![[BACK]](/cvsweb/icons/back.gif){kind=icon}
Return to gcc.texinfo CVS log ![[TXT]](/cvsweb/icons/text.gif){kind=icon}
![[DIR]](/cvsweb/icons/dir.gif){kind=icon}
Up to [GCC 1.x] / gcc|
Return to gcc.texinfo CVS log | Up to [GCC 1.x] / gcc |
1.1 root 1: \input texinfo @c -*-texinfo-*-
2:
3: @settitle Using and Porting GNU CC
4: @setfilename gcc.info
5:
6: @ifinfo
7: This file documents the use and the internals of the GNU compiler.
8:
1.1.1.5 root 9: Copyright (C) 1988, 1989 Free Software Foundation, Inc.
1.1 root 10:
11: Permission is granted to make and distribute verbatim copies of
12: this manual provided the copyright notice and this permission notice
13: are preserved on all copies.
14:
15: @ignore
16: Permission is granted to process this file through Tex and print the
17: results, provided the printed document carries copying permission
18: notice identical to this one except for the removal of this paragraph
19: (this paragraph not being relevant to the printed manual).
20:
21: @end ignore
22: Permission is granted to copy and distribute modified versions of this
23: manual under the conditions for verbatim copying, provided also that the
1.1.1.9 ! 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
1.1.1.9 ! root 45: @center last updated 21 February 1990
1.1 root 46: @sp 1
1.1.1.9 ! root 47: @center for version 1.37.1
1.1 root 48: @page
49: @vskip 0pt plus 1filll
1.1.1.5 root 50: Copyright @copyright{} 1988, 1989 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.
89: * Incompatibilities:: Incompatibilities of GNU CC.
90: * Extensions:: GNU extensions to the C language.
91: * Bugs:: How to report bugs (if you want to get them fixed).
92: * Portability:: Goals of GNU CC's portability features.
93: * Interface:: Function-call interface of GNU CC output.
94: * Passes:: Order of passes, what they do, and what each file is for.
95: * RTL:: The intermediate representation that most passes work on.
96: * Machine Desc:: How to write machine description instruction patterns.
97: * Machine Macros:: How to write the machine description C macros.
1.1.1.8 root 98: * Config:: Writing the @file{xm-@var{machine}.h} file.
1.1 root 99: @end menu
100:
101: @node Copying, Contributors, Top, Top
1.1.1.6 root 102: @unnumbered GNU GENERAL PUBLIC LICENSE
103: @center Version 1, February 1989
1.1 root 104:
1.1.1.6 root 105: @display
106: Copyright @copyright{} 1989 Free Software Foundation, Inc.
107: 675 Mass Ave, Cambridge, MA 02139, USA
108:
109: Everyone is permitted to copy and distribute verbatim copies
110: of this license document, but changing it is not allowed.
111: @end display
112:
113: @unnumberedsec Preamble
114:
115: The license agreements of most software companies try to keep users
116: at the mercy of those companies. By contrast, our General Public
117: License is intended to guarantee your freedom to share and change free
118: software---to make sure the software is free for all its users. The
119: General Public License applies to the Free Software Foundation's
120: software and to any other program whose authors commit to using it.
121: You can use it for your programs, too.
122:
123: When we speak of free software, we are referring to freedom, not
124: price. Specifically, the General Public License is designed to make
125: sure that you have the freedom to give away or sell copies of free
126: software, that you receive source code or can get it if you want it,
127: that you can change the software or use pieces of it in new free
128: programs; and that you know you can do these things.
129:
130: To protect your rights, we need to make restrictions that forbid
131: anyone to deny you these rights or to ask you to surrender the rights.
132: These restrictions translate to certain responsibilities for you if you
133: distribute copies of the software, or if you modify it.
134:
135: For example, if you distribute copies of a such a program, whether
136: gratis or for a fee, you must give the recipients all the rights that
137: you have. You must make sure that they, too, receive or can get the
1.1 root 138: source code. And you must tell them their rights.
139:
1.1.1.6 root 140: We protect your rights with two steps: (1) copyright the software, and
141: (2) offer you this license which gives you legal permission to copy,
142: distribute and/or modify the software.
143:
144: Also, for each author's protection and ours, we want to make certain
145: that everyone understands that there is no warranty for this free
146: software. If the software is modified by someone else and passed on, we
147: want its recipients to know that what they have is not the original, so
148: that any problems introduced by others will not reflect on the original
149: authors' reputations.
1.1 root 150:
1.1.1.6 root 151: The precise terms and conditions for copying, distribution and
152: modification follow.
1.1 root 153:
1.1.1.6 root 154: @iftex
155: @unnumberedsec TERMS AND CONDITIONS
156: @end iftex
157: @ifinfo
158: @center TERMS AND CONDITIONS
159: @end ifinfo
1.1 root 160:
1.1.1.6 root 161: @enumerate
1.1 root 162: @item
1.1.1.6 root 163: This License Agreement applies to any program or other work which
164: contains a notice placed by the copyright holder saying it may be
165: distributed under the terms of this General Public License. The
166: ``Program'', below, refers to any such program or work, and a ``work based
167: on the Program'' means either the Program or any work containing the
168: Program or a portion of it, either verbatim or with modifications. Each
169: licensee is addressed as ``you''.
170:
171: @item
172: You may copy and distribute verbatim copies of the Program's source
173: code as you receive it, in any medium, provided that you conspicuously and
174: appropriately publish on each copy an appropriate copyright notice and
175: disclaimer of warranty; keep intact all the notices that refer to this
176: General Public License and to the absence of any warranty; and give any
177: other recipients of the Program a copy of this General Public License
178: along with the Program. You may charge a fee for the physical act of
179: transferring a copy.
180:
181: @item
182: You may modify your copy or copies of the Program or any portion of
183: it, and copy and distribute such modifications under the terms of Paragraph
184: 1 above, provided that you also do the following:
1.1 root 185:
186: @itemize @bullet
187: @item
1.1.1.6 root 188: cause the modified files to carry prominent notices stating that
189: you changed the files and the date of any change; and
1.1 root 190:
191: @item
192: cause the whole of any work that you distribute or publish, that
1.1.1.6 root 193: in whole or in part contains the Program or any part thereof, either
194: with or without modifications, to be licensed at no charge to all
195: third parties under the terms of this General Public License (except
196: that you may choose to grant warranty protection to some or all
197: third parties, at your option).
198:
199: @item
200: If the modified program normally reads commands interactively when
201: run, you must cause it, when started running for such interactive use
202: in the simplest and most usual way, to print or display an
203: announcement including an appropriate copyright notice and a notice
204: that there is no warranty (or else, saying that you provide a
205: warranty) and that users may redistribute the program under these
206: conditions, and telling the user how to view a copy of this General
207: Public License.
208:
209: @item
210: You may charge a fee for the physical act of transferring a
211: copy, and you may at your option offer warranty protection in
212: exchange for a fee.
1.1 root 213: @end itemize
214:
1.1.1.6 root 215: Mere aggregation of another independent work with the Program (or its
1.1 root 216: derivative) on a volume of a storage or distribution medium does not bring
1.1.1.6 root 217: the other work under the scope of these terms.
1.1 root 218:
219: @item
1.1.1.6 root 220: You may copy and distribute the Program (or a portion or derivative of
221: it, under Paragraph 2) in object code or executable form under the terms of
222: Paragraphs 1 and 2 above provided that you also do one of the following:
1.1 root 223:
224: @itemize @bullet
225: @item
226: accompany it with the complete corresponding machine-readable
227: source code, which must be distributed under the terms of
228: Paragraphs 1 and 2 above; or,
229:
230: @item
231: accompany it with a written offer, valid for at least three
1.1.1.6 root 232: years, to give any third party free (except for a nominal charge
233: for the cost of distribution) a complete machine-readable copy of the
1.1 root 234: corresponding source code, to be distributed under the terms of
235: Paragraphs 1 and 2 above; or,
236:
237: @item
238: accompany it with the information you received as to where the
239: corresponding source code may be obtained. (This alternative is
240: allowed only for noncommercial distribution and only if you
241: received the program in object code or executable form alone.)
242: @end itemize
243:
1.1.1.6 root 244: Source code for a work means the preferred form of the work for making
245: modifications to it. For an executable file, complete source code means
246: all the source code for all modules it contains; but, as a special
247: exception, it need not include source code for modules which are standard
248: libraries that accompany the operating system on which the executable
249: file runs, or for standard header files or definitions files that
250: accompany that operating system.
251:
252: @item
253: You may not copy, modify, sublicense, distribute or transfer the
254: Program except as expressly provided under this General Public License.
255: Any attempt otherwise to copy, modify, sublicense, distribute or transfer
256: the Program is void, and will automatically terminate your rights to use
257: the Program under this License. However, parties who have received
258: copies, or rights to use copies, from you under this General Public
259: License will not have their licenses terminated so long as such parties
260: remain in full compliance.
261:
262: @item
263: By copying, distributing or modifying the Program (or any work based
264: on the Program) you indicate your acceptance of this license to do so,
265: and all its terms and conditions.
266:
267: @item
268: Each time you redistribute the Program (or any work based on the
269: Program), the recipient automatically receives a license from the original
270: licensor to copy, distribute or modify the Program subject to these
271: terms and conditions. You may not impose any further restrictions on the
272: recipients' exercise of the rights granted herein.
273:
274: @item
275: The Free Software Foundation may publish revised and/or new versions
276: of the General Public License from time to time. Such new versions will
277: be similar in spirit to the present version, but may differ in detail to
278: address new problems or concerns.
279:
280: Each version is given a distinguishing version number. If the Program
281: specifies a version number of the license which applies to it and ``any
282: later version'', you have the option of following the terms and conditions
283: either of that version or of any later version published by the Free
284: Software Foundation. If the Program does not specify a version number of
285: the license, you may choose any version ever published by the Free Software
286: Foundation.
287:
288: @item
289: If you wish to incorporate parts of the Program into other free
290: programs whose distribution conditions are different, write to the author
291: to ask for permission. For software which is copyrighted by the Free
292: Software Foundation, write to the Free Software Foundation; we sometimes
293: make exceptions for this. Our decision will be guided by the two goals
294: of preserving the free status of all derivatives of our free software and
295: of promoting the sharing and reuse of software generally.
296:
297: @iftex
298: @heading NO WARRANTY
299: @end iftex
300: @ifinfo
301: @center NO WARRANTY
302: @end ifinfo
303:
304: @item
305: BECAUSE THE PROGRAM IS LICENSED FREE OF CHARGE, THERE IS NO WARRANTY
306: FOR THE PROGRAM, TO THE EXTENT PERMITTED BY APPLICABLE LAW. EXCEPT WHEN
307: OTHERWISE STATED IN WRITING THE COPYRIGHT HOLDERS AND/OR OTHER PARTIES
308: PROVIDE THE PROGRAM ``AS IS'' WITHOUT WARRANTY OF ANY KIND, EITHER EXPRESSED
309: OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
310: MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. THE ENTIRE RISK AS
311: TO THE QUALITY AND PERFORMANCE OF THE PROGRAM IS WITH YOU. SHOULD THE
312: PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF ALL NECESSARY SERVICING,
313: REPAIR OR CORRECTION.
314:
315: @item
316: IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING WILL
317: ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MAY MODIFY AND/OR
318: REDISTRIBUTE THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES,
319: INCLUDING ANY GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES
320: ARISING OUT OF THE USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT
321: LIMITED TO LOSS OF DATA OR DATA BEING RENDERED INACCURATE OR LOSSES
322: SUSTAINED BY YOU OR THIRD PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE
323: WITH ANY OTHER PROGRAMS), EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN
324: ADVISED OF THE POSSIBILITY OF SUCH DAMAGES.
1.1 root 325: @end enumerate
326:
1.1.1.6 root 327: @iftex
328: @heading END OF TERMS AND CONDITIONS
329: @end iftex
330: @ifinfo
331: @center END OF TERMS AND CONDITIONS
332: @end ifinfo
333:
334: @page
335: @unnumberedsec Appendix: How to Apply These Terms to Your New Programs
336:
337: If you develop a new program, and you want it to be of the greatest
338: possible use to humanity, the best way to achieve this is to make it
339: free software which everyone can redistribute and change under these
340: terms.
341:
342: To do so, attach the following notices to the program. It is safest to
343: attach them to the start of each source file to most effectively convey
344: the exclusion of warranty; and each file should have at least the
345: ``copyright'' line and a pointer to where the full notice is found.
346:
347: @smallexample
348: @var{one line to give the program's name and a brief idea of what it does.}
349: Copyright (C) 19@var{yy} @var{name of author}
350:
351: This program is free software; you can redistribute it and/or modify
352: it under the terms of the GNU General Public License as published by
353: the Free Software Foundation; either version 1, or (at your option)
354: any later version.
355:
356: This program is distributed in the hope that it will be useful,
357: but WITHOUT ANY WARRANTY; without even the implied warranty of
358: MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
359: GNU General Public License for more details.
360:
361: You should have received a copy of the GNU General Public License
362: along with this program; if not, write to the Free Software
363: Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
364: @end smallexample
365:
366: Also add information on how to contact you by electronic and paper mail.
367:
368: If the program is interactive, make it output a short notice like this
369: when it starts in an interactive mode:
370:
371: @smallexample
372: Gnomovision version 69, Copyright (C) 19@var{yy} @var{name of author}
373: Gnomovision comes with ABSOLUTELY NO WARRANTY; for details type `show w'.
374: This is free software, and you are welcome to redistribute it
375: under certain conditions; type `show c' for details.
376: @end smallexample
377:
378: The hypothetical commands `show w' and `show c' should show the
379: appropriate parts of the General Public License. Of course, the
380: commands you use may be called something other than `show w' and `show
381: c'; they could even be mouse-clicks or menu items---whatever suits your
382: program.
383:
384: You should also get your employer (if you work as a programmer) or your
385: school, if any, to sign a ``copyright disclaimer'' for the program, if
386: necessary. Here a sample; alter the names:
387:
388: @example
389: Yoyodyne, Inc., hereby disclaims all copyright interest in the
390: program `Gnomovision' (a program to direct compilers to make passes
391: at assemblers) written by James Hacker.
392:
393: @var{signature of Ty Coon}, 1 April 1989
394: Ty Coon, President of Vice
395: @end example
396:
397: That's all there is to it!
1.1 root 398:
1.1.1.9 ! root 399: @node Contributors, Boycott, Copying, Top
1.1 root 400: @unnumbered Contributors to GNU CC
401:
402: In addition to Richard Stallman, several people have written parts
403: of GNU CC.
404:
405: @itemize @bullet
406: @item
407: The idea of using RTL and some of the optimization ideas came from the
408: U. of Arizona Portable Optimizer, written by Jack Davidson and
409: Christopher Fraser. See ``Register Allocation and Exhaustive Peephole
410: Optimization'', Software Practice and Experience 14 (9), Sept. 1984,
411: 857-866.
412:
413: @item
414: Paul Rubin wrote most of the preprocessor.
415:
416: @item
1.1.1.6 root 417: Leonard Tower wrote parts of the parser, RTL generator, and RTL
1.1 root 418: definitions, and of the Vax machine description.
419:
420: @item
421: Ted Lemon wrote parts of the RTL reader and printer.
422:
423: @item
424: Jim Wilson implemented loop strength reduction and some other
425: loop optimizations.
426:
427: @item
428: Nobuyuki Hikichi of Software Research Associates, Tokyo, contributed
1.1.1.8 root 429: the support for the Sony NEWS machine.
1.1 root 430:
431: @item
432: Charles LaBrec contributed the support for the Integrated Solutions
433: 68020 system.
434:
435: @item
436: Michael Tiemann of MCC wrote most of the description of the National
437: Semiconductor 32000 series cpu. He also wrote the code for inline
438: function integration and for the SPARC cpu and Motorola 88000 cpu
439: and part of the Sun FPA support.
440:
441: @item
442: Jan Stein of the Chalmers Computer Society provided support for
443: Genix, as well as part of the 32000 machine description.
444:
445: @item
446: Randy Smith finished the Sun FPA support.
447:
448: @item
449: Robert Brown implemented the support for Encore 32000 systems.
450:
451: @item
452: David Kashtan of SRI adapted GNU CC to the Vomit-Making System.
453:
454: @item
455: Alex Crain provided changes for the 3b1.
456:
457: @item
458: Greg Satz and Chris Hanson assisted in making GNU CC work on HP-UX for
459: the 9000 series 300.
460:
461: @item
462: William Schelter did most of the work on the Intel 80386 support.
1.1.1.5 root 463:
464: @item
465: Christopher Smith did the port for Convex machines.
466:
467: @item
468: Paul Petersen wrote the machine description for the Alliant FX/8.
1.1.1.7 root 469:
470: @item
1.1.1.8 root 471: Alain Lichnewsky ported GNU CC to the Mips cpu.
472:
473: @item
474: Devon Bowen, Dale Wiles and Kevin Zachmann ported GNU CC to the Tahoe.
475:
476: @item
477: Jonathan Stone wrote the machine description for the Pyramid computer.
1.1 root 478: @end itemize
479:
1.1.1.9 ! root 480: @node Boycott, Options, Contributors, Top
! 481: @chapter Protect Your Freedom---Fight ``Look And Feel''
! 482:
! 483: Ashton-Tate, Apple, Lotus and Xerox are trying to create a new form of
! 484: legal monopoly: a copyright on a class of user interfaces. These
! 485: monopolies would cause serious problems for users and developers of
! 486: computer software and systems.
! 487:
! 488: Until three years ago, the law seemed clear: no one could restrict
! 489: others from using a user interface; programmers were free to implement
! 490: any interface they chose. Imitating interfaces, sometimes with changes,
! 491: was standard practice in the computer field. The interfaces we know
! 492: evolved gradually in this way; for example, the Macintosh user interface
! 493: drew ideas from the Xerox interface, which in turn drew on work done at
! 494: Stanford and SRI. 1-2-3 imitated VisiCalc, and dBase imitated a database
! 495: program from JPL.
! 496:
! 497: Most computer companies, and nearly all computer users, were happy with
! 498: this state of affairs. The companies that are suing say it does not
! 499: offer ``enough incentive'' to develop their products, but they must have
! 500: considered it ``enough'' when they made their decision to do so. It
! 501: seems they are not satisfied with the opportunity to continue to compete
! 502: in the marketplace---not even with a head start.
! 503:
! 504: If Xerox, Lotus, Apple and Ashton-Tate are permitted to make law through
! 505: the courts, the precedent will hobble the software industry:
! 506:
! 507: @itemize @bullet
! 508: @item
! 509: Gratuitous incompatibilites will burden users. Imagine if each
! 510: car manufacturer had to arrange the pedals in a different order.
! 511:
! 512: @item
! 513: Software will become and remain more expensive. Users will be
! 514: ``locked in'' to proprietary interfaces, for which there is no real
! 515: competition.
! 516:
! 517: @item
! 518: Large companies have an unfair advantage wherever lawsuits become
! 519: commonplace. Since they can easily afford to sue, they can intimidate
! 520: small companies with threats even when they don't really have a case.
! 521:
! 522: @item
! 523: User interface improvements will come slower, since incremental
! 524: evolution through creative imitation will no longer be permitted.
! 525:
! 526: @item
! 527: Even Apple, etc., will find it harder to make improvements if
! 528: they can no longer adapt the good ideas that others introduce, for
! 529: fear of weakening their own legal positions. Some users suggest that
! 530: this stagnation may already have started.
! 531: @end itemize
! 532:
! 533: Here are some things you can do to protect your freedom to write
! 534: programs:
! 535:
! 536: @itemize @bullet
! 537: @item
! 538: Don't buy from Xerox, Lotus, Apple or Ashton-Tate. Buy from their
! 539: competitors or from the defendants they are suing.
! 540:
! 541: @item
! 542: Don't develop software to work with the systems made by these companies.
! 543:
! 544: @item
! 545: Port your existing software to competing systems, so that you encourage
! 546: users to switch.
! 547:
! 548: @item
! 549: Write letters to company presidents to let them know their conduct
! 550: is unacceptable.
! 551:
! 552: @item
! 553: Tell your friends and colleagues about this issue and how it threatens
! 554: to ruin the computer industry.
! 555:
! 556: @item
! 557: Join the League for Programming Freedom. Phone (617) 492-0023 or write to:
! 558:
! 559: @example
! 560: League for Programming Freedom
! 561: 1 Kendall Square #143
! 562: P.O. Box 9171
! 563: Cambridge, MA 02139 league@@prep.ai.mit.edu
! 564: @end example
! 565:
! 566: @item
! 567: Above all, don't work for the look-and-feel plaintiffs, and don't
! 568: accept contracts from them.
! 569:
! 570: @item
! 571: Write to or phone your elected representatives to show them how
! 572: important this issue is.
! 573:
! 574: @example
! 575: Senator So and So Representative So and So
! 576: United States Senate House of Representatives
! 577: Washington, DC 20510 Washington, DC 20515
! 578: @end example
! 579:
! 580: You can phone senators and representatives at (202) 225-3121.
! 581: @end itemize
! 582:
! 583: Express your opinion! You can make a difference.
! 584:
! 585: @node Options, Installation, Boycott, Top
1.1 root 586: @chapter GNU CC Command Options
587:
588: The GNU C compiler uses a command syntax much like the Unix C compiler.
589: The @code{gcc} program accepts options and file names as operands.
590: Multiple single-letter options may @emph{not} be grouped: @samp{-dr} is
1.1.1.8 root 591: very different from @w{@samp{-d -r}}.
1.1 root 592:
593: When you invoke GNU CC, it normally does preprocessing, compilation,
594: assembly and linking. File names which end in @samp{.c} are taken as C
1.1.1.5 root 595: source to be preprocessed and compiled; file names ending in @samp{.i}
596: are taken as preprocessor output to be compiled; compiler output files
597: plus any input files with names ending in @samp{.s} are assembled; then
598: the resulting object files, plus any other input files, are linked
599: together to produce an executable.
1.1 root 600:
601: Command options allow you to stop this process at an intermediate stage.
602: For example, the @samp{-c} option says not to run the linker. Then the
603: output consists of object files output by the assembler.
604:
1.1.1.5 root 605: Other command options are passed on to one stage of processing. Some
606: options control the preprocessor and others the compiler itself. Yet
607: other options control the assembler and linker; these are not documented
608: here, but you rarely need to use any of them.
1.1 root 609:
610: Here are the options to control the overall compilation process, including
611: those that say whether to link, whether to assemble, and so on.
612:
613: @table @samp
614: @item -o @var{file}
615: Place output in file @var{file}. This applies regardless to whatever
616: sort of output is being produced, whether it be an executable file,
617: an object file, an assembler file or preprocessed C code.
618:
619: If @samp{-o} is not specified, the default is to put an executable file
620: in @file{a.out}, the object file @file{@var{source}.c} in
621: @file{@var{source}.o}, an assembler file in @file{@var{source}.s}, and
622: preprocessed C on standard output.@refill
623:
624: @item -c
625: Compile or assemble the source files, but do not link. Produce object
626: files with names made by replacing @samp{.c} or @samp{.s} with
627: @samp{.o} at the end of the input file names. Do nothing at all for
628: object files specified as input.
629:
630: @item -S
631: Compile into assembler code but do not assemble. The assembler output
632: file name is made by replacing @samp{.c} with @samp{.s} at the end of
633: the input file name. Do nothing at all for assembler source files or
634: object files specified as input.
635:
636: @item -E
637: Run only the C preprocessor. Preprocess all the C source files
638: specified and output the results to standard output.
639:
640: @item -v
641: Compiler driver program prints the commands it executes as it runs
642: the preprocessor, compiler proper, assembler and linker. Some of
643: these are directed to print their own version numbers.
644:
1.1.1.5 root 645: @item -pipe
646: Use pipes rather than temporary files for communication between the
647: various stages of compilation. This fails to work on some systems
648: where the assembler is unable to read from a pipe; but the GNU
649: assembler has no trouble.
650:
1.1 root 651: @item -B@var{prefix}
652: Compiler driver program tries @var{prefix} as a prefix for each
653: program it tries to run. These programs are @file{cpp}, @file{cc1},
654: @file{as} and @file{ld}.
655:
656: For each subprogram to be run, the compiler driver first tries the
657: @samp{-B} prefix, if any. If that name is not found, or if @samp{-B}
658: was not specified, the driver tries two standard prefixes, which are
659: @file{/usr/lib/gcc-} and @file{/usr/local/lib/gcc-}. If neither of
660: those results in a file name that is found, the unmodified program
661: name is searched for using the directories specified in your
662: @samp{PATH} environment variable.
663:
664: The run-time support file @file{gnulib} is also searched for using
665: the @samp{-B} prefix, if needed. If it is not found there, the two
666: standard prefixes above are tried, and that is all. The file is left
667: out of the link if it is not found by those means. Most of the time,
668: on most machines, you can do without it.
1.1.1.5 root 669:
670: You can get a similar result from the environment variable;
671: @code{GCC_EXEC_PREFIX} if it is defined, its value is used as a prefix
672: in the same way. If both the @samp{-B} option and the
673: @code{GCC_EXEC_PREFIX} variable are present, the @samp{-B} option is
674: used first and the environment variable value second.
1.1.1.8 root 675:
676: @item -b@var{prefix}
677: The argument @var{prefix} is used as a second prefix for the compiler
678: executables and libraries. This prefix is optional: the compiler tries
679: each file first with it, then without it. This prefix follows the
680: prefix specified with @samp{-B} or the default prefixes.
681:
682: Thus, @samp{-bvax- -Bcc/} in the presence of environment variable
683: @code{GCC_EXEC_PREFIX} with definition @file{/u/foo/} causes GNU CC to
684: try the following file names for the preprocessor executable:
685:
686: @example
687: cc/vax-cpp
688: cc/cpp
689: /u/foo/vax-cpp
690: /u/foo/cpp
691: /usr/local/lib/gcc-vax-cpp
692: /usr/local/lib/gcc-cpp
693: /usr/lib/gcc-vax-cpp
694: /usr/lib/gcc-cpp
695: @end example
1.1 root 696: @end table
697:
698: These options control the details of C compilation itself.
699:
700: @table @samp
701: @item -ansi
702: Support all ANSI standard C programs.
703:
704: This turns off certain features of GNU C that are incompatible with
705: ANSI C, such as the @code{asm}, @code{inline} and @code{typeof}
706: keywords, and predefined macros such as @code{unix} and @code{vax}
707: that identify the type of system you are using. It also enables the
708: undesirable and rarely used ANSI trigraph feature.
709:
1.1.1.8 root 710: The alternate keywords @code{__asm__}, @code{__inline__} and
711: @code{__typeof__} continue to work despite @samp{-ansi}. You would not
1.1.1.7 root 712: want to use them in an ANSI C program, of course, but it useful to put
713: them in header files that might be included in compilations done with
1.1.1.8 root 714: @samp{-ansi}. Alternate predefined macros such as @code{__unix__} and
715: @code{__vax__} are also available, with or without @samp{-ansi}.
1.1.1.7 root 716:
1.1 root 717: The @samp{-ansi} option does not cause non-ANSI programs to be
718: rejected gratuitously. For that, @samp{-pedantic} is required in
719: addition to @samp{-ansi}.
720:
721: The macro @code{__STRICT_ANSI__} is predefined when the @samp{-ansi}
722: option is used. Some header files may notice this macro and refrain
723: from declaring certain functions or defining certain macros that the
1.1.1.7 root 724: ANSI standard doesn't call for; this is to avoid interfering with any
725: programs that might use these names for other things.
1.1 root 726:
727: @item -traditional
728: Attempt to support some aspects of traditional C compilers.
729: Specifically:
730:
731: @itemize @bullet
732: @item
733: All @code{extern} declarations take effect globally even if they
734: are written inside of a function definition. This includes implicit
735: declarations of functions.
736:
737: @item
738: The keywords @code{typeof}, @code{inline}, @code{signed}, @code{const}
739: and @code{volatile} are not recognized.@refill
740:
741: @item
742: Comparisons between pointers and integers are always allowed.
743:
744: @item
745: Integer types @code{unsigned short} and @code{unsigned char} promote
746: to @code{unsigned int}.
747:
748: @item
749: Out-of-range floating point literals are not an error.
750:
751: @item
1.1.1.8 root 752: String ``constants'' are not necessarily constant; they are stored in
753: writable space, and identical looking constants are allocated
754: separately.
755:
756: @item
1.1.1.2 root 757: All automatic variables not declared @code{register} are preserved by
758: @code{longjmp}. Ordinarily, GNU C follows ANSI C: automatic variables
759: not declared @code{volatile} may be clobbered.
760:
761: @item
1.1 root 762: In the preprocessor, comments convert to nothing at all, rather than
763: to a space. This allows traditional token concatenation.
764:
765: @item
766: In the preprocessor, macro arguments are recognized within string
767: constants in a macro definition (and their values are stringified,
768: though without additional quote marks, when they appear in such a
769: context). The preprocessor always considers a string constant to end
770: at a newline.
771:
772: @item
773: The predefined macro @code{__STDC__} is not defined when you use
774: @samp{-traditional}, but @code{__GNUC__} is (since the GNU extensions
775: which @code{__GNUC__} indicates are not affected by
776: @samp{-traditional}). If you need to write header files that work
777: differently depending on whether @samp{-traditional} is in use, by
778: testing both of these predefined macros you can distinguish four
779: situations: GNU C, traditional GNU C, other ANSI C compilers, and
780: other old C compilers.
781: @end itemize
782:
783: @item -O
784: Optimize. Optimizing compilation takes somewhat more time, and a lot
785: more memory for a large function.
786:
787: Without @samp{-O}, the compiler's goal is to reduce the cost of
788: compilation and to make debugging produce the expected results.
789: Statements are independent: if you stop the program with a breakpoint
790: between statements, you can then assign a new value to any variable or
791: change the program counter to any other statement in the function and
792: get exactly the results you would expect from the source code.
793:
794: Without @samp{-O}, only variables declared @code{register} are
795: allocated in registers. The resulting compiled code is a little worse
796: than produced by PCC without @samp{-O}.
797:
798: With @samp{-O}, the compiler tries to reduce code size and execution
799: time.
800:
801: Some of the @samp{-f} options described below turn specific kinds of
802: optimization on or off.
803:
804: @item -g
805: Produce debugging information in the operating system's native format
806: (for DBX or SDB). GDB also can work with this debugging information.
807:
808: Unlike most other C compilers, GNU CC allows you to use @samp{-g} with
809: @samp{-O}. The shortcuts taken by optimized code may occasionally
810: produce surprising results: some variables you declared may not exist
811: at all; flow of control may briefly move where you did not expect it;
812: some statements may not be executed because they compute constant
813: results or their values were already at hand; some statements may
814: execute in different places because they were moved out of loops.
815: Nevertheless it proves possible to debug optimized output. This makes
816: it reasonable to use the optimizer for programs that might have bugs.
817:
818: @item -gg
1.1.1.8 root 819: Produce debugging information in the old GDB format. This is obsolete.
1.1 root 820:
821: @item -w
822: Inhibit all warning messages.
823:
824: @item -W
825: Print extra warning messages for these events:
826:
827: @itemize @bullet
828: @item
829: An automatic variable is used without first being initialized.
830:
831: These warnings are possible only in optimizing compilation,
832: because they require data flow information that is computed only
1.1.1.6 root 833: when optimizing. If you don't specify @samp{-O}, you simply won't
834: get these warnings.
835:
836: These warnings occur only for variables that are candidates for
837: register allocation. Therefore, they do not occur for a variable that
838: is declared @code{volatile}, or whose address is taken, or whose size
839: is other than 1, 2, 4 or 8 bytes. Also, they do not occur for
840: structures, unions or arrays, even when they are in registers.
841:
842: Note that there may be no warning about a variable that is used only
843: to compute a value that itself is never used, because such
844: computations may be deleted by data flow analysis before the warnings
845: are printed.
1.1 root 846:
847: These warnings are made optional because GNU CC is not smart
848: enough to see all the reasons why the code might be correct
849: despite appearing to have an error. Here is one example of how
850: this can happen:
851:
852: @example
853: @{
854: int x;
855: switch (y)
856: @{
857: case 1: x = 1;
858: break;
859: case 2: x = 4;
860: break;
861: case 3: x = 5;
862: @}
863: foo (x);
864: @}
865: @end example
866:
867: @noindent
868: If the value of @code{y} is always 1, 2 or 3, then @code{x} is
869: always initialized, but GNU CC doesn't know this. Here is
870: another common case:
871:
872: @example
873: @{
874: int save_y;
875: if (change_y) save_y = y, y = new_y;
876: @dots{}
877: if (change_y) y = save_y;
878: @}
879: @end example
880:
881: @noindent
882: This has no bug because @code{save_y} is used only if it is set.
883:
1.1.1.5 root 884: Some spurious warnings can be avoided if you declare as
885: @code{volatile} all the functions you use that never return.
886: @xref{Function Attributes}.
887:
1.1 root 888: @item
889: A nonvolatile automatic variable might be changed by a call to
890: @code{longjmp}. These warnings as well are possible only in
891: optimizing compilation.
892:
893: The compiler sees only the calls to @code{setjmp}. It cannot know
894: where @code{longjmp} will be called; in fact, a signal handler could
895: call it at any point in the code. As a result, you may get a warning
896: even when there is in fact no problem because @code{longjmp} cannot
897: in fact be called at the place which would cause a problem.
898:
899: @item
900: A function can return either with or without a value. (Falling
901: off the end of the function body is considered returning without
1.1.1.6 root 902: a value.) For example, this function would evoke such a
1.1 root 903: warning:
904:
905: @example
906: foo (a)
907: @{
908: if (a > 0)
909: return a;
910: @}
911: @end example
912:
913: Spurious warnings can occur because GNU CC does not realize that
914: certain functions (including @code{abort} and @code{longjmp})
915: will never return.
1.1.1.4 root 916:
917: @item
918: An expression-statement contains no side effects.
1.1 root 919: @end itemize
920:
921: In the future, other useful warnings may also be enabled by this
922: option.
923:
924: @item -Wimplicit
925: Warn whenever a function is implicitly declared.
926:
927: @item -Wreturn-type
928: Warn whenever a function is defined with a return-type that defaults
929: to @code{int}. Also warn about any @code{return} statement with no
930: return-value in a function whose return-type is not @code{void}.
931:
932: @item -Wunused
1.1.1.5 root 933: Warn whenever a local variable is unused aside from its declaration,
1.1.1.8 root 934: whenever a function is declared static but never defined, and whenever
935: a statement computes a result that is explicitly not used.
1.1.1.7 root 936:
937: @item -Wswitch
938: Warn whenever a @code{switch} statement has an index of enumeral type
939: and lacks a @code{case} for one or more of the named codes of that
940: enumeration. (The presence of a @code{default} label prevents this
941: warning.) @code{case} labels outside the enumeration range also
942: provoke warnings when this option is used.
943:
1.1 root 944: @item -Wcomment
945: Warn whenever a comment-start sequence @samp{/*} appears in a comment.
946:
1.1.1.7 root 947: @item -Wtrigraphs
948: Warn if any trigraphs are encountered (assuming they are enabled).
949:
1.1 root 950: @item -Wall
1.1.1.8 root 951: All of the above @samp{-W} options combined. These are all the
952: options which pertain to usage that we recommend avoiding and that we
953: believe is easy to avoid, even in conjunction with macros.
954:
955: The other @samp{-W@dots{}} options below are not implied by @samp{-Wall}
956: because certain kinds of useful macros are almost impossible to write
957: without causing those warnings.
958:
959: @item -Wshadow
960: Warn whenever a local variable shadows another local variable.
961:
962: @item -Wid-clash-@var{len}
963: Warn whenever two distinct identifiers match in the first @var{len}
964: characters. This may help you prepare a program that will compile
965: with certain obsolete, brain-damaged compilers.
966:
967: @item -Wpointer-arith
968: Warn about anything that depends on the ``size of'' a function type or
969: of @code{void}. GNU C assigns these types a size of 1, for
970: convenience in calculations with @code{void *} pointers and pointers
971: to functions.
1.1 root 972:
1.1.1.6 root 973: @item -Wcast-qual
974: Warn whenever a pointer is cast so as to remove a type qualifier from
975: the target type. For example, warn if a @code{const char *} is cast
976: to an ordinary @code{char *}.
977:
1.1 root 978: @item -Wwrite-strings
979: Give string constants the type @code{const char[@var{length}]} so that
980: copying the address of one into a non-@code{const} @code{char *}
981: pointer will get a warning. These warnings will help you find at
982: compile time code that can try to write into a string constant, but
983: only if you have been very careful about using @code{const} in
984: declarations and prototypes. Otherwise, it will just be a nuisance;
985: this is why we did not make @samp{-Wall} request these warnings.
986:
987: @item -p
988: Generate extra code to write profile information suitable for the
989: analysis program @code{prof}.
990:
991: @item -pg
992: Generate extra code to write profile information suitable for the
993: analysis program @code{gprof}.
994:
1.1.1.6 root 995: @item -a
1.1.1.8 root 996: Generate extra code to write profile information for basic blocks, which
997: will record the number of times each basic block is executed. This data
998: could be analyzed by a program like @code{tcov}. Note, however, that
999: the format of the data is not what @code{tcov} expects. Eventually GNU
1.1.1.6 root 1000: @code{gprof} should be extended to process this data.
1001:
1.1 root 1002: @item -l@var{library}
1003: Search a standard list of directories for a library named
1004: @var{library}, which is actually a file named
1005: @file{lib@var{library}.a}. The linker uses this file as if it
1006: had been specified precisely by name.
1007:
1008: The directories searched include several standard system directories
1009: plus any that you specify with @samp{-L}.
1010:
1011: Normally the files found this way are library files---archive files
1012: whose members are object files. The linker handles an archive file by
1013: scanning through it for members which define symbols that have so far
1014: been referenced but not defined. But if the file that is found is an
1015: ordinary object file, it is linked in the usual fashion. The only
1016: difference between using an @samp{-l} option and specifying a file name
1017: is that @samp{-l} searches several directories.
1018:
1019: @item -L@var{dir}
1020: Add directory @var{dir} to the list of directories to be searched
1021: for @samp{-l}.
1022:
1023: @item -nostdlib
1.1.1.9 ! root 1024: Don't use the standard system libraries and startup files when linking.
! 1025: Only the files you specify will be passed to the linker.
1.1 root 1026:
1027: @item -m@var{machinespec}
1028: Machine-dependent option specifying something about the type of target
1029: machine. These options are defined by the macro
1030: @code{TARGET_SWITCHES} in the machine description. The default for
1031: the options is also defined by that macro, which enables you to change
1032: the defaults.@refill
1033:
1034: These are the @samp{-m} options defined in the 68000 machine
1035: description:
1036:
1037: @table @samp
1038: @item -m68020
1039: @itemx -mc68020
1040: Generate output for a 68020 (rather than a 68000). This is the
1041: default if you use the unmodified sources.
1042:
1043: @item -m68000
1044: @item -mc68000
1045: Generate output for a 68000 (rather than a 68020).
1046:
1047: @item -m68881
1048: Generate output containing 68881 instructions for floating point.
1049: This is the default if you use the unmodified sources.
1050:
1051: @item -mfpa
1052: Generate output containing Sun FPA instructions for floating point.
1053:
1054: @item -msoft-float
1055: Generate output containing library calls for floating point.
1056:
1057: @item -mshort
1058: Consider type @code{int} to be 16 bits wide, like @code{short int}.
1059:
1060: @item -mnobitfield
1061: Do not use the bit-field instructions. @samp{-m68000} implies
1062: @samp{-mnobitfield}.
1063:
1064: @item -mbitfield
1065: Do use the bit-field instructions. @samp{-m68020} implies
1066: @samp{-mbitfield}. This is the default if you use the unmodified
1067: sources.
1068:
1069: @item -mrtd
1070: Use a different function-calling convention, in which functions
1071: that take a fixed number of arguments return with the @code{rtd}
1072: instruction, which pops their arguments while returning. This
1073: saves one instruction in the caller since there is no need to pop
1074: the arguments there.
1075:
1076: This calling convention is incompatible with the one normally
1077: used on Unix, so you cannot use it if you need to call libraries
1078: compiled with the Unix compiler.
1079:
1080: Also, you must provide function prototypes for all functions that
1081: take variable numbers of arguments (including @code{printf});
1082: otherwise incorrect code will be generated for calls to those
1083: functions.
1084:
1085: In addition, seriously incorrect code will result if you call a
1086: function with too many arguments. (Normally, extra arguments are
1087: harmlessly ignored.)
1088:
1089: The @code{rtd} instruction is supported by the 68010 and 68020
1090: processors, but not by the 68000.
1091: @end table
1092:
1093: These @samp{-m} options are defined in the Vax machine description:
1094:
1095: @table @samp
1096: @item -munix
1097: Do not output certain jump instructions (@code{aobleq} and so on)
1098: that the Unix assembler for the Vax cannot handle across long
1099: ranges.
1100:
1101: @item -mgnu
1102: Do output those jump instructions, on the assumption that you
1103: will assemble with the GNU assembler.
1104:
1105: @item -mg
1106: Output code for g-format floating point numbers instead of d-format.
1107: @end table
1108:
1.1.1.5 root 1109: These @samp{-m} switches are supported on the Sparc:
1110:
1111: @table @samp
1112: @item -mfpu
1113: Generate output containing floating point instructions. This is the
1114: default if you use the unmodified sources.
1115:
1.1.1.9 ! root 1116: @ignore
1.1.1.5 root 1117: @item -msoft-float
1118: Generate output containing library calls for floating point.
1119:
1.1.1.9 ! root 1120: @end ignore
1.1.1.5 root 1121: @item -mno-epilogue
1.1.1.6 root 1122: Generate separate return instructions for @code{return} statements.
1123: This has both advantages and disadvantages; I don't recall what they
1124: are.
1.1.1.5 root 1125: @end table
1126:
1127: These @samp{-m} options are defined in the Convex machine description:
1128:
1129: @table @samp
1130: @item -mc1
1131: Generate output for a C1. This is the default when the compiler is
1132: configured for a C1.
1133:
1134: @item -mc2
1135: Generate output for a C2. This is the default when the compiler is
1136: configured for a C2.
1137:
1138: @item -margcount
1139: Generate code which puts an argument count in the word preceding each
1140: argument list. Some nonportable Convex and Vax programs need this
1141: word. (Debuggers don't; this info is in the symbol table.)
1142:
1143: @item -mnoargcount
1144: Omit the argument count word. This is the default if you use the
1145: unmodified sources.
1146: @end table
1147:
1.1 root 1148: @item -f@var{flag}
1.1.1.4 root 1149: Specify machine-independent flags. Most flags have both positive and
1150: negative forms; the negative form of @samp{-ffoo} would be
1151: @samp{-fno-foo}. In the table below, only one of the forms is
1152: listed---the one which is not the default. You can figure out the
1153: other form by either removing @samp{no-} or adding it.
1.1 root 1154:
1155: @table @samp
1.1.1.6 root 1156: @item -fpcc-struct-return
1157: Use the same convention for returning @code{struct} and @code{union}
1158: values that is used by the usual C compiler on your system. This
1159: convention is less efficient for small structures, and on many
1160: machines it fails to be reentrant; but it has the advantage of
1161: allowing intercallability between GCC-compiled code and PCC-compiled
1162: code.
1163:
1.1 root 1164: @item -ffloat-store
1165: Do not store floating-point variables in registers. This
1166: prevents undesirable excess precision on machines such as the
1167: 68000 where the floating registers (of the 68881) keep more
1168: precision than a @code{double} is supposed to have.
1169:
1170: For most programs, the excess precision does only good, but a few
1171: programs rely on the precise definition of IEEE floating point.
1172: Use @samp{-ffloat-store} for such programs.
1173:
1174: @item -fno-asm
1175: Do not recognize @code{asm}, @code{inline} or @code{typeof} as a
1.1.1.7 root 1176: keyword. These words may then be used as identifiers. You can
1.1.1.8 root 1177: use @code{__asm__}, @code{__inline__} and @code{__typeof__} instead.
1.1 root 1178:
1179: @item -fno-defer-pop
1180: Always pop the arguments to each function call as soon as that
1181: function returns. Normally the compiler (when optimizing) lets
1182: arguments accumulate on the stack for several function calls and
1183: pops them all at once.
1184:
1185: @item -fstrength-reduce
1186: Perform the optimizations of loop strength reduction and
1187: elimination of iteration variables.
1188:
1189: @item -fcombine-regs
1190: Allow the combine pass to combine an instruction that copies one
1191: register into another. This might or might not produce better
1192: code when used in addition to @samp{-O}. I am interested in
1193: hearing about the difference this makes.
1194:
1195: @item -fforce-mem
1196: Force memory operands to be copied into registers before doing
1197: arithmetic on them. This may produce better code by making all
1198: memory references potential common subexpressions. When they are
1199: not common subexpressions, instruction combination should
1200: eliminate the separate register-load. I am interested in hearing
1201: about the difference this makes.
1202:
1203: @item -fforce-addr
1204: Force memory address constants to be copied into registers before
1205: doing arithmetic on them. This may produce better code just as
1206: @samp{-fforce-mem} may. I am interested in hearing about the
1207: difference this makes.
1208:
1209: @item -fomit-frame-pointer
1210: Don't keep the frame pointer in a register for functions that
1211: don't need one. This avoids the instructions to save, set up and
1212: restore frame pointers; it also makes an extra register available
1213: in many functions. @strong{It also makes debugging impossible.}
1214:
1215: On some machines, such as the Vax, this flag has no effect,
1216: because the standard calling sequence automatically handles the
1217: frame pointer and nothing is saved by pretending it doesn't
1218: exist. The machine-description macro
1219: @code{FRAME_POINTER_REQUIRED} controls whether a target machine
1220: supports this flag. @xref{Registers}.@refill
1221:
1222: @item -finline-functions
1223: Integrate all simple functions into their callers. The compiler
1224: heuristically decides which functions are simple enough to be
1225: worth integrating in this way.
1226:
1227: If all calls to a given function are integrated, and the function
1228: is declared @code{static}, then the function is normally not
1229: output as assembler code in its own right.
1230:
1.1.1.6 root 1231: @item -fcaller-saves
1232: Enable values to be allocated in registers that will be clobbered by
1233: function calls, by emitting extra instructions to save and restore the
1234: registers around such calls. Such allocation is done only when it
1235: seems to result in better code than would otherwise be produced.
1236:
1237: This option is enabled by default on certain machines, usually those
1238: which have no call-preserved registers to use instead.
1239:
1.1 root 1240: @item -fkeep-inline-functions
1241: Even if all calls to a given function are integrated, and the
1242: function is declared @code{static}, nevertheless output a
1243: separate run-time callable version of the function.
1244:
1245: @item -fwritable-strings
1.1.1.8 root 1246: Store string constants in the writable data segment and don't uniquize
1247: them. This is for compatibility with old programs which assume they can
1248: write into string constants. @samp{-traditional} also has this effect.
1249:
1250: Writing into string constants is a very bad idea; ``constants'' should
1251: be constant.
1.1 root 1252:
1.1.1.4 root 1253: @item -fcond-mismatch
1254: Allow conditional expressions with mismatched types in the second and
1255: third arguments. The value of such an expression is void.
1256:
1.1 root 1257: @item -fno-function-cse
1258: Do not put function addresses in registers; make each instruction
1259: that calls a constant function contain the function's address
1260: explicitly.
1261:
1262: This option results in less efficient code, but some strange
1263: hacks that alter the assembler output may be confused by the
1264: optimizations performed when this option is not used.
1265:
1266: @item -fvolatile
1267: Consider all memory references through pointers to be volatile.
1268:
1.1.1.4 root 1269: @item -fshared-data
1270: Requests that the data and non-@code{const} variables of this
1271: compilation be shared data rather than private data. The distinction
1272: makes sense only on certain operating systems, where shared data is
1273: shared between processes running the same program, while private data
1274: exists in one copy per process.
1275:
1.1 root 1276: @item -funsigned-char
1.1.1.4 root 1277: Let the type @code{char} be the unsigned, like @code{unsigned char}.
1.1 root 1278:
1279: Each kind of machine has a default for what @code{char} should
1280: be. It is either like @code{unsigned char} by default or like
1281: @code{signed char} by default. (Actually, at present, the
1282: default is always signed.)
1283:
1284: The type @code{char} is always a distinct type from either
1285: @code{signed char} or @code{unsigned char}, even though its
1286: behavior is always just like one of those two.
1287:
1.1.1.4 root 1288: Note that this is equivalent to @samp{-fno-signed-char}, which is the
1289: negative form of @samp{-fsigned-char}.
1290:
1.1 root 1291: @item -fsigned-char
1292: Let the type @code{char} be signed, like @code{signed char}.
1293:
1.1.1.4 root 1294: Note that this is equivalent to @samp{-fno-unsigned-char}, which is
1295: the negative form of @samp{-funsigned-char}.
1296:
1.1.1.8 root 1297: @item -fdelayed-branch
1298: If supported for the target machine, attempt to reorder instructions
1299: to exploit instruction slots available after delayed branch
1300: instructions.
1301:
1.1 root 1302: @item -ffixed-@var{reg}
1303: Treat the register named @var{reg} as a fixed register; generated
1304: code should never refer to it (except perhaps as a stack pointer,
1305: frame pointer or in some other fixed role).
1306:
1307: @var{reg} must be the name of a register. The register names
1308: accepted are machine-specific and are defined in the
1309: @code{REGISTER_NAMES} macro in the machine description macro
1310: file.
1311:
1.1.1.4 root 1312: This flag does not have a negative form, because it specifies a
1313: three-way choice.
1314:
1.1 root 1315: @item -fcall-used-@var{reg}
1316: Treat the register named @var{reg} as an allocatable register
1317: that is clobbered by function calls. It may be allocated for
1318: temporaries or variables that do not live across a call.
1319: Functions compiled this way will not save and restore the
1320: register @var{reg}.
1321:
1322: Use of this flag for a register that has a fixed pervasive role
1323: in the machine's execution model, such as the stack pointer or
1324: frame pointer, will produce disastrous results.
1325:
1.1.1.4 root 1326: This flag does not have a negative form, because it specifies a
1327: three-way choice.
1328:
1.1 root 1329: @item -fcall-saved-@var{reg}
1330: Treat the register named @var{reg} as an allocatable register
1331: saved by functions. It may be allocated even for temporaries or
1332: variables that live across a call. Functions compiled this way
1333: will save and restore the register @var{reg} if they use it.
1334:
1335: Use of this flag for a register that has a fixed pervasive role
1336: in the machine's execution model, such as the stack pointer or
1337: frame pointer, will produce disastrous results.
1338:
1339: A different sort of disaster will result from the use of this
1340: flag for a register in which function values may be returned.
1.1.1.4 root 1341:
1342: This flag does not have a negative form, because it specifies a
1343: three-way choice.
1.1 root 1344: @end table
1345:
1346: @item -d@var{letters}
1347: Says to make debugging dumps at times specified by @var{letters}.
1348: Here are the possible letters:
1349:
1350: @table @samp
1351: @item r
1352: Dump after RTL generation.
1353: @item j
1354: Dump after first jump optimization.
1355: @item s
1356: Dump after CSE (including the jump optimization that sometimes
1357: follows CSE).
1358: @item L
1359: Dump after loop optimization.
1360: @item f
1361: Dump after flow analysis.
1362: @item c
1363: Dump after instruction combination.
1364: @item l
1365: Dump after local register allocation.
1366: @item g
1367: Dump after global register allocation.
1.1.1.8 root 1368: @item d
1369: Dump after delayed branch scheduling.
1370: @item J
1371: Dump after last jump optimization.
1.1 root 1372: @item m
1373: Print statistics on memory usage, at the end of the run.
1374: @end table
1375:
1376: @item -pedantic
1377: Issue all the warnings demanded by strict ANSI standard C; reject
1378: all programs that use forbidden extensions.
1379:
1380: Valid ANSI standard C programs should compile properly with or without
1381: this option (though a rare few will require @samp{-ansi}). However,
1382: without this option, certain GNU extensions and traditional C features
1383: are supported as well. With this option, they are rejected. There is
1384: no reason to @i{use} this option; it exists only to satisfy pedants.
1.1.1.5 root 1385:
1.1.1.8 root 1386: @samp{-pedantic} does not cause warning messages for use of the
1387: alternate keywords whose names begin and end with @samp{__}.
1388: @xref{Alternate Keywords}.
1389:
1.1.1.5 root 1390: @item -static
1391: On Suns running version 4, this prevents linking with the shared
1392: libraries. (@samp{-g} has the same effect.)
1.1 root 1393: @end table
1394:
1395: These options control the C preprocessor, which is run on each C source
1396: file before actual compilation. If you use the @samp{-E} option, nothing
1397: is done except C preprocessing. Some of these options make sense only
1398: together with @samp{-E} because they request preprocessor output that is
1399: not suitable for actual compilation.
1400:
1401: @table @samp
1402: @item -C
1403: Tell the preprocessor not to discard comments. Used with the
1404: @samp{-E} option.
1405:
1406: @item -I@var{dir}
1407: Search directory @var{dir} for include files.
1408:
1409: @item -I-
1410: Any directories specified with @samp{-I} options before the @samp{-I-}
1411: option are searched only for the case of @samp{#include "@var{file}"};
1412: they are not searched for @samp{#include <@var{file}>}.
1413:
1414: If additional directories are specified with @samp{-I} options after
1415: the @samp{-I-}, these directories are searched for all @samp{#include}
1416: directives. (Ordinarily @emph{all} @samp{-I} directories are used
1417: this way.)
1418:
1419: In addition, the @samp{-I-} option inhibits the use of the current
1.1.1.8 root 1420: directory (where the current input file came from) as the first search
1421: directory for @samp{#include "@var{file}"}. There is no way to override
1422: this effect of @samp{-I-}. With @samp{-I.} you can specify searching
1423: the directory which was current when the compiler was invoked. That is
1424: not exactly the same as what the preprocessor does by default, but it is
1425: often satisfactory.
1426:
1427: @samp{-I-} does not inhibit the use of the standard system directories
1428: for header files. Thus, @samp{-I-} and @samp{-nostdinc} are
1429: independent.
1430:
1431: @item -i @var{file}
1432: Process @var{file} as input, discarding the resulting output, before
1433: processing the regular input file. Because the output generated from
1434: @var{file} is discarded, the only effect of @samp{-i @var{file}} is to
1435: make the macros defined in @var{file} available for use in the main
1436: input.
1.1 root 1437:
1438: @item -nostdinc
1439: Do not search the standard system directories for header files. Only
1440: the directories you have specified with @samp{-I} options (and the
1441: current directory, if appropriate) are searched.
1442:
1443: Between @samp{-nostdinc} and @samp{-I-}, you can eliminate all
1444: directories from the search path except those you specify.
1445:
1446: @item -M
1447: Tell the preprocessor to output a rule suitable for @code{make}
1448: describing the dependencies of each source file. For each source
1449: file, the preprocessor outputs one @code{make}-rule whose target is
1450: the object file name for that source file and whose dependencies are
1451: all the files @samp{#include}d in it. This rule may be a single line
1452: or may be continued with @samp{\}-newline if it is long.
