloading, which would result in
3376: additional following ``store'' instructions. On most machines, these
3377: instructions would alter the condition code before there was time to
3378: test it. This problem doesn't arise for ordinary ``test'' and
3379: ``compare'' instructions because they don't have any output operands.
3380:
1.1.1.7 root 3381: If you are writing a header file that should be includable in ANSI C
1.1.1.8 root 3382: programs, write @code{__asm__} instead of @code{asm}. @xref{Alternate
1.1.1.7 root 3383: Keywords}.
3384:
1.1.1.8 root 3385: @node Asm Labels, Explicit Reg Vars, Extended Asm, Extensions
1.1 root 3386: @section Controlling Names Used in Assembler Code
3387:
1.1.1.8 root 3388: You can specify the name to be used in the assembler code for a C
3389: function or variable by writing the @code{asm} (or @code{__asm__})
3390: keyword after the declarator as follows:
1.1 root 3391:
3392: @example
3393: int foo asm ("myfoo") = 2;
3394: @end example
3395:
3396: @noindent
3397: This specifies that the name to be used for the variable @code{foo} in
3398: the assembler code should be @samp{myfoo} rather than the usual
3399: @samp{_foo}.
3400:
3401: On systems where an underscore is normally prepended to the name of a C
3402: function or variable, this feature allows you to define names for the
3403: linker that do not start with an underscore.
3404:
3405: You cannot use @code{asm} in this way in a function @emph{definition}; but
3406: you can get the same effect by writing a declaration for the function
3407: before its definition and putting @code{asm} there, like this:
3408:
3409: @example
3410: extern func () asm ("FUNC");
3411:
3412: func (x, y)
3413: int x, y;
3414: @dots{}
3415: @end example
3416:
3417: It is up to you to make sure that the assembler names you choose do not
3418: conflict with any other assembler symbols. Also, you must not use a
3419: register name; that would produce completely invalid assembler code. GNU
3420: CC does not as yet have the ability to store static variables in registers.
3421: Perhaps that will be added.
3422:
1.1.1.8 root 3423: @node Explicit Reg Vars, Alternate Keywords, Asm Labels, Extensions
3424: @section Variables in Specified Registers
3425:
3426: GNU C allows you to put a few global variables into specified hardware
3427: registers. You can also specify the register in which an ordinary
3428: register variable should be allocated.
3429:
3430: @itemize @bullet
3431: @item
3432: Global register variables reserve registers throughout the program.
3433: This may be useful in programs such as programming language
3434: interpreters which have a couple of global variables that are accessed
3435: very often.
3436:
3437: @item
3438: Local register variables in specific registers do not reserve the
3439: registers. The compiler's data flow analysis is capable of
3440: determining where the specified registers contain live values, and
3441: where they are available for other uses. These local variables are
3442: sometimes convenient for use with the extended @code{asm} feature
3443: (@pxref{Extended Asm}).
3444: @end itemize
3445:
3446: @menu
3447: * Global Reg Vars::
3448: * Local Reg Vars::
3449: @end menu
1.1.1.5 root 3450:
1.1.1.8 root 3451: @node Global Reg Vars, Local Reg Vars, Explicit Reg Vars, Explicit Reg Vars
3452: @subsection Defining Global Register Variables
1.1.1.5 root 3453:
3454: You can define a global register variable in GNU C like this:
3455:
3456: @example
3457: register int *foo asm ("a5");
3458: @end example
3459:
3460: @noindent
3461: Here @code{a5} is the name of the register which should be used. Choose a
3462: register which is normally saved and restored by function calls on your
3463: machine, so that library routines will not clobber it.
3464:
3465: Naturally the register name is cpu-dependent, so you would need to
3466: conditionalize your program according to cpu type. The register
3467: @code{a5} would be a good choice on a 68000 for a variable of pointer
3468: type. On machines with register windows, be sure to choose a ``global''
1.1.1.8 root 3469: register that is not affected magically by the function call mechanism.
1.1.1.5 root 3470:
3471: In addition, operating systems on one type of cpu may differ in how they
3472: name the registers; then you would need additional conditionals. For
3473: example, some 68000 operating systems call this register @code{%a5}.
3474:
3475: Eventually there may be a way of asking the compiler to choose a register
3476: automatically, but first we need to figure out how it should choose and
1.1.1.6 root 3477: how to enable you to guide the choice. No solution is evident.
1.1.1.5 root 3478:
3479: Defining a global register variable in a certain register reserves that
3480: register entirely for this use, at least within the current compilation.
3481: The register will not be allocated for any other purpose in the functions
3482: in the current compilation. The register will not be saved and restored by
3483: these functions. Stores into this register are never deleted even if they
3484: would appear to be dead, but references may be deleted or moved or
3485: simplified.
3486:
3487: It is not safe to access the global register variables from signal
3488: handlers, or from more than one thread of control, because the system
3489: library routines may temporarily use the register for other things (unless
3490: you recompile them specially for the task at hand).
3491:
3492: It is not safe for one function that uses a global register variable to
3493: call another such function @code{foo} by way of a third function
3494: @code{lose} that was compiled without knowledge of this variable (i.e. in a
3495: different source file in which the variable wasn't declared). This is
3496: because @code{lose} might save the register and put some other value there.
3497: For example, you can't expect a global register variable to be available in
3498: the comparison-function that you pass to @code{qsort}, since @code{qsort}
3499: might have put something else in that register. (If you are prepared to
3500: recompile @code{qsort} with the same global register variable, you can
3501: solve this problem.)
3502:
3503: If you want to recompile @code{qsort} or other source files which do not
3504: actually use your global register variable, so that they will not use that
3505: register for any other purpose, then it suffices to specify the compiler
3506: option @samp{-ffixed-@var{reg}}. You need not actually add a global
3507: register declaration to their source code.
3508:
3509: A function which can alter the value of a global register variable cannot
3510: safely be called from a function compiled without this variable, because it
3511: could clobber the value the caller expects to find there on return.
3512: Therefore, the function which is the entry point into the part of the
3513: program that uses the global register variable must explicitly save and
3514: restore the value which belongs to its caller.
3515:
3516: On most machines, @code{longjmp} will restore to each global register
3517: variable the value it had at the time of the @code{setjmp}. On some
3518: machines, however, @code{longjmp} will not change the value of global
3519: register variables. To be portable, the function that called @code{setjmp}
3520: should make other arrangements to save the values of the global register
1.1.1.10! root 3521: variables, and to restore them in a @code{longjmp}. This way, the same
1.1.1.5 root 3522: thing will happen regardless of what @code{longjmp} does.
3523:
3524: All global register variable declarations must precede all function
3525: definitions. If such a declaration could appear after function
3526: definitions, the declaration would be too late to prevent the register from
3527: being used for other purposes in the preceding functions.
3528:
1.1.1.6 root 3529: Global register variables may not have initial values, because an
3530: executable file has no means to supply initial contents for a register.
3531:
1.1.1.9 root 3532: @node Local Reg Vars,, Global Reg Vars, Explicit Reg Vars
1.1.1.8 root 3533: @subsection Specifying Registers for Local Variables
3534:
3535: You can define a local register variable with a specified register
3536: like this:
3537:
3538: @example
3539: register int *foo asm ("a5");
3540: @end example
3541:
3542: @noindent
3543: Here @code{a5} is the name of the register which should be used. Note
3544: that this is the same syntax used for defining global register
3545: variables, but for a local variable it would appear within a function.
3546:
3547: Naturally the register name is cpu-dependent, but this is not a
3548: problem, since specific registers are most often useful with explicit
3549: assembler instructions (@pxref{Extended Asm}). Both of these things
3550: generally require that you conditionalize your program according to
3551: cpu type.
3552:
3553: In addition, operating systems on one type of cpu may differ in how they
3554: name the registers; then you would need additional conditionals. For
3555: example, some 68000 operating systems call this register @code{%a5}.
3556:
3557: Eventually there may be a way of asking the compiler to choose a register
3558: automatically, but first we need to figure out how it should choose and
3559: how to enable you to guide the choice. No solution is evident.
3560:
3561: Defining such a register variable does not reserve the register; it
1.1.1.10! root 3562: remains available for other uses in places where flow control determines
! 3563: the variable's value is not live. However, these registers are made
! 3564: unavailable for use in the reload pass. I would not be surprised if
! 3565: excessive use of this feature leaves the compiler too few available
1.1.1.8 root 3566: registers to compile certain functions.
3567:
3568: @node Alternate Keywords,, Explicit Reg Vars, Extensions
1.1.1.7 root 3569: @section Alternate Keywords
3570:
3571: The option @samp{-traditional} disables certain keywords; @samp{-ansi}
3572: disables certain others. This causes trouble when you want to use GNU C
3573: extensions, or ANSI C features, in a general-purpose header file that
3574: should be usable by all programs, including ANSI C programs and traditional
3575: ones. The keywords @code{asm}, @code{typeof} and @code{inline} cannot be
3576: used since they won't work in a program compiled with @samp{-ansi}, while
3577: the keywords @code{const}, @code{volatile}, @code{signed}, @code{typeof}
3578: and @code{inline} won't work in a program compiled with
3579: @samp{-traditional}.@refill
3580:
1.1.1.8 root 3581: The way to solve these problems is to put @samp{__} at the beginning and
3582: end of each problematical keyword. For example, use @code{__asm__}
3583: instead of @code{asm}, @code{__const__} instead of @code{const}, and
3584: @code{__inline__} instead of @code{inline}.
1.1.1.7 root 3585:
3586: Other C compilers won't accept these alternative keywords; if you want to
3587: compile with another compiler, you can define the alternate keywords as
3588: macros to replace them with the customary keywords. It looks like this:
3589:
3590: @example
3591: #ifndef __GNUC__
1.1.1.8 root 3592: #define __asm__ asm
1.1.1.7 root 3593: #endif
3594: @end example
3595:
1.1 root 3596: @node Bugs, Portability, Extensions, Top
3597: @chapter Reporting Bugs
3598:
3599: Your bug reports play an essential role in making GNU CC reliable.
3600:
3601: Reporting a bug may help you by bringing a solution to your problem, or it
3602: may not. But in any case the important function of a bug report is to help
3603: the entire community by making the next version of GNU CC work better. Bug
3604: reports are your contribution to the maintenance of GNU CC.
3605:
3606: In order for a bug report to serve its purpose, you must include the
3607: information that makes for fixing the bug.
3608:
3609: @menu
3610: * Criteria: Bug Criteria. Have you really found a bug?
3611: * Reporting: Bug Reporting. How to report a bug effectively.
1.1.1.10! root 3612: * Known: Trouble. A list of known problems.
! 3613: * Help: Service. How to get help.
1.1 root 3614: @end menu
3615:
3616: @node Bug Criteria, Bug Reporting, Bugs, Bugs
3617: @section Have You Found a Bug?
3618:
3619: If you are not sure whether you have found a bug, here are some guidelines:
3620:
3621: @itemize @bullet
3622: @item
3623: If the compiler gets a fatal signal, for any input whatever, that is a
3624: compiler bug. Reliable compilers never crash.
3625:
3626: @item
3627: If the compiler produces invalid assembly code, for any input whatever
3628: (except an @code{asm} statement), that is a compiler bug, unless the
3629: compiler reports errors (not just warnings) which would ordinarily
3630: prevent the assembler from being run.
3631:
3632: @item
3633: If the compiler produces valid assembly code that does not correctly
3634: execute the input source code, that is a compiler bug.
3635:
3636: However, you must double-check to make sure, because you may have run
3637: into an incompatibility between GNU C and traditional C
3638: (@pxref{Incompatibilities}). These incompatibilities might be considered
3639: bugs, but they are inescapable consequences of valuable features.
3640:
3641: Or you may have a program whose behavior is undefined, which happened
3642: by chance to give the desired results with another C compiler.
3643:
3644: For example, in many nonoptimizing compilers, you can write @samp{x;}
3645: at the end of a function instead of @samp{return x;}, with the same
1.1.1.8 root 3646: results. But the value of the function is undefined if @code{return}
1.1 root 3647: is omitted; it is not a bug when GNU CC produces different results.
3648:
3649: Problems often result from expressions with two increment operators,
1.1.1.8 root 3650: as in @code{f (*p++, *p++)}. Your previous compiler might have
1.1 root 3651: interpreted that expression the way you intended; GNU CC might
1.1.1.8 root 3652: interpret it another way. Neither compiler is wrong. The bug is
3653: in your code.
1.1 root 3654:
3655: After you have localized the error to a single source line, it should
3656: be easy to check for these things. If your program is correct and
3657: well defined, you have found a compiler bug.
3658:
3659: @item
3660: If the compiler produces an error message for valid input, that is a
3661: compiler bug.
3662:
3663: Note that the following is not valid input, and the error message for
3664: it is not a bug:
3665:
3666: @example
3667: int foo (char);
3668:
3669: int
3670: foo (x)
3671: char x;
3672: @{ @dots{} @}
3673: @end example
3674:
3675: @noindent
3676: The prototype says to pass a @code{char}, while the definition says to
3677: pass an @code{int} and treat the value as a @code{char}. This is what
3678: the ANSI standard says, and it makes sense.
3679:
3680: @item
3681: If the compiler does not produce an error message for invalid input,
3682: that is a compiler bug. However, you should note that your idea of
3683: ``invalid input'' might be my idea of ``an extension'' or ``support
3684: for traditional practice''.
3685:
3686: @item
3687: If you are an experienced user of C compilers, your suggestions
3688: for improvement of GNU CC are welcome in any case.
3689: @end itemize
3690:
3691: @node Bug Reporting,, Bug Criteria, Bugs
3692: @section How to Report Bugs
3693:
3694: Send bug reports for GNU C to one of these addresses:
3695:
3696: @example
3697: bug-gcc@@prep.ai.mit.edu
3698: @{ucbvax|mit-eddie|uunet@}!prep.ai.mit.edu!bug-gcc
3699: @end example
3700:
1.1.1.8 root 3701: @strong{Do not send bug reports to @samp{info-gcc}, or to the newsgroup
3702: @samp{gnu.gcc}.} Most users of GNU CC do not want to receive bug
3703: reports. Those that do, have asked to be on @samp{bug-gcc}.
3704:
3705: The mailing list @samp{bug-gcc} has a newsgroup which serves as a
3706: repeater. The mailing list and the newsgroup carry exactly the same
3707: messages. Often people think of posting bug reports to the newsgroup
3708: instead of mailing them. This appears to work, but it has one problem
3709: which can be crucial: a newsgroup posting does not contain a mail path
3710: back to the sender. Thus, if I need to ask for more information, I
3711: may be unable to reach you. For this reason, it is better to send bug
3712: reports to the mailing list.
3713:
3714: As a last resort, send bug reports on paper to:
1.1 root 3715:
3716: @example
3717: GNU Compiler Bugs
3718: 545 Tech Sq
3719: Cambridge, MA 02139
3720: @end example
3721:
3722: The fundamental principle of reporting bugs usefully is this:
1.1.1.8 root 3723: @strong{report all the facts}. If you are not sure whether to state a
3724: fact or leave it out, state it!
1.1 root 3725:
3726: Often people omit facts because they think they know what causes the
3727: problem and they conclude that some details don't matter. Thus, you might
3728: assume that the name of the variable you use in an example does not matter.
3729: Well, probably it doesn't, but one cannot be sure. Perhaps the bug is a
3730: stray memory reference which happens to fetch from the location where that
3731: name is stored in memory; perhaps, if the name were different, the contents
3732: of that location would fool the compiler into doing the right thing despite
1.1.1.8 root 3733: the bug. Play it safe and give a specific, complete example. That is the
3734: easiest thing for you to do, and the most helpful.
1.1 root 3735:
1.1.1.8 root 3736: Keep in mind that the purpose of a bug report is to enable me to fix
3737: the bug if it is not known. It isn't very important what happens if
3738: the bug is already known. Therefore, always write your bug reports on
3739: the assumption that the bug is not known.
3740:
3741: Sometimes people give a few sketchy facts and ask, ``Does this ring a
3742: bell?'' Those bug reports are useless, and I urge everyone to
3743: @emph{refuse to respond to them} except to chide the sender to report
3744: bugs properly.
3745:
3746: To enable me to fix the bug, you should include all these things:
1.1 root 3747:
3748: @itemize @bullet
3749: @item
3750: The version of GNU CC. You can get this by running it with the
3751: @samp{-v} option.
3752:
3753: Without this, I won't know whether there is any point in looking for
3754: the bug in the current version of GNU CC.
3755:
3756: @item
3757: A complete input file that will reproduce the bug. If the bug is in
3758: the C preprocessor, send me a source file and any header files that it
3759: requires. If the bug is in the compiler proper (@file{cc1}), run your
3760: source file through the C preprocessor by doing @samp{gcc -E
3761: @var{sourcefile} > @var{outfile}}, then include the contents of
3762: @var{outfile} in the bug report. (Any @samp{-I}, @samp{-D} or
3763: @samp{-U} options that you used in actual compilation should also be
3764: used when doing this.)
3765:
3766: A single statement is not enough of an example. In order to compile
3767: it, it must be embedded in a function definition; and the bug might
3768: depend on the details of how this is done.
3769:
3770: Without a real example I can compile, all I can do about your bug
3771: report is wish you luck. It would be futile to try to guess how to
3772: provoke the bug. For example, bugs in register allocation and
3773: reloading frequently depend on every little detail of the function
3774: they happen in.
3775:
3776: @item
3777: The command arguments you gave GNU CC to compile that example and
3778: observe the bug. For example, did you use @samp{-O}? To guarantee
3779: you won't omit something important, list them all.
3780:
3781: If I were to try to guess the arguments, I would probably guess wrong
3782: and then I would not encounter the bug.
3783:
3784: @item
3785: The names of the files that you used for @file{tm.h} and @file{md}
3786: when you installed the compiler.
3787:
3788: @item
3789: The type of machine you are using, and the operating system name and
3790: version number.
3791:
3792: @item
3793: A description of what behavior you observe that you believe is
3794: incorrect. For example, ``It gets a fatal signal,'' or, ``There is an
3795: incorrect assembler instruction in the output.''
3796:
3797: Of course, if the bug is that the compiler gets a fatal signal, then I
3798: will certainly notice it. But if the bug is incorrect output, I might
3799: not notice unless it is glaringly wrong. I won't study all the
3800: assembler code from a 50-line C program just on the off chance that it
3801: might be wrong.
3802:
3803: Even if the problem you experience is a fatal signal, you should still
3804: say so explicitly. Suppose something strange is going on, such as,
3805: your copy of the compiler is out of synch, or you have encountered a
3806: bug in the C library on your system. (This has happened!) Your copy
3807: might crash and mine would not. If you @i{told} me to expect a crash,
3808: then when mine fails to crash, I would know that the bug was not
3809: happening for me. If you had not told me to expect a crash, then I
3810: would not be able to draw any conclusion from my observations.
3811:
1.1.1.8 root 3812: Often the observed symptom is incorrect output when your program is run.
3813: Sad to say, this is not enough information for me unless the program is
3814: short and simple. If you send me a large program, I don't have time to
3815: figure out how it would work if compiled correctly, much less which line
3816: of it was compiled wrong. So you will have to do that. Tell me which
3817: source line it is, and what incorrect result happens when that line is
3818: executed. A person who understands the test program can find this as
3819: easily as a bug in the program itself.
1.1 root 3820:
3821: @item
3822: If you send me examples of output from GNU CC, please use @samp{-g}
3823: when you make them. The debugging information includes source line
3824: numbers which are essential for correlating the output with the input.
3825:
3826: @item
3827: If you wish to suggest changes to the GNU CC source, send me context
3828: diffs. If you even discuss something in the GNU CC source, refer to
3829: it by context, not by line number.
3830:
3831: The line numbers in my development sources don't match those in your
3832: sources. Your line numbers would convey no useful information to me.
3833:
3834: @item
3835: Additional information from a debugger might enable me to find
3836: a problem on a machine which I do not have available myself.
3837: However, you need to think when you collect this information if
3838: you want it to have any chance of being useful.
3839:
3840: For example, many people send just a backtrace, but that is never
3841: useful by itself. A simple backtrace with arguments conveys little
3842: about GNU CC because the compiler is largely data-driven; the same
3843: functions are called over and over for different RTL insns, doing
3844: different things depending on the details of the insn.
3845:
3846: Most of the arguments listed in the backtrace are useless because they
3847: are pointers to RTL list structure. The numeric values of the
3848: pointers, which the debugger prints in the backtrace, have no
3849: significance whatever; all that matters is the contents of the objects
3850: they point to (and most of the contents are other such pointers).
3851:
3852: In addition, most compiler passes consist of one or more loops that
3853: scan the RTL insn sequence. The most vital piece of information about
1.1.1.8 root 3854: such a loop---which insn it has reached---is usually in a local variable,
1.1 root 3855: not in an argument.
3856:
3857: What you need to provide in addition to a backtrace are the values of
3858: the local variables for several stack frames up. When a local
3859: variable or an argument is an RTX, first print its value and then use
3860: the GDB command @code{pr} to print the RTL expression that it points
3861: to. (If GDB doesn't run on your machine, use your debugger to call
3862: the function @code{debug_rtx} with the RTX as an argument.) In
3863: general, whenever a variable is a pointer, its value is no use
3864: without the data it points to.
3865:
3866: In addition, include a debugging dump from just before the pass
3867: in which the crash happens. Most bugs involve a series of insns,
3868: not just one.
3869: @end itemize
3870:
3871: Here are some things that are not necessary:
3872:
3873: @itemize @bullet
3874: @item
3875: A description of the envelope of the bug.
3876:
3877: Often people who encounter a bug spend a lot of time investigating
3878: which changes to the input file will make the bug go away and which
3879: changes will not affect it.
3880:
3881: This is often time consuming and not very useful, because the way I
3882: will find the bug is by running a single example under the debugger
3883: with breakpoints, not by pure deduction from a series of examples.
1.1.1.8 root 3884: I recommend that you save your time for something else.
1.1 root 3885:
3886: Of course, if you can find a simpler example to report @emph{instead}
3887: of the original one, that is a convenience for me. Errors in the
3888: output will be easier to spot, running under the debugger will take
3889: less time, etc. Most GNU CC bugs involve just one function, so the
3890: most straightforward way to simplify an example is to delete all the
3891: function definitions except the one where the bug occurs. Those
3892: earlier in the file may be replaced by external declarations if the
1.1.1.8 root 3893: crucial function depends on them. (Exception: inline functions may
3894: affect compilation of functions defined later in the file.)
1.1 root 3895:
3896: However, simplification is not vital; if you don't want to do this,
1.1.1.8 root 3897: report the bug anyway and send me the entire test case you used.
1.1 root 3898:
3899: @item
3900: A patch for the bug.
3901:
3902: A patch for the bug does help me if it is a good one. But don't omit
1.1.1.8 root 3903: the necessary information, such as the test case, on the assumption that
3904: a patch is all I need. I might see problems with your patch and decide
3905: to fix the problem another way, or I might not understand it at all.
1.1 root 3906:
3907: Sometimes with a program as complicated as GNU CC it is very hard to
3908: construct an example that will make the program follow a certain path
3909: through the code. If you don't send me the example, I won't be able
3910: to construct one, so I won't be able to verify that the bug is fixed.
3911:
1.1.1.8 root 3912: And if I can't understand what bug you are trying to fix, or why your
3913: patch should be an improvement, I won't install it. A test case will
3914: help me to understand.
3915:
1.1 root 3916: @item
3917: A guess about what the bug is or what it depends on.
3918:
3919: Such guesses are usually wrong. Even I can't guess right about such
1.1.1.8 root 3920: things without first using the debugger to find the facts.
1.1 root 3921: @end itemize
3922:
3923: @node Portability, Interface, Bugs, Top
3924: @chapter GNU CC and Portability
3925:
3926: The main goal of GNU CC was to make a good, fast compiler for machines in
3927: the class that the GNU system aims to run on: 32-bit machines that address
3928: 8-bit bytes and have several general registers. Elegance, theoretical
3929: power and simplicity are only secondary.
3930:
3931: GNU CC gets most of the information about the target machine from a machine
3932: description which gives an algebraic formula for each of the machine's
3933: instructions. This is a very clean way to describe the target. But when
3934: the compiler needs information that is difficult to express in this
3935: fashion, I have not hesitated to define an ad-hoc parameter to the machine
3936: description. The purpose of portability is to reduce the total work needed
3937: on the compiler; it was not of interest for its own sake.
3938:
3939: GNU CC does not contain machine dependent code, but it does contain code
3940: that depends on machine parameters such as endianness (whether the most
3941: significant byte has the highest or lowest address of the bytes in a word)
3942: and the availability of autoincrement addressing. In the RTL-generation
3943: pass, it is often necessary to have multiple strategies for generating code
3944: for a particular kind of syntax tree, strategies that are usable for different
3945: combinations of parameters. Often I have not tried to address all possible
3946: cases, but only the common ones or only the ones that I have encountered.
3947: As a result, a new target may require additional strategies. You will know
3948: if this happens because the compiler will call @code{abort}. Fortunately,
3949: the new strategies can be added in a machine-independent fashion, and will
3950: affect only the target machines that need them.
3951:
3952: @node Interface, Passes, Portability, Top
3953: @chapter Interfacing to GNU CC Output
3954:
3955: GNU CC is normally configured to use the same function calling convention
3956: normally in use on the target system. This is done with the
3957: machine-description macros described (@pxref{Machine Macros}).
3958:
3959: However, returning of structure and union values is done differently on
3960: some target machines. As a result, functions compiled with PCC
3961: returning such types cannot be called from code compiled with GNU CC,
3962: and vice versa. This does not cause trouble often because few Unix
3963: library routines return structures or unions.
3964:
3965: GNU CC code returns structures and unions that are 1, 2, 4 or 8 bytes
3966: long in the same registers used for @code{int} or @code{double} return
3967: values. (GNU CC typically allocates variables of such types in
3968: registers also.) Structures and unions of other sizes are returned by
3969: storing them into an address passed by the caller (usually in a
3970: register). The machine-description macros @code{STRUCT_VALUE} and
3971: @code{STRUCT_INCOMING_VALUE} tell GNU CC where to pass this address.
3972:
3973: By contrast, PCC on most target machines returns structures and unions
3974: of any size by copying the data into an area of static storage, and then
3975: returning the address of that storage as if it were a pointer value.
3976: The caller must copy the data from that memory area to the place where
3977: the value is wanted. This is slower than the method used by GNU CC, and
3978: fails to be reentrant.
3979:
3980: On some target machines, such as RISC machines and the 80386, the
3981: standard system convention is to pass to the subroutine the address of
3982: where to return the value. On these machines, GNU CC has been
3983: configured to be compatible with the standard compiler, when this method
3984: is used. It may not be compatible for structures of 1, 2, 4 or 8 bytes.
3985:
3986: GNU CC uses the system's standard convention for passing arguments. On
3987: some machines, the first few arguments are passed in registers; in
3988: others, all are passed on the stack. It would be possible to use
3989: registers for argument passing on any machine, and this would probably
3990: result in a significant speedup. But the result would be complete
3991: incompatibility with code that follows the standard convention. So this
3992: change is practical only if you are switching to GNU CC as the sole C
3993: compiler for the system. We may implement register argument passing on
3994: certain machines once we have a complete GNU system so that we can
3995: compile the libraries with GNU CC.
3996:
3997: If you use @code{longjmp}, beware of automatic variables. ANSI C says that
3998: automatic variables that are not declared @code{volatile} have undefined
3999: values after a @code{longjmp}. And this is all GNU CC promises to do,
4000: because it is very difficult to restore register variables correctly, and
4001: one of GNU CC's features is that it can put variables in registers without
4002: your asking it to.
4003:
4004: If you want a variable to be unaltered by @code{longjmp}, and you don't
4005: want to write @code{volatile} because old C compilers don't accept it,
4006: just take the address of the variable. If a variable's address is ever
4007: taken, even if just to compute it and ignore it, then the variable cannot
4008: go in a register:
4009:
4010: @example
4011: @{
4012: int careful;
4013: &careful;
4014: @dots{}
4015: @}
4016: @end example
4017:
4018: Code compiled with GNU CC may call certain library routines. Most of
4019: them handle arithmetic for which there are no instructions. This
4020: includes multiply and divide on some machines, and floating point
4021: operations on any machine for which floating point support is disabled
4022: with @samp{-msoft-float}. Some standard parts of the C library, such as
4023: @code{bcopy} or @code{memcpy}, are also called automatically. The usual
4024: function call interface is used for calling the library routines.
4025:
4026: These library routines should be defined in the library @file{gnulib},
4027: which GNU CC automatically searches whenever it links a program. On
4028: machines that have multiply and divide instructions, if hardware
4029: floating point is in use, normally @file{gnulib} is not needed, but it
4030: is searched just in case.
4031:
4032: Each arithmetic function is defined in @file{gnulib.c} to use the
4033: corresponding C arithmetic operator. As long as the file is compiled
4034: with another C compiler, which supports all the C arithmetic operators,
4035: this file will work portably. However, @file{gnulib.c} does not work if
4036: compiled with GNU CC, because each arithmetic function would compile
4037: into a call to itself!
4038:
4039: @node Passes, RTL, Interface, Top
4040: @chapter Passes and Files of the Compiler
4041:
4042: The overall control structure of the compiler is in @file{toplev.c}. This
4043: file is responsible for initialization, decoding arguments, opening and
4044: closing files, and sequencing the passes.
4045:
4046: The parsing pass is invoked only once, to parse the entire input. The RTL
4047: intermediate code for a function is generated as the function is parsed, a
4048: statement at a time. Each statement is read in as a syntax tree and then
4049: converted to RTL; then the storage for the tree for the statement is
4050: reclaimed. Storage for types (and the expressions for their sizes),
4051: declarations, and a representation of the binding contours and how they nest,
4052: remains until the function is finished being compiled; these are all needed
4053: to output the debugging information.
4054:
4055: Each time the parsing pass reads a complete function definition or
4056: top-level declaration, it calls the function
4057: @code{rest_of_compilation} or @code{rest_of_decl_compilation} in
4058: @file{toplev.c}, which are responsible for all further processing
4059: necessary, ending with output of the assembler language. All other
4060: compiler passes run, in sequence, within @code{rest_of_compilation}.
4061: When that function returns from compiling a function definition, the
4062: storage used for that function definition's compilation is entirely
4063: freed, unless it is an inline function (@pxref{Inline}).
4064:
4065: Here is a list of all the passes of the compiler and their source files.
4066: Also included is a description of where debugging dumps can be requested
4067: with @samp{-d} options.
4068:
4069: @itemize @bullet
4070: @item
4071: Parsing. This pass reads the entire text of a function definition,
4072: constructing partial syntax trees. This and RTL generation are no longer
4073: truly separate passes (formerly they were), but it is easier to think
4074: of them as separate.
4075:
4076: The tree representation does not entirely follow C syntax, because it is
4077: intended to support other languages as well.
4078:
4079: C data type analysis is also done in this pass, and every tree node
4080: that represents an expression has a data type attached. Variables are
4081: represented as declaration nodes.
4082:
4083: Constant folding and associative-law simplifications are also done
4084: during this pass.
4085:
4086: The source files for parsing are @file{c-parse.y}, @file{c-decl.c},
4087: @file{c-typeck.c}, @file{c-convert.c}, @file{stor-layout.c},
4088: @file{fold-const.c}, and @file{tree.c}. The last three files are
4089: intended to be language-independent. There are also header files
4090: @file{c-parse.h}, @file{c-tree.h}, @file{tree.h} and @file{tree.def}.
4091: The last two define the format of the tree representation.@refill
4092:
4093: @item
4094: RTL generation. This is the conversion of syntax tree into RTL code.
4095: It is actually done statement-by-statement during parsing, but for
4096: most purposes it can be thought of as a separate pass.
4097:
4098: This is where the bulk of target-parameter-dependent code is found,
4099: since often it is necessary for strategies to apply only when certain
4100: standard kinds of instructions are available. The purpose of named
4101: instruction patterns is to provide this information to the RTL
4102: generation pass.
4103:
4104: Optimization is done in this pass for @code{if}-conditions that are
4105: comparisons, boolean operations or conditional expressions. Tail
4106: recursion is detected at this time also. Decisions are made about how
4107: best to arrange loops and how to output @code{switch} statements.
4108:
4109: The source files for RTL generation are @file{stmt.c}, @file{expr.c},
4110: @file{explow.c}, @file{expmed.c}, @file{optabs.c} and @file{emit-rtl.c}.
4111: Also, the file @file{insn-emit.c}, generated from the machine description
4112: by the program @code{genemit}, is used in this pass. The header files
4113: @file{expr.h} is used for communication within this pass.@refill
4114:
4115: The header files @file{insn-flags.h} and @file{insn-codes.h},
4116: generated from the machine description by the programs @code{genflags}
4117: and @code{gencodes}, tell this pass which standard names are available
4118: for use and which patterns correspond to them.@refill
4119:
4120: Aside from debugging information output, none of the following passes
4121: refers to the tree structure representation of the function (only
4122: part of which is saved).
4123:
4124: The decision of whether the function can and should be expanded inline
4125: in its subsequent callers is made at the end of rtl generation. The
4126: function must meet certain criteria, currently related to the size of
4127: the function and the types and number of parameters it has. Note that
4128: this function may contain loops, recursive calls to itself
4129: (tail-recursive functions can be inlined!), gotos, in short, all
4130: constructs supported by GNU CC.
4131:
4132: The option @samp{-dr} causes a debugging dump of the RTL code after
4133: this pass. This dump file's name is made by appending @samp{.rtl} to
4134: the input file name.
4135:
4136: @item
4137: Jump optimization. This pass simplifies jumps to the following
4138: instruction, jumps across jumps, and jumps to jumps. It deletes
4139: unreferenced labels and unreachable code, except that unreachable code
4140: that contains a loop is not recognized as unreachable in this pass.
4141: (Such loops are deleted later in the basic block analysis.)
4142:
4143: Jump optimization is performed two or three times. The first time is
4144: immediately following RTL generation. The second time is after CSE,
4145: but only if CSE says repeated jump optimization is needed. The
4146: last time is right before the final pass. That time, cross-jumping
4147: and deletion of no-op move instructions are done together with the
4148: optimizations described above.
4149:
4150: The source file of this pass is @file{jump.c}.
4151:
4152: The option @samp{-dj} causes a debugging dump of the RTL code after
4153: this pass is run for the first time. This dump file's name is made by
4154: appending @samp{.jump} to the input file name.
4155:
4156: @item
4157: Register scan. This pass finds the first and last use of each
4158: register, as a guide for common subexpression elimination. Its source
4159: is in @file{regclass.c}.
4160:
4161: @item
4162: Common subexpression elimination. This pass also does constant
4163: propagation. Its source file is @file{cse.c}. If constant
4164: propagation causes conditional jumps to become unconditional or to
4165: become no-ops, jump optimization is run again when CSE is finished.
4166:
4167: The option @samp{-ds} causes a debugging dump of the RTL code after
4168: this pass. This dump file's name is made by appending @samp{.cse} to
4169: the input file name.
4170:
4171: @item
1.1.1.8 root 4172: Loop optimization. This pass moves constant expressions out of loops,
4173: and optionally does strength-reduction as well. Its source file is
4174: @file{loop.c}.
1.1 root 4175:
4176: The option @samp{-dL} causes a debugging dump of the RTL code after
4177: this pass. This dump file's name is made by appending @samp{.loop} to
4178: the input file name.
4179:
4180: @item
4181: Stupid register allocation is performed at this point in a
4182: nonoptimizing compilation. It does a little data flow analysis as
4183: well. When stupid register allocation is in use, the next pass
4184: executed is the reloading pass; the others in between are skipped.
4185: The source file is @file{stupid.c}.
4186:
4187: @item
4188: Data flow analysis (@file{flow.c}). This pass divides the program
4189: into basic blocks (and in the process deletes unreachable loops); then
4190: it computes which pseudo-registers are live at each point in the
4191: program, and makes the first instruction that uses a value point at
4192: the instruction that computed the value.
4193:
4194: This pass also deletes computations whose results are never used, and
4195: combines memory references with add or subtract instructions to make
4196: autoincrement or autodecrement addressing.
4197:
4198: The option @samp{-df} causes a debugging dump of the RTL code after
4199: this pass. This dump file's name is made by appending @samp{.flow} to
4200: the input file name. If stupid register allocation is in use, this
4201: dump file reflects the full results of such allocation.
4202:
4203: @item
4204: Instruction combination (@file{combine.c}). This pass attempts to
4205: combine groups of two or three instructions that are related by data
4206: flow into single instructions. It combines the RTL expressions for
4207: the instructions by substitution, simplifies the result using algebra,
4208: and then attempts to match the result against the machine description.
4209:
4210: The option @samp{-dc} causes a debugging dump of the RTL code after
4211: this pass. This dump file's name is made by appending @samp{.combine}
4212: to the input file name.
4213:
4214: @item
4215: Register class preferencing. The RTL code is scanned to find out
4216: which register class is best for each pseudo register. The source
4217: file is @file{regclass.c}.
4218:
4219: @item
4220: Local register allocation (@file{local-alloc.c}). This pass allocates
4221: hard registers to pseudo registers that are used only within one basic
4222: block. Because the basic block is linear, it can use fast and
4223: powerful techniques to do a very good job.
4224:
4225: The option @samp{-dl} causes a debugging dump of the RTL code after
4226: this pass. This dump file's name is made by appending @samp{.lreg} to
4227: the input file name.
4228:
4229: @item
4230: Global register allocation (@file{global-alloc.c}). This pass
4231: allocates hard registers for the remaining pseudo registers (those
4232: whose life spans are not contained in one basic block).
4233:
4234: @item
4235: Reloading. This pass renumbers pseudo registers with the hardware
4236: registers numbers they were allocated. Pseudo registers that did not
4237: get hard registers are replaced with stack slots. Then it finds
4238: instructions that are invalid because a value has failed to end up in
4239: a register, or has ended up in a register of the wrong kind. It fixes
4240: up these instructions by reloading the problematical values
4241: temporarily into registers. Additional instructions are generated to
4242: do the copying.
4243:
4244: Source files are @file{reload.c} and @file{reload1.c}, plus the header
4245: @file{reload.h} used for communication between them.
4246:
4247: The option @samp{-dg} causes a debugging dump of the RTL code after
4248: this pass. This dump file's name is made by appending @samp{.greg} to
4249: the input file name.
4250:
4251: @item
4252: Jump optimization is repeated, this time including cross-jumping
1.1.1.5 root 4253: and deletion of no-op move instructions.
1.1 root 4254:
4255: The option @samp{-dJ} causes a debugging dump of the RTL code after
4256: this pass. This dump file's name is made by appending @samp{.jump2}
4257: to the input file name.
4258:
4259: @item
1.1.1.8 root 4260: Delayed branch scheduling may be done at this point. The source file
4261: name is @file{dbranch.c}.
4262:
4263: The option @samp{-dd} causes a debugging dump of the RTL code after
4264: this pass. This dump file's name is made by appending @samp{.dbr}
4265: to the input file name.
4266:
4267: @item
1.1 root 4268: Final. This pass outputs the assembler code for the function. It is
4269: also responsible for identifying spurious test and compare
1.1.1.5 root 4270: instructions. Machine-specific peephole optimizations are performed
4271: at the same time. The function entry and exit sequences are generated
1.1 root 4272: directly as assembler code in this pass; they never exist as RTL.
4273:
4274: The source files are @file{final.c} plus @file{insn-output.c}; the
4275: latter is generated automatically from the machine description by the
4276: tool @file{genoutput}. The header file @file{conditions.h} is used
4277: for communication between these files.
4278:
4279: @item
4280: Debugging information output. This is run after final because it must
4281: output the stack slot offsets for pseudo registers that did not get
4282: hard registers. Source files are @file{dbxout.c} for DBX symbol table
4283: format and @file{symout.c} for GDB's own symbol table format.