1453:
1454: @samp{-M} implies @samp{-E}.
1455:
1456: @item -MM
1457: Like @samp{-M} but the output mentions only the user-header files
1458: included with @samp{#include "@var{file}"}. System header files
1459: included with @samp{#include <@var{file}>} are omitted.
1460:
1461: @samp{-MM} implies @samp{-E}.
1462:
1463: @item -D@var{macro}
1.1.1.8 root 1464: Define macro @var{macro} with the string @samp{1} as its definition.
1.1 root 1465:
1466: @item -D@var{macro}=@var{defn}
1467: Define macro @var{macro} as @var{defn}.
1468:
1469: @item -U@var{macro}
1470: Undefine macro @var{macro}.
1471:
1.1.1.7 root 1472: @item -trigraphs
1.1 root 1473: Support ANSI C trigraphs. You don't want to know about this
1474: brain-damage. The @samp{-ansi} option also has this effect.
1475: @end table
1476:
1477: @node Installation, Trouble, Options, Top
1478: @chapter Installing GNU CC
1479:
1480: Here is the procedure for installing GNU CC on a Unix system.
1.1.1.8 root 1481:
1.1 root 1482: @menu
1.1.1.8 root 1483: * Other Dir:: Compiling in a separate directory (not where the source is).
1484: * Sun Install:: See below for installation on the Sun.
1485: * 3B1 Install:: See below for installation on the 3B1.
1.1 root 1486: * VMS Install:: See below for installation on VMS.
1.1.1.9 ! root 1487: * HPUX Install:: See below for installation on HPUX.
1.1 root 1488: @end menu
1489: @iftex
1.1.1.9 ! root 1490: See below for VMS systems, and modified procedures needed on Sun
! 1491: systems, 3b1 machines and HPUX. The following section says how to
! 1492: compile in a separate directory on Unix; here we assume you compile in
! 1493: the same directory that contains the source files.
1.1 root 1494: @end iftex
1495:
1496: @enumerate
1497: @item
1498: Edit @file{Makefile}. If you are using HPUX, or any form of system V,
1499: you must make a few changes described in comments at the beginning of
1.1.1.9 ! root 1500: the file. Genix requires changes also, and so does the Pyramid.
1.1 root 1501:
1502: @item
1503: On a Sequent system, go to the Berkeley universe.
1504:
1505: @item
1.1.1.2 root 1506: Choose configuration files. The easy way to do this is to run the
1.1.1.8 root 1507: command file @file{config.gcc} with a single argument, which specifies
1508: the type of machine (and in some cases which operating system).
1.1.1.4 root 1509:
1510: Here is a list of the possible arguments:
1511:
1512: @table @samp
1513: @item vax
1514: Vaxes running BSD.
1515: @item vms
1516: Vaxes running VMS.
1517: @item vax-sysv
1518: Vaxes running system V.
1519: @item i386-sysv
1520: Intel 386 PCs running system V.
1.1.1.5 root 1521: @item i386-sysv-gas
1522: Intel 386 PCs running system V, using the GNU assembler and GNU
1523: linker.
1.1.1.6 root 1524: @item sequent-i386
1.1.1.4 root 1525: Sequent with Intel 386 processors.
1.1.1.8 root 1526: @item i386-aix
1527: Intel 386 PCs or PS/2s running AIX.
1.1.1.4 root 1528: @item sun2
1529: Sun 2 running system version 2 or 3.
1530: @item sun3
1.1.1.5 root 1531: Sun 3 running system version 2 or 3, with 68881.
1.1.1.7 root 1532: Note there we do not provide a configuration file to use an FPA
1.1.1.8 root 1533: by default, because programs that establish signal handlers for
1.1.1.7 root 1534: floating point traps inherently cannot work with the FPA.
1.1.1.5 root 1535: @item sun3-nfp
1536: Sun 3 running system version 2 or 3, without 68881.
1.1.1.4 root 1537: @item sun4
1.1.1.8 root 1538: Sun 4 running system version 2 or 3. @xref{Incompatibilities},
1539: for calling convention incompatibilities on the Sun 4 (sparc).
1.1.1.4 root 1540: @item sun2-os4
1541: Sun 2 running system version 4.
1542: @item sun3-os4
1.1.1.5 root 1543: Sun 3 running system version 4, with 68881.
1544: @item sun3-nfp-os4
1545: Sun 3 running system version 4, without 68881.
1.1.1.4 root 1546: @item sun4-os4
1.1.1.8 root 1547: Sun 4 running system version 4. @xref{Incompatibilities},
1548: for calling convention incompatibilities on the Sun 4 (sparc).
1.1.1.4 root 1549: @item sun386
1550: Sun 386 (``roadrunner'').
1.1.1.5 root 1551: @item alliant
1.1.1.8 root 1552: Alliant FX/8 computer. Note that the standard installed C compiler in
1553: Concentrix 5.0 has a bug which prevent it from compiling GNU CC
1554: correctly. You can patch the compiler bug as follows:
1555:
1556: @example
1557: cp /bin/pcc ./pcc
1.1.1.9 ! root 1558: adb -w ./pcc - << EOF
1.1.1.8 root 1559: 15f6?w 6610
1560: EOF
1561: @end example
1562:
1563: Then you must use the @samp{-ip12} option when compiling GNU CC
1564: with the patched compiler, as shown here:
1565:
1566: @example
1567: make CC="./pcc -ip12" CFLAGS=-w
1568: @end example
1569:
1570: Note also that Alliant's version of DBX does not manage to work with the
1571: output from GNU CC.
1572: @item tahoe
1573: The tahoe computer (running BSD, and using DBX).
1574: @item decstation
1575: The DEC 3100 Mips machine (``pmax''). Note that GNU CC cannot generate
1576: debugging information in the unusual format used on the Mips.
1577: @item mips-sysv
1578: The Mips computer, RS series, with the System V environment as default.
1579: Note that GNU CC cannot generate debugging information in the unusual
1580: format used on the Mips.
1581: @item mips-bsd43
1582: The Mips computer, RS series, with the BSD 4.3 environment as default.
1583: Note that GNU CC cannot generate debugging information in the unusual
1584: format used on the Mips.
1.1.1.7 root 1585: @item mips
1.1.1.8 root 1586: The Mips computer, M series. Note that GNU CC cannot generate debugging
1587: information in the unusual format used on the Mips.
1588: @item iris
1589: The Mips computer, as delivered by Iris. Note that GNU CC cannot
1590: generate debugging information in the unusual format used on the Mips.
1.1.1.5 root 1591: @item convex-c1
1592: Convex C1 computer.
1593: @item convex-c2
1594: Convex C2 computer.
1.1.1.8 root 1595: @item pyramid
1596: Pyramid computer.
1.1.1.4 root 1597: @item hp9k320
1.1.1.7 root 1598: HP 9000 series 300 using HPUX assembler. Note there is no
1599: support in GNU CC for HP's debugger; thus, @samp{-g} is not
1600: available in this configuration.
1.1.1.8 root 1601: @item hp9k320-gas
1.1.1.4 root 1602: HP 9000 series 300 using GNU assembler, linker and debugger.
1.1.1.7 root 1603: This requires the HP-adapt package, which is available along with
1604: the GNU linker as part of the ``binutils'' distribution.
1605: This is on the GNU CC distribution tape.
1.1.1.8 root 1606: @item hp9k320-old
1607: HP 9000 series 300 using HPUX assembler, in operating system versions
1608: older than 6.5. Note there is no support in GNU CC for HP's debugger;
1609: thus, @samp{-g} is not available in this configuration.
1610: @item hp9k320-bsd
1611: HP 9000 series 300 running BSD.
1.1.1.4 root 1612: @item isi68
1.1.1.8 root 1613: ISI 68000 or 68020 system with a 68881.
1614: @item isi68-nfp
1615: ISI 68000 or 68020 system without a 68881.
1.1.1.4 root 1616: @item news800
1617: Sony NEWS 68020 system.
1.1.1.6 root 1618: @item next
1619: NeXT system.
1.1.1.7 root 1620: @item altos
1621: Altos 3068. Note that you must use the GNU assembler, linker and
1622: debugger, with COFF-encapsulation. Also, you must fix a kernel
1623: bug. Details in the file @file{ALTOS-README}.
1.1.1.4 root 1624: @item 3b1
1.1.1.8 root 1625: AT&T 3b1, a.k.a. 7300 PC. Note that special procedures are needed
1626: to compile GNU CC with this machine's standard C compiler, due to
1627: bugs in that compiler. @xref{3b1 Install}. You can bootstrap it
1628: more easily with previous versions of GNU CC if you have them.
1.1.1.9 ! root 1629: @item 3b1-gas
! 1630: AT&T 3b1 using the GNU assembler.
1.1.1.4 root 1631: @item sequent-ns32k
1632: Sequent containing ns32000 processors.
1633: @item encore
1634: Encore ns32000 system.
1635: @item genix
1636: National Semiconductor ns32000 system.
1637: @item 88000
1638: Motorola 88000 processor. This port is not finished.
1639: @end table
1.1.1.2 root 1640:
1.1.1.4 root 1641: Here we spell out what files need to be set up:
1.1 root 1642:
1643: @itemize @bullet
1644: @item
1645: Make a symbolic link named @file{config.h} to the top-level
1646: config file for the machine you are using (@pxref{Config}). This
1647: file is responsible for defining information about the host
1648: machine. It includes @file{tm.h}.
1649:
1.1.1.7 root 1650: The file is located in the subdirectory @file{config}. Its name
1651: should be @file{xm-@var{machine}.h}, with these exceptions:
1.1 root 1652:
1653: @table @file
1.1.1.3 root 1654: @item xm-vms.h
1.1 root 1655: for vaxen running VMS.
1.1.1.3 root 1656: @item xm-vaxv.h
1.1 root 1657: for vaxen running system V.
1.1.1.3 root 1658: @item xm-i386v.h
1.1 root 1659: for Intel 80386's running system V.
1.1.1.3 root 1660: @item xm-sun386i.h
1661: for Sun roadrunner running any version of the operating system.
1662: @item xm-hp9k320.h
1.1 root 1663: for the HP 9000 series 300.
1.1.1.4 root 1664: @item xm-genix.h
1.1 root 1665: for the ns32000 running Genix
1666: @end table
1667:
1668: If your system does not support symbolic links, you might want to
1669: set up @file{config.h} to contain a @samp{#include} command which
1670: refers to the appropriate file.
1671:
1672: @item
1673: Make a symbolic link named @file{tm.h} to the machine-description
1.1.1.7 root 1674: macro file for your machine. It should be in the subdirectory
1675: @file{config} and its name should be @file{tm-@var{machine}.h}.
1.1 root 1676:
1677: If your system is a 68000, don't use the file @file{tm-m68k.h}
1678: directly. Instead, use one of these files:
1679:
1680: @table @file
1681: @item tm-sun3.h
1.1.1.5 root 1682: for Sun 3 machines with 68881.
1683: @item tm-sun3-nfp.h
1684: for Sun 3 machines with no hardware floating point.
1.1.1.8 root 1685: @item tm-sun3os3.h
1686: for Sun 3 machines with 68881, running Sunos version 3.
1687: @item tm-sun3os3nf.h
1688: for Sun 3 machines with no hardware floating point, running Sunos
1689: version 3.
1.1 root 1690: @item tm-sun2.h
1691: for Sun 2 machines.
1692: @item tm-3b1.h
1693: for AT&T 3b1 (aka 7300 Unix PC).
1694: @item tm-isi68.h
1.1.1.3 root 1695: for Integrated Solutions systems. This file assumes you
1696: use the GNU assembler.
1.1.1.8 root 1697: @item tm-isi68-nfp.h
1698: for Integrated Solutions systems without a 68881. This file assumes you
1699: use the GNU assembler.
1.1 root 1700: @item tm-news800.h
1.1.1.8 root 1701: for Sony NEWS systems.
1.1 root 1702: @item tm-hp9k320.h
1703: for HPUX systems, if you are using GNU CC with the system's
1704: assembler and linker.
1705: @item tm-hp9k320g.h
1706: for HPUX systems, if you are using the GNU assembler, linker and
1707: other utilities. Not all of the pieces of GNU software needed
1708: for this mode of operation are as yet in distribution; full
1709: instructions will appear here in the future.@refill
1710: @end table
1711:
1712: For the vax, use @file{tm-vax.h} on BSD Unix, @file{tm-vaxv.h} on
1713: system V, or @file{tm-vms.h} on VMS.@refill
1714:
1715: For the Motorola 88000, use @file{tm-m88k.h}. The support for the
1.1.1.9 ! root 1716: 88000 does not currently work; it requires extensive changes which
! 1717: we hope to reconcile in version 2.
1.1 root 1718:
1719: For the 80386, don't use @file{tm-i386.h} directly. Use
1720: @file{tm-i386v.h} if the target machine is running system V,
1.1.1.5 root 1721: @file{tm-i386gas.h} if it is running system V but you are using the
1722: GNU assembler and linker, @file{tm-seq386.h} for a Sequent 386 system,
1723: or @file{tm-compaq.h} for a Compaq, or @file{tm-sun386i.h} for a Sun
1724: 386 system.
1.1 root 1725:
1.1.1.8 root 1726: For the Mips computer, there are five choices: @file{tm-mips.h} for the
1727: M series, @file{tm-mips-bsd.h} for the RS series with BSD,
1728: @file{tm-mips-sysv.h} for the RS series with System V, @file{tm-iris.h}
1729: for the Iris version of the machine, and @file{tm-decstatn.h} for the
1730: Decstation.
1731:
1.1 root 1732: For the 32000, use @file{tm-sequent.h} if you are using a Sequent
1733: machine, or @file{tm-encore.h} for an Encore machine, or
1.1.1.4 root 1734: @file{tm-genix.h} if you are using Genix version 3; otherwise, perhaps
1.1 root 1735: @file{tm-ns32k.h} will work for you.
1736:
1737: Note that Genix has bugs in @code{alloca} and @code{malloc}; you must
1738: get the compiled versions of these from GNU Emacs and edit GNU CC's
1739: @file{Makefile} to use them.
1740:
1741: Note that Encore systems are supported only under BSD.
1742:
1.1.1.6 root 1743: For Sparc (Sun 4) machines, use @file{tm-sparc.h} with operating system
1744: version 4, and @file{tm-sun4os3.h} with system version 3.
1745:
1.1 root 1746: @item
1747: Make a symbolic link named @file{md} to the machine description
1.1.1.7 root 1748: pattern file. It should be in the @file{config} subdirectory and its
1749: name should be @file{@var{machine}.md}; but @var{machine} is often not
1750: the same as the name used in the @file{tm.h} file because the
1751: @file{md} files are more general.
1.1 root 1752:
1753: @item
1754: Make a symbolic link named @file{aux-output.c} to the output
1.1.1.7 root 1755: subroutine file for your machine. It should be in the @file{config}
1756: subdirectory and its name should be @file{out-@var{machine}.c}.
1.1 root 1757: @end itemize
1758:
1759: @item
1760: Make sure the Bison parser generator is installed. (This is
1761: unnecessary if the Bison output files @file{c-parse.tab.c} and
1762: @file{cexp.c} are more recent than @file{c-parse.y} and @file{cexp.y}
1763: and you do not plan to change the @samp{.y} files.)
1764:
1.1.1.9 ! root 1765: Bison versions older than Sept 8, 1988 will produce incorrect output
1.1 root 1766: for @file{c-parse.tab.c}.
1767:
1768: @item
1769: Build the compiler. Just type @samp{make} in the compiler directory.
1770:
1.1.1.2 root 1771: Ignore any warnings you may see about ``statement not reached'' in the
1772: @file{insn-emit.c}; they are normal. Any other compilation errors may
1773: represent bugs in the port to your machine or operating system, and
1774: should be investigated and reported (@pxref{Bugs}).
1775:
1.1.1.9 ! root 1776: Some commercial compilers fail to compile GNU CC because they have bugs
! 1777: or limitations. For example, the Microsoft compiler is said to run out
! 1778: of macro space. Some Ultrix compilers run out of expression space; then
! 1779: you need to break up the statement where the problem happens.
1.1.1.7 root 1780:
1781: @item
1.1.1.5 root 1782: If you are using COFF-encapsulation, you must convert @file{gnulib} to
1783: a GNU-format library at this point. See the file @file{README-ENCAP}
1784: in the directory containing the GNU binary file utilities, for
1785: directions.
1786:
1787: @item
1.1 root 1788: Move the first-stage object files and executables into a subdirectory
1789: with this command:
1790:
1791: @example
1792: make stage1
1793: @end example
1794:
1795: The files are moved into a subdirectory named @file{stage1}.
1796: Once installation is complete, you may wish to delete these files
1797: with @code{rm -r stage1}.
1798:
1799: @item
1800: Recompile the compiler with itself, with this command:
1801:
1802: @example
1803: make CC=stage1/gcc CFLAGS="-g -O -Bstage1/"
1804: @end example
1805:
1806: On a 68000 or 68020 system lacking floating point hardware,
1807: unless you have selected a @file{tm.h} file that expects by default
1808: that there is no such hardware, do this instead:
1809:
1810: @example
1811: make CC=stage1/gcc CFLAGS="-g -O -Bstage1/ -msoft-float"
1812: @end example
1813:
1814: @item
1815: If you wish to test the compiler by compiling it with itself one more
1.1.1.7 root 1816: time, do this (in C shell):
1.1 root 1817:
1818: @example
1819: make stage2
1820: make CC=stage2/gcc CFLAGS="-g -O -Bstage2/"
1821: foreach file (*.o)
1822: cmp $file stage2/$file
1823: end
1824: @end example
1825:
1.1.1.7 root 1826: @noindent
1.1 root 1827: Aside from the @samp{-B} option, the options should be the same as
1828: when you made stage 2.
1829:
1.1.1.7 root 1830: The @code{foreach} command (written in C shell) will notify you if any of
1831: these stage 3 object files differs from those of stage 2. On BSD systems,
1832: any difference, no matter how innocuous, indicates that the stage 2
1833: compiler has compiled GNU CC incorrectly, and is therefore a potentially
1834: serious bug which you should investigate and report (@pxref{Bugs}).
1835:
1836: On systems that use COFF object files, bytes 5 to 8 will always be
1837: different, since it is a timestamp. On these systems, you can do the
1838: comparison as follows (in Bourne shell):
1839:
1840: @example
1841: for file in *.o; do
1842: echo $file
1843: tail +10 $file > foo1
1844: tail +10 stage2/$file > foo2
1845: cmp foo1 foo2
1846: done
1847: @end example
1848:
1.1 root 1849: @item
1850: Install the compiler driver, the compiler's passes and run-time support.
1851: You can use the following command:
1852:
1853: @example
1854: make install
1855: @end example
1856:
1857: @noindent
1858: This copies the files @file{cc1}, @file{cpp} and @file{gnulib} to
1859: files @file{gcc-cc1}, @file{gcc-cpp} and @file{gcc-gnulib} in
1860: directory @file{/usr/local/lib}, which is where the compiler driver
1861: program looks for them. It also copies the driver program @file{gcc}
1.1.1.6 root 1862: into the directory @file{/usr/local/bin}, so that it appears in typical
1.1 root 1863: execution search paths.@refill
1864:
1865: @strong{Warning: there is a bug in @code{alloca} in the Sun library.
1866: To avoid this bug, install the binaries of GNU CC that were compiled
1867: by GNU CC. They use @code{alloca} as a built-in function and never
1868: the one in the library.}
1869:
1870: @strong{Warning: the GNU CPP may not work for @file{ioctl.h},
1871: @file{ttychars.h} and other system header files unless the
1872: @samp{-traditional} option is used.} The bug is in the header files:
1873: at least on some machines, they rely on behavior that is incompatible
1874: with ANSI C. This behavior consists of substituting for macro
1875: argument names when they appear inside of character constants. The
1876: @samp{-traditional} option tells GNU CC to behave the way these
1877: headers expect.
1878:
1879: Because of this problem, you might prefer to configure GNU CC to use
1880: the system's own C preprocessor. To do so, make the file
1881: @file{/usr/local/lib/gcc-cpp} a link to @file{/lib/cpp}.
1882:
1883: Alternatively, on Sun systems and 4.3BSD at least, you can correct the
1884: include files by running the shell script @file{fixincludes}. This
1885: installs modified, corrected copies of the files @file{ioctl.h},
1886: @file{ttychars.h} and many others, in a special directory where only
1.1.1.2 root 1887: GNU CC will normally look for them. This script will work on various
1.1.1.6 root 1888: systems because it chooses the files by searching all the system
1.1.1.2 root 1889: headers for the problem cases that we know about.
1.1 root 1890: @end enumerate
1891:
1892: If you cannot install the compiler's passes and run-time support in
1893: @file{/usr/local/lib}, you can alternatively use the @samp{-B} option to
1894: specify a prefix by which they may be found. The compiler concatenates
1895: the prefix with the names @file{cpp}, @file{cc1} and @file{gnulib}.
1896: Thus, you can put the files in a directory @file{/usr/foo/gcc} and
1897: specify @samp{-B/usr/foo/gcc/} when you run GNU CC.
1898:
1899: Also, you can specify an alternative default directory for these files
1900: by setting the Make variable @code{libdir} when you make GNU CC.
1901:
1.1.1.8 root 1902: @node Other Dir, Sun Install, Installation, Installation
1903: @section Compilation in a Separate Directory
1.1 root 1904:
1.1.1.8 root 1905: If you wish to build the object files and executables in a directory
1906: other than the one containing the source files, here is what you must
1907: do differently:
1908:
1909: @enumerate
1910: @item
1911: Go to that directory before running @file{config.gcc}:
1912:
1913: @example
1914: mkdir gcc-sun3
1915: cd gcc-sun3
1916: @end example
1.1.1.4 root 1917:
1.1.1.8 root 1918: On systems that do not support symbolic links, this directory must be
1919: on the same file system as the source code directory.
1920:
1921: @item
1922: Specify where to find @file{config.gcc} when you run it:
1923:
1924: @example
1925: ../gcc-1.36/config.gcc @dots{}
1926: @end example
1927:
1928: @item
1929: Specify where to find the sources, as an argument to @file{config.gcc}:
1930:
1931: @example
1932: ../gcc-1.36/config.gcc -srcdir=../gcc-1.36 sun3
1933: @end example
1934:
1935: The @samp{-srcdir=@var{dir}} option is not needed when the source
1936: directory is the parent of the current directory, because
1937: @file{config.gcc} detects that case automatically.
1938: @end enumerate
1939:
1940: Now, you can run @code{make} in that directory. You need not repeat the
1941: configuration steps shown above, when ordinary source files change. You
1942: must, however, run @code{config.gcc} again when the configuration files
1943: change, if your system does not support symbolic links.
1944:
1945: @node Sun Install, 3b1 Install, Other Dir, Installation
1946: @section Installing GNU CC on the Sun
1947:
1948: Make sure the environment variable @code{FLOAT_OPTION} is not set when
1949: you compile @file{gnulib}. If this option were set to @code{f68881}
1950: when @file{gnulib} is compiled, the resulting code would demand to be
1951: linked with a special startup file and would not link properly without
1952: special pains.
1953:
1954: There is a bug in @code{alloca} in certain versions of the Sun library.
1955: To avoid this bug, install the binaries of GNU CC that were compiled by
1956: GNU CC. They use @code{alloca} as a built-in function and never the one
1957: in the library.
1958:
1959: Some versions of the Sun compiler crash when compiling GNU CC.
1960: The problem is a segmentation fault in cpp.
1961:
1962: This problem seems to be due to the bulk of data in the environment
1963: variables. You may be able to avoid it by using the following
1964: command to compile GNU CC with Sun CC:
1965:
1966: @example
1967: make CC="TERMCAP=x OBJS=x LIBFUNCS=x STAGESTUFF=x cc"
1968: @end example
1969:
1970: @node 3b1 Install, VMS Install, Sun Install, Installation
1971: @section Installing GNU CC on the 3b1
1972:
1973: Installing GNU CC on the 3b1 is difficult if you do not already have
1974: GNU CC running, due to bugs in the installed C compiler. However,
1975: the following procedure might work. We are unable to test it.
1.1 root 1976:
1977: @enumerate
1978: @item
1.1.1.8 root 1979: Comment out the @samp{#include "config.h"} line on line 37 of
1980: @file{cccp.c} and do @samp{make cpp}. This makes a preliminary version
1981: of GNU cpp.
1.1 root 1982:
1983: @item
1.1.1.8 root 1984: Save the old @file{/lib/cpp} and copy the preliminary GNU cpp to that
1985: file name.
1.1.1.5 root 1986:
1.1.1.8 root 1987: @item
1988: Undo your change in @file{cccp.c}, or reinstall the original version,
1989: and do @samp{make cpp} again.
1990:
1991: @item
1992: Copy this final version of GNU cpp into @file{/lib/cpp}.
1993:
1994: @item
1.1.1.9 ! root 1995: Replace every occurrence of @code{obstack_free} in @file{tree.c}
1.1.1.8 root 1996: with @code{_obstack_free}.
1997:
1998: @item
1999: Run @code{make} to get the first-stage GNU CC.
2000:
2001: @item
2002: Reinstall the original version of @file{/lib/cpp}.
2003:
2004: @item
2005: Now you can compile GNU CC with itself and install it in the normal
2006: fashion.
1.1 root 2007: @end enumerate
2008:
1.1.1.9 ! root 2009: If you have installed an earlier version of GCC, you can compile the
! 2010: newer version with that. However, you will run into trouble compiling
! 2011: @file{gnulib}, since that is normally compiled with CC. To solve the
! 2012: problem, uncomment this line in @file{Makefile}:
! 2013:
! 2014: @example
! 2015: CCLIBFLAGS = -B/usr/local/lib/gcc- -tp -Wp,-traditional
! 2016: @end example
! 2017:
! 2018: @node VMS Install, HPUX Install, 3B1 Install, Installation
1.1.1.8 root 2019: @section Installing GNU CC on VMS
2020:
2021: The VMS version of GNU CC is distributed in a backup saveset containing
2022: both source code and precompiled binaries.
2023:
2024: To install the @file{gcc} command so you can use the compiler easily, in
1.1 root 2025: the same manner as you use the VMS C compiler, you must install the VMS CLD
2026: file for GNU CC as follows:
2027:
2028: @enumerate
2029: @item
2030: Define the VMS logical names @samp{GNU_CC} and @samp{GNU_CC_INCLUDE}
2031: to point to the directories where the GNU CC executables
1.1.1.8 root 2032: (@file{gcc-cpp}, @file{gcc-cc1}, etc.) and the C include files are
1.1 root 2033: kept. This should be done with the commands:@refill
2034:
2035: @example
1.1.1.8 root 2036: $ assign /super /system disk:[gcc.] gnu_cc
2037: $ assign /super /system disk:[gcc.include.] gnu_cc_include
1.1 root 2038: @end example
2039:
2040: @noindent
2041: with the appropriate disk and directory names. These commands can be
2042: placed in your system startup file so they will be executed whenever
1.1.1.8 root 2043: the machine is rebooted. You may, if you choose, do this via the
2044: @file{GCC_INSTALL.COM} script in the @file{[GCC]} directory.
1.1 root 2045:
2046: @item
1.1.1.8 root 2047: Install the @file{GCC} command with the command line:
1.1 root 2048:
2049: @example
1.1.1.8 root 2050: $ set command /table=sys$library:dcltables gnu_cc:[000000]gcc
1.1 root 2051: @end example
2052:
1.1.1.7 root 2053: @item
2054: To install the help file, do the following:
2055:
2056: @example
2057: $ lib/help sys$library:helplib.hlb gcc.hlp
2058: @end example
2059:
1.1 root 2060: @noindent
2061: Now you can invoke the compiler with a command like @samp{gcc /verbose
2062: file.c}, which is equivalent to the command @samp{gcc -v -c file.c} in
2063: Unix.
2064: @end enumerate
2065:
1.1.1.8 root 2066: We try to put corresponding binaries and sources on the VMS distribution
2067: tape. But sometimes the binaries will be from an older version that the
2068: sources, because we don't always have time to update them. (Use the
2069: @samp{/verbose} option to determine the version number of the binaries and
2070: compare it with the source file @file{version.c} to tell whether this is
2071: so.) In this case, you should use the binaries you get to recompile the
2072: sources. If you must recompile, here is how:
2073:
2074: @enumerate
2075: @item
2076: Copy the file @file{tm-vms.h} to @file{tm.h}, @file{xm-vms.h} to
2077: @file{config.h}, @file{vax.md} to @file{md.} and @file{out-vax.c}
2078: to @file{aux-output.c}. The files to be copied are found in the
2079: subdirectory named @file{config}; they should be copied to the
2080: main directory of GNU CC.@refill
2081:
2082: @item
2083: Setup the logical names and command tables as defined above. In
2084: addition, define the vms logical name @samp{GNU_BISON} to point at the
2085: to the directories where the Bison executable is kept. This should be
2086: done with the command:@refill
2087:
2088: @example
2089: $ assign /super /system disk:[bison.] gnu_bison
2090: @end example
2091:
2092: You may, if you choose, use the @file{INSTALL_BISON.COM} script in the
2093: @file{[BISON]} directory.
2094:
2095: @item
2096: Install the @samp{BISON} command with the command line:@refill
2097:
2098: @example
2099: $ set command /table=sys$library:dcltables gnu_bison:[000000]bison
2100: @end example
2101:
2102: @item
2103: Type @samp{@@make} to do recompile everything.
2104:
2105: If you are compiling with a version of GNU CC older than 1.33, specify
2106: @samp{/DEFINE=("inline=")} as an option in all the compilations. This
2107: requires editing all the @code{gcc} commands in @file{make-cc1.com}.
2108: (The older versions had problems supporting @code{inline}.) Once you
2109: have a working 1.33 or newer GNU CC, you can change this file back.
2110: @end enumerate
2111:
1.1.1.5 root 2112: There is a known problem on VMS: @code{const} global variables don't
2113: work compatibly with the VMS C compiler; we don't know a way to get
2114: them to the linker properly.
2115:
1.1.1.7 root 2116: Note that GNU CC on VMS does not generate debugging information to
2117: describe the program's symbols. It is not straightforward to implement
2118: this, and we have no time to spend on it, but we might consent to
2119: install a very modular implementation if you write it. You will
2120: probably have to modify GAS as well as GNU CC.
2121:
1.1.1.9 ! root 2122: @node HPUX Install,, VMS Install, Installation
! 2123: @section Installing GNU CC on HPUX
! 2124:
! 2125: To install GNU CC on HPUX, you must start by editing the file
! 2126: @file{Makefile}. Search for the string @samp{HPUX} to find comments
! 2127: saying what to change. You need to change some variable definitions and
! 2128: (if you are using GAS) some lines in the rule for the target
! 2129: @samp{gnulib}.
! 2130:
! 2131: To compile with the HPUX C compiler, you must specify get the file
! 2132: @file{alloca.c} from GNU Emacs. Then, when you run @code{make}, use
! 2133: this argument:
! 2134:
! 2135: @example
! 2136: make ALLOCA=alloca.o
! 2137: @end example
! 2138:
! 2139: When recompiling GNU CC with itself, do not define @code{ALLOCA}.
! 2140: Instead, an @samp{-I} option needs to be added to @code{CFLAGS} as
! 2141: follows:
! 2142:
! 2143: @example
! 2144: make CC=stage1/gcc CFLAGS="-g -O -Bstage1/ -I../binutils/hp-include"
! 2145: @end example
! 2146:
1.1 root 2147: @node Trouble, Incompatibilities, Installation, Top
2148: @chapter Known Causes of Trouble with GNU CC.
2149:
2150: Here are some of the things that have caused trouble for people installing
2151: or using GNU CC.
2152:
2153: @itemize @bullet
2154: @item
2155: On certain systems, defining certain environment variables such as
1.1.1.8 root 2156: @code{CC} can interfere with the functioning of @code{make}.
1.1 root 2157:
2158: @item
2159: Cross compilation can run into trouble for certain machines because
2160: some target machines' assemblers require floating point numbers to be
2161: written as @emph{integer} constants in certain contexts.
2162:
2163: The compiler writes these integer constants by examining the floating
2164: point value as an integer and printing that integer, because this is
2165: simple to write and independent of the details of the floating point
2166: representation. But this does not work if the compiler is running on
2167: a different machine with an incompatible floating point format, or
2168: even a different byte-ordering.
2169:
1.1.1.5 root 2170: In addition, correct constant folding of floating point values
2171: requires representing them in the target machine's format.
2172: (The C standard does not quite require this, but in practice
2173: it is the only way to win.)
2174:
2175: It is now possible to overcome these problems by defining macros such
2176: as @code{REAL_VALUE_TYPE}. But doing so is a substantial amount of
2177: work for each target machine. @xref{Cross-compilation}.
1.1 root 2178:
2179: @item
2180: DBX rejects some files produced by GNU CC, though it accepts similar
2181: constructs in output from PCC. Until someone can supply a coherent
2182: description of what is valid DBX input and what is not, there is
2183: nothing I can do about these problems. You are on your own.
1.1.1.2 root 2184:
2185: @item
2186: Users often think it is a bug when GNU CC reports an error for code
2187: like this:
2188:
2189: @example
2190: int foo (short);
2191:
2192: int foo (x)
2193: short x;
2194: @{@dots{}@}
2195: @end example
2196:
1.1.1.4 root 2197: The error message is correct: this code really is erroneous, because the
2198: old-style non-prototype definition passes subword integers in their
2199: promoted types. In other words, the argument is really an @code{int},
2200: not a @code{short}. The correct prototype is this:
1.1.1.2 root 2201:
2202: @example
2203: int foo (int);
2204: @end example
2205:
2206: @item
2207: Users often think it is a bug when GNU CC reports an error for code
2208: like this:
2209:
2210: @example
2211: int foo (struct mumble *);
2212:
2213: struct mumble @{ @dots{} @};
2214:
2215: int foo (struct mumble *x)
2216: @{ @dots{} @}
2217: @end example
2218:
2219: This code really is erroneous, because the scope of @code{struct
2220: mumble} the prototype is limited to the argument list containing it.
2221: It does not refer to the @code{struct mumble} defined with file scope
2222: immediately below---they are two unrelated types with similar names in
2223: different scopes.
2224:
2225: But in the definition of @code{foo}, the file-scope type is used
2226: because that is available to be inherited. Thus, the definition and
2227: the prototype do not match, and you get an error.
2228:
2229: This behavior may seem silly, but it's what the ANSI standard
2230: specifies. It is easy enough for you to make your code work by moving
2231: the definition of @code{struct mumble} above the prototype. I don't
2232: think it's worth being incompatible for.
1.1 root 2233: @end itemize
2234:
2235: @node Incompatibilities, Extensions, Trouble, Top
2236: @chapter Incompatibilities of GNU CC
2237:
2238: There are several noteworthy incompatibilities between GNU C and most
1.1.1.9 ! root 2239: existing (non-ANSI) versions of C. The @samp{-traditional} option
! 2240: eliminates most of these incompatibilities, @emph{but not all}, by
! 2241: telling GNU C to behave like older C compilers.
1.1 root 2242:
2243: @itemize @bullet
2244: @item
2245: GNU CC normally makes string constants read-only. If several
2246: identical-looking string constants are used, GNU CC stores only one
2247: copy of the string.
2248:
2249: One consequence is that you cannot call @code{mktemp} with a string
2250: constant argument. The function @code{mktemp} always alters the
2251: string its argument points to.
2252:
2253: Another consequence is that @code{sscanf} does not work on some
2254: systems when passed a string constant as its format control string.
2255: This is because @code{sscanf} incorrectly tries to write into the
1.1.1.4 root 2256: string constant. Likewise @code{fscanf} and @code{scanf}.
1.1 root 2257:
2258: The best solution to these problems is to change the program to use
2259: @code{char}-array variables with initialization strings for these
2260: purposes instead of string constants. But if this is not possible,
2261: you can use the @samp{-fwritable-strings} flag, which directs GNU CC
2262: to handle string constants the same way most C compilers do.
1.1.1.8 root 2263: @samp{-traditional} also has this effect, among others.
1.1 root 2264:
2265: @item
2266: GNU CC does not substitute macro arguments when they appear inside of
2267: string constants. For example, the following macro in GNU CC
2268:
2269: @example
2270: #define foo(a) "a"
2271: @end example
2272:
2273: @noindent
1.1.1.8 root 2274: will produce output @code{"a"} regardless of what the argument @var{a} is.
1.1 root 2275:
2276: The @samp{-traditional} option directs GNU CC to handle such cases
2277: (among others) in the old-fashioned (non-ANSI) fashion.
2278:
2279: @item
2280: When you use @code{setjmp} and @code{longjmp}, the only automatic
2281: variables guaranteed to remain valid are those declared
2282: @code{volatile}. This is a consequence of automatic register
2283: allocation. Consider this function:
2284:
2285: @example
2286: jmp_buf j;
2287:
2288: foo ()
2289: @{
2290: int a, b;
2291:
2292: a = fun1 ();
2293: if (setjmp (j))
2294: return a;
2295:
2296: a = fun2 ();
2297: /* @r{@code{longjmp (j)} may be occur in @code{fun3}.} */
2298: return a + fun3 ();
2299: @}
2300: @end example
2301:
2302: Here @code{a} may or may not be restored to its first value when the
2303: @code{longjmp} occurs. If @code{a} is allocated in a register, then
2304: its first value is restored; otherwise, it keeps the last value stored
2305: in it.
2306:
2307: If you use the @samp{-W} option with the @samp{-O} option, you will
2308: get a warning when GNU CC thinks such a problem might be possible.
2309:
1.1.1.2 root 2310: The @samp{-traditional} option directs GNU C to put variables in
2311: the stack by default, rather than in registers, in functions that
2312: call @code{setjmp}. This results in the behavior found in
2313: traditional C compilers.
2314:
1.1 root 2315: @item
2316: Declarations of external variables and functions within a block apply
2317: only to the block containing the declaration. In other words, they
2318: have the same scope as any other declaration in the same place.
2319:
2320: In some other C compilers, a @code{extern} declaration affects all the
2321: rest of the file even if it happens within a block.
2322:
2323: The @samp{-traditional} option directs GNU C to treat all @code{extern}
2324: declarations as global, like traditional compilers.
2325:
2326: @item
2327: In traditional C, you can combine @code{long}, etc., with a typedef name,
2328: as shown here:
2329:
2330: @example
2331: typedef int foo;
2332: typedef long foo bar;
2333: @end example
2334:
2335: In ANSI C, this is not allowed: @code{long} and other type modifiers
2336: require an explicit @code{int}. Because this criterion is expressed
2337: by Bison grammar rules rather than C code, the @samp{-traditional}
2338: flag cannot alter it.
2339:
2340: @item
2341: PCC allows typedef names to be used as function parameters. The
2342: difficulty described immediately above applies here too.
2343:
2344: @item
2345: PCC allows whitespace in the middle of compound assignment operators
2346: such as @samp{+=}. GNU CC, following the ANSI standard, does not
2347: allow this. The difficulty described immediately above applies here
2348: too.
2349:
2350: @item
2351: GNU CC will flag unterminated character constants inside of preprocessor
2352: conditionals that fail. Some programs have English comments enclosed in
2353: conditionals that are guaranteed to fail; if these comments contain
2354: apostrophes, GNU CC will probably report an error. For example,
2355: this code would produce an error:
2356:
2357: @example
2358: #if 0
2359: You can't expect this to work.
2360: #endif
2361: @end example
2362:
2363: The best solution to such a problem is to put the text into an actual
2364: C comment delimited by @samp{/*@dots{}*/}. However,
2365: @samp{-traditional} suppresses these error messages.
2366:
2367: @item
2368: When compiling functions that return @code{float}, PCC converts it to
2369: a double. GNU CC actually returns a @code{float}. If you are concerned
2370: with PCC compatibility, you should declare your functions to return
2371: @code{double}; you might as well say what you mean.
2372:
2373: @item
2374: When compiling functions that return structures or unions, GNU CC
1.1.1.6 root 2375: output code normally uses a method different from that used on most
2376: versions of Unix. As a result, code compiled with GNU CC cannot call
2377: a structure-returning function compiled with PCC, and vice versa.
1.1 root 2378:
1.1.1.6 root 2379: The method used by GNU CC is as follows: a structure or union which is 1,
1.1 root 2380: 2, 4 or 8 bytes long is returned like a scalar. A structure or union
2381: with any other size is stored into an address supplied by the caller
2382: in a special, fixed register.
2383:
2384: PCC usually handles all sizes of structures and unions by returning
2385: the address of a block of static storage containing the value. This
1.1.1.6 root 2386: method is not used in GNU CC because it is slower and nonreentrant.
1.1.1.5 root 2387:
1.1.1.6 root 2388: You can tell GNU CC to use the PCC convention with the option
2389: @samp{-fpcc-struct-return}.
1.1.1.8 root 2390:
2391: @item
2392: On the Sparc, GNU CC uses an incompatible calling convention for
2393: structures. It passes them by including their contents in the argument
2394: list, whereas the standard compiler passes them effectively by
2395: reference.
2396:
2397: This really ought to be fixed, but such calling conventions are not
2398: yet supported in GNU CC, so it isn't straightforward to fix it.
2399:
2400: The convention for structure returning is also incompatible, and
2401: @samp{-fpcc-struct-return} does not help.
1.1 root 2402: @end itemize
2403:
2404: @node Extensions, Bugs, Incompatibilities, Top
2405: @chapter GNU Extensions to the C Language
2406:
2407: GNU C provides several language features not found in ANSI standard C.
2408: (The @samp{-pedantic} option directs GNU CC to print a warning message if
2409: any of these features is used.) To test for the availability of these
2410: features in conditional compilation, check for a predefined macro
2411: @code{__GNUC__}, which is always defined under GNU CC.
2412:
2413: @menu
2414: * Statement Exprs:: Putting statements and declarations inside expressions.
2415: * Naming Types:: Giving a name to the type of some expression.
1.1.1.9 ! root 2416: * Typeof:: @code{typeof}: referring to the type of an expression.
! 2417: * Lvalues:: Using @samp{?:}, @samp{,} and casts in lvalues.
! 2418: * Conditionals:: Omitting the middle operand of a @samp{?:} expression.
! 2419: * Zero-Length:: Zero-length arrays.
! 2420: * Variable-Length:: Arrays whose length is computed at run time.
! 2421: * Subscripting:: Any array can be subscripted, even if not an lvalue.
! 2422: * Pointer Arith:: Arithmetic on @code{void}-pointers and function pointers.
! 2423: * Initializers:: Non-constant initializers.
! 2424: * Constructors:: Constructor expressions give structures, unions
! 2425: or arrays as values.
1.1.1.5 root 2426: * Function Attributes:: Declaring that functions have no side effects,
1.1.1.9 ! root 2427: or that they can never return.
1.1 root 2428: * Dollar Signs:: Dollar sign is allowed in identifiers.
2429: * Alignment:: Inquiring about the alignment of a type or variable.
2430: * Inline:: Defining inline functions (as fast as macros).
1.1.1.9 ! root 2431: * Extended Asm:: Assembler instructions with C expressions as operands.
! 2432: (With them you can define ``built-in'' functions.)
! 2433: * Asm Labels:: Specifying the assembler name to use for a C symbol.
1.1.1.8 root 2434: * Explicit Reg Vars:: Defining variables residing in specified registers.
2435: * Alternate Keywords:: @code{__const__}, @code{__asm__}, etc., for header files.
1.1 root 2436: @end menu
2437:
2438: @node Statement Exprs, Naming Types, Extensions, Extensions
2439: @section Statements and Declarations inside of Expressions
2440:
2441: A compound statement in parentheses may appear inside an expression in GNU
2442: C. This allows you to declare variables within an expression. For
2443: example:
2444:
2445: @example
2446: (@{ int y = foo (); int z;
2447: if (y > 0) z = y;
2448: else z = - y;
2449: z; @})
2450: @end example
2451:
2452: @noindent
2453: is a valid (though slightly more complex than necessary) expression
2454: for the absolute value of @code{foo ()}.
2455:
2456: This feature is especially useful in making macro definitions ``safe'' (so
2457: that they evaluate each operand exactly once). For example, the
2458: ``maximum'' function is commonly defined as a macro in standard C as
2459: follows:
2460:
2461: @example
2462: #define max(a,b) ((a) > (b) ? (a) : (b))
2463: @end example
2464:
2465: @noindent
2466: But this definition computes either @var{a} or @var{b} twice, with bad
2467: results if the operand has side effects. In GNU C, if you know the
2468: type of the operands (here let's assume @code{int}), you can define
2469: the macro safely as follows:
2470:
2471: @example
2472: #define maxint(a,b) \
2473: (@{int _a = (a), _b = (b); _a > _b ? _a : _b; @})
2474: @end example
2475:
2476: Embedded statements are not allowed in constant expressions, such as
2477: the value of an enumeration constant, the width of a bit field, or
2478: the initial value of a static variable.
2479:
2480: If you don't know the type of the operand, you can still do this, but you
2481: must use @code{typeof} (@pxref{Typeof}) or type naming (@pxref{Naming
2482: Types}).
2483:
2484: @node Naming Types, Typeof, Statement Exprs, Extensions
2485: @section Naming an Expression's Type
2486:
2487: You can give a name to the type of an expression using a @code{typedef}
2488: declaration with an initializer. Here is how to define @var{name} as a
2489: type name for the type of @var{exp}:
2490:
2491: @example
2492: typedef @var{name} = @var{exp};
2493: @end example
2494:
2495: This is useful in conjunction with the statements-within-expressions
2496: feature. Here is how the two together can be used to define a safe
2497: ``maximum'' macro that operates on any arithmetic type:
2498:
2499: @example
2500: #define max(a,b) \
2501: (@{typedef _ta = (a), _tb = (b); \
2502: _ta _a = (a); _tb _b = (b); \
2503: _a > _b ? _a : _b; @})
2504: @end example
2505:
2506: The reason for using names that start with underscores for the local
2507: variables is to avoid conflicts with variable names that occur within the
2508: expressions that are substituted for @code{a} and @code{b}. Eventually we
2509: hope to design a new form of declaration syntax that allows you to declare
2510: variables whose scopes start only after their initializers; this will be a
2511: more reliable way to prevent such conflicts.
2512:
2513: @node Typeof, Lvalues, Naming Types, Extensions
2514: @section Referring to a Type with @code{typeof}
2515:
2516: Another way to refer to the type of an expression is with @code{typeof}.
2517: The syntax of using of this keyword looks like @code{sizeof}, but the
2518: construct acts semantically like a type name defined with @code{typedef}.
2519:
2520: There are two ways of writing the argument to @code{typeof}: with an
2521: expression or with a type. Here is an example with an expression:
2522:
2523: @example
2524: typeof (x[0](1))
2525: @end example
2526:
2527: @noindent
2528: This assumes that @code{x} is an array of functions; the type described
2529: is that of the values of the functions.
2530:
2531: Here is an example with a typename as the argument:
2532:
2533: @example
2534: typeof (int *)
2535: @end example
2536:
2537: @noindent
2538: Here the type described is that of pointers to @code{int}.
2539:
1.1.1.7 root 2540: If you are writing a header file that must work when included in ANSI C
1.1.1.8 root 2541: programs, write @code{__typeof__} instead of @code{typeof}.
1.1.1.7 root 2542: @xref{Alternate Keywords}.
2543:
1.1 root 2544: A @code{typeof}-construct can be used anywhere a typedef name could be
2545: used. For example, you can use it in a declaration, in a cast, or inside
2546: of @code{sizeof} or @code{typeof}.
2547:
2548: @itemize @bullet
2549: @item
2550: This declares @code{y} with the type of what @code{x} points to.
2551:
2552: @example
2553: typeof (*x) y;
2554: @end example
2555:
2556: @item
2557: This declares @code{y} as an array of such values.
2558:
2559: @example
2560: typeof (*x) y[4];
2561: @end example
2562:
2563: @item
2564: This declares @code{y} as an array of pointers to characters:
2565:
2566: @example
2567: typeof (typeof (char *)[4]) y;
2568: @end example
2569:
2570: @noindent
2571: It is equivalent to the following traditional C declaration:
2572:
2573: @example
2574: char *y[4];
2575: @end example
2576:
2577: To see the meaning of the declaration using @code{typeof}, and why it
2578: might be a useful way to write, let's rewrite it with these macros:
2579:
2580: @example
2581: #define pointer(T) typeof(T *)
2582: #define array(T, N) typeof(T [N])
2583: @end example
2584:
2585: @noindent
2586: Now the declaration can be rewritten this way:
2587:
2588: @example
2589: array (pointer (char), 4) y;
2590: @end example
2591:
2592: @noindent
1.1.1.8 root 2593: Thus, @code{array (pointer (char), 4)} is the type of arrays of 4
1.1 root 2594: pointers to @code{char}.
2595: @end itemize
2596:
2597: @node Lvalues, Conditionals, Typeof, Extensions
2598: @section Generalized Lvalues
2599:
2600: Compound expressions, conditional expressions and casts are allowed as
2601: lvalues provided their operands are lvalues. This means that you can take
2602: their addresses or store values into them.
2603:
2604: For example, a compound expression can be assigned, provided the last
2605: expression in the sequence is an lvalue. These two expressions are
2606: equivalent:
2607:
2608: @example
2609: (a, b) += 5
2610: a, (b += 5)
2611: @end example
2612:
2613: Similarly, the address of the compound expression can be taken. These two
2614: expressions are equivalent:
2615:
2616: @example
2617: &(a, b)
2618: a, &b
2619: @end example
2620:
2621: A conditional expression is a valid lvalue if its type is not void and the
2622: true and false branches are both valid lvalues. For example, these two
2623: expressions are equivalent:
2624:
2625: @example
2626: (a ? b : c) = 5
2627: (a ? b = 5 : (c = 5))
2628: @end example
2629:
2630: A cast is a valid lvalue if its operand is valid. Taking the address of
2631: the cast is the same as taking the address without a cast, except for the
2632: type of the result. For example, these two expressions are equivalent (but
1.1.1.8 root 2633: the second may be valid when the type of @code{a} does not permit a cast to
2634: @code{int *}).
1.1 root 2635:
2636: @example
2637: &(int *)a
2638: (int **)&a
2639: @end example
2640:
2641: A simple assignment whose left-hand side is a cast works by converting the
2642: right-hand side first to the specified type, then to the type of the inner
2643: left-hand side expression. After this is stored, the value is converter
2644: back to the specified type to become the value of the assignment. Thus, if
1.1.1.8 root 2645: @code{a} has type @code{char *}, the following two expressions are
1.1 root 2646: equivalent:
2647:
2648: @example
2649: (int)a = 5
2650: (int)(a = (char *)5)
2651: @end example
2652:
2653: An assignment-with-arithmetic operation such as @samp{+=} applied to a cast
2654: performs the arithmetic using the type resulting from the cast, and then
2655: continues as in the previous case. Therefore, these two expressions are
2656: equivalent:
2657:
2658: @example
2659: (int)a += 5
2660: (int)(a = (char *) ((int)a + 5))
2661: @end example
2662:
2663: @node Conditionals, Zero-Length, Lvalues, Extensions
2664: @section Conditional Expressions with Omitted Middle-Operands
2665:
2666: The middle operand in a conditional expression may be omitted. Then
2667: if the first operand is nonzero, its value is the value of the conditional
2668: expression.
2669:
2670: Therefore, the expression
2671:
2672: @example
2673: x ? : y
2674: @end example
2675:
2676: @noindent
2677: has the value of @code{x} if that is nonzero; otherwise, the value of
2678: @code{y}.
2679:
2680: This example is perfectly equivalent to
2681:
2682: @example
2683: x ? x : y
2684: @end example
2685:
2686: @noindent
2687: In this simple case, the ability to omit the middle operand is not
2688: especially useful. When it becomes useful is when the first operand does,
2689: or may (if it is a macro argument), contain a side effect. Then repeating
2690: the operand in the middle would perform the side effect twice. Omitting
2691: the middle operand uses the value already computed without the undesirable
2692: effects of recomputing it.
2693:
2694: @node Zero-Length, Variable-Length, Conditionals, Extensions
2695: @section Arrays of Length Zero
2696:
2697: Zero-length arrays are allowed in GNU C. They are very useful as the last
2698: element of a structure which is really a header for a variable-length
2699: object:
2700:
2701: @example
2702: struct line @{
2703: int length;
2704: char contents[0];
2705: @};
2706:
2707: @{
2708: struct line *thisline
2709: = (struct line *) malloc (sizeof (struct line) + this_length);
2710: thisline->length = this_length;
2711: @}
2712: @end example
2713:
2714: In standard C, you would have to give @code{contents} a length of 1, which
2715: means either you waste space or complicate the argument to @code{malloc}.
2716:
2717: @node Variable-Length, Subscripting, Zero-Length, Extensions
2718: @section Arrays of Variable Length
2719:
2720: Variable-length automatic arrays are allowed in GNU C. These arrays are
2721: declared like any other automatic arrays, but with a length that is not a
2722: constant expression. The storage is allocated at that time and
2723: deallocated when the brace-level is exited. For example:
2724:
2725: @example
2726: FILE *concat_fopen (char *s1, char *s2, char *mode)
2727: @{
2728: char str[strlen (s1) + strlen (s2) + 1];
2729: strcpy (str, s1);
2730: strcat (str, s2);
2731: return fopen (str, mode);
2732: @}
2733: @end example
2734:
1.1.1.7 root 2735: You can also use variable-length arrays as arguments to functions:
1.1 root 2736:
2737: @example
2738: struct entry
1.1.1.7 root 2739: tester (int len, char data[len])
1.1 root 2740: @{
1.1.1.7 root 2741: @dots{}
1.1 root 2742: @}
2743: @end example
2744:
2745: The length of an array is computed on entry to the brace-level where the
2746: array is declared and is remembered for the scope of the array in case you
2747: access it with @code{sizeof}.
2748:
2749: Jumping or breaking out of the scope of the array name will also deallocate
2750: the storage. Jumping into the scope is not allowed; you will get an error
2751: message for it.
2752:
2753: You can use the function @code{alloca} to get an effect much like
2754: variable-length arrays. The function @code{alloca} is available in
2755: many other C implementations (but not in all). On the other hand,
2756: variable-length arrays are more elegant.
2757:
2758: There are other differences between these two methods. Space allocated
2759: with @code{alloca} exists until the containing @emph{function} returns.
2760: The space for a variable-length array is deallocated as soon as the array
2761: name's scope ends. (If you use both variable-length arrays and
2762: @code{alloca} in the same function, deallocation of a variable-length array
2763: will also deallocate anything more recently allocated with @code{alloca}.)