4284: @end itemize
4285:
4286: Some additional files are used by all or many passes:
4287:
4288: @itemize @bullet
4289: @item
4290: Every pass uses @file{machmode.def}, which defines the machine modes.
4291:
4292: @item
4293: All the passes that work with RTL use the header files @file{rtl.h}
4294: and @file{rtl.def}, and subroutines in file @file{rtl.c}. The tools
4295: @code{gen*} also use these files to read and work with the machine
4296: description RTL.
4297:
4298: @item
4299: Several passes refer to the header file @file{insn-config.h} which
4300: contains a few parameters (C macro definitions) generated
4301: automatically from the machine description RTL by the tool
4302: @code{genconfig}.
4303:
4304: @item
4305: Several passes use the instruction recognizer, which consists of
4306: @file{recog.c} and @file{recog.h}, plus the files @file{insn-recog.c}
4307: and @file{insn-extract.c} that are generated automatically from the
4308: machine description by the tools @file{genrecog} and
4309: @file{genextract}.@refill
4310:
4311: @item
4312: Several passes use the header files @file{regs.h} which defines the
4313: information recorded about pseudo register usage, and @file{basic-block.h}
4314: which defines the information recorded about basic blocks.
4315:
4316: @item
4317: @file{hard-reg-set.h} defines the type @code{HARD_REG_SET}, a bit-vector
4318: with a bit for each hard register, and some macros to manipulate it.
4319: This type is just @code{int} if the machine has few enough hard registers;
4320: otherwise it is an array of @code{int} and some of the macros expand
4321: into loops.
4322: @end itemize
4323:
4324: @node RTL, Machine Desc, Passes, Top
4325: @chapter RTL Representation
4326:
4327: Most of the work of the compiler is done on an intermediate representation
4328: called register transfer language. In this language, the instructions to be
4329: output are described, pretty much one by one, in an algebraic form that
4330: describes what the instruction does.
4331:
4332: RTL is inspired by Lisp lists. It has both an internal form, made up of
4333: structures that point at other structures, and a textual form that is used
4334: in the machine description and in printed debugging dumps. The textual
4335: form uses nested parentheses to indicate the pointers in the internal form.
4336:
4337: @menu
4338: * RTL Objects:: Expressions vs vectors vs strings vs integers.
4339: * Accessors:: Macros to access expression operands or vector elts.
4340: * Flags:: Other flags in an RTL expression.
4341: * Machine Modes:: Describing the size and format of a datum.
4342: * Constants:: Expressions with constant values.
4343: * Regs and Memory:: Expressions representing register contents or memory.
4344: * Arithmetic:: Expressions representing arithmetic on other expressions.
4345: * Comparisons:: Expressions representing comparison of expressions.
4346: * Bit Fields:: Expressions representing bit-fields in memory or reg.
4347: * Conversions:: Extending, truncating, floating or fixing.
4348: * RTL Declarations:: Declaring volatility, constancy, etc.
4349: * Side Effects:: Expressions for storing in registers, etc.
4350: * Incdec:: Embedded side-effects for autoincrement addressing.
1.1.1.9 root 4351: * Assembler:: Representing @code{asm} with operands.
1.1 root 4352: * Insns:: Expression types for entire insns.
1.1.1.9 root 4353: * Calls:: RTL representation of function call insns.
1.1 root 4354: * Sharing:: Some expressions are unique; others *must* be copied.
4355: @end menu
4356:
4357: @node RTL Objects, Accessors, RTL, RTL
4358: @section RTL Object Types
4359:
4360: RTL uses four kinds of objects: expressions, integers, strings and vectors.
4361: Expressions are the most important ones. An RTL expression (``RTX'', for
4362: short) is a C structure, but it is usually referred to with a pointer; a
4363: type that is given the typedef name @code{rtx}.
4364:
4365: An integer is simply an @code{int}, and a string is a @code{char *}.
1.1.1.8 root 4366: Within RTL code, strings appear only inside @code{symbol_ref} expressions,
1.1 root 4367: but they appear in other contexts in the RTL expressions that make up
4368: machine descriptions. Their written form uses decimal digits.
4369:
4370: A string is a sequence of characters. In core it is represented as a
4371: @code{char *} in usual C fashion, and it is written in C syntax as well.
4372: However, strings in RTL may never be null. If you write an empty string in
4373: a machine description, it is represented in core as a null pointer rather
4374: than as a pointer to a null character. In certain contexts, these null
4375: pointers instead of strings are valid.
4376:
4377: A vector contains an arbitrary, specified number of pointers to
4378: expressions. The number of elements in the vector is explicitly present in
4379: the vector. The written form of a vector consists of square brackets
4380: (@samp{[@dots{}]}) surrounding the elements, in sequence and with
4381: whitespace separating them. Vectors of length zero are not created; null
4382: pointers are used instead.
4383:
4384: Expressions are classified by @dfn{expression codes} (also called RTX
4385: codes). The expression code is a name defined in @file{rtl.def}, which is
4386: also (in upper case) a C enumeration constant. The possible expression
4387: codes and their meanings are machine-independent. The code of an RTX can
4388: be extracted with the macro @code{GET_CODE (@var{x})} and altered with
4389: @code{PUT_CODE (@var{x}, @var{newcode})}.
4390:
4391: The expression code determines how many operands the expression contains,
4392: and what kinds of objects they are. In RTL, unlike Lisp, you cannot tell
4393: by looking at an operand what kind of object it is. Instead, you must know
4394: from its context---from the expression code of the containing expression.
1.1.1.8 root 4395: For example, in an expression of code @code{subreg}, the first operand is
1.1 root 4396: to be regarded as an expression and the second operand as an integer. In
1.1.1.8 root 4397: an expression of code @code{plus}, there are two operands, both of which
4398: are to be regarded as expressions. In a @code{symbol_ref} expression,
1.1 root 4399: there is one operand, which is to be regarded as a string.
4400:
4401: Expressions are written as parentheses containing the name of the
4402: expression type, its flags and machine mode if any, and then the operands
4403: of the expression (separated by spaces).
4404:
4405: Expression code names in the @samp{md} file are written in lower case,
4406: but when they appear in C code they are written in upper case. In this
1.1.1.8 root 4407: manual, they are shown as follows: @code{const_int}.
1.1 root 4408:
4409: In a few contexts a null pointer is valid where an expression is normally
1.1.1.4 root 4410: wanted. The written form of this is @code{(nil)}.
1.1 root 4411:
4412: @node Accessors, Flags, RTL Objects, RTL
4413: @section Access to Operands
4414:
4415: For each expression type @file{rtl.def} specifies the number of contained
4416: objects and their kinds, with four possibilities: @samp{e} for expression
4417: (actually a pointer to an expression), @samp{i} for integer, @samp{s} for
4418: string, and @samp{E} for vector of expressions. The sequence of letters
4419: for an expression code is called its @dfn{format}. Thus, the format of
1.1.1.8 root 4420: @code{subreg} is @samp{ei}.@refill
1.1 root 4421:
4422: Two other format characters are used occasionally: @samp{u} and @samp{0}.
4423: @samp{u} is equivalent to @samp{e} except that it is printed differently in
4424: debugging dumps, and @samp{0} means a slot whose contents do not fit any
4425: normal category. @samp{0} slots are not printed at all in dumps, and are
4426: often used in special ways by small parts of the compiler.@refill
4427:
4428: There are macros to get the number of operands and the format of an
4429: expression code:
4430:
4431: @table @code
4432: @item GET_RTX_LENGTH (@var{code})
4433: Number of operands of an RTX of code @var{code}.
4434:
4435: @item GET_RTX_FORMAT (@var{code})
4436: The format of an RTX of code @var{code}, as a C string.
4437: @end table
4438:
4439: Operands of expressions are accessed using the macros @code{XEXP},
4440: @code{XINT} and @code{XSTR}. Each of these macros takes two arguments: an
4441: expression-pointer (RTX) and an operand number (counting from zero).
4442: Thus,@refill
4443:
4444: @example
4445: XEXP (@var{x}, 2)
4446: @end example
4447:
4448: @noindent
4449: accesses operand 2 of expression @var{x}, as an expression.
4450:
4451: @example
4452: XINT (@var{x}, 2)
4453: @end example
4454:
4455: @noindent
4456: accesses the same operand as an integer. @code{XSTR}, used in the same
4457: fashion, would access it as a string.
4458:
4459: Any operand can be accessed as an integer, as an expression or as a string.
4460: You must choose the correct method of access for the kind of value actually
4461: stored in the operand. You would do this based on the expression code of
4462: the containing expression. That is also how you would know how many
4463: operands there are.
4464:
1.1.1.8 root 4465: For example, if @var{x} is a @code{subreg} expression, you know that it has
1.1 root 4466: two operands which can be correctly accessed as @code{XEXP (@var{x}, 0)}
4467: and @code{XINT (@var{x}, 1)}. If you did @code{XINT (@var{x}, 0)}, you
4468: would get the address of the expression operand but cast as an integer;
4469: that might occasionally be useful, but it would be cleaner to write
4470: @code{(int) XEXP (@var{x}, 0)}. @code{XEXP (@var{x}, 1)} would also
4471: compile without error, and would return the second, integer operand cast as
4472: an expression pointer, which would probably result in a crash when
4473: accessed. Nothing stops you from writing @code{XEXP (@var{x}, 28)} either,
4474: but this will access memory past the end of the expression with
4475: unpredictable results.@refill
4476:
4477: Access to operands which are vectors is more complicated. You can use the
4478: macro @code{XVEC} to get the vector-pointer itself, or the macros
4479: @code{XVECEXP} and @code{XVECLEN} to access the elements and length of a
4480: vector.
4481:
4482: @table @code
4483: @item XVEC (@var{exp}, @var{idx})
4484: Access the vector-pointer which is operand number @var{idx} in @var{exp}.
4485:
4486: @item XVECLEN (@var{exp}, @var{idx})
4487: Access the length (number of elements) in the vector which is
4488: in operand number @var{idx} in @var{exp}. This value is an @code{int}.
4489:
4490: @item XVECEXP (@var{exp}, @var{idx}, @var{eltnum})
4491: Access element number @var{eltnum} in the vector which is
4492: in operand number @var{idx} in @var{exp}. This value is an RTX.
4493:
4494: It is up to you to make sure that @var{eltnum} is not negative
4495: and is less than @code{XVECLEN (@var{exp}, @var{idx})}.
4496: @end table
4497:
4498: All the macros defined in this section expand into lvalues and therefore
4499: can be used to assign the operands, lengths and vector elements as well as
4500: to access them.
4501:
4502: @node Flags, Machine Modes, Accessors, RTL
4503: @section Flags in an RTL Expression
4504:
4505: RTL expressions contain several flags (one-bit bit-fields) that are used
4506: in certain types of expression. Most often they are accessed with the
4507: following macros:
4508:
4509: @table @code
1.1.1.10! root 4510: @item EXTERNAL_SYMBOL_P (@var{x})
! 4511: In a @code{symbol_ref} expression, nonzero if it corresponds to a variable
! 4512: declared extern in the users code. Zero for all other variables. Stored in
! 4513: the @code{volatil} field and printed as @samp{/v}.
! 4514:
1.1 root 4515: @item MEM_VOLATILE_P (@var{x})
1.1.1.8 root 4516: In @code{mem} expressions, nonzero for volatile memory references.
1.1 root 4517: Stored in the @code{volatil} field and printed as @samp{/v}.
4518:
4519: @item MEM_IN_STRUCT_P (@var{x})
1.1.1.8 root 4520: In @code{mem} expressions, nonzero for reference to an entire
1.1 root 4521: structure, union or array, or to a component of one. Zero for
4522: references to a scalar variable or through a pointer to a scalar.
4523: Stored in the @code{in_struct} field and printed as @samp{/s}.
4524:
4525: @item REG_USER_VAR_P (@var{x})
1.1.1.8 root 4526: In a @code{reg}, nonzero if it corresponds to a variable present in
1.1 root 4527: the user's source code. Zero for temporaries generated internally by
4528: the compiler. Stored in the @code{volatil} field and printed as
4529: @samp{/v}.
4530:
4531: @item REG_FUNCTION_VALUE_P (@var{x})
1.1.1.8 root 4532: Nonzero in a @code{reg} if it is the place in which this function's
1.1 root 4533: value is going to be returned. (This happens only in a hard
4534: register.) Stored in the @code{integrated} field and printed as
4535: @samp{/i}.
4536:
4537: The same hard register may be used also for collecting the values of
4538: functions called by this one, but @code{REG_FUNCTION_VALUE_P} is zero
4539: in this kind of use.
4540:
4541: @item RTX_UNCHANGING_P (@var{x})
1.1.1.8 root 4542: Nonzero in a @code{reg} or @code{mem} if the value is not changed
1.1 root 4543: explicitly by the current function. (If it is a memory reference then
4544: it may be changed by other functions or by aliasing.) Stored in the
4545: @code{unchanging} field and printed as @samp{/u}.
4546:
4547: @item RTX_INTEGRATED_P (@var{insn})
4548: Nonzero in an insn if it resulted from an in-line function call.
4549: Stored in the @code{integrated} field and printed as @samp{/i}. This
4550: may be deleted; nothing currently depends on it.
4551:
4552: @item INSN_DELETED_P (@var{insn})
4553: In an insn, nonzero if the insn has been deleted. Stored in the
4554: @code{volatil} field and printed as @samp{/v}.
4555:
4556: @item CONSTANT_POOL_ADDRESS_P (@var{x})
1.1.1.8 root 4557: Nonzero in a @code{symbol_ref} if it refers to part of the current
1.1 root 4558: function's ``constants pool''. These are addresses close to the
4559: beginning of the function, and GNU CC assumes they can be addressed
4560: directly (perhaps with the help of base registers). Stored in the
4561: @code{unchanging} field and printed as @samp{/u}.
4562: @end table
4563:
4564: These are the fields which the above macros refer to:
4565:
4566: @table @code
4567: @item used
4568: This flag is used only momentarily, at the end of RTL generation for a
4569: function, to count the number of times an expression appears in insns.
4570: Expressions that appear more than once are copied, according to the
4571: rules for shared structure (@pxref{Sharing}).
4572:
4573: @item volatil
1.1.1.10! root 4574: This flag is used in @code{mem},@code{symbol_ref} and @code{reg} expressions
! 4575: and in insns. In RTL dump files, it is printed as @samp{/v}.
1.1 root 4576:
1.1.1.8 root 4577: In a @code{mem} expression, it is 1 if the memory reference is volatile.
1.1 root 4578: Volatile memory references may not be deleted, reordered or combined.
4579:
1.1.1.8 root 4580: In a @code{reg} expression, it is 1 if the value is a user-level variable.
1.1 root 4581: 0 indicates an internal compiler temporary.
4582:
1.1.1.10! root 4583: In a @code{symbol_ref} expression, it is 1 if the symbol is declared
! 4584: @code{extern}.
! 4585:
1.1 root 4586: In an insn, 1 means the insn has been deleted.
4587:
4588: @item in_struct
1.1.1.8 root 4589: This flag is used in @code{mem} expressions. It is 1 if the memory
1.1 root 4590: datum referred to is all or part of a structure or array; 0 if it is (or
4591: might be) a scalar variable. A reference through a C pointer has 0
4592: because the pointer might point to a scalar variable.
4593:
4594: This information allows the compiler to determine something about possible
4595: cases of aliasing.
4596:
4597: In an RTL dump, this flag is represented as @samp{/s}.
4598:
4599: @item unchanging
1.1.1.8 root 4600: This flag is used in @code{reg} and @code{mem} expressions. 1 means
1.1 root 4601: that the value of the expression never changes (at least within the
4602: current function).
4603:
4604: In an RTL dump, this flag is represented as @samp{/u}.
4605:
4606: @item integrated
4607: In some kinds of expressions, including insns, this flag means the
4608: rtl was produced by procedure integration.
4609:
1.1.1.8 root 4610: In a @code{reg} expression, this flag indicates the register
1.1 root 4611: containing the value to be returned by the current function. On
4612: machines that pass parameters in registers, the same register number
4613: may be used for parameters as well, but this flag is not set on such
4614: uses.
4615: @end table
4616:
4617: @node Machine Modes, Constants, Flags, RTL
4618: @section Machine Modes
4619:
4620: A machine mode describes a size of data object and the representation used
4621: for it. In the C code, machine modes are represented by an enumeration
4622: type, @code{enum machine_mode}, defined in @file{machmode.def}. Each RTL
4623: expression has room for a machine mode and so do certain kinds of tree
4624: expressions (declarations and types, to be precise).
4625:
4626: In debugging dumps and machine descriptions, the machine mode of an RTL
4627: expression is written after the expression code with a colon to separate
4628: them. The letters @samp{mode} which appear at the end of each machine mode
1.1.1.8 root 4629: name are omitted. For example, @code{(reg:SI 38)} is a @code{reg}
1.1 root 4630: expression with machine mode @code{SImode}. If the mode is
4631: @code{VOIDmode}, it is not written at all.
4632:
4633: Here is a table of machine modes.
4634:
4635: @table @code
4636: @item QImode
4637: ``Quarter-Integer'' mode represents a single byte treated as an integer.
4638:
4639: @item HImode
4640: ``Half-Integer'' mode represents a two-byte integer.
4641:
1.1.1.7 root 4642: @item PSImode
4643: ``Partial Single Integer'' mode represents an integer which occupies
4644: four bytes but which doesn't really use all four. On some machines,
4645: this is the right mode to use for pointers.
4646:
1.1 root 4647: @item SImode
4648: ``Single Integer'' mode represents a four-byte integer.
4649:
1.1.1.7 root 4650: @item PDImode
4651: ``Partial Double Integer'' mode represents an integer which occupies
4652: eight bytes but which doesn't really use all eight. On some machines,
4653: this is the right mode to use for certain pointers.
4654:
1.1 root 4655: @item DImode
4656: ``Double Integer'' mode represents an eight-byte integer.
4657:
4658: @item TImode
4659: ``Tetra Integer'' (?) mode represents a sixteen-byte integer.
4660:
4661: @item SFmode
4662: ``Single Floating'' mode represents a single-precision (four byte) floating
4663: point number.
4664:
4665: @item DFmode
4666: ``Double Floating'' mode represents a double-precision (eight byte) floating
4667: point number.
4668:
1.1.1.7 root 4669: @item XFmode
4670: ``Extended Floating'' mode represents a triple-precision (twelve byte)
4671: floating point number. This mode is used for IEEE extended floating
4672: point.
4673:
1.1 root 4674: @item TFmode
4675: ``Tetra Floating'' mode represents a quadruple-precision (sixteen byte)
4676: floating point number.
4677:
4678: @item BLKmode
4679: ``Block'' mode represents values that are aggregates to which none of
4680: the other modes apply. In RTL, only memory references can have this mode,
4681: and only if they appear in string-move or vector instructions. On machines
4682: which have no such instructions, @code{BLKmode} will not appear in RTL.
4683:
4684: @item VOIDmode
4685: Void mode means the absence of a mode or an unspecified mode.
1.1.1.8 root 4686: For example, RTL expressions of code @code{const_int} have mode
1.1 root 4687: @code{VOIDmode} because they can be taken to have whatever mode the context
4688: requires. In debugging dumps of RTL, @code{VOIDmode} is expressed by
4689: the absence of any mode.
4690:
4691: @item EPmode
4692: ``Entry Pointer'' mode is intended to be used for function variables in
4693: Pascal and other block structured languages. Such values contain
4694: both a function address and a static chain pointer for access to
4695: automatic variables of outer levels. This mode is only partially
4696: implemented since C does not use it.
4697:
4698: @item CSImode@r{, @dots{}}
4699: ``Complex Single Integer'' mode stands for a complex number represented
4700: as a pair of @code{SImode} integers. Any of the integer and floating modes
4701: may have @samp{C} prefixed to its name to obtain a complex number mode.
4702: For example, there are @code{CQImode}, @code{CSFmode}, and @code{CDFmode}.
4703: Since C does not support complex numbers, these machine modes are only
4704: partially implemented.
4705:
4706: @item BImode
4707: This is the machine mode of a bit-field in a structure. It is used
4708: only in the syntax tree, never in RTL, and in the syntax tree it appears
4709: only in declaration nodes. In C, it appears only in @code{FIELD_DECL}
4710: nodes for structure fields defined with a bit size.
4711: @end table
4712:
4713: The machine description defines @code{Pmode} as a C macro which expands
4714: into the machine mode used for addresses. Normally this is @code{SImode}.
4715:
4716: The only modes which a machine description @i{must} support are
4717: @code{QImode}, @code{SImode}, @code{SFmode} and @code{DFmode}. The
4718: compiler will attempt to use @code{DImode} for two-word structures and
1.1.1.7 root 4719: unions, but this can be prevented by overriding the definition of
4720: @code{MAX_FIXED_MODE_SIZE}. Likewise, you can arrange for the C type
4721: @code{short int} to avoid using @code{HImode}. In the long term it
4722: might be desirable to make the set of available machine modes
4723: machine-dependent and eliminate all assumptions about specific machine
4724: modes or their uses from the machine-independent code of the compiler.
1.1 root 4725:
1.1.1.4 root 4726: To help begin this process, the machine modes are divided into mode
4727: classes. These are represented by the enumeration type @code{enum
4728: mode_class} defined in @file{rtl.h}. The possible mode classes are:
4729:
4730: @table @code
4731: @item MODE_INT
4732: Integer modes. By default these are @code{QImode}, @code{HImode},
4733: @code{SImode}, @code{DImode}, @code{TImode}, and also @code{BImode}.
4734:
4735: @item MODE_FLOAT
4736: Floating-point modes. By default these are @code{QFmode},
4737: @code{HFmode}, @code{SFmode}, @code{DFmode} and @code{TFmode}, but the
4738: MC68881 also defines @code{XFmode} to be an 80-bit extended-precision
4739: floating-point mode.
4740:
4741: @item MODE_COMPLEX_INT
4742: Complex integer modes. By default these are @code{CQImode},
4743: @code{CHImode}, @code{CSImode}, @code{CDImode} and @code{CTImode}.
4744:
4745: @item MODE_COMPLEX_FLOAT
4746: Complex floating-point modes. By default these are @code{CQFmode},
4747: @code{CHFmode}, @code{CSFmode}, @code{CDFmode} and @code{CTFmode},
4748:
4749: @item MODE_FUNCTION
4750: Algol or Pascal function variables including a static chain.
4751: (These are not currently implemented).
4752:
4753: @item MODE_RANDOM
4754: This is a catchall mode class for modes which don't fit into the above
4755: classes. Currently @code{VOIDmode}, @code{BLKmode} and @code{EPmode}
4756: are in @code{MODE_RANDOM}.
4757: @end table
4758:
1.1 root 4759: Here are some C macros that relate to machine modes:
4760:
4761: @table @code
4762: @item GET_MODE (@var{x})
4763: Returns the machine mode of the RTX @var{x}.
4764:
4765: @item PUT_MODE (@var{x}, @var{newmode})
4766: Alters the machine mode of the RTX @var{x} to be @var{newmode}.
4767:
1.1.1.4 root 4768: @item NUM_MACHINE_MODES
4769: Stands for the number of machine modes available on the target
4770: machine. This is one greater than the largest numeric value of any
4771: machine mode.
4772:
4773: @item GET_MODE_NAME (@var{m})
4774: Returns the name of mode @var{m} as a string.
4775:
4776: @item GET_MODE_CLASS (@var{m})
4777: Returns the mode class of mode @var{m}.
4778:
1.1 root 4779: @item GET_MODE_SIZE (@var{m})
4780: Returns the size in bytes of a datum of mode @var{m}.
4781:
4782: @item GET_MODE_BITSIZE (@var{m})
4783: Returns the size in bits of a datum of mode @var{m}.
4784:
4785: @item GET_MODE_UNIT_SIZE (@var{m})
4786: Returns the size in bits of the subunits of a datum of mode @var{m}.
4787: This is the same as @code{GET_MODE_SIZE} except in the case of
4788: complex modes and @code{EPmode}. For them, the unit size is the
4789: size of the real or imaginary part, or the size of the function
4790: pointer or the context pointer.
4791: @end table
4792:
4793: @node Constants, Regs and Memory, Machine Modes, RTL
4794: @section Constant Expression Types
4795:
4796: The simplest RTL expressions are those that represent constant values.
4797:
4798: @table @code
4799: @item (const_int @var{i})
4800: This type of expression represents the integer value @var{i}. @var{i}
4801: is customarily accessed with the macro @code{INTVAL} as in
4802: @code{INTVAL (@var{exp})}, which is equivalent to @code{XINT (@var{exp}, 0)}.
4803:
4804: There is only one expression object for the integer value zero;
4805: it is the value of the variable @code{const0_rtx}. Likewise, the
4806: only expression for integer value one is found in @code{const1_rtx}.
1.1.1.8 root 4807: Any attempt to create an expression of code @code{const_int} and
1.1 root 4808: value zero or one will return @code{const0_rtx} or @code{const1_rtx}
4809: as appropriate.
4810:
4811: @item (const_double:@var{m} @var{i0} @var{i1})
1.1.1.6 root 4812: Represents a 64-bit constant of mode @var{m}. All floating point
1.1 root 4813: constants are represented in this way, and so are 64-bit @code{DImode}
4814: integer constants.
4815:
4816: The two integers @var{i0} and @var{i1} together contain the bits of
4817: the value. If the constant is floating point (either single or double
4818: precision), then they represent a @code{double}. To convert them to a
4819: @code{double}, do
4820:
4821: @example
4822: union @{ double d; int i[2];@} u;
1.1.1.8 root 4823: u.i[0] = CONST_DOUBLE_LOW(x);
4824: u.i[1] = CONST_DOUBLE_HIGH(x);
1.1 root 4825: @end example
4826:
4827: @noindent
4828: and then refer to @code{u.d}.
4829:
4830: The global variables @code{dconst0_rtx} and @code{fconst0_rtx} hold
1.1.1.8 root 4831: @code{const_double} expressions with value 0, in modes @code{DFmode}
1.1.1.7 root 4832: and @code{SFmode}, respectively. The macro @code{CONST0_RTX
1.1.1.8 root 4833: (@var{mode})} refers to a @code{const_double} expression with value 0
1.1.1.7 root 4834: in mode @var{mode}. The mode @var{mode} must be of mode class
4835: @code{MODE_FLOAT}.
1.1 root 4836:
4837: @item (symbol_ref @var{symbol})
4838: Represents the value of an assembler label for data. @var{symbol} is
4839: a string that describes the name of the assembler label. If it starts
4840: with a @samp{*}, the label is the rest of @var{symbol} not including
4841: the @samp{*}. Otherwise, the label is @var{symbol}, prefixed with
4842: @samp{_}.
4843:
4844: @item (label_ref @var{label})
4845: Represents the value of an assembler label for code. It contains one
1.1.1.8 root 4846: operand, an expression, which must be a @code{code_label} that appears
1.1 root 4847: in the instruction sequence to identify the place where the label
4848: should go.
4849:
4850: The reason for using a distinct expression type for code label
4851: references is so that jump optimization can distinguish them.
4852:
4853: @item (const @var{exp})
4854: Represents a constant that is the result of an assembly-time
4855: arithmetic computation. The operand, @var{exp}, is an expression that
1.1.1.8 root 4856: contains only constants (@code{const_int}, @code{symbol_ref} and
4857: @code{label_ref} expressions) combined with @code{plus} and
4858: @code{minus}. However, not all combinations are valid, since the
1.1 root 4859: assembler cannot do arbitrary arithmetic on relocatable symbols.
4860: @end table
4861:
4862: @node Regs and Memory, Arithmetic, Constants, RTL
4863: @section Registers and Memory
4864:
4865: Here are the RTL expression types for describing access to machine
4866: registers and to main memory.
4867:
4868: @table @code
4869: @item (reg:@var{m} @var{n})
4870: For small values of the integer @var{n} (less than
4871: @code{FIRST_PSEUDO_REGISTER}), this stands for a reference to machine
4872: register number @var{n}: a @dfn{hard register}. For larger values of
4873: @var{n}, it stands for a temporary value or @dfn{pseudo register}.
4874: The compiler's strategy is to generate code assuming an unlimited
4875: number of such pseudo registers, and later convert them into hard
4876: registers or into memory references.
4877:
4878: The symbol @code{FIRST_PSEUDO_REGISTER} is defined by the machine
4879: description, since the number of hard registers on the machine is an
4880: invariant characteristic of the machine. Note, however, that not
4881: all of the machine registers must be general registers. All the
4882: machine registers that can be used for storage of data are given
4883: hard register numbers, even those that can be used only in certain
4884: instructions or can hold only certain types of data.
4885:
4886: Each pseudo register number used in a function's RTL code is
1.1.1.8 root 4887: represented by a unique @code{reg} expression.
1.1 root 4888:
4889: @var{m} is the machine mode of the reference. It is necessary because
4890: machines can generally refer to each register in more than one mode.
4891: For example, a register may contain a full word but there may be
4892: instructions to refer to it as a half word or as a single byte, as
4893: well as instructions to refer to it as a floating point number of
4894: various precisions.
4895:
4896: Even for a register that the machine can access in only one mode,
4897: the mode must always be specified.
4898:
4899: A hard register may be accessed in various modes throughout one
4900: function, but each pseudo register is given a natural mode
4901: and is accessed only in that mode. When it is necessary to describe
1.1.1.8 root 4902: an access to a pseudo register using a nonnatural mode, a @code{subreg}
1.1 root 4903: expression is used.
4904:
1.1.1.8 root 4905: A @code{reg} expression with a machine mode that specifies more than
1.1 root 4906: one word of data may actually stand for several consecutive registers.
4907: If in addition the register number specifies a hardware register, then
4908: it actually represents several consecutive hardware registers starting
4909: with the specified one.
4910:
1.1.1.8 root 4911: Such multi-word hardware register @code{reg} expressions must not be live
1.1 root 4912: across the boundary of a basic block. The lifetime analysis pass does not
4913: know how to record properly that several consecutive registers are
4914: actually live there, and therefore register allocation would be confused.
4915: The CSE pass must go out of its way to make sure the situation does
4916: not arise.
4917:
4918: @item (subreg:@var{m} @var{reg} @var{wordnum})
1.1.1.8 root 4919: @code{subreg} expressions are used to refer to a register in a machine
1.1 root 4920: mode other than its natural one, or to refer to one register of
1.1.1.8 root 4921: a multi-word @code{reg} that actually refers to several registers.
1.1 root 4922:
4923: Each pseudo-register has a natural mode. If it is necessary to
4924: operate on it in a different mode---for example, to perform a fullword
1.1.1.8 root 4925: move instruction on a pseudo-register that contains a single
4926: byte---the pseudo-register must be enclosed in a @code{subreg}. In
4927: such a case, @var{wordnum} is zero.
1.1 root 4928:
1.1.1.8 root 4929: The other use of @code{subreg} is to extract the individual registers
1.1 root 4930: of a multi-register value. Machine modes such as @code{DImode} and
4931: @code{EPmode} indicate values longer than a word, values which usually
4932: require two consecutive registers. To access one of the registers,
1.1.1.8 root 4933: use a @code{subreg} with mode @code{SImode} and a @var{wordnum} that
1.1 root 4934: says which register.
4935:
4936: The compilation parameter @code{WORDS_BIG_ENDIAN}, if defined, says
4937: that word number zero is the most significant part; otherwise, it is
4938: the least significant part.
4939:
4940: Between the combiner pass and the reload pass, it is possible to have
1.1.1.8 root 4941: a @code{subreg} which contains a @code{mem} instead of a @code{reg} as
1.1 root 4942: its first operand. The reload pass eliminates these cases by
1.1.1.8 root 4943: reloading the @code{mem} into a suitable register.
1.1 root 4944:
4945: Note that it is not valid to access a @code{DFmode} value in @code{SFmode}
1.1.1.8 root 4946: using a @code{subreg}. On some machines the most significant part of a
1.1 root 4947: @code{DFmode} value does not have the same format as a single-precision
4948: floating value.
4949:
4950: @item (cc0)
4951: This refers to the machine's condition code register. It has no
1.1.1.10! root 4952: operands and may not have a machine mode. There are two ways to use it:
! 4953:
! 4954: @itemize @bullet
! 4955: @item
! 4956: To stand for a complete set of condition code flags. This is best on
! 4957: most machines, where each comparison sets the entire series of flags.
! 4958:
! 4959: With this technique, @code{(cc0)} may be validly used in only two
! 4960: contexts: as the destination of an assignment (in test and compare
! 4961: instructions) and in comparison operators comparing against zero
! 4962: (@code{const_int} with value zero; that is to say, @code{const0_rtx}).
! 4963:
! 4964: @item
! 4965: To stand for a single flag that is the result of a single condition.
! 4966: This is useful on machines that have only a single flag bit, and in
! 4967: which comparison instructions must specify the condition to test.
! 4968:
! 4969: With this technique, @code{(cc0)} may be validly used in only two
! 4970: contexts: as the destination of an assignment (in test and compare
! 4971: instructions) where the source is a comparison operator, and as the
! 4972: first operand of @code{if_then_else} (in a conditional branch).
! 4973: @end itemize
1.1 root 4974:
1.1.1.8 root 4975: There is only one expression object of code @code{cc0}; it is the
1.1 root 4976: value of the variable @code{cc0_rtx}. Any attempt to create an
1.1.1.8 root 4977: expression of code @code{cc0} will return @code{cc0_rtx}.
1.1 root 4978:
4979: One special thing about the condition code register is that
4980: instructions can set it implicitly. On many machines, nearly all
4981: instructions set the condition code based on the value that they
4982: compute or store. It is not necessary to record these actions
4983: explicitly in the RTL because the machine description includes a
4984: prescription for recognizing the instructions that do so (by means of
4985: the macro @code{NOTICE_UPDATE_CC}). Only instructions whose sole
4986: purpose is to set the condition code, and instructions that use the
4987: condition code, need mention @code{(cc0)}.
4988:
1.1.1.10! root 4989: In some cases, better code may result from recognizing combinations or
! 4990: peepholes that include instructions that set the condition codes, even
! 4991: in cases where some reloading is inevitable. For examples, search for
! 4992: @samp{addcc} and @samp{andcc} in @file{sparc.md}.
! 4993:
1.1 root 4994: @item (pc)
4995: This represents the machine's program counter. It has no operands and
4996: may not have a machine mode. @code{(pc)} may be validly used only in
4997: certain specific contexts in jump instructions.
4998:
1.1.1.8 root 4999: There is only one expression object of code @code{pc}; it is the value
1.1 root 5000: of the variable @code{pc_rtx}. Any attempt to create an expression of
1.1.1.8 root 5001: code @code{pc} will return @code{pc_rtx}.
1.1 root 5002:
5003: All instructions that do not jump alter the program counter implicitly
5004: by incrementing it, but there is no need to mention this in the RTL.
5005:
5006: @item (mem:@var{m} @var{addr})
5007: This RTX represents a reference to main memory at an address
5008: represented by the expression @var{addr}. @var{m} specifies how large
5009: a unit of memory is accessed.
5010: @end table
5011:
5012: @node Arithmetic, Comparisons, Regs and Memory, RTL
5013: @section RTL Expressions for Arithmetic
5014:
5015: @table @code
5016: @item (plus:@var{m} @var{x} @var{y})
5017: Represents the sum of the values represented by @var{x} and @var{y}
5018: carried out in machine mode @var{m}. This is valid only if
5019: @var{x} and @var{y} both are valid for mode @var{m}.
5020:
5021: @item (minus:@var{m} @var{x} @var{y})
1.1.1.8 root 5022: Like @code{plus} but represents subtraction.
1.1 root 5023:
1.1.1.6 root 5024: @item (compare @var{x} @var{y})
1.1 root 5025: Represents the result of subtracting @var{y} from @var{x}
5026: for purposes of comparison. The absence of a machine mode
1.1.1.8 root 5027: in the @code{compare} expression indicates that the result is
1.1 root 5028: computed without overflow, as if with infinite precision.
5029:
5030: Of course, machines can't really subtract with infinite precision.
5031: However, they can pretend to do so when only the sign of the
5032: result will be used, which is the case when the result is stored
5033: in @code{(cc0)}. And that is the only way this kind of expression
5034: may validly be used: as a value to be stored in the condition codes.
5035:
5036: @item (neg:@var{m} @var{x})
5037: Represents the negation (subtraction from zero) of the value
5038: represented by @var{x}, carried out in mode @var{m}. @var{x} must be
5039: valid for mode @var{m}.
5040:
5041: @item (mult:@var{m} @var{x} @var{y})
5042: Represents the signed product of the values represented by @var{x} and
5043: @var{y} carried out in machine mode @var{m}. If
5044: @var{x} and @var{y} are both valid for mode @var{m}, this is ordinary
5045: size-preserving multiplication. Alternatively, both @var{x} and @var{y}
5046: may be valid for a different, narrower mode. This represents the
5047: kind of multiplication that generates a product wider than the operands.
5048: Widening multiplication and same-size multiplication are completely
5049: distinct and supported by different machine instructions; machines may
5050: support one but not the other.@refill
5051:
1.1.1.8 root 5052: @code{mult} may be used for floating point multiplication as well.
1.1 root 5053: Then @var{m} is a floating point machine mode.
5054:
5055: @item (umult:@var{m} @var{x} @var{y})
1.1.1.8 root 5056: Like @code{mult} but represents unsigned multiplication. It may be
5057: used in both same-size and widening forms, like @code{mult}.
5058: @code{umult} is used only for fixed-point multiplication.
1.1 root 5059:
5060: @item (div:@var{m} @var{x} @var{y})
5061: Represents the quotient in signed division of @var{x} by @var{y},
5062: carried out in machine mode @var{m}. If @var{m} is a floating-point
5063: mode, it represents the exact quotient; otherwise, the integerized
5064: quotient. If @var{x} and @var{y} are both valid for mode @var{m},
5065: this is ordinary size-preserving division. Some machines have
5066: division instructions in which the operands and quotient widths are
1.1.1.8 root 5067: not all the same; such instructions are represented by @code{div}
1.1 root 5068: expressions in which the machine modes are not all the same.
5069:
5070: @item (udiv:@var{m} @var{x} @var{y})
1.1.1.8 root 5071: Like @code{div} but represents unsigned division.
1.1 root 5072:
5073: @item (mod:@var{m} @var{x} @var{y})
5074: @itemx (umod:@var{m} @var{x} @var{y})
1.1.1.8 root 5075: Like @code{div} and @code{udiv} but represent the remainder instead of
1.1 root 5076: the quotient.
5077:
5078: @item (not:@var{m} @var{x})
5079: Represents the bitwise complement of the value represented by @var{x},
5080: carried out in mode @var{m}, which must be a fixed-point machine mode.
5081: @var{x} must be valid for mode @var{m}, which must be a fixed-point mode.
5082:
5083: @item (and:@var{m} @var{x} @var{y})
5084: Represents the bitwise logical-and of the values represented by
5085: @var{x} and @var{y}, carried out in machine mode @var{m}. This is
5086: valid only if @var{x} and @var{y} both are valid for mode @var{m},
5087: which must be a fixed-point mode.
5088:
5089: @item (ior:@var{m} @var{x} @var{y})
5090: Represents the bitwise inclusive-or of the values represented by
5091: @var{x} and @var{y}, carried out in machine mode @var{m}. This is
5092: valid only if @var{x} and @var{y} both are valid for mode @var{m},
5093: which must be a fixed-point mode.
5094:
5095: @item (xor:@var{m} @var{x} @var{y})
5096: Represents the bitwise exclusive-or of the values represented by
5097: @var{x} and @var{y}, carried out in machine mode @var{m}. This is
5098: valid only if @var{x} and @var{y} both are valid for mode @var{m},
5099: which must be a fixed-point mode.
5100:
5101: @item (lshift:@var{m} @var{x} @var{c})
5102: Represents the result of logically shifting @var{x} left by @var{c}
5103: places. @var{x} must be valid for the mode @var{m}, a fixed-point
5104: machine mode. @var{c} must be valid for a fixed-point mode;
5105: which mode is determined by the mode called for in the machine
5106: description entry for the left-shift instruction. For example,
5107: on the Vax, the mode of @var{c} is @code{QImode} regardless of @var{m}.