2764:
2765: @node Subscripting, Pointer Arith, Variable-Length, Extensions
2766: @section Non-Lvalue Arrays May Have Subscripts
2767:
2768: Subscripting is allowed on arrays that are not lvalues, even though the
2769: unary @samp{&} operator is not. For example, this is valid in GNU C though
2770: not valid in other C dialects:
2771:
2772: @example
2773: struct foo @{int a[4];@};
2774:
2775: struct foo f();
2776:
2777: bar (int index)
2778: @{
2779: return f().a[index];
2780: @}
2781: @end example
2782:
2783: @node Pointer Arith, Initializers, Subscripting, Extensions
2784: @section Arithmetic on @code{void}-Pointers and Function Pointers
2785:
2786: In GNU C, addition and subtraction operations are supported on pointers to
2787: @code{void} and on pointers to functions. This is done by treating the
2788: size of a @code{void} or of a function as 1.
2789:
2790: A consequence of this is that @code{sizeof} is also allowed on @code{void}
2791: and on function types, and returns 1.
2792:
1.1.1.8 root 2793: The option @samp{-Wpointer-arith} requests a warning if these extensions
2794: are used.
2795:
1.1 root 2796: @node Initializers, Constructors, Pointer Arith, Extensions
2797: @section Non-Constant Initializers
2798:
1.1.1.8 root 2799: The elements of an aggregate initializer for an automatic variable are
2800: not required to be constant expressions in GNU C. Here is an example of
2801: an initializer with run-time varying elements:
1.1 root 2802:
2803: @example
2804: foo (float f, float g)
2805: @{
2806: float beat_freqs[2] = @{ f-g, f+g @};
2807: @dots{}
2808: @}
2809: @end example
2810:
1.1.1.5 root 2811: @node Constructors, Function Attributes, Initializers, Extensions
1.1 root 2812: @section Constructor Expressions
2813:
2814: GNU C supports constructor expressions. A constructor looks like a cast
2815: containing an initializer. Its value is an object of the type specified in
2816: the cast, containing the elements specified in the initializer. The type
2817: must be a structure, union or array type.
2818:
2819: Assume that @code{struct foo} and @code{structure} are declared as shown:
2820:
2821: @example
2822: struct foo @{int a; char b[2];@} structure;
2823: @end example
2824:
2825: @noindent
1.1.1.8 root 2826: Here is an example of constructing a @code{struct foo} with a constructor:
1.1 root 2827:
2828: @example
2829: structure = ((struct foo) @{x + y, 'a', 0@});
2830: @end example
2831:
2832: @noindent
2833: This is equivalent to writing the following:
2834:
2835: @example
2836: @{
2837: struct foo temp = @{x + y, 'a', 0@};
2838: structure = temp;
2839: @}
2840: @end example
2841:
2842: You can also construct an array. If all the elements of the constructor
2843: are (made up of) simple constant expressions, suitable for use in
2844: initializers, then the constructor is an lvalue and can be coerced to a
2845: pointer to its first element, as shown here:
2846:
2847: @example
2848: char **foo = (char *[]) @{ "x", "y", "z" @};
2849: @end example
2850:
2851: Array constructors whose elements are not simple constants are not very
2852: useful, because the constructor is not an lvalue. There are only two valid
2853: ways to use it: to subscript it, or initialize an array variable with it.
2854: The former is probably slower than a @code{switch} statement, while the
2855: latter does the same thing an ordinary C initializer would do.
2856:
2857: @example
2858: output = ((int[]) @{ 2, x, 28 @}) [input];
2859: @end example
2860:
1.1.1.8 root 2861: @node Function Attributes, Dollar Signs, Constructors, Extensions
1.1.1.5 root 2862: @section Declaring Attributes of Functions
2863:
2864: In GNU C, you declare certain things about functions called in your program
2865: which help the compiler optimize function calls.
2866:
2867: A few functions, such as @code{abort} and @code{exit}, cannot return.
2868: These functions should be declared @code{volatile}. For example,
2869:
2870: @example
2871: extern volatile void abort ();
2872: @end example
2873:
2874: @noindent
2875: tells the compiler that it can assume that @code{abort} will not return.
2876: This makes slightly better code, but more importantly it helps avoid
2877: spurious warnings of uninitialized variables.
2878:
2879: Many functions do not examine any values except their arguments, and
2880: have no effects except the return value. Such a function can be subject
2881: to common subexpression elimination and loop optimization just as an
2882: arithmetic operator would be. These functions should be declared
2883: @code{const}. For example,
2884:
2885: @example
2886: extern const void square ();
2887: @end example
2888:
2889: @noindent
2890: says that the hypothetical function @code{square} is safe to call
2891: fewer times than the program says.
2892:
2893: Note that a function that has pointer arguments and examines the data
2894: pointed to must @emph{not} be declared @code{const}. Likewise, a
2895: function that calls a non-@code{const} function must not be
2896: @code{const}.
2897:
2898: Some people object to this feature, claiming that ANSI C's @code{#pragma}
2899: should be used instead. There are two reasons I did not do this.
2900:
2901: @enumerate
2902: @item
2903: It is impossible to generate @code{#pragma} commands from a macro.
2904:
2905: @item
2906: The @code{#pragma} command is just as likely as these keywords to mean
2907: something else in another compiler.
2908: @end enumerate
2909:
2910: These two reasons apply to @emph{any} application whatever: as far as
2911: I can see, @code{#pragma} is never useful.
2912:
2913: @node Dollar Signs, Alignment, Function Attributes, Extensions
1.1 root 2914: @section Dollar Signs in Identifier Names
2915:
2916: In GNU C, you may use dollar signs in identifier names. This is because
2917: many traditional C implementations allow such identifiers.
2918:
1.1.1.9 ! root 2919: Dollar signs are allowed if you specify @samp{-traditional}; they are
! 2920: not allowed if you specify @samp{-ansi}. Whether they are allowed by
! 2921: default depends on the target machine; usually, they are not.
! 2922:
1.1 root 2923: @node Alignment, Inline, Dollar Signs, Extensions
2924: @section Inquiring about the Alignment of a Type or Variable
2925:
1.1.1.8 root 2926: The keyword @code{__alignof__} allows you to inquire about how an object
1.1 root 2927: is aligned, or the minimum alignment usually required by a type. Its
2928: syntax is just like @code{sizeof}.
2929:
2930: For example, if the target machine requires a @code{double} value to be
1.1.1.8 root 2931: aligned on an 8-byte boundary, then @code{__alignof__ (double)} is 8.
2932: This is true on many RISC machines. On more traditional machine
2933: designs, @code{__alignof__ (double)} is 4 or even 2.
1.1 root 2934:
2935: Some machines never actually require alignment; they allow reference to any
1.1.1.8 root 2936: data type even at an odd addresses. For these machines, @code{__alignof__}
1.1 root 2937: reports the @emph{recommended} alignment of a type.
2938:
1.1.1.8 root 2939: When the operand of @code{__alignof__} is an lvalue rather than a type, the
1.1 root 2940: value is the largest alignment that the lvalue is known to have. It may
2941: have this alignment as a result of its data type, or because it is part of
2942: a structure and inherits alignment from that structure. For example, after
2943: this declaration:
2944:
2945: @example
2946: struct foo @{ int x; char y; @} foo1;
2947: @end example
2948:
2949: @noindent
1.1.1.8 root 2950: the value of @code{__alignof__ (foo1.y)} is probably 2 or 4, the same as
2951: @code{__alignof__ (int)}, even though the data type of @code{foo1.y}
2952: does not itself demand any alignment.@refill
1.1 root 2953:
2954: @node Inline, Extended Asm, Alignment, Extensions
2955: @section An Inline Function is As Fast As a Macro
2956:
2957: By declaring a function @code{inline}, you can direct GNU CC to integrate
2958: that function's code into the code for its callers. This makes execution
2959: faster by eliminating the function-call overhead; in addition, if any of
2960: the actual argument values are constant, their known values may permit
2961: simplifications at compile time so that not all of the inline function's
2962: code needs to be included.
2963:
2964: To declare a function inline, use the @code{inline} keyword in its
2965: declaration, like this:
2966:
2967: @example
2968: inline int
2969: inc (int *a)
2970: @{
2971: (*a)++;
2972: @}
2973: @end example
2974:
1.1.1.7 root 2975: (If you are writing a header file to be included in ANSI C programs, write
1.1.1.8 root 2976: @code{__inline__} instead of @code{inline}. @xref{Alternate Keywords}.)
1.1.1.7 root 2977:
2978: You can also make all ``simple enough'' functions inline with the option
2979: @samp{-finline-functions}. Note that certain usages in a function
2980: definition can make it unsuitable for inline substitution.
1.1 root 2981:
2982: When a function is both inline and @code{static}, if all calls to the
1.1.1.8 root 2983: function are integrated into the caller, and the function's address is
2984: never used, then the function's own assembler code is never referenced.
2985: In this case, GNU CC does not actually output assembler code for the
2986: function, unless you specify the option @samp{-fkeep-inline-functions}.
2987: Some calls cannot be integrated for various reasons (in particular,
2988: calls that precede the function's definition cannot be integrated, and
2989: neither can recursive calls within the definition). If there is a
2990: nonintegrated call, then the function is compiled to assembler code as
2991: usual. The function must also be compiled as usual if the program
2992: refers to its address, because that can't be inlined.
1.1 root 2993:
2994: When an inline function is not @code{static}, then the compiler must assume
2995: that there may be calls from other source files; since a global symbol can
2996: be defined only once in any program, the function must not be defined in
2997: the other source files, so the calls therein cannot be integrated.
2998: Therefore, a non-@code{static} inline function is always compiled on its
2999: own in the usual fashion.
3000:
1.1.1.8 root 3001: If you specify both @code{inline} and @code{extern} in the function
3002: definition, then the definition is used only for inlining. In no case
3003: is the function compiled on its own, not even if you refer to its
3004: address explicitly. Such an address becomes an external reference, as
3005: if you had only declared the function, and had not defined it.
3006:
3007: This combination of @code{inline} and @code{extern} has almost the
3008: effect of a macro. The way to use it is to put a function definition in
3009: a header file with these keywords, and put another copy of the
3010: definition (lacking @code{inline} and @code{extern}) in a library file.
3011: The definition in the header file will cause most calls to the function
3012: to be inlined. If any uses of the function remain, they will refer to
3013: the single copy in the library.
3014:
1.1 root 3015: @node Extended Asm, Asm Labels, Inline, Extensions
3016: @section Assembler Instructions with C Expression Operands
3017:
3018: In an assembler instruction using @code{asm}, you can now specify the
3019: operands of the instruction using C expressions. This means no more
3020: guessing which registers or memory locations will contain the data you want
3021: to use.
3022:
3023: You must specify an assembler instruction template much like what appears
3024: in a machine description, plus an operand constraint string for each
3025: operand.
3026:
3027: For example, here is how to use the 68881's @code{fsinx} instruction:
3028:
3029: @example
3030: asm ("fsinx %1,%0" : "=f" (result) : "f" (angle));
3031: @end example
3032:
3033: @noindent
3034: Here @code{angle} is the C expression for the input operand while
3035: @code{result} is that of the output operand. Each has @samp{"f"} as its
3036: operand constraint, saying that a floating-point register is required. The
1.1.1.5 root 3037: @samp{=} in @samp{=f} indicates that the operand is an output; all output
1.1.1.4 root 3038: operands' constraints must use @samp{=}. The constraints use the same
3039: language used in the machine description (@pxref{Constraints}).
1.1 root 3040:
3041: Each operand is described by an operand-constraint string followed by the C
3042: expression in parentheses. A colon separates the assembler template from
3043: the first output operand, and another separates the last output operand
3044: from the first input, if any. Commas separate output operands and separate
1.1.1.4 root 3045: inputs. The total number of operands is limited to the maximum number of
1.1 root 3046: operands in any instruction pattern in the machine description.
3047:
1.1.1.4 root 3048: If there are no output operands, and there are input operands, then there
3049: must be two consecutive colons surrounding the place where the output
3050: operands would go.
3051:
1.1 root 3052: Output operand expressions must be lvalues; the compiler can check this.
3053: The input operands need not be lvalues. The compiler cannot check whether
3054: the operands have data types that are reasonable for the instruction being
3055: executed. It does not parse the assembler instruction template and does
3056: not know what it means, or whether it is valid assembler input. The
3057: extended @code{asm} feature is most often used for machine instructions
3058: that the compiler itself does not know exist.
3059:
3060: The output operands must be write-only; GNU CC will assume that the values
3061: in these operands before the instruction are dead and need not be
1.1.1.8 root 3062: generated. Extended asm does not support input-output or read-write
3063: operands. For this reason, the constraint character @samp{+}, which
3064: indicates such an operand, may not be used.
3065:
3066: When the assembler instruction has a read-write operand, or an operand
3067: in which only some of the bits are to be changed, you must logically
1.1 root 3068: split its function into two separate operands, one input operand and one
3069: write-only output operand. The connection between them is expressed by
3070: constraints which say they need to be in the same location when the
1.1.1.8 root 3071: instruction executes. You can use the same C expression for both
3072: operands, or different expressions. For example, here we write the
3073: (fictitious) @samp{combine} instruction with @code{bar} as its read-only
3074: source operand and @code{foo} as its read-write destination:
1.1 root 3075:
3076: @example
3077: asm ("combine %2,%0" : "=r" (foo) : "0" (foo), "g" (bar));
3078: @end example
3079:
3080: @noindent
3081: The constraint @samp{"0"} for operand 1 says that it must occupy the same
1.1.1.5 root 3082: location as operand 0. A digit in constraint is allowed only in an input
3083: operand, and it must refer to an output operand.
1.1 root 3084:
3085: Only a digit in the constraint can guarantee that one operand will be in
3086: the same place as another. The mere fact that @code{foo} is the value of
3087: both operands is not enough to guarantee that they will be in the same
3088: place in the generated assembler code. The following would not work:
3089:
3090: @example
3091: asm ("combine %2,%0" : "=r" (foo) : "r" (foo), "g" (bar));
3092: @end example
3093:
3094: Various optimizations or reloading could cause operands 0 and 1 to be in
3095: different registers; GNU CC knows no reason not to do so. For example, the
3096: compiler might find a copy of the value of @code{foo} in one register and
3097: use it for operand 1, but generate the output operand 0 in a different
3098: register (copying it afterward to @code{foo}'s own address). Of course,
3099: since the register for operand 1 is not even mentioned in the assembler
3100: code, the result will not work, but GNU CC can't tell that.
3101:
3102: Unless an output operand has the @samp{&} constraint modifier, GNU CC may
3103: allocate it in the same register as an unrelated input operand, on the
3104: assumption that the inputs are consumed before the outputs are produced.
3105: This assumption may be false if the assembler code actually consists of
3106: more than one instruction. In such a case, use @samp{&} for each output
3107: operand that may not overlap an input. @xref{Modifiers}.
3108:
1.1.1.4 root 3109: Some instructions clobber specific hard registers. To describe this, write
3110: a third colon after the input operands, followed by the names of the
3111: clobbered hard registers (given as strings). Here is a realistic example
3112: for the vax:
1.1 root 3113:
3114: @example
3115: asm volatile ("movc3 %0,%1,%2"
3116: : /* no outputs */
3117: : "g" (from), "g" (to), "g" (count)
3118: : "r0", "r1", "r2", "r3", "r4", "r5");
3119: @end example
3120:
1.1.1.4 root 3121: You can put multiple assembler instructions together in a single @code{asm}
1.1.1.7 root 3122: template, separated either with newlines (written as @samp{\n}) or with
3123: semicolons if the assembler allows such semicolons. The GNU assembler
3124: allows semicolons and all Unix assemblers seem to do so. The input
3125: operands are guaranteed not to use any of the clobbered registers, and
3126: neither will the output operands' addresses, so you can read and write the
3127: clobbered registers as many times as you like. Here is an example of
3128: multiple instructions in a template; it assumes that the subroutine
3129: @code{_foo} accepts arguments in registers 9 and 10:
1.1.1.4 root 3130:
3131: @example
3132: asm ("movl %0,r9;movl %1,r10;call _foo"
3133: : /* no outputs */
3134: : "g" (from), "g" (to)
3135: : "r9", "r10");
3136: @end example
3137:
1.1.1.7 root 3138: If you want to test the condition code produced by an assembler instruction,
3139: you must include a branch and a label in the @code{asm} construct, as follows:
3140:
3141: @example
3142: asm ("clr %0;frob %1;beq 0f;mov #1,%0;0:"
3143: : "g" (result)
3144: : "g" (input));
3145: @end example
3146:
3147: @noindent
3148: This assumes your assembler supports local labels, as the GNU assembler
3149: and most Unix assemblers do.
3150:
1.1 root 3151: Usually the most convenient way to use these @code{asm} instructions is to
3152: encapsulate them in macros that look like functions. For example,
3153:
3154: @example
3155: #define sin(x) \
3156: (@{ double __value, __arg = (x); \
3157: asm ("fsinx %1,%0": "=f" (__value): "f" (__arg)); \
3158: __value; @})
3159: @end example
3160:
3161: @noindent
3162: Here the variable @code{__arg} is used to make sure that the instruction
3163: operates on a proper @code{double} value, and to accept only those
3164: arguments @code{x} which can convert automatically to a @code{double}.
3165:
3166: Another way to make sure the instruction operates on the correct data type
3167: is to use a cast in the @code{asm}. This is different from using a
3168: variable @code{__arg} in that it converts more different types. For
3169: example, if the desired type were @code{int}, casting the argument to
3170: @code{int} would accept a pointer with no complaint, while assigning the
3171: argument to an @code{int} variable named @code{__arg} would warn about
3172: using a pointer unless the caller explicitly casts it.
3173:
1.1.1.4 root 3174: If an @code{asm} has output operands, GNU CC assumes for optimization
3175: purposes that the instruction has no side effects except to change the
3176: output operands. This does not mean that instructions with a side effect
3177: cannot be used, but you must be careful, because the compiler may eliminate
3178: them if the output operands aren't used, or move them out of loops, or
3179: replace two with one if they constitute a common subexpression. Also, if
3180: your instruction does have a side effect on a variable that otherwise
3181: appears not to change, the old value of the variable may be reused later if
3182: it happens to be found in a register.
1.1 root 3183:
3184: You can prevent an @code{asm} instruction from being deleted, moved or
3185: combined by writing the keyword @code{volatile} after the @code{asm}. For
3186: example:
3187:
3188: @example
3189: #define set_priority(x) \
3190: asm volatile ("set_priority %0": /* no outputs */ : "g" (x))
3191: @end example
3192:
1.1.1.7 root 3193: @noindent
3194: (However, an instruction without output operands will not be deleted
3195: or moved, regardless, unless it is unreachable.)
1.1.1.4 root 3196:
1.1 root 3197: It is a natural idea to look for a way to give access to the condition
3198: code left by the assembler instruction. However, when we attempted to
3199: implement this, we found no way to make it work reliably. The problem
3200: is that output operands might need reloading, which would result in
3201: additional following ``store'' instructions. On most machines, these
3202: instructions would alter the condition code before there was time to
3203: test it. This problem doesn't arise for ordinary ``test'' and
3204: ``compare'' instructions because they don't have any output operands.
3205:
1.1.1.7 root 3206: If you are writing a header file that should be includable in ANSI C
1.1.1.8 root 3207: programs, write @code{__asm__} instead of @code{asm}. @xref{Alternate
1.1.1.7 root 3208: Keywords}.
3209:
1.1.1.8 root 3210: @node Asm Labels, Explicit Reg Vars, Extended Asm, Extensions
1.1 root 3211: @section Controlling Names Used in Assembler Code
3212:
1.1.1.8 root 3213: You can specify the name to be used in the assembler code for a C
3214: function or variable by writing the @code{asm} (or @code{__asm__})
3215: keyword after the declarator as follows:
1.1 root 3216:
3217: @example
3218: int foo asm ("myfoo") = 2;
3219: @end example
3220:
3221: @noindent
3222: This specifies that the name to be used for the variable @code{foo} in
3223: the assembler code should be @samp{myfoo} rather than the usual
3224: @samp{_foo}.
3225:
3226: On systems where an underscore is normally prepended to the name of a C
3227: function or variable, this feature allows you to define names for the
3228: linker that do not start with an underscore.
3229:
3230: You cannot use @code{asm} in this way in a function @emph{definition}; but
3231: you can get the same effect by writing a declaration for the function
3232: before its definition and putting @code{asm} there, like this:
3233:
3234: @example
3235: extern func () asm ("FUNC");
3236:
3237: func (x, y)
3238: int x, y;
3239: @dots{}
3240: @end example
3241:
3242: It is up to you to make sure that the assembler names you choose do not
3243: conflict with any other assembler symbols. Also, you must not use a
3244: register name; that would produce completely invalid assembler code. GNU
3245: CC does not as yet have the ability to store static variables in registers.
3246: Perhaps that will be added.
3247:
1.1.1.8 root 3248: @node Explicit Reg Vars, Alternate Keywords, Asm Labels, Extensions
3249: @section Variables in Specified Registers
3250:
3251: GNU C allows you to put a few global variables into specified hardware
3252: registers. You can also specify the register in which an ordinary
3253: register variable should be allocated.
3254:
3255: @itemize @bullet
3256: @item
3257: Global register variables reserve registers throughout the program.
3258: This may be useful in programs such as programming language
3259: interpreters which have a couple of global variables that are accessed
3260: very often.
3261:
3262: @item
3263: Local register variables in specific registers do not reserve the
3264: registers. The compiler's data flow analysis is capable of
3265: determining where the specified registers contain live values, and
3266: where they are available for other uses. These local variables are
3267: sometimes convenient for use with the extended @code{asm} feature
3268: (@pxref{Extended Asm}).
3269: @end itemize
3270:
3271: @menu
3272: * Global Reg Vars::
3273: * Local Reg Vars::
3274: @end menu
1.1.1.5 root 3275:
1.1.1.8 root 3276: @node Global Reg Vars, Local Reg Vars, Explicit Reg Vars, Explicit Reg Vars
3277: @subsection Defining Global Register Variables
1.1.1.5 root 3278:
3279: You can define a global register variable in GNU C like this:
3280:
3281: @example
3282: register int *foo asm ("a5");
3283: @end example
3284:
3285: @noindent
3286: Here @code{a5} is the name of the register which should be used. Choose a
3287: register which is normally saved and restored by function calls on your
3288: machine, so that library routines will not clobber it.
3289:
3290: Naturally the register name is cpu-dependent, so you would need to
3291: conditionalize your program according to cpu type. The register
3292: @code{a5} would be a good choice on a 68000 for a variable of pointer
3293: type. On machines with register windows, be sure to choose a ``global''
1.1.1.8 root 3294: register that is not affected magically by the function call mechanism.
1.1.1.5 root 3295:
3296: In addition, operating systems on one type of cpu may differ in how they
3297: name the registers; then you would need additional conditionals. For
3298: example, some 68000 operating systems call this register @code{%a5}.
3299:
3300: Eventually there may be a way of asking the compiler to choose a register
3301: automatically, but first we need to figure out how it should choose and
1.1.1.6 root 3302: how to enable you to guide the choice. No solution is evident.
1.1.1.5 root 3303:
3304: Defining a global register variable in a certain register reserves that
3305: register entirely for this use, at least within the current compilation.
3306: The register will not be allocated for any other purpose in the functions
3307: in the current compilation. The register will not be saved and restored by
3308: these functions. Stores into this register are never deleted even if they
3309: would appear to be dead, but references may be deleted or moved or
3310: simplified.
3311:
3312: It is not safe to access the global register variables from signal
3313: handlers, or from more than one thread of control, because the system
3314: library routines may temporarily use the register for other things (unless
3315: you recompile them specially for the task at hand).
3316:
3317: It is not safe for one function that uses a global register variable to
3318: call another such function @code{foo} by way of a third function
3319: @code{lose} that was compiled without knowledge of this variable (i.e. in a
3320: different source file in which the variable wasn't declared). This is
3321: because @code{lose} might save the register and put some other value there.
3322: For example, you can't expect a global register variable to be available in
3323: the comparison-function that you pass to @code{qsort}, since @code{qsort}
3324: might have put something else in that register. (If you are prepared to
3325: recompile @code{qsort} with the same global register variable, you can
3326: solve this problem.)
3327:
3328: If you want to recompile @code{qsort} or other source files which do not
3329: actually use your global register variable, so that they will not use that
3330: register for any other purpose, then it suffices to specify the compiler
3331: option @samp{-ffixed-@var{reg}}. You need not actually add a global
3332: register declaration to their source code.
3333:
3334: A function which can alter the value of a global register variable cannot
3335: safely be called from a function compiled without this variable, because it
3336: could clobber the value the caller expects to find there on return.
3337: Therefore, the function which is the entry point into the part of the
3338: program that uses the global register variable must explicitly save and
3339: restore the value which belongs to its caller.
3340:
3341: On most machines, @code{longjmp} will restore to each global register
3342: variable the value it had at the time of the @code{setjmp}. On some
3343: machines, however, @code{longjmp} will not change the value of global
3344: register variables. To be portable, the function that called @code{setjmp}
3345: should make other arrangements to save the values of the global register
1.1.1.9 ! root 3346: variables, and to restore them in a @code{longjmp}. This way, the the same
1.1.1.5 root 3347: thing will happen regardless of what @code{longjmp} does.
3348:
3349: All global register variable declarations must precede all function
3350: definitions. If such a declaration could appear after function
3351: definitions, the declaration would be too late to prevent the register from
3352: being used for other purposes in the preceding functions.
3353:
1.1.1.6 root 3354: Global register variables may not have initial values, because an
3355: executable file has no means to supply initial contents for a register.
3356:
1.1.1.9 ! root 3357: @node Local Reg Vars,, Global Reg Vars, Explicit Reg Vars
1.1.1.8 root 3358: @subsection Specifying Registers for Local Variables
3359:
3360: You can define a local register variable with a specified register
3361: like this:
3362:
3363: @example
3364: register int *foo asm ("a5");
3365: @end example
3366:
3367: @noindent
3368: Here @code{a5} is the name of the register which should be used. Note
3369: that this is the same syntax used for defining global register
3370: variables, but for a local variable it would appear within a function.
3371:
3372: Naturally the register name is cpu-dependent, but this is not a
3373: problem, since specific registers are most often useful with explicit
3374: assembler instructions (@pxref{Extended Asm}). Both of these things
3375: generally require that you conditionalize your program according to
3376: cpu type.
3377:
3378: In addition, operating systems on one type of cpu may differ in how they
3379: name the registers; then you would need additional conditionals. For
3380: example, some 68000 operating systems call this register @code{%a5}.
3381:
3382: Eventually there may be a way of asking the compiler to choose a register
3383: automatically, but first we need to figure out how it should choose and
3384: how to enable you to guide the choice. No solution is evident.
3385:
3386: Defining such a register variable does not reserve the register; it
3387: remains available for other uses in places where flow control
3388: determines the variable's value is not live. However, these registers
3389: made unavailable for use in the reload pass. I would not be surprised
3390: if excessive use of this feature leaves the compiler too few available
3391: registers to compile certain functions.
3392:
3393: @node Alternate Keywords,, Explicit Reg Vars, Extensions
1.1.1.7 root 3394: @section Alternate Keywords
3395:
3396: The option @samp{-traditional} disables certain keywords; @samp{-ansi}
3397: disables certain others. This causes trouble when you want to use GNU C
3398: extensions, or ANSI C features, in a general-purpose header file that
3399: should be usable by all programs, including ANSI C programs and traditional
3400: ones. The keywords @code{asm}, @code{typeof} and @code{inline} cannot be
3401: used since they won't work in a program compiled with @samp{-ansi}, while
3402: the keywords @code{const}, @code{volatile}, @code{signed}, @code{typeof}
3403: and @code{inline} won't work in a program compiled with
3404: @samp{-traditional}.@refill
3405:
1.1.1.8 root 3406: The way to solve these problems is to put @samp{__} at the beginning and
3407: end of each problematical keyword. For example, use @code{__asm__}
3408: instead of @code{asm}, @code{__const__} instead of @code{const}, and
3409: @code{__inline__} instead of @code{inline}.
1.1.1.7 root 3410:
3411: Other C compilers won't accept these alternative keywords; if you want to
3412: compile with another compiler, you can define the alternate keywords as
3413: macros to replace them with the customary keywords. It looks like this:
3414:
3415: @example
3416: #ifndef __GNUC__
1.1.1.8 root 3417: #define __asm__ asm
1.1.1.7 root 3418: #endif
3419: @end example
3420:
1.1 root 3421: @node Bugs, Portability, Extensions, Top
3422: @chapter Reporting Bugs
3423:
3424: Your bug reports play an essential role in making GNU CC reliable.
3425:
3426: Reporting a bug may help you by bringing a solution to your problem, or it
3427: may not. But in any case the important function of a bug report is to help
3428: the entire community by making the next version of GNU CC work better. Bug
3429: reports are your contribution to the maintenance of GNU CC.
3430:
3431: In order for a bug report to serve its purpose, you must include the
3432: information that makes for fixing the bug.
3433:
3434: @menu
3435: * Criteria: Bug Criteria. Have you really found a bug?
3436: * Reporting: Bug Reporting. How to report a bug effectively.
3437: @end menu
3438:
3439: @node Bug Criteria, Bug Reporting, Bugs, Bugs
3440: @section Have You Found a Bug?
3441:
3442: If you are not sure whether you have found a bug, here are some guidelines:
3443:
3444: @itemize @bullet
3445: @item
3446: If the compiler gets a fatal signal, for any input whatever, that is a
3447: compiler bug. Reliable compilers never crash.
3448:
3449: @item
3450: If the compiler produces invalid assembly code, for any input whatever
3451: (except an @code{asm} statement), that is a compiler bug, unless the
3452: compiler reports errors (not just warnings) which would ordinarily
3453: prevent the assembler from being run.
3454:
3455: @item
3456: If the compiler produces valid assembly code that does not correctly
3457: execute the input source code, that is a compiler bug.
3458:
3459: However, you must double-check to make sure, because you may have run
3460: into an incompatibility between GNU C and traditional C
3461: (@pxref{Incompatibilities}). These incompatibilities might be considered
3462: bugs, but they are inescapable consequences of valuable features.
3463:
3464: Or you may have a program whose behavior is undefined, which happened
3465: by chance to give the desired results with another C compiler.
3466:
3467: For example, in many nonoptimizing compilers, you can write @samp{x;}
3468: at the end of a function instead of @samp{return x;}, with the same
1.1.1.8 root 3469: results. But the value of the function is undefined if @code{return}
1.1 root 3470: is omitted; it is not a bug when GNU CC produces different results.
3471:
3472: Problems often result from expressions with two increment operators,
1.1.1.8 root 3473: as in @code{f (*p++, *p++)}. Your previous compiler might have
1.1 root 3474: interpreted that expression the way you intended; GNU CC might
1.1.1.8 root 3475: interpret it another way. Neither compiler is wrong. The bug is
3476: in your code.
1.1 root 3477:
3478: After you have localized the error to a single source line, it should
3479: be easy to check for these things. If your program is correct and
3480: well defined, you have found a compiler bug.
3481:
3482: @item
3483: If the compiler produces an error message for valid input, that is a
3484: compiler bug.
3485:
3486: Note that the following is not valid input, and the error message for
3487: it is not a bug:
3488:
3489: @example
3490: int foo (char);
3491:
3492: int
3493: foo (x)
3494: char x;
3495: @{ @dots{} @}
3496: @end example
3497:
3498: @noindent
3499: The prototype says to pass a @code{char}, while the definition says to
3500: pass an @code{int} and treat the value as a @code{char}. This is what
3501: the ANSI standard says, and it makes sense.
3502:
3503: @item
3504: If the compiler does not produce an error message for invalid input,
3505: that is a compiler bug. However, you should note that your idea of
3506: ``invalid input'' might be my idea of ``an extension'' or ``support
3507: for traditional practice''.
3508:
3509: @item
3510: If you are an experienced user of C compilers, your suggestions
3511: for improvement of GNU CC are welcome in any case.
3512: @end itemize
3513:
3514: @node Bug Reporting,, Bug Criteria, Bugs
3515: @section How to Report Bugs
3516:
3517: Send bug reports for GNU C to one of these addresses:
3518:
3519: @example
3520: bug-gcc@@prep.ai.mit.edu
3521: @{ucbvax|mit-eddie|uunet@}!prep.ai.mit.edu!bug-gcc
3522: @end example
3523:
1.1.1.8 root 3524: @strong{Do not send bug reports to @samp{info-gcc}, or to the newsgroup
3525: @samp{gnu.gcc}.} Most users of GNU CC do not want to receive bug
3526: reports. Those that do, have asked to be on @samp{bug-gcc}.
3527:
3528: The mailing list @samp{bug-gcc} has a newsgroup which serves as a
3529: repeater. The mailing list and the newsgroup carry exactly the same
3530: messages. Often people think of posting bug reports to the newsgroup
3531: instead of mailing them. This appears to work, but it has one problem
3532: which can be crucial: a newsgroup posting does not contain a mail path
3533: back to the sender. Thus, if I need to ask for more information, I
3534: may be unable to reach you. For this reason, it is better to send bug
3535: reports to the mailing list.
3536:
3537: As a last resort, send bug reports on paper to:
1.1 root 3538:
3539: @example
3540: GNU Compiler Bugs
3541: 545 Tech Sq
3542: Cambridge, MA 02139
3543: @end example
3544:
3545: The fundamental principle of reporting bugs usefully is this:
1.1.1.8 root 3546: @strong{report all the facts}. If you are not sure whether to state a
3547: fact or leave it out, state it!
1.1 root 3548:
3549: Often people omit facts because they think they know what causes the
3550: problem and they conclude that some details don't matter. Thus, you might
3551: assume that the name of the variable you use in an example does not matter.
3552: Well, probably it doesn't, but one cannot be sure. Perhaps the bug is a
3553: stray memory reference which happens to fetch from the location where that
3554: name is stored in memory; perhaps, if the name were different, the contents
3555: of that location would fool the compiler into doing the right thing despite
1.1.1.8 root 3556: the bug. Play it safe and give a specific, complete example. That is the
3557: easiest thing for you to do, and the most helpful.
1.1 root 3558:
1.1.1.8 root 3559: Keep in mind that the purpose of a bug report is to enable me to fix
3560: the bug if it is not known. It isn't very important what happens if
3561: the bug is already known. Therefore, always write your bug reports on
3562: the assumption that the bug is not known.
3563:
3564: Sometimes people give a few sketchy facts and ask, ``Does this ring a
3565: bell?'' Those bug reports are useless, and I urge everyone to
3566: @emph{refuse to respond to them} except to chide the sender to report
3567: bugs properly.
3568:
3569: To enable me to fix the bug, you should include all these things:
1.1 root 3570:
3571: @itemize @bullet
3572: @item
3573: The version of GNU CC. You can get this by running it with the
3574: @samp{-v} option.
3575:
3576: Without this, I won't know whether there is any point in looking for
3577: the bug in the current version of GNU CC.
3578:
3579: @item
3580: A complete input file that will reproduce the bug. If the bug is in
3581: the C preprocessor, send me a source file and any header files that it
3582: requires. If the bug is in the compiler proper (@file{cc1}), run your
3583: source file through the C preprocessor by doing @samp{gcc -E
3584: @var{sourcefile} > @var{outfile}}, then include the contents of
3585: @var{outfile} in the bug report. (Any @samp{-I}, @samp{-D} or
3586: @samp{-U} options that you used in actual compilation should also be
3587: used when doing this.)
3588:
3589: A single statement is not enough of an example. In order to compile
3590: it, it must be embedded in a function definition; and the bug might
3591: depend on the details of how this is done.
3592:
3593: Without a real example I can compile, all I can do about your bug
3594: report is wish you luck. It would be futile to try to guess how to
3595: provoke the bug. For example, bugs in register allocation and
3596: reloading frequently depend on every little detail of the function
3597: they happen in.
3598:
3599: @item
3600: The command arguments you gave GNU CC to compile that example and
3601: observe the bug. For example, did you use @samp{-O}? To guarantee
3602: you won't omit something important, list them all.
3603:
3604: If I were to try to guess the arguments, I would probably guess wrong
3605: and then I would not encounter the bug.
3606:
3607: @item
3608: The names of the files that you used for @file{tm.h} and @file{md}
3609: when you installed the compiler.
3610:
3611: @item
3612: The type of machine you are using, and the operating system name and
3613: version number.
3614:
3615: @item
3616: A description of what behavior you observe that you believe is
3617: incorrect. For example, ``It gets a fatal signal,'' or, ``There is an
3618: incorrect assembler instruction in the output.''
3619:
3620: Of course, if the bug is that the compiler gets a fatal signal, then I
3621: will certainly notice it. But if the bug is incorrect output, I might
3622: not notice unless it is glaringly wrong. I won't study all the
3623: assembler code from a 50-line C program just on the off chance that it
3624: might be wrong.
3625:
3626: Even if the problem you experience is a fatal signal, you should still
3627: say so explicitly. Suppose something strange is going on, such as,
3628: your copy of the compiler is out of synch, or you have encountered a
3629: bug in the C library on your system. (This has happened!) Your copy
3630: might crash and mine would not. If you @i{told} me to expect a crash,
3631: then when mine fails to crash, I would know that the bug was not
3632: happening for me. If you had not told me to expect a crash, then I
3633: would not be able to draw any conclusion from my observations.
3634:
1.1.1.8 root 3635: Often the observed symptom is incorrect output when your program is run.
3636: Sad to say, this is not enough information for me unless the program is
3637: short and simple. If you send me a large program, I don't have time to
3638: figure out how it would work if compiled correctly, much less which line
3639: of it was compiled wrong. So you will have to do that. Tell me which
3640: source line it is, and what incorrect result happens when that line is
3641: executed. A person who understands the test program can find this as
3642: easily as a bug in the program itself.
1.1 root 3643:
3644: @item
3645: If you send me examples of output from GNU CC, please use @samp{-g}
3646: when you make them. The debugging information includes source line
3647: numbers which are essential for correlating the output with the input.
3648:
3649: @item
3650: If you wish to suggest changes to the GNU CC source, send me context
3651: diffs. If you even discuss something in the GNU CC source, refer to
3652: it by context, not by line number.
3653:
3654: The line numbers in my development sources don't match those in your
3655: sources. Your line numbers would convey no useful information to me.
3656:
3657: @item
3658: Additional information from a debugger might enable me to find
3659: a problem on a machine which I do not have available myself.
3660: However, you need to think when you collect this information if
3661: you want it to have any chance of being useful.
3662:
3663: For example, many people send just a backtrace, but that is never
3664: useful by itself. A simple backtrace with arguments conveys little
3665: about GNU CC because the compiler is largely data-driven; the same
3666: functions are called over and over for different RTL insns, doing
3667: different things depending on the details of the insn.
3668:
3669: Most of the arguments listed in the backtrace are useless because they
3670: are pointers to RTL list structure. The numeric values of the
3671: pointers, which the debugger prints in the backtrace, have no
3672: significance whatever; all that matters is the contents of the objects
3673: they point to (and most of the contents are other such pointers).
3674:
3675: In addition, most compiler passes consist of one or more loops that
3676: scan the RTL insn sequence. The most vital piece of information about
1.1.1.8 root 3677: such a loop---which insn it has reached---is usually in a local variable,
1.1 root 3678: not in an argument.
3679:
3680: What you need to provide in addition to a backtrace are the values of
3681: the local variables for several stack frames up. When a local
3682: variable or an argument is an RTX, first print its value and then use
3683: the GDB command @code{pr} to print the RTL expression that it points
3684: to. (If GDB doesn't run on your machine, use your debugger to call
3685: the function @code{debug_rtx} with the RTX as an argument.) In
3686: general, whenever a variable is a pointer, its value is no use
3687: without the data it points to.
3688:
3689: In addition, include a debugging dump from just before the pass
3690: in which the crash happens. Most bugs involve a series of insns,
3691: not just one.
3692: @end itemize
3693:
3694: Here are some things that are not necessary:
3695:
3696: @itemize @bullet
3697: @item
3698: A description of the envelope of the bug.
3699:
3700: Often people who encounter a bug spend a lot of time investigating
3701: which changes to the input file will make the bug go away and which
3702: changes will not affect it.
3703:
3704: This is often time consuming and not very useful, because the way I
3705: will find the bug is by running a single example under the debugger
3706: with breakpoints, not by pure deduction from a series of examples.
1.1.1.8 root 3707: I recommend that you save your time for something else.
1.1 root 3708:
3709: Of course, if you can find a simpler example to report @emph{instead}
3710: of the original one, that is a convenience for me. Errors in the
3711: output will be easier to spot, running under the debugger will take
3712: less time, etc. Most GNU CC bugs involve just one function, so the
3713: most straightforward way to simplify an example is to delete all the
3714: function definitions except the one where the bug occurs. Those
3715: earlier in the file may be replaced by external declarations if the
1.1.1.8 root 3716: crucial function depends on them. (Exception: inline functions may
3717: affect compilation of functions defined later in the file.)
1.1 root 3718:
3719: However, simplification is not vital; if you don't want to do this,
1.1.1.8 root 3720: report the bug anyway and send me the entire test case you used.
1.1 root 3721:
3722: @item
3723: A patch for the bug.
3724:
3725: A patch for the bug does help me if it is a good one. But don't omit
1.1.1.8 root 3726: the necessary information, such as the test case, on the assumption that
3727: a patch is all I need. I might see problems with your patch and decide
3728: to fix the problem another way, or I might not understand it at all.
1.1 root 3729:
3730: Sometimes with a program as complicated as GNU CC it is very hard to
3731: construct an example that will make the program follow a certain path
3732: through the code. If you don't send me the example, I won't be able
3733: to construct one, so I won't be able to verify that the bug is fixed.
3734:
1.1.1.8 root 3735: And if I can't understand what bug you are trying to fix, or why your
3736: patch should be an improvement, I won't install it. A test case will
3737: help me to understand.
3738:
1.1 root 3739: @item
3740: A guess about what the bug is or what it depends on.
3741:
3742: Such guesses are usually wrong. Even I can't guess right about such
1.1.1.8 root 3743: things without first using the debugger to find the facts.
1.1 root 3744: @end itemize
3745:
3746: @node Portability, Interface, Bugs, Top
3747: @chapter GNU CC and Portability
3748:
3749: The main goal of GNU CC was to make a good, fast compiler for machines in
3750: the class that the GNU system aims to run on: 32-bit machines that address
3751: 8-bit bytes and have several general registers. Elegance, theoretical
3752: power and simplicity are only secondary.
3753:
3754: GNU CC gets most of the information about the target machine from a machine
3755: description which gives an algebraic formula for each of the machine's
3756: instructions. This is a very clean way to describe the target. But when
3757: the compiler needs information that is difficult to express in this
3758: fashion, I have not hesitated to define an ad-hoc parameter to the machine
3759: description. The purpose of portability is to reduce the total work needed
3760: on the compiler; it was not of interest for its own sake.
3761:
3762: GNU CC does not contain machine dependent code, but it does contain code
3763: that depends on machine parameters such as endianness (whether the most
3764: significant byte has the highest or lowest address of the bytes in a word)
3765: and the availability of autoincrement addressing. In the RTL-generation
3766: pass, it is often necessary to have multiple strategies for generating code
3767: for a particular kind of syntax tree, strategies that are usable for different
3768: combinations of parameters. Often I have not tried to address all possible
3769: cases, but only the common ones or only the ones that I have encountered.
3770: As a result, a new target may require additional strategies. You will know
3771: if this happens because the compiler will call @code{abort}. Fortunately,
3772: the new strategies can be added in a machine-independent fashion, and will
3773: affect only the target machines that need them.
3774:
3775: @node Interface, Passes, Portability, Top
3776: @chapter Interfacing to GNU CC Output
3777:
3778: GNU CC is normally configured to use the same function calling convention
3779: normally in use on the target system. This is done with the
3780: machine-description macros described (@pxref{Machine Macros}).
3781:
3782: However, returning of structure and union values is done differently on
3783: some target machines. As a result, functions compiled with PCC
3784: returning such types cannot be called from code compiled with GNU CC,
3785: and vice versa. This does not cause trouble often because few Unix
3786: library routines return structures or unions.
3787:
3788: GNU CC code returns structures and unions that are 1, 2, 4 or 8 bytes
3789: long in the same registers used for @code{int} or @code{double} return
3790: values. (GNU CC typically allocates variables of such types in
3791: registers also.) Structures and unions of other sizes are returned by
3792: storing them into an address passed by the caller (usually in a
3793: register). The machine-description macros @code{STRUCT_VALUE} and
3794: @code{STRUCT_INCOMING_VALUE} tell GNU CC where to pass this address.
3795:
3796: By contrast, PCC on most target machines returns structures and unions
3797: of any size by copying the data into an area of static storage, and then
3798: returning the address of that storage as if it were a pointer value.
3799: The caller must copy the data from that memory area to the place where
3800: the value is wanted. This is slower than the method used by GNU CC, and
3801: fails to be reentrant.
3802:
3803: On some target machines, such as RISC machines and the 80386, the
3804: standard system convention is to pass to the subroutine the address of
3805: where to return the value. On these machines, GNU CC has been
3806: configured to be compatible with the standard compiler, when this method
3807: is used. It may not be compatible for structures of 1, 2, 4 or 8 bytes.
3808:
3809: GNU CC uses the system's standard convention for passing arguments. On
3810: some machines, the first few arguments are passed in registers; in
3811: others, all are passed on the stack. It would be possible to use
3812: registers for argument passing on any machine, and this would probably
3813: result in a significant speedup. But the result would be complete
3814: incompatibility with code that follows the standard convention. So this
3815: change is practical only if you are switching to GNU CC as the sole C
3816: compiler for the system. We may implement register argument passing on
3817: certain machines once we have a complete GNU system so that we can
3818: compile the libraries with GNU CC.
3819:
3820: If you use @code{longjmp}, beware of automatic variables. ANSI C says that
3821: automatic variables that are not declared @code{volatile} have undefined
3822: values after a @code{longjmp}. And this is all GNU CC promises to do,
3823: because it is very difficult to restore register variables correctly, and
3824: one of GNU CC's features is that it can put variables in registers without
3825: your asking it to.
3826:
3827: If you want a variable to be unaltered by @code{longjmp}, and you don't
3828: want to write @code{volatile} because old C compilers don't accept it,
3829: just take the address of the variable. If a variable's address is ever
3830: taken, even if just to compute it and ignore it, then the variable cannot
3831: go in a register:
3832:
3833: @example
3834: @{
3835: int careful;
3836: &careful;
3837: @dots{}
3838: @}
3839: @end example
3840:
3841: Code compiled with GNU CC may call certain library routines. Most of
3842: them handle arithmetic for which there are no instructions. This
3843: includes multiply and divide on some machines, and floating point
3844: operations on any machine for which floating point support is disabled
3845: with @samp{-msoft-float}. Some standard parts of the C library, such as
3846: @code{bcopy} or @code{memcpy}, are also called automatically. The usual
3847: function call interface is used for calling the library routines.
3848:
3849: These library routines should be defined in the library @file{gnulib},
3850: which GNU CC automatically searches whenever it links a program. On
3851: machines that have multiply and divide instructions, if hardware
3852: floating point is in use, normally @file{gnulib} is not needed, but it
3853: is searched just in case.
3854:
3855: Each arithmetic function is defined in @file{gnulib.c} to use the
3856: corresponding C arithmetic operator. As long as the file is compiled
3857: with another C compiler, which supports all the C arithmetic operators,
3858: this file will work portably. However, @file{gnulib.c} does not work if
3859: compiled with GNU CC, because each arithmetic function would compile
3860: into a call to itself!
3861:
3862: @node Passes, RTL, Interface, Top
3863: @chapter Passes and Files of the Compiler
3864:
3865: The overall control structure of the compiler is in @file{toplev.c}. This
3866: file is responsible for initialization, decoding arguments, opening and
3867: closing files, and sequencing the passes.
3868:
3869: The parsing pass is invoked only once, to parse the entire input. The RTL
3870: intermediate code for a function is generated as the function is parsed, a
3871: statement at a time. Each statement is read in as a syntax tree and then
3872: converted to RTL; then the storage for the tree for the statement is
3873: reclaimed. Storage for types (and the expressions for their sizes),
3874: declarations, and a representation of the binding contours and how they nest,
3875: remains until the function is finished being compiled; these are all needed
3876: to output the debugging information.
3877:
3878: Each time the parsing pass reads a complete function definition or
3879: top-level declaration, it calls the function
3880: @code{rest_of_compilation} or @code{rest_of_decl_compilation} in
3881: @file{toplev.c}, which are responsible for all further processing
3882: necessary, ending with output of the assembler language. All other
3883: compiler passes run, in sequence, within @code{rest_of_compilation}.
3884: When that function returns from compiling a function definition, the
3885: storage used for that function definition's compilation is entirely
3886: freed, unless it is an inline function (@pxref{Inline}).
3887:
3888: Here is a list of all the passes of the compiler and their source files.
3889: Also included is a description of where debugging dumps can be requested
3890: with @samp{-d} options.
3891:
3892: @itemize @bullet
3893: @item
3894: Parsing. This pass reads the entire text of a function definition,
3895: constructing partial syntax trees. This and RTL generation are no longer
3896: truly separate passes (formerly they were), but it is easier to think
3897: of them as separate.
3898:
3899: The tree representation does not entirely follow C syntax, because it is
3900: intended to support other languages as well.
3901:
3902: C data type analysis is also done in this pass, and every tree node
3903: that represents an expression has a data type attached. Variables are
3904: represented as declaration nodes.
3905:
3906: Constant folding and associative-law simplifications are also done
3907: during this pass.
3908:
3909: The source files for parsing are @file{c-parse.y}, @file{c-decl.c},
3910: @file{c-typeck.c}, @file{c-convert.c}, @file{stor-layout.c},
3911: @file{fold-const.c}, and @file{tree.c}. The last three files are
3912: intended to be language-independent. There are also header files
3913: @file{c-parse.h}, @file{c-tree.h}, @file{tree.h} and @file{tree.def}.
3914: The last two define the format of the tree representation.@refill
3915:
3916: @item
3917: RTL generation. This is the conversion of syntax tree into RTL code.
3918: It is actually done statement-by-statement during parsing, but for
3919: most purposes it can be thought of as a separate pass.
3920:
3921: This is where the bulk of target-parameter-dependent code is found,
3922: since often it is necessary for strategies to apply only when certain
3923: standard kinds of instructions are available. The purpose of named
3924: instruction patterns is to provide this information to the RTL
3925: generation pass.
3926:
3927: Optimization is done in this pass for @code{if}-conditions that are
3928: comparisons, boolean operations or conditional expressions. Tail
3929: recursion is detected at this time also. Decisions are made about how
3930: best to arrange loops and how to output @code{switch} statements.
3931:
3932: The source files for RTL generation are @file{stmt.c}, @file{expr.c},
3933: @file{explow.c}, @file{expmed.c}, @file{optabs.c} and @file{emit-rtl.c}.
3934: Also, the file @file{insn-emit.c}, generated from the machine description
3935: by the program @code{genemit}, is used in this pass. The header files
3936: @file{expr.h} is used for communication within this pass.@refill
3937:
3938: The header files @file{insn-flags.h} and @file{insn-codes.h},
3939: generated from the machine description by the programs @code{genflags}
3940: and @code{gencodes}, tell this pass which standard names are available
3941: for use and which patterns correspond to them.@refill
3942:
3943: Aside from debugging information output, none of the following passes
3944: refers to the tree structure representation of the function (only
3945: part of which is saved).
3946:
3947: The decision of whether the function can and should be expanded inline
3948: in its subsequent callers is made at the end of rtl generation. The
3949: function must meet certain criteria, currently related to the size of
3950: the function and the types and number of parameters it has. Note that
3951: this function may contain loops, recursive calls to itself
3952: (tail-recursive functions can be inlined!), gotos, in short, all
3953: constructs supported by GNU CC.
3954:
3955: The option @samp{-dr} causes a debugging dump of the RTL code after
3956: this pass. This dump file's name is made by appending @samp{.rtl} to
3957: the input file name.
3958:
3959: @item
3960: Jump optimization. This pass simplifies jumps to the following
3961: instruction, jumps across jumps, and jumps to jumps. It deletes
3962: unreferenced labels and unreachable code, except that unreachable code
3963: that contains a loop is not recognized as unreachable in this pass.
3964: (Such loops are deleted later in the basic block analysis.)
3965:
3966: Jump optimization is performed two or three times. The first time is
3967: immediately following RTL generation. The second time is after CSE,
3968: but only if CSE says repeated jump optimization is needed. The
3969: last time is right before the final pass. That time, cross-jumping
3970: and deletion of no-op move instructions are done together with the
3971: optimizations described above.
3972:
3973: The source file of this pass is @file{jump.c}.
3974:
3975: The option @samp{-dj} causes a debugging dump of the RTL code after
3976: this pass is run for the first time. This dump file's name is made by
3977: appending @samp{.jump} to the input file name.
3978:
3979: @item
3980: Register scan. This pass finds the first and last use of each
3981: register, as a guide for common subexpression elimination. Its source
3982: is in @file{regclass.c}.
3983:
3984: @item
3985: Common subexpression elimination. This pass also does constant
3986: propagation. Its source file is @file{cse.c}. If constant
3987: propagation causes conditional jumps to become unconditional or to
3988: become no-ops, jump optimization is run again when CSE is finished.
3989:
3990: The option @samp{-ds} causes a debugging dump of the RTL code after
3991: this pass. This dump file's name is made by appending @samp{.cse} to
3992: the input file name.
3993:
3994: @item
1.1.1.8 root 3995: Loop optimization. This pass moves constant expressions out of loops,
3996: and optionally does strength-reduction as well. Its source file is
3997: @file{loop.c}.
1.1 root 3998:
3999: The option @samp{-dL} causes a debugging dump of the RTL code after
4000: this pass. This dump file's name is made by appending @samp{.loop} to
4001: the input file name.
4002:
4003: @item
4004: Stupid register allocation is performed at this point in a
4005: nonoptimizing compilation. It does a little data flow analysis as
4006: well. When stupid register allocation is in use, the next pass
4007: executed is the reloading pass; the others in between are skipped.
4008: The source file is @file{stupid.c}.
4009:
4010: @item
4011: Data flow analysis (@file{flow.c}). This pass divides the program
4012: into basic blocks (and in the process deletes unreachable loops); then
4013: it computes which pseudo-registers are live at each point in the
4014: program, and makes the first instruction that uses a value point at
4015: the instruction that computed the value.
4016:
4017: This pass also deletes computations whose results are never used, and
4018: combines memory references with add or subtract instructions to make
4019: autoincrement or autodecrement addressing.
4020:
4021: The option @samp{-df} causes a debugging dump of the RTL code after
4022: this pass. This dump file's name is made by appending @samp{.flow} to
4023: the input file name. If stupid register allocation is in use, this
4024: dump file reflects the full results of such allocation.
4025:
4026: @item
4027: Instruction combination (@file{combine.c}). This pass attempts to
4028: combine groups of two or three instructions that are related by data
4029: flow into single instructions. It combines the RTL expressions for
4030: the instructions by substitution, simplifies the result using algebra,
4031: and then attempts to match the result against the machine description.
4032:
4033: The option @samp{-dc} causes a debugging dump of the RTL code after
4034: this pass. This dump file's name is made by appending @samp{.combine}
4035: to the input file name.
4036:
4037: @item
4038: Register class preferencing. The RTL code is scanned to find out
4039: which register class is best for each pseudo register. The source
4040: file is @file{regclass.c}.
4041:
4042: @item
4043: Local register allocation (@file{local-alloc.c}). This pass allocates
4044: hard registers to pseudo registers that are used only within one basic
4045: block. Because the basic block is linear, it can use fast and
4046: powerful techniques to do a very good job.
4047:
4048: The option @samp{-dl} causes a debugging dump of the RTL code after
4049: this pass. This dump file's name is made by appending @samp{.lreg} to
4050: the input file name.
4051:
4052: @item
4053: Global register allocation (@file{global-alloc.c}). This pass
4054: allocates hard registers for the remaining pseudo registers (those
4055: whose life spans are not contained in one basic block).
4056:
4057: @item
4058: Reloading. This pass renumbers pseudo registers with the hardware
4059: registers numbers they were allocated. Pseudo registers that did not
4060: get hard registers are replaced with stack slots. Then it finds
4061: instructions that are invalid because a value has failed to end up in
4062: a register, or has ended up in a register of the wrong kind. It fixes
4063: up these instructions by reloading the problematical values
4064: temporarily into registers. Additional instructions are generated to
4065: do the copying.
4066:
4067: Source files are @file{reload.c} and @file{reload1.c}, plus the header
4068: @file{reload.h} used for communication between them.
4069:
4070: The option @samp{-dg} causes a debugging dump of the RTL code after
4071: this pass. This dump file's name is made by appending @samp{.greg} to
4072: the input file name.
4073:
4074: @item
4075: Jump optimization is repeated, this time including cross-jumping
1.1.1.5 root 4076: and deletion of no-op move instructions.
1.1 root 4077:
4078: The option @samp{-dJ} causes a debugging dump of the RTL code after
4079: this pass. This dump file's name is made by appending @samp{.jump2}
4080: to the input file name.
4081:
4082: @item
1.1.1.8 root 4083: Delayed branch scheduling may be done at this point. The source file
4084: name is @file{dbranch.c}.
4085:
4086: The option @samp{-dd} causes a debugging dump of the RTL code after
4087: this pass. This dump file's name is made by appending @samp{.dbr}
4088: to the input file name.
4089:
4090: @item
1.1 root 4091: Final. This pass outputs the assembler code for the function. It is
4092: also responsible for identifying spurious test and compare
1.1.1.5 root 4093: instructions. Machine-specific peephole optimizations are performed
4094: at the same time. The function entry and exit sequences are generated
1.1 root 4095: directly as assembler code in this pass; they never exist as RTL.
4096:
4097: The source files are @file{final.c} plus @file{insn-output.c}; the
4098: latter is generated automatically from the machine description by the
4099: tool @file{genoutput}. The header file @file{conditions.h} is used
4100: for communication between these files.
4101:
4102: @item
4103: Debugging information output. This is run after final because it must
4104: output the stack slot offsets for pseudo registers that did not get
4105: hard registers. Source files are @file{dbxout.c} for DBX symbol table
4106: format and @file{symout.c} for GDB's own symbol table format.
4107: @end itemize
4108:
4109: Some additional files are used by all or many passes:
4110:
4111: @itemize @bullet
4112: @item
4113: Every pass uses @file{machmode.def}, which defines the machine modes.
4114:
4115: @item
4116: All the passes that work with RTL use the header files @file{rtl.h}
4117: and @file{rtl.def}, and subroutines in file @file{rtl.c}. The tools
4118: @code{gen*} also use these files to read and work with the machine
4119: description RTL.
4120:
4121: @item
4122: Several passes refer to the header file @file{insn-config.h} which
4123: contains a few parameters (C macro definitions) generated
4124: automatically from the machine description RTL by the tool
4125: @code{genconfig}.
4126:
4127: @item
4128: Several passes use the instruction recognizer, which consists of
4129: @file{recog.c} and @file{recog.h}, plus the files @file{insn-recog.c}
4130: and @file{insn-extract.c} that are generated automatically from the
4131: machine description by the tools @file{genrecog} and
4132: @file{genextract}.@refill
4133:
4134: @item
4135: Several passes use the header files @file{regs.h} which defines the
4136: information recorded about pseudo register usage, and @file{basic-block.h}
4137: which defines the information recorded about basic blocks.
4138:
4139: @item
4140: @file{hard-reg-set.h} defines the type @code{HARD_REG_SET}, a bit-vector
4141: with a bit for each hard register, and some macros to manipulate it.
4142: This type is just @code{int} if the machine has few enough hard registers;
4143: otherwise it is an array of @code{int} and some of the macros expand
4144: into loops.
4145: @end itemize
4146:
4147: @node RTL, Machine Desc, Passes, Top
4148: @chapter RTL Representation
4149:
4150: Most of the work of the compiler is done on an intermediate representation
4151: called register transfer language. In this language, the instructions to be
4152: output are described, pretty much one by one, in an algebraic form that
4153: describes what the instruction does.
4154:
4155: RTL is inspired by Lisp lists. It has both an internal form, made up of
4156: structures that point at other structures, and a textual form that is used
4157: in the machine description and in printed debugging dumps. The textual
4158: form uses nested parentheses to indicate the pointers in the internal form.
4159:
4160: @menu
4161: * RTL Objects:: Expressions vs vectors vs strings vs integers.
4162: * Accessors:: Macros to access expression operands or vector elts.
4163: * Flags:: Other flags in an RTL expression.
4164: * Machine Modes:: Describing the size and format of a datum.
4165: * Constants:: Expressions with constant values.
4166: * Regs and Memory:: Expressions representing register contents or memory.
4167: * Arithmetic:: Expressions representing arithmetic on other expressions.
4168: * Comparisons:: Expressions representing comparison of expressions.
4169: * Bit Fields:: Expressions representing bit-fields in memory or reg.
4170: * Conversions:: Extending, truncating, floating or fixing.
4171: * RTL Declarations:: Declaring volatility, constancy, etc.
4172: * Side Effects:: Expressions for storing in registers, etc.
4173: * Incdec:: Embedded side-effects for autoincrement addressing.
1.1.1.9 ! root 4174: * Assembler:: Representing @code{asm} with operands.
1.1 root 4175: * Insns:: Expression types for entire insns.
1.1.1.9 ! root 4176: * Calls:: RTL representation of function call insns.
1.1 root 4177: * Sharing:: Some expressions are unique; others *must* be copied.
4178: @end menu
4179:
4180: @node RTL Objects, Accessors, RTL, RTL
4181: @section RTL Object Types
4182:
4183: RTL uses four kinds of objects: expressions, integers, strings and vectors.
4184: Expressions are the most important ones. An RTL expression (``RTX'', for
4185: short) is a C structure, but it is usually referred to with a pointer; a
4186: type that is given the typedef name @code{rtx}.
4187:
4188: An integer is simply an @code{int}, and a string is a @code{char *}.
1.1.1.8 root 4189: Within RTL code, strings appear only inside @code{symbol_ref} expressions,
1.1 root 4190: but they appear in other contexts in the RTL expressions that make up
4191: machine descriptions. Their written form uses decimal digits.
4192:
4193: A string is a sequence of characters. In core it is represented as a
4194: @code{char *} in usual C fashion, and it is written in C syntax as well.
4195: However, strings in RTL may never be null. If you write an empty string in
4196: a machine description, it is represented in core as a null pointer rather
4197: than as a pointer to a null character. In certain contexts, these null
4198: pointers instead of strings are valid.
4199:
4200: A vector contains an arbitrary, specified number of pointers to
4201: expressions. The number of elements in the vector is explicitly present in
4202: the vector. The written form of a vector consists of square brackets
4203: (@samp{[@dots{}]}) surrounding the elements, in sequence and with
4204: whitespace separating them. Vectors of length zero are not created; null
4205: pointers are used instead.
4206:
4207: Expressions are classified by @dfn{expression codes} (also called RTX
4208: codes). The expression code is a name defined in @file{rtl.def}, which is
4209: also (in upper case) a C enumeration constant. The possible expression
4210: codes and their meanings are machine-independent. The code of an RTX can
4211: be extracted with the macro @code{GET_CODE (@var{x})} and altered with
4212: @code{PUT_CODE (@var{x}, @var{newcode})}.
4213:
4214: The expression code determines how many operands the expression contains,
4215: and what kinds of objects they are. In RTL, unlike Lisp, you cannot tell
4216: by looking at an operand what kind of object it is. Instead, you must know
4217: from its context---from the expression code of the containing expression.
1.1.1.8 root 4218: For example, in an expression of code @code{subreg}, the first operand is
1.1 root 4219: to be regarded as an expression and the second operand as an integer. In
1.1.1.8 root 4220: an expression of code @code{plus}, there are two operands, both of which
4221: are to be regarded as expressions. In a @code{symbol_ref} expression,
1.1 root 4222: there is one operand, which is to be regarded as a string.
4223:
4224: Expressions are written as parentheses containing the name of the
4225: expression type, its flags and machine mode if any, and then the operands
4226: of the expression (separated by spaces).
4227:
4228: Expression code names in the @samp{md} file are written in lower case,
4229: but when they appear in C code they are written in upper case. In this
1.1.1.8 root 4230: manual, they are shown as follows: @code{const_int}.
1.1 root 4231:
4232: In a few contexts a null pointer is valid where an expression is normally
1.1.1.4 root 4233: wanted. The written form of this is @code{(nil)}.
1.1 root 4234:
4235: @node Accessors, Flags, RTL Objects, RTL
4236: @section Access to Operands
4237:
4238: For each expression type @file{rtl.def} specifies the number of contained
4239: objects and their kinds, with four possibilities: @samp{e} for expression
4240: (actually a pointer to an expression), @samp{i} for integer, @samp{s} for
4241: string, and @samp{E} for vector of expressions. The sequence of letters
4242: for an expression code is called its @dfn{format}. Thus, the format of
1.1.1.8 root 4243: @code{subreg} is @samp{ei}.@refill
1.1 root 4244:
4245: Two other format characters are used occasionally: @samp{u} and @samp{0}.
4246: @samp{u} is equivalent to @samp{e} except that it is printed differently in
4247: debugging dumps, and @samp{0} means a slot whose contents do not fit any
4248: normal category. @samp{0} slots are not printed at all in dumps, and are
4249: often used in special ways by small parts of the compiler.@refill
4250:
4251: There are macros to get the number of operands and the format of an
4252: expression code:
4253:
4254: @table @code
4255: @item GET_RTX_LENGTH (@var{code})
4256: Number of operands of an RTX of code @var{code}.
4257:
4258: @item GET_RTX_FORMAT (@var{code})
4259: The format of an RTX of code @var{code}, as a C string.
4260: @end table
4261:
4262: Operands of expressions are accessed using the macros @code{XEXP},
4263: @code{XINT} and @code{XSTR}. Each of these macros takes two arguments: an
4264: expression-pointer (RTX) and an operand number (counting from zero).
4265: Thus,@refill
4266:
4267: @example
4268: XEXP (@var{x}, 2)
4269: @end example
4270:
4271: @noindent
4272: accesses operand 2 of expression @var{x}, as an expression.
4273:
4274: @example
4275: XINT (@var{x}, 2)
4276: @end example
4277:
4278: @noindent
4279: accesses the same operand as an integer. @code{XSTR}, used in the same
4280: fashion, would access it as a string.
4281:
4282: Any operand can be accessed as an integer, as an expression or as a string.
4283: You must choose the correct method of access for the kind of value actually
4284: stored in the operand. You would do this based on the expression code of
4285: the containing expression. That is also how you would know how many
4286: operands there are.
4287:
1.1.1.8 root 4288: For example, if @var{x} is a @code{subreg} expression, you know that it has
1.1 root 4289: two operands which can be correctly accessed as @code{XEXP (@var{x}, 0)}
4290: and @code{XINT (@var{x}, 1)}. If you did @code{XINT (@var{x}, 0)}, you
4291: would get the address of the expression operand but cast as an integer;
4292: that might occasionally be useful, but it would be cleaner to write
4293: @code{(int) XEXP (@var{x}, 0)}. @code{XEXP (@var{x}, 1)} would also
4294: compile without error, and would return the second, integer operand cast as
4295: an expression pointer, which would probably result in a crash when
4296: accessed. Nothing stops you from writing @code{XEXP (@var{x}, 28)} either,
4297: but this will access memory past the end of the expression with
4298: unpredictable results.@refill
4299:
4300: Access to operands which are vectors is more complicated. You can use the
4301: macro @code{XVEC} to get the vector-pointer itself, or the macros
4302: @code{XVECEXP} and @code{XVECLEN} to access the elements and length of a
4303: vector.
4304:
4305: @table @code
4306: @item XVEC (@var{exp}, @var{idx})
4307: Access the vector-pointer which is operand number @var{idx} in @var{exp}.
4308:
4309: @item XVECLEN (@var{exp}, @var{idx})
4310: Access the length (number of elements) in the vector which is
4311: in operand number @var{idx} in @var{exp}. This value is an @code{int}.
4312:
4313: @item XVECEXP (@var{exp}, @var{idx}, @var{eltnum})
4314: Access element number @var{eltnum} in the vector which is
4315: in operand number @var{idx} in @var{exp}. This value is an RTX.
4316:
4317: It is up to you to make sure that @var{eltnum} is not negative
4318: and is less than @code{XVECLEN (@var{exp}, @var{idx})}.
4319: @end table
4320:
4321: All the macros defined in this section expand into lvalues and therefore
4322: can be used to assign the operands, lengths and vector elements as well as
4323: to access them.
4324:
4325: @node Flags, Machine Modes, Accessors, RTL
4326: @section Flags in an RTL Expression
4327:
4328: RTL expressions contain several flags (one-bit bit-fields) that are used
4329: in certain types of expression. Most often they are accessed with the
4330: following macros:
4331:
4332: @table @code
4333: @item MEM_VOLATILE_P (@var{x})
1.1.1.8 root 4334: In @code{mem} expressions, nonzero for volatile memory references.
1.1 root 4335: Stored in the @code{volatil} field and printed as @samp{/v}.
4336:
4337: @item MEM_IN_STRUCT_P (@var{x})
1.1.1.8 root 4338: In @code{mem} expressions, nonzero for reference to an entire
1.1 root 4339: structure, union or array, or to a component of one. Zero for
4340: references to a scalar variable or through a pointer to a scalar.
4341: Stored in the @code{in_struct} field and printed as @samp{/s}.
4342:
4343: @item REG_USER_VAR_P (@var{x})
1.1.1.8 root 4344: In a @code{reg}, nonzero if it corresponds to a variable present in
1.1 root 4345: the user's source code. Zero for temporaries generated internally by
4346: the compiler. Stored in the @code{volatil} field and printed as
4347: @samp{/v}.
4348:
4349: @item REG_FUNCTION_VALUE_P (@var{x})
1.1.1.8 root 4350: Nonzero in a @code{reg} if it is the place in which this function's
1.1 root 4351: value is going to be returned. (This happens only in a hard
4352: register.) Stored in the @code{integrated} field and printed as
4353: @samp{/i}.
4354:
4355: The same hard register may be used also for collecting the values of
4356: functions called by this one, but @code{REG_FUNCTION_VALUE_P} is zero
4357: in this kind of use.
4358:
4359: @item RTX_UNCHANGING_P (@var{x})
1.1.1.8 root 4360: Nonzero in a @code{reg} or @code{mem} if the value is not changed
1.1 root 4361: explicitly by the current function. (If it is a memory reference then
4362: it may be changed by other functions or by aliasing.) Stored in the
4363: @code{unchanging} field and printed as @samp{/u}.
4364:
4365: @item RTX_INTEGRATED_P (@var{insn})
4366: Nonzero in an insn if it resulted from an in-line function call.
4367: Stored in the @code{integrated} field and printed as @samp{/i}. This
4368: may be deleted; nothing currently depends on it.
4369:
4370: @item INSN_DELETED_P (@var{insn})
4371: In an insn, nonzero if the insn has been deleted. Stored in the
4372: @code{volatil} field and printed as @samp{/v}.
4373:
4374: @item CONSTANT_POOL_ADDRESS_P (@var{x})
1.1.1.8 root 4375: Nonzero in a @code{symbol_ref} if it refers to part of the current
1.1 root 4376: function's ``constants pool''. These are addresses close to the
4377: beginning of the function, and GNU CC assumes they can be addressed
4378: directly (perhaps with the help of base registers). Stored in the
4379: @code{unchanging} field and printed as @samp{/u}.
4380: @end table
4381:
4382: These are the fields which the above macros refer to:
4383:
4384: @table @code
4385: @item used
4386: This flag is used only momentarily, at the end of RTL generation for a
4387: function, to count the number of times an expression appears in insns.
4388: Expressions that appear more than once are copied, according to the
4389: rules for shared structure (@pxref{Sharing}).
4390:
4391: @item volatil
1.1.1.8 root 4392: This flag is used in @code{mem} and @code{reg} expressions and in insns.
1.1 root 4393: In RTL dump files, it is printed as @samp{/v}.
4394:
1.1.1.8 root 4395: In a @code{mem} expression, it is 1 if the memory reference is volatile.
1.1 root 4396: Volatile memory references may not be deleted, reordered or combined.
4397:
1.1.1.8 root 4398: In a @code{reg} expression, it is 1 if the value is a user-level variable.
1.1 root 4399: 0 indicates an internal compiler temporary.
4400:
4401: In an insn, 1 means the insn has been deleted.
4402:
4403: @item in_struct
1.1.1.8 root 4404: This flag is used in @code{mem} expressions. It is 1 if the memory
1.1 root 4405: datum referred to is all or part of a structure or array; 0 if it is (or
4406: might be) a scalar variable. A reference through a C pointer has 0
4407: because the pointer might point to a scalar variable.
4408:
4409: This information allows the compiler to determine something about possible
4410: cases of aliasing.
4411:
4412: In an RTL dump, this flag is represented as @samp{/s}.
4413:
4414: @item unchanging
1.1.1.8 root 4415: This flag is used in @code{reg} and @code{mem} expressions. 1 means
1.1 root 4416: that the value of the expression never changes (at least within the
4417: current function).
4418:
4419: In an RTL dump, this flag is represented as @samp{/u}.
4420:
4421: @item integrated
4422: In some kinds of expressions, including insns, this flag means the
4423: rtl was produced by procedure integration.
4424:
1.1.1.8 root 4425: In a @code{reg} expression, this flag indicates the register
1.1 root 4426: containing the value to be returned by the current function. On
4427: machines that pass parameters in registers, the same register number
4428: may be used for parameters as well, but this flag is not set on such
4429: uses.
4430: @end table
4431:
4432: @node Machine Modes, Constants, Flags, RTL
4433: @section Machine Modes
4434:
4435: A machine mode describes a size of data object and the representation used
4436: for it. In the C code, machine modes are represented by an enumeration
4437: type, @code{enum machine_mode}, defined in @file{machmode.def}. Each RTL
4438: expression has room for a machine mode and so do certain kinds of tree
4439: expressions (declarations and types, to be precise).
4440:
4441: In debugging dumps and machine descriptions, the machine mode of an RTL
4442: expression is written after the expression code with a colon to separate
4443: them. The letters @samp{mode} which appear at the end of each machine mode
1.1.1.8 root 4444: name are omitted. For example, @code{(reg:SI 38)} is a @code{reg}
1.1 root 4445: expression with machine mode @code{SImode}. If the mode is
4446: @code{VOIDmode}, it is not written at all.
4447:
4448: Here is a table of machine modes.
4449:
4450: @table @code
4451: @item QImode
4452: ``Quarter-Integer'' mode represents a single byte treated as an integer.
4453:
4454: @item HImode
4455: ``Half-Integer'' mode represents a two-byte integer.
4456:
1.1.1.7 root 4457: @item PSImode
4458: ``Partial Single Integer'' mode represents an integer which occupies
4459: four bytes but which doesn't really use all four. On some machines,
4460: this is the right mode to use for pointers.
4461:
1.1 root 4462: @item SImode
4463: ``Single Integer'' mode represents a four-byte integer.
4464:
1.1.1.7 root 4465: @item PDImode
4466: ``Partial Double Integer'' mode represents an integer which occupies
4467: eight bytes but which doesn't really use all eight. On some machines,
4468: this is the right mode to use for certain pointers.
4469:
1.1 root 4470: @item DImode
4471: ``Double Integer'' mode represents an eight-byte integer.
4472:
4473: @item TImode
4474: ``Tetra Integer'' (?) mode represents a sixteen-byte integer.
4475:
4476: @item SFmode
4477: ``Single Floating'' mode represents a single-precision (four byte) floating
4478: point number.
4479:
4480: @item DFmode
4481: ``Double Floating'' mode represents a double-precision (eight byte) floating
4482: point number.
4483:
1.1.1.7 root 4484: @item XFmode
4485: ``Extended Floating'' mode represents a triple-precision (twelve byte)
4486: floating point number. This mode is used for IEEE extended floating
4487: point.
4488:
1.1 root 4489: @item TFmode
4490: ``Tetra Floating'' mode represents a quadruple-precision (sixteen byte)
4491: floating point number.
4492:
4493: @item BLKmode
4494: ``Block'' mode represents values that are aggregates to which none of
4495: the other modes apply. In RTL, only memory references can have this mode,
4496: and only if they appear in string-move or vector instructions. On machines
4497: which have no such instructions, @code{BLKmode} will not appear in RTL.
4498:
4499: @item VOIDmode
4500: Void mode means the absence of a mode or an unspecified mode.
1.1.1.8 root 4501: For example, RTL expressions of code @code{const_int} have mode
1.1 root 4502: @code{VOIDmode} because they can be taken to have whatever mode the context
4503: requires. In debugging dumps of RTL, @code{VOIDmode} is expressed by
4504: the absence of any mode.
4505:
4506: @item EPmode
4507: ``Entry Pointer'' mode is intended to be used for function variables in
4508: Pascal and other block structured languages. Such values contain
4509: both a function address and a static chain pointer for access to
4510: automatic variables of outer levels. This mode is only partially
4511: implemented since C does not use it.
4512:
4513: @item CSImode@r{, @dots{}}
4514: ``Complex Single Integer'' mode stands for a complex number represented
4515: as a pair of @code{SImode} integers. Any of the integer and floating modes
4516: may have @samp{C} prefixed to its name to obtain a complex number mode.
4517: For example, there are @code{CQImode}, @code{CSFmode}, and @code{CDFmode}.
4518: Since C does not support complex numbers, these machine modes are only
4519: partially implemented.
4520:
4521: @item BImode
4522: This is the machine mode of a bit-field in a structure. It is used
4523: only in the syntax tree, never in RTL, and in the syntax tree it appears
4524: only in declaration nodes. In C, it appears only in @code{FIELD_DECL}
4525: nodes for structure fields defined with a bit size.
4526: @end table
4527:
4528: The machine description defines @code{Pmode} as a C macro which expands
4529: into the machine mode used for addresses. Normally this is @code{SImode}.
4530:
4531: The only modes which a machine description @i{must} support are
4532: @code{QImode}, @code{SImode}, @code{SFmode} and @code{DFmode}. The
4533: compiler will attempt to use @code{DImode} for two-word structures and
1.1.1.7 root 4534: unions, but this can be prevented by overriding the definition of
4535: @code{MAX_FIXED_MODE_SIZE}. Likewise, you can arrange for the C type
4536: @code{short int} to avoid using @code{HImode}. In the long term it
4537: might be desirable to make the set of available machine modes
4538: machine-dependent and eliminate all assumptions about specific machine
4539: modes or their uses from the machine-independent code of the compiler.
1.1 root 4540:
1.1.1.4 root 4541: To help begin this process, the machine modes are divided into mode
4542: classes. These are represented by the enumeration type @code{enum
4543: mode_class} defined in @file{rtl.h}. The possible mode classes are:
4544:
4545: @table @code
4546: @item MODE_INT
4547: Integer modes. By default these are @code{QImode}, @code{HImode},
4548: @code{SImode}, @code{DImode}, @code{TImode}, and also @code{BImode}.
4549:
4550: @item MODE_FLOAT
4551: Floating-point modes. By default these are @code{QFmode},
4552: @code{HFmode}, @code{SFmode}, @code{DFmode} and @code{TFmode}, but the
4553: MC68881 also defines @code{XFmode} to be an 80-bit extended-precision
4554: floating-point mode.
4555:
4556: @item MODE_COMPLEX_INT
4557: Complex integer modes. By default these are @code{CQImode},
4558: @code{CHImode}, @code{CSImode}, @code{CDImode} and @code{CTImode}.
4559:
4560: @item MODE_COMPLEX_FLOAT
4561: Complex floating-point modes. By default these are @code{CQFmode},
4562: @code{CHFmode}, @code{CSFmode}, @code{CDFmode} and @code{CTFmode},
4563:
4564: @item MODE_FUNCTION
4565: Algol or Pascal function variables including a static chain.
4566: (These are not currently implemented).
4567:
4568: @item MODE_RANDOM
4569: This is a catchall mode class for modes which don't fit into the above
4570: classes. Currently @code{VOIDmode}, @code{BLKmode} and @code{EPmode}
4571: are in @code{MODE_RANDOM}.
4572: @end table
4573:
1.1 root 4574: Here are some C macros that relate to machine modes:
4575:
4576: @table @code
4577: @item GET_MODE (@var{x})
4578: Returns the machine mode of the RTX @var{x}.
4579:
4580: @item PUT_MODE (@var{x}, @var{newmode})
4581: Alters the machine mode of the RTX @var{x} to be @var{newmode}.
4582:
1.1.1.4 root 4583: @item NUM_MACHINE_MODES
4584: Stands for the number of machine modes available on the target
4585: machine. This is one greater than the largest numeric value of any
4586: machine mode.
4587:
4588: @item GET_MODE_NAME (@var{m})
4589: Returns the name of mode @var{m} as a string.
4590:
4591: @item GET_MODE_CLASS (@var{m})
4592: Returns the mode class of mode @var{m}.
4593:
1.1 root 4594: @item GET_MODE_SIZE (@var{m})
4595: Returns the size in bytes of a datum of mode @var{m}.
4596:
4597: @item GET_MODE_BITSIZE (@var{m})
4598: Returns the size in bits of a datum of mode @var{m}.
4599:
4600: @item GET_MODE_UNIT_SIZE (@var{m})
4601: Returns the size in bits of the subunits of a datum of mode @var{m}.
4602: This is the same as @code{GET_MODE_SIZE} except in the case of
4603: complex modes and @code{EPmode}. For them, the unit size is the
4604: size of the real or imaginary part, or the size of the function
4605: pointer or the context pointer.
4606: @end table
4607:
4608: @node Constants, Regs and Memory, Machine Modes, RTL
4609: @section Constant Expression Types
4610:
4611: The simplest RTL expressions are those that represent constant values.
4612:
4613: @table @code
4614: @item (const_int @var{i})
4615: This type of expression represents the integer value @var{i}. @var{i}
4616: is customarily accessed with the macro @code{INTVAL} as in
4617: @code{INTVAL (@var{exp})}, which is equivalent to @code{XINT (@var{exp}, 0)}.
4618:
4619: There is only one expression object for the integer value zero;
4620: it is the value of the variable @code{const0_rtx}. Likewise, the
4621: only expression for integer value one is found in @code{const1_rtx}.
1.1.1.8 root 4622: Any attempt to create an expression of code @code{const_int} and
1.1 root 4623: value zero or one will return @code{const0_rtx} or @code{const1_rtx}
4624: as appropriate.
4625:
4626: @item (const_double:@var{m} @var{i0} @var{i1})
1.1.1.6 root 4627: Represents a 64-bit constant of mode @var{m}. All floating point
1.1 root 4628: constants are represented in this way, and so are 64-bit @code{DImode}
4629: integer constants.
4630:
4631: The two integers @var{i0} and @var{i1} together contain the bits of
4632: the value. If the constant is floating point (either single or double
4633: precision), then they represent a @code{double}. To convert them to a
4634: @code{double}, do
4635:
4636: @example
4637: union @{ double d; int i[2];@} u;
1.1.1.8 root 4638: u.i[0] = CONST_DOUBLE_LOW(x);
4639: u.i[1] = CONST_DOUBLE_HIGH(x);
1.1 root 4640: @end example
4641:
4642: @noindent
4643: and then refer to @code{u.d}.
4644:
4645: The global variables @code{dconst0_rtx} and @code{fconst0_rtx} hold
1.1.1.8 root 4646: @code{const_double} expressions with value 0, in modes @code{DFmode}
1.1.1.7 root 4647: and @code{SFmode}, respectively. The macro @code{CONST0_RTX
1.1.1.8 root 4648: (@var{mode})} refers to a @code{const_double} expression with value 0
1.1.1.7 root 4649: in mode @var{mode}. The mode @var{mode} must be of mode class
4650: @code{MODE_FLOAT}.
1.1 root 4651:
4652: @item (symbol_ref @var{symbol})
4653: Represents the value of an assembler label for data. @var{symbol} is
4654: a string that describes the name of the assembler label. If it starts
4655: with a @samp{*}, the label is the rest of @var{symbol} not including
4656: the @samp{*}. Otherwise, the label is @var{symbol}, prefixed with
4657: @samp{_}.
4658:
4659: @item (label_ref @var{label})
4660: Represents the value of an assembler label for code. It contains one
1.1.1.8 root 4661: operand, an expression, which must be a @code{code_label} that appears
1.1 root 4662: in the instruction sequence to identify the place where the label
4663: should go.
4664:
4665: The reason for using a distinct expression type for code label
4666: references is so that jump optimization can distinguish them.
4667:
4668: @item (const @var{exp})
4669: Represents a constant that is the result of an assembly-time
4670: arithmetic computation. The operand, @var{exp}, is an expression that
1.1.1.8 root 4671: contains only constants (@code{const_int}, @code{symbol_ref} and
4672: @code{label_ref} expressions) combined with @code{plus} and
4673: @code{minus}. However, not all combinations are valid, since the
1.1 root 4674: assembler cannot do arbitrary arithmetic on relocatable symbols.
4675: @end table
4676:
4677: @node Regs and Memory, Arithmetic, Constants, RTL
4678: @section Registers and Memory
4679:
4680: Here are the RTL expression types for describing access to machine
4681: registers and to main memory.
4682:
4683: @table @code
4684: @item (reg:@var{m} @var{n})
4685: For small values of the integer @var{n} (less than
4686: @code{FIRST_PSEUDO_REGISTER}), this stands for a reference to machine
4687: register number @var{n}: a @dfn{hard register}. For larger values of
4688: @var{n}, it stands for a temporary value or @dfn{pseudo register}.
4689: The compiler's strategy is to generate code assuming an unlimited
4690: number of such pseudo registers, and later convert them into hard
4691: registers or into memory references.
4692:
4693: The symbol @code{FIRST_PSEUDO_REGISTER} is defined by the machine
4694: description, since the number of hard registers on the machine is an
4695: invariant characteristic of the machine. Note, however, that not
4696: all of the machine registers must be general registers. All the
4697: machine registers that can be used for storage of data are given
4698: hard register numbers, even those that can be used only in certain
4699: instructions or can hold only certain types of data.
4700:
4701: Each pseudo register number used in a function's RTL code is
1.1.1.8 root 4702: represented by a unique @code{reg} expression.
1.1 root 4703:
4704: @var{m} is the machine mode of the reference. It is necessary because
4705: machines can generally refer to each register in more than one mode.
4706: For example, a register may contain a full word but there may be
4707: instructions to refer to it as a half word or as a single byte, as
4708: well as instructions to refer to it as a floating point number of
4709: various precisions.
4710:
4711: Even for a register that the machine can access in only one mode,
4712: the mode must always be specified.
4713:
4714: A hard register may be accessed in various modes throughout one
4715: function, but each pseudo register is given a natural mode
4716: and is accessed only in that mode. When it is necessary to describe
1.1.1.8 root 4717: an access to a pseudo register using a nonnatural mode, a @code{subreg}
1.1 root 4718: expression is used.
4719:
1.1.1.8 root 4720: A @code{reg} expression with a machine mode that specifies more than
1.1 root 4721: one word of data may actually stand for several consecutive registers.
4722: If in addition the register number specifies a hardware register, then
4723: it actually represents several consecutive hardware registers starting
4724: with the specified one.
4725:
1.1.1.8 root 4726: Such multi-word hardware register @code{reg} expressions must not be live
1.1 root 4727: across the boundary of a basic block. The lifetime analysis pass does not
4728: know how to record properly that several consecutive registers are
4729: actually live there, and therefore register allocation would be confused.
4730: The CSE pass must go out of its way to make sure the situation does
4731: not arise.
4732:
4733: @item (subreg:@var{m} @var{reg} @var{wordnum})
1.1.1.8 root 4734: @code{subreg} expressions are used to refer to a register in a machine
1.1 root 4735: mode other than its natural one, or to refer to one register of
1.1.1.8 root 4736: a multi-word @code{reg} that actually refers to several registers.
1.1 root 4737:
4738: Each pseudo-register has a natural mode. If it is necessary to
4739: operate on it in a different mode---for example, to perform a fullword
1.1.1.8 root 4740: move instruction on a pseudo-register that contains a single
4741: byte---the pseudo-register must be enclosed in a @code{subreg}. In
4742: such a case, @var{wordnum} is zero.
1.1 root 4743:
1.1.1.8 root 4744: The other use of @code{subreg} is to extract the individual registers
1.1 root 4745: of a multi-register value. Machine modes such as @code{DImode} and
4746: @code{EPmode} indicate values longer than a word, values which usually
4747: require two consecutive registers. To access one of the registers,
1.1.1.8 root 4748: use a @code{subreg} with mode @code{SImode} and a @var{wordnum} that
1.1 root 4749: says which register.
4750:
4751: The compilation parameter @code{WORDS_BIG_ENDIAN}, if defined, says
4752: that word number zero is the most significant part; otherwise, it is
4753: the least significant part.
4754:
4755: Between the combiner pass and the reload pass, it is possible to have
1.1.1.8 root 4756: a @code{subreg} which contains a @code{mem} instead of a @code{reg} as
1.1 root 4757: its first operand. The reload pass eliminates these cases by
1.1.1.8 root 4758: reloading the @code{mem} into a suitable register.
1.1 root 4759:
4760: Note that it is not valid to access a @code{DFmode} value in @code{SFmode}
1.1.1.8 root 4761: using a @code{subreg}. On some machines the most significant part of a
1.1 root 4762: @code{DFmode} value does not have the same format as a single-precision
4763: floating value.
4764:
4765: @item (cc0)
4766: This refers to the machine's condition code register. It has no
4767: operands and may not have a machine mode. It may be validly used in
4768: only two contexts: as the destination of an assignment (in test and
4769: compare instructions) and in comparison operators comparing against
1.1.1.8 root 4770: zero (@code{const_int} with value zero; that is to say,
1.1 root 4771: @code{const0_rtx}).
4772:
1.1.1.8 root 4773: There is only one expression object of code @code{cc0}; it is the
1.1 root 4774: value of the variable @code{cc0_rtx}. Any attempt to create an
1.1.1.8 root 4775: expression of code @code{cc0} will return @code{cc0_rtx}.
1.1 root 4776:
4777: One special thing about the condition code register is that
4778: instructions can set it implicitly. On many machines, nearly all
4779: instructions set the condition code based on the value that they
4780: compute or store. It is not necessary to record these actions
4781: explicitly in the RTL because the machine description includes a
4782: prescription for recognizing the instructions that do so (by means of
4783: the macro @code{NOTICE_UPDATE_CC}). Only instructions whose sole
4784: purpose is to set the condition code, and instructions that use the
4785: condition code, need mention @code{(cc0)}.
4786:
4787: @item (pc)
4788: This represents the machine's program counter. It has no operands and
4789: may not have a machine mode. @code{(pc)} may be validly used only in
4790: certain specific contexts in jump instructions.
4791:
1.1.1.8 root 4792: There is only one expression object of code @code{pc}; it is the value
1.1 root 4793: of the variable @code{pc_rtx}. Any attempt to create an expression of
1.1.1.8 root 4794: code @code{pc} will return @code{pc_rtx}.
1.1 root 4795:
4796: All instructions that do not jump alter the program counter implicitly
4797: by incrementing it, but there is no need to mention this in the RTL.
4798:
4799: @item (mem:@var{m} @var{addr})
4800: This RTX represents a reference to main memory at an address
4801: represented by the expression @var{addr}. @var{m} specifies how large
4802: a unit of memory is accessed.
4803: @end table
4804:
4805: @node Arithmetic, Comparisons, Regs and Memory, RTL
4806: @section RTL Expressions for Arithmetic
4807:
4808: @table @code
4809: @item (plus:@var{m} @var{x} @var{y})
4810: Represents the sum of the values represented by @var{x} and @var{y}
4811: carried out in machine mode @var{m}. This is valid only if
4812: @var{x} and @var{y} both are valid for mode @var{m}.
4813:
4814: @item (minus:@var{m} @var{x} @var{y})
1.1.1.8 root 4815: Like @code{plus} but represents subtraction.
1.1 root 4816:
1.1.1.6 root 4817: @item (compare @var{x} @var{y})
1.1 root 4818: Represents the result of subtracting @var{y} from @var{x}
4819: for purposes of comparison. The absence of a machine mode
1.1.1.8 root 4820: in the @code{compare} expression indicates that the result is
1.1 root 4821: computed without overflow, as if with infinite precision.
4822:
4823: Of course, machines can't really subtract with infinite precision.
4824: However, they can pretend to do so when only the sign of the
4825: result will be used, which is the case when the result is stored
4826: in @code{(cc0)}. And that is the only way this kind of expression
4827: may validly be used: as a value to be stored in the condition codes.
4828:
4829: @item (neg:@var{m} @var{x})
4830: Represents the negation (subtraction from zero) of the value
4831: represented by @var{x}, carried out in mode @var{m}. @var{x} must be
4832: valid for mode @var{m}.
4833:
4834: @item (mult:@var{m} @var{x} @var{y})
4835: Represents the signed product of the values represented by @var{x} and
4836: @var{y} carried out in machine mode @var{m}. If
4837: @var{x} and @var{y} are both valid for mode @var{m}, this is ordinary
4838: size-preserving multiplication. Alternatively, both @var{x} and @var{y}
4839: may be valid for a different, narrower mode. This represents the
4840: kind of multiplication that generates a product wider than the operands.
4841: Widening multiplication and same-size multiplication are completely
4842: distinct and supported by different machine instructions; machines may
4843: support one but not the other.@refill
4844:
1.1.1.8 root 4845: @code{mult} may be used for floating point multiplication as well.
1.1 root 4846: Then @var{m} is a floating point machine mode.
4847:
4848: @item (umult:@var{m} @var{x} @var{y})
1.1.1.8 root 4849: Like @code{mult} but represents unsigned multiplication. It may be
4850: used in both same-size and widening forms, like @code{mult}.
4851: @code{umult} is used only for fixed-point multiplication.
1.1 root 4852:
4853: @item (div:@var{m} @var{x} @var{y})
4854: Represents the quotient in signed division of @var{x} by @var{y},
4855: carried out in machine mode @var{m}. If @var{m} is a floating-point
4856: mode, it represents the exact quotient; otherwise, the integerized
4857: quotient. If @var{x} and @var{y} are both valid for mode @var{m},
4858: this is ordinary size-preserving division. Some machines have
4859: division instructions in which the operands and quotient widths are
1.1.1.8 root 4860: not all the same; such instructions are represented by @code{div}
1.1 root 4861: expressions in which the machine modes are not all the same.
4862:
4863: @item (udiv:@var{m} @var{x} @var{y})
1.1.1.8 root 4864: Like @code{div} but represents unsigned division.
1.1 root 4865:
4866: @item (mod:@var{m} @var{x} @var{y})
4867: @itemx (umod:@var{m} @var{x} @var{y})
1.1.1.8 root 4868: Like @code{div} and @code{udiv} but represent the remainder instead of
1.1 root 4869: the quotient.
4870:
4871: @item (not:@var{m} @var{x})
4872: Represents the bitwise complement of the value represented by @var{x},
4873: carried out in mode @var{m}, which must be a fixed-point machine mode.
4874: @var{x} must be valid for mode @var{m}, which must be a fixed-point mode.
4875:
4876: @item (and:@var{m} @var{x} @var{y})
4877: Represents the bitwise logical-and of the values represented by
4878: @var{x} and @var{y}, carried out in machine mode @var{m}. This is
4879: valid only if @var{x} and @var{y} both are valid for mode @var{m},
4880: which must be a fixed-point mode.
4881:
4882: @item (ior:@var{m} @var{x} @var{y})
4883: Represents the bitwise inclusive-or of the values represented by
4884: @var{x} and @var{y}, carried out in machine mode @var{m}. This is
4885: valid only if @var{x} and @var{y} both are valid for mode @var{m},
4886: which must be a fixed-point mode.
4887:
4888: @item (xor:@var{m} @var{x} @var{y})
4889: Represents the bitwise exclusive-or of the values represented by
4890: @var{x} and @var{y}, carried out in machine mode @var{m}. This is
4891: valid only if @var{x} and @var{y} both are valid for mode @var{m},
4892: which must be a fixed-point mode.
4893:
4894: @item (lshift:@var{m} @var{x} @var{c})
4895: Represents the result of logically shifting @var{x} left by @var{c}
4896: places. @var{x} must be valid for the mode @var{m}, a fixed-point
4897: machine mode. @var{c} must be valid for a fixed-point mode;
4898: which mode is determined by the mode called for in the machine
4899: description entry for the left-shift instruction. For example,
4900: on the Vax, the mode of @var{c} is @code{QImode} regardless of @var{m}.
4901:
4902: On some machines, negative values of @var{c} may be meaningful; this
4903: is why logical left shift and arithmetic left shift are distinguished.
4904: For example, Vaxes have no right-shift instructions, and right shifts
4905: are represented as left-shift instructions whose counts happen
4906: to be negative constants or else computed (in a previous instruction)
4907: by negation.
4908:
4909: @item (ashift:@var{m} @var{x} @var{c})
1.1.1.8 root 4910: Like @code{lshift} but for arithmetic left shift.
1.1 root 4911:
4912: @item (lshiftrt:@var{m} @var{x} @var{c})
4913: @itemx (ashiftrt:@var{m} @var{x} @var{c})
1.1.1.8 root 4914: Like @code{lshift} and @code{ashift} but for right shift.
1.1 root 4915:
4916: @item (rotate:@var{m} @var{x} @var{c})
4917: @itemx (rotatert:@var{m} @var{x} @var{c})
4918: Similar but represent left and right rotate.
4919:
4920: @item (abs:@var{m} @var{x})
4921: Represents the absolute value of @var{x}, computed in mode @var{m}.
4922: @var{x} must be valid for @var{m}.
4923:
4924: @item (sqrt:@var{m} @var{x})
4925: Represents the square root of @var{x}, computed in mode @var{m}.
4926: @var{x} must be valid for @var{m}. Most often @var{m} will be
4927: a floating point mode.
4928:
4929: @item (ffs:@var{m} @var{x})
4930: Represents the one plus the index of the least significant 1-bit in
4931: @var{x}, represented as an integer of mode @var{m}. (The value is
4932: zero if @var{x} is zero.) The mode of @var{x} need not be @var{m};
4933: depending on the target machine, various mode combinations may be
4934: valid.
4935: @end table
4936:
4937: @node Comparisons, Bit Fields, Arithmetic, RTL
4938: @section Comparison Operations
4939:
4940: Comparison operators test a relation on two operands and are considered to
4941: represent the value 1 if the relation holds, or zero if it does not. The
4942: mode of the comparison is determined by the operands; they must both be
4943: valid for a common machine mode. A comparison with both operands constant
4944: would be invalid as the machine mode could not be deduced from it, but such
4945: a comparison should never exist in RTL due to constant folding.
4946:
4947: Inequality comparisons come in two flavors, signed and unsigned. Thus,
1.1.1.8 root 4948: there are distinct expression codes @code{gt} and @code{gtu} for signed and
1.1 root 4949: unsigned greater-than. These can produce different results for the same
4950: pair of integer values: for example, 1 is signed greater-than -1 but not
4951: unsigned greater-than, because -1 when regarded as unsigned is actually
4952: @code{0xffffffff} which is greater than 1.
4953:
4954: The signed comparisons are also used for floating point values. Floating
4955: point comparisons are distinguished by the machine modes of the operands.
4956:
4957: The comparison operators may be used to compare the condition codes
4958: @code{(cc0)} against zero, as in @code{(eq (cc0) (const_int 0))}. Such a
4959: construct actually refers to the result of the preceding instruction in
4960: which the condition codes were set. The above example stands for 1 if the
4961: condition codes were set to say ``zero'' or ``equal'', 0 otherwise.
4962: Although the same comparison operators are used for this as may be used in
4963: other contexts on actual data, no confusion can result since the machine
4964: description would never allow both kinds of uses in the same context.
4965:
4966: @table @code
4967: @item (eq @var{x} @var{y})
4968: 1 if the values represented by @var{x} and @var{y} are equal,
4969: otherwise 0.
4970:
4971: @item (ne @var{x} @var{y})
4972: 1 if the values represented by @var{x} and @var{y} are not equal,
4973: otherwise 0.
4974:
4975: @item (gt @var{x} @var{y})
4976: 1 if the @var{x} is greater than @var{y}. If they are fixed-point,
4977: the comparison is done in a signed sense.
4978:
4979: @item (gtu @var{x} @var{y})
1.1.1.8 root 4980: Like @code{gt} but does unsigned comparison, on fixed-point numbers only.
1.1 root 4981:
4982: @item (lt @var{x} @var{y})
4983: @item (ltu @var{x} @var{y})
1.1.1.8 root 4984: Like @code{gt} and @code{gtu} but test for ``less than''.
1.1 root 4985:
4986: @item (ge @var{x} @var{y})
4987: @item (geu @var{x} @var{y})
1.1.1.8 root 4988: Like @code{gt} and @code{gtu} but test for ``greater than or equal''.
1.1 root 4989:
4990: @item (le @var{x} @var{y})
4991: @item (leu @var{x} @var{y})
1.1.1.8 root 4992: Like @code{gt} and @code{gtu} but test for ``less than or equal''.
1.1 root 4993:
4994: @item (if_then_else @var{cond} @var{then} @var{else})
4995: This is not a comparison operation but is listed here because it is
4996: always used in conjunction with a comparison operation. To be
4997: precise, @var{cond} is a comparison expression. This expression
4998: represents a choice, according to @var{cond}, between the value
4999: represented by @var{then} and the one represented by @var{else}.
5000:
1.1.1.8 root 5001: On most machines, @code{if_then_else} expressions are valid only
1.1 root 5002: to express conditional jumps.
5003: @end table
5004:
5005: @node Bit Fields, Conversions, Comparisons, RTL
5006: @section Bit-fields
5007:
5008: Special expression codes exist to represent bit-field instructions.
5009: These types of expressions are lvalues in RTL; they may appear
5010: on the left side of a assignment, indicating insertion of a value
5011: into the specified bit field.
5012:
5013: @table @code
5014: @item (sign_extract:SI @var{loc} @var{size} @var{pos})
5015: This represents a reference to a sign-extended bit-field contained or
5016: starting in @var{loc} (a memory or register reference). The bit field
5017: is @var{size} bits wide and starts at bit @var{pos}. The compilation
5018: option @code{BITS_BIG_ENDIAN} says which end of the memory unit
5019: @var{pos} counts from.
5020:
5021: Which machine modes are valid for @var{loc} depends on the machine,
5022: but typically @var{loc} should be a single byte when in memory
5023: or a full word in a register.
5024:
5025: @item (zero_extract:SI @var{loc} @var{size} @var{pos})
1.1.1.8 root 5026: Like @code{sign_extract} but refers to an unsigned or zero-extended
1.1 root 5027: bit field. The same sequence of bits are extracted, but they
5028: are filled to an entire word with zeros instead of by sign-extension.
5029: @end table
5030:
5031: @node Conversions, RTL Declarations, Bit Fields, RTL
5032: @section Conversions
5033:
5034: All conversions between machine modes must be represented by
5035: explicit conversion operations. For example, an expression
5036: which is the sum of a byte and a full word cannot be written as
1.1.1.8 root 5037: @code{(plus:SI (reg:QI 34) (reg:SI 80))} because the @code{plus}
1.1 root 5038: operation requires two operands of the same machine mode.
5039: Therefore, the byte-sized operand is enclosed in a conversion
5040: operation, as in
5041:
5042: @example
5043: (plus:SI (sign_extend:SI (reg:QI 34)) (reg:SI 80))
5044: @end example
5045:
5046: The conversion operation is not a mere placeholder, because there
5047: may be more than one way of converting from a given starting mode
5048: to the desired final mode. The conversion operation code says how
5049: to do it.
5050:
5051: @table @code
5052: @item (sign_extend:@var{m} @var{x})
5053: Represents the result of sign-extending the value @var{x}
5054: to machine mode @var{m}. @var{m} must be a fixed-point mode
5055: and @var{x} a fixed-point value of a mode narrower than @var{m}.
5056:
5057: @item (zero_extend:@var{m} @var{x})
5058: Represents the result of zero-extending the value @var{x}
5059: to machine mode @var{m}. @var{m} must be a fixed-point mode
5060: and @var{x} a fixed-point value of a mode narrower than @var{m}.
5061:
5062: @item (float_extend:@var{m} @var{x})
5063: Represents the result of extending the value @var{x}
5064: to machine mode @var{m}. @var{m} must be a floating point mode
5065: and @var{x} a floating point value of a mode narrower than @var{m}.
5066:
5067: @item (truncate:@var{m} @var{x})
5068: Represents the result of truncating the value @var{x}
5069: to machine mode @var{m}. @var{m} must be a fixed-point mode
5070: and @var{x} a fixed-point value of a mode wider than @var{m}.
5071:
5072: @item (float_truncate:@var{m} @var{x})
5073: Represents the result of truncating the value @var{x}
5074: to machine mode @var{m}. @var{m} must be a floating point mode
5075: and @var{x} a floating point value of a mode wider than @var{m}.
5076:
5077: @item (float:@var{m} @var{x})
5078: Represents the result of converting fixed point value @var{x},
5079: regarded as signed, to floating point mode @var{m}.
5080:
5081: @item (unsigned_float:@var{m} @var{x})
5082: Represents the result of converting fixed point value @var{x},
5083: regarded as unsigned, to floating point mode @var{m}.
5084:
5085: @item (fix:@var{m} @var{x})
5086: When @var{m} is a fixed point mode, represents the result of
5087: converting floating point value @var{x} to mode @var{m}, regarded as
5088: signed. How rounding is done is not specified, so this operation may
5089: be used validly in compiling C code only for integer-valued operands.
5090:
5091: @item (unsigned_fix:@var{m} @var{x})
5092: Represents the result of converting floating point value @var{x} to
5093: fixed point mode @var{m}, regarded as unsigned. How rounding is done
5094: is not specified.
5095:
5096: @item (fix:@var{m} @var{x})
5097: When @var{m} is a floating point mode, represents the result of
5098: converting floating point value @var{x} (valid for mode @var{m}) to an
5099: integer, still represented in floating point mode @var{m}, by rounding
5100: towards zero.
5101: @end table
5102:
5103: @node RTL Declarations, Side Effects, Conversions, RTL
5104: @section Declarations
5105:
5106: Declaration expression codes do not represent arithmetic operations
5107: but rather state assertions about their operands.
5108:
5109: @table @code
5110: @item (strict_low_part (subreg:@var{m} (reg:@var{n} @var{r}) 0))
5111: This expression code is used in only one context: operand 0 of a
1.1.1.8 root 5112: @code{set} expression. In addition, the operand of this expression
5113: must be a @code{subreg} expression.
1.1 root 5114:
1.1.1.8 root 5115: The presence of @code{strict_low_part} says that the part of the
1.1 root 5116: register which is meaningful in mode @var{n}, but is not part of
5117: mode @var{m}, is not to be altered. Normally, an assignment to such
5118: a subreg is allowed to have undefined effects on the rest of the
5119: register when @var{m} is less than a word.
5120: @end table
5121:
5122: @node Side Effects, Incdec, RTL Declarations, RTL
5123: @section Side Effect Expressions
5124:
5125: The expression codes described so far represent values, not actions.
5126: But machine instructions never produce values; they are meaningful
5127: only for their side effects on the state of the machine. Special
5128: expression codes are used to represent side effects.
5129:
5130: The body of an instruction is always one of these side effect codes;
5131: the codes described above, which represent values, appear only as
5132: the operands of these.
5133:
5134: @table @code
5135: @item (set @var{lval} @var{x})
5136: Represents the action of storing the value of @var{x} into the place
5137: represented by @var{lval}. @var{lval} must be an expression
1.1.1.8 root 5138: representing a place that can be stored in: @code{reg} (or
5139: @code{subreg} or @code{strict_low_part}), @code{mem}, @code{pc} or
5140: @code{cc0}.@refill
1.1 root 5141:
1.1.1.8 root 5142: If @var{lval} is a @code{reg}, @code{subreg} or @code{mem}, it has a
1.1 root 5143: machine mode; then @var{x} must be valid for that mode.@refill
5144:
1.1.1.8 root 5145: If @var{lval} is a @code{reg} whose machine mode is less than the full
1.1 root 5146: width of the register, then it means that the part of the register
5147: specified by the machine mode is given the specified value and the
5148: rest of the register receives an undefined value. Likewise, if
1.1.1.8 root 5149: @var{lval} is a @code{subreg} whose machine mode is narrower than
1.1 root 5150: @code{SImode}, the rest of the register can be changed in an undefined way.
5151:
1.1.1.8 root 5152: If @var{lval} is a @code{strict_low_part} of a @code{subreg}, then the
1.1 root 5153: part of the register specified by the machine mode of the
1.1.1.8 root 5154: @code{subreg} is given the value @var{x} and the rest of the register
1.1 root 5155: is not changed.@refill
5156:
5157: If @var{lval} is @code{(cc0)}, it has no machine mode, and @var{x} may
5158: have any mode. This represents a ``test'' or ``compare'' instruction.@refill
5159:
5160: If @var{lval} is @code{(pc)}, we have a jump instruction, and the
5161: possibilities for @var{x} are very limited. It may be a
1.1.1.8 root 5162: @code{label_ref} expression (unconditional jump). It may be an
5163: @code{if_then_else} (conditional jump), in which case either the
1.1 root 5164: second or the third operand must be @code{(pc)} (for the case which
1.1.1.8 root 5165: does not jump) and the other of the two must be a @code{label_ref}
5166: (for the case which does jump). @var{x} may also be a @code{mem} or
5167: @code{(plus:SI (pc) @var{y})}, where @var{y} may be a @code{reg} or a
5168: @code{mem}; these unusual patterns are used to represent jumps through
1.1 root 5169: branch tables.@refill
5170:
5171: @item (return)
5172: Represents a return from the current function, on machines where this
5173: can be done with one instruction, such as Vaxes. On machines where a
5174: multi-instruction ``epilogue'' must be executed in order to return
5175: from the function, returning is done by jumping to a label which
1.1.1.8 root 5176: precedes the epilogue, and the @code{return} expression code is never
1.1 root 5177: used.