5108:
5109: On some machines, negative values of @var{c} may be meaningful; this
5110: is why logical left shift and arithmetic left shift are distinguished.
5111: For example, Vaxes have no right-shift instructions, and right shifts
5112: are represented as left-shift instructions whose counts happen
5113: to be negative constants or else computed (in a previous instruction)
5114: by negation.
5115:
5116: @item (ashift:@var{m} @var{x} @var{c})
1.1.1.8 root 5117: Like @code{lshift} but for arithmetic left shift.
1.1 root 5118:
5119: @item (lshiftrt:@var{m} @var{x} @var{c})
5120: @itemx (ashiftrt:@var{m} @var{x} @var{c})
1.1.1.8 root 5121: Like @code{lshift} and @code{ashift} but for right shift.
1.1 root 5122:
5123: @item (rotate:@var{m} @var{x} @var{c})
5124: @itemx (rotatert:@var{m} @var{x} @var{c})
5125: Similar but represent left and right rotate.
5126:
5127: @item (abs:@var{m} @var{x})
5128: Represents the absolute value of @var{x}, computed in mode @var{m}.
5129: @var{x} must be valid for @var{m}.
5130:
5131: @item (sqrt:@var{m} @var{x})
5132: Represents the square root of @var{x}, computed in mode @var{m}.
5133: @var{x} must be valid for @var{m}. Most often @var{m} will be
5134: a floating point mode.
5135:
5136: @item (ffs:@var{m} @var{x})
1.1.1.10! root 5137: Represents one plus the index of the least significant 1-bit in
1.1 root 5138: @var{x}, represented as an integer of mode @var{m}. (The value is
5139: zero if @var{x} is zero.) The mode of @var{x} need not be @var{m};
5140: depending on the target machine, various mode combinations may be
5141: valid.
5142: @end table
5143:
5144: @node Comparisons, Bit Fields, Arithmetic, RTL
5145: @section Comparison Operations
5146:
1.1.1.10! root 5147: Comparison operators test a relation on two operands and are considered
! 5148: to represent a machine-dependent nonzero value (@code{STORE_FLAG_VALUE})
! 5149: if the relation holds, or zero if it does not. The mode of the
! 5150: comparison is determined by the operands; they must both be valid for a
! 5151: common machine mode. A comparison with both operands constant would be
! 5152: invalid as the machine mode could not be deduced from it, but such a
! 5153: comparison should never exist in RTL due to constant folding.
1.1 root 5154:
5155: Inequality comparisons come in two flavors, signed and unsigned. Thus,
1.1.1.8 root 5156: there are distinct expression codes @code{gt} and @code{gtu} for signed and
1.1 root 5157: unsigned greater-than. These can produce different results for the same
5158: pair of integer values: for example, 1 is signed greater-than -1 but not
5159: unsigned greater-than, because -1 when regarded as unsigned is actually
5160: @code{0xffffffff} which is greater than 1.
5161:
5162: The signed comparisons are also used for floating point values. Floating
5163: point comparisons are distinguished by the machine modes of the operands.
5164:
5165: The comparison operators may be used to compare the condition codes
5166: @code{(cc0)} against zero, as in @code{(eq (cc0) (const_int 0))}. Such a
5167: construct actually refers to the result of the preceding instruction in
5168: which the condition codes were set. The above example stands for 1 if the
5169: condition codes were set to say ``zero'' or ``equal'', 0 otherwise.
5170: Although the same comparison operators are used for this as may be used in
5171: other contexts on actual data, no confusion can result since the machine
5172: description would never allow both kinds of uses in the same context.
5173:
5174: @table @code
5175: @item (eq @var{x} @var{y})
5176: 1 if the values represented by @var{x} and @var{y} are equal,
5177: otherwise 0.
5178:
5179: @item (ne @var{x} @var{y})
5180: 1 if the values represented by @var{x} and @var{y} are not equal,
5181: otherwise 0.
5182:
5183: @item (gt @var{x} @var{y})
5184: 1 if the @var{x} is greater than @var{y}. If they are fixed-point,
5185: the comparison is done in a signed sense.
5186:
5187: @item (gtu @var{x} @var{y})
1.1.1.8 root 5188: Like @code{gt} but does unsigned comparison, on fixed-point numbers only.
1.1 root 5189:
5190: @item (lt @var{x} @var{y})
5191: @item (ltu @var{x} @var{y})
1.1.1.8 root 5192: Like @code{gt} and @code{gtu} but test for ``less than''.
1.1 root 5193:
5194: @item (ge @var{x} @var{y})
5195: @item (geu @var{x} @var{y})
1.1.1.8 root 5196: Like @code{gt} and @code{gtu} but test for ``greater than or equal''.
1.1 root 5197:
5198: @item (le @var{x} @var{y})
5199: @item (leu @var{x} @var{y})
1.1.1.8 root 5200: Like @code{gt} and @code{gtu} but test for ``less than or equal''.
1.1 root 5201:
5202: @item (if_then_else @var{cond} @var{then} @var{else})
5203: This is not a comparison operation but is listed here because it is
5204: always used in conjunction with a comparison operation. To be
5205: precise, @var{cond} is a comparison expression. This expression
5206: represents a choice, according to @var{cond}, between the value
5207: represented by @var{then} and the one represented by @var{else}.
5208:
1.1.1.8 root 5209: On most machines, @code{if_then_else} expressions are valid only
1.1 root 5210: to express conditional jumps.
5211: @end table
5212:
5213: @node Bit Fields, Conversions, Comparisons, RTL
5214: @section Bit-fields
5215:
5216: Special expression codes exist to represent bit-field instructions.
5217: These types of expressions are lvalues in RTL; they may appear
1.1.1.10! root 5218: on the left side of an assignment, indicating insertion of a value
1.1 root 5219: into the specified bit field.
5220:
5221: @table @code
5222: @item (sign_extract:SI @var{loc} @var{size} @var{pos})
5223: This represents a reference to a sign-extended bit-field contained or
5224: starting in @var{loc} (a memory or register reference). The bit field
5225: is @var{size} bits wide and starts at bit @var{pos}. The compilation
5226: option @code{BITS_BIG_ENDIAN} says which end of the memory unit
5227: @var{pos} counts from.
5228:
5229: Which machine modes are valid for @var{loc} depends on the machine,
5230: but typically @var{loc} should be a single byte when in memory
5231: or a full word in a register.
5232:
5233: @item (zero_extract:SI @var{loc} @var{size} @var{pos})
1.1.1.8 root 5234: Like @code{sign_extract} but refers to an unsigned or zero-extended
1.1 root 5235: bit field. The same sequence of bits are extracted, but they
5236: are filled to an entire word with zeros instead of by sign-extension.
5237: @end table
5238:
5239: @node Conversions, RTL Declarations, Bit Fields, RTL
5240: @section Conversions
5241:
5242: All conversions between machine modes must be represented by
5243: explicit conversion operations. For example, an expression
5244: which is the sum of a byte and a full word cannot be written as
1.1.1.8 root 5245: @code{(plus:SI (reg:QI 34) (reg:SI 80))} because the @code{plus}
1.1 root 5246: operation requires two operands of the same machine mode.
5247: Therefore, the byte-sized operand is enclosed in a conversion
5248: operation, as in
5249:
5250: @example
5251: (plus:SI (sign_extend:SI (reg:QI 34)) (reg:SI 80))
5252: @end example
5253:
5254: The conversion operation is not a mere placeholder, because there
5255: may be more than one way of converting from a given starting mode
5256: to the desired final mode. The conversion operation code says how
5257: to do it.
5258:
5259: @table @code
5260: @item (sign_extend:@var{m} @var{x})
5261: Represents the result of sign-extending the value @var{x}
5262: to machine mode @var{m}. @var{m} must be a fixed-point mode
5263: and @var{x} a fixed-point value of a mode narrower than @var{m}.
5264:
5265: @item (zero_extend:@var{m} @var{x})
5266: Represents the result of zero-extending the value @var{x}
5267: to machine mode @var{m}. @var{m} must be a fixed-point mode
5268: and @var{x} a fixed-point value of a mode narrower than @var{m}.
5269:
5270: @item (float_extend:@var{m} @var{x})
5271: Represents the result of extending the value @var{x}
5272: to machine mode @var{m}. @var{m} must be a floating point mode
5273: and @var{x} a floating point value of a mode narrower than @var{m}.
5274:
5275: @item (truncate:@var{m} @var{x})
5276: Represents the result of truncating the value @var{x}
5277: to machine mode @var{m}. @var{m} must be a fixed-point mode
5278: and @var{x} a fixed-point value of a mode wider than @var{m}.
5279:
5280: @item (float_truncate:@var{m} @var{x})
5281: Represents the result of truncating the value @var{x}
5282: to machine mode @var{m}. @var{m} must be a floating point mode
5283: and @var{x} a floating point value of a mode wider than @var{m}.
5284:
5285: @item (float:@var{m} @var{x})
5286: Represents the result of converting fixed point value @var{x},
5287: regarded as signed, to floating point mode @var{m}.
5288:
5289: @item (unsigned_float:@var{m} @var{x})
5290: Represents the result of converting fixed point value @var{x},
5291: regarded as unsigned, to floating point mode @var{m}.
5292:
5293: @item (fix:@var{m} @var{x})
5294: When @var{m} is a fixed point mode, represents the result of
5295: converting floating point value @var{x} to mode @var{m}, regarded as
5296: signed. How rounding is done is not specified, so this operation may
5297: be used validly in compiling C code only for integer-valued operands.
5298:
5299: @item (unsigned_fix:@var{m} @var{x})
5300: Represents the result of converting floating point value @var{x} to
5301: fixed point mode @var{m}, regarded as unsigned. How rounding is done
5302: is not specified.
5303:
5304: @item (fix:@var{m} @var{x})
5305: When @var{m} is a floating point mode, represents the result of
5306: converting floating point value @var{x} (valid for mode @var{m}) to an
5307: integer, still represented in floating point mode @var{m}, by rounding
5308: towards zero.
5309: @end table
5310:
5311: @node RTL Declarations, Side Effects, Conversions, RTL
5312: @section Declarations
5313:
5314: Declaration expression codes do not represent arithmetic operations
5315: but rather state assertions about their operands.
5316:
5317: @table @code
5318: @item (strict_low_part (subreg:@var{m} (reg:@var{n} @var{r}) 0))
5319: This expression code is used in only one context: operand 0 of a
1.1.1.8 root 5320: @code{set} expression. In addition, the operand of this expression
5321: must be a @code{subreg} expression.
1.1 root 5322:
1.1.1.8 root 5323: The presence of @code{strict_low_part} says that the part of the
1.1 root 5324: register which is meaningful in mode @var{n}, but is not part of
5325: mode @var{m}, is not to be altered. Normally, an assignment to such
5326: a subreg is allowed to have undefined effects on the rest of the
5327: register when @var{m} is less than a word.
5328: @end table
5329:
5330: @node Side Effects, Incdec, RTL Declarations, RTL
5331: @section Side Effect Expressions
5332:
5333: The expression codes described so far represent values, not actions.
5334: But machine instructions never produce values; they are meaningful
5335: only for their side effects on the state of the machine. Special
5336: expression codes are used to represent side effects.
5337:
5338: The body of an instruction is always one of these side effect codes;
5339: the codes described above, which represent values, appear only as
5340: the operands of these.
5341:
5342: @table @code
5343: @item (set @var{lval} @var{x})
5344: Represents the action of storing the value of @var{x} into the place
5345: represented by @var{lval}. @var{lval} must be an expression
1.1.1.8 root 5346: representing a place that can be stored in: @code{reg} (or
5347: @code{subreg} or @code{strict_low_part}), @code{mem}, @code{pc} or
5348: @code{cc0}.@refill
1.1 root 5349:
1.1.1.8 root 5350: If @var{lval} is a @code{reg}, @code{subreg} or @code{mem}, it has a
1.1 root 5351: machine mode; then @var{x} must be valid for that mode.@refill
5352:
1.1.1.8 root 5353: If @var{lval} is a @code{reg} whose machine mode is less than the full
1.1 root 5354: width of the register, then it means that the part of the register
5355: specified by the machine mode is given the specified value and the
5356: rest of the register receives an undefined value. Likewise, if
1.1.1.8 root 5357: @var{lval} is a @code{subreg} whose machine mode is narrower than
1.1 root 5358: @code{SImode}, the rest of the register can be changed in an undefined way.
5359:
1.1.1.8 root 5360: If @var{lval} is a @code{strict_low_part} of a @code{subreg}, then the
1.1 root 5361: part of the register specified by the machine mode of the
1.1.1.8 root 5362: @code{subreg} is given the value @var{x} and the rest of the register
1.1 root 5363: is not changed.@refill
5364:
5365: If @var{lval} is @code{(cc0)}, it has no machine mode, and @var{x} may
5366: have any mode. This represents a ``test'' or ``compare'' instruction.@refill
5367:
5368: If @var{lval} is @code{(pc)}, we have a jump instruction, and the
5369: possibilities for @var{x} are very limited. It may be a
1.1.1.8 root 5370: @code{label_ref} expression (unconditional jump). It may be an
5371: @code{if_then_else} (conditional jump), in which case either the
1.1 root 5372: second or the third operand must be @code{(pc)} (for the case which
1.1.1.8 root 5373: does not jump) and the other of the two must be a @code{label_ref}
5374: (for the case which does jump). @var{x} may also be a @code{mem} or
5375: @code{(plus:SI (pc) @var{y})}, where @var{y} may be a @code{reg} or a
5376: @code{mem}; these unusual patterns are used to represent jumps through
1.1 root 5377: branch tables.@refill
5378:
5379: @item (return)
5380: Represents a return from the current function, on machines where this
5381: can be done with one instruction, such as Vaxes. On machines where a
5382: multi-instruction ``epilogue'' must be executed in order to return
5383: from the function, returning is done by jumping to a label which
1.1.1.8 root 5384: precedes the epilogue, and the @code{return} expression code is never
1.1 root 5385: used.
5386:
5387: @item (call @var{function} @var{nargs})
1.1.1.8 root 5388: Represents a function call. @var{function} is a @code{mem} expression
1.1 root 5389: whose address is the address of the function to be called.
5390: @var{nargs} is an expression which can be used for two purposes: on
5391: some machines it represents the number of bytes of stack argument; on
5392: others, it represents the number of argument registers.
5393:
5394: Each machine has a standard machine mode which @var{function} must
5395: have. The machine description defines macro @code{FUNCTION_MODE} to
5396: expand into the requisite mode name. The purpose of this mode is to
5397: specify what kind of addressing is allowed, on machines where the
5398: allowed kinds of addressing depend on the machine mode being
5399: addressed.
5400:
5401: @item (clobber @var{x})
5402: Represents the storing or possible storing of an unpredictable,
1.1.1.8 root 5403: undescribed value into @var{x}, which must be a @code{reg} or
5404: @code{mem} expression.
1.1 root 5405:
5406: One place this is used is in string instructions that store standard
5407: values into particular hard registers. It may not be worth the
5408: trouble to describe the values that are stored, but it is essential to
5409: inform the compiler that the registers will be altered, lest it
5410: attempt to keep data in them across the string instruction.
5411:
5412: @var{x} may also be null---a null C pointer, no expression at all.
5413: Such a @code{(clobber (null))} expression means that all memory
5414: locations must be presumed clobbered.
5415:
5416: Note that the machine description classifies certain hard registers as
5417: ``call-clobbered''. All function call instructions are assumed by
5418: default to clobber these registers, so there is no need to use
1.1.1.8 root 5419: @code{clobber} expressions to indicate this fact. Also, each function
1.1.1.6 root 5420: call is assumed to have the potential to alter any memory location,
5421: unless the function is declared @code{const}.
1.1 root 5422:
1.1.1.8 root 5423: When a @code{clobber} expression for a register appears inside a
5424: @code{parallel} with other side effects, GNU CC guarantees that the
1.1.1.4 root 5425: register is unoccupied both before and after that insn. Therefore, it
5426: is safe for the assembler code produced by the insn to use the
5427: register as a temporary. You can clobber either a specific hard
5428: register or a pseudo register; in the latter case, GNU CC will
5429: allocate a hard register that is available there for use as a
5430: temporary.
5431:
1.1.1.8 root 5432: If you clobber a pseudo register in this way, use a pseudo register
5433: which appears nowhere else---generate a new one each time. Otherwise,
5434: you may confuse CSE.
5435:
5436: There is one other known use for clobbering a pseudo register in a
5437: @code{parallel}: when one of the input operands of the insn is also
5438: clobbered by the insn. In this case, using the same pseudo register in
5439: the clobber and elsewhere in the insn produces the expected results.
5440:
1.1 root 5441: @item (use @var{x})
5442: Represents the use of the value of @var{x}. It indicates that the
5443: value in @var{x} at this point in the program is needed, even though
5444: it may not be apparent why this is so. Therefore, the compiler will
1.1.1.4 root 5445: not attempt to delete previous instructions whose only effect is to
1.1.1.8 root 5446: store a value in @var{x}. @var{x} must be a @code{reg} expression.
1.1 root 5447:
5448: @item (parallel [@var{x0} @var{x1} @dots{}])
5449: Represents several side effects performed in parallel. The square
1.1.1.8 root 5450: brackets stand for a vector; the operand of @code{parallel} is a
1.1 root 5451: vector of expressions. @var{x0}, @var{x1} and so on are individual
1.1.1.8 root 5452: side effect expressions---expressions of code @code{set}, @code{call},
5453: @code{return}, @code{clobber} or @code{use}.@refill
1.1 root 5454:
5455: ``In parallel'' means that first all the values used in the individual
5456: side-effects are computed, and second all the actual side-effects are
5457: performed. For example,
5458:
5459: @example
5460: (parallel [(set (reg:SI 1) (mem:SI (reg:SI 1)))
5461: (set (mem:SI (reg:SI 1)) (reg:SI 1))])
5462: @end example
5463:
5464: @noindent
5465: says unambiguously that the values of hard register 1 and the memory
5466: location addressed by it are interchanged. In both places where
5467: @code{(reg:SI 1)} appears as a memory address it refers to the value
1.1.1.4 root 5468: in register 1 @emph{before} the execution of the insn.
5469:
1.1.1.8 root 5470: It follows that it is @emph{incorrect} to use @code{parallel} and
5471: expect the result of one @code{set} to be available for the next one.
1.1.1.4 root 5472: For example, people sometimes attempt to represent a jump-if-zero
5473: instruction this way:
5474:
5475: @example
5476: (parallel [(set (cc0) (reg:SI 34))
1.1.1.9 root 5477: (set (pc) (if_then_else
5478: (eq (cc0) (const_int 0))
5479: (label_ref @dots{})
5480: (pc)))])
1.1.1.4 root 5481: @end example
5482:
5483: @noindent
5484: But this is incorrect, because it says that the jump condition depends
5485: on the condition code value @emph{before} this instruction, not on the
5486: new value that is set by this instruction.
1.1 root 5487:
1.1.1.5 root 5488: Peephole optimization, which takes place in together with final assembly
1.1.1.8 root 5489: code output, can produce insns whose patterns consist of a @code{parallel}
1.1 root 5490: whose elements are the operands needed to output the resulting
1.1.1.8 root 5491: assembler code---often @code{reg}, @code{mem} or constant expressions.
1.1 root 5492: This would not be well-formed RTL at any other stage in compilation,
5493: but it is ok then because no further optimization remains to be done.
1.1.1.4 root 5494: However, the definition of the macro @code{NOTICE_UPDATE_CC} must
5495: deal with such insns if you define any peephole optimizations.
1.1 root 5496:
5497: @item (sequence [@var{insns} @dots{}])
5498: Represents a sequence of insns. Each of the @var{insns} that appears
5499: in the vector is suitable for appearing in the chain of insns, so it
1.1.1.8 root 5500: must be an @code{insn}, @code{jump_insn}, @code{call_insn},
5501: @code{code_label}, @code{barrier} or @code{note}.
1.1 root 5502:
1.1.1.8 root 5503: A @code{sequence} RTX never appears in an actual insn. It represents
5504: the sequence of insns that result from a @code{define_expand}
1.1 root 5505: @emph{before} those insns are passed to @code{emit_insn} to insert
5506: them in the chain of insns. When actually inserted, the individual
1.1.1.8 root 5507: sub-insns are separated out and the @code{sequence} is forgotten.
1.1 root 5508: @end table
5509:
5510: Three expression codes appear in place of a side effect, as the body of an
5511: insn, though strictly speaking they do not describe side effects as such:
5512:
5513: @table @code
5514: @item (asm_input @var{s})
5515: Represents literal assembler code as described by the string @var{s}.
5516:
5517: @item (addr_vec:@var{m} [@var{lr0} @var{lr1} @dots{}])
5518: Represents a table of jump addresses. The vector elements @var{lr0},
1.1.1.8 root 5519: etc., are @code{label_ref} expressions. The mode @var{m} specifies
1.1 root 5520: how much space is given to each address; normally @var{m} would be
5521: @code{Pmode}.
5522:
5523: @item (addr_diff_vec:@var{m} @var{base} [@var{lr0} @var{lr1} @dots{}])
5524: Represents a table of jump addresses expressed as offsets from
1.1.1.8 root 5525: @var{base}. The vector elements @var{lr0}, etc., are @code{label_ref}
1.1 root 5526: expressions and so is @var{base}. The mode @var{m} specifies how much
5527: space is given to each address-difference.@refill
5528: @end table
5529:
5530: @node Incdec, Assembler, Side Effects, RTL
5531: @section Embedded Side-Effects on Addresses
5532:
5533: Four special side-effect expression codes appear as memory addresses.
5534:
5535: @table @code
5536: @item (pre_dec:@var{m} @var{x})
5537: Represents the side effect of decrementing @var{x} by a standard
5538: amount and represents also the value that @var{x} has after being
1.1.1.8 root 5539: decremented. @var{x} must be a @code{reg} or @code{mem}, but most
5540: machines allow only a @code{reg}. @var{m} must be the machine mode
1.1 root 5541: for pointers on the machine in use. The amount @var{x} is decremented
5542: by is the length in bytes of the machine mode of the containing memory
5543: reference of which this expression serves as the address. Here is an
5544: example of its use:@refill
5545:
5546: @example
5547: (mem:DF (pre_dec:SI (reg:SI 39)))
5548: @end example
5549:
5550: @noindent
5551: This says to decrement pseudo register 39 by the length of a @code{DFmode}
5552: value and use the result to address a @code{DFmode} value.
5553:
5554: @item (pre_inc:@var{m} @var{x})
5555: Similar, but specifies incrementing @var{x} instead of decrementing it.
5556:
5557: @item (post_dec:@var{m} @var{x})
1.1.1.8 root 5558: Represents the same side effect as @code{pre_dec} but a different
1.1 root 5559: value. The value represented here is the value @var{x} has @i{before}
5560: being decremented.
5561:
5562: @item (post_inc:@var{m} @var{x})
5563: Similar, but specifies incrementing @var{x} instead of decrementing it.
5564: @end table
5565:
5566: These embedded side effect expressions must be used with care. Instruction
5567: patterns may not use them. Until the @samp{flow} pass of the compiler,
5568: they may occur only to represent pushes onto the stack. The @samp{flow}
5569: pass finds cases where registers are incremented or decremented in one
5570: instruction and used as an address shortly before or after; these cases are
5571: then transformed to use pre- or post-increment or -decrement.
5572:
5573: Explicit popping of the stack could be represented with these embedded
5574: side effect operators, but that would not be safe; the instruction
5575: combination pass could move the popping past pushes, thus changing
5576: the meaning of the code.
5577:
5578: An instruction that can be represented with an embedded side effect
1.1.1.8 root 5579: could also be represented using @code{parallel} containing an additional
5580: @code{set} to describe how the address register is altered. This is not
1.1 root 5581: done because machines that allow these operations at all typically
5582: allow them wherever a memory address is called for. Describing them as
5583: additional parallel stores would require doubling the number of entries
5584: in the machine description.
5585:
5586: @node Assembler, Insns, IncDec, RTL
5587: @section Assembler Instructions as Expressions
5588:
1.1.1.8 root 5589: The RTX code @code{asm_operands} represents a value produced by a
1.1 root 5590: user-specified assembler instruction. It is used to represent
5591: an @code{asm} statement with arguments. An @code{asm} statement with
5592: a single output operand, like this:
5593:
5594: @example
1.1.1.6 root 5595: asm ("foo %1,%2,%0" : "=a" (outputvar) : "g" (x + y), "di" (*z));
1.1 root 5596: @end example
5597:
5598: @noindent
1.1.1.8 root 5599: is represented using a single @code{asm_operands} RTX which represents
1.1 root 5600: the value that is stored in @code{outputvar}:
5601:
5602: @example
5603: (set @var{rtx-for-outputvar}
5604: (asm_operands "foo %1,%2,%0" "a" 0
5605: [@var{rtx-for-addition-result} @var{rtx-for-*z}]
5606: [(asm_input:@var{m1} "g")
5607: (asm_input:@var{m2} "di")]))
5608: @end example
5609:
5610: @noindent
1.1.1.8 root 5611: Here the operands of the @code{asm_operands} RTX are the assembler
1.1 root 5612: template string, the output-operand's constraint, the index-number of the
5613: output operand among the output operands specified, a vector of input
5614: operand RTX's, and a vector of input-operand modes and constraints. The
5615: mode @var{m1} is the mode of the sum @code{x+y}; @var{m2} is that of
5616: @code{*z}.
5617:
5618: When an @code{asm} statement has multiple output values, its insn has
1.1.1.8 root 5619: several such @code{set} RTX's inside of a @code{parallel}. Each @code{set}
5620: contains a @code{asm_operands}; all of these share the same assembler
1.1 root 5621: template and vectors, but each contains the constraint for the respective
5622: output operand. They are also distinguished by the output-operand index
5623: number, which is 0, 1, @dots{} for successive output operands.
5624:
5625: @node Insns, Calls, Assembler, RTL
5626: @section Insns
5627:
5628: The RTL representation of the code for a function is a doubly-linked
5629: chain of objects called @dfn{insns}. Insns are expressions with
5630: special codes that are used for no other purpose. Some insns are
5631: actual instructions; others represent dispatch tables for @code{switch}
5632: statements; others represent labels to jump to or various sorts of
5633: declarative information.
5634:
5635: In addition to its own specific data, each insn must have a unique id-number
5636: that distinguishes it from all other insns in the current function, and
5637: chain pointers to the preceding and following insns. These three fields
5638: occupy the same position in every insn, independent of the expression code
5639: of the insn. They could be accessed with @code{XEXP} and @code{XINT},
5640: but instead three special macros are always used:
5641:
5642: @table @code
5643: @item INSN_UID (@var{i})
5644: Accesses the unique id of insn @var{i}.
5645:
5646: @item PREV_INSN (@var{i})
5647: Accesses the chain pointer to the insn preceding @var{i}.
5648: If @var{i} is the first insn, this is a null pointer.
5649:
5650: @item NEXT_INSN (@var{i})
5651: Accesses the chain pointer to the insn following @var{i}.
5652: If @var{i} is the last insn, this is a null pointer.
5653: @end table
5654:
5655: The @code{NEXT_INSN} and @code{PREV_INSN} pointers must always
1.1.1.6 root 5656: correspond: if @var{insn} is not the first insn,
1.1 root 5657:
5658: @example
5659: NEXT_INSN (PREV_INSN (@var{insn})) == @var{insn}
5660: @end example
5661:
5662: @noindent
5663: is always true.
5664:
5665: Every insn has one of the following six expression codes:
5666:
1.1.1.8 root 5667: @table @code
1.1 root 5668: @item insn
1.1.1.8 root 5669: The expression code @code{insn} is used for instructions that do not jump
5670: and do not do function calls. Insns with code @code{insn} have four
1.1 root 5671: additional fields beyond the three mandatory ones listed above.
5672: These four are described in a table below.
5673:
5674: @item jump_insn
1.1.1.8 root 5675: The expression code @code{jump_insn} is used for instructions that may jump
5676: (or, more generally, may contain @code{label_ref} expressions).
5677: @code{jump_insn} insns have the same extra fields as @code{insn} insns,
1.1.1.10! root 5678: accessed in the same way. If there is an instruction to return from the
! 5679: current function, it is recorded as a @code{jump_insn}.
1.1 root 5680:
5681: @item call_insn
1.1.1.8 root 5682: The expression code @code{call_insn} is used for instructions that may do
1.1 root 5683: function calls. It is important to distinguish these instructions because
5684: they imply that certain registers and memory locations may be altered
5685: unpredictably.
5686:
1.1.1.8 root 5687: @code{call_insn} insns have the same extra fields as @code{insn} insns,
1.1 root 5688: accessed in the same way.
5689:
5690: @item code_label
1.1.1.8 root 5691: A @code{code_label} insn represents a label that a jump insn can jump to.
1.1 root 5692: It contains one special field of data in addition to the three standard ones.
5693: It is used to hold the @dfn{label number}, a number that identifies this
5694: label uniquely among all the labels in the compilation (not just in the
5695: current function). Ultimately, the label is represented in the assembler
5696: output as an assembler label @samp{L@var{n}} where @var{n} is the label number.
5697:
5698: @item barrier
5699: Barriers are placed in the instruction stream after unconditional
5700: jump instructions to indicate that the jumps are unconditional.
5701: They contain no information beyond the three standard fields.
5702:
5703: @item note
1.1.1.8 root 5704: @code{note} insns are used to represent additional debugging and
1.1 root 5705: declarative information. They contain two nonstandard fields, an
5706: integer which is accessed with the macro @code{NOTE_LINE_NUMBER} and a
5707: string accessed with @code{NOTE_SOURCE_FILE}.
5708:
5709: If @code{NOTE_LINE_NUMBER} is positive, the note represents the
5710: position of a source line and @code{NOTE_SOURCE_FILE} is the source file name
5711: that the line came from. These notes control generation of line
5712: number data in the assembler output.
5713:
5714: Otherwise, @code{NOTE_LINE_NUMBER} is not really a line number but a
5715: code with one of the following values (and @code{NOTE_SOURCE_FILE}
5716: must contain a null pointer):
5717:
5718: @table @code
5719: @item NOTE_INSN_DELETED
5720: Such a note is completely ignorable. Some passes of the compiler
5721: delete insns by altering them into notes of this kind.
5722:
5723: @item NOTE_INSN_BLOCK_BEG
5724: @itemx NOTE_INSN_BLOCK_END
5725: These types of notes indicate the position of the beginning and end
5726: of a level of scoping of variable names. They control the output
5727: of debugging information.
5728:
5729: @item NOTE_INSN_LOOP_BEG
5730: @itemx NOTE_INSN_LOOP_END
5731: These types of notes indicate the position of the beginning and end
5732: of a @code{while} or @code{for} loop. They enable the loop optimizer
5733: to find loops quickly.
1.1.1.6 root 5734: @item NOTE_INSN_FUNCTION_END
5735: Appears near the end of the function body, just before the label that
5736: @code{return} statements jump to (on machine where a single instruction
5737: does not suffice for returning). This note may be deleted by jump
5738: optimization.
5739: @item NOTE_INSN_SETJMP
5740: Appears following each call to @code{setjmp} or a related function.
1.1.1.7 root 5741:
1.1.1.10! root 5742: @item NOTE_INSN_LOOP_CONT
1.1.1.7 root 5743: Appears at the place in a loop that @code{continue} statements jump to.
1.1 root 5744: @end table
1.1.1.7 root 5745:
5746: These codes are printed symbolically when they appear in debugging dumps.
1.1 root 5747: @end table
5748:
1.1.1.6 root 5749: The machine mode of an insn is normally zero (@code{VOIDmode}), but the
5750: reload pass sets it to @code{QImode} if the insn needs reloading.
5751:
1.1.1.8 root 5752: Here is a table of the extra fields of @code{insn}, @code{jump_insn}
5753: and @code{call_insn} insns:
1.1 root 5754:
5755: @table @code
5756: @item PATTERN (@var{i})
5757: An expression for the side effect performed by this insn.
5758:
1.1.1.6 root 5759: @item INSN_CODE (@var{i})
5760: An integer that says which pattern in the machine description matches
5761: this insn, or -1 if the matching has not yet been attempted.
5762:
5763: Such matching is never attempted and this field is not used on an insn
1.1.1.8 root 5764: whose pattern consists of a single @code{use}, @code{clobber},
5765: @code{asm}, @code{addr_vec} or @code{addr_diff_vec} expression.
1.1 root 5766:
5767: @item LOG_LINKS (@var{i})
1.1.1.8 root 5768: A list (chain of @code{insn_list} expressions) of previous ``related''
1.1 root 5769: insns: insns which store into registers values that are used for the
5770: first time in this insn. (An additional constraint is that neither a
5771: jump nor a label may come between the related insns). This list is
5772: set up by the flow analysis pass; it is a null pointer until then.
5773:
1.1.1.6 root 5774: @item REG_NOTES (@var{i})
1.1.1.8 root 5775: A list (chain of @code{expr_list} expressions) giving information
1.1.1.6 root 5776: about the usage of registers in this insn. This list is set up by the
5777: flow analysis pass; it is a null pointer until then.
1.1 root 5778: @end table
5779:
1.1.1.8 root 5780: The @code{LOG_LINKS} field of an insn is a chain of @code{insn_list}
1.1 root 5781: expressions. Each of these has two operands: the first is an insn,
1.1.1.8 root 5782: and the second is another @code{insn_list} expression (the next one in
5783: the chain). The last @code{insn_list} in the chain has a null pointer
1.1 root 5784: as second operand. The significant thing about the chain is which
1.1.1.8 root 5785: insns appear in it (as first operands of @code{insn_list}
1.1 root 5786: expressions). Their order is not significant.
5787:
5788: The @code{REG_NOTES} field of an insn is a similar chain but of
1.1.1.8 root 5789: @code{expr_list} expressions instead of @code{insn_list}. There are
1.1.1.5 root 5790: several kinds of register notes, which are distinguished by the machine
1.1.1.8 root 5791: mode of the @code{expr_list}, which in a register note is really
1.1.1.5 root 5792: understood as being an @code{enum reg_note}. The first operand @var{op}
1.1.1.8 root 5793: of the @code{expr_list} is data whose meaning depends on the kind of
1.1.1.5 root 5794: note. Here are the kinds of register note:
1.1 root 5795:
5796: @table @code
5797: @item REG_DEAD
5798: The register @var{op} dies in this insn; that is to say, altering the
5799: value immediately after this insn would not affect the future behavior
5800: of the program.
5801:
5802: @item REG_INC
5803: The register @var{op} is incremented (or decremented; at this level
5804: there is no distinction) by an embedded side effect inside this insn.
1.1.1.8 root 5805: This means it appears in a @code{post_inc}, @code{pre_inc},
5806: @code{post_dec} or @code{pre_dec} RTX.
1.1 root 5807:
5808: @item REG_EQUIV
5809: The register that is set by this insn will be equal to @var{op} at run
5810: time, and could validly be replaced in all its occurrences by
5811: @var{op}. (``Validly'' here refers to the data flow of the program;
5812: simple replacement may make some insns invalid.)
5813:
5814: The value which the insn explicitly copies into the register may look
5815: different from @var{op}, but they will be equal at run time.
5816:
5817: For example, when a constant is loaded into a register that is never
5818: assigned any other value, this kind of note is used.
5819:
5820: When a parameter is copied into a pseudo-register at entry to a function,
5821: a note of this kind records that the register is equivalent to the stack
5822: slot where the parameter was passed. Although in this case the register
5823: may be set by other insns, it is still valid to replace the register
5824: by the stack slot throughout the function.
5825:
5826: @item REG_EQUAL
5827: The register that is set by this insn will be equal to @var{op} at run
5828: time at the end of this insn (but not necessarily elsewhere in the
5829: function).
5830:
5831: The RTX @var{op} is typically an arithmetic expression. For example,
5832: when a sequence of insns such as a library call is used to perform an
5833: arithmetic operation, this kind of note is attached to the insn that
5834: produces or copies the final value. It tells the CSE pass how to
5835: think of that value.
5836:
5837: @item REG_RETVAL
5838: This insn copies the value of a library call, and @var{op} is the
5839: first insn that was generated to set up the arguments for the library
5840: call.
5841:
5842: Flow analysis uses this note to delete all of a library call whose
5843: result is dead.
5844:
5845: @item REG_WAS_0
5846: The register @var{op} contained zero before this insn. You can rely
5847: on this note if it is present; its absence implies nothing.
5848:
5849: @item REG_LIBCALL
5850: This is the inverse of @code{REG_RETVAL}: it is placed on the first
5851: insn of a library call, and it points to the last one.
5852:
5853: Loop optimization uses this note to move an entire library call out
5854: of a loop when its value is constant.
5855:
5856: @item REG_NONNEG
5857: The register @var{op} is known to have nonnegative value when this
5858: insn is reached.
5859: @end table
5860:
1.1.1.8 root 5861: For convenience, the machine mode in an @code{insn_list} or
5862: @code{expr_list} is printed using these symbolic codes in debugging dumps.
1.1.1.7 root 5863:
1.1.1.8 root 5864: The only difference between the expression codes @code{insn_list} and
5865: @code{expr_list} is that the first operand of an @code{insn_list} is
1.1 root 5866: assumed to be an insn and is printed in debugging dumps as the insn's
1.1.1.8 root 5867: unique id; the first operand of an @code{expr_list} is printed in the
1.1.1.7 root 5868: ordinary way as an expression.
1.1 root 5869:
5870: @node Calls, Sharing, Insns, RTL
5871: @section RTL Representation of Function-Call Insns
5872:
1.1.1.8 root 5873: Insns that call subroutines have the RTL expression code @code{call_insn}.
1.1 root 5874: These insns must satisfy special rules, and their bodies must use a special
1.1.1.8 root 5875: RTL expression code, @code{call}.
1.1 root 5876:
1.1.1.8 root 5877: A @code{call} expression has two operands, as follows:
1.1 root 5878:
5879: @example
1.1.1.6 root 5880: (call (mem:@var{fm} @var{addr}) @var{nbytes})
1.1 root 5881: @end example
5882:
5883: @noindent
5884: Here @var{nbytes} is an operand that represents the number of bytes of
5885: argument data being passed to the subroutine, @var{fm} is a machine mode
5886: (which must equal as the definition of the @code{FUNCTION_MODE} macro in
5887: the machine description) and @var{addr} represents the address of the
5888: subroutine.
5889:
1.1.1.8 root 5890: For a subroutine that returns no value, the @code{call} RTX as shown above
1.1 root 5891: is the entire body of the insn.
5892:
5893: For a subroutine that returns a value whose mode is not @code{BLKmode},
5894: the value is returned in a hard register. If this register's number is
5895: @var{r}, then the body of the call insn looks like this:
5896:
5897: @example
5898: (set (reg:@var{m} @var{r})
1.1.1.9 root 5899: (call (mem:@var{fm} @var{addr}) @var{nbytes}))
1.1 root 5900: @end example
5901:
5902: @noindent
5903: This RTL expression makes it clear (to the optimizer passes) that the
5904: appropriate register receives a useful value in this insn.
5905:
5906: Immediately after RTL generation, if the value of the subroutine is
5907: actually used, this call insn is always followed closely by an insn which
5908: refers to the register @var{r}. This remains true through all the
5909: optimizer passes until cross jumping occurs.
5910:
5911: The following insn has one of two forms. Either it copies the value into a
5912: pseudo-register, like this:
5913:
5914: @example
5915: (set (reg:@var{m} @var{p}) (reg:@var{m} @var{r}))
5916: @end example
5917:
5918: @noindent
5919: or (in the case where the calling function will simply return whatever
5920: value the call produced, and no operation is needed to do this):
5921:
5922: @example
5923: (use (reg:@var{m} @var{r}))
5924: @end example
5925:
5926: @noindent
5927: Between the call insn and this following insn there may intervene only a
1.1.1.8 root 5928: stack-adjustment insn (and perhaps some @code{note} insns).