5178:
5179: @item (call @var{function} @var{nargs})
1.1.1.8 root 5180: Represents a function call. @var{function} is a @code{mem} expression
1.1 root 5181: whose address is the address of the function to be called.
5182: @var{nargs} is an expression which can be used for two purposes: on
5183: some machines it represents the number of bytes of stack argument; on
5184: others, it represents the number of argument registers.
5185:
5186: Each machine has a standard machine mode which @var{function} must
5187: have. The machine description defines macro @code{FUNCTION_MODE} to
5188: expand into the requisite mode name. The purpose of this mode is to
5189: specify what kind of addressing is allowed, on machines where the
5190: allowed kinds of addressing depend on the machine mode being
5191: addressed.
5192:
5193: @item (clobber @var{x})
5194: Represents the storing or possible storing of an unpredictable,
1.1.1.8 root 5195: undescribed value into @var{x}, which must be a @code{reg} or
5196: @code{mem} expression.
1.1 root 5197:
5198: One place this is used is in string instructions that store standard
5199: values into particular hard registers. It may not be worth the
5200: trouble to describe the values that are stored, but it is essential to
5201: inform the compiler that the registers will be altered, lest it
5202: attempt to keep data in them across the string instruction.
5203:
5204: @var{x} may also be null---a null C pointer, no expression at all.
5205: Such a @code{(clobber (null))} expression means that all memory
5206: locations must be presumed clobbered.
5207:
5208: Note that the machine description classifies certain hard registers as
5209: ``call-clobbered''. All function call instructions are assumed by
5210: default to clobber these registers, so there is no need to use
1.1.1.8 root 5211: @code{clobber} expressions to indicate this fact. Also, each function
1.1.1.6 root 5212: call is assumed to have the potential to alter any memory location,
5213: unless the function is declared @code{const}.
1.1 root 5214:
1.1.1.8 root 5215: When a @code{clobber} expression for a register appears inside a
5216: @code{parallel} with other side effects, GNU CC guarantees that the
1.1.1.4 root 5217: register is unoccupied both before and after that insn. Therefore, it
5218: is safe for the assembler code produced by the insn to use the
5219: register as a temporary. You can clobber either a specific hard
5220: register or a pseudo register; in the latter case, GNU CC will
5221: allocate a hard register that is available there for use as a
5222: temporary.
5223:
1.1.1.8 root 5224: If you clobber a pseudo register in this way, use a pseudo register
5225: which appears nowhere else---generate a new one each time. Otherwise,
5226: you may confuse CSE.
5227:
5228: There is one other known use for clobbering a pseudo register in a
5229: @code{parallel}: when one of the input operands of the insn is also
5230: clobbered by the insn. In this case, using the same pseudo register in
5231: the clobber and elsewhere in the insn produces the expected results.
5232:
1.1 root 5233: @item (use @var{x})
5234: Represents the use of the value of @var{x}. It indicates that the
5235: value in @var{x} at this point in the program is needed, even though
5236: it may not be apparent why this is so. Therefore, the compiler will
1.1.1.4 root 5237: not attempt to delete previous instructions whose only effect is to
1.1.1.8 root 5238: store a value in @var{x}. @var{x} must be a @code{reg} expression.
1.1 root 5239:
5240: @item (parallel [@var{x0} @var{x1} @dots{}])
5241: Represents several side effects performed in parallel. The square
1.1.1.8 root 5242: brackets stand for a vector; the operand of @code{parallel} is a
1.1 root 5243: vector of expressions. @var{x0}, @var{x1} and so on are individual
1.1.1.8 root 5244: side effect expressions---expressions of code @code{set}, @code{call},
5245: @code{return}, @code{clobber} or @code{use}.@refill
1.1 root 5246:
5247: ``In parallel'' means that first all the values used in the individual
5248: side-effects are computed, and second all the actual side-effects are
5249: performed. For example,
5250:
5251: @example
5252: (parallel [(set (reg:SI 1) (mem:SI (reg:SI 1)))
5253: (set (mem:SI (reg:SI 1)) (reg:SI 1))])
5254: @end example
5255:
5256: @noindent
5257: says unambiguously that the values of hard register 1 and the memory
5258: location addressed by it are interchanged. In both places where
5259: @code{(reg:SI 1)} appears as a memory address it refers to the value
1.1.1.4 root 5260: in register 1 @emph{before} the execution of the insn.
5261:
1.1.1.8 root 5262: It follows that it is @emph{incorrect} to use @code{parallel} and
5263: expect the result of one @code{set} to be available for the next one.
1.1.1.4 root 5264: For example, people sometimes attempt to represent a jump-if-zero
5265: instruction this way:
5266:
5267: @example
5268: (parallel [(set (cc0) (reg:SI 34))
1.1.1.9 ! root 5269: (set (pc) (if_then_else
! 5270: (eq (cc0) (const_int 0))
! 5271: (label_ref @dots{})
! 5272: (pc)))])
1.1.1.4 root 5273: @end example
5274:
5275: @noindent
5276: But this is incorrect, because it says that the jump condition depends
5277: on the condition code value @emph{before} this instruction, not on the
5278: new value that is set by this instruction.
1.1 root 5279:
1.1.1.5 root 5280: Peephole optimization, which takes place in together with final assembly
1.1.1.8 root 5281: code output, can produce insns whose patterns consist of a @code{parallel}
1.1 root 5282: whose elements are the operands needed to output the resulting
1.1.1.8 root 5283: assembler code---often @code{reg}, @code{mem} or constant expressions.
1.1 root 5284: This would not be well-formed RTL at any other stage in compilation,
5285: but it is ok then because no further optimization remains to be done.
1.1.1.4 root 5286: However, the definition of the macro @code{NOTICE_UPDATE_CC} must
5287: deal with such insns if you define any peephole optimizations.
1.1 root 5288:
5289: @item (sequence [@var{insns} @dots{}])
5290: Represents a sequence of insns. Each of the @var{insns} that appears
5291: in the vector is suitable for appearing in the chain of insns, so it
1.1.1.8 root 5292: must be an @code{insn}, @code{jump_insn}, @code{call_insn},
5293: @code{code_label}, @code{barrier} or @code{note}.
1.1 root 5294:
1.1.1.8 root 5295: A @code{sequence} RTX never appears in an actual insn. It represents
5296: the sequence of insns that result from a @code{define_expand}
1.1 root 5297: @emph{before} those insns are passed to @code{emit_insn} to insert
5298: them in the chain of insns. When actually inserted, the individual
1.1.1.8 root 5299: sub-insns are separated out and the @code{sequence} is forgotten.
1.1 root 5300: @end table
5301:
5302: Three expression codes appear in place of a side effect, as the body of an
5303: insn, though strictly speaking they do not describe side effects as such:
5304:
5305: @table @code
5306: @item (asm_input @var{s})
5307: Represents literal assembler code as described by the string @var{s}.
5308:
5309: @item (addr_vec:@var{m} [@var{lr0} @var{lr1} @dots{}])
5310: Represents a table of jump addresses. The vector elements @var{lr0},
1.1.1.8 root 5311: etc., are @code{label_ref} expressions. The mode @var{m} specifies
1.1 root 5312: how much space is given to each address; normally @var{m} would be
5313: @code{Pmode}.
5314:
5315: @item (addr_diff_vec:@var{m} @var{base} [@var{lr0} @var{lr1} @dots{}])
5316: Represents a table of jump addresses expressed as offsets from
1.1.1.8 root 5317: @var{base}. The vector elements @var{lr0}, etc., are @code{label_ref}
1.1 root 5318: expressions and so is @var{base}. The mode @var{m} specifies how much
5319: space is given to each address-difference.@refill
5320: @end table
5321:
5322: @node Incdec, Assembler, Side Effects, RTL
5323: @section Embedded Side-Effects on Addresses
5324:
5325: Four special side-effect expression codes appear as memory addresses.
5326:
5327: @table @code
5328: @item (pre_dec:@var{m} @var{x})
5329: Represents the side effect of decrementing @var{x} by a standard
5330: amount and represents also the value that @var{x} has after being
1.1.1.8 root 5331: decremented. @var{x} must be a @code{reg} or @code{mem}, but most
5332: machines allow only a @code{reg}. @var{m} must be the machine mode
1.1 root 5333: for pointers on the machine in use. The amount @var{x} is decremented
5334: by is the length in bytes of the machine mode of the containing memory
5335: reference of which this expression serves as the address. Here is an
5336: example of its use:@refill
5337:
5338: @example
5339: (mem:DF (pre_dec:SI (reg:SI 39)))
5340: @end example
5341:
5342: @noindent
5343: This says to decrement pseudo register 39 by the length of a @code{DFmode}
5344: value and use the result to address a @code{DFmode} value.
5345:
5346: @item (pre_inc:@var{m} @var{x})
5347: Similar, but specifies incrementing @var{x} instead of decrementing it.
5348:
5349: @item (post_dec:@var{m} @var{x})
1.1.1.8 root 5350: Represents the same side effect as @code{pre_dec} but a different
1.1 root 5351: value. The value represented here is the value @var{x} has @i{before}
5352: being decremented.
5353:
5354: @item (post_inc:@var{m} @var{x})
5355: Similar, but specifies incrementing @var{x} instead of decrementing it.
5356: @end table
5357:
5358: These embedded side effect expressions must be used with care. Instruction
5359: patterns may not use them. Until the @samp{flow} pass of the compiler,
5360: they may occur only to represent pushes onto the stack. The @samp{flow}
5361: pass finds cases where registers are incremented or decremented in one
5362: instruction and used as an address shortly before or after; these cases are
5363: then transformed to use pre- or post-increment or -decrement.
5364:
5365: Explicit popping of the stack could be represented with these embedded
5366: side effect operators, but that would not be safe; the instruction
5367: combination pass could move the popping past pushes, thus changing
5368: the meaning of the code.
5369:
5370: An instruction that can be represented with an embedded side effect
1.1.1.8 root 5371: could also be represented using @code{parallel} containing an additional
5372: @code{set} to describe how the address register is altered. This is not
1.1 root 5373: done because machines that allow these operations at all typically
5374: allow them wherever a memory address is called for. Describing them as
5375: additional parallel stores would require doubling the number of entries
5376: in the machine description.
5377:
5378: @node Assembler, Insns, IncDec, RTL
5379: @section Assembler Instructions as Expressions
5380:
1.1.1.8 root 5381: The RTX code @code{asm_operands} represents a value produced by a
1.1 root 5382: user-specified assembler instruction. It is used to represent
5383: an @code{asm} statement with arguments. An @code{asm} statement with
5384: a single output operand, like this:
5385:
5386: @example
1.1.1.6 root 5387: asm ("foo %1,%2,%0" : "=a" (outputvar) : "g" (x + y), "di" (*z));
1.1 root 5388: @end example
5389:
5390: @noindent
1.1.1.8 root 5391: is represented using a single @code{asm_operands} RTX which represents
1.1 root 5392: the value that is stored in @code{outputvar}:
5393:
5394: @example
5395: (set @var{rtx-for-outputvar}
5396: (asm_operands "foo %1,%2,%0" "a" 0
5397: [@var{rtx-for-addition-result} @var{rtx-for-*z}]
5398: [(asm_input:@var{m1} "g")
5399: (asm_input:@var{m2} "di")]))
5400: @end example
5401:
5402: @noindent
1.1.1.8 root 5403: Here the operands of the @code{asm_operands} RTX are the assembler
1.1 root 5404: template string, the output-operand's constraint, the index-number of the
5405: output operand among the output operands specified, a vector of input
5406: operand RTX's, and a vector of input-operand modes and constraints. The
5407: mode @var{m1} is the mode of the sum @code{x+y}; @var{m2} is that of
5408: @code{*z}.
5409:
5410: When an @code{asm} statement has multiple output values, its insn has
1.1.1.8 root 5411: several such @code{set} RTX's inside of a @code{parallel}. Each @code{set}
5412: contains a @code{asm_operands}; all of these share the same assembler
1.1 root 5413: template and vectors, but each contains the constraint for the respective
5414: output operand. They are also distinguished by the output-operand index
5415: number, which is 0, 1, @dots{} for successive output operands.
5416:
5417: @node Insns, Calls, Assembler, RTL
5418: @section Insns
5419:
5420: The RTL representation of the code for a function is a doubly-linked
5421: chain of objects called @dfn{insns}. Insns are expressions with
5422: special codes that are used for no other purpose. Some insns are
5423: actual instructions; others represent dispatch tables for @code{switch}
5424: statements; others represent labels to jump to or various sorts of
5425: declarative information.
5426:
5427: In addition to its own specific data, each insn must have a unique id-number
5428: that distinguishes it from all other insns in the current function, and
5429: chain pointers to the preceding and following insns. These three fields
5430: occupy the same position in every insn, independent of the expression code
5431: of the insn. They could be accessed with @code{XEXP} and @code{XINT},
5432: but instead three special macros are always used:
5433:
5434: @table @code
5435: @item INSN_UID (@var{i})
5436: Accesses the unique id of insn @var{i}.
5437:
5438: @item PREV_INSN (@var{i})
5439: Accesses the chain pointer to the insn preceding @var{i}.
5440: If @var{i} is the first insn, this is a null pointer.
5441:
5442: @item NEXT_INSN (@var{i})
5443: Accesses the chain pointer to the insn following @var{i}.
5444: If @var{i} is the last insn, this is a null pointer.
5445: @end table
5446:
5447: The @code{NEXT_INSN} and @code{PREV_INSN} pointers must always
1.1.1.6 root 5448: correspond: if @var{insn} is not the first insn,
1.1 root 5449:
5450: @example
5451: NEXT_INSN (PREV_INSN (@var{insn})) == @var{insn}
5452: @end example
5453:
5454: @noindent
5455: is always true.
5456:
5457: Every insn has one of the following six expression codes:
5458:
1.1.1.8 root 5459: @table @code
1.1 root 5460: @item insn
1.1.1.8 root 5461: The expression code @code{insn} is used for instructions that do not jump
5462: and do not do function calls. Insns with code @code{insn} have four
1.1 root 5463: additional fields beyond the three mandatory ones listed above.
5464: These four are described in a table below.
5465:
5466: @item jump_insn
1.1.1.8 root 5467: The expression code @code{jump_insn} is used for instructions that may jump
5468: (or, more generally, may contain @code{label_ref} expressions).
5469: @code{jump_insn} insns have the same extra fields as @code{insn} insns,
1.1 root 5470: accessed in the same way.
5471:
5472: @item call_insn
1.1.1.8 root 5473: The expression code @code{call_insn} is used for instructions that may do
1.1 root 5474: function calls. It is important to distinguish these instructions because
5475: they imply that certain registers and memory locations may be altered
5476: unpredictably.
5477:
1.1.1.8 root 5478: @code{call_insn} insns have the same extra fields as @code{insn} insns,
1.1 root 5479: accessed in the same way.
5480:
5481: @item code_label
1.1.1.8 root 5482: A @code{code_label} insn represents a label that a jump insn can jump to.
1.1 root 5483: It contains one special field of data in addition to the three standard ones.
5484: It is used to hold the @dfn{label number}, a number that identifies this
5485: label uniquely among all the labels in the compilation (not just in the
5486: current function). Ultimately, the label is represented in the assembler
5487: output as an assembler label @samp{L@var{n}} where @var{n} is the label number.
5488:
5489: @item barrier
5490: Barriers are placed in the instruction stream after unconditional
5491: jump instructions to indicate that the jumps are unconditional.
5492: They contain no information beyond the three standard fields.
5493:
5494: @item note
1.1.1.8 root 5495: @code{note} insns are used to represent additional debugging and
1.1 root 5496: declarative information. They contain two nonstandard fields, an
5497: integer which is accessed with the macro @code{NOTE_LINE_NUMBER} and a
5498: string accessed with @code{NOTE_SOURCE_FILE}.
5499:
5500: If @code{NOTE_LINE_NUMBER} is positive, the note represents the
5501: position of a source line and @code{NOTE_SOURCE_FILE} is the source file name
5502: that the line came from. These notes control generation of line
5503: number data in the assembler output.
5504:
5505: Otherwise, @code{NOTE_LINE_NUMBER} is not really a line number but a
5506: code with one of the following values (and @code{NOTE_SOURCE_FILE}
5507: must contain a null pointer):
5508:
5509: @table @code
5510: @item NOTE_INSN_DELETED
5511: Such a note is completely ignorable. Some passes of the compiler
5512: delete insns by altering them into notes of this kind.
5513:
5514: @item NOTE_INSN_BLOCK_BEG
5515: @itemx NOTE_INSN_BLOCK_END
5516: These types of notes indicate the position of the beginning and end
5517: of a level of scoping of variable names. They control the output
5518: of debugging information.
5519:
5520: @item NOTE_INSN_LOOP_BEG
5521: @itemx NOTE_INSN_LOOP_END
5522: These types of notes indicate the position of the beginning and end
5523: of a @code{while} or @code{for} loop. They enable the loop optimizer
5524: to find loops quickly.
1.1.1.6 root 5525: @item NOTE_INSN_FUNCTION_END
5526: Appears near the end of the function body, just before the label that
5527: @code{return} statements jump to (on machine where a single instruction
5528: does not suffice for returning). This note may be deleted by jump
5529: optimization.
5530: @item NOTE_INSN_SETJMP
5531: Appears following each call to @code{setjmp} or a related function.
1.1.1.7 root 5532:
5533: @item NOTE_INSN_LOOP_BEG
5534: Appears at the place in a loop that @code{continue} statements jump to.
1.1 root 5535: @end table
1.1.1.7 root 5536:
5537: These codes are printed symbolically when they appear in debugging dumps.
1.1 root 5538: @end table
5539:
1.1.1.6 root 5540: The machine mode of an insn is normally zero (@code{VOIDmode}), but the
5541: reload pass sets it to @code{QImode} if the insn needs reloading.
5542:
1.1.1.8 root 5543: Here is a table of the extra fields of @code{insn}, @code{jump_insn}
5544: and @code{call_insn} insns:
1.1 root 5545:
5546: @table @code
5547: @item PATTERN (@var{i})
5548: An expression for the side effect performed by this insn.
5549:
1.1.1.6 root 5550: @item INSN_CODE (@var{i})
5551: An integer that says which pattern in the machine description matches
5552: this insn, or -1 if the matching has not yet been attempted.
5553:
5554: Such matching is never attempted and this field is not used on an insn
1.1.1.8 root 5555: whose pattern consists of a single @code{use}, @code{clobber},
5556: @code{asm}, @code{addr_vec} or @code{addr_diff_vec} expression.
1.1 root 5557:
5558: @item LOG_LINKS (@var{i})
1.1.1.8 root 5559: A list (chain of @code{insn_list} expressions) of previous ``related''
1.1 root 5560: insns: insns which store into registers values that are used for the
5561: first time in this insn. (An additional constraint is that neither a
5562: jump nor a label may come between the related insns). This list is
5563: set up by the flow analysis pass; it is a null pointer until then.
5564:
1.1.1.6 root 5565: @item REG_NOTES (@var{i})
1.1.1.8 root 5566: A list (chain of @code{expr_list} expressions) giving information
1.1.1.6 root 5567: about the usage of registers in this insn. This list is set up by the
5568: flow analysis pass; it is a null pointer until then.
1.1 root 5569: @end table
5570:
1.1.1.8 root 5571: The @code{LOG_LINKS} field of an insn is a chain of @code{insn_list}
1.1 root 5572: expressions. Each of these has two operands: the first is an insn,
1.1.1.8 root 5573: and the second is another @code{insn_list} expression (the next one in
5574: the chain). The last @code{insn_list} in the chain has a null pointer
1.1 root 5575: as second operand. The significant thing about the chain is which
1.1.1.8 root 5576: insns appear in it (as first operands of @code{insn_list}
1.1 root 5577: expressions). Their order is not significant.
5578:
5579: The @code{REG_NOTES} field of an insn is a similar chain but of
1.1.1.8 root 5580: @code{expr_list} expressions instead of @code{insn_list}. There are
1.1.1.5 root 5581: several kinds of register notes, which are distinguished by the machine
1.1.1.8 root 5582: mode of the @code{expr_list}, which in a register note is really
1.1.1.5 root 5583: understood as being an @code{enum reg_note}. The first operand @var{op}
1.1.1.8 root 5584: of the @code{expr_list} is data whose meaning depends on the kind of
1.1.1.5 root 5585: note. Here are the kinds of register note:
1.1 root 5586:
5587: @table @code
5588: @item REG_DEAD
5589: The register @var{op} dies in this insn; that is to say, altering the
5590: value immediately after this insn would not affect the future behavior
5591: of the program.
5592:
5593: @item REG_INC
5594: The register @var{op} is incremented (or decremented; at this level
5595: there is no distinction) by an embedded side effect inside this insn.
1.1.1.8 root 5596: This means it appears in a @code{post_inc}, @code{pre_inc},
5597: @code{post_dec} or @code{pre_dec} RTX.
1.1 root 5598:
5599: @item REG_EQUIV
5600: The register that is set by this insn will be equal to @var{op} at run
5601: time, and could validly be replaced in all its occurrences by
5602: @var{op}. (``Validly'' here refers to the data flow of the program;
5603: simple replacement may make some insns invalid.)
5604:
5605: The value which the insn explicitly copies into the register may look
5606: different from @var{op}, but they will be equal at run time.
5607:
5608: For example, when a constant is loaded into a register that is never
5609: assigned any other value, this kind of note is used.
5610:
5611: When a parameter is copied into a pseudo-register at entry to a function,
5612: a note of this kind records that the register is equivalent to the stack
5613: slot where the parameter was passed. Although in this case the register
5614: may be set by other insns, it is still valid to replace the register
5615: by the stack slot throughout the function.
5616:
5617: @item REG_EQUAL
5618: The register that is set by this insn will be equal to @var{op} at run
5619: time at the end of this insn (but not necessarily elsewhere in the
5620: function).
5621:
5622: The RTX @var{op} is typically an arithmetic expression. For example,
5623: when a sequence of insns such as a library call is used to perform an
5624: arithmetic operation, this kind of note is attached to the insn that
5625: produces or copies the final value. It tells the CSE pass how to
5626: think of that value.
5627:
5628: @item REG_RETVAL
5629: This insn copies the value of a library call, and @var{op} is the
5630: first insn that was generated to set up the arguments for the library
5631: call.
5632:
5633: Flow analysis uses this note to delete all of a library call whose
5634: result is dead.
5635:
5636: @item REG_WAS_0
5637: The register @var{op} contained zero before this insn. You can rely
5638: on this note if it is present; its absence implies nothing.
5639:
5640: @item REG_LIBCALL
5641: This is the inverse of @code{REG_RETVAL}: it is placed on the first
5642: insn of a library call, and it points to the last one.
5643:
5644: Loop optimization uses this note to move an entire library call out
5645: of a loop when its value is constant.
5646:
5647: @item REG_NONNEG
5648: The register @var{op} is known to have nonnegative value when this
5649: insn is reached.
5650: @end table
5651:
1.1.1.8 root 5652: For convenience, the machine mode in an @code{insn_list} or
5653: @code{expr_list} is printed using these symbolic codes in debugging dumps.
1.1.1.7 root 5654:
1.1.1.8 root 5655: The only difference between the expression codes @code{insn_list} and
5656: @code{expr_list} is that the first operand of an @code{insn_list} is
1.1 root 5657: assumed to be an insn and is printed in debugging dumps as the insn's
1.1.1.8 root 5658: unique id; the first operand of an @code{expr_list} is printed in the
1.1.1.7 root 5659: ordinary way as an expression.
1.1 root 5660:
5661: @node Calls, Sharing, Insns, RTL
5662: @section RTL Representation of Function-Call Insns
5663:
1.1.1.8 root 5664: Insns that call subroutines have the RTL expression code @code{call_insn}.
1.1 root 5665: These insns must satisfy special rules, and their bodies must use a special
1.1.1.8 root 5666: RTL expression code, @code{call}.
1.1 root 5667:
1.1.1.8 root 5668: A @code{call} expression has two operands, as follows:
1.1 root 5669:
5670: @example
1.1.1.6 root 5671: (call (mem:@var{fm} @var{addr}) @var{nbytes})
1.1 root 5672: @end example
5673:
5674: @noindent
5675: Here @var{nbytes} is an operand that represents the number of bytes of
5676: argument data being passed to the subroutine, @var{fm} is a machine mode
5677: (which must equal as the definition of the @code{FUNCTION_MODE} macro in
5678: the machine description) and @var{addr} represents the address of the
5679: subroutine.
5680:
1.1.1.8 root 5681: For a subroutine that returns no value, the @code{call} RTX as shown above
1.1 root 5682: is the entire body of the insn.
5683:
5684: For a subroutine that returns a value whose mode is not @code{BLKmode},
5685: the value is returned in a hard register. If this register's number is
5686: @var{r}, then the body of the call insn looks like this:
5687:
5688: @example
5689: (set (reg:@var{m} @var{r})
1.1.1.9 ! root 5690: (call (mem:@var{fm} @var{addr}) @var{nbytes}))
1.1 root 5691: @end example
5692:
5693: @noindent
5694: This RTL expression makes it clear (to the optimizer passes) that the
5695: appropriate register receives a useful value in this insn.
5696:
5697: Immediately after RTL generation, if the value of the subroutine is
5698: actually used, this call insn is always followed closely by an insn which
5699: refers to the register @var{r}. This remains true through all the
5700: optimizer passes until cross jumping occurs.
5701:
5702: The following insn has one of two forms. Either it copies the value into a
5703: pseudo-register, like this:
5704:
5705: @example
5706: (set (reg:@var{m} @var{p}) (reg:@var{m} @var{r}))
5707: @end example
5708:
5709: @noindent
5710: or (in the case where the calling function will simply return whatever
5711: value the call produced, and no operation is needed to do this):
5712:
5713: @example
5714: (use (reg:@var{m} @var{r}))
5715: @end example
5716:
5717: @noindent
5718: Between the call insn and this following insn there may intervene only a
1.1.1.8 root 5719: stack-adjustment insn (and perhaps some @code{note} insns).
1.1 root 5720:
5721: When a subroutine returns a @code{BLKmode} value, it is handled by
5722: passing to the subroutine the address of a place to store the value.
5723: So the call insn itself does not ``return'' any value, and it has the
5724: same RTL form as a call that returns nothing.
5725:
5726: @node Sharing,, Calls, RTL
5727: @section Structure Sharing Assumptions
5728:
5729: The compiler assumes that certain kinds of RTL expressions are unique;
5730: there do not exist two distinct objects representing the same value.
5731: In other cases, it makes an opposite assumption: that no RTL expression
5732: object of a certain kind appears in more than one place in the
5733: containing structure.
5734:
5735: These assumptions refer to a single function; except for the RTL
5736: objects that describe global variables and external functions,
5737: no RTL objects are common to two functions.
5738:
5739: @itemize @bullet
5740: @item
1.1.1.8 root 5741: Each pseudo-register has only a single @code{reg} object to represent it,
1.1 root 5742: and therefore only a single machine mode.
5743:
5744: @item
1.1.1.8 root 5745: For any symbolic label, there is only one @code{symbol_ref} object
1.1 root 5746: referring to it.
5747:
5748: @item
1.1.1.8 root 5749: There is only one @code{const_int} expression with value zero,
1.1 root 5750: and only one with value one.
5751:
5752: @item
1.1.1.8 root 5753: There is only one @code{pc} expression.
1.1 root 5754:
5755: @item
1.1.1.8 root 5756: There is only one @code{cc0} expression.
1.1 root 5757:
5758: @item
1.1.1.8 root 5759: There is only one @code{const_double} expression with mode
1.1 root 5760: @code{SFmode} and value zero, and only one with mode @code{DFmode} and
5761: value zero.
5762:
5763: @item
1.1.1.8 root 5764: No @code{label_ref} appears in more than one place in the RTL
1.1 root 5765: structure; in other words, it is safe to do a tree-walk of all the
1.1.1.8 root 5766: insns in the function and assume that each time a @code{label_ref} is
1.1 root 5767: seen it is distinct from all others that are seen.
5768:
5769: @item
1.1.1.8 root 5770: Only one @code{mem} object is normally created for each static
1.1 root 5771: variable or stack slot, so these objects are frequently shared in all
5772: the places they appear. However, separate but equal objects for these
5773: variables are occasionally made.
5774:
5775: @item
1.1.1.5 root 5776: When a single @code{asm} statement has multiple output operands,
5777: a distinct @code{asm_operands} RTX is made for each output operand.
5778: However, these all share the vector which contains the sequence of
5779: input operands. Because this sharing is used later on to test whether
5780: two @code{asm_operands} RTX's come from the same statement, the sharing
5781: must be guaranteed to be preserved.
5782:
5783: @item
1.1 root 5784: No RTL object appears in more than one place in the RTL structure
5785: except as described above. Many passes of the compiler rely on this
5786: by assuming that they can modify RTL objects in place without unwanted
5787: side-effects on other insns.
5788:
5789: @item
5790: During initial RTL generation, shared structure is freely introduced.
5791: After all the RTL for a function has been generated, all shared
5792: structure is copied by @code{unshare_all_rtl} in @file{emit-rtl.c},
5793: after which the above rules are guaranteed to be followed.
5794:
5795: @item
5796: During the combiner pass, shared structure with an insn can exist
5797: temporarily. However, the shared structure is copied before the
5798: combiner is finished with the insn. This is done by
1.1.1.8 root 5799: @code{copy_substitutions} in @file{combine.c}.
1.1 root 5800: @end itemize
5801:
5802: @node Machine Desc, Machine Macros, RTL, Top
5803: @chapter Machine Descriptions
5804:
5805: A machine description has two parts: a file of instruction patterns
5806: (@file{.md} file) and a C header file of macro definitions.
5807:
5808: The @file{.md} file for a target machine contains a pattern for each
5809: instruction that the target machine supports (or at least each instruction
5810: that is worth telling the compiler about). It may also contain comments.
5811: A semicolon causes the rest of the line to be a comment, unless the semicolon
5812: is inside a quoted string.
5813:
5814: See the next chapter for information on the C header file.
5815:
5816: @menu
5817: * Patterns:: How to write instruction patterns.
1.1.1.8 root 5818: * Example:: An explained example of a @code{define_insn} pattern.
1.1 root 5819: * RTL Template:: The RTL template defines what insns match a pattern.
5820: * Output Template:: The output template says how to make assembler code
5821: from such an insn.
5822: * Output Statement:: For more generality, write C code to output
5823: the assembler code.
5824: * Constraints:: When not all operands are general operands.
5825: * Standard Names:: Names mark patterns to use for code generation.
5826: * Pattern Ordering:: When the order of patterns makes a difference.
5827: * Dependent Patterns:: Having one pattern may make you need another.
5828: * Jump Patterns:: Special considerations for patterns for jump insns.
5829: * Peephole Definitions::Defining machine-specific peephole optimizations.
5830: * Expander Definitions::Generating a sequence of several RTL insns
5831: for a standard operation.
5832: @end menu
5833:
5834: @node Patterns, Example, Machine Desc, Machine Desc
5835: @section Everything about Instruction Patterns
5836:
5837: Each instruction pattern contains an incomplete RTL expression, with pieces
5838: to be filled in later, operand constraints that restrict how the pieces can
5839: be filled in, and an output pattern or C code to generate the assembler
1.1.1.8 root 5840: output, all wrapped up in a @code{define_insn} expression.
1.1 root 5841:
1.1.1.8 root 5842: A @code{define_insn} is an RTL expression containing four or five operands:
1.1 root 5843:
5844: @enumerate
5845: @item
5846: An optional name. The presence of a name indicate that this instruction
5847: pattern can perform a certain standard job for the RTL-generation
5848: pass of the compiler. This pass knows certain names and will use
5849: the instruction patterns with those names, if the names are defined
5850: in the machine description.
5851:
5852: The absence of a name is indicated by writing an empty string
5853: where the name should go. Nameless instruction patterns are never
5854: used for generating RTL code, but they may permit several simpler insns
5855: to be combined later on.
5856:
5857: Names that are not thus known and used in RTL-generation have no
5858: effect; they are equivalent to no name at all.
5859:
5860: @item
5861: The @dfn{RTL template} (@pxref{RTL Template}) is a vector of
5862: incomplete RTL expressions which show what the instruction should look
1.1.1.8 root 5863: like. It is incomplete because it may contain @code{match_operand}
5864: and @code{match_dup} expressions that stand for operands of the
1.1 root 5865: instruction.
5866:
5867: If the vector has only one element, that element is what the
5868: instruction should look like. If the vector has multiple elements,
1.1.1.8 root 5869: then the instruction looks like a @code{parallel} expression
1.1 root 5870: containing that many elements as described.
5871:
5872: @item
5873: A condition. This is a string which contains a C expression that is
5874: the final test to decide whether an insn body matches this pattern.
5875:
5876: For a named pattern, the condition (if present) may not depend on
5877: the data in the insn being matched, but only the target-machine-type
5878: flags. The compiler needs to test these conditions during
5879: initialization in order to learn exactly which named instructions are
5880: available in a particular run.
5881:
5882: For nameless patterns, the condition is applied only when matching an
5883: individual insn, and only after the insn has matched the pattern's
5884: recognition template. The insn's operands may be found in the vector
5885: @code{operands}.
5886:
5887: @item
5888: The @dfn{output template}: a string that says how to output matching
5889: insns as assembler code. @samp{%} in this string specifies where
5890: to substitute the value of an operand. @xref{Output Template}.
5891:
5892: When simple substitution isn't general enough, you can specify a piece
5893: of C code to compute the output. @xref{Output Statement}.
5894:
5895: @item
5896: Optionally, some @dfn{machine-specific information}. The meaning
5897: of this information is defined only by an individual machine description;
5898: typically it might say whether this insn alters the condition codes,
5899: or how many bytes of output it generates.
5900:
5901: This operand is written as a string containing a C initializer
5902: (complete with braces) for the structure type @code{INSN_MACHINE_INFO},
5903: whose definition is up to you (@pxref{Misc}).
5904: @end enumerate
5905:
5906: @node Example, RTL Template, Patterns, Machine Desc
1.1.1.8 root 5907: @section Example of @code{define_insn}
1.1 root 5908:
5909: Here is an actual example of an instruction pattern, for the 68000/68020.
5910:
5911: @example
5912: (define_insn "tstsi"
5913: [(set (cc0)
5914: (match_operand:SI 0 "general_operand" "rm"))]
5915: ""
5916: "*
5917: @{ if (TARGET_68020 || ! ADDRESS_REG_P (operands[0]))
5918: return \"tstl %0\";
5919: return \"cmpl #0,%0\"; @}")
5920: @end example
5921:
5922: This is an instruction that sets the condition codes based on the value of
5923: a general operand. It has no condition, so any insn whose RTL description
5924: has the form shown may be handled according to this pattern. The name
5925: @samp{tstsi} means ``test a @code{SImode} value'' and tells the RTL generation
5926: pass that, when it is necessary to test such a value, an insn to do so
5927: can be constructed using this pattern.
5928:
5929: The output control string is a piece of C code which chooses which
5930: output template to return based on the kind of operand and the specific
5931: type of CPU for which code is being generated.
5932:
5933: @samp{"rm"} is an operand constraint. Its meaning is explained below.
5934:
5935: @node RTL Template, Output Template, Example, Machine Desc
5936: @section RTL Template for Generating and Recognizing Insns
5937:
5938: The RTL template is used to define which insns match the particular pattern
5939: and how to find their operands. For named patterns, the RTL template also
5940: says how to construct an insn from specified operands.
5941:
5942: Construction involves substituting specified operands into a copy of the
5943: template. Matching involves determining the values that serve as the
5944: operands in the insn being matched. Both of these activities are
5945: controlled by special expression types that direct matching and
5946: substitution of the operands.
5947:
5948: @table @code
1.1.1.8 root 5949: @item (match_operand:@var{m} @var{n} @var{pred} @var{constraint})
1.1 root 5950: This expression is a placeholder for operand number @var{n} of
5951: the insn. When constructing an insn, operand number @var{n}
5952: will be substituted at this point. When matching an insn, whatever
5953: appears at this position in the insn will be taken as operand
1.1.1.8 root 5954: number @var{n}; but it must satisfy @var{pred} or this instruction
1.1 root 5955: pattern will not match at all.
5956:
5957: Operand numbers must be chosen consecutively counting from zero in
1.1.1.8 root 5958: each instruction pattern. There may be only one @code{match_operand}
1.1 root 5959: expression in the pattern for each operand number. Usually operands
1.1.1.8 root 5960: are numbered in the order of appearance in @code{match_operand}
1.1 root 5961: expressions.
5962:
1.1.1.8 root 5963: @var{pred} is a string that is the name of a C function that accepts
5964: two arguments, an expression and a machine mode. During matching, the
5965: function will be called with the putative operand as the expression
5966: and @var{m} as the mode argument. If it returns zero, this
5967: instruction pattern fails to match. @var{pred} may be an empty
5968: string; then it means no test is to be done on the operand,
5969: so anything which occurs in this position is valid.
5970:
5971: @var{constraint} controls reloading and the choice of the best register
5972: class to use for a value, as explained later (@pxref{Constraints}).
5973:
5974: People are often unclear on the difference between the constraint and the
5975: predicate. The predicate helps decide whether a given insn matches the
5976: pattern. The constraint plays no role in this decision; instead, it
5977: controls various decisions in the case of an insn which does match.
5978:
5979: Most often, @var{pred} is @code{"general_operand"}. This function checks
5980: that the putative operand is either a constant, a register or a memory
5981: reference, and that it is valid for mode @var{m}.
1.1 root 5982:
1.1.1.8 root 5983: For an operand that must be a register, @var{pred} should be
1.1 root 5984: @code{"register_operand"}. It would be valid to use
5985: @code{"general_operand"}, since the reload pass would copy any
5986: non-register operands through registers, but this would make GNU CC do
5987: extra work, and it would prevent the register allocator from doing the
5988: best possible job.
5989:
1.1.1.8 root 5990: For an operand that must be a constant, either @var{pred} should be
1.1 root 5991: @code{"immediate_operand"}, or the instruction pattern's extra
5992: condition should check for constants, or both. You cannot expect the
5993: constraints to do this work! If the constraints allow only constants,
5994: but the predicate allows something else, the compiler will crash when
5995: that case arises.
5996:
5997: @item (match_dup @var{n})
5998: This expression is also a placeholder for operand number @var{n}.
5999: It is used when the operand needs to appear more than once in the
6000: insn.
6001:
1.1.1.8 root 6002: In construction, @code{match_dup} behaves exactly like
6003: @code{match_operand}: the operand is substituted into the insn being
6004: constructed. But in matching, @code{match_dup} behaves differently.
1.1 root 6005: It assumes that operand number @var{n} has already been determined by
1.1.1.8 root 6006: a @code{match_operand} appearing earlier in the recognition template,
1.1 root 6007: and it matches only an identical-looking expression.
6008:
1.1.1.4 root 6009: @item (match_operator:@var{m} @var{n} "@var{predicate}" [@var{operands}@dots{}])
6010: This pattern is a kind of placeholder for a variable RTL expression
6011: code.
6012:
6013: When constructing an insn, it stands for an RTL expression whose
6014: expression code is taken from that of operand @var{n}, and whose
6015: operands are constructed from the patterns @var{operands}.
6016:
6017: When matching an expression, it matches an expression if the function
6018: @var{predicate} returns nonzero on that expression @emph{and} the
6019: patterns @var{operands} match the operands of the expression.
6020:
6021: Suppose that the function @code{commutative_operator} is defined as
6022: follows, to match any expression whose operator is one of the six
6023: commutative arithmetic operators of RTL and whose mode is @var{mode}:
6024:
6025: @example
6026: int
6027: commutative_operator (x, mode)
6028: rtx x;
6029: enum machine_mode mode;
6030: @{
6031: enum rtx_code code = GET_CODE (x);
6032: if (GET_MODE (x) != mode)
6033: return 0;
6034: return (code == PLUS || code == MULT || code == UMULT
6035: || code == AND || code == IOR || code == XOR);
6036: @}
6037: @end example
6038:
6039: Then the following pattern will match any RTL expression consisting
6040: of a commutative operator applied to two general operands:
6041:
6042: @example
6043: (match_operator:SI 2 "commutative_operator"
6044: [(match_operand:SI 3 "general_operand" "g")
6045: (match_operand:SI 4 "general_operand" "g")])
6046: @end example
6047:
6048: Here the vector @code{[@var{operands}@dots{}]} contains two patterns
6049: because the expressions to be matched all contain two operands.
6050:
6051: When this pattern does match, the two operands of the commutative
6052: operator are recorded as operands 3 and 4 of the insn. (This is done
1.1.1.8 root 6053: by the two instances of @code{match_operand}.) Operand 2 of the insn
1.1.1.4 root 6054: will be the entire commutative expression: use @code{GET_CODE
6055: (operands[2])} to see which commutative operator was used.
6056:
1.1.1.8 root 6057: The machine mode @var{m} of @code{match_operator} works like that of
6058: @code{match_operand}: it is passed as the second argument to the
1.1.1.4 root 6059: predicate function, and that function is solely responsible for
6060: deciding whether the expression to be matched ``has'' that mode.
6061:
6062: When constructing an insn, argument 2 of the gen-function will specify
6063: the operation (i.e. the expression code) for the expression to be
6064: made. It should be an RTL expression, whose expression code is copied
6065: into a new expression whose operands are arguments 3 and 4 of the
6066: gen-function. The subexpressions of argument 2 are not used;
6067: only its expression code matters.
6068:
1.1.1.8 root 6069: There is no way to specify constraints in @code{match_operator}. The
6070: operand of the insn which corresponds to the @code{match_operator}
1.1.1.4 root 6071: never has any constraints because it is never reloaded as a whole.
6072: However, if parts of its @var{operands} are matched by
1.1.1.8 root 6073: @code{match_operand} patterns, those parts may have constraints of
1.1.1.4 root 6074: their own.
6075:
1.1 root 6076: @item (address (match_operand:@var{m} @var{n} "address_operand" ""))
6077: This complex of expressions is a placeholder for an operand number
6078: @var{n} in a ``load address'' instruction: an operand which specifies
6079: a memory location in the usual way, but for which the actual operand
6080: value used is the address of the location, not the contents of the
6081: location.
6082:
1.1.1.8 root 6083: @code{address} expressions never appear in RTL code, only in machine
1.1 root 6084: descriptions. And they are used only in machine descriptions that do
6085: not use the operand constraint feature. When operand constraints are
6086: in use, the letter @samp{p} in the constraint serves this purpose.
6087:
6088: @var{m} is the machine mode of the @emph{memory location being
6089: addressed}, not the machine mode of the address itself. That mode is
6090: always the same on a given target machine (it is @code{Pmode}, which
6091: normally is @code{SImode}), so there is no point in mentioning it;
1.1.1.8 root 6092: thus, no machine mode is written in the @code{address} expression. If
1.1 root 6093: some day support is added for machines in which addresses of different
6094: kinds of objects appear differently or are used differently (such as
6095: the PDP-10), different formats would perhaps need different machine
1.1.1.8 root 6096: modes and these modes might be written in the @code{address}
1.1 root 6097: expression.
6098: @end table
6099:
6100: @node Output Template, Output Statement, RTL Template, Machine Desc
6101: @section Output Templates and Operand Substitution
6102:
1.1.1.6 root 6103: The @dfn{output template} is a string which specifies how to output the
6104: assembler code for an instruction pattern. Most of the template is a
6105: fixed string which is output literally. The character @samp{%} is used
6106: to specify where to substitute an operand; it can also be used to
6107: identify places where different variants of the assembler require
1.1 root 6108: different syntax.
6109:
6110: In the simplest case, a @samp{%} followed by a digit @var{n} says to output
6111: operand @var{n} at that point in the string.
6112:
6113: @samp{%} followed by a letter and a digit says to output an operand in an
6114: alternate fashion. Four letters have standard, built-in meanings described
6115: below. The machine description macro @code{PRINT_OPERAND} can define
6116: additional letters with nonstandard meanings.
6117:
6118: @samp{%c@var{digit}} can be used to substitute an operand that is a
6119: constant value without the syntax that normally indicates an immediate
6120: operand.
6121:
6122: @samp{%n@var{digit}} is like @samp{%c@var{digit}} except that the value of
6123: the constant is negated before printing.
6124:
6125: @samp{%a@var{digit}} can be used to substitute an operand as if it were a
6126: memory reference, with the actual operand treated as the address. This may
6127: be useful when outputting a ``load address'' instruction, because often the
6128: assembler syntax for such an instruction requires you to write the operand
6129: as if it were a memory reference.
6130:
6131: @samp{%l@var{digit}} is used to substitute a @code{label_ref} into a jump
6132: instruction.
6133:
6134: @samp{%} followed by a punctuation character specifies a substitution that
6135: does not use an operand. Only one case is standard: @samp{%%} outputs a
6136: @samp{%} into the assembler code. Other nonstandard cases can be
1.1.1.8 root 6137: defined in the @code{PRINT_OPERAND} macro. You must also define
6138: which punctuation characters are valid with the
6139: @code{PRINT_OPERAND_PUNCT_VALID_P} macro.
1.1 root 6140:
6141: The template may generate multiple assembler instructions. Write the text
6142: for the instructions, with @samp{\;} between them.
6143:
1.1.1.6 root 6144: When the RTL contains two operands which are required by constraint to match
1.1 root 6145: each other, the output template must refer only to the lower-numbered operand.
6146: Matching operands are not always identical, and the rest of the compiler
6147: arranges to put the proper RTL expression for printing into the lower-numbered
6148: operand.
6149:
6150: One use of nonstandard letters or punctuation following @samp{%} is to
6151: distinguish between different assembler languages for the same machine; for
6152: example, Motorola syntax versus MIT syntax for the 68000. Motorola syntax
6153: requires periods in most opcode names, while MIT syntax does not. For
6154: example, the opcode @samp{movel} in MIT syntax is @samp{move.l} in Motorola
6155: syntax. The same file of patterns is used for both kinds of output syntax,
6156: but the character sequence @samp{%.} is used in each place where Motorola
6157: syntax wants a period. The @code{PRINT_OPERAND} macro for Motorola syntax
6158: defines the sequence to output a period; the macro for MIT syntax defines
6159: it to do nothing.
6160:
6161: @node Output Statement, Constraints, Output Template, Machine Desc
6162: @section C Statements for Generating Assembler Output
6163:
6164: Often a single fixed template string cannot produce correct and efficient
6165: assembler code for all the cases that are recognized by a single
6166: instruction pattern. For example, the opcodes may depend on the kinds of
6167: operands; or some unfortunate combinations of operands may require extra
6168: machine instructions.
6169:
6170: If the output control string starts with a @samp{*}, then it is not an
6171: output template but rather a piece of C program that should compute a
6172: template. It should execute a @code{return} statement to return the
6173: template-string you want. Most such templates use C string literals, which
6174: require doublequote characters to delimit them. To include these
6175: doublequote characters in the string, prefix each one with @samp{\}.
6176:
6177: The operands may be found in the array @code{operands}, whose C data type
6178: is @code{rtx []}.
6179:
6180: It is possible to output an assembler instruction and then go on to output
6181: or compute more of them, using the subroutine @code{output_asm_insn}. This
6182: receives two arguments: a template-string and a vector of operands. The
6183: vector may be @code{operands}, or it may be another array of @code{rtx}
6184: that you declare locally and initialize yourself.
6185:
6186: When an insn pattern has multiple alternatives in its constraints, often
1.1.1.5 root 6187: the appearance of the assembler code is determined mostly by which alternative
1.1 root 6188: was matched. When this is so, the C code can test the variable
6189: @code{which_alternative}, which is the ordinal number of the alternative
6190: that was actually satisfied (0 for the first, 1 for the second alternative,
6191: etc.).
6192:
6193: For example, suppose there are two opcodes for storing zero, @samp{clrreg}
6194: for registers and @samp{clrmem} for memory locations. Here is how
6195: a pattern could use @code{which_alternative} to choose between them:
6196:
6197: @example
6198: (define_insn ""
6199: [(set (match_operand:SI 0 "general_operand" "r,m")
6200: (const_int 0))]
6201: ""
6202: "*
6203: return (which_alternative == 0
6204: ? \"clrreg %0\" : \"clrmem %0\");
6205: ")
6206: @end example
6207:
6208: @node Constraints, Standard Names, Output Statement, Machine Desc
6209: @section Operand Constraints
6210:
1.1.1.8 root 6211: Each @code{match_operand} in an instruction pattern can specify a
1.1 root 6212: constraint for the type of operands allowed. Constraints can say whether
6213: an operand may be in a register, and which kinds of register; whether the
6214: operand can be a memory reference, and which kinds of address; whether the
6215: operand may be an immediate constant, and which possible values it may
6216: have. Constraints can also require two operands to match.
6217:
6218: @menu
6219: * Simple Constraints:: Basic use of constraints.
6220: * Multi-Alternative:: When an insn has two alternative constraint-patterns.
6221: * Class Preferences:: Constraints guide which hard register to put things in.
6222: * Modifiers:: More precise control over effects of constraints.
6223: * No Constraints:: Describing a clean machine without constraints.
6224: @end menu
6225:
6226: @node Simple Constraints, Multi-Alternative, Constraints, Constraints
6227: @subsection Simple Constraints
6228:
6229: The simplest kind of constraint is a string full of letters, each of
6230: which describes one kind of operand that is permitted. Here are
6231: the letters that are allowed:
6232:
6233: @table @asis
6234: @item @samp{m}
6235: A memory operand is allowed, with any kind of address that the machine
6236: supports in general.
6237:
6238: @item @samp{o}
6239: A memory operand is allowed, but only if the address is
1.1.1.8 root 6240: @dfn{offsettable}. This means that adding a small integer (actually,
1.1 root 6241: the width in bytes of the operand, as determined by its machine mode)
6242: may be added to the address and the result is also a valid memory
6243: address.
6244:
1.1.1.8 root 6245: For example, an address which is constant is offsettable; so is an
1.1 root 6246: address that is the sum of a register and a constant (as long as a
6247: slightly larger constant is also within the range of address-offsets
6248: supported by the machine); but an autoincrement or autodecrement
1.1.1.8 root 6249: address is not offsettable. More complicated indirect/indexed
6250: addresses may or may not be offsettable depending on the other
1.1 root 6251: addressing modes that the machine supports.
6252:
6253: Note that in an output operand which can be matched by another
6254: operand, the constraint letter @samp{o} is valid only when accompanied
6255: by both @samp{<} (if the target machine has predecrement addressing)
6256: and @samp{>} (if the target machine has preincrement addressing).
6257:
6258: When the constraint letter @samp{o} is used, the reload pass may
1.1.1.8 root 6259: generate instructions which copy a nonoffsettable address into an index
1.1 root 6260: register. The idea is that the register can be used as a replacement
1.1.1.8 root 6261: offsettable address. But this method requires that there be patterns
1.1 root 6262: to copy any kind of address into a register. Auto-increment
6263: and auto-decrement addresses are an exception; there need not be an
6264: instruction that can copy such an address into a register, because
6265: reload handles these cases specially.
6266:
6267: Most older machine designs have ``load address'' instructions which do
6268: just what is needed here. Some RISC machines do not advertise such
6269: instructions, but the possible addresses on these machines are very
6270: limited, so it is easy to fake them.
6271:
6272: @item @samp{<}
6273: A memory operand with autodecrement addressing (either predecrement or
6274: postdecrement) is allowed.
6275:
6276: @item @samp{>}
6277: A memory operand with autoincrement addressing (either preincrement or
6278: postincrement) is allowed.
6279:
6280: @item @samp{r}
6281: A register operand is allowed provided that it is in a general
6282: register.
6283:
6284: @item @samp{d}, @samp{a}, @samp{f}, @dots{}
6285: Other letters can be defined in machine-dependent fashion to stand for
6286: particular classes of registers. @samp{d}, @samp{a} and @samp{f} are
6287: defined on the 68000/68020 to stand for data, address and floating
6288: point registers.
6289:
6290: @item @samp{i}
6291: An immediate integer operand (one with constant value) is allowed.
6292: This includes symbolic constants whose values will be known only at
6293: assembly time.
6294:
6295: @item @samp{n}
6296: An immediate integer operand with a known numeric value is allowed.
6297: Many systems cannot support assembly-time constants for operands less
6298: than a word wide. Constraints for these operands should use @samp{n}
6299: rather than @samp{i}.
6300:
6301: @item @samp{I}, @samp{J}, @samp{K}, @dots{}
6302: Other letters in the range @samp{I} through @samp{M} may be defined in
6303: a machine-dependent fashion to permit immediate integer operands with
6304: explicit integer values in specified ranges. For example, on the
6305: 68000, @samp{I} is defined to stand for the range of values 1 to 8.
6306: This is the range permitted as a shift count in the shift
6307: instructions.
6308:
6309: @item @samp{F}
1.1.1.8 root 6310: An immediate floating operand (expression code @code{const_double}) is
1.1 root 6311: allowed.
6312:
6313: @item @samp{G}, @samp{H}
6314: @samp{G} and @samp{H} may be defined in a machine-dependent fashion to
6315: permit immediate floating operands in particular ranges of values.
6316:
6317: @item @samp{s}
6318: An immediate integer operand whose value is not an explicit integer is
6319: allowed.
6320:
6321: This might appear strange; if an insn allows a constant operand with a
6322: value not known at compile time, it certainly must allow any known
6323: value. So why use @samp{s} instead of @samp{i}? Sometimes it allows
6324: better code to be generated.
6325:
6326: For example, on the 68000 in a fullword instruction it is possible to
1.1.1.9 ! root 6327: use an immediate operand; but if the immediate value is between -128
! 6328: and 127, better code results from loading the value into a register and
1.1 root 6329: using the register. This is because the load into the register can be
6330: done with a @samp{moveq} instruction. We arrange for this to happen
6331: by defining the letter @samp{K} to mean ``any integer outside the
1.1.1.9 ! root 6332: range -128 to 127'', and then specifying @samp{Ks} in the operand
1.1 root 6333: constraints.
6334:
6335: @item @samp{g}
6336: Any register, memory or immediate integer operand is allowed, except for
6337: registers that are not general registers.
6338:
6339: @item @samp{@var{n}} (a digit)
6340: An operand that matches operand number @var{n} is allowed.
6341: If a digit is used together with letters, the digit should come last.
6342:
6343: This is called a @dfn{matching constraint} and what it really means is
6344: that the assembler has only a single operand that fills two roles
6345: considered separate in the RTL insn. For example, an add insn has two
6346: input operands and one output operand in the RTL, but on most machines
6347: an add instruction really has only two operands, one of them an
6348: input-output operand.
6349:
6350: Matching constraints work only in circumstances like that add insn.
6351: More precisely, the matching constraint must appear in an input-only
6352: operand and the operand that it matches must be an output-only operand
1.1.1.5 root 6353: with a lower number. Thus, operand @var{n} must have @samp{=} in its
6354: constraint.
1.1 root 6355:
6356: For operands to match in a particular case usually means that they
6357: are identical-looking RTL expressions. But in a few special cases
6358: specific kinds of dissimilarity are allowed. For example, @code{*x}
6359: as an input operand will match @code{*x++} as an output operand.
6360: For proper results in such cases, the output template should always
6361: use the output-operand's number when printing the operand.
6362:
6363: @item @samp{p}
6364: An operand that is a valid memory address is allowed. This is
6365: for ``load address'' and ``push address'' instructions.
6366:
1.1.1.8 root 6367: @samp{p} in the constraint must be accompanies by @code{address_operand}
6368: as the predicate in the @code{match_operand}.
1.1 root 6369: @end table
6370:
6371: In order to have valid assembler code, each operand must satisfy
6372: its constraint. But a failure to do so does not prevent the pattern
6373: from applying to an insn. Instead, it directs the compiler to modify
6374: the code so that the constraint will be satisfied. Usually this is
6375: done by copying an operand into a register.
6376:
6377: Contrast, therefore, the two instruction patterns that follow:
6378:
6379: @example
6380: (define_insn ""
6381: [(set (match_operand:SI 0 "general_operand" "r")
6382: (plus:SI (match_dup 0)
6383: (match_operand:SI 1 "general_operand" "r")))]
6384: ""
6385: "@dots{}")
6386: @end example
6387:
6388: @noindent
6389: which has two operands, one of which must appear in two places, and
6390:
6391: @example
6392: (define_insn ""
6393: [(set (match_operand:SI 0 "general_operand" "r")
6394: (plus:SI (match_operand:SI 1 "general_operand" "0")
6395: (match_operand:SI 2 "general_operand" "r")))]
6396: ""
6397: "@dots{}")
6398: @end example
6399:
6400: @noindent
6401: which has three operands, two of which are required by a constraint to be
6402: identical. If we are considering an insn of the form
6403:
6404: @example
6405: (insn @var{n} @var{prev} @var{next}
6406: (set (reg:SI 3)
6407: (plus:SI (reg:SI 6) (reg:SI 109)))
6408: @dots{})
6409: @end example
6410:
6411: @noindent
6412: the first pattern would not apply at all, because this insn does not
6413: contain two identical subexpressions in the right place. The pattern would
6414: say, ``That does not look like an add instruction; try other patterns.''
6415: The second pattern would say, ``Yes, that's an add instruction, but there
6416: is something wrong with it.'' It would direct the reload pass of the
6417: compiler to generate additional insns to make the constraint true. The
6418: results might look like this:
6419:
6420: @example
6421: (insn @var{n2} @var{prev} @var{n}
6422: (set (reg:SI 3) (reg:SI 6))
6423: @dots{})
6424:
6425: (insn @var{n} @var{n2} @var{next}
6426: (set (reg:SI 3)
6427: (plus:SI (reg:SI 3) (reg:SI 109)))
6428: @dots{})
6429: @end example
6430:
6431: It is up to you to make sure that each operand, in each pattern, has
6432: constraints that can handle any RTL expression that could be present for
6433: that operand. (When multiple alternatives are in use, each pattern must,
6434: for each possible combination of operand expressions, have at least one
6435: alternative which can handle that combination of operands.) The
6436: constraints don't need to @emph{allow} any possible operand---when this is
6437: the case, they do not constrain---but they must at least point the way to
6438: reloading any possible operand so that it will fit.
6439:
6440: @itemize @bullet
6441: @item
6442: If the constraint accepts whatever operands the predicate permits,
6443: there is no problem: reloading is never necessary for this operand.
6444:
6445: For example, an operand whose constraints permit everything except
6446: registers is safe provided its predicate rejects registers.
6447:
6448: An operand whose predicate accepts only constant values is safe
6449: provided its constraints include the letter @samp{i}. If any possible
6450: constant value is accepted, then nothing less than @samp{i} will do;
1.1.1.5 root 6451: if the predicate is more selective, then the constraints may also be
1.1 root 6452: more selective.
6453:
6454: @item
6455: Any operand expression can be reloaded by copying it into a register.
6456: So if an operand's constraints allow some kind of register, it is
6457: certain to be safe. It need not permit all classes of registers; the
6458: compiler knows how to copy a register into another register of the
6459: proper class in order to make an instruction valid.
6460:
6461: @item
1.1.1.8 root 6462: A nonoffsettable memory reference can be reloaded by copying the
1.1 root 6463: address into a register. So if the constraint uses the letter
6464: @samp{o}, all memory references are taken care of.
6465:
6466: @item
1.1.1.8 root 6467: A constant operand can be reloaded by allocating space in memory to
6468: hold it as preinitialized data. Then the memory reference can be used
6469: in place of the constant. So if the constraint uses the letters
6470: @samp{o} or @samp{m}, constant operands are not a problem.
1.1 root 6471: @end itemize
6472:
6473: If the operand's predicate can recognize registers, but the constraint does
6474: not permit them, it can make the compiler crash. When this operand happens
6475: to be a register, the reload pass will be stymied, because it does not know
6476: how to copy a register temporarily into memory.
6477:
6478: @node Multi-Alternative, Class Preferences, Simple Constraints, Constraints
6479: @subsection Multiple Alternative Constraints
6480:
6481: Sometimes a single instruction has multiple alternative sets of possible
6482: operands. For example, on the 68000, a logical-or instruction can combine
6483: register or an immediate value into memory, or it can combine any kind of
6484: operand into a register; but it cannot combine one memory location into
6485: another.
6486:
6487: These constraints are represented as multiple alternatives. An alternative
6488: can be described by a series of letters for each operand. The overall
6489: constraint for an operand is made from the letters for this operand
6490: from the first alternative, a comma, the letters for this operand from
6491: the second alternative, a comma, and so on until the last alternative.
6492: Here is how it is done for fullword logical-or on the 68000:
6493:
6494: @example
6495: (define_insn "iorsi3"
1.1.1.9 ! root 6496: [(set (match_operand:SI 0 "general_operand" "=m,d")
! 6497: (ior:SI (match_operand:SI 1 "general_operand" "%0,0")
1.1 root 6498: (match_operand:SI 2 "general_operand" "dKs,dmKs")))]
6499: @dots{})
6500: @end example
6501:
6502: The first alternative has @samp{m} (memory) for operand 0, @samp{0} for
1.1.1.9 ! root 6503: operand 1 (meaning it must match operand 0), and @samp{dKs} for operand
! 6504: 2. The second alternative has @samp{d} (data register) for operand 0,
! 6505: @samp{0} for operand 1, and @samp{dmKs} for operand 2. The @samp{=} and
! 6506: @samp{%} in the constraints apply to all the alternatives; their meaning
1.1 root 6507: is explained in the next section.
6508:
6509: If all the operands fit any one alternative, the instruction is valid.
6510: Otherwise, for each alternative, the compiler counts how many instructions
6511: must be added to copy the operands so that that alternative applies.
6512: The alternative requiring the least copying is chosen. If two alternatives
6513: need the same amount of copying, the one that comes first is chosen.
6514: These choices can be altered with the @samp{?} and @samp{!} characters:
6515:
6516: @table @samp
6517: @item ?
6518: Disparage slightly the alternative that the @samp{?} appears in,
6519: as a choice when no alternative applies exactly. The compiler regards
6520: this alternative as one unit more costly for each @samp{?} that appears
6521: in it.
6522:
6523: @item !
6524: Disparage severely the alternative that the @samp{!} appears in.
6525: When operands must be copied into registers, the compiler will
6526: never choose this alternative as the one to strive for.
6527: @end table
6528:
1.1.1.5 root 6529: When an insn pattern has multiple alternatives in its constraints, often
6530: the appearance of the assembler code is determined mostly by which
1.1 root 6531: alternative was matched. When this is so, the C code for writing the
6532: assembler code can use the variable @code{which_alternative}, which is
1.1.1.5 root 6533: the ordinal number of the alternative that was actually satisfied (0 for
6534: the first, 1 for the second alternative, etc.). For example:
1.1 root 6535:
6536: @example
6537: (define_insn ""
6538: [(set (match_operand:SI 0 "general_operand" "r,m")
6539: (const_int 0))]
6540: ""
6541: "*
6542: return (which_alternative == 0
6543: ? \"clrreg %0\" : \"clrmem %0\");
6544: ")
6545: @end example
6546:
6547: @node Class Preferences, Modifiers, Multi-Alternative, Constraints
6548: @subsection Register Class Preferences
6549:
6550: The operand constraints have another function: they enable the compiler
6551: to decide which kind of hardware register a pseudo register is best
6552: allocated to. The compiler examines the constraints that apply to the
6553: insns that use the pseudo register, looking for the machine-dependent
6554: letters such as @samp{d} and @samp{a} that specify classes of registers.
6555: The pseudo register is put in whichever class gets the most ``votes''.
6556: The constraint letters @samp{g} and @samp{r} also vote: they vote in
6557: favor of a general register. The machine description says which registers
6558: are considered general.
6559:
6560: Of course, on some machines all registers are equivalent, and no register
6561: classes are defined. Then none of this complexity is relevant.
6562:
6563: @node Modifiers, No Constraints, Class Preferences, Constraints
6564: @subsection Constraint Modifier Characters
6565:
6566: @table @samp
6567: @item =
6568: Means that this operand is write-only for this instruction: the previous
6569: value is discarded and replaced by output data.
6570:
6571: @item +
6572: Means that this operand is both read and written by the instruction.
6573:
6574: When the compiler fixes up the operands to satisfy the constraints,
6575: it needs to know which operands are inputs to the instruction and
6576: which are outputs from it. @samp{=} identifies an output; @samp{+}
6577: identifies an operand that is both input and output; all other operands
6578: are assumed to be input only.
6579:
6580: @item &
6581: Means (in a particular alternative) that this operand is written
6582: before the instruction is finished using the input operands.
6583: Therefore, this operand may not lie in a register that is used as an
6584: input operand or as part of any memory address.
6585:
6586: @samp{&} applies only to the alternative in which it is written. In
6587: constraints with multiple alternatives, sometimes one alternative
6588: requires @samp{&} while others do not. See, for example, the
6589: @samp{movdf} insn of the 68000.
6590:
6591: @samp{&} does not obviate the need to write @samp{=}.
6592:
6593: @item %
6594: Declares the instruction to be commutative for this operand and the
6595: following operand. This means that the compiler may interchange the
6596: two operands if that is the cheapest way to make all operands fit the
6597: constraints. This is often used in patterns for addition instructions
6598: that really have only two operands: the result must go in one of the
6599: arguments. Here for example, is how the 68000 halfword-add
6600: instruction is defined:
6601:
6602: @example
6603: (define_insn "addhi3"
6604: [(set (match_operand:HI 0 "general_operand" "=m,r")
6605: (plus:HI (match_operand:HI 1 "general_operand" "%0,0")
6606: (match_operand:HI 2 "general_operand" "di,g")))]
6607: @dots{})
6608: @end example
6609:
6610: Note that in previous versions of GNU CC the @samp{%} constraint
6611: modifier always applied to operands 1 and 2 regardless of which
6612: operand it was written in. The usual custom was to write it in
6613: operand 0. Now it must be in operand 1 if the operands to be
6614: exchanged are 1 and 2.
6615:
6616: @item #
6617: Says that all following characters, up to the next comma, are to be
6618: ignored as a constraint. They are significant only for choosing
6619: register preferences.
6620:
6621: @item *
6622: Says that the following character should be ignored when choosing
6623: register preferences. @samp{*} has no effect on the meaning of the
6624: constraint as a constraint.
6625:
6626: Here is an example: the 68000 has an instruction to sign-extend a
6627: halfword in a data register, and can also sign-extend a value by
6628: copying it into an address register. While either kind of register is
6629: acceptable, the constraints on an address-register destination are
6630: less strict, so it is best if register allocation makes an address
6631: register its goal. Therefore, @samp{*} is used so that the @samp{d}
6632: constraint letter (for data register) is ignored when computing
6633: register preferences.
6634:
6635: @example
6636: (define_insn "extendhisi2"
6637: [(set (match_operand:SI 0 "general_operand" "=*d,a")
6638: (sign_extend:SI
6639: (match_operand:HI 1 "general_operand" "0,g")))]
6640: @dots{})
6641: @end example
6642: @end table
6643:
6644: @node No Constraints,, Modifiers, Constraints
6645: @subsection Not Using Constraints
6646:
6647: Some machines are so clean that operand constraints are not required. For
6648: example, on the Vax, an operand valid in one context is valid in any other
6649: context. On such a machine, every operand constraint would be @samp{g},
6650: excepting only operands of ``load address'' instructions which are
6651: written as if they referred to a memory location's contents but actual
6652: refer to its address. They would have constraint @samp{p}.
6653:
6654: For such machines, instead of writing @samp{g} and @samp{p} for all
6655: the constraints, you can choose to write a description with empty constraints.
1.1.1.8 root 6656: Then you write @samp{""} for the constraint in every @code{match_operand}.
6657: Address operands are identified by writing an @code{address} expression
6658: around the @code{match_operand}, not by their constraints.
1.1 root 6659:
6660: When the machine description has just empty constraints, certain parts
1.1.1.6 root 6661: of compilation are skipped, making the compiler faster. However,
6662: few machines actually do not need constraints; all machine descriptions
6663: now in existence use constraints.
1.1 root 6664:
6665: @node Standard Names, Pattern Ordering, Constraints, Machine Desc
6666: @section Standard Names for Patterns Used in Generation
6667:
6668: Here is a table of the instruction names that are meaningful in the RTL
6669: generation pass of the compiler. Giving one of these names to an
6670: instruction pattern tells the RTL generation pass that it can use the
6671: pattern in to accomplish a certain task.
6672:
6673: @table @asis
6674: @item @samp{mov@var{m}}
1.1.1.8 root 6675: Here @var{m} stands for a two-letter machine mode name, in lower case.
6676: This instruction pattern moves data with that machine mode from operand
6677: 1 to operand 0. For example, @samp{movsi} moves full-word data.
1.1 root 6678:
1.1.1.8 root 6679: If operand 0 is a @code{subreg} with mode @var{m} of a register whose
6680: own mode is wider than @var{m}, the effect of this instruction is
1.1 root 6681: to store the specified value in the part of the register that corresponds
6682: to mode @var{m}. The effect on the rest of the register is undefined.
6683:
6684: This class of patterns is special in several ways. First of all, each
6685: of these names @emph{must} be defined, because there is no other way
6686: to copy a datum from one place to another.
6687:
6688: Second, these patterns are not used solely in the RTL generation pass.
6689: Even the reload pass can generate move insns to copy values from stack
1.1.1.8 root 6690: slots into temporary registers. When it does so, one of the operands is
6691: a hard register and the other is an operand that can need to be reloaded
6692: into a register.
6693:
6694: Therefore, when given such a pair of operands, the pattern must generate
6695: RTL which needs no reloading and needs no temporary registers---no
6696: registers other than the operands. For example, if you support the
6697: pattern with a @code{define_expand}, then in such a case the
6698: @code{define_expand} mustn't call @code{force_reg} or any other such
6699: function which might generate new pseudo registers.
1.1 root 6700:
6701: This requirement exists even for subword modes on a RISC machine where
6702: fetching those modes from memory normally requires several insns and
6703: some temporary registers. Look in @file{spur.md} to see how the
1.1.1.8 root 6704: requirement can be satisfied.
1.1 root 6705:
6706: The variety of operands that have reloads depends on the rest of the
6707: machine description, but typically on a RISC machine these can only be
6708: pseudo registers that did not get hard registers, while on other
6709: machines explicit memory references will get optional reloads.
6710:
1.1.1.9 ! root 6711: The constraints on a @samp{move@var{m}} must allow any hard register to
! 6712: be moved to any other hard register (provided that
! 6713: @code{HARD_REGNO_MODE_OK} permits mode @var{m} in both registers).
! 6714:
! 6715: It is obligatory to support floating point @samp{move@var{m}}
! 6716: instructions into and out of any registers that can hold fixed point
! 6717: values, because unions and structures (which have modes @code{SImode} or
! 6718: @code{DImode}) can be in those registers and they may have floating
! 6719: point members.
! 6720:
! 6721: There may also be a need to support fixed point @samp{move@var{m}}
! 6722: instructions in and out of floating point registers. Unfortunately, I
! 6723: have forgotten why this was so, and I don't know whether it is still
! 6724: true. If @code{HARD_REGNO_MODE_OK} rejects fixed point values in
! 6725: floating point registers, then the constraints of the fixed point
! 6726: @samp{move@var{m}} instructions must be designed to avoid ever trying to
! 6727: reload into a floating point register.
1.1 root 6728:
6729: @item @samp{movstrict@var{m}}
1.1.1.8 root 6730: Like @samp{mov@var{m}} except that if operand 0 is a @code{subreg}
1.1 root 6731: with mode @var{m} of a register whose natural mode is wider,
6732: the @samp{movstrict@var{m}} instruction is guaranteed not to alter
6733: any of the register except the part which belongs to mode @var{m}.
6734:
6735: @item @samp{add@var{m}3}
6736: Add operand 2 and operand 1, storing the result in operand 0. All operands
6737: must have mode @var{m}. This can be used even on two-address machines, by
6738: means of constraints requiring operands 1 and 0 to be the same location.
6739:
6740: @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}
6741: Similar, for other arithmetic operations.
6742:
6743: There are special considerations for register classes for logical-and
6744: instructions, affecting also the macro @code{PREFERRED_RELOAD_CLASS}.
6745: They apply not only to the patterns with these standard names, but to
6746: any patterns that will match such an instruction. @xref{Register
6747: Classes}.
6748:
6749: @item @samp{mulhisi3}
6750: Multiply operands 1 and 2, which have mode @code{HImode}, and store
6751: a @code{SImode} product in operand 0.
6752:
6753: @item @samp{mulqihi3}, @samp{mulsidi3}
6754: Similar widening-multiplication instructions of other widths.
6755:
6756: @item @samp{umulqihi3}, @samp{umulhisi3}, @samp{umulsidi3}
6757: Similar widening-multiplication instructions that do unsigned
6758: multiplication.
6759:
6760: @item @samp{divmod@var{m}4}
6761: Signed division that produces both a quotient and a remainder.
6762: Operand 1 is divided by operand 2 to produce a quotient stored
6763: in operand 0 and a remainder stored in operand 3.
6764:
6765: @item @samp{udivmod@var{m}4}
6766: Similar, but does unsigned division.
6767:
6768: @item @samp{ashl@var{m}3}
6769: Arithmetic-shift operand 1 left by a number of bits specified by
6770: operand 2, and store the result in operand 0. Operand 2 has
6771: mode @code{SImode}, not mode @var{m}.
6772:
6773: @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}
6774: Other shift and rotate instructions.
6775:
6776: Logical and arithmetic left shift are the same. Machines that do not
6777: allow negative shift counts often have only one instruction for
6778: shifting left. On such machines, you should define a pattern named
6779: @samp{ashl@var{m}3} and leave @samp{lshl@var{m}3} undefined.
6780:
6781: There are special considerations for register classes for shift
6782: instructions, affecting also the macro @code{PREFERRED_RELOAD_CLASS}.
6783: They apply not only to the patterns with these standard names, but to
6784: any patterns that will match such an instruction. @xref{Register
6785: Classes}.
6786:
6787: @item @samp{neg@var{m}2}
6788: Negate operand 1 and store the result in operand 0.
6789:
6790: @item @samp{abs@var{m}2}
6791: Store the absolute value of operand 1 into operand 0.
6792:
6793: @item @samp{sqrt@var{m}2}
6794: Store the square root of operand 1 into operand 0.
6795:
6796: @item @samp{ffs@var{m}2}
6797: Store into operand 0 one plus the index of the least significant 1-bit
6798: of operand 1. If operand 1 is zero, store zero. @var{m} is the mode
6799: of operand 0; operand 1's mode is specified by the instruction
6800: pattern, and the compiler will convert the operand to that mode before
6801: generating the instruction.
6802:
6803: @item @samp{one_cmpl@var{m}2}
6804: Store the bitwise-complement of operand 1 into operand 0.
6805:
6806: @item @samp{cmp@var{m}}
6807: Compare operand 0 and operand 1, and set the condition codes.
6808: The RTL pattern should look like this:
6809:
6810: @example
1.1.1.6 root 6811: (set (cc0) (compare (match_operand:@var{m} 0 @dots{})
6812: (match_operand:@var{m} 1 @dots{})))
1.1 root 6813: @end example
6814:
6815: Each such definition in the machine description, for integer mode
6816: @var{m}, must have a corresponding @samp{tst@var{m}} pattern, because
6817: optimization can simplify the compare into a test when operand 1 is
6818: zero.
6819:
6820: @item @samp{tst@var{m}}
6821: Compare operand 0 against zero, and set the condition codes.
6822: The RTL pattern should look like this:
6823:
6824: @example
6825: (set (cc0) (match_operand:@var{m} 0 @dots{}))
6826: @end example
6827:
6828: @item @samp{movstr@var{m}}
6829: Block move instruction. The addresses of the destination and source
6830: strings are the first two operands, and both are in mode @code{Pmode}.
6831: The number of bytes to move is the third operand, in mode @var{m}.
1.1.1.5 root 6832: The fourth operand is the known shared alignment of the source and
6833: destination, in the form of a @code{const_int} rtx.
1.1 root 6834:
6835: @item @samp{cmpstr@var{m}}
6836: Block compare instruction, with operands like @samp{movstr@var{m}}
6837: except that the two memory blocks are compared byte by byte
6838: in lexicographic order. The effect of the instruction is to set
6839: the condition codes.
6840:
6841: @item @samp{float@var{m}@var{n}2}
1.1.1.9 ! root 6842: Convert signed integer operand 1 (valid for fixed point mode @var{m}) to
! 6843: floating point mode @var{n} and store in operand 0 (which has mode
! 6844: @var{n}).
! 6845:
! 6846: @item @samp{floatuns@var{m}@var{n}2}
! 6847: Convert unsigned integer operand 1 (valid for fixed point mode @var{m})
! 6848: to floating point mode @var{n} and store in operand 0 (which has mode
! 6849: @var{n}).
1.1 root 6850:
6851: @item @samp{fix@var{m}@var{n}2}
6852: Convert operand 1 (valid for floating point mode @var{m}) to fixed
6853: point mode @var{n} as a signed number and store in operand 0 (which
6854: has mode @var{n}). This instruction's result is defined only when
6855: the value of operand 1 is an integer.
6856:
6857: @item @samp{fixuns@var{m}@var{n}2}
6858: Convert operand 1 (valid for floating point mode @var{m}) to fixed
6859: point mode @var{n} as an unsigned number and store in operand 0 (which
6860: has mode @var{n}). This instruction's result is defined only when the
6861: value of operand 1 is an integer.
6862:
6863: @item @samp{ftrunc@var{m}2}
6864: Convert operand 1 (valid for floating point mode @var{m}) to an
6865: integer value, still represented in floating point mode @var{m}, and
6866: store it in operand 0 (valid for floating point mode @var{m}).
6867:
6868: @item @samp{fix_trunc@var{m}@var{n}2}
6869: Like @samp{fix@var{m}@var{n}2} but works for any floating point value
6870: of mode @var{m} by converting the value to an integer.
6871:
6872: @item @samp{fixuns_trunc@var{m}@var{n}2}
6873: Like @samp{fixuns@var{m}@var{n}2} but works for any floating point
6874: value of mode @var{m} by converting the value to an integer.
6875:
6876: @item @samp{trunc@var{m}@var{n}}
6877: Truncate operand 1 (valid for mode @var{m}) to mode @var{n} and
6878: store in operand 0 (which has mode @var{n}). Both modes must be fixed
6879: point or both floating point.
6880:
6881: @item @samp{extend@var{m}@var{n}}
6882: Sign-extend operand 1 (valid for mode @var{m}) to mode @var{n} and
6883: store in operand 0 (which has mode @var{n}). Both modes must be fixed
6884: point or both floating point.
6885:
6886: @item @samp{zero_extend@var{m}@var{n}}
6887: Zero-extend operand 1 (valid for mode @var{m}) to mode @var{n} and
6888: store in operand 0 (which has mode @var{n}). Both modes must be fixed
6889: point.
6890:
6891: @item @samp{extv}
6892: Extract a bit-field from operand 1 (a register or memory operand),
6893: where operand 2 specifies the width in bits and operand 3 the starting
6894: bit, and store it in operand 0. Operand 0 must have @code{Simode}.
6895: Operand 1 may have mode @code{QImode} or @code{SImode}; often
6896: @code{SImode} is allowed only for registers. Operands 2 and 3 must be
6897: valid for @code{SImode}.
6898:
6899: The RTL generation pass generates this instruction only with constants
6900: for operands 2 and 3.
6901:
6902: The bit-field value is sign-extended to a full word integer
6903: before it is stored in operand 0.
6904:
6905: @item @samp{extzv}
6906: Like @samp{extv} except that the bit-field value is zero-extended.
6907:
6908: @item @samp{insv}
6909: Store operand 3 (which must be valid for @code{SImode}) into a
6910: bit-field in operand 0, where operand 1 specifies the width in bits
6911: and operand 2 the starting bit. Operand 0 may have mode @code{QImode}
6912: or @code{SImode}; often @code{SImode} is allowed only for registers.
6913: Operands 1 and 2 must be valid for @code{SImode}.
6914:
6915: The RTL generation pass generates this instruction only with constants
6916: for operands 1 and 2.
6917:
6918: @item @samp{s@var{cond}}
6919: Store zero or nonzero in the operand according to the condition codes.
6920: Value stored is nonzero iff the condition @var{cond} is true.
6921: @var{cond} is the name of a comparison operation expression code, such
1.1.1.8 root 6922: as @code{eq}, @code{lt} or @code{leu}.
1.1 root 6923:
6924: You specify the mode that the operand must have when you write the
6925: @code{match_operand} expression. The compiler automatically sees
6926: which mode you have used and supplies an operand of that mode.
6927:
1.1.1.8 root 6928: The value stored for a true condition must have 1 as its low bit, or
6929: else must be negative. Otherwise the instruction is not suitable and
6930: must be omitted from the machine description. You must tell the
6931: compiler exactly which value is stored by defining the macro
6932: @code{STORE_FLAG_VALUE}.
1.1 root 6933:
6934: @item @samp{b@var{cond}}
1.1.1.8 root 6935: Conditional branch instruction. Operand 0 is a @code{label_ref}
1.1 root 6936: that refers to the label to jump to. Jump if the condition codes
6937: meet condition @var{cond}.
6938:
6939: @item @samp{call}
6940: Subroutine call instruction returning no value. Operand 0 is the
6941: function to call; operand 1 is the number of bytes of arguments pushed
1.1.1.8 root 6942: (in mode @code{SImode}, except it is normally a @code{const_int});
1.1 root 6943: operand 2 is the number of registers used as operands.
6944:
6945: On most machines, operand 2 is not actually stored into the RTL
6946: pattern. It is supplied for the sake of some RISC machines which need
6947: to put this information into the assembler code; they can put it in
6948: the RTL instead of operand 1.
6949:
1.1.1.8 root 6950: Operand 0 should be a @code{mem} RTX whose address is the address of
1.1 root 6951: the function.
6952:
6953: @item @samp{call_value}
6954: Subroutine call instruction returning a value. Operand 0 is the hard
6955: register in which the value is returned. There are three more
6956: operands, the same as the three operands of the @samp{call}
6957: instruction (but with numbers increased by one).
6958:
6959: Subroutines that return @code{BLKmode} objects use the @samp{call}
6960: insn.
6961:
6962: @item @samp{return}
6963: Subroutine return instruction. This instruction pattern name should be
6964: defined only if a single instruction can do all the work of returning
6965: from a function.
6966:
1.1.1.8 root 6967: @item @samp{nop}
6968: No-op instruction. This instruction pattern name should always be defined
6969: to output a no-op in assembler code. @code{(const_int 0)} will do as an
6970: RTL pattern.
6971:
1.1 root 6972: @item @samp{casesi}
6973: Instruction to jump through a dispatch table, including bounds checking.
6974: This instruction takes five operands:
6975:
6976: @enumerate
6977: @item
6978: The index to dispatch on, which has mode @code{SImode}.
6979:
6980: @item
6981: The lower bound for indices in the table, an integer constant.
6982:
6983: @item
1.1.1.6 root 6984: The total range of indices in the table---the largest index
6985: minus the smallest one (both inclusive).
1.1 root 6986:
6987: @item
6988: A label to jump to if the index has a value outside the bounds.
6989: (If the machine-description macro @code{CASE_DROPS_THROUGH} is defined,
6990: then an out-of-bounds index drops through to the code following
6991: the jump table instead of jumping to this label. In that case,
6992: this label is not actually used by the @samp{casesi} instruction,
6993: but it is always provided as an operand.)
6994:
6995: @item
6996: A label that precedes the table itself.
6997: @end enumerate
6998:
1.1.1.8 root 6999: The table is a @code{addr_vec} or @code{addr_diff_vec} inside of a
7000: @code{jump_insn}. The number of elements in the table is one plus the
1.1 root 7001: difference between the upper bound and the lower bound.
7002:
7003: @item @samp{tablejump}
7004: Instruction to jump to a variable address. This is a low-level
7005: capability which can be used to implement a dispatch table when there
7006: is no @samp{casesi} pattern.
7007:
7008: This pattern requires two operands: the address or offset, and a label
7009: which should immediately precede the jump table. If the macro
7010: @code{CASE_VECTOR_PC_RELATIVE} is defined then the first operand is an
7011: absolute address to jump to; otherwise, it is an offset which counts
7012: from the address of the table.
7013:
7014: The @samp{tablejump} insn is always the last insn before the jump
7015: table it uses. Its assembler code normally has no need to use the
7016: second operand, but you should incorporate it in the RTL pattern so
7017: that the jump optimizer will not delete the table as unreachable code.
7018: @end table
7019:
7020: @node Pattern Ordering, Dependent Patterns, Standard Names, Machine Desc
7021: @section When the Order of Patterns Matters
7022:
7023: Sometimes an insn can match more than one instruction pattern. Then the
7024: pattern that appears first in the machine description is the one used.
7025: Therefore, more specific patterns (patterns that will match fewer things)
7026: and faster instructions (those that will produce better code when they
7027: do match) should usually go first in the description.
7028:
7029: In some cases the effect of ordering the patterns can be used to hide
7030: a pattern when it is not valid. For example, the 68000 has an
7031: instruction for converting a fullword to floating point and another
7032: for converting a byte to floating point. An instruction converting
7033: an integer to floating point could match either one. We put the
7034: pattern to convert the fullword first to make sure that one will
7035: be used rather than the other. (Otherwise a large integer might
7036: be generated as a single-byte immediate quantity, which would not work.)
7037: Instead of using this pattern ordering it would be possible to make the
7038: pattern for convert-a-byte smart enough to deal properly with any
7039: constant value.
7040:
7041: @node Dependent Patterns, Jump Patterns, Pattern Ordering, Machine Desc
7042: @section Interdependence of Patterns
7043:
7044: Every machine description must have a named pattern for each of the
7045: conditional branch names @samp{b@var{cond}}. The recognition template
7046: must always have the form
7047:
7048: @example
7049: (set (pc)
7050: (if_then_else (@var{cond} (cc0) (const_int 0))
7051: (label_ref (match_operand 0 "" ""))
7052: (pc)))
7053: @end example
7054:
7055: @noindent
7056: In addition, every machine description must have an anonymous pattern
7057: for each of the possible reverse-conditional branches. These patterns
7058: look like
7059:
7060: @example
7061: (set (pc)
7062: (if_then_else (@var{cond} (cc0) (const_int 0))
7063: (pc)
7064: (label_ref (match_operand 0 "" ""))))
7065: @end example
7066:
7067: @noindent
7068: They are necessary because jump optimization can turn direct-conditional
7069: branches into reverse-conditional branches.
7070:
7071: The compiler does more with RTL than just create it from patterns
7072: and recognize the patterns: it can perform arithmetic expression codes
7073: when constant values for their operands can be determined. As a result,
7074: sometimes having one pattern can require other patterns. For example, the
7075: Vax has no `and' instruction, but it has `and not' instructions. Here
7076: is the definition of one of them:
7077:
7078: @example
7079: (define_insn "andcbsi2"
7080: [(set (match_operand:SI 0 "general_operand" "")
7081: (and:SI (match_dup 0)
7082: (not:SI (match_operand:SI
7083: 1 "general_operand" ""))))]
7084: ""
7085: "bicl2 %1,%0")
7086: @end example
7087:
7088: @noindent
7089: If operand 1 is an explicit integer constant, an instruction constructed
7090: using that pattern can be simplified into an `and' like this:
7091:
7092: @example
7093: (set (reg:SI 41)
7094: (and:SI (reg:SI 41)
7095: (const_int 0xffff7fff)))
7096: @end example
7097:
7098: @noindent
7099: (where the integer constant is the one's complement of what
7100: appeared in the original instruction).
7101:
7102: To avoid a fatal error, the compiler must have a pattern that recognizes
7103: such an instruction. Here is what is used:
7104:
7105: @example
7106: (define_insn ""
7107: [(set (match_operand:SI 0 "general_operand" "")
7108: (and:SI (match_dup 0)
7109: (match_operand:SI 1 "general_operand" "")))]
7110: "GET_CODE (operands[1]) == CONST_INT"
7111: "*
7112: @{ operands[1]
7113: = gen_rtx (CONST_INT, VOIDmode, ~INTVAL (operands[1]));
7114: return \"bicl2 %1,%0\";
7115: @}")
7116: @end example
7117:
7118: @noindent
7119: Whereas a pattern to match a general `and' instruction is impossible to
7120: support on the Vax, this pattern is possible because it matches only a
7121: constant second argument: a special case that can be output as an `and not'
7122: instruction.
7123:
7124: A ``compare'' instruction whose RTL looks like this:
7125:
7126: @example
1.1.1.6 root 7127: (set (cc0) (compare @var{operand} (const_int 0)))
1.1 root 7128: @end example
7129:
7130: @noindent
7131: may be simplified by optimization into a ``test'' like this:
7132:
7133: @example
7134: (set (cc0) @var{operand})
7135: @end example
7136:
7137: @noindent
7138: So in the machine description, each ``compare'' pattern for an integer
7139: mode must have a corresponding ``test'' pattern that will match the
7140: result of such simplification.
7141:
7142: In some cases machines support instructions identical except for the
7143: machine mode of one or more operands. For example, there may be
7144: ``sign-extend halfword'' and ``sign-extend byte'' instructions whose
7145: patterns are
7146:
7147: @example
7148: (set (match_operand:SI 0 @dots{})
7149: (extend:SI (match_operand:HI 1 @dots{})))
7150:
7151: (set (match_operand:SI 0 @dots{})
7152: (extend:SI (match_operand:QI 1 @dots{})))
7153: @end example
7154:
7155: @noindent
7156: Constant integers do not specify a machine mode, so an instruction to
7157: extend a constant value could match either pattern. The pattern it
7158: actually will match is the one that appears first in the file. For correct
7159: results, this must be the one for the widest possible mode (@code{HImode},
7160: here). If the pattern matches the @code{QImode} instruction, the results
7161: will be incorrect if the constant value does not actually fit that mode.
7162:
7163: Such instructions to extend constants are rarely generated because they are
7164: optimized away, but they do occasionally happen in nonoptimized
7165: compilations.
7166:
7167: When an instruction has the constraint letter @samp{o}, the reload
1.1.1.8 root 7168: pass may generate instructions which copy a nonoffsettable address into
1.1 root 7169: an index register. The idea is that the register can be used as a
1.1.1.8 root 7170: replacement offsettable address. In order for these generated
1.1 root 7171: instructions to work, there must be patterns to copy any kind of valid
7172: address into a register.
7173:
7174: Most older machine designs have ``load address'' instructions which do
7175: just what is needed here. Some RISC machines do not advertise such
7176: instructions, but the possible addresses on these machines are very
7177: limited, so it is easy to fake them.
7178:
7179: Auto-increment and auto-decrement addresses are an exception; there
7180: need not be an instruction that can copy such an address into a
7181: register, because reload handles these cases in a different manner.
7182:
7183: @node Jump Patterns, Peephole Definitions, Dependent Patterns, Machine Desc
7184: @section Defining Jump Instruction Patterns
7185:
7186: GNU CC assumes that the machine has a condition code. A comparison insn
7187: sets the condition code, recording the results of both signed and unsigned
7188: comparison of the given operands. A separate branch insn tests the
7189: condition code and branches or not according its value. The branch insns
7190: come in distinct signed and unsigned flavors. Many common machines, such
7191: as the Vax, the 68000 and the 32000, work this way.
7192:
7193: Some machines have distinct signed and unsigned compare instructions, and
7194: only one set of conditional branch instructions. The easiest way to handle
7195: these machines is to treat them just like the others until the final stage
7196: where assembly code is written. At this time, when outputting code for the
7197: compare instruction, peek ahead at the following branch using
7198: @code{NEXT_INSN (insn)}. (The variable @code{insn} refers to the insn
7199: being output, in the output-writing code in an instruction pattern.) If
7200: the RTL says that is an unsigned branch, output an unsigned compare;
7201: otherwise output a signed compare. When the branch itself is output, you
7202: can treat signed and unsigned branches identically.
7203:
7204: The reason you can do this is that GNU CC always generates a pair of
7205: consecutive RTL insns, one to set the condition code and one to test it,
7206: and keeps the pair inviolate until the end.
7207:
7208: To go with this technique, you must define the machine-description macro
7209: @code{NOTICE_UPDATE_CC} to do @code{CC_STATUS_INIT}; in other words, no
7210: compare instruction is superfluous.
7211:
7212: Some machines have compare-and-branch instructions and no condition code.
7213: A similar technique works for them. When it is time to ``output'' a
7214: compare instruction, record its operands in two static variables. When
7215: outputting the branch-on-condition-code instruction that follows, actually
7216: output a compare-and-branch instruction that uses the remembered operands.
7217:
7218: It also works to define patterns for compare-and-branch instructions.
7219: In optimizing compilation, the pair of compare and branch instructions
1.1.1.5 root 7220: will be combined according to these patterns. But this does not happen
1.1 root 7221: if optimization is not requested. So you must use one of the solutions
7222: above in addition to any special patterns you define.
7223:
7224: @node Peephole Definitions, Expander Definitions, Jump Patterns, Machine Desc
7225: @section Defining Machine-Specific Peephole Optimizers
7226:
7227: In addition to instruction patterns the @file{md} file may contain
7228: definitions of machine-specific peephole optimizations.
7229:
7230: The combiner does not notice certain peephole optimizations when the data
7231: flow in the program does not suggest that it should try them. For example,
7232: sometimes two consecutive insns related in purpose can be combined even
7233: though the second one does not appear to use a register computed in the
7234: first one. A machine-specific peephole optimizer can detect such
7235: opportunities.
7236:
7237: A definition looks like this:
7238:
7239: @example
7240: (define_peephole
7241: [@var{insn-pattern-1}
7242: @var{insn-pattern-2}
7243: @dots{}]
7244: "@var{condition}"
7245: "@var{template}"
7246: "@var{machine-specific info}")
7247: @end example
7248:
7249: @noindent
7250: The last string operand may be omitted if you are not using any
7251: machine-specific information in this machine description. If present,
1.1.1.8 root 7252: it must obey the same rules as in a @code{define_insn}.
1.1 root 7253:
7254: In this skeleton, @var{insn-pattern-1} and so on are patterns to match
1.1.1.5 root 7255: consecutive insns. The optimization applies to a sequence of insns when
7256: @var{insn-pattern-1} matches the first one, @var{insn-pattern-2} matches
7257: the next, and so on.@refill
1.1 root 7258:
1.1.1.8 root 7259: Each of the insns matched by a peephole must also match a
7260: @code{define_insn}. Peepholes are checked only at the last stage just
7261: before code generation, and only optionally. Therefore, any insn which
7262: would match a peephole but no @code{define_insn} will cause a crash in code
7263: generation in an unoptimized compilation, or at various optimization
7264: stages.
1.1 root 7265:
1.1.1.5 root 7266: The operands of the insns are matched with @code{match_operands} and
7267: @code{match_dup}, as usual. What is not usual is that the operand numbers
7268: apply to all the insn patterns in the definition. So, you can check for
7269: identical operands in two insns by using @code{match_operand} in one insn
7270: and @code{match_dup} in the other.
1.1 root 7271:
7272: The operand constraints used in @code{match_operand} patterns do not have
1.1.1.8 root 7273: any direct effect on the applicability of the peephole, but they will
7274: be validated afterward, so make sure your constraints are general enough
7275: to apply whenever the peephole matches. If the peephole matches
7276: but the constraints are not satisfied, the compiler will crash.
7277:
7278: It is safe to omit constraints in all the operands of the peephole; or
7279: you can write constraints which serve as a double-check on the criteria
7280: previously tested.
1.1 root 7281:
1.1.1.5 root 7282: Once a sequence of insns matches the patterns, the @var{condition} is
7283: checked. This is a C expression which makes the final decision whether to
7284: perform the optimization (we do so if the expression is nonzero). If
1.1 root 7285: @var{condition} is omitted (in other words, the string is empty) then the
1.1.1.5 root 7286: optimization is applied to every sequence of insns that matches the
1.1 root 7287: patterns.
7288:
1.1.1.5 root 7289: The defined peephole optimizations are applied after register allocation
7290: is complete. Therefore, the peephole definition can check which
7291: operands have ended up in which kinds of registers, just by looking at
7292: the operands.
1.1 root 7293:
7294: The way to refer to the operands in @var{condition} is to write
7295: @code{operands[@var{i}]} for operand number @var{i} (as matched by
7296: @code{(match_operand @var{i} @dots{})}). Use the variable @code{insn} to
7297: refer to the last of the insns being matched; use @code{PREV_INSN} to find
1.1.1.8 root 7298: the preceding insns (but be careful to skip over any @code{note} insns that
1.1 root 7299: intervene).@refill
7300:
7301: When optimizing computations with intermediate results, you can use
7302: @var{condition} to match only when the intermediate results are not used
7303: elsewhere. Use the C expression @code{dead_or_set_p (@var{insn},
7304: @var{op})}, where @var{insn} is the insn in which you expect the value to
7305: be used for the last time (from the value of @code{insn}, together with use
7306: of @code{PREV_INSN}), and @var{op} is the intermediate value (from
7307: @code{operands[@var{i}]}).@refill
7308:
1.1.1.5 root 7309: Applying the optimization means replacing the sequence of insns with one
7310: new insn. The @var{template} controls ultimate output of assembler code
7311: for this combined insn. It works exactly like the template of a
7312: @code{define_insn}. Operand numbers in this template are the same ones
7313: used in matching the original sequence of insns.
1.1 root 7314:
7315: The result of a defined peephole optimizer does not need to match any of
1.1.1.5 root 7316: the insn patterns in the machine description; it does not even have an
7317: opportunity to match them. The peephole optimizer definition itself serves
7318: as the insn pattern to control how the insn is output.
7319:
7320: Defined peephole optimizers are run as assembler code is being output,
7321: so the insns they produce are never combined or rearranged in any way.
1.1 root 7322:
7323: Here is an example, taken from the 68000 machine description:
7324:
7325: @example
7326: (define_peephole
7327: [(set (reg:SI 15) (plus:SI (reg:SI 15) (const_int 4)))
7328: (set (match_operand:DF 0 "register_operand" "f")
7329: (match_operand:DF 1 "register_operand" "ad"))]
7330: "FP_REG_P (operands[0]) && ! FP_REG_P (operands[1])"
7331: "*
7332: @{
7333: rtx xoperands[2];
7334: xoperands[1] = gen_rtx (REG, SImode, REGNO (operands[1]) + 1);
7335: #ifdef MOTOROLA
7336: output_asm_insn (\"move.l %1,(sp)\", xoperands);
7337: output_asm_insn (\"move.l %1,-(sp)\", operands);
7338: return \"fmove.d (sp)+,%0\";
7339: #else
7340: output_asm_insn (\"movel %1,sp@@\", xoperands);
7341: output_asm_insn (\"movel %1,sp@@-\", operands);
7342: return \"fmoved sp@@+,%0\";
7343: #endif
7344: @}
7345: ")
7346: @end example
7347:
7348: The effect of this optimization is to change
7349:
7350: @example
7351: jbsr _foobar
7352: addql #4,sp
7353: movel d1,sp@@-
7354: movel d0,sp@@-
7355: fmoved sp@@+,fp0
7356: @end example
7357:
7358: @noindent
7359: into
7360:
7361: @example
7362: jbsr _foobar
7363: movel d1,sp@@
7364: movel d0,sp@@-
7365: fmoved sp@@+,fp0
7366: @end example
7367:
1.1.1.5 root 7368: @ignore
7369: If a peephole matches a sequence including one or more jump insns, you must
7370: take account of the flags such as @code{CC_REVERSED} which specify that the
7371: condition codes are represented in an unusual manner. The compiler
7372: automatically alters any ordinary conditional jumps which occur in such
7373: situations, but the compiler cannot alter jumps which have been replaced by
7374: peephole optimizations. So it is up to you to alter the assembler code
7375: that the peephole produces. Supply C code to write the assembler output,
7376: and in this C code check the condition code status flags and change the
7377: assembler code as appropriate.
7378: @end ignore
7379:
1.1.1.8 root 7380: @var{insn-pattern-1} and so on look @emph{almost} like the second
7381: operand of @code{define_insn}. There is one important difference: the
7382: second operand of @code{define_insn} consists of one or more RTX's
7383: enclosed in square brackets. Usually, there is only one: then the same
7384: action can be written as an element of a @code{define_peephole}. But
7385: when there are multiple actions in a @code{define_insn}, they are
7386: implicitly enclosed in a @code{parallel}. Then you must explicitly
7387: write the @code{parallel}, and the square brackets within it, in the
7388: @code{define_peephole}. Thus, if an insn pattern looks like this,
7389:
7390: @example
7391: (define_insn "divmodsi4"
7392: [(set (match_operand:SI 0 "general_operand" "=d")
7393: (div:SI (match_operand:SI 1 "general_operand" "0")
7394: (match_operand:SI 2 "general_operand" "dmsK")))
7395: (set (match_operand:SI 3 "general_operand" "=d")
7396: (mod:SI (match_dup 1) (match_dup 2)))]
7397: "TARGET_68020"
7398: "divsl%.l %2,%3:%0")
7399: @end example
7400:
7401: @noindent
7402: then the way to mention this insn in a peephole is as follows:
7403:
7404: @example
7405: (define_peephole
7406: [@dots{}
7407: (parallel
7408: [(set (match_operand:SI 0 "general_operand" "=d")
7409: (div:SI (match_operand:SI 1 "general_operand" "0")
7410: (match_operand:SI 2 "general_operand" "dmsK")))
7411: (set (match_operand:SI 3 "general_operand" "=d")
7412: (mod:SI (match_dup 1) (match_dup 2)))])
7413: @dots{}]
7414: @dots{})
7415: @end example
7416:
1.1 root 7417: @node Expander Definitions,, Peephole Definitions, Machine Desc
7418: @section Defining RTL Sequences for Code Generation
7419:
7420: On some target machines, some standard pattern names for RTL generation
7421: cannot be handled with single insn, but a sequence of RTL insns can
7422: represent them. For these target machines, you can write a
1.1.1.8 root 7423: @code{define_expand} to specify how to generate the sequence of RTL.
1.1 root 7424:
1.1.1.8 root 7425: A @code{define_expand} is an RTL expression that looks almost like a
7426: @code{define_insn}; but, unlike the latter, a @code{define_expand} is used
1.1 root 7427: only for RTL generation and it can produce more than one RTL insn.
7428:
1.1.1.8 root 7429: A @code{define_expand} RTX has four operands:
1.1 root 7430:
7431: @itemize @bullet
7432: @item
1.1.1.8 root 7433: The name. Each @code{define_expand} must have a name, since the only
1.1 root 7434: use for it is to refer to it by name.
7435:
7436: @item
7437: The RTL template. This is just like the RTL template for a
1.1.1.8 root 7438: @code{define_peephole} in that it is a vector of RTL expressions
1.1 root 7439: each being one insn.
7440:
7441: @item
7442: The condition, a string containing a C expression. This expression is
7443: used to express how the availability of this pattern depends on
7444: subclasses of target machine, selected by command-line options when
7445: GNU CC is run. This is just like the condition of a
1.1.1.8 root 7446: @code{define_insn} that has a standard name.
1.1 root 7447:
7448: @item
7449: The preparation statements, a string containing zero or more C
7450: statements which are to be executed before RTL code is generated from
7451: the RTL template.
7452:
7453: Usually these statements prepare temporary registers for use as
7454: internal operands in the RTL template, but they can also generate RTL
1.1.1.8 root 7455: insns directly by calling routines such as @code{emit_insn}, etc.
1.1 root 7456: Any such insns precede the ones that come from the RTL template.
7457: @end itemize
7458:
1.1.1.8 root 7459: Every RTL insn emitted by a @code{define_expand} must match some
7460: @code{define_insn} in the machine description. Otherwise, the compiler
7461: will crash when trying to generate code for the insn or trying to optimize
7462: it.
7463:
1.1 root 7464: The RTL template, in addition to controlling generation of RTL insns,
7465: also describes the operands that need to be specified when this pattern
7466: is used. In particular, it gives a predicate for each operand.
7467:
7468: A true operand, which need to be specified in order to generate RTL from
1.1.1.8 root 7469: the pattern, should be described with a @code{match_operand} in its first
1.1 root 7470: occurrence in the RTL template. This enters information on the operand's
7471: predicate into the tables that record such things. GNU CC uses the
7472: information to preload the operand into a register if that is required for
7473: valid RTL code. If the operand is referred to more than once, subsequent
1.1.1.8 root 7474: references should use @code{match_dup}.
1.1 root 7475:
7476: The RTL template may also refer to internal ``operands'' which are
7477: temporary registers or labels used only within the sequence made by the
1.1.1.8 root 7478: @code{define_expand}. Internal operands are substituted into the RTL
7479: template with @code{match_dup}, never with @code{match_operand}. The
1.1 root 7480: values of the internal operands are not passed in as arguments by the
7481: compiler when it requests use of this pattern. Instead, they are computed
7482: within the pattern, in the preparation statements. These statements
7483: compute the values and store them into the appropriate elements of
1.1.1.8 root 7484: @code{operands} so that @code{match_dup} can find them.
1.1 root 7485:
7486: There are two special macros defined for use in the preparation statements:
7487: @code{DONE} and @code{FAIL}. Use them with a following semicolon,
7488: as a statement.
7489:
7490: @table @code
7491: @item DONE
7492: Use the @code{DONE} macro to end RTL generation for the pattern. The
7493: only RTL insns resulting from the pattern on this occasion will be
7494: those already emitted by explicit calls to @code{emit_insn} within the
7495: preparation statements; the RTL template will not be generated.
7496:
7497: @item FAIL
7498: Make the pattern fail on this occasion. When a pattern fails, it means
7499: that the pattern was not truly available. The calling routines in the
7500: compiler will try other strategies for code generation using other patterns.
7501:
7502: Failure is currently supported only for binary operations (addition,
7503: multiplication, shifting, etc.).
7504:
7505: Do not emit any insns explicitly with @code{emit_insn} before failing.
7506: @end table
7507:
7508: Here is an example, the definition of left-shift for the SPUR chip:
7509:
7510: @example
7511: (define_expand "ashlsi3"
7512: [(set (match_operand:SI 0 "register_operand" "")
7513: (ashift:SI
7514: (match_operand:SI 1 "register_operand" "")
7515: (match_operand:SI 2 "nonmemory_operand" "")))]
7516: ""
7517: "
7518: @{
7519: if (GET_CODE (operands[2]) != CONST_INT
7520: || (unsigned) INTVAL (operands[2]) > 3)
7521: FAIL;
7522: @}")
7523: @end example
7524:
7525: @noindent
1.1.1.8 root 7526: This example uses @code{define_expand} so that it can generate an RTL insn
1.1 root 7527: for shifting when the shift-count is in the supported range of 0 to 3 but
7528: fail in other cases where machine insns aren't available. When it fails,
7529: the compiler tries another strategy using different patterns (such as, a
7530: library call).
7531:
7532: If the compiler were able to handle nontrivial condition-strings in
1.1.1.8 root 7533: patterns with names, then it would be possible to use a
7534: @code{define_insn} in that case. Here is another case (zero-extension
7535: on the 68000) which makes more use of the power of @code{define_expand}:
1.1 root 7536:
7537: @example
7538: (define_expand "zero_extendhisi2"
7539: [(set (match_operand:SI 0 "general_operand" "")
7540: (const_int 0))
7541: (set (strict_low_part
7542: (subreg:HI
1.1.1.8 root 7543: (match_dup 0)
1.1 root 7544: 0))
7545: (match_operand:HI 1 "general_operand" ""))]
7546: ""
7547: "operands[1] = make_safe_from (operands[1], operands[0]);")
7548: @end example
7549:
7550: @noindent
7551: Here two RTL insns are generated, one to clear the entire output operand
7552: and the other to copy the input operand into its low half. This sequence
7553: is incorrect if the input operand refers to [the old value of] the output
7554: operand, so the preparation statement makes sure this isn't so. The
7555: function @code{make_safe_from} copies the @code{operands[1]} into a
7556: temporary register if it refers to @code{operands[0]}. It does this
7557: by emitting another RTL insn.
7558:
7559: Finally, a third example shows the use of an internal operand.
1.1.1.8 root 7560: Zero-extension on the SPUR chip is done by @code{and}-ing the result
1.1 root 7561: against a halfword mask. But this mask cannot be represented by a
1.1.1.8 root 7562: @code{const_int} because the constant value is too large to be legitimate
1.1 root 7563: on this machine. So it must be copied into a register with
1.1.1.8 root 7564: @code{force_reg} and then the register used in the @code{and}.
1.1 root 7565:
7566: @example
7567: (define_expand "zero_extendhisi2"
7568: [(set (match_operand:SI 0 "register_operand" "")
7569: (and:SI (subreg:SI
7570: (match_operand:HI 1 "register_operand" "")
7571: 0)
7572: (match_dup 2)))]
7573: ""
7574: "operands[2]
7575: = force_reg (SImode, gen_rtx (CONST_INT,
7576: VOIDmode, 65535)); ")
7577: @end example
7578:
1.1.1.8 root 7579: @strong{Note:} If the @code{define_expand} is used to serve a standard
7580: binary or unary arithmetic operation, then the last insn it generates
7581: must not be a @code{code_label}, @code{barrier} or @code{note}. It must
7582: be an @code{insn}, @code{jump_insn} or @code{call_insn}.
7583:
1.1 root 7584: @node Machine Macros, Config, Machine Desc, Top
7585: @chapter Machine Description Macros
7586:
7587: The other half of the machine description is a C header file conventionally
7588: given the name @file{tm-@var{machine}.h}. The file @file{tm.h} should be a
7589: link to it. The header file @file{config.h} includes @file{tm.h} and most
7590: compiler source files include @file{config.h}.