1.1 root 5929:
5930: When a subroutine returns a @code{BLKmode} value, it is handled by
5931: passing to the subroutine the address of a place to store the value.
5932: So the call insn itself does not ``return'' any value, and it has the
5933: same RTL form as a call that returns nothing.
5934:
5935: @node Sharing,, Calls, RTL
5936: @section Structure Sharing Assumptions
5937:
5938: The compiler assumes that certain kinds of RTL expressions are unique;
5939: there do not exist two distinct objects representing the same value.
5940: In other cases, it makes an opposite assumption: that no RTL expression
5941: object of a certain kind appears in more than one place in the
5942: containing structure.
5943:
5944: These assumptions refer to a single function; except for the RTL
5945: objects that describe global variables and external functions,
5946: no RTL objects are common to two functions.
5947:
5948: @itemize @bullet
5949: @item
1.1.1.8 root 5950: Each pseudo-register has only a single @code{reg} object to represent it,
1.1 root 5951: and therefore only a single machine mode.
5952:
5953: @item
1.1.1.8 root 5954: For any symbolic label, there is only one @code{symbol_ref} object
1.1 root 5955: referring to it.
5956:
5957: @item
1.1.1.8 root 5958: There is only one @code{const_int} expression with value zero,
1.1 root 5959: and only one with value one.
5960:
5961: @item
1.1.1.8 root 5962: There is only one @code{pc} expression.
1.1 root 5963:
5964: @item
1.1.1.8 root 5965: There is only one @code{cc0} expression.
1.1 root 5966:
5967: @item
1.1.1.8 root 5968: There is only one @code{const_double} expression with mode
1.1 root 5969: @code{SFmode} and value zero, and only one with mode @code{DFmode} and
5970: value zero.
5971:
5972: @item
1.1.1.8 root 5973: No @code{label_ref} appears in more than one place in the RTL
1.1 root 5974: structure; in other words, it is safe to do a tree-walk of all the
1.1.1.8 root 5975: insns in the function and assume that each time a @code{label_ref} is
1.1 root 5976: seen it is distinct from all others that are seen.
5977:
5978: @item
1.1.1.8 root 5979: Only one @code{mem} object is normally created for each static
1.1 root 5980: variable or stack slot, so these objects are frequently shared in all
5981: the places they appear. However, separate but equal objects for these
5982: variables are occasionally made.
5983:
5984: @item
1.1.1.5 root 5985: When a single @code{asm} statement has multiple output operands,
5986: a distinct @code{asm_operands} RTX is made for each output operand.
5987: However, these all share the vector which contains the sequence of
5988: input operands. Because this sharing is used later on to test whether
5989: two @code{asm_operands} RTX's come from the same statement, the sharing
5990: must be guaranteed to be preserved.
5991:
5992: @item
1.1 root 5993: No RTL object appears in more than one place in the RTL structure
5994: except as described above. Many passes of the compiler rely on this
5995: by assuming that they can modify RTL objects in place without unwanted
5996: side-effects on other insns.
5997:
5998: @item
5999: During initial RTL generation, shared structure is freely introduced.
6000: After all the RTL for a function has been generated, all shared
6001: structure is copied by @code{unshare_all_rtl} in @file{emit-rtl.c},
6002: after which the above rules are guaranteed to be followed.
6003:
6004: @item
6005: During the combiner pass, shared structure with an insn can exist
6006: temporarily. However, the shared structure is copied before the
6007: combiner is finished with the insn. This is done by
1.1.1.8 root 6008: @code{copy_substitutions} in @file{combine.c}.
1.1 root 6009: @end itemize
6010:
6011: @node Machine Desc, Machine Macros, RTL, Top
6012: @chapter Machine Descriptions
6013:
6014: A machine description has two parts: a file of instruction patterns
6015: (@file{.md} file) and a C header file of macro definitions.
6016:
6017: The @file{.md} file for a target machine contains a pattern for each
6018: instruction that the target machine supports (or at least each instruction
6019: that is worth telling the compiler about). It may also contain comments.
6020: A semicolon causes the rest of the line to be a comment, unless the semicolon
6021: is inside a quoted string.
6022:
6023: See the next chapter for information on the C header file.
6024:
6025: @menu
6026: * Patterns:: How to write instruction patterns.
1.1.1.8 root 6027: * Example:: An explained example of a @code{define_insn} pattern.
1.1 root 6028: * RTL Template:: The RTL template defines what insns match a pattern.
6029: * Output Template:: The output template says how to make assembler code
6030: from such an insn.
6031: * Output Statement:: For more generality, write C code to output
6032: the assembler code.
6033: * Constraints:: When not all operands are general operands.
6034: * Standard Names:: Names mark patterns to use for code generation.
6035: * Pattern Ordering:: When the order of patterns makes a difference.
6036: * Dependent Patterns:: Having one pattern may make you need another.
6037: * Jump Patterns:: Special considerations for patterns for jump insns.
6038: * Peephole Definitions::Defining machine-specific peephole optimizations.
6039: * Expander Definitions::Generating a sequence of several RTL insns
6040: for a standard operation.
6041: @end menu
6042:
6043: @node Patterns, Example, Machine Desc, Machine Desc
6044: @section Everything about Instruction Patterns
6045:
6046: Each instruction pattern contains an incomplete RTL expression, with pieces
6047: to be filled in later, operand constraints that restrict how the pieces can
6048: be filled in, and an output pattern or C code to generate the assembler
1.1.1.8 root 6049: output, all wrapped up in a @code{define_insn} expression.
1.1 root 6050:
1.1.1.8 root 6051: A @code{define_insn} is an RTL expression containing four or five operands:
1.1 root 6052:
6053: @enumerate
6054: @item
6055: An optional name. The presence of a name indicate that this instruction
6056: pattern can perform a certain standard job for the RTL-generation
6057: pass of the compiler. This pass knows certain names and will use
6058: the instruction patterns with those names, if the names are defined
6059: in the machine description.
6060:
6061: The absence of a name is indicated by writing an empty string
6062: where the name should go. Nameless instruction patterns are never
6063: used for generating RTL code, but they may permit several simpler insns
6064: to be combined later on.
6065:
6066: Names that are not thus known and used in RTL-generation have no
6067: effect; they are equivalent to no name at all.
6068:
6069: @item
6070: The @dfn{RTL template} (@pxref{RTL Template}) is a vector of
6071: incomplete RTL expressions which show what the instruction should look
1.1.1.8 root 6072: like. It is incomplete because it may contain @code{match_operand}
6073: and @code{match_dup} expressions that stand for operands of the
1.1 root 6074: instruction.
6075:
1.1.1.10! root 6076: If the vector has only one element, that element is the template for the
! 6077: instruction pattern. If the vector has multiple elements, then the
! 6078: instruction pattern is a @code{parallel} expression containing the
! 6079: elements described.
1.1 root 6080:
6081: @item
6082: A condition. This is a string which contains a C expression that is
6083: the final test to decide whether an insn body matches this pattern.
6084:
6085: For a named pattern, the condition (if present) may not depend on
6086: the data in the insn being matched, but only the target-machine-type
6087: flags. The compiler needs to test these conditions during
6088: initialization in order to learn exactly which named instructions are
6089: available in a particular run.
6090:
6091: For nameless patterns, the condition is applied only when matching an
6092: individual insn, and only after the insn has matched the pattern's
6093: recognition template. The insn's operands may be found in the vector
6094: @code{operands}.
6095:
6096: @item
6097: The @dfn{output template}: a string that says how to output matching
6098: insns as assembler code. @samp{%} in this string specifies where
6099: to substitute the value of an operand. @xref{Output Template}.
6100:
6101: When simple substitution isn't general enough, you can specify a piece
6102: of C code to compute the output. @xref{Output Statement}.
6103:
6104: @item
6105: Optionally, some @dfn{machine-specific information}. The meaning
6106: of this information is defined only by an individual machine description;
6107: typically it might say whether this insn alters the condition codes,
6108: or how many bytes of output it generates.
6109:
6110: This operand is written as a string containing a C initializer
6111: (complete with braces) for the structure type @code{INSN_MACHINE_INFO},
6112: whose definition is up to you (@pxref{Misc}).
6113: @end enumerate
6114:
6115: @node Example, RTL Template, Patterns, Machine Desc
1.1.1.8 root 6116: @section Example of @code{define_insn}
1.1 root 6117:
6118: Here is an actual example of an instruction pattern, for the 68000/68020.
6119:
6120: @example
6121: (define_insn "tstsi"
6122: [(set (cc0)
6123: (match_operand:SI 0 "general_operand" "rm"))]
6124: ""
6125: "*
6126: @{ if (TARGET_68020 || ! ADDRESS_REG_P (operands[0]))
6127: return \"tstl %0\";
6128: return \"cmpl #0,%0\"; @}")
6129: @end example
6130:
6131: This is an instruction that sets the condition codes based on the value of
6132: a general operand. It has no condition, so any insn whose RTL description
6133: has the form shown may be handled according to this pattern. The name
6134: @samp{tstsi} means ``test a @code{SImode} value'' and tells the RTL generation
6135: pass that, when it is necessary to test such a value, an insn to do so
6136: can be constructed using this pattern.
6137:
6138: The output control string is a piece of C code which chooses which
6139: output template to return based on the kind of operand and the specific
6140: type of CPU for which code is being generated.
6141:
6142: @samp{"rm"} is an operand constraint. Its meaning is explained below.
6143:
6144: @node RTL Template, Output Template, Example, Machine Desc
6145: @section RTL Template for Generating and Recognizing Insns
6146:
6147: The RTL template is used to define which insns match the particular pattern
6148: and how to find their operands. For named patterns, the RTL template also
6149: says how to construct an insn from specified operands.
6150:
6151: Construction involves substituting specified operands into a copy of the
6152: template. Matching involves determining the values that serve as the
6153: operands in the insn being matched. Both of these activities are
6154: controlled by special expression types that direct matching and
6155: substitution of the operands.
6156:
6157: @table @code
1.1.1.8 root 6158: @item (match_operand:@var{m} @var{n} @var{pred} @var{constraint})
1.1 root 6159: This expression is a placeholder for operand number @var{n} of
6160: the insn. When constructing an insn, operand number @var{n}
6161: will be substituted at this point. When matching an insn, whatever
6162: appears at this position in the insn will be taken as operand
1.1.1.8 root 6163: number @var{n}; but it must satisfy @var{pred} or this instruction
1.1 root 6164: pattern will not match at all.
6165:
6166: Operand numbers must be chosen consecutively counting from zero in
1.1.1.8 root 6167: each instruction pattern. There may be only one @code{match_operand}
1.1 root 6168: expression in the pattern for each operand number. Usually operands
1.1.1.8 root 6169: are numbered in the order of appearance in @code{match_operand}
1.1 root 6170: expressions.
6171:
1.1.1.8 root 6172: @var{pred} is a string that is the name of a C function that accepts
6173: two arguments, an expression and a machine mode. During matching, the
6174: function will be called with the putative operand as the expression
6175: and @var{m} as the mode argument. If it returns zero, this
6176: instruction pattern fails to match. @var{pred} may be an empty
6177: string; then it means no test is to be done on the operand,
6178: so anything which occurs in this position is valid.
6179:
6180: @var{constraint} controls reloading and the choice of the best register
6181: class to use for a value, as explained later (@pxref{Constraints}).
6182:
6183: People are often unclear on the difference between the constraint and the
6184: predicate. The predicate helps decide whether a given insn matches the
6185: pattern. The constraint plays no role in this decision; instead, it
6186: controls various decisions in the case of an insn which does match.
6187:
6188: Most often, @var{pred} is @code{"general_operand"}. This function checks
6189: that the putative operand is either a constant, a register or a memory
6190: reference, and that it is valid for mode @var{m}.
1.1 root 6191:
1.1.1.8 root 6192: For an operand that must be a register, @var{pred} should be
1.1 root 6193: @code{"register_operand"}. It would be valid to use
6194: @code{"general_operand"}, since the reload pass would copy any
6195: non-register operands through registers, but this would make GNU CC do
6196: extra work, and it would prevent the register allocator from doing the
6197: best possible job.
6198:
1.1.1.8 root 6199: For an operand that must be a constant, either @var{pred} should be
1.1 root 6200: @code{"immediate_operand"}, or the instruction pattern's extra
6201: condition should check for constants, or both. You cannot expect the
6202: constraints to do this work! If the constraints allow only constants,
6203: but the predicate allows something else, the compiler will crash when
6204: that case arises.
6205:
6206: @item (match_dup @var{n})
6207: This expression is also a placeholder for operand number @var{n}.
6208: It is used when the operand needs to appear more than once in the
6209: insn.
6210:
1.1.1.8 root 6211: In construction, @code{match_dup} behaves exactly like
6212: @code{match_operand}: the operand is substituted into the insn being
6213: constructed. But in matching, @code{match_dup} behaves differently.
1.1 root 6214: It assumes that operand number @var{n} has already been determined by
1.1.1.8 root 6215: a @code{match_operand} appearing earlier in the recognition template,
1.1 root 6216: and it matches only an identical-looking expression.
6217:
1.1.1.4 root 6218: @item (match_operator:@var{m} @var{n} "@var{predicate}" [@var{operands}@dots{}])
6219: This pattern is a kind of placeholder for a variable RTL expression
6220: code.
6221:
6222: When constructing an insn, it stands for an RTL expression whose
6223: expression code is taken from that of operand @var{n}, and whose
6224: operands are constructed from the patterns @var{operands}.
6225:
6226: When matching an expression, it matches an expression if the function
6227: @var{predicate} returns nonzero on that expression @emph{and} the
6228: patterns @var{operands} match the operands of the expression.
6229:
6230: Suppose that the function @code{commutative_operator} is defined as
6231: follows, to match any expression whose operator is one of the six
6232: commutative arithmetic operators of RTL and whose mode is @var{mode}:
6233:
6234: @example
6235: int
6236: commutative_operator (x, mode)
6237: rtx x;
6238: enum machine_mode mode;
6239: @{
6240: enum rtx_code code = GET_CODE (x);
6241: if (GET_MODE (x) != mode)
6242: return 0;
6243: return (code == PLUS || code == MULT || code == UMULT
6244: || code == AND || code == IOR || code == XOR);
6245: @}
6246: @end example
6247:
6248: Then the following pattern will match any RTL expression consisting
6249: of a commutative operator applied to two general operands:
6250:
6251: @example
6252: (match_operator:SI 2 "commutative_operator"
6253: [(match_operand:SI 3 "general_operand" "g")
6254: (match_operand:SI 4 "general_operand" "g")])
6255: @end example
6256:
6257: Here the vector @code{[@var{operands}@dots{}]} contains two patterns
6258: because the expressions to be matched all contain two operands.
6259:
6260: When this pattern does match, the two operands of the commutative
6261: operator are recorded as operands 3 and 4 of the insn. (This is done
1.1.1.8 root 6262: by the two instances of @code{match_operand}.) Operand 2 of the insn
1.1.1.4 root 6263: will be the entire commutative expression: use @code{GET_CODE
6264: (operands[2])} to see which commutative operator was used.
6265:
1.1.1.8 root 6266: The machine mode @var{m} of @code{match_operator} works like that of
6267: @code{match_operand}: it is passed as the second argument to the
1.1.1.4 root 6268: predicate function, and that function is solely responsible for
6269: deciding whether the expression to be matched ``has'' that mode.
6270:
6271: When constructing an insn, argument 2 of the gen-function will specify
6272: the operation (i.e. the expression code) for the expression to be
6273: made. It should be an RTL expression, whose expression code is copied
6274: into a new expression whose operands are arguments 3 and 4 of the
6275: gen-function. The subexpressions of argument 2 are not used;
6276: only its expression code matters.
6277:
1.1.1.8 root 6278: There is no way to specify constraints in @code{match_operator}. The
6279: operand of the insn which corresponds to the @code{match_operator}
1.1.1.4 root 6280: never has any constraints because it is never reloaded as a whole.
6281: However, if parts of its @var{operands} are matched by
1.1.1.8 root 6282: @code{match_operand} patterns, those parts may have constraints of
1.1.1.4 root 6283: their own.
6284:
1.1 root 6285: @item (address (match_operand:@var{m} @var{n} "address_operand" ""))
6286: This complex of expressions is a placeholder for an operand number
6287: @var{n} in a ``load address'' instruction: an operand which specifies
6288: a memory location in the usual way, but for which the actual operand
6289: value used is the address of the location, not the contents of the
6290: location.
6291:
1.1.1.8 root 6292: @code{address} expressions never appear in RTL code, only in machine
1.1 root 6293: descriptions. And they are used only in machine descriptions that do
6294: not use the operand constraint feature. When operand constraints are
6295: in use, the letter @samp{p} in the constraint serves this purpose.
6296:
6297: @var{m} is the machine mode of the @emph{memory location being
6298: addressed}, not the machine mode of the address itself. That mode is
6299: always the same on a given target machine (it is @code{Pmode}, which
6300: normally is @code{SImode}), so there is no point in mentioning it;
1.1.1.8 root 6301: thus, no machine mode is written in the @code{address} expression. If
1.1 root 6302: some day support is added for machines in which addresses of different
6303: kinds of objects appear differently or are used differently (such as
6304: the PDP-10), different formats would perhaps need different machine
1.1.1.8 root 6305: modes and these modes might be written in the @code{address}
1.1 root 6306: expression.
6307: @end table
6308:
6309: @node Output Template, Output Statement, RTL Template, Machine Desc
6310: @section Output Templates and Operand Substitution
6311:
1.1.1.6 root 6312: The @dfn{output template} is a string which specifies how to output the
6313: assembler code for an instruction pattern. Most of the template is a
6314: fixed string which is output literally. The character @samp{%} is used
6315: to specify where to substitute an operand; it can also be used to
6316: identify places where different variants of the assembler require
1.1 root 6317: different syntax.
6318:
6319: In the simplest case, a @samp{%} followed by a digit @var{n} says to output
6320: operand @var{n} at that point in the string.
6321:
6322: @samp{%} followed by a letter and a digit says to output an operand in an
6323: alternate fashion. Four letters have standard, built-in meanings described
6324: below. The machine description macro @code{PRINT_OPERAND} can define
6325: additional letters with nonstandard meanings.
6326:
6327: @samp{%c@var{digit}} can be used to substitute an operand that is a
6328: constant value without the syntax that normally indicates an immediate
6329: operand.
6330:
6331: @samp{%n@var{digit}} is like @samp{%c@var{digit}} except that the value of
6332: the constant is negated before printing.
6333:
6334: @samp{%a@var{digit}} can be used to substitute an operand as if it were a
6335: memory reference, with the actual operand treated as the address. This may
6336: be useful when outputting a ``load address'' instruction, because often the
6337: assembler syntax for such an instruction requires you to write the operand
6338: as if it were a memory reference.
6339:
6340: @samp{%l@var{digit}} is used to substitute a @code{label_ref} into a jump
6341: instruction.
6342:
6343: @samp{%} followed by a punctuation character specifies a substitution that
6344: does not use an operand. Only one case is standard: @samp{%%} outputs a
6345: @samp{%} into the assembler code. Other nonstandard cases can be
1.1.1.8 root 6346: defined in the @code{PRINT_OPERAND} macro. You must also define
6347: which punctuation characters are valid with the
6348: @code{PRINT_OPERAND_PUNCT_VALID_P} macro.
1.1 root 6349:
6350: The template may generate multiple assembler instructions. Write the text
6351: for the instructions, with @samp{\;} between them.
6352:
1.1.1.6 root 6353: When the RTL contains two operands which are required by constraint to match
1.1 root 6354: each other, the output template must refer only to the lower-numbered operand.
6355: Matching operands are not always identical, and the rest of the compiler
6356: arranges to put the proper RTL expression for printing into the lower-numbered
6357: operand.
6358:
6359: One use of nonstandard letters or punctuation following @samp{%} is to
6360: distinguish between different assembler languages for the same machine; for
6361: example, Motorola syntax versus MIT syntax for the 68000. Motorola syntax
6362: requires periods in most opcode names, while MIT syntax does not. For
6363: example, the opcode @samp{movel} in MIT syntax is @samp{move.l} in Motorola
6364: syntax. The same file of patterns is used for both kinds of output syntax,
6365: but the character sequence @samp{%.} is used in each place where Motorola
6366: syntax wants a period. The @code{PRINT_OPERAND} macro for Motorola syntax
6367: defines the sequence to output a period; the macro for MIT syntax defines
6368: it to do nothing.
6369:
6370: @node Output Statement, Constraints, Output Template, Machine Desc
6371: @section C Statements for Generating Assembler Output
6372:
6373: Often a single fixed template string cannot produce correct and efficient
6374: assembler code for all the cases that are recognized by a single
6375: instruction pattern. For example, the opcodes may depend on the kinds of
6376: operands; or some unfortunate combinations of operands may require extra
6377: machine instructions.
6378:
6379: If the output control string starts with a @samp{*}, then it is not an
6380: output template but rather a piece of C program that should compute a
6381: template. It should execute a @code{return} statement to return the
6382: template-string you want. Most such templates use C string literals, which
6383: require doublequote characters to delimit them. To include these
6384: doublequote characters in the string, prefix each one with @samp{\}.
6385:
6386: The operands may be found in the array @code{operands}, whose C data type
6387: is @code{rtx []}.
6388:
6389: It is possible to output an assembler instruction and then go on to output
6390: or compute more of them, using the subroutine @code{output_asm_insn}. This
6391: receives two arguments: a template-string and a vector of operands. The
6392: vector may be @code{operands}, or it may be another array of @code{rtx}
6393: that you declare locally and initialize yourself.
6394:
6395: When an insn pattern has multiple alternatives in its constraints, often
1.1.1.5 root 6396: the appearance of the assembler code is determined mostly by which alternative
1.1 root 6397: was matched. When this is so, the C code can test the variable
6398: @code{which_alternative}, which is the ordinal number of the alternative
6399: that was actually satisfied (0 for the first, 1 for the second alternative,
6400: etc.).
6401:
6402: For example, suppose there are two opcodes for storing zero, @samp{clrreg}
6403: for registers and @samp{clrmem} for memory locations. Here is how
6404: a pattern could use @code{which_alternative} to choose between them:
6405:
6406: @example
6407: (define_insn ""
6408: [(set (match_operand:SI 0 "general_operand" "r,m")
6409: (const_int 0))]
6410: ""
6411: "*
6412: return (which_alternative == 0
6413: ? \"clrreg %0\" : \"clrmem %0\");
6414: ")
6415: @end example
6416:
6417: @node Constraints, Standard Names, Output Statement, Machine Desc
6418: @section Operand Constraints
6419:
1.1.1.8 root 6420: Each @code{match_operand} in an instruction pattern can specify a
1.1 root 6421: constraint for the type of operands allowed. Constraints can say whether
6422: an operand may be in a register, and which kinds of register; whether the
6423: operand can be a memory reference, and which kinds of address; whether the
6424: operand may be an immediate constant, and which possible values it may
6425: have. Constraints can also require two operands to match.
6426:
6427: @menu
6428: * Simple Constraints:: Basic use of constraints.
6429: * Multi-Alternative:: When an insn has two alternative constraint-patterns.
6430: * Class Preferences:: Constraints guide which hard register to put things in.
6431: * Modifiers:: More precise control over effects of constraints.
6432: * No Constraints:: Describing a clean machine without constraints.
6433: @end menu
6434:
6435: @node Simple Constraints, Multi-Alternative, Constraints, Constraints
6436: @subsection Simple Constraints
6437:
6438: The simplest kind of constraint is a string full of letters, each of
6439: which describes one kind of operand that is permitted. Here are
6440: the letters that are allowed:
6441:
6442: @table @asis
6443: @item @samp{m}
6444: A memory operand is allowed, with any kind of address that the machine
6445: supports in general.
6446:
6447: @item @samp{o}
6448: A memory operand is allowed, but only if the address is
1.1.1.8 root 6449: @dfn{offsettable}. This means that adding a small integer (actually,
1.1 root 6450: the width in bytes of the operand, as determined by its machine mode)
6451: may be added to the address and the result is also a valid memory
6452: address.
6453:
1.1.1.8 root 6454: For example, an address which is constant is offsettable; so is an
1.1 root 6455: address that is the sum of a register and a constant (as long as a
6456: slightly larger constant is also within the range of address-offsets
6457: supported by the machine); but an autoincrement or autodecrement
1.1.1.8 root 6458: address is not offsettable. More complicated indirect/indexed
6459: addresses may or may not be offsettable depending on the other
1.1 root 6460: addressing modes that the machine supports.
6461:
6462: Note that in an output operand which can be matched by another
6463: operand, the constraint letter @samp{o} is valid only when accompanied
6464: by both @samp{<} (if the target machine has predecrement addressing)
6465: and @samp{>} (if the target machine has preincrement addressing).
6466:
6467: When the constraint letter @samp{o} is used, the reload pass may
1.1.1.8 root 6468: generate instructions which copy a nonoffsettable address into an index
1.1 root 6469: register. The idea is that the register can be used as a replacement
1.1.1.8 root 6470: offsettable address. But this method requires that there be patterns
1.1 root 6471: to copy any kind of address into a register. Auto-increment
6472: and auto-decrement addresses are an exception; there need not be an
6473: instruction that can copy such an address into a register, because
6474: reload handles these cases specially.
6475:
6476: Most older machine designs have ``load address'' instructions which do
6477: just what is needed here. Some RISC machines do not advertise such
6478: instructions, but the possible addresses on these machines are very
6479: limited, so it is easy to fake them.
6480:
6481: @item @samp{<}
6482: A memory operand with autodecrement addressing (either predecrement or
6483: postdecrement) is allowed.
6484:
6485: @item @samp{>}
6486: A memory operand with autoincrement addressing (either preincrement or
6487: postincrement) is allowed.
6488:
6489: @item @samp{r}
6490: A register operand is allowed provided that it is in a general
6491: register.
6492:
6493: @item @samp{d}, @samp{a}, @samp{f}, @dots{}
6494: Other letters can be defined in machine-dependent fashion to stand for
6495: particular classes of registers. @samp{d}, @samp{a} and @samp{f} are
6496: defined on the 68000/68020 to stand for data, address and floating
6497: point registers.
6498:
6499: @item @samp{i}
6500: An immediate integer operand (one with constant value) is allowed.
6501: This includes symbolic constants whose values will be known only at
6502: assembly time.
6503:
6504: @item @samp{n}
6505: An immediate integer operand with a known numeric value is allowed.
6506: Many systems cannot support assembly-time constants for operands less
6507: than a word wide. Constraints for these operands should use @samp{n}
6508: rather than @samp{i}.
6509:
6510: @item @samp{I}, @samp{J}, @samp{K}, @dots{}
6511: Other letters in the range @samp{I} through @samp{M} may be defined in
6512: a machine-dependent fashion to permit immediate integer operands with
6513: explicit integer values in specified ranges. For example, on the
6514: 68000, @samp{I} is defined to stand for the range of values 1 to 8.
6515: This is the range permitted as a shift count in the shift
6516: instructions.
6517:
6518: @item @samp{F}
1.1.1.8 root 6519: An immediate floating operand (expression code @code{const_double}) is
1.1 root 6520: allowed.
6521:
6522: @item @samp{G}, @samp{H}
6523: @samp{G} and @samp{H} may be defined in a machine-dependent fashion to
6524: permit immediate floating operands in particular ranges of values.
6525:
6526: @item @samp{s}
6527: An immediate integer operand whose value is not an explicit integer is
6528: allowed.
6529:
6530: This might appear strange; if an insn allows a constant operand with a
6531: value not known at compile time, it certainly must allow any known
6532: value. So why use @samp{s} instead of @samp{i}? Sometimes it allows
6533: better code to be generated.
6534:
6535: For example, on the 68000 in a fullword instruction it is possible to
1.1.1.9 root 6536: use an immediate operand; but if the immediate value is between -128
6537: and 127, better code results from loading the value into a register and
1.1 root 6538: using the register. This is because the load into the register can be
6539: done with a @samp{moveq} instruction. We arrange for this to happen
6540: by defining the letter @samp{K} to mean ``any integer outside the
1.1.1.9 root 6541: range -128 to 127'', and then specifying @samp{Ks} in the operand
1.1 root 6542: constraints.
6543:
6544: @item @samp{g}
6545: Any register, memory or immediate integer operand is allowed, except for
6546: registers that are not general registers.
6547:
6548: @item @samp{@var{n}} (a digit)
6549: An operand that matches operand number @var{n} is allowed.
6550: If a digit is used together with letters, the digit should come last.
6551:
6552: This is called a @dfn{matching constraint} and what it really means is
6553: that the assembler has only a single operand that fills two roles
6554: considered separate in the RTL insn. For example, an add insn has two
6555: input operands and one output operand in the RTL, but on most machines
6556: an add instruction really has only two operands, one of them an
6557: input-output operand.
6558:
6559: Matching constraints work only in circumstances like that add insn.
6560: More precisely, the matching constraint must appear in an input-only
6561: operand and the operand that it matches must be an output-only operand
1.1.1.5 root 6562: with a lower number. Thus, operand @var{n} must have @samp{=} in its
6563: constraint.
1.1 root 6564:
6565: For operands to match in a particular case usually means that they
6566: are identical-looking RTL expressions. But in a few special cases
6567: specific kinds of dissimilarity are allowed. For example, @code{*x}
6568: as an input operand will match @code{*x++} as an output operand.
6569: For proper results in such cases, the output template should always
6570: use the output-operand's number when printing the operand.
6571:
6572: @item @samp{p}
6573: An operand that is a valid memory address is allowed. This is
6574: for ``load address'' and ``push address'' instructions.
6575:
1.1.1.8 root 6576: @samp{p} in the constraint must be accompanies by @code{address_operand}
6577: as the predicate in the @code{match_operand}.
1.1 root 6578: @end table
6579:
6580: In order to have valid assembler code, each operand must satisfy
6581: its constraint. But a failure to do so does not prevent the pattern
6582: from applying to an insn. Instead, it directs the compiler to modify
6583: the code so that the constraint will be satisfied. Usually this is
6584: done by copying an operand into a register.
6585:
6586: Contrast, therefore, the two instruction patterns that follow:
6587:
6588: @example
6589: (define_insn ""
6590: [(set (match_operand:SI 0 "general_operand" "r")
6591: (plus:SI (match_dup 0)
6592: (match_operand:SI 1 "general_operand" "r")))]
6593: ""
6594: "@dots{}")
6595: @end example
6596:
6597: @noindent
6598: which has two operands, one of which must appear in two places, and
6599:
6600: @example
6601: (define_insn ""
6602: [(set (match_operand:SI 0 "general_operand" "r")
6603: (plus:SI (match_operand:SI 1 "general_operand" "0")
6604: (match_operand:SI 2 "general_operand" "r")))]
6605: ""
6606: "@dots{}")
6607: @end example
6608:
6609: @noindent
6610: which has three operands, two of which are required by a constraint to be
6611: identical. If we are considering an insn of the form
6612:
6613: @example
6614: (insn @var{n} @var{prev} @var{next}
6615: (set (reg:SI 3)
6616: (plus:SI (reg:SI 6) (reg:SI 109)))
6617: @dots{})
6618: @end example
6619:
6620: @noindent
6621: the first pattern would not apply at all, because this insn does not
6622: contain two identical subexpressions in the right place. The pattern would
6623: say, ``That does not look like an add instruction; try other patterns.''
6624: The second pattern would say, ``Yes, that's an add instruction, but there
6625: is something wrong with it.'' It would direct the reload pass of the
6626: compiler to generate additional insns to make the constraint true. The
6627: results might look like this:
6628:
6629: @example
6630: (insn @var{n2} @var{prev} @var{n}
6631: (set (reg:SI 3) (reg:SI 6))
6632: @dots{})
6633:
6634: (insn @var{n} @var{n2} @var{next}
6635: (set (reg:SI 3)
6636: (plus:SI (reg:SI 3) (reg:SI 109)))
6637: @dots{})
6638: @end example
6639:
6640: It is up to you to make sure that each operand, in each pattern, has
6641: constraints that can handle any RTL expression that could be present for
6642: that operand. (When multiple alternatives are in use, each pattern must,
6643: for each possible combination of operand expressions, have at least one
6644: alternative which can handle that combination of operands.) The
6645: constraints don't need to @emph{allow} any possible operand---when this is
6646: the case, they do not constrain---but they must at least point the way to
6647: reloading any possible operand so that it will fit.
6648:
6649: @itemize @bullet
6650: @item
6651: If the constraint accepts whatever operands the predicate permits,
6652: there is no problem: reloading is never necessary for this operand.
6653:
6654: For example, an operand whose constraints permit everything except
6655: registers is safe provided its predicate rejects registers.
6656:
6657: An operand whose predicate accepts only constant values is safe
6658: provided its constraints include the letter @samp{i}. If any possible
6659: constant value is accepted, then nothing less than @samp{i} will do;
1.1.1.5 root 6660: if the predicate is more selective, then the constraints may also be
1.1 root 6661: more selective.
6662:
6663: @item
6664: Any operand expression can be reloaded by copying it into a register.
6665: So if an operand's constraints allow some kind of register, it is
6666: certain to be safe. It need not permit all classes of registers; the
6667: compiler knows how to copy a register into another register of the
6668: proper class in order to make an instruction valid.
6669:
6670: @item
1.1.1.8 root 6671: A nonoffsettable memory reference can be reloaded by copying the
1.1 root 6672: address into a register. So if the constraint uses the letter
6673: @samp{o}, all memory references are taken care of.
6674:
6675: @item
1.1.1.8 root 6676: A constant operand can be reloaded by allocating space in memory to
6677: hold it as preinitialized data. Then the memory reference can be used
6678: in place of the constant. So if the constraint uses the letters
6679: @samp{o} or @samp{m}, constant operands are not a problem.
1.1 root 6680: @end itemize
6681:
6682: If the operand's predicate can recognize registers, but the constraint does
6683: not permit them, it can make the compiler crash. When this operand happens
6684: to be a register, the reload pass will be stymied, because it does not know
6685: how to copy a register temporarily into memory.
6686:
6687: @node Multi-Alternative, Class Preferences, Simple Constraints, Constraints
6688: @subsection Multiple Alternative Constraints
6689:
6690: Sometimes a single instruction has multiple alternative sets of possible
6691: operands. For example, on the 68000, a logical-or instruction can combine
6692: register or an immediate value into memory, or it can combine any kind of
6693: operand into a register; but it cannot combine one memory location into
6694: another.
6695:
6696: These constraints are represented as multiple alternatives. An alternative
6697: can be described by a series of letters for each operand. The overall
6698: constraint for an operand is made from the letters for this operand
6699: from the first alternative, a comma, the letters for this operand from
6700: the second alternative, a comma, and so on until the last alternative.
6701: Here is how it is done for fullword logical-or on the 68000:
6702:
6703: @example
6704: (define_insn "iorsi3"
1.1.1.9 root 6705: [(set (match_operand:SI 0 "general_operand" "=m,d")
6706: (ior:SI (match_operand:SI 1 "general_operand" "%0,0")
1.1 root 6707: (match_operand:SI 2 "general_operand" "dKs,dmKs")))]
6708: @dots{})
6709: @end example
6710:
6711: The first alternative has @samp{m} (memory) for operand 0, @samp{0} for
1.1.1.9 root 6712: operand 1 (meaning it must match operand 0), and @samp{dKs} for operand
6713: 2. The second alternative has @samp{d} (data register) for operand 0,
6714: @samp{0} for operand 1, and @samp{dmKs} for operand 2. The @samp{=} and
6715: @samp{%} in the constraints apply to all the alternatives; their meaning
1.1 root 6716: is explained in the next section.
6717:
6718: If all the operands fit any one alternative, the instruction is valid.
6719: Otherwise, for each alternative, the compiler counts how many instructions
6720: must be added to copy the operands so that that alternative applies.
6721: The alternative requiring the least copying is chosen. If two alternatives
6722: need the same amount of copying, the one that comes first is chosen.
6723: These choices can be altered with the @samp{?} and @samp{!} characters:
6724:
6725: @table @samp
6726: @item ?
6727: Disparage slightly the alternative that the @samp{?} appears in,
6728: as a choice when no alternative applies exactly. The compiler regards
6729: this alternative as one unit more costly for each @samp{?} that appears
6730: in it.
6731:
6732: @item !
6733: Disparage severely the alternative that the @samp{!} appears in.
6734: When operands must be copied into registers, the compiler will
6735: never choose this alternative as the one to strive for.
6736: @end table
6737:
1.1.1.5 root 6738: When an insn pattern has multiple alternatives in its constraints, often
6739: the appearance of the assembler code is determined mostly by which
1.1 root 6740: alternative was matched. When this is so, the C code for writing the
6741: assembler code can use the variable @code{which_alternative}, which is
1.1.1.5 root 6742: the ordinal number of the alternative that was actually satisfied (0 for
6743: the first, 1 for the second alternative, etc.). For example:
1.1 root 6744:
6745: @example
6746: (define_insn ""
6747: [(set (match_operand:SI 0 "general_operand" "r,m")
6748: (const_int 0))]
6749: ""
6750: "*
6751: return (which_alternative == 0
6752: ? \"clrreg %0\" : \"clrmem %0\");
6753: ")
6754: @end example
6755:
6756: @node Class Preferences, Modifiers, Multi-Alternative, Constraints
6757: @subsection Register Class Preferences
6758:
6759: The operand constraints have another function: they enable the compiler
6760: to decide which kind of hardware register a pseudo register is best
6761: allocated to. The compiler examines the constraints that apply to the
6762: insns that use the pseudo register, looking for the machine-dependent
6763: letters such as @samp{d} and @samp{a} that specify classes of registers.
6764: The pseudo register is put in whichever class gets the most ``votes''.
6765: The constraint letters @samp{g} and @samp{r} also vote: they vote in
6766: favor of a general register. The machine description says which registers
6767: are considered general.
6768:
6769: Of course, on some machines all registers are equivalent, and no register
6770: classes are defined. Then none of this complexity is relevant.
6771:
6772: @node Modifiers, No Constraints, Class Preferences, Constraints
6773: @subsection Constraint Modifier Characters
6774:
6775: @table @samp
6776: @item =
6777: Means that this operand is write-only for this instruction: the previous
6778: value is discarded and replaced by output data.
6779:
6780: @item +
6781: Means that this operand is both read and written by the instruction.
6782:
6783: When the compiler fixes up the operands to satisfy the constraints,
6784: it needs to know which operands are inputs to the instruction and
6785: which are outputs from it. @samp{=} identifies an output; @samp{+}
6786: identifies an operand that is both input and output; all other operands
6787: are assumed to be input only.
6788:
6789: @item &
6790: Means (in a particular alternative) that this operand is written
6791: before the instruction is finished using the input operands.
6792: Therefore, this operand may not lie in a register that is used as an
6793: input operand or as part of any memory address.
6794:
6795: @samp{&} applies only to the alternative in which it is written. In
6796: constraints with multiple alternatives, sometimes one alternative
6797: requires @samp{&} while others do not. See, for example, the
6798: @samp{movdf} insn of the 68000.
6799:
6800: @samp{&} does not obviate the need to write @samp{=}.
6801:
6802: @item %
6803: Declares the instruction to be commutative for this operand and the
6804: following operand. This means that the compiler may interchange the
6805: two operands if that is the cheapest way to make all operands fit the
6806: constraints. This is often used in patterns for addition instructions
6807: that really have only two operands: the result must go in one of the
6808: arguments. Here for example, is how the 68000 halfword-add
6809: instruction is defined:
6810:
6811: @example
6812: (define_insn "addhi3"
6813: [(set (match_operand:HI 0 "general_operand" "=m,r")
6814: (plus:HI (match_operand:HI 1 "general_operand" "%0,0")
6815: (match_operand:HI 2 "general_operand" "di,g")))]
6816: @dots{})
6817: @end example
6818:
6819: Note that in previous versions of GNU CC the @samp{%} constraint
6820: modifier always applied to operands 1 and 2 regardless of which
6821: operand it was written in. The usual custom was to write it in
6822: operand 0. Now it must be in operand 1 if the operands to be
6823: exchanged are 1 and 2.
6824:
6825: @item #
6826: Says that all following characters, up to the next comma, are to be
6827: ignored as a constraint. They are significant only for choosing
6828: register preferences.
6829:
6830: @item *
6831: Says that the following character should be ignored when choosing
6832: register preferences. @samp{*} has no effect on the meaning of the
6833: constraint as a constraint.
6834:
6835: Here is an example: the 68000 has an instruction to sign-extend a
6836: halfword in a data register, and can also sign-extend a value by
6837: copying it into an address register. While either kind of register is
6838: acceptable, the constraints on an address-register destination are
6839: less strict, so it is best if register allocation makes an address
6840: register its goal. Therefore, @samp{*} is used so that the @samp{d}
6841: constraint letter (for data register) is ignored when computing
6842: register preferences.
6843:
6844: @example
6845: (define_insn "extendhisi2"
6846: [(set (match_operand:SI 0 "general_operand" "=*d,a")
6847: (sign_extend:SI
6848: (match_operand:HI 1 "general_operand" "0,g")))]
6849: @dots{})
6850: @end example
6851: @end table
6852:
6853: @node No Constraints,, Modifiers, Constraints
6854: @subsection Not Using Constraints
6855:
6856: Some machines are so clean that operand constraints are not required. For
6857: example, on the Vax, an operand valid in one context is valid in any other
6858: context. On such a machine, every operand constraint would be @samp{g},
6859: excepting only operands of ``load address'' instructions which are
6860: written as if they referred to a memory location's contents but actual
6861: refer to its address. They would have constraint @samp{p}.
6862:
6863: For such machines, instead of writing @samp{g} and @samp{p} for all
6864: the constraints, you can choose to write a description with empty constraints.
1.1.1.8 root 6865: Then you write @samp{""} for the constraint in every @code{match_operand}.
6866: Address operands are identified by writing an @code{address} expression
6867: around the @code{match_operand}, not by their constraints.
1.1 root 6868:
6869: When the machine description has just empty constraints, certain parts
1.1.1.6 root 6870: of compilation are skipped, making the compiler faster. However,
6871: few machines actually do not need constraints; all machine descriptions
6872: now in existence use constraints.
1.1 root 6873:
6874: @node Standard Names, Pattern Ordering, Constraints, Machine Desc
6875: @section Standard Names for Patterns Used in Generation
6876:
6877: Here is a table of the instruction names that are meaningful in the RTL
6878: generation pass of the compiler. Giving one of these names to an
6879: instruction pattern tells the RTL generation pass that it can use the
6880: pattern in to accomplish a certain task.
6881:
6882: @table @asis
6883: @item @samp{mov@var{m}}
1.1.1.8 root 6884: Here @var{m} stands for a two-letter machine mode name, in lower case.
6885: This instruction pattern moves data with that machine mode from operand
6886: 1 to operand 0. For example, @samp{movsi} moves full-word data.
1.1 root 6887:
1.1.1.8 root 6888: If operand 0 is a @code{subreg} with mode @var{m} of a register whose
6889: own mode is wider than @var{m}, the effect of this instruction is
1.1 root 6890: to store the specified value in the part of the register that corresponds
6891: to mode @var{m}. The effect on the rest of the register is undefined.
6892:
6893: This class of patterns is special in several ways. First of all, each
6894: of these names @emph{must} be defined, because there is no other way
6895: to copy a datum from one place to another.
6896:
6897: Second, these patterns are not used solely in the RTL generation pass.
6898: Even the reload pass can generate move insns to copy values from stack
1.1.1.8 root 6899: slots into temporary registers. When it does so, one of the operands is
6900: a hard register and the other is an operand that can need to be reloaded
6901: into a register.
6902:
6903: Therefore, when given such a pair of operands, the pattern must generate
6904: RTL which needs no reloading and needs no temporary registers---no
6905: registers other than the operands. For example, if you support the
6906: pattern with a @code{define_expand}, then in such a case the
6907: @code{define_expand} mustn't call @code{force_reg} or any other such
6908: function which might generate new pseudo registers.
1.1 root 6909:
6910: This requirement exists even for subword modes on a RISC machine where
6911: fetching those modes from memory normally requires several insns and
6912: some temporary registers. Look in @file{spur.md} to see how the
1.1.1.8 root 6913: requirement can be satisfied.
1.1 root 6914:
6915: The variety of operands that have reloads depends on the rest of the
6916: machine description, but typically on a RISC machine these can only be
6917: pseudo registers that did not get hard registers, while on other
6918: machines explicit memory references will get optional reloads.
6919:
1.1.1.9 root 6920: The constraints on a @samp{move@var{m}} must allow any hard register to
6921: be moved to any other hard register (provided that
6922: @code{HARD_REGNO_MODE_OK} permits mode @var{m} in both registers).
6923:
6924: It is obligatory to support floating point @samp{move@var{m}}
6925: instructions into and out of any registers that can hold fixed point
6926: values, because unions and structures (which have modes @code{SImode} or
6927: @code{DImode}) can be in those registers and they may have floating
6928: point members.
6929:
6930: There may also be a need to support fixed point @samp{move@var{m}}
6931: instructions in and out of floating point registers. Unfortunately, I
6932: have forgotten why this was so, and I don't know whether it is still
6933: true. If @code{HARD_REGNO_MODE_OK} rejects fixed point values in
6934: floating point registers, then the constraints of the fixed point
6935: @samp{move@var{m}} instructions must be designed to avoid ever trying to
6936: reload into a floating point register.
1.1 root 6937:
6938: @item @samp{movstrict@var{m}}
1.1.1.8 root 6939: Like @samp{mov@var{m}} except that if operand 0 is a @code{subreg}
1.1 root 6940: with mode @var{m} of a register whose natural mode is wider,
6941: the @samp{movstrict@var{m}} instruction is guaranteed not to alter
6942: any of the register except the part which belongs to mode @var{m}.
6943:
6944: @item @samp{add@var{m}3}
6945: Add operand 2 and operand 1, storing the result in operand 0. All operands
6946: must have mode @var{m}. This can be used even on two-address machines, by
6947: means of constraints requiring operands 1 and 0 to be the same location.
6948:
6949: @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}
6950: Similar, for other arithmetic operations.
6951:
6952: There are special considerations for register classes for logical-and
6953: instructions, affecting also the macro @code{PREFERRED_RELOAD_CLASS}.
6954: They apply not only to the patterns with these standard names, but to
6955: any patterns that will match such an instruction. @xref{Register
6956: Classes}.
6957:
6958: @item @samp{mulhisi3}
6959: Multiply operands 1 and 2, which have mode @code{HImode}, and store
6960: a @code{SImode} product in operand 0.
6961:
6962: @item @samp{mulqihi3}, @samp{mulsidi3}
6963: Similar widening-multiplication instructions of other widths.
6964:
6965: @item @samp{umulqihi3}, @samp{umulhisi3}, @samp{umulsidi3}
6966: Similar widening-multiplication instructions that do unsigned
6967: multiplication.
6968:
6969: @item @samp{divmod@var{m}4}
6970: Signed division that produces both a quotient and a remainder.
6971: Operand 1 is divided by operand 2 to produce a quotient stored
6972: in operand 0 and a remainder stored in operand 3.
6973:
6974: @item @samp{udivmod@var{m}4}
6975: Similar, but does unsigned division.
6976:
6977: @item @samp{ashl@var{m}3}
6978: Arithmetic-shift operand 1 left by a number of bits specified by
6979: operand 2, and store the result in operand 0. Operand 2 has
6980: mode @code{SImode}, not mode @var{m}.
6981:
6982: @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}
6983: Other shift and rotate instructions.
6984:
6985: Logical and arithmetic left shift are the same. Machines that do not
6986: allow negative shift counts often have only one instruction for
6987: shifting left. On such machines, you should define a pattern named
6988: @samp{ashl@var{m}3} and leave @samp{lshl@var{m}3} undefined.
6989:
6990: There are special considerations for register classes for shift
6991: instructions, affecting also the macro @code{PREFERRED_RELOAD_CLASS}.
6992: They apply not only to the patterns with these standard names, but to
6993: any patterns that will match such an instruction. @xref{Register
6994: Classes}.
6995:
6996: @item @samp{neg@var{m}2}
6997: Negate operand 1 and store the result in operand 0.
6998:
6999: @item @samp{abs@var{m}2}
7000: Store the absolute value of operand 1 into operand 0.
7001:
7002: @item @samp{sqrt@var{m}2}
7003: Store the square root of operand 1 into operand 0.
7004:
7005: @item @samp{ffs@var{m}2}
7006: Store into operand 0 one plus the index of the least significant 1-bit
7007: of operand 1. If operand 1 is zero, store zero. @var{m} is the mode
7008: of operand 0; operand 1's mode is specified by the instruction
7009: pattern, and the compiler will convert the operand to that mode before
7010: generating the instruction.
7011:
7012: @item @samp{one_cmpl@var{m}2}
7013: Store the bitwise-complement of operand 1 into operand 0.
7014:
7015: @item @samp{cmp@var{m}}
7016: Compare operand 0 and operand 1, and set the condition codes.
7017: The RTL pattern should look like this:
7018:
7019: @example
1.1.1.6 root 7020: (set (cc0) (compare (match_operand:@var{m} 0 @dots{})
7021: (match_operand:@var{m} 1 @dots{})))
1.1 root 7022: @end example
7023:
7024: Each such definition in the machine description, for integer mode
7025: @var{m}, must have a corresponding @samp{tst@var{m}} pattern, because
7026: optimization can simplify the compare into a test when operand 1 is
7027: zero.
7028:
7029: @item @samp{tst@var{m}}
7030: Compare operand 0 against zero, and set the condition codes.
7031: The RTL pattern should look like this:
7032:
7033: @example
7034: (set (cc0) (match_operand:@var{m} 0 @dots{}))
7035: @end example
7036:
7037: @item @samp{movstr@var{m}}
7038: Block move instruction. The addresses of the destination and source
7039: strings are the first two operands, and both are in mode @code{Pmode}.
7040: The number of bytes to move is the third operand, in mode @var{m}.
1.1.1.5 root 7041: The fourth operand is the known shared alignment of the source and
7042: destination, in the form of a @code{const_int} rtx.
1.1 root 7043:
7044: @item @samp{cmpstr@var{m}}
7045: Block compare instruction, with operands like @samp{movstr@var{m}}
7046: except that the two memory blocks are compared byte by byte
7047: in lexicographic order. The effect of the instruction is to set
7048: the condition codes.
7049:
7050: @item @samp{float@var{m}@var{n}2}
1.1.1.9 root 7051: Convert signed integer operand 1 (valid for fixed point mode @var{m}) to
7052: floating point mode @var{n} and store in operand 0 (which has mode
7053: @var{n}).
7054:
7055: @item @samp{floatuns@var{m}@var{n}2}
7056: Convert unsigned integer operand 1 (valid for fixed point mode @var{m})
7057: to floating point mode @var{n} and store in operand 0 (which has mode
7058: @var{n}).
1.1 root 7059:
7060: @item @samp{fix@var{m}@var{n}2}
7061: Convert operand 1 (valid for floating point mode @var{m}) to fixed
7062: point mode @var{n} as a signed number and store in operand 0 (which
7063: has mode @var{n}). This instruction's result is defined only when
7064: the value of operand 1 is an integer.
7065:
7066: @item @samp{fixuns@var{m}@var{n}2}
7067: Convert operand 1 (valid for floating point mode @var{m}) to fixed
7068: point mode @var{n} as an unsigned number and store in operand 0 (which
7069: has mode @var{n}). This instruction's result is defined only when the
7070: value of operand 1 is an integer.
7071:
7072: @item @samp{ftrunc@var{m}2}
7073: Convert operand 1 (valid for floating point mode @var{m}) to an
7074: integer value, still represented in floating point mode @var{m}, and
7075: store it in operand 0 (valid for floating point mode @var{m}).
7076:
7077: @item @samp{fix_trunc@var{m}@var{n}2}
7078: Like @samp{fix@var{m}@var{n}2} but works for any floating point value
7079: of mode @var{m} by converting the value to an integer.
7080:
7081: @item @samp{fixuns_trunc@var{m}@var{n}2}
7082: Like @samp{fixuns@var{m}@var{n}2} but works for any floating point
7083: value of mode @var{m} by converting the value to an integer.
7084:
7085: @item @samp{trunc@var{m}@var{n}}
7086: Truncate operand 1 (valid for mode @var{m}) to mode @var{n} and
7087: store in operand 0 (which has mode @var{n}). Both modes must be fixed
7088: point or both floating point.
7089:
7090: @item @samp{extend@var{m}@var{n}}
7091: Sign-extend operand 1 (valid for mode @var{m}) to mode @var{n} and
7092: store in operand 0 (which has mode @var{n}). Both modes must be fixed
7093: point or both floating point.
7094:
7095: @item @samp{zero_extend@var{m}@var{n}}
7096: Zero-extend operand 1 (valid for mode @var{m}) to mode @var{n} and
7097: store in operand 0 (which has mode @var{n}). Both modes must be fixed
7098: point.
7099:
7100: @item @samp{extv}
7101: Extract a bit-field from operand 1 (a register or memory operand),
7102: where operand 2 specifies the width in bits and operand 3 the starting
1.1.1.10! root 7103: bit, and store it in operand 0. Operand 0 must have @code{SImode}.
1.1 root 7104: Operand 1 may have mode @code{QImode} or @code{SImode}; often
7105: @code{SImode} is allowed only for registers. Operands 2 and 3 must be
7106: valid for @code{SImode}.
7107:
7108: The RTL generation pass generates this instruction only with constants
7109: for operands 2 and 3.
7110:
7111: The bit-field value is sign-extended to a full word integer
7112: before it is stored in operand 0.
7113:
7114: @item @samp{extzv}
7115: Like @samp{extv} except that the bit-field value is zero-extended.
7116:
7117: @item @samp{insv}
7118: Store operand 3 (which must be valid for @code{SImode}) into a
7119: bit-field in operand 0, where operand 1 specifies the width in bits
7120: and operand 2 the starting bit. Operand 0 may have mode @code{QImode}
7121: or @code{SImode}; often @code{SImode} is allowed only for registers.
7122: Operands 1 and 2 must be valid for @code{SImode}.
7123:
7124: The RTL generation pass generates this instruction only with constants
7125: for operands 1 and 2.
7126:
7127: @item @samp{s@var{cond}}
7128: Store zero or nonzero in the operand according to the condition codes.
7129: Value stored is nonzero iff the condition @var{cond} is true.
7130: @var{cond} is the name of a comparison operation expression code, such
1.1.1.8 root 7131: as @code{eq}, @code{lt} or @code{leu}.
1.1 root 7132:
7133: You specify the mode that the operand must have when you write the
7134: @code{match_operand} expression. The compiler automatically sees
7135: which mode you have used and supplies an operand of that mode.
7136:
1.1.1.8 root 7137: The value stored for a true condition must have 1 as its low bit, or
7138: else must be negative. Otherwise the instruction is not suitable and
7139: must be omitted from the machine description. You must tell the
7140: compiler exactly which value is stored by defining the macro
7141: @code{STORE_FLAG_VALUE}.
1.1 root 7142:
7143: @item @samp{b@var{cond}}
1.1.1.8 root 7144: Conditional branch instruction. Operand 0 is a @code{label_ref}
1.1 root 7145: that refers to the label to jump to. Jump if the condition codes
7146: meet condition @var{cond}.
7147:
7148: @item @samp{call}
7149: Subroutine call instruction returning no value. Operand 0 is the
7150: function to call; operand 1 is the number of bytes of arguments pushed
1.1.1.8 root 7151: (in mode @code{SImode}, except it is normally a @code{const_int});
1.1 root 7152: operand 2 is the number of registers used as operands.
7153:
7154: On most machines, operand 2 is not actually stored into the RTL
7155: pattern. It is supplied for the sake of some RISC machines which need
7156: to put this information into the assembler code; they can put it in
7157: the RTL instead of operand 1.
7158:
1.1.1.8 root 7159: Operand 0 should be a @code{mem} RTX whose address is the address of
1.1 root 7160: the function.
7161:
7162: @item @samp{call_value}
7163: Subroutine call instruction returning a value. Operand 0 is the hard
7164: register in which the value is returned. There are three more
7165: operands, the same as the three operands of the @samp{call}
7166: instruction (but with numbers increased by one).
7167:
7168: Subroutines that return @code{BLKmode} objects use the @samp{call}
7169: insn.
7170:
7171: @item @samp{return}
7172: Subroutine return instruction. This instruction pattern name should be
7173: defined only if a single instruction can do all the work of returning
7174: from a function.
7175:
1.1.1.8 root 7176: @item @samp{nop}
7177: No-op instruction. This instruction pattern name should always be defined
7178: to output a no-op in assembler code. @code{(const_int 0)} will do as an
7179: RTL pattern.
7180:
1.1 root 7181: @item @samp{casesi}
7182: Instruction to jump through a dispatch table, including bounds checking.
7183: This instruction takes five operands:
7184:
7185: @enumerate
7186: @item
7187: The index to dispatch on, which has mode @code{SImode}.
7188:
7189: @item
7190: The lower bound for indices in the table, an integer constant.
7191:
7192: @item
1.1.1.6 root 7193: The total range of indices in the table---the largest index
7194: minus the smallest one (both inclusive).
1.1 root 7195:
7196: @item
7197: A label to jump to if the index has a value outside the bounds.
7198: (If the machine-description macro @code{CASE_DROPS_THROUGH} is defined,
7199: then an out-of-bounds index drops through to the code following
7200: the jump table instead of jumping to this label. In that case,
7201: this label is not actually used by the @samp{casesi} instruction,
7202: but it is always provided as an operand.)
7203:
7204: @item
7205: A label that precedes the table itself.
7206: @end enumerate
7207:
1.1.1.8 root 7208: The table is a @code{addr_vec} or @code{addr_diff_vec} inside of a
7209: @code{jump_insn}. The number of elements in the table is one plus the
1.1 root 7210: difference between the upper bound and the lower bound.
7211:
7212: @item @samp{tablejump}
7213: Instruction to jump to a variable address. This is a low-level
7214: capability which can be used to implement a dispatch table when there
7215: is no @samp{casesi} pattern.
7216:
7217: This pattern requires two operands: the address or offset, and a label
7218: which should immediately precede the jump table. If the macro
7219: @code{CASE_VECTOR_PC_RELATIVE} is defined then the first operand is an
1.1.1.10! root 7220: offset that counts from the address of the table; otherwise, it is an
! 7221: absolute address to jump to.
1.1 root 7222:
7223: The @samp{tablejump} insn is always the last insn before the jump
7224: table it uses. Its assembler code normally has no need to use the
7225: second operand, but you should incorporate it in the RTL pattern so
7226: that the jump optimizer will not delete the table as unreachable code.
7227: @end table
7228:
7229: @node Pattern Ordering, Dependent Patterns, Standard Names, Machine Desc
7230: @section When the Order of Patterns Matters
7231:
7232: Sometimes an insn can match more than one instruction pattern. Then the
7233: pattern that appears first in the machine description is the one used.
7234: Therefore, more specific patterns (patterns that will match fewer things)
7235: and faster instructions (those that will produce better code when they
7236: do match) should usually go first in the description.
7237:
7238: In some cases the effect of ordering the patterns can be used to hide
7239: a pattern when it is not valid. For example, the 68000 has an
7240: instruction for converting a fullword to floating point and another
7241: for converting a byte to floating point. An instruction converting
7242: an integer to floating point could match either one. We put the
7243: pattern to convert the fullword first to make sure that one will
7244: be used rather than the other. (Otherwise a large integer might
7245: be generated as a single-byte immediate quantity, which would not work.)
7246: Instead of using this pattern ordering it would be possible to make the
7247: pattern for convert-a-byte smart enough to deal properly with any
7248: constant value.
7249:
7250: @node Dependent Patterns, Jump Patterns, Pattern Ordering, Machine Desc
7251: @section Interdependence of Patterns
7252:
7253: Every machine description must have a named pattern for each of the
7254: conditional branch names @samp{b@var{cond}}. The recognition template
7255: must always have the form
7256:
7257: @example
7258: (set (pc)
7259: (if_then_else (@var{cond} (cc0) (const_int 0))
7260: (label_ref (match_operand 0 "" ""))
7261: (pc)))
7262: @end example
7263:
7264: @noindent
7265: In addition, every machine description must have an anonymous pattern
7266: for each of the possible reverse-conditional branches. These patterns
7267: look like
7268:
7269: @example
7270: (set (pc)
7271: (if_then_else (@var{cond} (cc0) (const_int 0))
7272: (pc)
7273: (label_ref (match_operand 0 "" ""))))
7274: @end example
7275:
7276: @noindent
7277: They are necessary because jump optimization can turn direct-conditional
7278: branches into reverse-conditional branches.
7279:
7280: The compiler does more with RTL than just create it from patterns
7281: and recognize the patterns: it can perform arithmetic expression codes
7282: when constant values for their operands can be determined. As a result,
7283: sometimes having one pattern can require other patterns. For example, the
7284: Vax has no `and' instruction, but it has `and not' instructions. Here
7285: is the definition of one of them:
7286:
7287: @example
7288: (define_insn "andcbsi2"
7289: [(set (match_operand:SI 0 "general_operand" "")
7290: (and:SI (match_dup 0)
7291: (not:SI (match_operand:SI
7292: 1 "general_operand" ""))))]
7293: ""
7294: "bicl2 %1,%0")
7295: @end example
7296:
7297: @noindent
7298: If operand 1 is an explicit integer constant, an instruction constructed
7299: using that pattern can be simplified into an `and' like this:
7300:
7301: @example
7302: (set (reg:SI 41)
7303: (and:SI (reg:SI 41)
7304: (const_int 0xffff7fff)))
7305: @end example
7306:
7307: @noindent
7308: (where the integer constant is the one's complement of what
7309: appeared in the original instruction).
7310:
7311: To avoid a fatal error, the compiler must have a pattern that recognizes
7312: such an instruction. Here is what is used:
7313:
7314: @example
7315: (define_insn ""
7316: [(set (match_operand:SI 0 "general_operand" "")
7317: (and:SI (match_dup 0)
7318: (match_operand:SI 1 "general_operand" "")))]
7319: "GET_CODE (operands[1]) == CONST_INT"
7320: "*
7321: @{ operands[1]
7322: = gen_rtx (CONST_INT, VOIDmode, ~INTVAL (operands[1]));
7323: return \"bicl2 %1,%0\";
7324: @}")
7325: @end example
7326:
7327: @noindent
7328: Whereas a pattern to match a general `and' instruction is impossible to
7329: support on the Vax, this pattern is possible because it matches only a
7330: constant second argument: a special case that can be output as an `and not'
7331: instruction.
7332:
7333: A ``compare'' instruction whose RTL looks like this:
7334:
7335: @example
1.1.1.6 root 7336: (set (cc0) (compare @var{operand} (const_int 0)))
1.1 root 7337: @end example
7338:
7339: @noindent
7340: may be simplified by optimization into a ``test'' like this:
7341:
7342: @example
7343: (set (cc0) @var{operand})
7344: @end example
7345:
7346: @noindent
7347: So in the machine description, each ``compare'' pattern for an integer
7348: mode must have a corresponding ``test'' pattern that will match the
7349: result of such simplification.
7350:
7351: In some cases machines support instructions identical except for the
7352: machine mode of one or more operands. For example, there may be
7353: ``sign-extend halfword'' and ``sign-extend byte'' instructions whose
7354: patterns are
7355:
7356: @example
7357: (set (match_operand:SI 0 @dots{})
7358: (extend:SI (match_operand:HI 1 @dots{})))
7359:
7360: (set (match_operand:SI 0 @dots{})
7361: (extend:SI (match_operand:QI 1 @dots{})))
7362: @end example
7363:
7364: @noindent
7365: Constant integers do not specify a machine mode, so an instruction to
7366: extend a constant value could match either pattern. The pattern it
7367: actually will match is the one that appears first in the file. For correct
7368: results, this must be the one for the widest possible mode (@code{HImode},
7369: here). If the pattern matches the @code{QImode} instruction, the results
7370: will be incorrect if the constant value does not actually fit that mode.
7371:
7372: Such instructions to extend constants are rarely generated because they are
7373: optimized away, but they do occasionally happen in nonoptimized
7374: compilations.
7375:
7376: When an instruction has the constraint letter @samp{o}, the reload
1.1.1.8 root 7377: pass may generate instructions which copy a nonoffsettable address into
1.1 root 7378: an index register. The idea is that the register can be used as a
1.1.1.8 root 7379: replacement offsettable address. In order for these generated
1.1 root 7380: instructions to work, there must be patterns to copy any kind of valid
7381: address into a register.
7382:
7383: Most older machine designs have ``load address'' instructions which do
7384: just what is needed here. Some RISC machines do not advertise such
7385: instructions, but the possible addresses on these machines are very
7386: limited, so it is easy to fake them.
7387:
7388: Auto-increment and auto-decrement addresses are an exception; there
7389: need not be an instruction that can copy such an address into a
7390: register, because reload handles these cases in a different manner.
7391:
7392: @node Jump Patterns, Peephole Definitions, Dependent Patterns, Machine Desc
7393: @section Defining Jump Instruction Patterns
7394:
7395: GNU CC assumes that the machine has a condition code. A comparison insn
7396: sets the condition code, recording the results of both signed and unsigned
7397: comparison of the given operands. A separate branch insn tests the
7398: condition code and branches or not according its value. The branch insns
7399: come in distinct signed and unsigned flavors. Many common machines, such
7400: as the Vax, the 68000 and the 32000, work this way.
7401:
7402: Some machines have distinct signed and unsigned compare instructions, and
7403: only one set of conditional branch instructions. The easiest way to handle
7404: these machines is to treat them just like the others until the final stage
7405: where assembly code is written. At this time, when outputting code for the
7406: compare instruction, peek ahead at the following branch using
7407: @code{NEXT_INSN (insn)}. (The variable @code{insn} refers to the insn
7408: being output, in the output-writing code in an instruction pattern.) If
7409: the RTL says that is an unsigned branch, output an unsigned compare;
7410: otherwise output a signed compare. When the branch itself is output, you
7411: can treat signed and unsigned branches identically.
7412:
7413: The reason you can do this is that GNU CC always generates a pair of
7414: consecutive RTL insns, one to set the condition code and one to test it,
7415: and keeps the pair inviolate until the end.
7416:
7417: To go with this technique, you must define the machine-description macro
7418: @code{NOTICE_UPDATE_CC} to do @code{CC_STATUS_INIT}; in other words, no
7419: compare instruction is superfluous.
7420:
7421: Some machines have compare-and-branch instructions and no condition code.
7422: A similar technique works for them. When it is time to ``output'' a
7423: compare instruction, record its operands in two static variables. When
7424: outputting the branch-on-condition-code instruction that follows, actually
7425: output a compare-and-branch instruction that uses the remembered operands.
7426:
7427: It also works to define patterns for compare-and-branch instructions.
7428: In optimizing compilation, the pair of compare and branch instructions
1.1.1.5 root 7429: will be combined according to these patterns. But this does not happen
1.1 root 7430: if optimization is not requested. So you must use one of the solutions
7431: above in addition to any special patterns you define.
7432:
7433: @node Peephole Definitions, Expander Definitions, Jump Patterns, Machine Desc
7434: @section Defining Machine-Specific Peephole Optimizers
7435:
7436: In addition to instruction patterns the @file{md} file may contain
7437: definitions of machine-specific peephole optimizations.
7438:
7439: The combiner does not notice certain peephole optimizations when the data
7440: flow in the program does not suggest that it should try them. For example,
7441: sometimes two consecutive insns related in purpose can be combined even
7442: though the second one does not appear to use a register computed in the
7443: first one. A machine-specific peephole optimizer can detect such
7444: opportunities.
7445:
7446: A definition looks like this:
7447:
7448: @example
7449: (define_peephole
7450: [@var{insn-pattern-1}
7451: @var{insn-pattern-2}
7452: @dots{}]
7453: "@var{condition}"
7454: "@var{template}"
7455: "@var{machine-specific info}")
7456: @end example
7457:
7458: @noindent
7459: The last string operand may be omitted if you are not using any
7460: machine-specific information in this machine description. If present,
1.1.1.8 root 7461: it must obey the same rules as in a @code{define_insn}.
1.1 root 7462:
7463: In this skeleton, @var{insn-pattern-1} and so on are patterns to match
1.1.1.5 root 7464: consecutive insns. The optimization applies to a sequence of insns when
7465: @var{insn-pattern-1} matches the first one, @var{insn-pattern-2} matches
7466: the next, and so on.@refill
1.1 root 7467:
1.1.1.8 root 7468: Each of the insns matched by a peephole must also match a
7469: @code{define_insn}. Peepholes are checked only at the last stage just
7470: before code generation, and only optionally. Therefore, any insn which
7471: would match a peephole but no @code{define_insn} will cause a crash in code
7472: generation in an unoptimized compilation, or at various optimization
7473: stages.
1.1 root 7474:
1.1.1.5 root 7475: The operands of the insns are matched with @code{match_operands} and
7476: @code{match_dup}, as usual. What is not usual is that the operand numbers
7477: apply to all the insn patterns in the definition. So, you can check for
7478: identical operands in two insns by using @code{match_operand} in one insn
7479: and @code{match_dup} in the other.
1.1 root 7480:
7481: The operand constraints used in @code{match_operand} patterns do not have
1.1.1.8 root 7482: any direct effect on the applicability of the peephole, but they will
7483: be validated afterward, so make sure your constraints are general enough
7484: to apply whenever the peephole matches. If the peephole matches
7485: but the constraints are not satisfied, the compiler will crash.
7486:
7487: It is safe to omit constraints in all the operands of the peephole; or
7488: you can write constraints which serve as a double-check on the criteria
7489: previously tested.
1.1 root 7490:
1.1.1.5 root 7491: Once a sequence of insns matches the patterns, the @var{condition} is
7492: checked. This is a C expression which makes the final decision whether to
7493: perform the optimization (we do so if the expression is nonzero). If
1.1 root 7494: @var{condition} is omitted (in other words, the string is empty) then the
1.1.1.5 root 7495: optimization is applied to every sequence of insns that matches the
1.1 root 7496: patterns.
7497:
1.1.1.5 root 7498: The defined peephole optimizations are applied after register allocation
7499: is complete. Therefore, the peephole definition can check which
7500: operands have ended up in which kinds of registers, just by looking at
7501: the operands.
1.1 root 7502:
7503: The way to refer to the operands in @var{condition} is to write
7504: @code{operands[@var{i}]} for operand number @var{i} (as matched by
7505: @code{(match_operand @var{i} @dots{})}). Use the variable @code{insn} to
7506: refer to the last of the insns being matched; use @code{PREV_INSN} to find
1.1.1.8 root 7507: the preceding insns (but be careful to skip over any @code{note} insns that
1.1 root 7508: intervene).@refill
7509:
7510: When optimizing computations with intermediate results, you can use
7511: @var{condition} to match only when the intermediate results are not used
7512: elsewhere. Use the C expression @code{dead_or_set_p (@var{insn},
7513: @var{op})}, where @var{insn} is the insn in which you expect the value to
7514: be used for the last time (from the value of @code{insn}, together with use
7515: of @code{PREV_INSN}), and @var{op} is the intermediate value (from
7516: @code{operands[@var{i}]}).@refill
7517:
1.1.1.5 root 7518: Applying the optimization means replacing the sequence of insns with one
7519: new insn. The @var{template} controls ultimate output of assembler code
7520: for this combined insn. It works exactly like the template of a
7521: @code{define_insn}. Operand numbers in this template are the same ones
7522: used in matching the original sequence of insns.
1.1 root 7523:
7524: The result of a defined peephole optimizer does not need to match any of
1.1.1.5 root 7525: the insn patterns in the machine description; it does not even have an
7526: opportunity to match them. The peephole optimizer definition itself serves
7527: as the insn pattern to control how the insn is output.
7528:
7529: Defined peephole optimizers are run as assembler code is being output,
7530: so the insns they produce are never combined or rearranged in any way.
1.1 root 7531:
7532: Here is an example, taken from the 68000 machine description:
7533:
7534: @example
7535: (define_peephole
7536: [(set (reg:SI 15) (plus:SI (reg:SI 15) (const_int 4)))
7537: (set (match_operand:DF 0 "register_operand" "f")
7538: (match_operand:DF 1 "register_operand" "ad"))]
7539: "FP_REG_P (operands[0]) && ! FP_REG_P (operands[1])"
7540: "*
7541: @{
7542: rtx xoperands[2];
7543: xoperands[1] = gen_rtx (REG, SImode, REGNO (operands[1]) + 1);
7544: #ifdef MOTOROLA
7545: output_asm_insn (\"move.l %1,(sp)\", xoperands);
7546: output_asm_insn (\"move.l %1,-(sp)\", operands);
7547: return \"fmove.d (sp)+,%0\";
7548: #else
7549: output_asm_insn (\"movel %1,sp@@\", xoperands);
7550: output_asm_insn (\"movel %1,sp@@-\", operands);
7551: return \"fmoved sp@@+,%0\";
7552: #endif
7553: @}
7554: ")
7555: @end example
7556:
7557: The effect of this optimization is to change
7558:
7559: @example
7560: jbsr _foobar
7561: addql #4,sp
7562: movel d1,sp@@-
7563: movel d0,sp@@-
7564: fmoved sp@@+,fp0
7565: @end example
7566:
7567: @noindent
7568: into
7569:
7570: @example
7571: jbsr _foobar
7572: movel d1,sp@@
7573: movel d0,sp@@-
7574: fmoved sp@@+,fp0
7575: @end example
7576:
1.1.1.5 root 7577: @ignore
7578: If a peephole matches a sequence including one or more jump insns, you must
7579: take account of the flags such as @code{CC_REVERSED} which specify that the
7580: condition codes are represented in an unusual manner. The compiler
7581: automatically alters any ordinary conditional jumps which occur in such
7582: situations, but the compiler cannot alter jumps which have been replaced by
7583: peephole optimizations. So it is up to you to alter the assembler code
7584: that the peephole produces. Supply C code to write the assembler output,
7585: and in this C code check the condition code status flags and change the
7586: assembler code as appropriate.
7587: @end ignore
7588:
1.1.1.8 root 7589: @var{insn-pattern-1} and so on look @emph{almost} like the second
7590: operand of @code{define_insn}. There is one important difference: the
7591: second operand of @code{define_insn} consists of one or more RTX's
7592: enclosed in square brackets. Usually, there is only one: then the same
7593: action can be written as an element of a @code{define_peephole}. But
7594: when there are multiple actions in a @code{define_insn}, they are
7595: implicitly enclosed in a @code{parallel}. Then you must explicitly
7596: write the @code{parallel}, and the square brackets within it, in the
7597: @code{define_peephole}. Thus, if an insn pattern looks like this,
7598:
7599: @example
7600: (define_insn "divmodsi4"
7601: [(set (match_operand:SI 0 "general_operand" "=d")
7602: (div:SI (match_operand:SI 1 "general_operand" "0")
7603: (match_operand:SI 2 "general_operand" "dmsK")))
7604: (set (match_operand:SI 3 "general_operand" "=d")
7605: (mod:SI (match_dup 1) (match_dup 2)))]
7606: "TARGET_68020"
7607: "divsl%.l %2,%3:%0")
7608: @end example
7609:
7610: @noindent
7611: then the way to mention this insn in a peephole is as follows:
7612:
7613: @example
7614: (define_peephole
7615: [@dots{}
7616: (parallel
7617: [(set (match_operand:SI 0 "general_operand" "=d")
7618: (div:SI (match_operand:SI 1 "general_operand" "0")
7619: (match_operand:SI 2 "general_operand" "dmsK")))
7620: (set (match_operand:SI 3 "general_operand" "=d")
7621: (mod:SI (match_dup 1) (match_dup 2)))])
7622: @dots{}]
7623: @dots{})
7624: @end example
7625:
1.1 root 7626: @node Expander Definitions,, Peephole Definitions, Machine Desc
7627: @section Defining RTL Sequences for Code Generation
7628:
7629: On some target machines, some standard pattern names for RTL generation
7630: cannot be handled with single insn, but a sequence of RTL insns can
7631: represent them. For these target machines, you can write a
1.1.1.8 root 7632: @code{define_expand} to specify how to generate the sequence of RTL.
1.1 root 7633:
1.1.1.8 root 7634: A @code{define_expand} is an RTL expression that looks almost like a
7635: @code{define_insn}; but, unlike the latter, a @code{define_expand} is used
1.1 root 7636: only for RTL generation and it can produce more than one RTL insn.
7637:
1.1.1.8 root 7638: A @code{define_expand} RTX has four operands:
1.1 root 7639:
7640: @itemize @bullet
7641: @item
1.1.1.8 root 7642: The name. Each @code{define_expand} must have a name, since the only
1.1 root 7643: use for it is to refer to it by name.
7644:
7645: @item
7646: The RTL template. This is just like the RTL template for a
1.1.1.8 root 7647: @code{define_peephole} in that it is a vector of RTL expressions
1.1 root 7648: each being one insn.
7649:
7650: @item
7651: The condition, a string containing a C expression. This expression is
7652: used to express how the availability of this pattern depends on
7653: subclasses of target machine, selected by command-line options when
7654: GNU CC is run. This is just like the condition of a
1.1.1.8 root 7655: @code{define_insn} that has a standard name.
1.1 root 7656:
7657: @item
7658: The preparation statements, a string containing zero or more C
7659: statements which are to be executed before RTL code is generated from
7660: the RTL template.
7661:
7662: Usually these statements prepare temporary registers for use as
7663: internal operands in the RTL template, but they can also generate RTL
1.1.1.8 root 7664: insns directly by calling routines such as @code{emit_insn}, etc.
1.1 root 7665: Any such insns precede the ones that come from the RTL template.
7666: @end itemize
7667:
1.1.1.8 root 7668: Every RTL insn emitted by a @code{define_expand} must match some
7669: @code{define_insn} in the machine description. Otherwise, the compiler
7670: will crash when trying to generate code for the insn or trying to optimize
7671: it.
7672:
1.1 root 7673: The RTL template, in addition to controlling generation of RTL insns,
7674: also describes the operands that need to be specified when this pattern
7675: is used. In particular, it gives a predicate for each operand.
7676:
7677: A true operand, which need to be specified in order to generate RTL from
1.1.1.8 root 7678: the pattern, should be described with a @code{match_operand} in its first
1.1 root 7679: occurrence in the RTL template. This enters information on the operand's
7680: predicate into the tables that record such things. GNU CC uses the
7681: information to preload the operand into a register if that is required for
7682: valid RTL code. If the operand is referred to more than once, subsequent
1.1.1.8 root 7683: references should use @code{match_dup}.
1.1 root 7684:
7685: The RTL template may also refer to internal ``operands'' which are
7686: temporary registers or labels used only within the sequence made by the
1.1.1.8 root 7687: @code{define_expand}. Internal operands are substituted into the RTL
7688: template with @code{match_dup}, never with @code{match_operand}. The
1.1 root 7689: values of the internal operands are not passed in as arguments by the
7690: compiler when it requests use of this pattern. Instead, they are computed
7691: within the pattern, in the preparation statements. These statements
7692: compute the values and store them into the appropriate elements of
1.1.1.8 root 7693: @code{operands} so that @code{match_dup} can find them.
1.1 root 7694:
7695: There are two special macros defined for use in the preparation statements:
7696: @code{DONE} and @code{FAIL}. Use them with a following semicolon,
7697: as a statement.