7591:
7592: @menu
1.1.1.9 ! root 7593: * Run-time Target:: Defining @samp{-m} options like @samp{-m68000} and @samp{-m68020}.
1.1 root 7594: * Storage Layout:: Defining sizes and alignments of data types.
7595: * Registers:: Naming and describing the hardware registers.
7596: * Register Classes:: Defining the classes of hardware registers.
7597: * Stack Layout:: Defining which way the stack grows and by how much.
7598: * Library Names:: Specifying names of subroutines to call automatically.
7599: * Addressing Modes:: Defining addressing modes valid for memory operands.
1.1.1.8 root 7600: * Delayed Branch:: Do branches execute the following instruction?
1.1 root 7601: * Condition Code:: Defining how insns update the condition code.
7602: * Assembler Format:: Defining how to write insns and pseudo-ops to output.
1.1.1.5 root 7603: * Cross-compilation:: Handling floating point for cross-compilers.
1.1 root 7604: * Misc:: Everything else.
7605: @end menu
7606:
7607: @node Run-time Target, Storage Layout, Machine Macros, Machine Macros
7608: @section Run-time Target Specification
7609:
7610: @table @code
7611: @item CPP_PREDEFINES
7612: Define this to be a string constant containing @samp{-D} options to
7613: define the predefined macros that identify this machine and system.
7614: These macros will be predefined unless the @samp{-ansi} option is
7615: specified.
7616:
1.1.1.4 root 7617: In addition, a parallel set of macros are predefined, whose names are
7618: made by appending @samp{__} at the beginning and at the end. These
7619: @samp{__} macros are permitted by the ANSI standard, so they are
7620: predefined regardless of whether @samp{-ansi} is specified.
7621:
7622: For example, on the Sun, one can use the following value:
1.1 root 7623:
7624: @example
7625: "-Dmc68000 -Dsun -Dunix"
7626: @end example
7627:
1.1.1.8 root 7628: The result is to define the macros @code{__mc68000__}, @code{__sun__}
7629: and @code{__unix__} unconditionally, and the macros @code{mc68000},
7630: @code{sun} and @code{unix} provided @samp{-ansi} is not specified.
1.1.1.4 root 7631:
1.1 root 7632: @item CPP_SPEC
7633: A C string constant that tells the GNU CC driver program options to
7634: pass to CPP. It can also specify how to translate options you
7635: give to GNU CC into options for GNU CC to pass to the CPP.
7636:
7637: Do not define this macro if it does not need to do anything.
7638:
7639: @item CC1_SPEC
7640: A C string constant that tells the GNU CC driver program options to
7641: pass to CC1. It can also specify how to translate options you
7642: give to GNU CC into options for GNU CC to pass to the CC1.
7643:
7644: Do not define this macro if it does not need to do anything.
7645:
7646: @item extern int target_flags;
7647: This declaration should be present.
7648:
7649: @item TARGET_@dots{}
7650: This series of macros is to allow compiler command arguments to
7651: enable or disable the use of optional features of the target machine.
7652: For example, one machine description serves both the 68000 and
7653: the 68020; a command argument tells the compiler whether it should
7654: use 68020-only instructions or not. This command argument works
7655: by means of a macro @code{TARGET_68020} that tests a bit in
7656: @code{target_flags}.
7657:
7658: Define a macro @code{TARGET_@var{featurename}} for each such option.
7659: Its definition should test a bit in @code{target_flags}; for example:
7660:
7661: @example
7662: #define TARGET_68020 (target_flags & 1)
7663: @end example
7664:
7665: One place where these macros are used is in the condition-expressions
7666: of instruction patterns. Note how @code{TARGET_68020} appears
7667: frequently in the 68000 machine description file, @file{m68k.md}.
7668: Another place they are used is in the definitions of the other
7669: macros in the @file{tm-@var{machine}.h} file.
7670:
7671: @item TARGET_SWITCHES
7672: This macro defines names of command options to set and clear
7673: bits in @code{target_flags}. Its definition is an initializer
7674: with a subgrouping for each command option.
7675:
7676: Each subgrouping contains a string constant, that defines the option
7677: name, and a number, which contains the bits to set in
7678: @code{target_flags}. A negative number says to clear bits instead;
7679: the negative of the number is which bits to clear. The actual option
7680: name is made by appending @samp{-m} to the specified name.
7681:
7682: One of the subgroupings should have a null string. The number in
7683: this grouping is the default value for @code{target_flags}. Any
7684: target options act starting with that value.
7685:
7686: Here is an example which defines @samp{-m68000} and @samp{-m68020}
7687: with opposite meanings, and picks the latter as the default:
7688:
7689: @example
7690: #define TARGET_SWITCHES \
7691: @{ @{ "68020", 1@}, \
7692: @{ "68000", -1@}, \
7693: @{ "", 1@}@}
7694: @end example
7695:
7696: @item OVERRIDE_OPTIONS
7697: Sometimes certain combinations of command options do not make sense on
7698: a particular target machine. You can define a macro
7699: @code{OVERRIDE_OPTIONS} to take account of this. This macro, if
7700: defined, is executed once just after all the command options have been
7701: parsed.
7702: @end table
7703:
7704: @node Storage Layout, Registers, Run-time Target, Machine Macros
7705: @section Storage Layout
7706:
7707: Note that the definitions of the macros in this table which are sizes or
7708: alignments measured in bits do not need to be constant. They can be C
7709: expressions that refer to static variables, such as the @code{target_flags}.
7710: @xref{Run-time Target}.
7711:
7712: @table @code
7713: @item BITS_BIG_ENDIAN
7714: Define this macro if the most significant bit in a byte has the lowest
7715: number. This means that bit-field instructions count from the most
7716: significant bit. If the machine has no bit-field instructions, this
7717: macro is irrelevant.
7718:
1.1.1.8 root 7719: This macro does not affect the way structure fields are packed into
7720: bytes or words; that is controlled by @code{BYTES_BIG_ENDIAN}.
7721:
1.1 root 7722: @item BYTES_BIG_ENDIAN
7723: Define this macro if the most significant byte in a word has the
7724: lowest number.
7725:
7726: @item WORDS_BIG_ENDIAN
7727: Define this macro if, in a multiword object, the most significant
7728: word has the lowest number.
7729:
7730: @item BITS_PER_UNIT
7731: Number of bits in an addressable storage unit (byte); normally 8.
7732:
7733: @item BITS_PER_WORD
7734: Number of bits in a word; normally 32.
7735:
7736: @item UNITS_PER_WORD
7737: Number of storage units in a word; normally 4.
7738:
7739: @item POINTER_SIZE
7740: Width of a pointer, in bits.
7741:
7742: @item POINTER_BOUNDARY
7743: Alignment required for pointers stored in memory, in bits.
7744:
7745: @item PARM_BOUNDARY
1.1.1.7 root 7746: Normal alignment required for function parameters on the stack, in
7747: bits. All stack parameters receive least this much alignment
7748: regardless of data type. On most machines, this is the same as the
7749: size of an integer.
7750:
7751: @item MAX_PARM_BOUNDARY
7752: Largest alignment required for any stack parameters, in bits. If the
7753: data type of the parameter calls for more alignment than
7754: @code{PARM_BOUNDARY}, then it is given extra padding up to this limit.
7755:
7756: Don't define this macro if it would be equal to @code{PARM_BOUNDARY};
7757: in other words, if the alignment of a stack parameter should not
7758: depend on its data type (as is the case on most machines).
1.1 root 7759:
7760: @item STACK_BOUNDARY
7761: Define this macro if you wish to preserve a certain alignment for
7762: the stack pointer at all times. The definition is a C expression
7763: for the desired alignment (measured in bits).
7764:
7765: @item FUNCTION_BOUNDARY
7766: Alignment required for a function entry point, in bits.
7767:
7768: @item BIGGEST_ALIGNMENT
7769: Biggest alignment that any data type can require on this machine, in bits.
7770:
1.1.1.8 root 7771: @item CONSTANT_ALIGNMENT (@var{code}, @var{typealign})
7772: A C expression to compute the alignment for a constant. The argument
7773: @var{typealign} is the alignment required for the constant's data type.
7774: @var{code} is the tree code of the constant itself.
7775:
7776: If this macro is not defined, the default is to use @var{typealign}. If
7777: you do define this macro, the value must be a multiple of
7778: @var{typealign}.
7779:
7780: The purpose of defining this macro is usually to cause string constants
7781: to be word aligned so that @file{dhrystone} can be made to run faster.
7782:
1.1 root 7783: @item EMPTY_FIELD_BOUNDARY
7784: Alignment in bits to be given to a structure bit field that follows an
7785: empty field such as @code{int : 0;}.
7786:
7787: @item STRUCTURE_SIZE_BOUNDARY
7788: Number of bits which any structure or union's size must be a multiple of.
7789: Each structure or union's size is rounded up to a multiple of this.
7790:
7791: If you do not define this macro, the default is the same as
7792: @code{BITS_PER_UNIT}.
7793:
7794: @item STRICT_ALIGNMENT
7795: Define this if instructions will fail to work if given data not
7796: on the nominal alignment. If instructions will merely go slower
7797: in that case, do not define this macro.
7798:
7799: @item PCC_BITFIELD_TYPE_MATTERS
7800: Define this if you wish to imitate a certain bizarre behavior pattern
7801: of some instances of PCC: a bit field whose declared type is
7802: @code{int} has the same effect on the size and alignment of a
7803: structure as an actual @code{int} would have.
7804:
7805: Just what effect that is in GNU CC depends on other parameters, but on
7806: most machines it would force the structure's alignment and size to a
7807: multiple of 32 or @code{BIGGEST_ALIGNMENT} bits.
7808:
1.1.1.7 root 7809: @item MAX_FIXED_MODE_SIZE
7810: An integer expression for the largest integer machine mode that should
7811: actually be used. All integer machine modes of this size or smaller
7812: can be used for structures and unions with the appropriate sizes.
7813:
1.1 root 7814: @item CHECK_FLOAT_VALUE (@var{mode}, @var{value})
7815: A C statement to validate the value @var{value} (or type
7816: @code{double}) for mode @var{mode}. This means that you check whether
7817: @var{value} fits within the possible range of values for mode
7818: @var{mode} on this target machine. The mode @var{mode} is always
7819: @code{SFmode} or @code{DFmode}.
7820:
7821: If @var{value} is not valid, you should call @code{error} to print an
7822: error message and then assign some valid value to @var{value}.
7823: Allowing an invalid value to go through the compiler can produce
7824: incorrect assembler code which may even cause Unix assemblers to
7825: crash.
7826:
7827: This macro need not be defined if there is no work for it to do.
7828: @end table
7829:
7830: @node Registers, Register Classes, Storage Layout, Machine Macros
7831: @section Register Usage
7832:
7833: @table @code
7834: @item FIRST_PSEUDO_REGISTER
7835: Number of hardware registers known to the compiler. They receive
7836: numbers 0 through @code{FIRST_PSEUDO_REGISTER-1}; thus, the first
7837: pseudo register's number really is assigned the number
7838: @code{FIRST_PSEUDO_REGISTER}.
7839:
7840: @item FIXED_REGISTERS
7841: An initializer that says which registers are used for fixed purposes
7842: all throughout the compiled code and are therefore not available for
7843: general allocation. These would include the stack pointer, the frame
7844: pointer (except on machines where that can be used as a general
7845: register when no frame pointer is needed), the program counter on
7846: machines where that is considered one of the addressable registers,
7847: and any other numbered register with a standard use.
7848:
7849: This information is expressed as a sequence of numbers, separated by
7850: commas and surrounded by braces. The @var{n}th number is 1 if
7851: register @var{n} is fixed, 0 otherwise.
7852:
7853: The table initialized from this macro, and the table initialized by
7854: the following one, may be overridden at run time either automatically,
7855: by the actions of the macro @code{CONDITIONAL_REGISTER_USAGE}, or by
7856: the user with the command options @samp{-ffixed-@var{reg}},
7857: @samp{-fcall-used-@var{reg}} and @samp{-fcall-saved-@var{reg}}.
7858:
7859: @item CALL_USED_REGISTERS
7860: Like @code{FIXED_REGISTERS} but has 1 for each register that is
7861: clobbered (in general) by function calls as well as for fixed
7862: registers. This macro therefore identifies the registers that are not
7863: available for general allocation of values that must live across
7864: function calls.
7865:
7866: If a register has 0 in @code{CALL_USED_REGISTERS}, the compiler
7867: automatically saves it on function entry and restores it on function
7868: exit, if the register is used within the function.
7869:
1.1.1.6 root 7870: @item DEFAULT_CALLER_SAVES
1.1.1.8 root 7871: Define this macro if function calls on the target machine do not preserve
1.1.1.6 root 7872: any registers; in other words, if @code{CALL_USED_REGISTERS} has 1
7873: for all registers. This macro enables @samp{-fcaller-saves} by default.
7874: Eventually that option will be enabled by default on all machines and both
7875: the option and this macro will be eliminated.
7876:
1.1 root 7877: @item CONDITIONAL_REGISTER_USAGE
7878: Zero or more C statements that may conditionally modify two variables
7879: @code{fixed_regs} and @code{call_used_regs} (both of type @code{char
7880: []}) after they have been initialized from the two preceding macros.
7881:
7882: This is necessary in case the fixed or call-clobbered registers depend
7883: on target flags.
7884:
7885: You need not define this macro if it has no work to do.
7886:
7887: If the usage of an entire class of registers depends on the target
1.1.1.5 root 7888: flags, you may indicate this to GCC by using this macro to modify
1.1 root 7889: @code{fixed_regs} and @code{call_used_regs} to 1 for each of the
1.1.1.5 root 7890: registers in the classes which should not be used by GCC. Also define
1.1 root 7891: the macro @code{REG_CLASS_FROM_LETTER} to return @code{NO_REGS} if it
7892: is called with a letter for a class that shouldn't be used.
7893:
7894: (However, if this class is not included in @code{GENERAL_REGS} and all
7895: of the insn patterns whose constraints permit this class are
7896: controlled by target switches, then GCC will automatically avoid using
7897: these registers when the target switches are opposed to them.)
7898:
7899: @item OVERLAPPING_REGNO_P (@var{regno})
1.1.1.5 root 7900: If defined, this is a C expression whose value is nonzero if hard
7901: register number @var{regno} is an overlapping register. This means a
7902: hard register which overlaps a hard register with a different number.
7903: (Such overlap is undesirable, but occasionally it allows a machine to
7904: be supported which otherwise could not be.) This macro must return
7905: nonzero for @emph{all} the registers which overlap each other. GNU CC
7906: can use an overlapping register only in certain limited ways. It can
7907: be used for allocation within a basic block, and may be spilled for
7908: reloading; that is all.
1.1 root 7909:
7910: If this macro is not defined, it means that none of the hard registers
7911: overlap each other. This is the usual situation.
7912:
7913: @item INSN_CLOBBERS_REGNO_P (@var{insn}, @var{regno})
7914: If defined, this is a C expression whose value should be nonzero if
7915: the insn @var{insn} has the effect of mysteriously clobbering the
7916: contents of hard register number @var{regno}. By ``mysterious'' we
7917: mean that the insn's RTL expression doesn't describe such an effect.
7918:
7919: If this macro is not defined, it means that no insn clobbers registers
7920: mysteriously. This is the usual situation; all else being equal,
7921: it is best for the RTL expression to show all the activity.
7922:
7923: @item PRESERVE_DEATH_INFO_REGNO_P (@var{regno})
7924: If defined, this is a C expression whose value is nonzero if accurate
7925: @code{REG_DEAD} notes are needed for hard register number @var{regno}
7926: at the time of outputting the assembler code. When this is so, a few
7927: optimizations that take place after register allocation and could
7928: invalidate the death notes are not done when this register is
7929: involved.
7930:
1.1.1.8 root 7931: You would arrange to preserve death info for a register when some of the
7932: code in the machine description which is executed to write the assembler
7933: code looks at the death notes. This is necessary only when the actual
7934: hardware feature which GNU CC thinks of as a register is not actually a
7935: register of the usual sort. (It might, for example, be a hardware
7936: stack.)
1.1 root 7937:
7938: If this macro is not defined, it means that no death notes need to be
7939: preserved. This is the usual situation.
7940:
7941: @item HARD_REGNO_REGS (@var{regno}, @var{mode})
7942: A C expression for the number of consecutive hard registers, starting
7943: at register number @var{regno}, required to hold a value of mode
7944: @var{mode}.
7945:
7946: On a machine where all registers are exactly one word, a suitable
7947: definition of this macro is
7948:
7949: @example
7950: #define HARD_REGNO_NREGS(REGNO, MODE) \
7951: ((GET_MODE_SIZE (MODE) + UNITS_PER_WORD - 1) \
7952: / UNITS_PER_WORD))
7953: @end example
7954:
7955: @item HARD_REGNO_MODE_OK (@var{regno}, @var{mode})
7956: A C expression that is nonzero if it is permissible to store a value
7957: of mode @var{mode} in hard register number @var{regno} (or in several
7958: registers starting with that one). For a machine where all registers
7959: are equivalent, a suitable definition is
7960:
7961: @example
7962: #define HARD_REGNO_MODE_OK(REGNO, MODE) 1
7963: @end example
7964:
1.1.1.8 root 7965: It is not necessary for this macro to check for the numbers of fixed
7966: registers, because the allocation mechanism considers them to be always
7967: occupied.
7968:
7969: On some machines, double-precision values must be kept in even/odd
7970: register pairs. The way to implement that is to define this macro
7971: to reject odd register numbers for such modes.
7972:
7973: GNU CC assumes that it can always move values between registers and
7974: (suitably addressed) memory locations. If it is impossible to move a
7975: value of a certain mode between memory and certain registers, then
7976: @code{HARD_REGNO_MODE_OK} must not allow this mode in those registers.
1.1 root 7977:
7978: Many machines have special registers for floating point arithmetic.
7979: Often people assume that floating point machine modes are allowed only
7980: in floating point registers. This is not true. Any registers that
7981: can hold integers can safely @emph{hold} a floating point machine
7982: mode, whether or not floating arithmetic can be done on it in those
7983: registers.
7984:
1.1.1.9 ! root 7985: On some machines, though, the converse is true: fixed-point machine
! 7986: modes may not go in floating registers. This is true if the floating
! 7987: registers normalize any value stored in them, because storing a
! 7988: non-floating value there would garble it. In this case,
! 7989: @code{HARD_REGNO_MODE_OK} should reject fixed-point machine modes in
! 7990: floating registers. But if the floating registers do not automatically
! 7991: normalize, if you can store any bit pattern in one and retrieve it
! 7992: unchanged without a trap, then any machine mode may go in a floating
! 7993: register and this macro should say so.
! 7994:
! 7995: The primary significance of special floating registers is rather that
! 7996: they are the registers acceptable in floating point arithmetic
! 7997: instructions. However, this is of no concern to
! 7998: @code{HARD_REGNO_MODE_OK}. You handle it by writing the proper
! 7999: constraints for those instructions.
! 8000:
! 8001: On some machines, the floating registers are especially slow to access,
! 8002: so that it is better to store a value in a stack frame than in such a
! 8003: register if floating point arithmetic is not being done. As long as the
! 8004: floating registers are not in class @code{GENERAL_REGS}, they will not
! 8005: be used unless some insn's constraint asks for one.
1.1 root 8006:
8007: @item MODES_TIEABLE_P (@var{mode1}, @var{mode2})
8008: A C expression that is nonzero if it is desirable to choose register
8009: allocation so as to avoid move instructions between a value of mode
8010: @var{mode1} and a value of mode @var{mode2}.
8011:
8012: If @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode1})} and
8013: @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode2})} are ever different
8014: for any @var{r}, then @code{MODES_TIEABLE_P (@var{mode1},
8015: @var{mode2})} must be zero.
8016:
8017: @item PC_REGNUM
8018: If the program counter has a register number, define this as that
8019: register number. Otherwise, do not define it.
8020:
8021: @item STACK_POINTER_REGNUM
8022: The register number of the stack pointer register, which must also be
8023: a fixed register according to @code{FIXED_REGISTERS}. On many
8024: machines, the hardware determines which register this is.
8025:
8026: @item FRAME_POINTER_REGNUM
8027: The register number of the frame pointer register, which is used to
8028: access automatic variables in the stack frame. On some machines, the
8029: hardware determines which register this is. On other machines, you
8030: can choose any register you wish for this purpose.
8031:
8032: @item FRAME_POINTER_REQUIRED
1.1.1.9 ! root 8033: A C expression which is nonzero if a function must have and use a frame
! 8034: pointer. This expression is evaluated twice: at the beginning of
! 8035: generating RTL, and in the reload pass. If its value is nonzero at
! 8036: either time, then the function will have a frame pointer.
! 8037:
! 8038: The expression can in principle examine the current function and decide
! 8039: according to the facts, but on most machines the constant 0 or the
! 8040: constant 1 suffices. Use 0 when the machine allows code to be generated
! 8041: with no frame pointer, and doing so saves some time or space. Use 1
! 8042: when there is no possible advantage to avoiding a frame pointer.
1.1 root 8043:
1.1.1.5 root 8044: In certain cases, the compiler does not know how to produce valid code
8045: without a frame pointer. The compiler recognizes those cases and
8046: automatically gives the function a frame pointer regardless of what
1.1 root 8047: @code{FRAME_POINTER_REQUIRED} says. You don't need to worry about
8048: them.@refill
8049:
8050: In a function that does not require a frame pointer, the frame pointer
8051: register can be allocated for ordinary usage, unless you mark it as a
8052: fixed register. See @code{FIXED_REGISTERS} for more information.
8053:
8054: @item ARG_POINTER_REGNUM
8055: The register number of the arg pointer register, which is used to
8056: access the function's argument list. On some machines, this is the
8057: same as the frame pointer register. On some machines, the hardware
8058: determines which register this is. On other machines, you can choose
8059: any register you wish for this purpose. If this is not the same
8060: register as the frame pointer register, then you must mark it as a
8061: fixed register according to @code{FIXED_REGISTERS}.
8062:
8063: @item STATIC_CHAIN_REGNUM
8064: The register number used for passing a function's static chain
8065: pointer. This is needed for languages such as Pascal and Algol where
8066: functions defined within other functions can access the local
8067: variables of the outer functions; it is not currently used because C
8068: does not provide this feature, but you must define the macro.
8069:
8070: The static chain register need not be a fixed register.
8071:
8072: @item STRUCT_VALUE_REGNUM
8073: When a function's value's mode is @code{BLKmode}, the value is not
8074: returned according to @code{FUNCTION_VALUE}. Instead, the caller
8075: passes the address of a block of memory in which the value should be
8076: stored.
8077:
8078: If this value is passed in a register, then @code{STRUCT_VALUE_REGNUM}
8079: should be the number of that register.
8080:
8081: @item STRUCT_VALUE
8082: If the structure value address is not passed in a register, define
8083: @code{STRUCT_VALUE} as an expression returning an RTX for the place
1.1.1.8 root 8084: where the address is passed. If it returns a @code{mem} RTX, the
1.1 root 8085: address is passed as an ``invisible'' first argument.
8086:
8087: @item STRUCT_VALUE_INCOMING_REGNUM
8088: On some architectures the place where the structure value address
8089: is found by the called function is not the same place that the
8090: caller put it. This can be due to register windows, or it could
8091: be because the function prologue moves it to a different place.
8092:
8093: If the incoming location of the structure value address is in a
8094: register, define this macro as the register number.
8095:
8096: @item STRUCT_VALUE_INCOMING
8097: If the incoming location is not a register, define
8098: @code{STRUCT_VALUE_INCOMING} as an expression for an RTX for where the
8099: called function should find the value. If it should find the value on
1.1.1.8 root 8100: the stack, define this to create a @code{mem} which refers to the
8101: frame pointer. If the value is a @code{mem}, the compiler assumes it
1.1 root 8102: is for an invisible first argument, and leaves space for it when
8103: finding the first real argument.
8104:
8105: @item REG_ALLOC_ORDER
8106: If defined, an initializer for a vector of integers, containing the
8107: numbers of hard registers in the order in which the GNU CC should
8108: prefer to use them (from most preferred to least).
8109:
8110: If this macro is not defined, registers are used lowest numbered first
8111: (all else being equal).
8112:
8113: One use of this macro is on the 360, where the highest numbered
8114: registers must always be saved and the save-multiple-registers
8115: instruction supports only sequences of consecutive registers. This
8116: macro is defined to cause the highest numbered allocatable registers
8117: to be used first.
8118: @end table
8119:
8120: @node Register Classes, Stack Layout, Registers, Machine Macros
8121: @section Register Classes
8122:
8123: On many machines, the numbered registers are not all equivalent.
8124: For example, certain registers may not be allowed for indexed addressing;
8125: certain registers may not be allowed in some instructions. These machine
8126: restrictions are described to the compiler using @dfn{register classes}.
8127:
8128: You define a number of register classes, giving each one a name and saying
8129: which of the registers belong to it. Then you can specify register classes
8130: that are allowed as operands to particular instruction patterns.
8131:
8132: In general, each register will belong to several classes. In fact, one
8133: class must be named @code{ALL_REGS} and contain all the registers. Another
8134: class must be named @code{NO_REGS} and contain no registers. Often the
8135: union of two classes will be another class; however, this is not required.
8136:
8137: One of the classes must be named @code{GENERAL_REGS}. There is nothing
8138: terribly special about the name, but the operand constraint letters
8139: @samp{r} and @samp{g} specify this class. If @code{GENERAL_REGS} is
8140: the same as @code{ALL_REGS}, just define it as a macro which expands
8141: to @code{ALL_REGS}.
8142:
8143: The way classes other than @code{GENERAL_REGS} are specified in operand
8144: constraints is through machine-dependent operand constraint letters.
8145: You can define such letters to correspond to various classes, then use
8146: them in operand constraints.
8147:
8148: You should define a class for the union of two classes whenever some
8149: instruction allows both classes. For example, if an instruction allows
8150: either a floating-point (coprocessor) register or a general register for a
8151: certain operand, you should define a class @code{FLOAT_OR_GENERAL_REGS}
8152: which includes both of them. Otherwise you will get suboptimal code.
8153:
8154: You must also specify certain redundant information about the register
8155: classes: for each class, which classes contain it and which ones are
8156: contained in it; for each pair of classes, the largest class contained
8157: in their union.
8158:
1.1.1.8 root 8159: When a value occupying several consecutive registers is expected in a
8160: certain class, all the registers used must belong to that class.
8161: Therefore, register classes cannot be used to enforce a requirement for
8162: a register pair to start with an even-numbered register. The way to
8163: specify this requirement is with @code{HARD_REGNO_MODE_OK}.
8164:
1.1 root 8165: Register classes used for input-operands of bitwise-and or shift
8166: instructions have a special requirement: each such class must have, for
8167: each fixed-point machine mode, a subclass whose registers can transfer that
8168: mode to or from memory. For example, on some machines, the operations for
8169: single-byte values (@code{QImode}) are limited to certain registers. When
8170: this is so, each register class that is used in a bitwise-and or shift
8171: instruction must have a subclass consisting of registers from which
8172: single-byte values can be loaded or stored. This is so that
8173: @code{PREFERRED_RELOAD_CLASS} can always have a possible value to return.
8174:
8175: @table @code
8176: @item enum reg_class
8177: An enumeral type that must be defined with all the register class names
8178: as enumeral values. @code{NO_REGS} must be first. @code{ALL_REGS}
8179: must be the last register class, followed by one more enumeral value,
8180: @code{LIM_REG_CLASSES}, which is not a register class but rather
8181: tells how many classes there are.
8182:
8183: Each register class has a number, which is the value of casting
8184: the class name to type @code{int}. The number serves as an index
8185: in many of the tables described below.
8186:
8187: @item N_REG_CLASSES
8188: The number of distinct register classes, defined as follows:
8189:
8190: @example
8191: #define N_REG_CLASSES (int) LIM_REG_CLASSES
8192: @end example
8193:
8194: @item REG_CLASS_NAMES
8195: An initializer containing the names of the register classes as C string
8196: constants. These names are used in writing some of the debugging dumps.
8197:
8198: @item REG_CLASS_CONTENTS
8199: An initializer containing the contents of the register classes, as integers
8200: which are bit masks. The @var{n}th integer specifies the contents of class
8201: @var{n}. The way the integer @var{mask} is interpreted is that
8202: register @var{r} is in the class if @code{@var{mask} & (1 << @var{r})} is 1.
8203:
8204: When the machine has more than 32 registers, an integer does not suffice.
8205: Then the integers are replaced by sub-initializers, braced groupings containing
8206: several integers. Each sub-initializer must be suitable as an initializer
8207: for the type @code{HARD_REG_SET} which is defined in @file{hard-reg-set.h}.
8208:
8209: @item REGNO_REG_CLASS (@var{regno})
8210: A C expression whose value is a register class containing hard register
8211: @var{regno}. In general there is more that one such class; choose a class
8212: which is @dfn{minimal}, meaning that no smaller class also contains the
8213: register.
8214:
8215: @item BASE_REG_CLASS
8216: A macro whose definition is the name of the class to which a valid
8217: base register must belong. A base register is one used in an address
8218: which is the register value plus a displacement.
8219:
8220: @item INDEX_REG_CLASS
8221: A macro whose definition is the name of the class to which a valid
8222: index register must belong. An index register is one used in an
8223: address where its value is either multiplied by a scale factor or
8224: added to another register (as well as added to a displacement).
8225:
8226: @item REG_CLASS_FROM_LETTER (@var{char})
8227: A C expression which defines the machine-dependent operand constraint
8228: letters for register classes. If @var{char} is such a letter, the
8229: value should be the register class corresponding to it. Otherwise,
8230: the value should be @code{NO_REGS}.
8231:
8232: @item REGNO_OK_FOR_BASE_P (@var{num})
8233: A C expression which is nonzero if register number @var{num} is
8234: suitable for use as a base register in operand addresses. It may be
8235: either a suitable hard register or a pseudo register that has been
8236: allocated such a hard register.
8237:
8238: @item REGNO_OK_FOR_INDEX_P (@var{num})
8239: A C expression which is nonzero if register number @var{num} is
8240: suitable for use as an index register in operand addresses. It may be
8241: either a suitable hard register or a pseudo register that has been
8242: allocated such a hard register.
8243:
8244: The difference between an index register and a base register is that
8245: the index register may be scaled. If an address involves the sum of
8246: two registers, neither one of them scaled, then either one may be
8247: labeled the ``base'' and the other the ``index''; but whichever
8248: labeling is used must fit the machine's constraints of which registers
8249: may serve in each capacity. The compiler will try both labelings,
8250: looking for one that is valid, and will reload one or both registers
8251: only if neither labeling works.
8252:
8253: @item PREFERRED_RELOAD_CLASS (@var{x}, @var{class})
8254: A C expression that places additional restrictions on the register class
8255: to use when it is necessary to copy value @var{x} into a register in class
8256: @var{class}. The value is a register class; perhaps @var{class}, or perhaps
8257: another, smaller class. On many machines, the definition
8258:
8259: @example
8260: #define PREFERRED_RELOAD_CLASS(X,CLASS) CLASS
8261: @end example
8262:
8263: @noindent
8264: is safe.
8265:
8266: Sometimes returning a more restrictive class makes better code. For
8267: example, on the 68000, when @var{x} is an integer constant that is in range
8268: for a @samp{moveq} instruction, the value of this macro is always
8269: @code{DATA_REGS} as long as @var{class} includes the data registers.
8270: Requiring a data register guarantees that a @samp{moveq} will be used.
8271:
1.1.1.8 root 8272: If @var{x} is a @code{const_double}, by returning @code{NO_REGS}
1.1 root 8273: you can force @var{x} into a memory constant. This is useful on
8274: certain machines where immediate floating values cannot be loaded into
8275: certain kinds of registers.
8276:
8277: In a shift instruction or a bitwise-and instruction, the mode of @var{x},
8278: the value being reloaded, may not be the same as the mode of the
8279: instruction's operand. (They will both be fixed-point modes, however.) In
8280: such a case, @var{class} may not be a safe value to return. @var{class} is
8281: certainly valid for the instruction, but it may not be valid for reloading
8282: @var{x}. This problem can occur on machines such as the 68000 and 80386
8283: where some registers can handle full-word values but cannot handle
8284: single-byte values.
8285:
8286: On such machines, this macro must examine the mode of @var{x} and return a
8287: subclass of @var{class} which can handle loads and stores of that mode. On
8288: the 68000, where address registers cannot handle @code{QImode}, if @var{x}
8289: has @code{QImode} then you must return @code{DATA_REGS}. If @var{class} is
8290: @code{ADDR_REGS}, then there is no correct value to return; but the shift
8291: and bitwise-and instructions don't use @code{ADDR_REGS}, so this fatal case
8292: never arises.
8293:
8294: @item CLASS_MAX_NREGS (@var{class}, @var{mode})
8295: A C expression for the maximum number of consecutive registers
8296: of class @var{class} needed to hold a value of mode @var{mode}.
8297:
8298: This is closely related to the macro @code{HARD_REGNO_NREGS}.
8299: In fact, the value of the macro @code{CLASS_MAX_NREGS (@var{class}, @var{mode})}
8300: should be the maximum value of @code{HARD_REGNO_NREGS (@var{regno}, @var{mode})}
8301: for all @var{regno} values in the class @var{class}.
8302:
8303: This macro helps control the handling of multiple-word values
8304: in the reload pass.
8305: @end table
8306:
8307: Two other special macros describe which constants fit which constraint
8308: letters.
8309:
8310: @table @code
8311: @item CONST_OK_FOR_LETTER_P (@var{value}, @var{c})
8312: A C expression that defines the machine-dependent operand constraint letters
8313: that specify particular ranges of integer values. If @var{c} is one
8314: of those letters, the expression should check that @var{value}, an integer,
8315: is in the appropriate range and return 1 if so, 0 otherwise. If @var{c} is
8316: not one of those letters, the value should be 0 regardless of @var{value}.
8317:
8318: @item CONST_DOUBLE_OK_FOR_LETTER_P (@var{value}, @var{c})
8319: A C expression that defines the machine-dependent operand constraint
8320: letters that specify particular ranges of floating values. If @var{c} is
8321: one of those letters, the expression should check that @var{value}, an RTX
1.1.1.8 root 8322: of code @code{const_double}, is in the appropriate range and return 1 if
1.1 root 8323: so, 0 otherwise. If @var{c} is not one of those letters, the value should
8324: be 0 regardless of @var{value}.
8325: @end table
8326:
8327: @node Stack Layout, Library Names, Register Classes, Machine Macros
8328: @section Describing Stack Layout
8329:
8330: @table @code
8331: @item STACK_GROWS_DOWNWARD
8332: Define this macro if pushing a word onto the stack moves the stack
8333: pointer to a smaller address.
8334:
8335: When we say, ``define this macro if @dots{},'' it means that the
8336: compiler checks this macro only with @code{#ifdef} so the precise
8337: definition used does not matter.
8338:
8339: @item FRAME_GROWS_DOWNWARD
8340: Define this macro if the addresses of local variable slots are at negative
8341: offsets from the frame pointer.
8342:
8343: @item STARTING_FRAME_OFFSET
8344: Offset from the frame pointer to the first local variable slot to be allocated.
8345:
8346: If @code{FRAME_GROWS_DOWNWARD}, the next slot's offset is found by
8347: subtracting the length of the first slot from @code{STARTING_FRAME_OFFSET}.
8348: Otherwise, it is found by adding the length of the first slot to
8349: the value @code{STARTING_FRAME_OFFSET}.
8350:
8351: @item PUSH_ROUNDING (@var{npushed})
8352: A C expression that is the number of bytes actually pushed onto the
8353: stack when an instruction attempts to push @var{npushed} bytes.
8354:
8355: If the target machine does not have a push instruction, do not define
8356: this macro. That directs GNU CC to use an alternate strategy: to
8357: allocate the entire argument block and then store the arguments into
8358: it.
8359:
8360: On some machines, the definition
8361:
8362: @example
8363: #define PUSH_ROUNDING(BYTES) (BYTES)
8364: @end example
8365:
8366: @noindent
8367: will suffice. But on other machines, instructions that appear
8368: to push one byte actually push two bytes in an attempt to maintain
8369: alignment. Then the definition should be
8370:
8371: @example
8372: #define PUSH_ROUNDING(BYTES) (((BYTES) + 1) & ~1)
8373: @end example
8374:
8375: @item FIRST_PARM_OFFSET (@var{fundecl})
8376: Offset from the argument pointer register to the first argument's
8377: address. On some machines it may depend on the data type of the
8378: function. (In the next version of GNU CC, the argument will be
8379: changed to the function data type rather than its declaration.)
8380:
8381: @item FIRST_PARM_CALLER_OFFSET (@var{fundecl})
8382: Define this macro on machines where register parameters have shadow
8383: locations on the stack, at addresses below the nominal parameter.
8384: This matters because certain arguments cannot be passed on the stack.
8385: On these machines, such arguments must be stored into the shadow
8386: locations.
8387:
8388: This macro should expand into a C expression whose value is the offset
8389: of the first parameter's shadow location from the nominal stack
8390: pointer value. (That value is itself computed by adding the value of
8391: @code{STACK_POINTER_OFFSET} to the stack pointer register.)
8392:
1.1.1.9 ! root 8393: @item REG_PARM_STACK_SPACE
! 8394: Define this macro if functions should assume that stack space has been
! 8395: allocated for arguments even when their values are passed in
! 8396: registers.
! 8397:
! 8398: The actual allocation of such space would be done either by
! 8399: the call instruction or by the function prologue, or by
! 8400: defining FIRST_PARM_CALLER_OFFSET.
! 8401:
1.1.1.6 root 8402: @item STACK_ARGS_ADJUST (@var{size})
8403: Define this macro if the machine requires padding on the stack for
8404: certain function calls. This is padding on a per-function-call basis,
8405: not padding for individual arguments.
8406:
1.1.1.7 root 8407: The argument @var{size} will be a C variable of type @code{struct
8408: arg_data} which contains two fields, an integer named @code{constant}
8409: and an RTX named @code{var}. These together represent a size measured
8410: in bytes which is the sum of the integer and the RTX. Most of the
8411: time @code{var} is 0, which means that the size is simply the integer.
8412:
8413: The definition should be a C statement or compound statement
8414: which alters the variable supplied in whatever way you wish.
8415:
8416: Note that the value you leave in the variable @code{size} will
8417: ultimately be rounded up to a multiple of @code{STACK_BOUNDARY} bits.
8418:
8419: This macro is not fully implemented for machines which have push
8420: instructions (i.e., on which @code{PUSH_ROUNDING} is defined).
1.1.1.6 root 8421:
1.1 root 8422: @item RETURN_POPS_ARGS (@var{funtype})
8423: A C expression that should be 1 if a function pops its own arguments
8424: on returning, or 0 if the function pops no arguments and the caller
8425: must therefore pop them all after the function returns.
8426:
8427: @var{funtype} is a C variable whose value is a tree node that
8428: describes the function in question. Normally it is a node of type
8429: @code{FUNCTION_TYPE} that describes the data type of the function.
8430: From this it is possible to obtain the data types of the value and
8431: arguments (if known).
8432:
8433: When a call to a library function is being considered, @var{funtype}
8434: will contain an identifier node for the library function. Thus, if
8435: you need to distinguish among various library functions, you can do so
8436: by their names. Note that ``library function'' in this context means
8437: a function used to perform arithmetic, whose name is known specially
8438: in the compiler and was not mentioned in the C code being compiled.
8439:
8440: On the Vax, all functions always pop their arguments, so the
8441: definition of this macro is 1. On the 68000, using the standard
8442: calling convention, no functions pop their arguments, so the value of
8443: the macro is always 0 in this case. But an alternative calling
8444: convention is available in which functions that take a fixed number of
8445: arguments pop them but other functions (such as @code{printf}) pop
8446: nothing (the caller pops all). When this convention is in use,
8447: @var{funtype} is examined to determine whether a function takes a
8448: fixed number of arguments.
8449:
8450: @item FUNCTION_VALUE (@var{valtype}, @var{func})
8451: A C expression to create an RTX representing the place where a
8452: function returns a value of data type @var{valtype}. @var{valtype} is
8453: a tree node representing a data type. Write @code{TYPE_MODE
8454: (@var{valtype})} to get the machine mode used to represent that type.
8455: On many machines, only the mode is relevant. (Actually, on most
8456: machines, scalar values are returned in the same place regardless of
8457: mode).@refill
8458:
8459: If the precise function being called is known, @var{func} is a tree
8460: node (@code{FUNCTION_DECL}) for it; otherwise, @var{func} is a null
8461: pointer. This makes it possible to use a different value-returning
8462: convention for specific functions when all their calls are
8463: known.@refill
8464:
8465: @item FUNCTION_OUTGOING_VALUE (@var{valtype}, @var{func})
8466: Define this macro if the target machine has ``register windows''
8467: so that the register in which a function returns its value is not
8468: the same as the one in which the caller sees the value.
8469:
8470: For such machines, @code{FUNCTION_VALUE} computes the register in
8471: which the caller will see the value, and
8472: @code{FUNCTION_OUTGOING_VALUE} should be defined in a similar fashion
8473: to tell the function where to put the value.@refill
8474:
8475: If @code{FUNCTION_OUTGOING_VALUE} is not defined,
8476: @code{FUNCTION_VALUE} serves both purposes.@refill
8477:
1.1.1.7 root 8478: @item RETURN_IN_MEMORY (@var{type})
8479: A C expression which can inhibit the returning of certain function
8480: values in registers, based on the type of value. A nonzero value says
8481: to return the function value in memory, just as large structures are
8482: always returned. Here @var{type} will be a C expression of type
8483: @code{tree}, representing the data type of the value.
8484:
8485: Note that values of mode @code{BLKmode} are returned in memory
8486: regardless of this macro. Also, the option @samp{-fpcc-struct-return}
8487: takes effect regardless of this macro. On most systems, it is
8488: possible to leave the macro undefined; this causes a default
8489: definition to be used, whose value is the constant 0.
8490:
1.1 root 8491: @item LIBCALL_VALUE (@var{mode})
8492: A C expression to create an RTX representing the place where a library
8493: function returns a value of mode @var{mode}. If the precise function
8494: being called is known, @var{func} is a tree node
8495: (@code{FUNCTION_DECL}) for it; otherwise, @var{func} is a null
8496: pointer. This makes it possible to use a different value-returning
8497: convention for specific functions when all their calls are
8498: known.@refill
8499:
8500: Note that ``library function'' in this context means a compiler
8501: support routine, used to perform arithmetic, whose name is known
8502: specially by the compiler and was not mentioned in the C code being
8503: compiled.
8504:
8505: @item FUNCTION_VALUE_REGNO_P (@var{regno})
8506: A C expression that is nonzero if @var{regno} is the number of a hard
8507: register in which the values of called function may come back.
8508:
8509: A register whose use for returning values is limited to serving as the
8510: second of a pair (for a value of type @code{double}, say) need not be
8511: recognized by this macro. So for most machines, this definition
8512: suffices:
8513:
8514: @example
8515: #define FUNCTION_VALUE_REGNO_P(N) ((N) == 0)
8516: @end example
8517:
8518: If the machine has register windows, so that the caller and the called
8519: function use different registers for the return value, this macro
8520: should recognize only the caller's register numbers.
8521:
8522: @item FUNCTION_ARG (@var{cum}, @var{mode}, @var{type}, @var{named})
8523: A C expression that controls whether a function argument is passed
8524: in a register, and which register.
8525:
8526: The arguments are @var{cum}, which summarizes all the previous
8527: arguments; @var{mode}, the machine mode of the argument; @var{type},
8528: the data type of the argument as a tree node or 0 if that is not known
8529: (which happens for C support library functions); and @var{named},
8530: which is 1 for an ordinary argument and 0 for nameless arguments that
1.1.1.8 root 8531: correspond to @samp{@dots{}} in the called function's prototype.
1.1 root 8532:
1.1.1.8 root 8533: The value of the expression should either be a @code{reg} RTX for the
1.1 root 8534: hard register in which to pass the argument, or zero to pass the
8535: argument on the stack.
8536:
8537: For the Vax and 68000, where normally all arguments are pushed, zero
8538: suffices as a definition.
8539:
1.1.1.8 root 8540: The usual way to make the ANSI library @file{stdarg.h} work on a machine
8541: where some arguments are usually passed in registers, is to cause
8542: nameless arguments to be passed on the stack instead. This is done
8543: by making @code{FUNCTION_ARG} return 0 whenever @var{named} is 0.
8544:
1.1 root 8545: @item FUNCTION_INCOMING_ARG (@var{cum}, @var{mode}, @var{type}, @var{named})
8546: Define this macro if the target machine has ``register windows'', so
8547: that the register in which a function sees an arguments is not
8548: necessarily the same as the one in which the caller passed the
8549: argument.
8550:
8551: For such machines, @code{FUNCTION_ARG} computes the register in which
8552: the caller passes the value, and @code{FUNCTION_INCOMING_ARG} should
8553: be defined in a similar fashion to tell the function being called
8554: where the arguments will arrive.
8555:
8556: If @code{FUNCTION_INCOMING_ARG} is not defined, @code{FUNCTION_ARG}
8557: serves both purposes.@refill
8558:
8559: @item FUNCTION_ARG_PARTIAL_NREGS (@var{cum}, @var{mode}, @var{type}, @var{named})
8560: A C expression for the number of words, at the beginning of an
8561: argument, must be put in registers. The value must be zero for
8562: arguments that are passed entirely in registers or that are entirely
8563: pushed on the stack.
8564:
8565: On some machines, certain arguments must be passed partially in
8566: registers and partially in memory. On these machines, typically the
8567: first @var{n} words of arguments are passed in registers, and the rest
8568: on the stack. If a multi-word argument (a @code{double} or a
8569: structure) crosses that boundary, its first few words must be passed
8570: in registers and the rest must be pushed. This macro tells the
8571: compiler when this occurs, and how many of the words should go in
8572: registers.
8573:
8574: @code{FUNCTION_ARG} for these arguments should return the first
8575: register to be used by the caller for this argument; likewise
8576: @code{FUNCTION_INCOMING_ARG}, for the called function.
8577:
8578: @item CUMULATIVE_ARGS
8579: A C type for declaring a variable that is used as the first argument
8580: of @code{FUNCTION_ARG} and other related values. For some target
8581: machines, the type @code{int} suffices and can hold the number of
8582: bytes of argument so far.
8583:
8584: @item INIT_CUMULATIVE_ARGS (@var{cum}, @var{fntype})
8585: A C statement (sans semicolon) for initializing the variable @var{cum}
8586: for the state at the beginning of the argument list. The variable has
8587: type @code{CUMULATIVE_ARGS}. The value of @var{fntype} is the tree node
8588: for the data type of the function which will receive the args, or 0
8589: if the args are to a compiler support library function.
8590:
8591: @item FUNCTION_ARG_ADVANCE (@var{cum}, @var{mode}, @var{type}, @var{named})
1.1.1.7 root 8592: A C statement (sans semicolon) to update the summarizer variable
8593: @var{cum} to advance past an argument in the argument list. The
8594: values @var{mode}, @var{type} and @var{named} describe that argument.
8595: Once this is done, the variable @var{cum} is suitable for analyzing
8596: the @emph{following} argument with @code{FUNCTION_ARG}, etc.@refill
1.1 root 8597:
8598: @item FUNCTION_ARG_REGNO_P (@var{regno})
8599: A C expression that is nonzero if @var{regno} is the number of a hard
8600: register in which function arguments are sometimes passed. This does
8601: @emph{not} include implicit arguments such as the static chain and
8602: the structure-value address. On many machines, no registers can be
8603: used for this purpose since all function arguments are pushed on the
8604: stack.
8605:
8606: @item FUNCTION_ARG_PADDING (@var{mode}, @var{size})
8607: If defined, a C expression which determines whether, and in which direction,
8608: to pad out an argument with extra space. The value should be of type
8609: @code{enum direction}: either @code{upward} to pad above the argument,
8610: @code{downward} to pad below, or @code{none} to inhibit padding.
8611:
8612: The argument @var{size} is an RTX which describes the size of the
8613: argument, in bytes. It should be used only if @var{mode} is
8614: @code{BLKmode}. Otherwise, @var{size} is 0.
8615:
8616: This macro does not control the @emph{amount} of padding; that is
8617: always just enough to reach the next multiple of @code{PARM_BOUNDARY}.
8618:
8619: This macro has a default definition which is right for most systems.
8620: For little-endian machines, the default is to pad upward. For
8621: big-endian machines, the default is to pad downward for an argument of
8622: constant size shorter than an @code{int}, and upward otherwise.
8623:
8624: @item FUNCTION_PROLOGUE (@var{file}, @var{size})
8625: A C compound statement that outputs the assembler code for entry to a
8626: function. The prologue is responsible for setting up the stack frame,
8627: initializing the frame pointer register, saving registers that must be
8628: saved, and allocating @var{size} additional bytes of storage for the
8629: local variables. @var{size} is an integer. @var{file} is a stdio
8630: stream to which the assembler code should be output.
8631:
8632: The label for the beginning of the function need not be output by this
8633: macro. That has already been done when the macro is run.
8634:
8635: To determine which registers to save, the macro can refer to the array
8636: @code{regs_ever_live}: element @var{r} is nonzero if hard register
8637: @var{r} is used anywhere within the function. This implies the
8638: function prologue should save register @var{r}, but not if it is one
8639: of the call-used registers.
8640:
8641: On machines where functions may or may not have frame-pointers, the
8642: function entry code must vary accordingly; it must set up the frame
8643: pointer if one is wanted, and not otherwise. To determine whether a
8644: frame pointer is in wanted, the macro can refer to the variable
8645: @code{frame_pointer_needed}. The variable's value will be 1 at run
8646: time in a function that needs a frame pointer.
8647:
1.1.1.8 root 8648: On machines where arguments may be passed in registers, and not have
8649: stack space allocated, this macro must examine the variable
8650: @code{current_function_pretend_args_size}, and allocate that many bytes
8651: of uninitialized space on the stack just underneath the first argument
8652: arriving on the stack. (This may not be at the very end of the stack,
8653: if the calling sequence has pushed anything else since pushing the stack
8654: arguments. But usually, on such machines, nothing else has been pushed
8655: yet, because the function prologue itself does all the pushing.)
8656:
8657: This ``pretend argument'' space is allocated in functions that use the
8658: ANSI library @file{stdarg.h} to accept anonymous arguments of
8659: unspecified types; the last named argument is copied into the space, so
8660: that the anonymous arguments follow it consecutively.
8661:
1.1 root 8662: @item FUNCTION_PROFILER (@var{file}, @var{labelno})
8663: A C statement or compound statement to output to @var{file} some
8664: assembler code to call the profiling subroutine @code{mcount}.
8665: Before calling, the assembler code must load the address of a
8666: counter variable into a register where @code{mcount} expects to
8667: find the address. The name of this variable is @samp{LP} followed
8668: by the number @var{labelno}, so you would generate the name using
8669: @samp{LP%d} in a @code{fprintf}.
8670:
8671: The details of how the address should be passed to @code{mcount} are
8672: determined by your operating system environment, not by GNU CC. To
8673: figure them out, compile a small program for profiling using the
8674: system's installed C compiler and look at the assembler code that
8675: results.
8676:
1.1.1.6 root 8677: @item FUNCTION_BLOCK_PROFILER (@var{file}, @var{labelno})
8678: A C statement or compound statement to output to @var{file} some
8679: assembler code to initialize basic-block profiling for the current
8680: object module. This code should call the subroutine
8681: @code{__bb_init_func} once per object module, passing it as its sole
8682: argument the address of a block allocated in the object module.
8683:
8684: The name of the block is a local symbol made with this statement:
8685:
8686: @example
8687: ASM_GENERATE_INTERNAL_LABEL (@var{buffer}, "LPBX", 0);
8688: @end example
8689:
8690: Of course, since you are writing the definition of
8691: @code{ASM_GENERATE_INTERNAL_LABEL} as well as that of this macro, you
8692: can take a short cut in the definition of this macro and use the name
8693: that you know will result.
8694:
8695: The first word of this block is a flag which will be nonzero if the
8696: object module has already been initialized. So test this word first,
8697: and do not call @code{__bb_init_func} if the flag is nonzero.
8698:
8699: @item BLOCK_PROFILER (@var{file}, @var{blockno})
8700: A C statement or compound statement to increment the count associated
8701: with the basic block number @var{blockno}. Basic blocks are numbered
8702: separately from zero within each compilation. The count associated
8703: with block number @var{blockno} is at index @var{blockno} in a vector
8704: of words; the name of this array is a local symbol made with this
8705: statement:
8706:
8707: @example
8708: ASM_GENERATE_INTERNAL_LABEL (@var{buffer}, "LPBX", 2);
8709: @end example
8710:
8711: Of course, since you are writing the definition of
8712: @code{ASM_GENERATE_INTERNAL_LABEL} as well as that of this macro, you
8713: can take a short cut in the definition of this macro and use the name
8714: that you know will result.
8715:
1.1 root 8716: @item EXIT_IGNORES_STACK
8717: Define this macro as a C expression that is nonzero if the return
8718: instruction or the function epilogue ignores the value of the stack
8719: pointer; in other words, if it is safe to delete an instruction to
8720: adjust the stack pointer before a return from the function.
8721:
1.1.1.8 root 8722: Note that this macro's value is relevant only for functions for which
8723: frame pointers are maintained. It is never safe to delete a final
8724: stack adjustment in a function that has no frame pointer, and the
8725: compiler knows this regardless of @code{EXIT_IGNORES_STACK}.
1.1 root 8726:
8727: @item FUNCTION_EPILOGUE (@var{file}, @var{size})
8728: A C compound statement that outputs the assembler code for exit from a
8729: function. The epilogue is responsible for restoring the saved
8730: registers and stack pointer to their values when the function was
8731: called, and returning control to the caller. This macro takes the
8732: same arguments as the macro @code{FUNCTION_PROLOGUE}, and the
8733: registers to restore are determined from @code{regs_ever_live} and
8734: @code{CALL_USED_REGISTERS} in the same way.
8735:
8736: On some machines, there is a single instruction that does all the work
8737: of returning from the function. On these machines, give that
8738: instruction the name @samp{return} and do not define the macro
8739: @code{FUNCTION_EPILOGUE} at all.
8740:
8741: Do not define a pattern named @samp{return} if you want the
8742: @code{FUNCTION_EPILOGUE} to be used. If you want the target switches
8743: to control whether return instructions or epilogues are used, define a
8744: @samp{return} pattern with a validity condition that tests the target
8745: switches appropriately. If the @samp{return} pattern's validity
8746: condition is false, epilogues will be used.
8747:
8748: On machines where functions may or may not have frame-pointers, the
8749: function exit code must vary accordingly. Sometimes the code for
8750: these two cases is completely different. To determine whether a frame
8751: pointer is in wanted, the macro can refer to the variable
8752: @code{frame_pointer_needed}. The variable's value will be 1 at run
8753: time in a function that needs a frame pointer.