7698:
7699: @table @code
7700: @item DONE
7701: Use the @code{DONE} macro to end RTL generation for the pattern. The
7702: only RTL insns resulting from the pattern on this occasion will be
7703: those already emitted by explicit calls to @code{emit_insn} within the
7704: preparation statements; the RTL template will not be generated.
7705:
7706: @item FAIL
7707: Make the pattern fail on this occasion. When a pattern fails, it means
7708: that the pattern was not truly available. The calling routines in the
7709: compiler will try other strategies for code generation using other patterns.
7710:
7711: Failure is currently supported only for binary operations (addition,
7712: multiplication, shifting, etc.).
7713:
7714: Do not emit any insns explicitly with @code{emit_insn} before failing.
7715: @end table
7716:
7717: Here is an example, the definition of left-shift for the SPUR chip:
7718:
7719: @example
7720: (define_expand "ashlsi3"
7721: [(set (match_operand:SI 0 "register_operand" "")
7722: (ashift:SI
7723: (match_operand:SI 1 "register_operand" "")
7724: (match_operand:SI 2 "nonmemory_operand" "")))]
7725: ""
7726: "
7727: @{
7728: if (GET_CODE (operands[2]) != CONST_INT
7729: || (unsigned) INTVAL (operands[2]) > 3)
7730: FAIL;
7731: @}")
7732: @end example
7733:
7734: @noindent
1.1.1.8 root 7735: This example uses @code{define_expand} so that it can generate an RTL insn
1.1 root 7736: for shifting when the shift-count is in the supported range of 0 to 3 but
7737: fail in other cases where machine insns aren't available. When it fails,
7738: the compiler tries another strategy using different patterns (such as, a
7739: library call).
7740:
7741: If the compiler were able to handle nontrivial condition-strings in
1.1.1.8 root 7742: patterns with names, then it would be possible to use a
7743: @code{define_insn} in that case. Here is another case (zero-extension
7744: on the 68000) which makes more use of the power of @code{define_expand}:
1.1 root 7745:
7746: @example
7747: (define_expand "zero_extendhisi2"
7748: [(set (match_operand:SI 0 "general_operand" "")
7749: (const_int 0))
7750: (set (strict_low_part
7751: (subreg:HI
1.1.1.8 root 7752: (match_dup 0)
1.1 root 7753: 0))
7754: (match_operand:HI 1 "general_operand" ""))]
7755: ""
7756: "operands[1] = make_safe_from (operands[1], operands[0]);")
7757: @end example
7758:
7759: @noindent
7760: Here two RTL insns are generated, one to clear the entire output operand
7761: and the other to copy the input operand into its low half. This sequence
7762: is incorrect if the input operand refers to [the old value of] the output
7763: operand, so the preparation statement makes sure this isn't so. The
7764: function @code{make_safe_from} copies the @code{operands[1]} into a
7765: temporary register if it refers to @code{operands[0]}. It does this
7766: by emitting another RTL insn.
7767:
7768: Finally, a third example shows the use of an internal operand.
1.1.1.8 root 7769: Zero-extension on the SPUR chip is done by @code{and}-ing the result
1.1 root 7770: against a halfword mask. But this mask cannot be represented by a
1.1.1.8 root 7771: @code{const_int} because the constant value is too large to be legitimate
1.1 root 7772: on this machine. So it must be copied into a register with
1.1.1.8 root 7773: @code{force_reg} and then the register used in the @code{and}.
1.1 root 7774:
7775: @example
7776: (define_expand "zero_extendhisi2"
7777: [(set (match_operand:SI 0 "register_operand" "")
7778: (and:SI (subreg:SI
7779: (match_operand:HI 1 "register_operand" "")
7780: 0)
7781: (match_dup 2)))]
7782: ""
7783: "operands[2]
7784: = force_reg (SImode, gen_rtx (CONST_INT,
7785: VOIDmode, 65535)); ")
7786: @end example
7787:
1.1.1.8 root 7788: @strong{Note:} If the @code{define_expand} is used to serve a standard
7789: binary or unary arithmetic operation, then the last insn it generates
7790: must not be a @code{code_label}, @code{barrier} or @code{note}. It must
7791: be an @code{insn}, @code{jump_insn} or @code{call_insn}.
7792:
1.1 root 7793: @node Machine Macros, Config, Machine Desc, Top
7794: @chapter Machine Description Macros
7795:
7796: The other half of the machine description is a C header file conventionally
7797: given the name @file{tm-@var{machine}.h}. The file @file{tm.h} should be a
7798: link to it. The header file @file{config.h} includes @file{tm.h} and most
7799: compiler source files include @file{config.h}.
7800:
7801: @menu
1.1.1.9 root 7802: * Run-time Target:: Defining @samp{-m} options like @samp{-m68000} and @samp{-m68020}.
1.1 root 7803: * Storage Layout:: Defining sizes and alignments of data types.
7804: * Registers:: Naming and describing the hardware registers.
7805: * Register Classes:: Defining the classes of hardware registers.
7806: * Stack Layout:: Defining which way the stack grows and by how much.
7807: * Library Names:: Specifying names of subroutines to call automatically.
7808: * Addressing Modes:: Defining addressing modes valid for memory operands.
1.1.1.8 root 7809: * Delayed Branch:: Do branches execute the following instruction?
1.1 root 7810: * Condition Code:: Defining how insns update the condition code.
7811: * Assembler Format:: Defining how to write insns and pseudo-ops to output.
1.1.1.5 root 7812: * Cross-compilation:: Handling floating point for cross-compilers.
1.1 root 7813: * Misc:: Everything else.
7814: @end menu
7815:
7816: @node Run-time Target, Storage Layout, Machine Macros, Machine Macros
7817: @section Run-time Target Specification
7818:
7819: @table @code
7820: @item CPP_PREDEFINES
7821: Define this to be a string constant containing @samp{-D} options to
7822: define the predefined macros that identify this machine and system.
7823: These macros will be predefined unless the @samp{-ansi} option is
7824: specified.
7825:
1.1.1.4 root 7826: In addition, a parallel set of macros are predefined, whose names are
7827: made by appending @samp{__} at the beginning and at the end. These
7828: @samp{__} macros are permitted by the ANSI standard, so they are
7829: predefined regardless of whether @samp{-ansi} is specified.
7830:
7831: For example, on the Sun, one can use the following value:
1.1 root 7832:
7833: @example
7834: "-Dmc68000 -Dsun -Dunix"
7835: @end example
7836:
1.1.1.8 root 7837: The result is to define the macros @code{__mc68000__}, @code{__sun__}
7838: and @code{__unix__} unconditionally, and the macros @code{mc68000},
7839: @code{sun} and @code{unix} provided @samp{-ansi} is not specified.
1.1.1.4 root 7840:
1.1 root 7841: @item CPP_SPEC
7842: A C string constant that tells the GNU CC driver program options to
7843: pass to CPP. It can also specify how to translate options you
7844: give to GNU CC into options for GNU CC to pass to the CPP.
7845:
7846: Do not define this macro if it does not need to do anything.
7847:
7848: @item CC1_SPEC
7849: A C string constant that tells the GNU CC driver program options to
7850: pass to CC1. It can also specify how to translate options you
7851: give to GNU CC into options for GNU CC to pass to the CC1.
7852:
7853: Do not define this macro if it does not need to do anything.
7854:
7855: @item extern int target_flags;
7856: This declaration should be present.
7857:
7858: @item TARGET_@dots{}
7859: This series of macros is to allow compiler command arguments to
7860: enable or disable the use of optional features of the target machine.
7861: For example, one machine description serves both the 68000 and
7862: the 68020; a command argument tells the compiler whether it should
7863: use 68020-only instructions or not. This command argument works
7864: by means of a macro @code{TARGET_68020} that tests a bit in
7865: @code{target_flags}.
7866:
7867: Define a macro @code{TARGET_@var{featurename}} for each such option.
7868: Its definition should test a bit in @code{target_flags}; for example:
7869:
7870: @example
7871: #define TARGET_68020 (target_flags & 1)
7872: @end example
7873:
7874: One place where these macros are used is in the condition-expressions
7875: of instruction patterns. Note how @code{TARGET_68020} appears
7876: frequently in the 68000 machine description file, @file{m68k.md}.
7877: Another place they are used is in the definitions of the other
7878: macros in the @file{tm-@var{machine}.h} file.
7879:
7880: @item TARGET_SWITCHES
7881: This macro defines names of command options to set and clear
7882: bits in @code{target_flags}. Its definition is an initializer
7883: with a subgrouping for each command option.
7884:
7885: Each subgrouping contains a string constant, that defines the option
7886: name, and a number, which contains the bits to set in
7887: @code{target_flags}. A negative number says to clear bits instead;
7888: the negative of the number is which bits to clear. The actual option
7889: name is made by appending @samp{-m} to the specified name.
7890:
7891: One of the subgroupings should have a null string. The number in
7892: this grouping is the default value for @code{target_flags}. Any
7893: target options act starting with that value.
7894:
7895: Here is an example which defines @samp{-m68000} and @samp{-m68020}
7896: with opposite meanings, and picks the latter as the default:
7897:
7898: @example
7899: #define TARGET_SWITCHES \
7900: @{ @{ "68020", 1@}, \
7901: @{ "68000", -1@}, \
7902: @{ "", 1@}@}
7903: @end example
7904:
7905: @item OVERRIDE_OPTIONS
7906: Sometimes certain combinations of command options do not make sense on
7907: a particular target machine. You can define a macro
7908: @code{OVERRIDE_OPTIONS} to take account of this. This macro, if
7909: defined, is executed once just after all the command options have been
7910: parsed.
7911: @end table
7912:
7913: @node Storage Layout, Registers, Run-time Target, Machine Macros
7914: @section Storage Layout
7915:
7916: Note that the definitions of the macros in this table which are sizes or
7917: alignments measured in bits do not need to be constant. They can be C
7918: expressions that refer to static variables, such as the @code{target_flags}.
7919: @xref{Run-time Target}.
7920:
7921: @table @code
7922: @item BITS_BIG_ENDIAN
7923: Define this macro if the most significant bit in a byte has the lowest
7924: number. This means that bit-field instructions count from the most
7925: significant bit. If the machine has no bit-field instructions, this
7926: macro is irrelevant.
7927:
1.1.1.8 root 7928: This macro does not affect the way structure fields are packed into
7929: bytes or words; that is controlled by @code{BYTES_BIG_ENDIAN}.
7930:
1.1 root 7931: @item BYTES_BIG_ENDIAN
7932: Define this macro if the most significant byte in a word has the
7933: lowest number.
7934:
7935: @item WORDS_BIG_ENDIAN
7936: Define this macro if, in a multiword object, the most significant
7937: word has the lowest number.
7938:
7939: @item BITS_PER_UNIT
7940: Number of bits in an addressable storage unit (byte); normally 8.
7941:
7942: @item BITS_PER_WORD
7943: Number of bits in a word; normally 32.
7944:
7945: @item UNITS_PER_WORD
7946: Number of storage units in a word; normally 4.
7947:
7948: @item POINTER_SIZE
7949: Width of a pointer, in bits.
7950:
7951: @item POINTER_BOUNDARY
7952: Alignment required for pointers stored in memory, in bits.
7953:
7954: @item PARM_BOUNDARY
1.1.1.7 root 7955: Normal alignment required for function parameters on the stack, in
7956: bits. All stack parameters receive least this much alignment
7957: regardless of data type. On most machines, this is the same as the
7958: size of an integer.
7959:
7960: @item MAX_PARM_BOUNDARY
7961: Largest alignment required for any stack parameters, in bits. If the
7962: data type of the parameter calls for more alignment than
7963: @code{PARM_BOUNDARY}, then it is given extra padding up to this limit.
7964:
7965: Don't define this macro if it would be equal to @code{PARM_BOUNDARY};
7966: in other words, if the alignment of a stack parameter should not
7967: depend on its data type (as is the case on most machines).
1.1 root 7968:
7969: @item STACK_BOUNDARY
7970: Define this macro if you wish to preserve a certain alignment for
7971: the stack pointer at all times. The definition is a C expression
7972: for the desired alignment (measured in bits).
7973:
7974: @item FUNCTION_BOUNDARY
7975: Alignment required for a function entry point, in bits.
7976:
7977: @item BIGGEST_ALIGNMENT
7978: Biggest alignment that any data type can require on this machine, in bits.
7979:
1.1.1.8 root 7980: @item CONSTANT_ALIGNMENT (@var{code}, @var{typealign})
7981: A C expression to compute the alignment for a constant. The argument
7982: @var{typealign} is the alignment required for the constant's data type.
7983: @var{code} is the tree code of the constant itself.
7984:
7985: If this macro is not defined, the default is to use @var{typealign}. If
7986: you do define this macro, the value must be a multiple of
7987: @var{typealign}.
7988:
7989: The purpose of defining this macro is usually to cause string constants
7990: to be word aligned so that @file{dhrystone} can be made to run faster.
7991:
1.1 root 7992: @item EMPTY_FIELD_BOUNDARY
7993: Alignment in bits to be given to a structure bit field that follows an
7994: empty field such as @code{int : 0;}.
7995:
7996: @item STRUCTURE_SIZE_BOUNDARY
7997: Number of bits which any structure or union's size must be a multiple of.
7998: Each structure or union's size is rounded up to a multiple of this.
7999:
8000: If you do not define this macro, the default is the same as
8001: @code{BITS_PER_UNIT}.
8002:
8003: @item STRICT_ALIGNMENT
8004: Define this if instructions will fail to work if given data not
8005: on the nominal alignment. If instructions will merely go slower
8006: in that case, do not define this macro.
8007:
8008: @item PCC_BITFIELD_TYPE_MATTERS
8009: Define this if you wish to imitate a certain bizarre behavior pattern
8010: of some instances of PCC: a bit field whose declared type is
8011: @code{int} has the same effect on the size and alignment of a
8012: structure as an actual @code{int} would have.
8013:
1.1.1.10! root 8014: If the macro is defined, then its definition should be a C expression;
! 8015: a nonzero value for the expression enables PCC-compatible behavior.
! 8016:
1.1 root 8017: Just what effect that is in GNU CC depends on other parameters, but on
8018: most machines it would force the structure's alignment and size to a
8019: multiple of 32 or @code{BIGGEST_ALIGNMENT} bits.
8020:
1.1.1.7 root 8021: @item MAX_FIXED_MODE_SIZE
8022: An integer expression for the largest integer machine mode that should
8023: actually be used. All integer machine modes of this size or smaller
8024: can be used for structures and unions with the appropriate sizes.
8025:
1.1 root 8026: @item CHECK_FLOAT_VALUE (@var{mode}, @var{value})
8027: A C statement to validate the value @var{value} (or type
8028: @code{double}) for mode @var{mode}. This means that you check whether
8029: @var{value} fits within the possible range of values for mode
8030: @var{mode} on this target machine. The mode @var{mode} is always
8031: @code{SFmode} or @code{DFmode}.
8032:
8033: If @var{value} is not valid, you should call @code{error} to print an
8034: error message and then assign some valid value to @var{value}.
8035: Allowing an invalid value to go through the compiler can produce
8036: incorrect assembler code which may even cause Unix assemblers to
8037: crash.
8038:
8039: This macro need not be defined if there is no work for it to do.
8040: @end table
8041:
8042: @node Registers, Register Classes, Storage Layout, Machine Macros
8043: @section Register Usage
8044:
8045: @table @code
8046: @item FIRST_PSEUDO_REGISTER
8047: Number of hardware registers known to the compiler. They receive
8048: numbers 0 through @code{FIRST_PSEUDO_REGISTER-1}; thus, the first
8049: pseudo register's number really is assigned the number
8050: @code{FIRST_PSEUDO_REGISTER}.
8051:
8052: @item FIXED_REGISTERS
8053: An initializer that says which registers are used for fixed purposes
8054: all throughout the compiled code and are therefore not available for
8055: general allocation. These would include the stack pointer, the frame
8056: pointer (except on machines where that can be used as a general
8057: register when no frame pointer is needed), the program counter on
8058: machines where that is considered one of the addressable registers,
8059: and any other numbered register with a standard use.
8060:
8061: This information is expressed as a sequence of numbers, separated by
8062: commas and surrounded by braces. The @var{n}th number is 1 if
8063: register @var{n} is fixed, 0 otherwise.
8064:
8065: The table initialized from this macro, and the table initialized by
8066: the following one, may be overridden at run time either automatically,
8067: by the actions of the macro @code{CONDITIONAL_REGISTER_USAGE}, or by
8068: the user with the command options @samp{-ffixed-@var{reg}},
8069: @samp{-fcall-used-@var{reg}} and @samp{-fcall-saved-@var{reg}}.
8070:
8071: @item CALL_USED_REGISTERS
8072: Like @code{FIXED_REGISTERS} but has 1 for each register that is
8073: clobbered (in general) by function calls as well as for fixed
8074: registers. This macro therefore identifies the registers that are not
8075: available for general allocation of values that must live across
8076: function calls.
8077:
8078: If a register has 0 in @code{CALL_USED_REGISTERS}, the compiler
8079: automatically saves it on function entry and restores it on function
8080: exit, if the register is used within the function.
8081:
1.1.1.6 root 8082: @item DEFAULT_CALLER_SAVES
1.1.1.8 root 8083: Define this macro if function calls on the target machine do not preserve
1.1.1.6 root 8084: any registers; in other words, if @code{CALL_USED_REGISTERS} has 1
8085: for all registers. This macro enables @samp{-fcaller-saves} by default.
8086: Eventually that option will be enabled by default on all machines and both
8087: the option and this macro will be eliminated.
8088:
1.1 root 8089: @item CONDITIONAL_REGISTER_USAGE
8090: Zero or more C statements that may conditionally modify two variables
8091: @code{fixed_regs} and @code{call_used_regs} (both of type @code{char
8092: []}) after they have been initialized from the two preceding macros.
8093:
8094: This is necessary in case the fixed or call-clobbered registers depend
8095: on target flags.
8096:
8097: You need not define this macro if it has no work to do.
8098:
8099: If the usage of an entire class of registers depends on the target
1.1.1.5 root 8100: flags, you may indicate this to GCC by using this macro to modify
1.1 root 8101: @code{fixed_regs} and @code{call_used_regs} to 1 for each of the
1.1.1.5 root 8102: registers in the classes which should not be used by GCC. Also define
1.1 root 8103: the macro @code{REG_CLASS_FROM_LETTER} to return @code{NO_REGS} if it
8104: is called with a letter for a class that shouldn't be used.
8105:
8106: (However, if this class is not included in @code{GENERAL_REGS} and all
8107: of the insn patterns whose constraints permit this class are
8108: controlled by target switches, then GCC will automatically avoid using
8109: these registers when the target switches are opposed to them.)
8110:
8111: @item OVERLAPPING_REGNO_P (@var{regno})
1.1.1.5 root 8112: If defined, this is a C expression whose value is nonzero if hard
8113: register number @var{regno} is an overlapping register. This means a
8114: hard register which overlaps a hard register with a different number.
8115: (Such overlap is undesirable, but occasionally it allows a machine to
8116: be supported which otherwise could not be.) This macro must return
8117: nonzero for @emph{all} the registers which overlap each other. GNU CC
8118: can use an overlapping register only in certain limited ways. It can
8119: be used for allocation within a basic block, and may be spilled for
8120: reloading; that is all.
1.1 root 8121:
8122: If this macro is not defined, it means that none of the hard registers
8123: overlap each other. This is the usual situation.
8124:
8125: @item INSN_CLOBBERS_REGNO_P (@var{insn}, @var{regno})
8126: If defined, this is a C expression whose value should be nonzero if
8127: the insn @var{insn} has the effect of mysteriously clobbering the
8128: contents of hard register number @var{regno}. By ``mysterious'' we
8129: mean that the insn's RTL expression doesn't describe such an effect.
8130:
8131: If this macro is not defined, it means that no insn clobbers registers
8132: mysteriously. This is the usual situation; all else being equal,
8133: it is best for the RTL expression to show all the activity.
8134:
8135: @item PRESERVE_DEATH_INFO_REGNO_P (@var{regno})
8136: If defined, this is a C expression whose value is nonzero if accurate
8137: @code{REG_DEAD} notes are needed for hard register number @var{regno}
8138: at the time of outputting the assembler code. When this is so, a few
8139: optimizations that take place after register allocation and could
8140: invalidate the death notes are not done when this register is
8141: involved.
8142:
1.1.1.8 root 8143: You would arrange to preserve death info for a register when some of the
8144: code in the machine description which is executed to write the assembler
8145: code looks at the death notes. This is necessary only when the actual
8146: hardware feature which GNU CC thinks of as a register is not actually a
8147: register of the usual sort. (It might, for example, be a hardware
8148: stack.)
1.1 root 8149:
8150: If this macro is not defined, it means that no death notes need to be
8151: preserved. This is the usual situation.
8152:
1.1.1.10! root 8153: @item HARD_REGNO_NREGS (@var{regno}, @var{mode})
1.1 root 8154: A C expression for the number of consecutive hard registers, starting
8155: at register number @var{regno}, required to hold a value of mode
8156: @var{mode}.
8157:
8158: On a machine where all registers are exactly one word, a suitable
8159: definition of this macro is
8160:
8161: @example
8162: #define HARD_REGNO_NREGS(REGNO, MODE) \
8163: ((GET_MODE_SIZE (MODE) + UNITS_PER_WORD - 1) \
8164: / UNITS_PER_WORD))
8165: @end example
8166:
8167: @item HARD_REGNO_MODE_OK (@var{regno}, @var{mode})
8168: A C expression that is nonzero if it is permissible to store a value
8169: of mode @var{mode} in hard register number @var{regno} (or in several
8170: registers starting with that one). For a machine where all registers
8171: are equivalent, a suitable definition is
8172:
8173: @example
8174: #define HARD_REGNO_MODE_OK(REGNO, MODE) 1
8175: @end example
8176:
1.1.1.8 root 8177: It is not necessary for this macro to check for the numbers of fixed
8178: registers, because the allocation mechanism considers them to be always
8179: occupied.
8180:
8181: On some machines, double-precision values must be kept in even/odd
8182: register pairs. The way to implement that is to define this macro
8183: to reject odd register numbers for such modes.
8184:
8185: GNU CC assumes that it can always move values between registers and
8186: (suitably addressed) memory locations. If it is impossible to move a
8187: value of a certain mode between memory and certain registers, then
8188: @code{HARD_REGNO_MODE_OK} must not allow this mode in those registers.
1.1 root 8189:
8190: Many machines have special registers for floating point arithmetic.
8191: Often people assume that floating point machine modes are allowed only
8192: in floating point registers. This is not true. Any registers that
8193: can hold integers can safely @emph{hold} a floating point machine
8194: mode, whether or not floating arithmetic can be done on it in those
8195: registers.
8196:
1.1.1.9 root 8197: On some machines, though, the converse is true: fixed-point machine
8198: modes may not go in floating registers. This is true if the floating
8199: registers normalize any value stored in them, because storing a
8200: non-floating value there would garble it. In this case,
8201: @code{HARD_REGNO_MODE_OK} should reject fixed-point machine modes in
8202: floating registers. But if the floating registers do not automatically
8203: normalize, if you can store any bit pattern in one and retrieve it
8204: unchanged without a trap, then any machine mode may go in a floating
8205: register and this macro should say so.
8206:
8207: The primary significance of special floating registers is rather that
8208: they are the registers acceptable in floating point arithmetic
8209: instructions. However, this is of no concern to
8210: @code{HARD_REGNO_MODE_OK}. You handle it by writing the proper
8211: constraints for those instructions.
8212:
8213: On some machines, the floating registers are especially slow to access,
8214: so that it is better to store a value in a stack frame than in such a
8215: register if floating point arithmetic is not being done. As long as the
8216: floating registers are not in class @code{GENERAL_REGS}, they will not
8217: be used unless some insn's constraint asks for one.
1.1 root 8218:
8219: @item MODES_TIEABLE_P (@var{mode1}, @var{mode2})
8220: A C expression that is nonzero if it is desirable to choose register
8221: allocation so as to avoid move instructions between a value of mode
8222: @var{mode1} and a value of mode @var{mode2}.
8223:
8224: If @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode1})} and
8225: @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode2})} are ever different
8226: for any @var{r}, then @code{MODES_TIEABLE_P (@var{mode1},
8227: @var{mode2})} must be zero.
8228:
8229: @item PC_REGNUM
8230: If the program counter has a register number, define this as that
8231: register number. Otherwise, do not define it.
8232:
8233: @item STACK_POINTER_REGNUM
8234: The register number of the stack pointer register, which must also be
8235: a fixed register according to @code{FIXED_REGISTERS}. On many
8236: machines, the hardware determines which register this is.
8237:
8238: @item FRAME_POINTER_REGNUM
8239: The register number of the frame pointer register, which is used to
8240: access automatic variables in the stack frame. On some machines, the
8241: hardware determines which register this is. On other machines, you
8242: can choose any register you wish for this purpose.
8243:
8244: @item FRAME_POINTER_REQUIRED
1.1.1.9 root 8245: A C expression which is nonzero if a function must have and use a frame
8246: pointer. This expression is evaluated twice: at the beginning of
8247: generating RTL, and in the reload pass. If its value is nonzero at
8248: either time, then the function will have a frame pointer.
8249:
8250: The expression can in principle examine the current function and decide
8251: according to the facts, but on most machines the constant 0 or the
8252: constant 1 suffices. Use 0 when the machine allows code to be generated
8253: with no frame pointer, and doing so saves some time or space. Use 1
8254: when there is no possible advantage to avoiding a frame pointer.
1.1 root 8255:
1.1.1.5 root 8256: In certain cases, the compiler does not know how to produce valid code
8257: without a frame pointer. The compiler recognizes those cases and
8258: automatically gives the function a frame pointer regardless of what
1.1 root 8259: @code{FRAME_POINTER_REQUIRED} says. You don't need to worry about
8260: them.@refill
8261:
8262: In a function that does not require a frame pointer, the frame pointer
8263: register can be allocated for ordinary usage, unless you mark it as a
8264: fixed register. See @code{FIXED_REGISTERS} for more information.
8265:
8266: @item ARG_POINTER_REGNUM
8267: The register number of the arg pointer register, which is used to
8268: access the function's argument list. On some machines, this is the
8269: same as the frame pointer register. On some machines, the hardware
8270: determines which register this is. On other machines, you can choose
8271: any register you wish for this purpose. If this is not the same
8272: register as the frame pointer register, then you must mark it as a
8273: fixed register according to @code{FIXED_REGISTERS}.
8274:
8275: @item STATIC_CHAIN_REGNUM
8276: The register number used for passing a function's static chain
8277: pointer. This is needed for languages such as Pascal and Algol where
8278: functions defined within other functions can access the local
8279: variables of the outer functions; it is not currently used because C
8280: does not provide this feature, but you must define the macro.
8281:
8282: The static chain register need not be a fixed register.
8283:
8284: @item STRUCT_VALUE_REGNUM
8285: When a function's value's mode is @code{BLKmode}, the value is not
8286: returned according to @code{FUNCTION_VALUE}. Instead, the caller
8287: passes the address of a block of memory in which the value should be
8288: stored.
8289:
8290: If this value is passed in a register, then @code{STRUCT_VALUE_REGNUM}
8291: should be the number of that register.
8292:
8293: @item STRUCT_VALUE
8294: If the structure value address is not passed in a register, define
8295: @code{STRUCT_VALUE} as an expression returning an RTX for the place
1.1.1.8 root 8296: where the address is passed. If it returns a @code{mem} RTX, the
1.1 root 8297: address is passed as an ``invisible'' first argument.
8298:
8299: @item STRUCT_VALUE_INCOMING_REGNUM
8300: On some architectures the place where the structure value address
8301: is found by the called function is not the same place that the
8302: caller put it. This can be due to register windows, or it could
8303: be because the function prologue moves it to a different place.
8304:
8305: If the incoming location of the structure value address is in a
8306: register, define this macro as the register number.
8307:
8308: @item STRUCT_VALUE_INCOMING
8309: If the incoming location is not a register, define
8310: @code{STRUCT_VALUE_INCOMING} as an expression for an RTX for where the
8311: called function should find the value. If it should find the value on
1.1.1.8 root 8312: the stack, define this to create a @code{mem} which refers to the
8313: frame pointer. If the value is a @code{mem}, the compiler assumes it
1.1 root 8314: is for an invisible first argument, and leaves space for it when
8315: finding the first real argument.
8316:
8317: @item REG_ALLOC_ORDER
8318: If defined, an initializer for a vector of integers, containing the
8319: numbers of hard registers in the order in which the GNU CC should
8320: prefer to use them (from most preferred to least).
8321:
8322: If this macro is not defined, registers are used lowest numbered first
8323: (all else being equal).
8324:
8325: One use of this macro is on the 360, where the highest numbered
8326: registers must always be saved and the save-multiple-registers
8327: instruction supports only sequences of consecutive registers. This
8328: macro is defined to cause the highest numbered allocatable registers
8329: to be used first.
8330: @end table
8331:
8332: @node Register Classes, Stack Layout, Registers, Machine Macros
8333: @section Register Classes
8334:
8335: On many machines, the numbered registers are not all equivalent.
8336: For example, certain registers may not be allowed for indexed addressing;
8337: certain registers may not be allowed in some instructions. These machine
8338: restrictions are described to the compiler using @dfn{register classes}.
8339:
8340: You define a number of register classes, giving each one a name and saying
8341: which of the registers belong to it. Then you can specify register classes
8342: that are allowed as operands to particular instruction patterns.
8343:
8344: In general, each register will belong to several classes. In fact, one
8345: class must be named @code{ALL_REGS} and contain all the registers. Another
8346: class must be named @code{NO_REGS} and contain no registers. Often the
8347: union of two classes will be another class; however, this is not required.
8348:
8349: One of the classes must be named @code{GENERAL_REGS}. There is nothing
8350: terribly special about the name, but the operand constraint letters
8351: @samp{r} and @samp{g} specify this class. If @code{GENERAL_REGS} is
8352: the same as @code{ALL_REGS}, just define it as a macro which expands
8353: to @code{ALL_REGS}.
8354:
8355: The way classes other than @code{GENERAL_REGS} are specified in operand
8356: constraints is through machine-dependent operand constraint letters.
8357: You can define such letters to correspond to various classes, then use
8358: them in operand constraints.
8359:
8360: You should define a class for the union of two classes whenever some
8361: instruction allows both classes. For example, if an instruction allows
8362: either a floating-point (coprocessor) register or a general register for a
8363: certain operand, you should define a class @code{FLOAT_OR_GENERAL_REGS}
8364: which includes both of them. Otherwise you will get suboptimal code.
8365:
8366: You must also specify certain redundant information about the register
8367: classes: for each class, which classes contain it and which ones are
8368: contained in it; for each pair of classes, the largest class contained
8369: in their union.
8370:
1.1.1.8 root 8371: When a value occupying several consecutive registers is expected in a
8372: certain class, all the registers used must belong to that class.
8373: Therefore, register classes cannot be used to enforce a requirement for
8374: a register pair to start with an even-numbered register. The way to
8375: specify this requirement is with @code{HARD_REGNO_MODE_OK}.
8376:
1.1 root 8377: Register classes used for input-operands of bitwise-and or shift
8378: instructions have a special requirement: each such class must have, for
8379: each fixed-point machine mode, a subclass whose registers can transfer that
8380: mode to or from memory. For example, on some machines, the operations for
8381: single-byte values (@code{QImode}) are limited to certain registers. When
8382: this is so, each register class that is used in a bitwise-and or shift
8383: instruction must have a subclass consisting of registers from which
8384: single-byte values can be loaded or stored. This is so that
8385: @code{PREFERRED_RELOAD_CLASS} can always have a possible value to return.
8386:
8387: @table @code
8388: @item enum reg_class
8389: An enumeral type that must be defined with all the register class names
8390: as enumeral values. @code{NO_REGS} must be first. @code{ALL_REGS}
8391: must be the last register class, followed by one more enumeral value,
8392: @code{LIM_REG_CLASSES}, which is not a register class but rather
8393: tells how many classes there are.
8394:
8395: Each register class has a number, which is the value of casting
8396: the class name to type @code{int}. The number serves as an index
8397: in many of the tables described below.
8398:
8399: @item N_REG_CLASSES
8400: The number of distinct register classes, defined as follows:
8401:
8402: @example
8403: #define N_REG_CLASSES (int) LIM_REG_CLASSES
8404: @end example
8405:
8406: @item REG_CLASS_NAMES
8407: An initializer containing the names of the register classes as C string
8408: constants. These names are used in writing some of the debugging dumps.
8409:
8410: @item REG_CLASS_CONTENTS
8411: An initializer containing the contents of the register classes, as integers
8412: which are bit masks. The @var{n}th integer specifies the contents of class
8413: @var{n}. The way the integer @var{mask} is interpreted is that
8414: register @var{r} is in the class if @code{@var{mask} & (1 << @var{r})} is 1.
8415:
8416: When the machine has more than 32 registers, an integer does not suffice.
8417: Then the integers are replaced by sub-initializers, braced groupings containing
8418: several integers. Each sub-initializer must be suitable as an initializer
8419: for the type @code{HARD_REG_SET} which is defined in @file{hard-reg-set.h}.
8420:
8421: @item REGNO_REG_CLASS (@var{regno})
8422: A C expression whose value is a register class containing hard register
8423: @var{regno}. In general there is more that one such class; choose a class
8424: which is @dfn{minimal}, meaning that no smaller class also contains the
8425: register.
8426:
8427: @item BASE_REG_CLASS
8428: A macro whose definition is the name of the class to which a valid
8429: base register must belong. A base register is one used in an address
8430: which is the register value plus a displacement.
8431:
8432: @item INDEX_REG_CLASS
8433: A macro whose definition is the name of the class to which a valid
8434: index register must belong. An index register is one used in an
8435: address where its value is either multiplied by a scale factor or
8436: added to another register (as well as added to a displacement).
8437:
8438: @item REG_CLASS_FROM_LETTER (@var{char})
8439: A C expression which defines the machine-dependent operand constraint
8440: letters for register classes. If @var{char} is such a letter, the
8441: value should be the register class corresponding to it. Otherwise,
8442: the value should be @code{NO_REGS}.
8443:
8444: @item REGNO_OK_FOR_BASE_P (@var{num})
8445: A C expression which is nonzero if register number @var{num} is
8446: suitable for use as a base register in operand addresses. It may be
8447: either a suitable hard register or a pseudo register that has been
8448: allocated such a hard register.
8449:
8450: @item REGNO_OK_FOR_INDEX_P (@var{num})
8451: A C expression which is nonzero if register number @var{num} is
8452: suitable for use as an index register in operand addresses. It may be
8453: either a suitable hard register or a pseudo register that has been
8454: allocated such a hard register.
8455:
8456: The difference between an index register and a base register is that
8457: the index register may be scaled. If an address involves the sum of
8458: two registers, neither one of them scaled, then either one may be
8459: labeled the ``base'' and the other the ``index''; but whichever
8460: labeling is used must fit the machine's constraints of which registers
8461: may serve in each capacity. The compiler will try both labelings,
8462: looking for one that is valid, and will reload one or both registers
8463: only if neither labeling works.
8464:
8465: @item PREFERRED_RELOAD_CLASS (@var{x}, @var{class})
8466: A C expression that places additional restrictions on the register class
8467: to use when it is necessary to copy value @var{x} into a register in class
8468: @var{class}. The value is a register class; perhaps @var{class}, or perhaps
8469: another, smaller class. On many machines, the definition
8470:
8471: @example
8472: #define PREFERRED_RELOAD_CLASS(X,CLASS) CLASS
8473: @end example
8474:
8475: @noindent
8476: is safe.
8477:
8478: Sometimes returning a more restrictive class makes better code. For
8479: example, on the 68000, when @var{x} is an integer constant that is in range
8480: for a @samp{moveq} instruction, the value of this macro is always
8481: @code{DATA_REGS} as long as @var{class} includes the data registers.
8482: Requiring a data register guarantees that a @samp{moveq} will be used.
8483:
1.1.1.8 root 8484: If @var{x} is a @code{const_double}, by returning @code{NO_REGS}
1.1 root 8485: you can force @var{x} into a memory constant. This is useful on
8486: certain machines where immediate floating values cannot be loaded into
8487: certain kinds of registers.
8488:
8489: In a shift instruction or a bitwise-and instruction, the mode of @var{x},
8490: the value being reloaded, may not be the same as the mode of the
8491: instruction's operand. (They will both be fixed-point modes, however.) In
8492: such a case, @var{class} may not be a safe value to return. @var{class} is
8493: certainly valid for the instruction, but it may not be valid for reloading
8494: @var{x}. This problem can occur on machines such as the 68000 and 80386
8495: where some registers can handle full-word values but cannot handle
8496: single-byte values.
8497:
8498: On such machines, this macro must examine the mode of @var{x} and return a
8499: subclass of @var{class} which can handle loads and stores of that mode. On
8500: the 68000, where address registers cannot handle @code{QImode}, if @var{x}
8501: has @code{QImode} then you must return @code{DATA_REGS}. If @var{class} is
8502: @code{ADDR_REGS}, then there is no correct value to return; but the shift
8503: and bitwise-and instructions don't use @code{ADDR_REGS}, so this fatal case
8504: never arises.
8505:
8506: @item CLASS_MAX_NREGS (@var{class}, @var{mode})
8507: A C expression for the maximum number of consecutive registers
8508: of class @var{class} needed to hold a value of mode @var{mode}.
8509:
8510: This is closely related to the macro @code{HARD_REGNO_NREGS}.
8511: In fact, the value of the macro @code{CLASS_MAX_NREGS (@var{class}, @var{mode})}
8512: should be the maximum value of @code{HARD_REGNO_NREGS (@var{regno}, @var{mode})}
8513: for all @var{regno} values in the class @var{class}.
8514:
8515: This macro helps control the handling of multiple-word values
8516: in the reload pass.
8517: @end table
8518:
8519: Two other special macros describe which constants fit which constraint
8520: letters.
8521:
8522: @table @code
8523: @item CONST_OK_FOR_LETTER_P (@var{value}, @var{c})
8524: A C expression that defines the machine-dependent operand constraint letters
8525: that specify particular ranges of integer values. If @var{c} is one
8526: of those letters, the expression should check that @var{value}, an integer,
8527: is in the appropriate range and return 1 if so, 0 otherwise. If @var{c} is
8528: not one of those letters, the value should be 0 regardless of @var{value}.
8529:
8530: @item CONST_DOUBLE_OK_FOR_LETTER_P (@var{value}, @var{c})
8531: A C expression that defines the machine-dependent operand constraint
8532: letters that specify particular ranges of floating values. If @var{c} is
8533: one of those letters, the expression should check that @var{value}, an RTX
1.1.1.8 root 8534: of code @code{const_double}, is in the appropriate range and return 1 if
1.1 root 8535: so, 0 otherwise. If @var{c} is not one of those letters, the value should
8536: be 0 regardless of @var{value}.
8537: @end table
8538:
8539: @node Stack Layout, Library Names, Register Classes, Machine Macros
8540: @section Describing Stack Layout
8541:
8542: @table @code
8543: @item STACK_GROWS_DOWNWARD
8544: Define this macro if pushing a word onto the stack moves the stack
8545: pointer to a smaller address.
8546:
8547: When we say, ``define this macro if @dots{},'' it means that the
8548: compiler checks this macro only with @code{#ifdef} so the precise
8549: definition used does not matter.
8550:
8551: @item FRAME_GROWS_DOWNWARD
8552: Define this macro if the addresses of local variable slots are at negative
8553: offsets from the frame pointer.
8554:
8555: @item STARTING_FRAME_OFFSET
8556: Offset from the frame pointer to the first local variable slot to be allocated.
8557:
8558: If @code{FRAME_GROWS_DOWNWARD}, the next slot's offset is found by
8559: subtracting the length of the first slot from @code{STARTING_FRAME_OFFSET}.
8560: Otherwise, it is found by adding the length of the first slot to
8561: the value @code{STARTING_FRAME_OFFSET}.
8562:
8563: @item PUSH_ROUNDING (@var{npushed})
8564: A C expression that is the number of bytes actually pushed onto the
8565: stack when an instruction attempts to push @var{npushed} bytes.
8566:
8567: If the target machine does not have a push instruction, do not define
8568: this macro. That directs GNU CC to use an alternate strategy: to
8569: allocate the entire argument block and then store the arguments into
8570: it.
8571:
8572: On some machines, the definition
8573:
8574: @example
8575: #define PUSH_ROUNDING(BYTES) (BYTES)
8576: @end example
8577:
8578: @noindent
8579: will suffice. But on other machines, instructions that appear
8580: to push one byte actually push two bytes in an attempt to maintain
8581: alignment. Then the definition should be
8582:
8583: @example
8584: #define PUSH_ROUNDING(BYTES) (((BYTES) + 1) & ~1)
8585: @end example
8586:
8587: @item FIRST_PARM_OFFSET (@var{fundecl})
8588: Offset from the argument pointer register to the first argument's
8589: address. On some machines it may depend on the data type of the
8590: function. (In the next version of GNU CC, the argument will be
8591: changed to the function data type rather than its declaration.)
8592:
8593: @item FIRST_PARM_CALLER_OFFSET (@var{fundecl})
8594: Define this macro on machines where register parameters have shadow
8595: locations on the stack, at addresses below the nominal parameter.
8596: This matters because certain arguments cannot be passed on the stack.
8597: On these machines, such arguments must be stored into the shadow
8598: locations.
8599:
8600: This macro should expand into a C expression whose value is the offset
8601: of the first parameter's shadow location from the nominal stack
8602: pointer value. (That value is itself computed by adding the value of
8603: @code{STACK_POINTER_OFFSET} to the stack pointer register.)
8604:
1.1.1.9 root 8605: @item REG_PARM_STACK_SPACE
8606: Define this macro if functions should assume that stack space has been
8607: allocated for arguments even when their values are passed in
8608: registers.
8609:
8610: The actual allocation of such space would be done either by
8611: the call instruction or by the function prologue, or by
1.1.1.10! root 8612: defining @code{FIRST_PARM_CALLER_OFFSET}.
1.1.1.9 root 8613:
1.1.1.6 root 8614: @item STACK_ARGS_ADJUST (@var{size})
8615: Define this macro if the machine requires padding on the stack for
8616: certain function calls. This is padding on a per-function-call basis,
8617: not padding for individual arguments.
8618:
1.1.1.7 root 8619: The argument @var{size} will be a C variable of type @code{struct
8620: arg_data} which contains two fields, an integer named @code{constant}
8621: and an RTX named @code{var}. These together represent a size measured
8622: in bytes which is the sum of the integer and the RTX. Most of the
8623: time @code{var} is 0, which means that the size is simply the integer.
8624:
8625: The definition should be a C statement or compound statement
8626: which alters the variable supplied in whatever way you wish.
8627:
8628: Note that the value you leave in the variable @code{size} will
8629: ultimately be rounded up to a multiple of @code{STACK_BOUNDARY} bits.
8630:
8631: This macro is not fully implemented for machines which have push
8632: instructions (i.e., on which @code{PUSH_ROUNDING} is defined).
1.1.1.6 root 8633:
1.1 root 8634: @item RETURN_POPS_ARGS (@var{funtype})
8635: A C expression that should be 1 if a function pops its own arguments
8636: on returning, or 0 if the function pops no arguments and the caller
8637: must therefore pop them all after the function returns.
8638:
8639: @var{funtype} is a C variable whose value is a tree node that
8640: describes the function in question. Normally it is a node of type
8641: @code{FUNCTION_TYPE} that describes the data type of the function.
8642: From this it is possible to obtain the data types of the value and
8643: arguments (if known).
8644:
8645: When a call to a library function is being considered, @var{funtype}
8646: will contain an identifier node for the library function. Thus, if
8647: you need to distinguish among various library functions, you can do so
8648: by their names. Note that ``library function'' in this context means
8649: a function used to perform arithmetic, whose name is known specially
8650: in the compiler and was not mentioned in the C code being compiled.
8651:
8652: On the Vax, all functions always pop their arguments, so the
8653: definition of this macro is 1. On the 68000, using the standard
8654: calling convention, no functions pop their arguments, so the value of
8655: the macro is always 0 in this case. But an alternative calling
8656: convention is available in which functions that take a fixed number of
8657: arguments pop them but other functions (such as @code{printf}) pop
8658: nothing (the caller pops all). When this convention is in use,
8659: @var{funtype} is examined to determine whether a function takes a
8660: fixed number of arguments.
8661:
1.1.1.10! root 8662: When this macro returns nonzero, the macro @code{FRAME_POINTER_REQUIRED}
! 8663: must also return nonzero for proper operation.
! 8664:
1.1 root 8665: @item FUNCTION_VALUE (@var{valtype}, @var{func})
8666: A C expression to create an RTX representing the place where a
8667: function returns a value of data type @var{valtype}. @var{valtype} is
8668: a tree node representing a data type. Write @code{TYPE_MODE
8669: (@var{valtype})} to get the machine mode used to represent that type.
8670: On many machines, only the mode is relevant. (Actually, on most
8671: machines, scalar values are returned in the same place regardless of
8672: mode).@refill
8673:
8674: If the precise function being called is known, @var{func} is a tree
8675: node (@code{FUNCTION_DECL}) for it; otherwise, @var{func} is a null
8676: pointer. This makes it possible to use a different value-returning
8677: convention for specific functions when all their calls are
8678: known.@refill
8679:
8680: @item FUNCTION_OUTGOING_VALUE (@var{valtype}, @var{func})
8681: Define this macro if the target machine has ``register windows''
8682: so that the register in which a function returns its value is not
8683: the same as the one in which the caller sees the value.
8684:
8685: For such machines, @code{FUNCTION_VALUE} computes the register in
8686: which the caller will see the value, and
8687: @code{FUNCTION_OUTGOING_VALUE} should be defined in a similar fashion
8688: to tell the function where to put the value.@refill
8689:
8690: If @code{FUNCTION_OUTGOING_VALUE} is not defined,
8691: @code{FUNCTION_VALUE} serves both purposes.@refill
8692:
1.1.1.7 root 8693: @item RETURN_IN_MEMORY (@var{type})
8694: A C expression which can inhibit the returning of certain function
8695: values in registers, based on the type of value. A nonzero value says
8696: to return the function value in memory, just as large structures are
8697: always returned. Here @var{type} will be a C expression of type
8698: @code{tree}, representing the data type of the value.
8699:
8700: Note that values of mode @code{BLKmode} are returned in memory
8701: regardless of this macro. Also, the option @samp{-fpcc-struct-return}
8702: takes effect regardless of this macro. On most systems, it is
8703: possible to leave the macro undefined; this causes a default
8704: definition to be used, whose value is the constant 0.
8705:
1.1 root 8706: @item LIBCALL_VALUE (@var{mode})
8707: A C expression to create an RTX representing the place where a library
8708: function returns a value of mode @var{mode}. If the precise function
8709: being called is known, @var{func} is a tree node
8710: (@code{FUNCTION_DECL}) for it; otherwise, @var{func} is a null
8711: pointer. This makes it possible to use a different value-returning
8712: convention for specific functions when all their calls are
8713: known.@refill
8714:
8715: Note that ``library function'' in this context means a compiler
8716: support routine, used to perform arithmetic, whose name is known
8717: specially by the compiler and was not mentioned in the C code being
8718: compiled.
8719:
8720: @item FUNCTION_VALUE_REGNO_P (@var{regno})
8721: A C expression that is nonzero if @var{regno} is the number of a hard
8722: register in which the values of called function may come back.
8723:
8724: A register whose use for returning values is limited to serving as the
8725: second of a pair (for a value of type @code{double}, say) need not be
8726: recognized by this macro. So for most machines, this definition
8727: suffices:
8728:
8729: @example
8730: #define FUNCTION_VALUE_REGNO_P(N) ((N) == 0)
8731: @end example
8732:
8733: If the machine has register windows, so that the caller and the called
8734: function use different registers for the return value, this macro
8735: should recognize only the caller's register numbers.
8736:
8737: @item FUNCTION_ARG (@var{cum}, @var{mode}, @var{type}, @var{named})
8738: A C expression that controls whether a function argument is passed
8739: in a register, and which register.
8740:
8741: The arguments are @var{cum}, which summarizes all the previous
8742: arguments; @var{mode}, the machine mode of the argument; @var{type},
8743: the data type of the argument as a tree node or 0 if that is not known
8744: (which happens for C support library functions); and @var{named},
8745: which is 1 for an ordinary argument and 0 for nameless arguments that
1.1.1.8 root 8746: correspond to @samp{@dots{}} in the called function's prototype.
1.1 root 8747:
1.1.1.8 root 8748: The value of the expression should either be a @code{reg} RTX for the
1.1 root 8749: hard register in which to pass the argument, or zero to pass the
8750: argument on the stack.
8751:
8752: For the Vax and 68000, where normally all arguments are pushed, zero
8753: suffices as a definition.
8754:
1.1.1.8 root 8755: The usual way to make the ANSI library @file{stdarg.h} work on a machine
8756: where some arguments are usually passed in registers, is to cause
8757: nameless arguments to be passed on the stack instead. This is done
8758: by making @code{FUNCTION_ARG} return 0 whenever @var{named} is 0.
8759:
1.1 root 8760: @item FUNCTION_INCOMING_ARG (@var{cum}, @var{mode}, @var{type}, @var{named})
8761: Define this macro if the target machine has ``register windows'', so
8762: that the register in which a function sees an arguments is not
8763: necessarily the same as the one in which the caller passed the
8764: argument.
8765:
8766: For such machines, @code{FUNCTION_ARG} computes the register in which
8767: the caller passes the value, and @code{FUNCTION_INCOMING_ARG} should
8768: be defined in a similar fashion to tell the function being called
8769: where the arguments will arrive.
8770:
8771: If @code{FUNCTION_INCOMING_ARG} is not defined, @code{FUNCTION_ARG}
8772: serves both purposes.@refill
8773:
8774: @item FUNCTION_ARG_PARTIAL_NREGS (@var{cum}, @var{mode}, @var{type}, @var{named})
8775: A C expression for the number of words, at the beginning of an
8776: argument, must be put in registers. The value must be zero for
8777: arguments that are passed entirely in registers or that are entirely
8778: pushed on the stack.
8779:
8780: On some machines, certain arguments must be passed partially in
8781: registers and partially in memory. On these machines, typically the
8782: first @var{n} words of arguments are passed in registers, and the rest
8783: on the stack. If a multi-word argument (a @code{double} or a
8784: structure) crosses that boundary, its first few words must be passed
8785: in registers and the rest must be pushed. This macro tells the
8786: compiler when this occurs, and how many of the words should go in
8787: registers.
8788:
8789: @code{FUNCTION_ARG} for these arguments should return the first
8790: register to be used by the caller for this argument; likewise
8791: @code{FUNCTION_INCOMING_ARG}, for the called function.
8792:
8793: @item CUMULATIVE_ARGS
8794: A C type for declaring a variable that is used as the first argument
8795: of @code{FUNCTION_ARG} and other related values. For some target
8796: machines, the type @code{int} suffices and can hold the number of
8797: bytes of argument so far.
8798:
8799: @item INIT_CUMULATIVE_ARGS (@var{cum}, @var{fntype})
8800: A C statement (sans semicolon) for initializing the variable @var{cum}
8801: for the state at the beginning of the argument list. The variable has
8802: type @code{CUMULATIVE_ARGS}. The value of @var{fntype} is the tree node
8803: for the data type of the function which will receive the args, or 0
8804: if the args are to a compiler support library function.
8805:
8806: @item FUNCTION_ARG_ADVANCE (@var{cum}, @var{mode}, @var{type}, @var{named})
1.1.1.7 root 8807: A C statement (sans semicolon) to update the summarizer variable
8808: @var{cum} to advance past an argument in the argument list. The
8809: values @var{mode}, @var{type} and @var{named} describe that argument.
8810: Once this is done, the variable @var{cum} is suitable for analyzing
8811: the @emph{following} argument with @code{FUNCTION_ARG}, etc.@refill
1.1 root 8812:
8813: @item FUNCTION_ARG_REGNO_P (@var{regno})
8814: A C expression that is nonzero if @var{regno} is the number of a hard
8815: register in which function arguments are sometimes passed. This does
8816: @emph{not} include implicit arguments such as the static chain and
8817: the structure-value address. On many machines, no registers can be
8818: used for this purpose since all function arguments are pushed on the
8819: stack.
8820:
8821: @item FUNCTION_ARG_PADDING (@var{mode}, @var{size})
8822: If defined, a C expression which determines whether, and in which direction,
8823: to pad out an argument with extra space. The value should be of type
8824: @code{enum direction}: either @code{upward} to pad above the argument,
8825: @code{downward} to pad below, or @code{none} to inhibit padding.
8826:
8827: The argument @var{size} is an RTX which describes the size of the
8828: argument, in bytes. It should be used only if @var{mode} is
8829: @code{BLKmode}. Otherwise, @var{size} is 0.
8830:
8831: This macro does not control the @emph{amount} of padding; that is
8832: always just enough to reach the next multiple of @code{PARM_BOUNDARY}.
8833:
8834: This macro has a default definition which is right for most systems.
8835: For little-endian machines, the default is to pad upward. For
8836: big-endian machines, the default is to pad downward for an argument of
8837: constant size shorter than an @code{int}, and upward otherwise.
8838:
8839: @item FUNCTION_PROLOGUE (@var{file}, @var{size})
8840: A C compound statement that outputs the assembler code for entry to a
8841: function. The prologue is responsible for setting up the stack frame,
8842: initializing the frame pointer register, saving registers that must be
8843: saved, and allocating @var{size} additional bytes of storage for the
8844: local variables. @var{size} is an integer. @var{file} is a stdio
8845: stream to which the assembler code should be output.
8846:
8847: The label for the beginning of the function need not be output by this
8848: macro. That has already been done when the macro is run.
8849:
8850: To determine which registers to save, the macro can refer to the array
8851: @code{regs_ever_live}: element @var{r} is nonzero if hard register
8852: @var{r} is used anywhere within the function. This implies the
8853: function prologue should save register @var{r}, but not if it is one
8854: of the call-used registers.
8855:
8856: On machines where functions may or may not have frame-pointers, the
8857: function entry code must vary accordingly; it must set up the frame
8858: pointer if one is wanted, and not otherwise. To determine whether a
8859: frame pointer is in wanted, the macro can refer to the variable
8860: @code{frame_pointer_needed}. The variable's value will be 1 at run
8861: time in a function that needs a frame pointer.
8862:
1.1.1.10! root 8863: On machines where an argument may be passed partly in registers and
! 8864: partly in memory, this macro must examine the variable
1.1.1.8 root 8865: @code{current_function_pretend_args_size}, and allocate that many bytes
8866: of uninitialized space on the stack just underneath the first argument
8867: arriving on the stack. (This may not be at the very end of the stack,
8868: if the calling sequence has pushed anything else since pushing the stack
8869: arguments. But usually, on such machines, nothing else has been pushed
8870: yet, because the function prologue itself does all the pushing.)
8871:
1.1 root 8872: @item FUNCTION_PROFILER (@var{file}, @var{labelno})
8873: A C statement or compound statement to output to @var{file} some
8874: assembler code to call the profiling subroutine @code{mcount}.
8875: Before calling, the assembler code must load the address of a
8876: counter variable into a register where @code{mcount} expects to
8877: find the address. The name of this variable is @samp{LP} followed
8878: by the number @var{labelno}, so you would generate the name using
8879: @samp{LP%d} in a @code{fprintf}.
8880:
8881: The details of how the address should be passed to @code{mcount} are
8882: determined by your operating system environment, not by GNU CC. To
8883: figure them out, compile a small program for profiling using the
8884: system's installed C compiler and look at the assembler code that
8885: results.
8886:
1.1.1.6 root 8887: @item FUNCTION_BLOCK_PROFILER (@var{file}, @var{labelno})
8888: A C statement or compound statement to output to @var{file} some
8889: assembler code to initialize basic-block profiling for the current
8890: object module. This code should call the subroutine
8891: @code{__bb_init_func} once per object module, passing it as its sole
8892: argument the address of a block allocated in the object module.
8893:
8894: The name of the block is a local symbol made with this statement:
8895:
8896: @example
8897: ASM_GENERATE_INTERNAL_LABEL (@var{buffer}, "LPBX", 0);
8898: @end example
8899:
8900: Of course, since you are writing the definition of
8901: @code{ASM_GENERATE_INTERNAL_LABEL} as well as that of this macro, you
8902: can take a short cut in the definition of this macro and use the name
8903: that you know will result.
8904:
8905: The first word of this block is a flag which will be nonzero if the
8906: object module has already been initialized. So test this word first,
8907: and do not call @code{__bb_init_func} if the flag is nonzero.
8908:
8909: @item BLOCK_PROFILER (@var{file}, @var{blockno})
8910: A C statement or compound statement to increment the count associated
8911: with the basic block number @var{blockno}. Basic blocks are numbered
8912: separately from zero within each compilation. The count associated
8913: with block number @var{blockno} is at index @var{blockno} in a vector
8914: of words; the name of this array is a local symbol made with this
8915: statement:
8916:
8917: @example
8918: ASM_GENERATE_INTERNAL_LABEL (@var{buffer}, "LPBX", 2);
8919: @end example
8920:
8921: Of course, since you are writing the definition of
8922: @code{ASM_GENERATE_INTERNAL_LABEL} as well as that of this macro, you
8923: can take a short cut in the definition of this macro and use the name
8924: that you know will result.
8925:
1.1.1.10! root 8926: @item EXIT_IGNORE_STACK
1.1 root 8927: Define this macro as a C expression that is nonzero if the return
8928: instruction or the function epilogue ignores the value of the stack
8929: pointer; in other words, if it is safe to delete an instruction to
8930: adjust the stack pointer before a return from the function.
8931:
1.1.1.8 root 8932: Note that this macro's value is relevant only for functions for which
8933: frame pointers are maintained. It is never safe to delete a final
8934: stack adjustment in a function that has no frame pointer, and the
1.1.1.10! root 8935: compiler knows this regardless of @code{EXIT_IGNORE_STACK}.
1.1 root 8936:
8937: @item FUNCTION_EPILOGUE (@var{file}, @var{size})
8938: A C compound statement that outputs the assembler code for exit from a
8939: function. The epilogue is responsible for restoring the saved
8940: registers and stack pointer to their values when the function was
8941: called, and returning control to the caller. This macro takes the
8942: same arguments as the macro @code{FUNCTION_PROLOGUE}, and the
8943: registers to restore are determined from @code{regs_ever_live} and
8944: @code{CALL_USED_REGISTERS} in the same way.
8945:
8946: On some machines, there is a single instruction that does all the work
8947: of returning from the function. On these machines, give that
8948: instruction the name @samp{return} and do not define the macro
8949: @code{FUNCTION_EPILOGUE} at all.
8950:
8951: Do not define a pattern named @samp{return} if you want the
8952: @code{FUNCTION_EPILOGUE} to be used. If you want the target switches
8953: to control whether return instructions or epilogues are used, define a
8954: @samp{return} pattern with a validity condition that tests the target
8955: switches appropriately. If the @samp{return} pattern's validity
8956: condition is false, epilogues will be used.
8957:
8958: On machines where functions may or may not have frame-pointers, the
8959: function exit code must vary accordingly. Sometimes the code for
8960: these two cases is completely different. To determine whether a frame
8961: pointer is in wanted, the macro can refer to the variable
8962: @code{frame_pointer_needed}. The variable's value will be 1 at run
8963: time in a function that needs a frame pointer.
8964:
8965: On some machines, some functions pop their arguments on exit while
8966: others leave that for the caller to do. For example, the 68020 when
8967: given @samp{-mrtd} pops arguments in functions that take a fixed
8968: number of arguments.
8969:
8970: Your definition of the macro @code{RETURN_POPS_ARGS} decides which
8971: functions pop their own arguments. @code{FUNCTION_EPILOGUE} needs to
8972: know what was decided. The variable @code{current_function_pops_args}
8973: is nonzero if the function should pop its own arguments. If so, use
8974: the variable @code{current_function_args_size} as the number of bytes
8975: to pop.
8976:
8977: @item FIX_FRAME_POINTER_ADDRESS (@var{addr}, @var{depth})
8978: A C compound statement to alter a memory address that uses the frame
8979: pointer register so that it uses the stack pointer register instead.
8980: This must be done in the instructions that load parameter values into
8981: registers, when the reload pass determines that a frame pointer is not
8982: necessary for the function. @var{addr} will be a C variable name, and
8983: the updated address should be stored in that variable. @var{depth}
8984: will be the current depth of stack temporaries (number of bytes of
8985: arguments currently pushed). The change in offset between a
8986: frame-pointer-relative address and a stack-pointer-relative address
8987: must include @var{depth}.
8988:
8989: Even if your machine description specifies there will always be a
8990: frame pointer in the frame pointer register, you must still define
8991: @code{FIX_FRAME_POINTER_ADDRESS}, but the definition will never be
8992: executed at run time, so it may be empty.
1.1.1.8 root 8993:
8994: @item LONGJMP_RESTORE_FROM_STACK
8995: Define this macro if the @code{longjmp} function restores registers
8996: from the stack frames, rather than from those saved specifically by
8997: @code{setjmp}. Certain quantities must not be kept in registers
8998: across a call to @code{setjmp} on such machines.
1.1 root 8999: @end table
9000:
9001: @node Library Names, Addressing Modes, Stack Layout, Machine Macros
9002: @section Library Subroutine Names
9003:
9004: @table @code
1.1.1.5 root 9005: @item MULSI3_LIBCALL
9006: A C string constant giving the name of the function to call for
9007: multiplication of one signed full-word by another. If you do not
9008: define this macro, the default name is used, which is @code{__mulsi3},
9009: a function defined in @file{gnulib}.
9010:
9011: @item UMULSI3_LIBCALL
9012: A C string constant giving the name of the function to call for
9013: multiplication of one unsigned full-word by another. If you do not
9014: define this macro, the default name is used, which is
9015: @code{__umulsi3}, a function defined in @file{gnulib}.
9016:
9017: @item DIVSI3_LIBCALL
9018: A C string constant giving the name of the function to call for
9019: division of one signed full-word by another. If you do not define
9020: this macro, the default name is used, which is @code{__divsi3}, a
9021: function defined in @file{gnulib}.
9022:
1.1 root 9023: @item UDIVSI3_LIBCALL
9024: A C string constant giving the name of the function to call for
1.1.1.5 root 9025: division of one unsigned full-word by another. If you do not define
9026: this macro, the default name is used, which is @code{__udivsi3}, a
9027: function defined in @file{gnulib}.
9028:
9029: @item MODSI3_LIBCALL
9030: A C string constant giving the name of the function to call for the
9031: remainder in division of one signed full-word by another. If you do
9032: not define this macro, the default name is used, which is
9033: @code{__modsi3}, a function defined in @file{gnulib}.
1.1 root 9034:
9035: @item UMODSI3_LIBCALL
9036: A C string constant giving the name of the function to call for the
1.1.1.5 root 9037: remainder in division of one unsigned full-word by another. If you do
9038: not define this macro, the default name is used, which is
9039: @code{__umodsi3}, a function defined in @file{gnulib}.
1.1 root 9040:
9041: @item TARGET_MEM_FUNCTIONS
9042: Define this macro if GNU CC should generate calls to the System V
9043: (and ANSI C) library functions @code{memcpy} and @code{memset}
9044: rather than the BSD functions @code{bcopy} and @code{bzero}.
9045: @end table
9046:
1.1.1.8 root 9047: @node Addressing Modes, Delayed Branch, Library Names, Machine Macros
1.1 root 9048: @section Addressing Modes
9049:
9050: @table @code
9051: @item HAVE_POST_INCREMENT
9052: Define this macro if the machine supports post-increment addressing.
9053:
9054: @item HAVE_PRE_INCREMENT
9055: @itemx HAVE_POST_DECREMENT
9056: @itemx HAVE_PRE_DECREMENT
9057: Similar for other kinds of addressing.
9058:
9059: @item CONSTANT_ADDRESS_P (@var{x})
9060: A C expression that is 1 if the RTX @var{x} is a constant whose value
9061: is an integer. This includes integers whose values are not explicitly
1.1.1.8 root 9062: known, such as @code{symbol_ref} and @code{label_ref} expressions and
9063: @code{const} arithmetic expressions.
1.1 root 9064:
9065: On most machines, this can be defined as @code{CONSTANT_P (@var{x})},
9066: but a few machines are more restrictive in which constant addresses
9067: are supported.
9068:
9069: @item MAX_REGS_PER_ADDRESS
9070: A number, the maximum number of registers that can appear in a valid
1.1.1.10! root 9071: memory address. Note that it is up to you to specify a value equal to
! 9072: the maximum number that @code{go_if_legitimate_address} would ever
! 9073: accept.
1.1 root 9074:
9075: @item GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{label})
9076: A C compound statement with a conditional @code{goto @var{label};}
9077: executed if @var{x} (an RTX) is a legitimate memory address on the
9078: target machine for a memory operand of mode @var{mode}.
9079:
9080: It usually pays to define several simpler macros to serve as
9081: subroutines for this one. Otherwise it may be too complicated to
9082: understand.
9083:
9084: This macro must exist in two variants: a strict variant and a
9085: non-strict one. The strict variant is used in the reload pass. It
9086: must be defined so that any pseudo-register that has not been
9087: allocated a hard register is considered a memory reference. In
9088: contexts where some kind of register is required, a pseudo-register
9089: with no hard register must be rejected.
9090:
9091: The non-strict variant is used in other passes. It must be defined to
9092: accept all pseudo-registers in every context where some kind of
9093: register is required.
9094:
9095: Compiler source files that want to use the strict variant of this
9096: macro define the macro @code{REG_OK_STRICT}. You should use an
9097: @code{#ifdef REG_OK_STRICT} conditional to define the strict variant
9098: in that case and the non-strict variant otherwise.
9099:
9100: Typically among the subroutines used to define
9101: @code{GO_IF_LEGITIMATE_ADDRESS} are subroutines to check for
9102: acceptable registers for various purposes (one for base registers, one
9103: for index registers, and so on). Then only these subroutine macros
9104: need have two variants; the higher levels of macros may be the same
9105: whether strict or not.@refill
9106:
1.1.1.8 root 9107: Normally, constant addresses which are the sum of a @code{symbol_ref}
9108: and an integer are stored inside a @code{const} RTX to mark them as
9109: constant. Therefore, there is no need to recognize such sums as
9110: legitimate addresses.
9111:
9112: Usually @code{PRINT_OPERAND_ADDRESS} is not prepared to handle constant
9113: sums that are not marked with @code{const}. It assumes that a naked
9114: @code{plus} indicates indexing. If so, then you @emph{must} reject such
9115: naked constant sums as illegitimate addresses, so that none of them will
9116: be given to @code{PRINT_OPERAND_ADDRESS}.@refill
9117:
1.1 root 9118: @item REG_OK_FOR_BASE_P (@var{x})
1.1.1.5 root 9119: A C expression that is nonzero if @var{x} (assumed to be a @code{reg}
1.1 root 9120: RTX) is valid for use as a base register. For hard registers, it
9121: should always accept those which the hardware permits and reject the
9122: others. Whether the macro accepts or rejects pseudo registers must be
9123: controlled by @code{REG_OK_STRICT} as described above. This usually
9124: requires two variant definitions, of which @code{REG_OK_STRICT}
9125: controls the one actually used.
9126:
9127: @item REG_OK_FOR_INDEX_P (@var{x})
1.1.1.5 root 9128: A C expression that is nonzero if @var{x} (assumed to be a @code{reg}
1.1 root 9129: RTX) is valid for use as an index register.
9130:
9131: The difference between an index register and a base register is that
9132: the index register may be scaled. If an address involves the sum of
9133: two registers, neither one of them scaled, then either one may be
9134: labeled the ``base'' and the other the ``index''; but whichever
9135: labeling is used must fit the machine's constraints of which registers
9136: may serve in each capacity. The compiler will try both labelings,
9137: looking for one that is valid, and will reload one or both registers
9138: only if neither labeling works.
9139:
9140: @item LEGITIMIZE_ADDRESS (@var{x}, @var{oldx}, @var{mode}, @var{win})
9141: A C compound statement that attempts to replace @var{x} with a valid
9142: memory address for an operand of mode @var{mode}. @var{win} will be a
9143: C statement label elsewhere in the code; the macro definition may use
9144:
9145: @example
9146: GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{win});
9147: @end example
9148:
9149: @noindent
9150: to avoid further processing if the address has become legitimate.
9151:
9152: @var{x} will always be the result of a call to @code{break_out_memory_refs},
9153: and @var{oldx} will be the operand that was given to that function to produce
9154: @var{x}.
9155:
9156: The code generated by this macro should not alter the substructure of
9157: @var{x}. If it transforms @var{x} into a more legitimate form, it
9158: should assign @var{x} (which will always be a C variable) a new value.
9159:
9160: It is not necessary for this macro to come up with a legitimate
9161: address. The compiler has standard ways of doing so in all cases. In
9162: fact, it is safe for this macro to do nothing. But often a
9163: machine-dependent strategy can generate better code.
9164:
9165: @item GO_IF_MODE_DEPENDENT_ADDRESS (@var{addr}, @var{label})
9166: A C statement or compound statement with a conditional @code{goto
9167: @var{label};} executed if memory address @var{x} (an RTX) can have
9168: different meanings depending on the machine mode of the memory
9169: reference it is used for.
9170:
9171: Autoincrement and autodecrement addresses typically have mode-dependent
9172: effects because the amount of the increment or decrement is the size
9173: of the operand being addressed. Some machines have other mode-dependent
9174: addresses. Many RISC machines have no mode-dependent addresses.
9175:
9176: You may assume that @var{addr} is a valid address for the machine.
9177:
9178: @item LEGITIMATE_CONSTANT_P (@var{x})
9179: A C expression that is nonzero if @var{x} is a legitimate constant for
9180: an immediate operand on the target machine. You can assume that
1.1.1.8 root 9181: either @var{x} is a @code{const_double} or it satisfies
1.1 root 9182: @code{CONSTANT_P}, so you need not check these things. In fact,
9183: @samp{1} is a suitable definition for this macro on machines where any
1.1.1.8 root 9184: @code{const_double} is valid and anything @code{CONSTANT_P} is valid.@refill
9185: @end table
9186:
9187: @node Delayed Branch, Condition Code, Addressing Modes, Machine Macros
9188: @section Parameters for Delayed Branch Optimization
9189:
9190: @table @code
9191: @item HAVE_DELAYED_BRANCH
9192: Define this macro if the target machine has delayed branches, that is,
9193: a branch does not take effect immediately, and the actual branch
9194: instruction may be followed by one or more instructions that will be
9195: issued before the PC is actually changed.
9196:
9197: If defined, this allows a special scheduling pass to be run after the
9198: second jump optimization to attempt to reorder instructions to exploit
9199: this. Defining this macro also requires the definition of certain
9200: other macros described below.
9201:
9202: @item DBR_SLOTS_AFTER (@var{insn})
9203: This macro must be defined if @code{HAVE_DELAYED_BRANCH} is defined.
9204: Its definition should be a C expression returning the number of
9205: available delay slots following the instruction(s) output by the
9206: pattern for @var{insn}. The definition of ``slot'' is
9207: machine-dependent, and may denote instructions, bytes, or whatever.
9208:
9209: @item DBR_INSN_SLOTS (@var{insn})
9210: This macro must be defined if @code{HAVE_DELAYED_BRANCH} is defined.
9211: It should be a C expression returning the number of slots (typically
9212: the number of machine instructions) consumed by @var{insn}.
9213:
9214: You may assume that @var{insn} is truly an insn, not a note, label,
9215: barrier, dispatch table, @code{use}, or @code{clobber}.
9216:
9217: @item DBR_INSN_ELIGIBLE_P (@var{insn}, @var{dinsn})
9218: A C expression whose value is non-zero if it is legitimate to put
9219: @var{insn} in the delay slot following @var{dinsn}.
9220:
9221: You do not need to take account of data flow considerations in the
9222: definition of this macro, because the delayed branch optimizer always
9223: does that. This macro is needed only when certain insns may not be
9224: placed in certain delay slots for reasons not evident from the RTL
9225: expressions themselves. If there are no such problems, you don't need
9226: to define this macro.
9227:
9228: You may assume that @var{insn} is truly an insn, not a note, label,
9229: barrier, dispatch table, @code{use}, or @code{clobber}. You may
9230: assume that @var{dinsn} is a jump insn with a delay slot.
9231:
9232: @item DBR_OUTPUT_SEQEND(@var{file})
9233: A C statement, to be executed after all slot-filler instructions have
9234: been output. If necessary, call @code{dbr_sequence_length} to
9235: determine the number of slots filled in a sequence (zero if not
9236: currently outputting a sequence), to decide how many no-ops to output,
9237: or whatever.
9238:
9239: Don't define this macro if it has nothing to do, but it is helpful in
9240: reading assembly output if the extent of the delay sequence is made
9241: explicit (e.g. with white space).
9242:
9243: Note that output routines for instructions with delay slots must be
9244: prepared to deal with not being output as part of a sequence (i.e.
9245: when the scheduling pass is not run, or when no slot fillers could be
9246: found.) The variable @code{final_sequence} is null when not
9247: processing a sequence, otherwise it contains the @code{sequence} rtx
9248: being output.
1.1 root 9249: @end table
9250:
1.1.1.9 root 9251: @node Condition Code, Cross-compilation, Delayed Branch, Machine Macros
1.1.1.8 root 9252: @section Condition Code Information
9253:
9254: The file @file{conditions.h} defines a variable @code{cc_status} to
9255: describe how the condition code was computed (in case the interpretation of
9256: the condition code depends on the instruction that it was set by). This
9257: variable contains the RTL expressions on which the condition code is
9258: currently based, and several standard flags.
9259:
9260: Sometimes additional machine-specific flags must be defined in the machine
9261: description header file. It can also add additional machine-specific
9262: information by defining @code{CC_STATUS_MDEP}.
9263:
9264: @table @code
9265: @item CC_STATUS_MDEP
9266: C code for a data type which is used for declaring the @code{mdep}
9267: component of @code{cc_status}. It defaults to @code{int}.
9268:
9269: @item CC_STATUS_MDEP_INIT
1.1.1.9 root 9270: A C expression to initialize the @code{mdep} field to ``empty''.
9271: The default definition does nothing, since most machines don't use
9272: the field anyway. If you want to use the field, you should probably
9273: define this macro to initialize it.
1.1.1.8 root 9274:
9275: @item NOTICE_UPDATE_CC (@var{exp}, @var{insn})
9276: A C compound statement to set the components of @code{cc_status}
9277: appropriately for an insn @var{insn} whose body is @var{exp}. It is
9278: this macro's responsibility to recognize insns that set the condition
9279: code as a byproduct of other activity as well as those that explicitly
9280: set @code{(cc0)}.
9281:
9282: If there are insn that do not set the condition code but do alter
9283: other machine registers, this macro must check to see whether they
9284: invalidate the expressions that the condition code is recorded as
9285: reflecting. For example, on the 68000, insns that store in address
9286: registers do not set the condition code, which means that usually
9287: @code{NOTICE_UPDATE_CC} can leave @code{cc_status} unaltered for such
9288: insns. But suppose that the previous insn set the condition code
9289: based on location @samp{a4@@(102)} and the current insn stores a new
9290: value in @samp{a4}. Although the condition code is not changed by
9291: this, it will no longer be true that it reflects the contents of
9292: @samp{a4@@(102)}. Therefore, @code{NOTICE_UPDATE_CC} must alter
9293: @code{cc_status} in this case to say that nothing is known about the
9294: condition code value.
9295:
9296: The definition of @code{NOTICE_UPDATE_CC} must be prepared to deal
9297: with the results of peephole optimization: insns whose patterns are
9298: @code{parallel} RTXs containing various @code{reg}, @code{mem} or
9299: constants which are just the operands. The RTL structure of these
9300: insns is not sufficient to indicate what the insns actually do. What
9301: @code{NOTICE_UPDATE_CC} should do when it sees one is just to run
9302: @code{CC_STATUS_INIT}.
9303: @end table
9304:
9305: @node Cross-compilation, Misc, Condition Code, Machine Macros
1.1.1.5 root 9306: @section Cross Compilation and Floating-Point Format
9307:
1.1.1.9 root 9308: While all modern machines use 2's complement representation for integers,
1.1.1.5 root 9309: there are a variety of representations for floating point numbers. This
9310: means that in a cross-compiler the representation of floating point numbers
9311: in the compiled program may be different from that used in the machine
9312: doing the compilation.
9313:
9314: Because different representation systems may offer different amounts of
9315: range and precision, the cross compiler cannot safely use the host
9316: machine's floating point arithmetic. Therefore, floating point constants
9317: must be represented in the target machine's format. This means that the
9318: cross compiler cannot use @code{atof} to parse a floating point constant;
9319: it must have its own special routine to use instead. Also, constant
9320: folding must emulate the target machine's arithmetic (or must not be done
9321: at all).
9322:
9323: The macros in the following table should be defined only if you are cross
9324: compiling between different floating point formats.
9325:
9326: Otherwise, don't define them. Then default definitions will be set up which
9327: use @code{double} as the data type, @code{==} to test for equality, etc.
9328:
9329: You don't need to worry about how many times you use an operand of any
9330: of these macros. The compiler never uses operands which have side effects.
9331:
9332: @table @code
9333: @item REAL_VALUE_TYPE
9334: A macro for the C data type to be used to hold a floating point value
9335: in the target machine's format. Typically this would be a
9336: @code{struct} containing an array of @code{int}.
9337:
9338: @item REAL_VALUES_EQUAL (@var{x}, @var{y})
9339: A macro for a C expression which compares for equality the two values,
9340: @var{x} and @var{y}, both of type @code{REAL_VALUE_TYPE}.
9341:
9342: @item REAL_VALUES_LESS (@var{x}, @var{y})
9343: A macro for a C expression which tests whether @var{x} is less than
9344: @var{y}, both values being of type @code{REAL_VALUE_TYPE} and
9345: interpreted as floating point numbers in the target machine's
9346: representation.
9347:
9348: @item REAL_VALUE_LDEXP (@var{x}, @var{scale})
9349: A macro for a C expression which performs the standard library
9350: function @code{ldexp}, but using the target machine's floating point
9351: representation. Both @var{x} and the value of the expression have
9352: type @code{REAL_VALUE_TYPE}. The second argument, @var{scale}, is an
9353: integer.
9354:
9355: @item REAL_VALUE_ATOF (@var{string})
9356: A macro for a C expression which converts @var{string}, an expression
9357: of type @code{char *}, into a floating point number in the target
9358: machine's representation. The value has type @code{REAL_VALUE_TYPE}.
9359: @end table
9360:
9361: Define the following additional macros if you want to make floating
9362: point constant folding work while cross compiling. If you don't
9363: define them, cross compilation is still possible, but constant folding
9364: will not happen for floating point values.
9365:
9366: @table @code
9367: @item REAL_ARITHMETIC (@var{output}, @var{code}, @var{x}, @var{y})
9368: A macro for a C statement which calculates an arithmetic operation of
9369: the two floating point values @var{x} and @var{y}, both of type
9370: @code{REAL_VALUE_TYPE} in the target machine's representation, to
9371: produce a result of the same type and representation which is stored
9372: in @var{output} (which will be a variable).
9373:
9374: The operation to be performed is specified by @var{code}, a tree code
9375: which will always be one of the following: @code{PLUS_EXPR},
9376: @code{MINUS_EXPR}, @code{MULT_EXPR}, @code{RDIV_EXPR},
9377: @code{MAX_EXPR}, @code{MIN_EXPR}.@refill
9378:
9379: The expansion of this macro is responsible for checking for overflow.
9380: If overflow happens, the macro expansion should execute the statement
9381: @code{return 0;}, which indicates the inability to perform the
9382: arithmetic operation requested.
9383:
9384: @item REAL_VALUE_NEGATE (@var{x})
9385: A macro for a C expression which returns the negative of the floating
9386: point value @var{x}. Both @var{x} and the value of the expression
9387: have type @code{REAL_VALUE_TYPE} and are in the target machine's
9388: floating point representation.
9389:
9390: There is no way for this macro to report overflow, since overflow
9391: can't happen in the negation operation.
9392:
9393: @item REAL_VALUE_TO_INT (@var{low}, @var{high}, @var{x})
9394: A macro for a C expression which converts a floating point value
9395: @var{x} into a double-precision integer which is then stored into
9396: @var{low} and @var{high}, two variables of type @var{int}.
9397:
9398: @item REAL_VALUE_FROM_INT (@var{x}, @var{low}, @var{high})
9399: A macro for a C expression which converts a double-precision integer
9400: found in @var{low} and @var{high}, two variables of type @var{int},
9401: into a floating point value which is then stored into @var{x}.
9402: @end table
9403:
1.1.1.8 root 9404: @node Misc, Assembler Format, Cross-compilation, Machine Macros
1.1 root 9405: @section Miscellaneous Parameters
9406:
9407: @table @code
9408: @item CASE_VECTOR_MODE
9409: An alias for a machine mode name. This is the machine mode that
9410: elements of a jump-table should have.
9411:
9412: @item CASE_VECTOR_PC_RELATIVE
9413: Define this macro if jump-tables should contain relative addresses.
9414:
9415: @item CASE_DROPS_THROUGH
9416: Define this if control falls through a @code{case} insn when the index
9417: value is out of range. This means the specified default-label is
9418: actually ignored by the @code{case} insn proper.
9419:
9420: @item IMPLICIT_FIX_EXPR
9421: An alias for a tree code that should be used by default for conversion
9422: of floating point values to fixed point. Normally,
9423: @code{FIX_ROUND_EXPR} is used.@refill
9424:
9425: @item FIXUNS_TRUNC_LIKE_FIX_TRUNC
9426: Define this macro if the same instructions that convert a floating
9427: point number to a signed fixed point number also convert validly to an
9428: unsigned one.
9429:
9430: @item EASY_DIV_EXPR
9431: An alias for a tree code that is the easiest kind of division to
9432: compile code for in the general case. It may be
9433: @code{TRUNC_DIV_EXPR}, @code{FLOOR_DIV_EXPR}, @code{CEIL_DIV_EXPR} or
9434: @code{ROUND_DIV_EXPR}. These four division operators differ in how
9435: they round the result to an integer. @code{EASY_DIV_EXPR} is used
9436: when it is permissible to use any of those kinds of division and the
9437: choice should be made on the basis of efficiency.@refill
9438:
9439: @item DEFAULT_SIGNED_CHAR
9440: An expression whose value is 1 or 0, according to whether the type
9441: @code{char} should be signed or unsigned by default. The user can
9442: always override this default with the options @samp{-fsigned-char}
9443: and @samp{-funsigned-char}.
9444:
9445: @item SCCS_DIRECTIVE
9446: Define this if the preprocessor should ignore @code{#sccs} directives
9447: and print no error message.
9448:
1.1.1.7 root 9449: @item HAVE_VPRINTF
9450: Define this if the library function @code{vprintf} is available on your
9451: system.
1.1 root 9452:
9453: @item MOVE_MAX
9454: The maximum number of bytes that a single instruction can move quickly
9455: from memory to memory.
9456:
9457: @item INT_TYPE_SIZE
9458: A C expression for the size in bits of the type @code{int} on the
1.1.1.8 root 9459: target machine. If you don't define this, the default is one word.
9460:
9461: @item SHORT_TYPE_SIZE
9462: A C expression for the size in bits of the type @code{short} on the
9463: target machine. If you don't define this, the default is half a word.
9464: (If this would be less than one storage unit, it is rounded up to one
9465: unit.)
9466:
9467: @item LONG_TYPE_SIZE
9468: A C expression for the size in bits of the type @code{long} on the
9469: target machine. If you don't define this, the default is one word.
9470:
9471: @item LONG_LONG_TYPE_SIZE
9472: A C expression for the size in bits of the type @code{long long} on the
9473: target machine. If you don't define this, the default is two
9474: words.
9475:
9476: @item CHAR_TYPE_SIZE
9477: A C expression for the size in bits of the type @code{char} on the
9478: target machine. If you don't define this, the default is one quarter
9479: of a word. (If this would be less than one storage unit, it is rounded up
9480: to one unit.)
9481:
9482: @item FLOAT_TYPE_SIZE
9483: A C expression for the size in bits of the type @code{float} on the
9484: target machine. If you don't define this, the default is one word.
9485:
9486: @item DOUBLE_TYPE_SIZE
9487: A C expression for the size in bits of the type @code{double} on the
9488: target machine. If you don't define this, the default is two
9489: words.
9490:
9491: @item LONG_DOUBLE_TYPE_SIZE
9492: A C expression for the size in bits of the type @code{long double} on
9493: the target machine. If you don't define this, the default is two
9494: words.
1.1 root 9495:
9496: @item SLOW_BYTE_ACCESS
9497: Define this macro as a C expression which is nonzero if accessing less
9498: than a word of memory (i.e. a @code{char} or a @code{short}) is slow
9499: (requires more than one instruction).
9500:
9501: @item SLOW_ZERO_EXTEND
9502: Define this macro if zero-extension (of a @code{char} or @code{short}
9503: to an @code{int}) can be done faster if the destination is a register
9504: that is known to be zero.
9505:
9506: If you define this macro, you must have instruction patterns that
9507: recognize RTL structures like this:
9508:
9509: @example
9510: (set (strict-low-part (subreg:QI (reg:SI @dots{}) 0)) @dots{})
9511: @end example
9512:
9513: @noindent
9514: and likewise for @code{HImode}.
9515:
9516: @item SHIFT_COUNT_TRUNCATED
9517: Define this macro if shift instructions ignore all but the lowest few
9518: bits of the shift count. It implies that a sign-extend or zero-extend
9519: instruction for the shift count can be omitted.
9520:
9521: @item TRULY_NOOP_TRUNCATION (@var{outprec}, @var{inprec})
9522: A C expression which is nonzero if on this machine it is safe to
9523: ``convert'' an integer of @var{inprec} bits to one of @var{outprec}
9524: bits (where @var{outprec} is smaller than @var{inprec}) by merely
9525: operating on it as if it had only @var{outprec} bits.
9526:
9527: On many machines, this expression can be 1.
9528:
9529: @item NO_FUNCTION_CSE
9530: Define this macro if it is as good or better to call a constant
9531: function address than to call an address kept in a register.
9532:
9533: @item PROMOTE_PROTOTYPES
9534: Define this macro if an argument declared as @code{char} or
9535: @code{short} in a prototype should actually be passed as an
9536: @code{int}. In addition to avoiding errors in certain cases of
9537: mismatch, it also makes for better code on certain machines.
9538:
9539: @item STORE_FLAG_VALUE
9540: A C expression for the value stored by a store-flag instruction
9541: (@code{s@var{cond}}) when the condition is true. This is usually 1 or
1.1.1.9 root 9542: -1; it is required to be an odd number or a negative number.
1.1 root 9543:
9544: Do not define @code{STORE_FLAG_VALUE} if the machine has no store-flag
9545: instructions.
9546:
9547: @item Pmode
9548: An alias for the machine mode for pointers. Normally the definition
9549: can be
9550:
9551: @example
9552: #define Pmode SImode
9553: @end example
9554:
9555: @item FUNCTION_MODE
9556: An alias for the machine mode used for memory references to functions
1.1.1.8 root 9557: being called, in @code{call} RTL expressions. On most machines this
1.1 root 9558: should be @code{QImode}.
9559:
9560: @item INSN_MACHINE_INFO
9561: This macro should expand into a C structure type to use for the
9562: machine-dependent info field specified with the optional last argument
1.1.1.8 root 9563: in @code{define_insn} and @code{define_peephole} patterns. For example,
9564: it might expand into @code{struct machine_info}; then it would be up
1.1 root 9565: to you to define this structure in the @file{tm.h} file.
9566:
9567: You do not need to define this macro if you do not write the optional
9568: last argument in any of the patterns in the machine description.
9569:
1.1.1.8 root 9570: @item DEFAULT_MACHINE_INFO
9571: This macro should expand into a C initializer to use to initialize
9572: the machine-dependent info for one insn pattern. It is used for patterns
9573: that do not specify the machine-dependent info.
9574:
9575: If you do not define this macro, zero is used.
9576:
1.1 root 9577: @item CONST_COSTS (@var{x}, @var{code})
9578: A part of a C @code{switch} statement that describes the relative
9579: costs of constant RTL expressions. It must contain @code{case} labels
1.1.1.8 root 9580: for expression codes @code{const_int}, @code{const}, @code{symbol_ref}, @code{label_ref}
9581: and @code{const_double}. Each case must ultimately reach a
1.1 root 9582: @code{return} statement to return the relative cost of the use of that
9583: kind of constant value in an expression. The cost may depend on the
9584: precise value of the constant, which is available for examination in
9585: @var{x}.
9586:
9587: @var{code} is the expression code---redundant, since it can be
9588: obtained with @code{GET_CODE (@var{x})}.
9589:
9590: @item DOLLARS_IN_IDENTIFIERS
9591: Define this to be nonzero if the character @samp{$} should be allowed
9592: by default in identifier names.
9593: @end table
9594:
1.1.1.8 root 9595: @node Assembler Format,, Misc, Machine Macros
1.1 root 9596: @section Output of Assembler Code
9597:
9598: @table @code
9599: @item ASM_SPEC
9600: A C string constant that tells the GNU CC driver program options to
9601: pass to the assembler. It can also specify how to translate options
9602: you give to GNU CC into options for GNU CC to pass to the assembler.
9603: See the file @file{tm-sun3.h} for an example of this.
9604:
9605: Do not define this macro if it does not need to do anything.
9606:
9607: @item LINK_SPEC
9608: A C string constant that tells the GNU CC driver program options to
9609: pass to the linker. It can also specify how to translate options you
9610: give to GNU CC into options for GNU CC to pass to the linker.
9611:
9612: Do not define this macro if it does not need to do anything.
9613:
9614: @item LIB_SPEC
9615: Another C string constant used much like @code{LINK_SPEC}. The difference
9616: between the two is that @code{LIBS_SPEC} is used at the end of the
9617: command given to the linker.
9618:
9619: If this macro is not defined, a default is provided that
9620: loads the standard C library from the usual place. See @file{gcc.c}.
9621:
9622: @item STARTFILE_SPEC
9623: Another C string constant used much like @code{LINK_SPEC}. The
9624: difference between the two is that @code{STARTFILE_SPEC} is used at
9625: the very beginning of the command given to the linker.
9626:
9627: If this macro is not defined, a default is provided that loads the
9628: standard C startup file from the usual place. See @file{gcc.c}.
9629:
1.1.1.7 root 9630: @item STANDARD_EXEC_PREFIX
9631: Define this macro as a C string constant if you wish to override the
9632: standard choice of @file{/usr/local/lib/gcc-} as the default prefix to
9633: try when searching for the executable files of the compiler.
9634:
9635: The prefix specified by the @samp{-B} option, if any, is tried before
9636: the default prefix. After the default prefix, if the executable is
9637: not found that way, @file{/usr/lib/gcc-} is tried next; then the
9638: directories in your search path for shell commands are searched.
9639:
1.1.1.4 root 9640: @item STANDARD_STARTFILE_PREFIX
9641: Define this macro as a C string constant if you wish to override the
1.1.1.7 root 9642: standard choice of @file{/usr/local/lib/} as the default prefix to try
9643: when searching for startup files such as @file{crt0.o}.
9644:
9645: In this search, all the prefixes tried for executable files are tried
9646: first. Then comes the default startfile prefix specified by this
9647: macro, followed by the prefixes @file{/lib/} and @file{/usr/lib/} as
9648: last resorts.
1.1.1.4 root 9649:
1.1 root 9650: @item ASM_FILE_START (@var{stream})
9651: A C expression which outputs to the stdio stream @var{stream}
9652: some appropriate text to go at the start of an assembler file.
9653:
9654: Normally this macro is defined to output a line containing
9655: @samp{#NO_APP}, which is a comment that has no effect on most
9656: assemblers but tells the GNU assembler that it can save time by not
9657: checking for certain assembler constructs.
9658:
9659: On systems that use SDB, it is necessary to output certain commands;
9660: see @file{tm-attasm.h}.
9661:
1.1.1.8 root 9662: @item ASM_FILE_END (@var{stream})
9663: A C expression which outputs to the stdio stream @var{stream}
9664: some appropriate text to go at the end of an assembler file.
9665:
9666: If this macro is not defined, the default is to output nothing
9667: special at the end of the file. Most systems don't require any
9668: definition.
9669:
9670: On systems that use SDB, it is necessary to output certain commands;
9671: see @file{tm-attasm.h}.
9672:
9673: @item ASM_IDENTIFY_GCC (@var{file})
9674: A C statement to output assembler commands which will identify
9675: the object file as having been compiled with GNU CC (or another
9676: GNU compiler).
9677:
9678: If you don't define this macro, the string @samp{gcc_compiled.:}
9679: is output. This string is calculated to define a symbol which,
9680: on BSD systems, will never be defined for any other reason.
9681: GDB checks for the presence of this symbol when reading the
9682: symbol table of an executable.
9683:
9684: On non-BSD systems, you must arrange communication with GDB in
9685: some other fashion. If GDB is not used on your system, you can
9686: define this macro with an empty body.
9687:
1.1 root 9688: @item ASM_APP_ON
9689: A C string constant for text to be output before each @code{asm}
9690: statement or group of consecutive ones. Normally this is
9691: @code{"#APP"}, which is a comment that has no effect on most
9692: assemblers but tells the GNU assembler that it must check the lines
9693: that follow for all valid assembler constructs.
9694:
9695: @item ASM_APP_OFF
9696: A C string constant for text to be output after each @code{asm}
9697: statement or group of consecutive ones. Normally this is
9698: @code{"#NO_APP"}, which tells the GNU assembler to resume making the
9699: time-saving assumptions that are valid for ordinary compiler output.
9700:
9701: @item TEXT_SECTION_ASM_OP
9702: A C string constant for the assembler operation that should precede
9703: instructions and read-only data. Normally @code{".text"} is right.
9704:
9705: @item DATA_SECTION_ASM_OP
9706: A C string constant for the assembler operation to identify the
9707: following data as writable initialized data. Normally @code{".data"}
9708: is right.
9709:
1.1.1.8 root 9710: @item EXTRA_SECTIONS
9711: A list of names for sections other than the standard two, which are
9712: @code{in_text} and @code{in_data}. You need not define this macro
9713: on a system with no other sections (that GCC needs to use).
9714:
9715: @item EXTRA_SECTION_FUNCTIONS
9716: One or more functions to be defined in @file{varasm.c}. These
9717: functions should do jobs analogous to those of @code{text_section} and
9718: @code{data_section}, for your additional sections. Do not define this
9719: macro if you do not define @code{EXTRA_SECTIONS}.
9720:
9721: @item SELECT_SECTION (@var{exp})
9722: A C statement or statements to switch to the appropriate section for
9723: output of @var{exp}. You can assume that @var{exp} is either a
9724: @code{VAR_DECL} node or a constant of some sort. Select the section
9725: by calling @code{text_section} or one of the alternatives for other
9726: sections.
9727:
9728: Do not define this macro if you use only the standard two sections
9729: and put all read-only variables and constants in the text section.
9730:
9731: @item SELECT_RTX_SECTION (@var{mode}, @var{rtx})
9732: A C statement or statements to switch to the appropriate section for
9733: output of @var{rtx} in mode @var{mode}. You can assume that @var{rtx}
9734: is some kind of constant in RTL. The argument @var{mode} is redundant
9735: except in the case of a @code{const_int} rtx. Select the section by
9736: calling @code{text_section} or one of the alternatives for other
9737: sections.
9738:
9739: Do not define this macro if you use only the standard two sections and
9740: put all constants in the text section.
9741:
1.1 root 9742: @item REGISTER_NAMES
9743: A C initializer containing the assembler's names for the machine
9744: registers, each one as a C string constant. This is what translates
9745: register numbers in the compiler into assembler language.
9746:
9747: @item DBX_REGISTER_NUMBER (@var{regno})
9748: A C expression that returns the DBX register number for the compiler
9749: register number @var{regno}. In simple cases, the value of this
9750: expression may be @var{regno} itself. But sometimes there are some
9751: registers that the compiler knows about and DBX does not, or vice
9752: versa. In such cases, some register may need to have one number in
9753: the compiler and another for DBX.
9754:
9755: @item DBX_DEBUGGING_INFO
9756: Define this macro if GNU CC should produce debugging output for DBX
9757: in response to the @samp{-g} option.
9758:
9759: @item SDB_DEBUGGING_INFO
9760: Define this macro if GNU CC should produce debugging output for SDB
9761: in response to the @samp{-g} option.
9762:
9763: @item PUT_SDB_@var{op}
9764: Define these macros to override the assembler syntax for the special
9765: SDB assembler directives. See @file{sdbout.c} for a list of these
9766: macros and their arguments. If the standard syntax is used, you need
9767: not define them yourself.
9768:
9769: @item SDB_GENERATE_FAKE
9770: Define this macro to override the usual method of constructing a dummy
9771: name for anonymous structure and union types. See @file{sdbout.c} for
1.1.1.9 root 9772: more information.
1.1 root 9773:
9774: @item DBX_NO_XREFS
9775: Define this macro if DBX on your system does not support the construct
9776: @samp{xs@var{tagname}}. On some systems, this construct is used to
9777: describe a forward reference to a structure named @var{tagname}.
9778: On other systems, this construct is not supported at all.
9779:
9780: @item DBX_CONTIN_LENGTH
9781: A symbol name in DBX-format debugging information is normally
9782: continued (split into two separate @code{.stabs} directives) when it
9783: exceeds a certain length (by default, 80 characters). On some
9784: operating systems, DBX requires this splitting; on others, splitting
9785: must not be done. You can inhibit splitting by defining this macro
9786: with the value zero. You can override the default splitting-length by
9787: defining this macro as an expression for the length you desire.
9788:
9789: @item DBX_CONTIN_CHAR
9790: Normally continuation is indicated by adding a @samp{\} character to
9791: the end of a @code{.stabs} string when a continuation follows. To use
9792: a different character instead, define this macro as a character
9793: constant for the character you want to use. Do not define this macro
9794: if backslash is correct for your system.
9795:
1.1.1.8 root 9796: @item DBX_STATIC_STAB_DATA_SECTION
9797: Define this macro if it is necessary to go to the data section before
9798: outputting the @samp{.stabs} pseudo-op for a non-global static
9799: variable.
9800:
1.1 root 9801: @item ASM_OUTPUT_LABEL (@var{stream}, @var{name})
9802: A C statement (sans semicolon) to output to the stdio stream
1.1.1.8 root 9803: @var{stream} the assembler definition of a label named @var{name}.
9804: Use the expression @code{assemble_name (@var{stream}, @var{name})} to
9805: output the name itself; before and after that, output the additional
1.1 root 9806: assembler syntax for defining the name, and a newline.
9807:
9808: @item ASM_DECLARE_FUNCTION_NAME (@var{stream}, @var{name}, @var{decl})
9809: A C statement (sans semicolon) to output to the stdio stream
9810: @var{stream} any text necessary for declaring the name @var{name} of a
9811: function which is being defined. This macro is responsible for
9812: outputting the label definition (perhaps using
9813: @code{ASM_OUTPUT_LABEL}). The argument @var{decl} is the
9814: @code{FUNCTION_DECL} tree node representing the function.
9815:
9816: If this macro is not defined, then the function name is defined in the
9817: usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}).
9818:
9819: @item ASM_GLOBALIZE_LABEL (@var{stream}, @var{name})
9820: A C statement (sans semicolon) to output to the stdio stream
9821: @var{stream} some commands that will make the label @var{name} global;
9822: that is, available for reference from other files. Use the expression
9823: @code{assemble_name (@var{stream}, @var{name})} to output the name
9824: itself; before and after that, output the additional assembler syntax
9825: for making that name global, and a newline.
9826:
1.1.1.8 root 9827: @item ASM_OUTPUT_EXTERNAL (@var{stream}, @var{decl}, @var{name})
1.1 root 9828: A C statement (sans semicolon) to output to the stdio stream
9829: @var{stream} any text necessary for declaring the name of an external
9830: symbol named @var{name} which is referenced in this compilation but
9831: not defined. The value of @var{decl} is the tree node for the
9832: declaration.
9833:
9834: This macro need not be defined if it does not need to output anything.
9835: The GNU assembler and most Unix assemblers don't require anything.
9836:
9837: @item ASM_OUTPUT_LABELREF (@var{stream}, @var{name})
1.1.1.8 root 9838: A C statement to output to the stdio stream @var{stream} a reference
9839: in assembler syntax to a label named @var{name}. The character
9840: @samp{_} should be added to the front of the name, if that is
9841: customary on your operating system, as it is in most Berkeley Unix
9842: systems. This macro is used in @code{assemble_name}.
1.1 root 9843:
9844: @item ASM_GENERATE_INTERNAL_LABEL (@var{string}, @var{prefix}, @var{num})
1.1.1.8 root 9845: A C statement to store into the string @var{string} a label whose name
9846: is made from the string @var{prefix} and the number @var{num}.
1.1 root 9847:
9848: This string, when output subsequently by @code{ASM_OUTPUT_LABELREF},
9849: should produce the same output that @code{ASM_OUTPUT_INTERNAL_LABEL}
9850: would produce with the same @var{prefix} and @var{num}.
9851:
9852: @item ASM_OUTPUT_INTERNAL_LABEL (@var{stream}, @var{prefix}, @var{num})
9853: A C statement to output to the stdio stream @var{stream} a label whose
9854: name is made from the string @var{prefix} and the number @var{num}.
9855: These labels are used for internal purposes, and there is no reason
9856: for them to appear in the symbol table of the object file. On many
9857: systems, the letter @samp{L} at the beginning of a label has this
9858: effect. The usual definition of this macro is as follows:
9859:
9860: @example
9861: fprintf (@var{stream}, "L%s%d:\n", @var{prefix}, @var{num})
9862: @end example
9863:
9864: @item ASM_OUTPUT_CASE_LABEL (@var{stream}, @var{prefix}, @var{num}, @var{table})
9865: Define this if the label before a jump-table needs to be output
9866: specially. The first three arguments are the same as for
9867: @code{ASM_OUTPUT_INTERNAL_LABEL}; the fourth argument is the
1.1.1.8 root 9868: jump-table which follows (a @code{jump_insn} containing an
9869: @code{addr_vec} or @code{addr_diff_vec}).
1.1 root 9870:
9871: This feature is used on system V to output a @code{swbeg} statement
9872: for the table.
9873:
9874: If this macro is not defined, these labels are output with
9875: @code{ASM_OUTPUT_INTERNAL_LABEL}.
9876:
9877: @item ASM_OUTPUT_CASE_END (@var{stream}, @var{num}, @var{table})
1.1.1.8 root 9878: Define this if something special must be output at the end of a
9879: jump-table. The definition should be a C statement to be executed
9880: after the assembler code for the table is written. It should write
9881: the appropriate code to stdio stream @var{stream}. The argument
9882: @var{table} is the jump-table insn, and @var{num} is the label-number
9883: of the preceding label.
1.1 root 9884:
9885: If this macro is not defined, nothing special is output at the end of
9886: the jump-table.
9887:
1.1.1.4 root 9888: @item ASM_OUTPUT_ALIGN_CODE (@var{file})
9889: A C expression to output text to align the location counter in the way
9890: that is desirable at a point in the code that is reached only by
9891: jumping.
9892:
9893: This macro need not be defined if you don't want any special alignment
9894: to be done at such a time. Most machine descriptions do not currently
9895: define the macro.
9896:
1.1 root 9897: @item ASM_FORMAT_PRIVATE_NAME (@var{outvar}, @var{name}, @var{number})
9898: A C expression to assign to @var{outvar} (which is a variable of type
9899: @code{char *}) a newly allocated string made from the string
9900: @var{name} and the number @var{number}, with some suitable punctuation
9901: added. Use @code{alloca} to get space for the string.
9902:
9903: This string will be used as the argument to @code{ASM_OUTPUT_LABELREF}
9904: to produce an assembler label for an internal static variable whose
9905: name is @var{name}. Therefore, the string must be such as to result
9906: in valid assembler code. The argument @var{number} is different each
9907: time this macro is executed; it prevents conflicts between
9908: similarly-named internal static variables in different scopes.
9909:
9910: Ideally this string should not be a valid C identifier, to prevent any
9911: conflict with the user's own symbols. Most assemblers allow periods
9912: or percent signs in assembler symbols; putting at least one of these
9913: between the name and the number will suffice.
9914:
9915: @item ASM_OUTPUT_REG_PUSH (@var{stream}, @var{regno})
9916: A C expression to output to @var{stream} some assembler code
9917: which will push hard register number @var{regno} onto the stack.
9918: The code need not be optimal, since this macro is used only when
9919: profiling.
9920:
9921: @item ASM_OUTPUT_REG_POP (@var{stream}, @var{regno})
9922: A C expression to output to @var{stream} some assembler code
9923: which will pop hard register number @var{regno} off of the stack.
9924: The code need not be optimal, since this macro is used only when
9925: profiling.
9926:
9927: @item ASM_OUTPUT_ADDR_DIFF_ELT (@var{stream}, @var{value}, @var{rel})
9928: This macro should be provided on machines where the addresses
9929: in a dispatch table are relative to the table's own address.
9930:
9931: The definition should be a C statement to output to the stdio stream
9932: @var{stream} an assembler pseudo-instruction to generate a difference
9933: between two labels. @var{value} and @var{rel} are the numbers of two
9934: internal labels. The definitions of these labels are output using
9935: @code{ASM_OUTPUT_INTERNAL_LABEL}, and they must be printed in the same
9936: way here. For example,
9937:
9938: @example
9939: fprintf (@var{stream}, "\t.word L%d-L%d\n",
9940: @var{value}, @var{rel})
9941: @end example
9942:
9943: @item ASM_OUTPUT_ADDR_VEC_ELT (@var{stream}, @var{value})
9944: This macro should be provided on machines where the addresses
9945: in a dispatch table are absolute.
9946:
9947: The definition should be a C statement to output to the stdio stream
9948: @var{stream} an assembler pseudo-instruction to generate a reference to
9949: a label. @var{value} is the number of an internal label whose
9950: definition is output using @code{ASM_OUTPUT_INTERNAL_LABEL}.
9951: For example,
9952:
9953: @example
9954: fprintf (@var{stream}, "\t.word L%d\n", @var{value})
9955: @end example
9956:
9957: @item ASM_OUTPUT_DOUBLE (@var{stream}, @var{value})
9958: A C statement to output to the stdio stream @var{stream} an assembler
9959: instruction to assemble a @code{double} constant whose value is
9960: @var{value}. @var{value} will be a C expression of type
9961: @code{double}.
9962:
9963: @item ASM_OUTPUT_FLOAT (@var{stream}, @var{value})
9964: A C statement to output to the stdio stream @var{stream} an assembler
9965: instruction to assemble a @code{float} constant whose value is
9966: @var{value}. @var{value} will be a C expression of type @code{float}.
9967:
9968: @item ASM_OUTPUT_INT (@var{stream}, @var{exp})
9969: @itemx ASM_OUTPUT_SHORT (@var{stream}, @var{exp})
9970: @itemx ASM_OUTPUT_CHAR (@var{stream}, @var{exp})
9971: A C statement to output to the stdio stream @var{stream} an assembler
9972: instruction to assemble a @code{int}, @code{short} or @code{char}
1.1.1.10! root 9973: constant whose value is @var{value}. The argument @var{exp} will be an
! 9974: RTL expression which represents a constant value. Use
! 9975: @samp{output_addr_const (@var{stream}, @var{exp})} to output this value
! 9976: as an assembler expression.@refill
1.1 root 9977:
1.1.1.8 root 9978: @item ASM_OUTPUT_DOUBLE_INT (@var{stream}, @var{exp})
9979: A C statement to output to the stdio stream @var{stream} an assembler
9980: instruction to assemble a @code{long long} constant whose value is
9981: @var{exp}. The argument @var{exp} will be an RTL expression which
9982: represents a constant value. It may be a @code{const_double} RTX,
9983: or it may be an ordinary single-precision constant. In the latter
9984: case, you should zero-extend it.
9985:
1.1 root 9986: @item ASM_OUTPUT_BYTE (@var{stream}, @var{value})
9987: A C statement to output to the stdio stream @var{stream} an assembler
9988: instruction to assemble a single byte containing the number @var{value}.
9989:
9990: @item ASM_OUTPUT_ASCII (@var{stream}, @var{ptr}, @var{len})
9991: A C statement to output to the stdio stream @var{stream} an assembler
9992: instruction to assemble a string constant containing the @var{len}
9993: bytes at @var{ptr}. @var{ptr} will be a C expression of type
9994: @code{char *} and @var{len} a C expression of type @code{int}.
9995:
9996: If the assembler has a @code{.ascii} pseudo-op as found in the
9997: Berkeley Unix assembler, do not define the macro
9998: @code{ASM_OUTPUT_ASCII}.
9999:
10000: @item ASM_OUTPUT_SKIP (@var{stream}, @var{nbytes})
10001: A C statement to output to the stdio stream @var{stream} an assembler
10002: instruction to advance the location counter by @var{nbytes} bytes.
10003: @var{nbytes} will be a C expression of type @code{int}.
10004:
10005: @item ASM_OUTPUT_ALIGN (@var{stream}, @var{power})
10006: A C statement to output to the stdio stream @var{stream} an assembler
10007: instruction to advance the location counter to a multiple of 2 to the
10008: @var{power} bytes. @var{power} will be a C expression of type @code{int}.
10009:
1.1.1.7 root 10010: @item ASM_OUTPUT_COMMON (@var{stream}, @var{name}, @var{size}, @var{rounded})
1.1 root 10011: A C statement (sans semicolon) to output to the stdio stream
1.1.1.7 root 10012: @var{stream} the assembler definition of a common-label named
10013: @var{name} whose size is @var{size} bytes. The variable @var{rounded}
10014: is the size rounded up to whatever alignment the caller wants.
10015:
10016: Use the expression @code{assemble_name (@var{stream}, @var{name})} to
10017: output the name itself; before and after that, output the additional
10018: assembler syntax for defining the name, and a newline.
1.1 root 10019:
10020: This macro controls how the assembler definitions of uninitialized
10021: global variables are output.
10022:
1.1.1.7 root 10023: @item ASM_OUTPUT_LOCAL (@var{stream}, @var{name}, @var{size}, @var{rounded})
1.1 root 10024: A C statement (sans semicolon) to output to the stdio stream
10025: @var{stream} the assembler definition of a local-common-label named
1.1.1.7 root 10026: @var{name} whose size is @var{size} bytes. The variable @var{rounded}
10027: is the size rounded up to whatever alignment the caller wants.
10028:
10029: Use the expression @code{assemble_name (@var{stream}, @var{name})} to
10030: output the name itself; before and after that, output the additional
10031: assembler syntax for defining the name, and a newline.
1.1 root 10032:
10033: This macro controls how the assembler definitions of uninitialized
10034: static variables are output.
10035:
1.1.1.8 root 10036: @item ASM_OUTPUT_SOURCE_FILENAME (@var{stream}, @var{name})
10037: A C statment to output DBX or SDB debugging information which indicates
10038: that filename @var{name} is the current source file to the stdio stream
10039: @var{stream}.
10040:
10041: This macro need not be defined if the standard form of debugging
10042: information for the debugger in use is appropriate.
10043:
1.1 root 10044: @item ASM_OUTPUT_SOURCE_LINE (@var{stream}, @var{line})
10045: A C statment to output DBX or SDB debugging information before code
10046: for line number @var{line} of the current source file to the
10047: stdio stream @var{stream}.
10048:
10049: This macro need not be defined if the standard form of debugging
10050: information for the debugger in use is appropriate.
10051:
10052: @item ASM_OUTPUT_IDENT (@var{stream}, @var{string})
10053: A C statement to output something to the assembler file to handle a
10054: @samp{#ident} directive containing the text @var{string}. If this
1.1.1.7 root 10055: macro is not defined, nothing is output for a @samp{#ident} directive.
1.1 root 10056:
10057: @item TARGET_BELL
10058: A C constant expression for the integer value for escape sequence
10059: @samp{\a}.
10060:
10061: @item TARGET_BS
10062: @itemx TARGET_TAB
10063: @itemx TARGET_NEWLINE
10064: C constant expressions for the integer values for escape sequences
10065: @samp{\b}, @samp{\t} and @samp{\n}.
10066:
10067: @item TARGET_VT
10068: @itemx TARGET_FF
10069: @itemx TARGET_CR
10070: C constant expressions for the integer values for escape sequences
10071: @samp{\v}, @samp{\f} and @samp{\r}.
10072:
10073: @item ASM_OUTPUT_OPCODE (@var{stream}, @var{ptr})
10074: Define this macro if you are using an unusual assembler that
10075: requires different names for the machine instructions.
10076:
10077: The definition is a C statement or statements which output an
10078: assembler instruction opcode to the stdio stream @var{stream}. The
10079: macro-operand @var{ptr} is a variable of type @code{char *} which
10080: points to the opcode name in its ``internal'' form---the form that is
10081: written in the machine description. The definition should output the
10082: opcode name to @var{stream}, performing any translation you desire, and
10083: increment the variable @var{ptr} to point at the end of the opcode
10084: so that it will not be output twice.
10085:
10086: In fact, your macro definition may process less than the entire opcode
10087: name, or more than the opcode name; but if you want to process text
10088: that includes @samp{%}-sequences to substitute operands, you must take
10089: care of the substitution yourself. Just be sure to increment
10090: @var{ptr} over whatever text should not be output normally.
10091:
1.1.1.8 root 10092: If you need to look at the operand values, they can be found as the
10093: elements of @code{recog_operand}.
10094:
1.1 root 10095: If the macro definition does nothing, the instruction is output
10096: in the usual way.
10097:
10098: @item FINAL_PRESCAN_INSN (@var{insn}, @var{opvec}, @var{noperands})
10099: If defined, a C statement to be executed just prior to the output of
10100: assembler code for @var{insn}, to modify the extracted operands so
10101: they will be output differently.
10102:
10103: Here the argument @var{opvec} is the vector containing the operands
10104: extracted from @var{insn}, and @var{noperands} is the number of
10105: elements of the vector which contain meaningful data for this insn.
10106: The contents of this vector are what will be used to convert the insn
10107: template into assembler code, so you can change the assembler output
10108: by changing the contents of the vector.
10109:
10110: This macro is useful when various assembler syntaxes share a single
10111: file of instruction patterns; by defining this macro differently, you
10112: can cause a large class of instructions to be output differently (such
10113: as with rearranged operands). Naturally, variations in assembler
10114: syntax affecting individual insn patterns ought to be handled by
10115: writing conditional output routines in those patterns.
10116:
10117: If this macro is not defined, it is equivalent to a null statement.
10118:
10119: @item PRINT_OPERAND (@var{stream}, @var{x}, @var{code})
10120: A C compound statement to output to stdio stream @var{stream} the
10121: assembler syntax for an instruction operand @var{x}. @var{x} is an
10122: RTL expression.
10123:
10124: @var{code} is a value that can be used to specify one of several ways
10125: of printing the operand. It is used when identical operands must be
10126: printed differently depending on the context. @var{code} comes from
10127: the @samp{%} specification that was used to request printing of the
10128: operand. If the specification was just @samp{%@var{digit}} then
10129: @var{code} is 0; if the specification was @samp{%@var{ltr}
10130: @var{digit}} then @var{code} is the ASCII code for @var{ltr}.
10131:
10132: If @var{x} is a register, this macro should print the register's name.
10133: The names can be found in an array @code{reg_names} whose type is
10134: @code{char *[]}. @code{reg_names} is initialized from
10135: @code{REGISTER_NAMES}.
10136:
10137: When the machine description has a specification @samp{%@var{punct}}
10138: (a @samp{%} followed by a punctuation character), this macro is called
10139: with a null pointer for @var{x} and the punctuation character for
10140: @var{code}.
10141:
1.1.1.8 root 10142: @item PRINT_OPERAND_PUNCT_VALID_P (@var{code})
10143: A C expression which evaluates to true if @var{code} is a valid
10144: punctuation character for use in the @code{PRINT_OPERAND} macro. If
10145: @code{PRINT_OPERAND_PUNCT_VALID_P} is not defined, it means that no
10146: punctuation characters (except for the standard one, @samp{%}) are used
10147: in this way.
10148:
1.1 root 10149: @item PRINT_OPERAND_ADDRESS (@var{stream}, @var{x})
10150: A C compound statement to output to stdio stream @var{stream} the
10151: assembler syntax for an instruction operand that is a memory reference
10152: whose address is @var{x}. @var{x} is an RTL expression.
10153:
10154: @item ASM_OPEN_PAREN
10155: @itemx ASM_CLOSE_PAREN
10156: These macros are defined as C string constant, describing the syntax
10157: in the assembler for grouping arithmetic expressions. The following
10158: definitions are correct for most assemblers:
10159:
10160: @example
10161: #define ASM_OPEN_PAREN "("
10162: #define ASM_CLOSE_PAREN ")"
10163: @end example
10164: @end table
10165:
10166: @node Config,, Machine Macros, Top
10167: @chapter The Configuration File
10168:
1.1.1.3 root 10169: The configuration file @file{xm-@var{machine}.h} contains macro definitions
10170: that describe the machine and system on which the compiler is running.
10171: Most of the values in it are actually the same on all machines that GNU CC
10172: runs on, so large parts of all configuration files are identical. But
1.1 root 10173: there are some macros that vary:
10174:
10175: @table @code
10176: @item FAILURE_EXIT_CODE
10177: A C expression for the status code to be returned when the compiler
10178: exits after serious errors.
10179:
10180: @item SUCCESS_EXIT_CODE
10181: A C expression for the status code to be returned when the compiler
10182: exits without serious errors.
1.1.1.10! root 10183:
! 10184: @item USE_C_ALLOCA
! 10185: Define this macro to indicate that the compiler is running with the
! 10186: @code{alloca} implemented in C. This version of @code{alloca} can be
! 10187: found in the file @file{alloca.c}; to use it, you must also alter the
! 10188: @file{Makefile} variable @code{ALLOCA}.
! 10189:
! 10190: This macro, unlike most, describes the machine that the compiler is
! 10191: running on, rather than the one the compiler is compiling for.
! 10192: Therefore, it should be set in the @file{xm-@var{machine}.h} file
! 10193: rather than in the @file{tm-@var{machine}.h} file.
! 10194:
! 10195: If you do define this macro, you should probably do it as follows:
! 10196:
! 10197: @example
! 10198: #ifndef __GNUC__
! 10199: #define USE_C_ALLOCA
! 10200: #else
! 10201: #define alloca __builtin_alloca
! 10202: #endif
! 10203: @end example
! 10204:
! 10205: @noindent
! 10206: so that when the compiler is compiled with GNU CC it uses the more
! 10207: efficient built-in @code{alloca} function.
1.1 root 10208: @end table
10209:
1.1.1.3 root 10210: In addition, configuration files for system V define @code{bcopy},
10211: @code{bzero} and @code{bcmp} as aliases. Some files define @code{alloca}
10212: as a macro when compiled with GNU CC, in order to take advantage of the
10213: benefit of GNU CC's built-in @code{alloca}.
10214:
1.1 root 10215: @contents
10216: @bye
This archive runs on limited infrastructure. Preserving old code on modern bandwidth. Automated agents are requested to crawl responsibly.