8754:
8755: On some machines, some functions pop their arguments on exit while
8756: others leave that for the caller to do. For example, the 68020 when
8757: given @samp{-mrtd} pops arguments in functions that take a fixed
8758: number of arguments.
8759:
8760: Your definition of the macro @code{RETURN_POPS_ARGS} decides which
8761: functions pop their own arguments. @code{FUNCTION_EPILOGUE} needs to
8762: know what was decided. The variable @code{current_function_pops_args}
8763: is nonzero if the function should pop its own arguments. If so, use
8764: the variable @code{current_function_args_size} as the number of bytes
8765: to pop.
8766:
8767: @item FIX_FRAME_POINTER_ADDRESS (@var{addr}, @var{depth})
8768: A C compound statement to alter a memory address that uses the frame
8769: pointer register so that it uses the stack pointer register instead.
8770: This must be done in the instructions that load parameter values into
8771: registers, when the reload pass determines that a frame pointer is not
8772: necessary for the function. @var{addr} will be a C variable name, and
8773: the updated address should be stored in that variable. @var{depth}
8774: will be the current depth of stack temporaries (number of bytes of
8775: arguments currently pushed). The change in offset between a
8776: frame-pointer-relative address and a stack-pointer-relative address
8777: must include @var{depth}.
8778:
8779: Even if your machine description specifies there will always be a
8780: frame pointer in the frame pointer register, you must still define
8781: @code{FIX_FRAME_POINTER_ADDRESS}, but the definition will never be
8782: executed at run time, so it may be empty.
1.1.1.8 root 8783:
8784: @item LONGJMP_RESTORE_FROM_STACK
8785: Define this macro if the @code{longjmp} function restores registers
8786: from the stack frames, rather than from those saved specifically by
8787: @code{setjmp}. Certain quantities must not be kept in registers
8788: across a call to @code{setjmp} on such machines.
1.1 root 8789: @end table
8790:
8791: @node Library Names, Addressing Modes, Stack Layout, Machine Macros
8792: @section Library Subroutine Names
8793:
8794: @table @code
1.1.1.5 root 8795: @item MULSI3_LIBCALL
8796: A C string constant giving the name of the function to call for
8797: multiplication of one signed full-word by another. If you do not
8798: define this macro, the default name is used, which is @code{__mulsi3},
8799: a function defined in @file{gnulib}.
8800:
8801: @item UMULSI3_LIBCALL
8802: A C string constant giving the name of the function to call for
8803: multiplication of one unsigned full-word by another. If you do not
8804: define this macro, the default name is used, which is
8805: @code{__umulsi3}, a function defined in @file{gnulib}.
8806:
8807: @item DIVSI3_LIBCALL
8808: A C string constant giving the name of the function to call for
8809: division of one signed full-word by another. If you do not define
8810: this macro, the default name is used, which is @code{__divsi3}, a
8811: function defined in @file{gnulib}.
8812:
1.1 root 8813: @item UDIVSI3_LIBCALL
8814: A C string constant giving the name of the function to call for
1.1.1.5 root 8815: division of one unsigned full-word by another. If you do not define
8816: this macro, the default name is used, which is @code{__udivsi3}, a
8817: function defined in @file{gnulib}.
8818:
8819: @item MODSI3_LIBCALL
8820: A C string constant giving the name of the function to call for the
8821: remainder in division of one signed full-word by another. If you do
8822: not define this macro, the default name is used, which is
8823: @code{__modsi3}, a function defined in @file{gnulib}.
1.1 root 8824:
8825: @item UMODSI3_LIBCALL
8826: A C string constant giving the name of the function to call for the
1.1.1.5 root 8827: remainder in division of one unsigned full-word by another. If you do
8828: not define this macro, the default name is used, which is
8829: @code{__umodsi3}, a function defined in @file{gnulib}.
1.1 root 8830:
8831: @item TARGET_MEM_FUNCTIONS
8832: Define this macro if GNU CC should generate calls to the System V
8833: (and ANSI C) library functions @code{memcpy} and @code{memset}
8834: rather than the BSD functions @code{bcopy} and @code{bzero}.
8835: @end table
8836:
1.1.1.8 root 8837: @node Addressing Modes, Delayed Branch, Library Names, Machine Macros
1.1 root 8838: @section Addressing Modes
8839:
8840: @table @code
8841: @item HAVE_POST_INCREMENT
8842: Define this macro if the machine supports post-increment addressing.
8843:
8844: @item HAVE_PRE_INCREMENT
8845: @itemx HAVE_POST_DECREMENT
8846: @itemx HAVE_PRE_DECREMENT
8847: Similar for other kinds of addressing.
8848:
8849: @item CONSTANT_ADDRESS_P (@var{x})
8850: A C expression that is 1 if the RTX @var{x} is a constant whose value
8851: is an integer. This includes integers whose values are not explicitly
1.1.1.8 root 8852: known, such as @code{symbol_ref} and @code{label_ref} expressions and
8853: @code{const} arithmetic expressions.
1.1 root 8854:
8855: On most machines, this can be defined as @code{CONSTANT_P (@var{x})},
8856: but a few machines are more restrictive in which constant addresses
8857: are supported.
8858:
8859: @item MAX_REGS_PER_ADDRESS
8860: A number, the maximum number of registers that can appear in a valid
8861: memory address.
8862:
8863: @item GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{label})
8864: A C compound statement with a conditional @code{goto @var{label};}
8865: executed if @var{x} (an RTX) is a legitimate memory address on the
8866: target machine for a memory operand of mode @var{mode}.
8867:
8868: It usually pays to define several simpler macros to serve as
8869: subroutines for this one. Otherwise it may be too complicated to
8870: understand.
8871:
8872: This macro must exist in two variants: a strict variant and a
8873: non-strict one. The strict variant is used in the reload pass. It
8874: must be defined so that any pseudo-register that has not been
8875: allocated a hard register is considered a memory reference. In
8876: contexts where some kind of register is required, a pseudo-register
8877: with no hard register must be rejected.
8878:
8879: The non-strict variant is used in other passes. It must be defined to
8880: accept all pseudo-registers in every context where some kind of
8881: register is required.
8882:
8883: Compiler source files that want to use the strict variant of this
8884: macro define the macro @code{REG_OK_STRICT}. You should use an
8885: @code{#ifdef REG_OK_STRICT} conditional to define the strict variant
8886: in that case and the non-strict variant otherwise.
8887:
8888: Typically among the subroutines used to define
8889: @code{GO_IF_LEGITIMATE_ADDRESS} are subroutines to check for
8890: acceptable registers for various purposes (one for base registers, one
8891: for index registers, and so on). Then only these subroutine macros
8892: need have two variants; the higher levels of macros may be the same
8893: whether strict or not.@refill
8894:
1.1.1.8 root 8895: Normally, constant addresses which are the sum of a @code{symbol_ref}
8896: and an integer are stored inside a @code{const} RTX to mark them as
8897: constant. Therefore, there is no need to recognize such sums as
8898: legitimate addresses.
8899:
8900: Usually @code{PRINT_OPERAND_ADDRESS} is not prepared to handle constant
8901: sums that are not marked with @code{const}. It assumes that a naked
8902: @code{plus} indicates indexing. If so, then you @emph{must} reject such
8903: naked constant sums as illegitimate addresses, so that none of them will
8904: be given to @code{PRINT_OPERAND_ADDRESS}.@refill
8905:
1.1 root 8906: @item REG_OK_FOR_BASE_P (@var{x})
1.1.1.5 root 8907: A C expression that is nonzero if @var{x} (assumed to be a @code{reg}
1.1 root 8908: RTX) is valid for use as a base register. For hard registers, it
8909: should always accept those which the hardware permits and reject the
8910: others. Whether the macro accepts or rejects pseudo registers must be
8911: controlled by @code{REG_OK_STRICT} as described above. This usually
8912: requires two variant definitions, of which @code{REG_OK_STRICT}
8913: controls the one actually used.
8914:
8915: @item REG_OK_FOR_INDEX_P (@var{x})
1.1.1.5 root 8916: A C expression that is nonzero if @var{x} (assumed to be a @code{reg}
1.1 root 8917: RTX) is valid for use as an index register.
8918:
8919: The difference between an index register and a base register is that
8920: the index register may be scaled. If an address involves the sum of
8921: two registers, neither one of them scaled, then either one may be
8922: labeled the ``base'' and the other the ``index''; but whichever
8923: labeling is used must fit the machine's constraints of which registers
8924: may serve in each capacity. The compiler will try both labelings,
8925: looking for one that is valid, and will reload one or both registers
8926: only if neither labeling works.
8927:
8928: @item LEGITIMIZE_ADDRESS (@var{x}, @var{oldx}, @var{mode}, @var{win})
8929: A C compound statement that attempts to replace @var{x} with a valid
8930: memory address for an operand of mode @var{mode}. @var{win} will be a
8931: C statement label elsewhere in the code; the macro definition may use
8932:
8933: @example
8934: GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{win});
8935: @end example
8936:
8937: @noindent
8938: to avoid further processing if the address has become legitimate.
8939:
8940: @var{x} will always be the result of a call to @code{break_out_memory_refs},
8941: and @var{oldx} will be the operand that was given to that function to produce
8942: @var{x}.
8943:
8944: The code generated by this macro should not alter the substructure of
8945: @var{x}. If it transforms @var{x} into a more legitimate form, it
8946: should assign @var{x} (which will always be a C variable) a new value.
8947:
8948: It is not necessary for this macro to come up with a legitimate
8949: address. The compiler has standard ways of doing so in all cases. In
8950: fact, it is safe for this macro to do nothing. But often a
8951: machine-dependent strategy can generate better code.
8952:
8953: @item GO_IF_MODE_DEPENDENT_ADDRESS (@var{addr}, @var{label})
8954: A C statement or compound statement with a conditional @code{goto
8955: @var{label};} executed if memory address @var{x} (an RTX) can have
8956: different meanings depending on the machine mode of the memory
8957: reference it is used for.
8958:
8959: Autoincrement and autodecrement addresses typically have mode-dependent
8960: effects because the amount of the increment or decrement is the size
8961: of the operand being addressed. Some machines have other mode-dependent
8962: addresses. Many RISC machines have no mode-dependent addresses.
8963:
8964: You may assume that @var{addr} is a valid address for the machine.
8965:
8966: @item LEGITIMATE_CONSTANT_P (@var{x})
8967: A C expression that is nonzero if @var{x} is a legitimate constant for
8968: an immediate operand on the target machine. You can assume that
1.1.1.8 root 8969: either @var{x} is a @code{const_double} or it satisfies
1.1 root 8970: @code{CONSTANT_P}, so you need not check these things. In fact,
8971: @samp{1} is a suitable definition for this macro on machines where any
1.1.1.8 root 8972: @code{const_double} is valid and anything @code{CONSTANT_P} is valid.@refill
8973: @end table
8974:
8975: @node Delayed Branch, Condition Code, Addressing Modes, Machine Macros
8976: @section Parameters for Delayed Branch Optimization
8977:
8978: @table @code
8979: @item HAVE_DELAYED_BRANCH
8980: Define this macro if the target machine has delayed branches, that is,
8981: a branch does not take effect immediately, and the actual branch
8982: instruction may be followed by one or more instructions that will be
8983: issued before the PC is actually changed.
8984:
8985: If defined, this allows a special scheduling pass to be run after the
8986: second jump optimization to attempt to reorder instructions to exploit
8987: this. Defining this macro also requires the definition of certain
8988: other macros described below.
8989:
8990: @item DBR_SLOTS_AFTER (@var{insn})
8991: This macro must be defined if @code{HAVE_DELAYED_BRANCH} is defined.
8992: Its definition should be a C expression returning the number of
8993: available delay slots following the instruction(s) output by the
8994: pattern for @var{insn}. The definition of ``slot'' is
8995: machine-dependent, and may denote instructions, bytes, or whatever.
8996:
8997: @item DBR_INSN_SLOTS (@var{insn})
8998: This macro must be defined if @code{HAVE_DELAYED_BRANCH} is defined.
8999: It should be a C expression returning the number of slots (typically
9000: the number of machine instructions) consumed by @var{insn}.
9001:
9002: You may assume that @var{insn} is truly an insn, not a note, label,
9003: barrier, dispatch table, @code{use}, or @code{clobber}.
9004:
9005: @item DBR_INSN_ELIGIBLE_P (@var{insn}, @var{dinsn})
9006: A C expression whose value is non-zero if it is legitimate to put
9007: @var{insn} in the delay slot following @var{dinsn}.
9008:
9009: You do not need to take account of data flow considerations in the
9010: definition of this macro, because the delayed branch optimizer always
9011: does that. This macro is needed only when certain insns may not be
9012: placed in certain delay slots for reasons not evident from the RTL
9013: expressions themselves. If there are no such problems, you don't need
9014: to define this macro.
9015:
9016: You may assume that @var{insn} is truly an insn, not a note, label,
9017: barrier, dispatch table, @code{use}, or @code{clobber}. You may
9018: assume that @var{dinsn} is a jump insn with a delay slot.
9019:
9020: @item DBR_OUTPUT_SEQEND(@var{file})
9021: A C statement, to be executed after all slot-filler instructions have
9022: been output. If necessary, call @code{dbr_sequence_length} to
9023: determine the number of slots filled in a sequence (zero if not
9024: currently outputting a sequence), to decide how many no-ops to output,
9025: or whatever.
9026:
9027: Don't define this macro if it has nothing to do, but it is helpful in
9028: reading assembly output if the extent of the delay sequence is made
9029: explicit (e.g. with white space).
9030:
9031: Note that output routines for instructions with delay slots must be
9032: prepared to deal with not being output as part of a sequence (i.e.
9033: when the scheduling pass is not run, or when no slot fillers could be
9034: found.) The variable @code{final_sequence} is null when not
9035: processing a sequence, otherwise it contains the @code{sequence} rtx
9036: being output.
1.1 root 9037: @end table
9038:
1.1.1.9 ! root 9039: @node Condition Code, Cross-compilation, Delayed Branch, Machine Macros
1.1.1.8 root 9040: @section Condition Code Information
9041:
9042: The file @file{conditions.h} defines a variable @code{cc_status} to
9043: describe how the condition code was computed (in case the interpretation of
9044: the condition code depends on the instruction that it was set by). This
9045: variable contains the RTL expressions on which the condition code is
9046: currently based, and several standard flags.
9047:
9048: Sometimes additional machine-specific flags must be defined in the machine
9049: description header file. It can also add additional machine-specific
9050: information by defining @code{CC_STATUS_MDEP}.
9051:
9052: @table @code
9053: @item CC_STATUS_MDEP
9054: C code for a data type which is used for declaring the @code{mdep}
9055: component of @code{cc_status}. It defaults to @code{int}.
9056:
9057: @item CC_STATUS_MDEP_INIT
1.1.1.9 ! root 9058: A C expression to initialize the @code{mdep} field to ``empty''.
! 9059: The default definition does nothing, since most machines don't use
! 9060: the field anyway. If you want to use the field, you should probably
! 9061: define this macro to initialize it.
1.1.1.8 root 9062:
9063: @item NOTICE_UPDATE_CC (@var{exp}, @var{insn})
9064: A C compound statement to set the components of @code{cc_status}
9065: appropriately for an insn @var{insn} whose body is @var{exp}. It is
9066: this macro's responsibility to recognize insns that set the condition
9067: code as a byproduct of other activity as well as those that explicitly
9068: set @code{(cc0)}.
9069:
9070: If there are insn that do not set the condition code but do alter
9071: other machine registers, this macro must check to see whether they
9072: invalidate the expressions that the condition code is recorded as
9073: reflecting. For example, on the 68000, insns that store in address
9074: registers do not set the condition code, which means that usually
9075: @code{NOTICE_UPDATE_CC} can leave @code{cc_status} unaltered for such
9076: insns. But suppose that the previous insn set the condition code
9077: based on location @samp{a4@@(102)} and the current insn stores a new
9078: value in @samp{a4}. Although the condition code is not changed by
9079: this, it will no longer be true that it reflects the contents of
9080: @samp{a4@@(102)}. Therefore, @code{NOTICE_UPDATE_CC} must alter
9081: @code{cc_status} in this case to say that nothing is known about the
9082: condition code value.
9083:
9084: The definition of @code{NOTICE_UPDATE_CC} must be prepared to deal
9085: with the results of peephole optimization: insns whose patterns are
9086: @code{parallel} RTXs containing various @code{reg}, @code{mem} or
9087: constants which are just the operands. The RTL structure of these
9088: insns is not sufficient to indicate what the insns actually do. What
9089: @code{NOTICE_UPDATE_CC} should do when it sees one is just to run
9090: @code{CC_STATUS_INIT}.
9091: @end table
9092:
9093: @node Cross-compilation, Misc, Condition Code, Machine Macros
1.1.1.5 root 9094: @section Cross Compilation and Floating-Point Format
9095:
1.1.1.9 ! root 9096: While all modern machines use 2's complement representation for integers,
1.1.1.5 root 9097: there are a variety of representations for floating point numbers. This
9098: means that in a cross-compiler the representation of floating point numbers
9099: in the compiled program may be different from that used in the machine
9100: doing the compilation.
9101:
9102: Because different representation systems may offer different amounts of
9103: range and precision, the cross compiler cannot safely use the host
9104: machine's floating point arithmetic. Therefore, floating point constants
9105: must be represented in the target machine's format. This means that the
9106: cross compiler cannot use @code{atof} to parse a floating point constant;
9107: it must have its own special routine to use instead. Also, constant
9108: folding must emulate the target machine's arithmetic (or must not be done
9109: at all).
9110:
9111: The macros in the following table should be defined only if you are cross
9112: compiling between different floating point formats.
9113:
9114: Otherwise, don't define them. Then default definitions will be set up which
9115: use @code{double} as the data type, @code{==} to test for equality, etc.
9116:
9117: You don't need to worry about how many times you use an operand of any
9118: of these macros. The compiler never uses operands which have side effects.
9119:
9120: @table @code
9121: @item REAL_VALUE_TYPE
9122: A macro for the C data type to be used to hold a floating point value
9123: in the target machine's format. Typically this would be a
9124: @code{struct} containing an array of @code{int}.
9125:
9126: @item REAL_VALUES_EQUAL (@var{x}, @var{y})
9127: A macro for a C expression which compares for equality the two values,
9128: @var{x} and @var{y}, both of type @code{REAL_VALUE_TYPE}.
9129:
9130: @item REAL_VALUES_LESS (@var{x}, @var{y})
9131: A macro for a C expression which tests whether @var{x} is less than
9132: @var{y}, both values being of type @code{REAL_VALUE_TYPE} and
9133: interpreted as floating point numbers in the target machine's
9134: representation.
9135:
9136: @item REAL_VALUE_LDEXP (@var{x}, @var{scale})
9137: A macro for a C expression which performs the standard library
9138: function @code{ldexp}, but using the target machine's floating point
9139: representation. Both @var{x} and the value of the expression have
9140: type @code{REAL_VALUE_TYPE}. The second argument, @var{scale}, is an
9141: integer.
9142:
9143: @item REAL_VALUE_ATOF (@var{string})
9144: A macro for a C expression which converts @var{string}, an expression
9145: of type @code{char *}, into a floating point number in the target
9146: machine's representation. The value has type @code{REAL_VALUE_TYPE}.
9147: @end table
9148:
9149: Define the following additional macros if you want to make floating
9150: point constant folding work while cross compiling. If you don't
9151: define them, cross compilation is still possible, but constant folding
9152: will not happen for floating point values.
9153:
9154: @table @code
9155: @item REAL_ARITHMETIC (@var{output}, @var{code}, @var{x}, @var{y})
9156: A macro for a C statement which calculates an arithmetic operation of
9157: the two floating point values @var{x} and @var{y}, both of type
9158: @code{REAL_VALUE_TYPE} in the target machine's representation, to
9159: produce a result of the same type and representation which is stored
9160: in @var{output} (which will be a variable).
9161:
9162: The operation to be performed is specified by @var{code}, a tree code
9163: which will always be one of the following: @code{PLUS_EXPR},
9164: @code{MINUS_EXPR}, @code{MULT_EXPR}, @code{RDIV_EXPR},
9165: @code{MAX_EXPR}, @code{MIN_EXPR}.@refill
9166:
9167: The expansion of this macro is responsible for checking for overflow.
9168: If overflow happens, the macro expansion should execute the statement
9169: @code{return 0;}, which indicates the inability to perform the
9170: arithmetic operation requested.
9171:
9172: @item REAL_VALUE_NEGATE (@var{x})
9173: A macro for a C expression which returns the negative of the floating
9174: point value @var{x}. Both @var{x} and the value of the expression
9175: have type @code{REAL_VALUE_TYPE} and are in the target machine's
9176: floating point representation.
9177:
9178: There is no way for this macro to report overflow, since overflow
9179: can't happen in the negation operation.
9180:
9181: @item REAL_VALUE_TO_INT (@var{low}, @var{high}, @var{x})
9182: A macro for a C expression which converts a floating point value
9183: @var{x} into a double-precision integer which is then stored into
9184: @var{low} and @var{high}, two variables of type @var{int}.
9185:
9186: @item REAL_VALUE_FROM_INT (@var{x}, @var{low}, @var{high})
9187: A macro for a C expression which converts a double-precision integer
9188: found in @var{low} and @var{high}, two variables of type @var{int},
9189: into a floating point value which is then stored into @var{x}.
9190: @end table
9191:
1.1.1.8 root 9192: @node Misc, Assembler Format, Cross-compilation, Machine Macros
1.1 root 9193: @section Miscellaneous Parameters
9194:
9195: @table @code
9196: @item CASE_VECTOR_MODE
9197: An alias for a machine mode name. This is the machine mode that
9198: elements of a jump-table should have.
9199:
9200: @item CASE_VECTOR_PC_RELATIVE
9201: Define this macro if jump-tables should contain relative addresses.
9202:
9203: @item CASE_DROPS_THROUGH
9204: Define this if control falls through a @code{case} insn when the index
9205: value is out of range. This means the specified default-label is
9206: actually ignored by the @code{case} insn proper.
9207:
9208: @item IMPLICIT_FIX_EXPR
9209: An alias for a tree code that should be used by default for conversion
9210: of floating point values to fixed point. Normally,
9211: @code{FIX_ROUND_EXPR} is used.@refill
9212:
9213: @item FIXUNS_TRUNC_LIKE_FIX_TRUNC
9214: Define this macro if the same instructions that convert a floating
9215: point number to a signed fixed point number also convert validly to an
9216: unsigned one.
9217:
9218: @item EASY_DIV_EXPR
9219: An alias for a tree code that is the easiest kind of division to
9220: compile code for in the general case. It may be
9221: @code{TRUNC_DIV_EXPR}, @code{FLOOR_DIV_EXPR}, @code{CEIL_DIV_EXPR} or
9222: @code{ROUND_DIV_EXPR}. These four division operators differ in how
9223: they round the result to an integer. @code{EASY_DIV_EXPR} is used
9224: when it is permissible to use any of those kinds of division and the
9225: choice should be made on the basis of efficiency.@refill
9226:
9227: @item DEFAULT_SIGNED_CHAR
9228: An expression whose value is 1 or 0, according to whether the type
9229: @code{char} should be signed or unsigned by default. The user can
9230: always override this default with the options @samp{-fsigned-char}
9231: and @samp{-funsigned-char}.
9232:
9233: @item SCCS_DIRECTIVE
9234: Define this if the preprocessor should ignore @code{#sccs} directives
9235: and print no error message.
9236:
1.1.1.7 root 9237: @item HAVE_VPRINTF
9238: Define this if the library function @code{vprintf} is available on your
9239: system.
1.1 root 9240:
9241: @item MOVE_MAX
9242: The maximum number of bytes that a single instruction can move quickly
9243: from memory to memory.
9244:
9245: @item INT_TYPE_SIZE
9246: A C expression for the size in bits of the type @code{int} on the
1.1.1.8 root 9247: target machine. If you don't define this, the default is one word.
9248:
9249: @item SHORT_TYPE_SIZE
9250: A C expression for the size in bits of the type @code{short} on the
9251: target machine. If you don't define this, the default is half a word.
9252: (If this would be less than one storage unit, it is rounded up to one
9253: unit.)
9254:
9255: @item LONG_TYPE_SIZE
9256: A C expression for the size in bits of the type @code{long} on the
9257: target machine. If you don't define this, the default is one word.
9258:
9259: @item LONG_LONG_TYPE_SIZE
9260: A C expression for the size in bits of the type @code{long long} on the
9261: target machine. If you don't define this, the default is two
9262: words.
9263:
9264: @item CHAR_TYPE_SIZE
9265: A C expression for the size in bits of the type @code{char} on the
9266: target machine. If you don't define this, the default is one quarter
9267: of a word. (If this would be less than one storage unit, it is rounded up
9268: to one unit.)
9269:
9270: @item FLOAT_TYPE_SIZE
9271: A C expression for the size in bits of the type @code{float} on the
9272: target machine. If you don't define this, the default is one word.
9273:
9274: @item DOUBLE_TYPE_SIZE
9275: A C expression for the size in bits of the type @code{double} on the
9276: target machine. If you don't define this, the default is two
9277: words.
9278:
9279: @item LONG_DOUBLE_TYPE_SIZE
9280: A C expression for the size in bits of the type @code{long double} on
9281: the target machine. If you don't define this, the default is two
9282: words.
1.1 root 9283:
9284: @item SLOW_BYTE_ACCESS
9285: Define this macro as a C expression which is nonzero if accessing less
9286: than a word of memory (i.e. a @code{char} or a @code{short}) is slow
9287: (requires more than one instruction).
9288:
9289: @item SLOW_ZERO_EXTEND
9290: Define this macro if zero-extension (of a @code{char} or @code{short}
9291: to an @code{int}) can be done faster if the destination is a register
9292: that is known to be zero.
9293:
9294: If you define this macro, you must have instruction patterns that
9295: recognize RTL structures like this:
9296:
9297: @example
9298: (set (strict-low-part (subreg:QI (reg:SI @dots{}) 0)) @dots{})
9299: @end example
9300:
9301: @noindent
9302: and likewise for @code{HImode}.
9303:
9304: @item SHIFT_COUNT_TRUNCATED
9305: Define this macro if shift instructions ignore all but the lowest few
9306: bits of the shift count. It implies that a sign-extend or zero-extend
9307: instruction for the shift count can be omitted.
9308:
9309: @item TRULY_NOOP_TRUNCATION (@var{outprec}, @var{inprec})
9310: A C expression which is nonzero if on this machine it is safe to
9311: ``convert'' an integer of @var{inprec} bits to one of @var{outprec}
9312: bits (where @var{outprec} is smaller than @var{inprec}) by merely
9313: operating on it as if it had only @var{outprec} bits.
9314:
9315: On many machines, this expression can be 1.
9316:
9317: @item NO_FUNCTION_CSE
9318: Define this macro if it is as good or better to call a constant
9319: function address than to call an address kept in a register.
9320:
9321: @item PROMOTE_PROTOTYPES
9322: Define this macro if an argument declared as @code{char} or
9323: @code{short} in a prototype should actually be passed as an
9324: @code{int}. In addition to avoiding errors in certain cases of
9325: mismatch, it also makes for better code on certain machines.
9326:
9327: @item STORE_FLAG_VALUE
9328: A C expression for the value stored by a store-flag instruction
9329: (@code{s@var{cond}}) when the condition is true. This is usually 1 or
1.1.1.9 ! root 9330: -1; it is required to be an odd number or a negative number.
1.1 root 9331:
9332: Do not define @code{STORE_FLAG_VALUE} if the machine has no store-flag
9333: instructions.
9334:
9335: @item Pmode
9336: An alias for the machine mode for pointers. Normally the definition
9337: can be
9338:
9339: @example
9340: #define Pmode SImode
9341: @end example
9342:
9343: @item FUNCTION_MODE
9344: An alias for the machine mode used for memory references to functions
1.1.1.8 root 9345: being called, in @code{call} RTL expressions. On most machines this
1.1 root 9346: should be @code{QImode}.
9347:
9348: @item INSN_MACHINE_INFO
9349: This macro should expand into a C structure type to use for the
9350: machine-dependent info field specified with the optional last argument
1.1.1.8 root 9351: in @code{define_insn} and @code{define_peephole} patterns. For example,
9352: it might expand into @code{struct machine_info}; then it would be up
1.1 root 9353: to you to define this structure in the @file{tm.h} file.
9354:
9355: You do not need to define this macro if you do not write the optional
9356: last argument in any of the patterns in the machine description.
9357:
1.1.1.8 root 9358: @item DEFAULT_MACHINE_INFO
9359: This macro should expand into a C initializer to use to initialize
9360: the machine-dependent info for one insn pattern. It is used for patterns
9361: that do not specify the machine-dependent info.
9362:
9363: If you do not define this macro, zero is used.
9364:
1.1 root 9365: @item CONST_COSTS (@var{x}, @var{code})
9366: A part of a C @code{switch} statement that describes the relative
9367: costs of constant RTL expressions. It must contain @code{case} labels
1.1.1.8 root 9368: for expression codes @code{const_int}, @code{const}, @code{symbol_ref}, @code{label_ref}
9369: and @code{const_double}. Each case must ultimately reach a
1.1 root 9370: @code{return} statement to return the relative cost of the use of that
9371: kind of constant value in an expression. The cost may depend on the
9372: precise value of the constant, which is available for examination in
9373: @var{x}.
9374:
9375: @var{code} is the expression code---redundant, since it can be
9376: obtained with @code{GET_CODE (@var{x})}.
9377:
9378: @item DOLLARS_IN_IDENTIFIERS
9379: Define this to be nonzero if the character @samp{$} should be allowed
9380: by default in identifier names.
1.1.1.7 root 9381:
9382: @item USE_C_ALLOCA
9383: Define this macro to indicate that the compiler is running with the
9384: @code{alloca} implemented in C. This version of @code{alloca} can be
1.1.1.9 ! root 9385: found in the file @file{alloca.c}; to use it, you must also alter the
! 9386: @file{Makefile} variable @code{ALLOCA}.
1.1.1.7 root 9387:
9388: This macro, unlike most, describes the machine that the compiler is
9389: running on, rather than the one the compiler is compiling for.
9390: Therefore, it should be set in the @file{xm-@var{machine}.h} file
1.1.1.8 root 9391: rather than in the @file{tm-@var{machine}.h} file.
1.1.1.7 root 9392:
9393: If you do define this macro, you should probably do it as follows:
9394:
9395: @example
9396: #ifndef __GNUC__
9397: #define USE_C_ALLOCA
9398: #else
1.1.1.8 root 9399: #define alloca __builtin_alloca
1.1.1.7 root 9400: #endif
9401: @end example
9402:
9403: @noindent
9404: so that when the compiler is compiled with GNU CC it uses the more
9405: efficient built-in @code{alloca} function.
1.1 root 9406: @end table
9407:
1.1.1.8 root 9408: @node Assembler Format,, Misc, Machine Macros
1.1 root 9409: @section Output of Assembler Code
9410:
9411: @table @code
9412: @item ASM_SPEC
9413: A C string constant that tells the GNU CC driver program options to
9414: pass to the assembler. It can also specify how to translate options
9415: you give to GNU CC into options for GNU CC to pass to the assembler.
9416: See the file @file{tm-sun3.h} for an example of this.
9417:
9418: Do not define this macro if it does not need to do anything.
9419:
9420: @item LINK_SPEC
9421: A C string constant that tells the GNU CC driver program options to
9422: pass to the linker. It can also specify how to translate options you
9423: give to GNU CC into options for GNU CC to pass to the linker.
9424:
9425: Do not define this macro if it does not need to do anything.
9426:
9427: @item LIB_SPEC
9428: Another C string constant used much like @code{LINK_SPEC}. The difference
9429: between the two is that @code{LIBS_SPEC} is used at the end of the
9430: command given to the linker.
9431:
9432: If this macro is not defined, a default is provided that
9433: loads the standard C library from the usual place. See @file{gcc.c}.
9434:
9435: @item STARTFILE_SPEC
9436: Another C string constant used much like @code{LINK_SPEC}. The
9437: difference between the two is that @code{STARTFILE_SPEC} is used at
9438: the very beginning of the command given to the linker.
9439:
9440: If this macro is not defined, a default is provided that loads the
9441: standard C startup file from the usual place. See @file{gcc.c}.
9442:
1.1.1.7 root 9443: @item STANDARD_EXEC_PREFIX
9444: Define this macro as a C string constant if you wish to override the
9445: standard choice of @file{/usr/local/lib/gcc-} as the default prefix to
9446: try when searching for the executable files of the compiler.
9447:
9448: The prefix specified by the @samp{-B} option, if any, is tried before
9449: the default prefix. After the default prefix, if the executable is
9450: not found that way, @file{/usr/lib/gcc-} is tried next; then the
9451: directories in your search path for shell commands are searched.
9452:
1.1.1.4 root 9453: @item STANDARD_STARTFILE_PREFIX
9454: Define this macro as a C string constant if you wish to override the
1.1.1.7 root 9455: standard choice of @file{/usr/local/lib/} as the default prefix to try
9456: when searching for startup files such as @file{crt0.o}.
9457:
9458: In this search, all the prefixes tried for executable files are tried
9459: first. Then comes the default startfile prefix specified by this
9460: macro, followed by the prefixes @file{/lib/} and @file{/usr/lib/} as
9461: last resorts.
1.1.1.4 root 9462:
1.1 root 9463: @item ASM_FILE_START (@var{stream})
9464: A C expression which outputs to the stdio stream @var{stream}
9465: some appropriate text to go at the start of an assembler file.
9466:
9467: Normally this macro is defined to output a line containing
9468: @samp{#NO_APP}, which is a comment that has no effect on most
9469: assemblers but tells the GNU assembler that it can save time by not
9470: checking for certain assembler constructs.
9471:
9472: On systems that use SDB, it is necessary to output certain commands;
9473: see @file{tm-attasm.h}.
9474:
1.1.1.8 root 9475: @item ASM_FILE_END (@var{stream})
9476: A C expression which outputs to the stdio stream @var{stream}
9477: some appropriate text to go at the end of an assembler file.
9478:
9479: If this macro is not defined, the default is to output nothing
9480: special at the end of the file. Most systems don't require any
9481: definition.
9482:
9483: On systems that use SDB, it is necessary to output certain commands;
9484: see @file{tm-attasm.h}.
9485:
9486: @item ASM_IDENTIFY_GCC (@var{file})
9487: A C statement to output assembler commands which will identify
9488: the object file as having been compiled with GNU CC (or another
9489: GNU compiler).
9490:
9491: If you don't define this macro, the string @samp{gcc_compiled.:}
9492: is output. This string is calculated to define a symbol which,
9493: on BSD systems, will never be defined for any other reason.
9494: GDB checks for the presence of this symbol when reading the
9495: symbol table of an executable.
9496:
9497: On non-BSD systems, you must arrange communication with GDB in
9498: some other fashion. If GDB is not used on your system, you can
9499: define this macro with an empty body.
9500:
1.1 root 9501: @item ASM_APP_ON
9502: A C string constant for text to be output before each @code{asm}
9503: statement or group of consecutive ones. Normally this is
9504: @code{"#APP"}, which is a comment that has no effect on most
9505: assemblers but tells the GNU assembler that it must check the lines
9506: that follow for all valid assembler constructs.
9507:
9508: @item ASM_APP_OFF
9509: A C string constant for text to be output after each @code{asm}
9510: statement or group of consecutive ones. Normally this is
9511: @code{"#NO_APP"}, which tells the GNU assembler to resume making the
9512: time-saving assumptions that are valid for ordinary compiler output.
9513:
9514: @item TEXT_SECTION_ASM_OP
9515: A C string constant for the assembler operation that should precede
9516: instructions and read-only data. Normally @code{".text"} is right.
9517:
9518: @item DATA_SECTION_ASM_OP
9519: A C string constant for the assembler operation to identify the
9520: following data as writable initialized data. Normally @code{".data"}
9521: is right.
9522:
1.1.1.8 root 9523: @item EXTRA_SECTIONS
9524: A list of names for sections other than the standard two, which are
9525: @code{in_text} and @code{in_data}. You need not define this macro
9526: on a system with no other sections (that GCC needs to use).
9527:
9528: @item EXTRA_SECTION_FUNCTIONS
9529: One or more functions to be defined in @file{varasm.c}. These
9530: functions should do jobs analogous to those of @code{text_section} and
9531: @code{data_section}, for your additional sections. Do not define this
9532: macro if you do not define @code{EXTRA_SECTIONS}.
9533:
9534: @item SELECT_SECTION (@var{exp})
9535: A C statement or statements to switch to the appropriate section for
9536: output of @var{exp}. You can assume that @var{exp} is either a
9537: @code{VAR_DECL} node or a constant of some sort. Select the section
9538: by calling @code{text_section} or one of the alternatives for other
9539: sections.
9540:
9541: Do not define this macro if you use only the standard two sections
9542: and put all read-only variables and constants in the text section.
9543:
9544: @item SELECT_RTX_SECTION (@var{mode}, @var{rtx})
9545: A C statement or statements to switch to the appropriate section for
9546: output of @var{rtx} in mode @var{mode}. You can assume that @var{rtx}
9547: is some kind of constant in RTL. The argument @var{mode} is redundant
9548: except in the case of a @code{const_int} rtx. Select the section by
9549: calling @code{text_section} or one of the alternatives for other
9550: sections.
9551:
9552: Do not define this macro if you use only the standard two sections and
9553: put all constants in the text section.
9554:
1.1 root 9555: @item REGISTER_NAMES
9556: A C initializer containing the assembler's names for the machine
9557: registers, each one as a C string constant. This is what translates
9558: register numbers in the compiler into assembler language.
9559:
9560: @item DBX_REGISTER_NUMBER (@var{regno})
9561: A C expression that returns the DBX register number for the compiler
9562: register number @var{regno}. In simple cases, the value of this
9563: expression may be @var{regno} itself. But sometimes there are some
9564: registers that the compiler knows about and DBX does not, or vice
9565: versa. In such cases, some register may need to have one number in
9566: the compiler and another for DBX.
9567:
9568: @item DBX_DEBUGGING_INFO
9569: Define this macro if GNU CC should produce debugging output for DBX
9570: in response to the @samp{-g} option.
9571:
9572: @item SDB_DEBUGGING_INFO
9573: Define this macro if GNU CC should produce debugging output for SDB
9574: in response to the @samp{-g} option.
9575:
9576: @item PUT_SDB_@var{op}
9577: Define these macros to override the assembler syntax for the special
9578: SDB assembler directives. See @file{sdbout.c} for a list of these
9579: macros and their arguments. If the standard syntax is used, you need
9580: not define them yourself.
9581:
9582: @item SDB_GENERATE_FAKE
9583: Define this macro to override the usual method of constructing a dummy
9584: name for anonymous structure and union types. See @file{sdbout.c} for
1.1.1.9 ! root 9585: more information.
1.1 root 9586:
9587: @item DBX_NO_XREFS
9588: Define this macro if DBX on your system does not support the construct
9589: @samp{xs@var{tagname}}. On some systems, this construct is used to
9590: describe a forward reference to a structure named @var{tagname}.
9591: On other systems, this construct is not supported at all.
9592:
9593: @item DBX_CONTIN_LENGTH
9594: A symbol name in DBX-format debugging information is normally
9595: continued (split into two separate @code{.stabs} directives) when it
9596: exceeds a certain length (by default, 80 characters). On some
9597: operating systems, DBX requires this splitting; on others, splitting
9598: must not be done. You can inhibit splitting by defining this macro
9599: with the value zero. You can override the default splitting-length by
9600: defining this macro as an expression for the length you desire.
9601:
9602: @item DBX_CONTIN_CHAR
9603: Normally continuation is indicated by adding a @samp{\} character to
9604: the end of a @code{.stabs} string when a continuation follows. To use
9605: a different character instead, define this macro as a character
9606: constant for the character you want to use. Do not define this macro
9607: if backslash is correct for your system.
9608:
1.1.1.8 root 9609: @item DBX_STATIC_STAB_DATA_SECTION
9610: Define this macro if it is necessary to go to the data section before
9611: outputting the @samp{.stabs} pseudo-op for a non-global static
9612: variable.
9613:
1.1 root 9614: @item ASM_OUTPUT_LABEL (@var{stream}, @var{name})
9615: A C statement (sans semicolon) to output to the stdio stream
1.1.1.8 root 9616: @var{stream} the assembler definition of a label named @var{name}.
9617: Use the expression @code{assemble_name (@var{stream}, @var{name})} to
9618: output the name itself; before and after that, output the additional
1.1 root 9619: assembler syntax for defining the name, and a newline.
9620:
9621: @item ASM_DECLARE_FUNCTION_NAME (@var{stream}, @var{name}, @var{decl})
9622: A C statement (sans semicolon) to output to the stdio stream
9623: @var{stream} any text necessary for declaring the name @var{name} of a
9624: function which is being defined. This macro is responsible for
9625: outputting the label definition (perhaps using
9626: @code{ASM_OUTPUT_LABEL}). The argument @var{decl} is the
9627: @code{FUNCTION_DECL} tree node representing the function.
9628:
9629: If this macro is not defined, then the function name is defined in the
9630: usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}).
9631:
9632: @item ASM_GLOBALIZE_LABEL (@var{stream}, @var{name})
9633: A C statement (sans semicolon) to output to the stdio stream
9634: @var{stream} some commands that will make the label @var{name} global;
9635: that is, available for reference from other files. Use the expression
9636: @code{assemble_name (@var{stream}, @var{name})} to output the name
9637: itself; before and after that, output the additional assembler syntax
9638: for making that name global, and a newline.
9639:
1.1.1.8 root 9640: @item ASM_OUTPUT_EXTERNAL (@var{stream}, @var{decl}, @var{name})
1.1 root 9641: A C statement (sans semicolon) to output to the stdio stream
9642: @var{stream} any text necessary for declaring the name of an external
9643: symbol named @var{name} which is referenced in this compilation but
9644: not defined. The value of @var{decl} is the tree node for the
9645: declaration.
9646:
9647: This macro need not be defined if it does not need to output anything.
9648: The GNU assembler and most Unix assemblers don't require anything.
9649:
9650: @item ASM_OUTPUT_LABELREF (@var{stream}, @var{name})
1.1.1.8 root 9651: A C statement to output to the stdio stream @var{stream} a reference
9652: in assembler syntax to a label named @var{name}. The character
9653: @samp{_} should be added to the front of the name, if that is
9654: customary on your operating system, as it is in most Berkeley Unix
9655: systems. This macro is used in @code{assemble_name}.
1.1 root 9656:
9657: @item ASM_GENERATE_INTERNAL_LABEL (@var{string}, @var{prefix}, @var{num})
1.1.1.8 root 9658: A C statement to store into the string @var{string} a label whose name
9659: is made from the string @var{prefix} and the number @var{num}.
1.1 root 9660:
9661: This string, when output subsequently by @code{ASM_OUTPUT_LABELREF},
9662: should produce the same output that @code{ASM_OUTPUT_INTERNAL_LABEL}
9663: would produce with the same @var{prefix} and @var{num}.
9664:
9665: @item ASM_OUTPUT_INTERNAL_LABEL (@var{stream}, @var{prefix}, @var{num})
9666: A C statement to output to the stdio stream @var{stream} a label whose
9667: name is made from the string @var{prefix} and the number @var{num}.
9668: These labels are used for internal purposes, and there is no reason
9669: for them to appear in the symbol table of the object file. On many
9670: systems, the letter @samp{L} at the beginning of a label has this
9671: effect. The usual definition of this macro is as follows:
9672:
9673: @example
9674: fprintf (@var{stream}, "L%s%d:\n", @var{prefix}, @var{num})
9675: @end example
9676:
9677: @item ASM_OUTPUT_CASE_LABEL (@var{stream}, @var{prefix}, @var{num}, @var{table})
9678: Define this if the label before a jump-table needs to be output
9679: specially. The first three arguments are the same as for
9680: @code{ASM_OUTPUT_INTERNAL_LABEL}; the fourth argument is the
1.1.1.8 root 9681: jump-table which follows (a @code{jump_insn} containing an
9682: @code{addr_vec} or @code{addr_diff_vec}).
1.1 root 9683:
9684: This feature is used on system V to output a @code{swbeg} statement
9685: for the table.
9686:
9687: If this macro is not defined, these labels are output with
9688: @code{ASM_OUTPUT_INTERNAL_LABEL}.
9689:
9690: @item ASM_OUTPUT_CASE_END (@var{stream}, @var{num}, @var{table})
1.1.1.8 root 9691: Define this if something special must be output at the end of a
9692: jump-table. The definition should be a C statement to be executed
9693: after the assembler code for the table is written. It should write
9694: the appropriate code to stdio stream @var{stream}. The argument
9695: @var{table} is the jump-table insn, and @var{num} is the label-number
9696: of the preceding label.
1.1 root 9697:
9698: If this macro is not defined, nothing special is output at the end of
9699: the jump-table.
9700:
1.1.1.4 root 9701: @item ASM_OUTPUT_ALIGN_CODE (@var{file})
9702: A C expression to output text to align the location counter in the way
9703: that is desirable at a point in the code that is reached only by
9704: jumping.
9705:
9706: This macro need not be defined if you don't want any special alignment
9707: to be done at such a time. Most machine descriptions do not currently
9708: define the macro.
9709:
1.1 root 9710: @item ASM_FORMAT_PRIVATE_NAME (@var{outvar}, @var{name}, @var{number})
9711: A C expression to assign to @var{outvar} (which is a variable of type
9712: @code{char *}) a newly allocated string made from the string
9713: @var{name} and the number @var{number}, with some suitable punctuation
9714: added. Use @code{alloca} to get space for the string.
9715:
9716: This string will be used as the argument to @code{ASM_OUTPUT_LABELREF}
9717: to produce an assembler label for an internal static variable whose
9718: name is @var{name}. Therefore, the string must be such as to result
9719: in valid assembler code. The argument @var{number} is different each
9720: time this macro is executed; it prevents conflicts between
9721: similarly-named internal static variables in different scopes.
9722:
9723: Ideally this string should not be a valid C identifier, to prevent any
9724: conflict with the user's own symbols. Most assemblers allow periods
9725: or percent signs in assembler symbols; putting at least one of these
9726: between the name and the number will suffice.
9727:
9728: @item ASM_OUTPUT_REG_PUSH (@var{stream}, @var{regno})
9729: A C expression to output to @var{stream} some assembler code
9730: which will push hard register number @var{regno} onto the stack.
9731: The code need not be optimal, since this macro is used only when
9732: profiling.
9733:
9734: @item ASM_OUTPUT_REG_POP (@var{stream}, @var{regno})
9735: A C expression to output to @var{stream} some assembler code
9736: which will pop hard register number @var{regno} off of the stack.
9737: The code need not be optimal, since this macro is used only when
9738: profiling.
9739:
9740: @item ASM_OUTPUT_ADDR_DIFF_ELT (@var{stream}, @var{value}, @var{rel})
9741: This macro should be provided on machines where the addresses
9742: in a dispatch table are relative to the table's own address.
9743:
9744: The definition should be a C statement to output to the stdio stream
9745: @var{stream} an assembler pseudo-instruction to generate a difference
9746: between two labels. @var{value} and @var{rel} are the numbers of two
9747: internal labels. The definitions of these labels are output using
9748: @code{ASM_OUTPUT_INTERNAL_LABEL}, and they must be printed in the same
9749: way here. For example,
9750:
9751: @example
9752: fprintf (@var{stream}, "\t.word L%d-L%d\n",
9753: @var{value}, @var{rel})
9754: @end example
9755:
9756: @item ASM_OUTPUT_ADDR_VEC_ELT (@var{stream}, @var{value})
9757: This macro should be provided on machines where the addresses
9758: in a dispatch table are absolute.
9759:
9760: The definition should be a C statement to output to the stdio stream
9761: @var{stream} an assembler pseudo-instruction to generate a reference to
9762: a label. @var{value} is the number of an internal label whose
9763: definition is output using @code{ASM_OUTPUT_INTERNAL_LABEL}.
9764: For example,
9765:
9766: @example
9767: fprintf (@var{stream}, "\t.word L%d\n", @var{value})
9768: @end example
9769:
9770: @item ASM_OUTPUT_DOUBLE (@var{stream}, @var{value})
9771: A C statement to output to the stdio stream @var{stream} an assembler
9772: instruction to assemble a @code{double} constant whose value is
9773: @var{value}. @var{value} will be a C expression of type
9774: @code{double}.
9775:
9776: @item ASM_OUTPUT_FLOAT (@var{stream}, @var{value})
9777: A C statement to output to the stdio stream @var{stream} an assembler
9778: instruction to assemble a @code{float} constant whose value is
9779: @var{value}. @var{value} will be a C expression of type @code{float}.
9780:
9781: @item ASM_OUTPUT_INT (@var{stream}, @var{exp})
9782: @itemx ASM_OUTPUT_SHORT (@var{stream}, @var{exp})
9783: @itemx ASM_OUTPUT_CHAR (@var{stream}, @var{exp})
9784: A C statement to output to the stdio stream @var{stream} an assembler
9785: instruction to assemble a @code{int}, @code{short} or @code{char}
9786: constant whose value is @var{value}. The argument @var{exp} will be
9787: an RTL expression which represents a constant value. Use
9788: @samp{output_addr_const (@var{exp})} to output this value as an
9789: assembler expression.@refill
9790:
1.1.1.8 root 9791: @item ASM_OUTPUT_DOUBLE_INT (@var{stream}, @var{exp})
9792: A C statement to output to the stdio stream @var{stream} an assembler
9793: instruction to assemble a @code{long long} constant whose value is
9794: @var{exp}. The argument @var{exp} will be an RTL expression which
9795: represents a constant value. It may be a @code{const_double} RTX,
9796: or it may be an ordinary single-precision constant. In the latter
9797: case, you should zero-extend it.
9798:
1.1 root 9799: @item ASM_OUTPUT_BYTE (@var{stream}, @var{value})
9800: A C statement to output to the stdio stream @var{stream} an assembler
9801: instruction to assemble a single byte containing the number @var{value}.
9802:
9803: @item ASM_OUTPUT_ASCII (@var{stream}, @var{ptr}, @var{len})
9804: A C statement to output to the stdio stream @var{stream} an assembler
9805: instruction to assemble a string constant containing the @var{len}
9806: bytes at @var{ptr}. @var{ptr} will be a C expression of type
9807: @code{char *} and @var{len} a C expression of type @code{int}.
9808:
9809: If the assembler has a @code{.ascii} pseudo-op as found in the
9810: Berkeley Unix assembler, do not define the macro
9811: @code{ASM_OUTPUT_ASCII}.
9812:
9813: @item ASM_OUTPUT_SKIP (@var{stream}, @var{nbytes})
9814: A C statement to output to the stdio stream @var{stream} an assembler
9815: instruction to advance the location counter by @var{nbytes} bytes.
9816: @var{nbytes} will be a C expression of type @code{int}.
9817:
9818: @item ASM_OUTPUT_ALIGN (@var{stream}, @var{power})
9819: A C statement to output to the stdio stream @var{stream} an assembler
9820: instruction to advance the location counter to a multiple of 2 to the
9821: @var{power} bytes. @var{power} will be a C expression of type @code{int}.
9822:
1.1.1.7 root 9823: @item ASM_OUTPUT_COMMON (@var{stream}, @var{name}, @var{size}, @var{rounded})
1.1 root 9824: A C statement (sans semicolon) to output to the stdio stream
1.1.1.7 root 9825: @var{stream} the assembler definition of a common-label named
9826: @var{name} whose size is @var{size} bytes. The variable @var{rounded}
9827: is the size rounded up to whatever alignment the caller wants.
9828:
9829: Use the expression @code{assemble_name (@var{stream}, @var{name})} to
9830: output the name itself; before and after that, output the additional
9831: assembler syntax for defining the name, and a newline.
1.1 root 9832:
9833: This macro controls how the assembler definitions of uninitialized
9834: global variables are output.
9835:
1.1.1.7 root 9836: @item ASM_OUTPUT_LOCAL (@var{stream}, @var{name}, @var{size}, @var{rounded})
1.1 root 9837: A C statement (sans semicolon) to output to the stdio stream
9838: @var{stream} the assembler definition of a local-common-label named
1.1.1.7 root 9839: @var{name} whose size is @var{size} bytes. The variable @var{rounded}
9840: is the size rounded up to whatever alignment the caller wants.
9841:
9842: Use the expression @code{assemble_name (@var{stream}, @var{name})} to
9843: output the name itself; before and after that, output the additional
9844: assembler syntax for defining the name, and a newline.
1.1 root 9845:
9846: This macro controls how the assembler definitions of uninitialized
9847: static variables are output.
9848:
1.1.1.8 root 9849: @item ASM_OUTPUT_SOURCE_FILENAME (@var{stream}, @var{name})
9850: A C statment to output DBX or SDB debugging information which indicates
9851: that filename @var{name} is the current source file to the stdio stream
9852: @var{stream}.
9853:
9854: This macro need not be defined if the standard form of debugging
9855: information for the debugger in use is appropriate.
9856:
1.1 root 9857: @item ASM_OUTPUT_SOURCE_LINE (@var{stream}, @var{line})
9858: A C statment to output DBX or SDB debugging information before code
9859: for line number @var{line} of the current source file to the
9860: stdio stream @var{stream}.
9861:
9862: This macro need not be defined if the standard form of debugging
9863: information for the debugger in use is appropriate.
9864:
9865: @item ASM_OUTPUT_IDENT (@var{stream}, @var{string})
9866: A C statement to output something to the assembler file to handle a
9867: @samp{#ident} directive containing the text @var{string}. If this
1.1.1.7 root 9868: macro is not defined, nothing is output for a @samp{#ident} directive.
1.1 root 9869:
9870: @item TARGET_BELL
9871: A C constant expression for the integer value for escape sequence
9872: @samp{\a}.
9873:
9874: @item TARGET_BS
9875: @itemx TARGET_TAB
9876: @itemx TARGET_NEWLINE
9877: C constant expressions for the integer values for escape sequences
9878: @samp{\b}, @samp{\t} and @samp{\n}.
9879:
9880: @item TARGET_VT
9881: @itemx TARGET_FF
9882: @itemx TARGET_CR
9883: C constant expressions for the integer values for escape sequences
9884: @samp{\v}, @samp{\f} and @samp{\r}.
9885:
9886: @item ASM_OUTPUT_OPCODE (@var{stream}, @var{ptr})
9887: Define this macro if you are using an unusual assembler that
9888: requires different names for the machine instructions.
9889:
9890: The definition is a C statement or statements which output an
9891: assembler instruction opcode to the stdio stream @var{stream}. The
9892: macro-operand @var{ptr} is a variable of type @code{char *} which
9893: points to the opcode name in its ``internal'' form---the form that is
9894: written in the machine description. The definition should output the
9895: opcode name to @var{stream}, performing any translation you desire, and
9896: increment the variable @var{ptr} to point at the end of the opcode
9897: so that it will not be output twice.
9898:
9899: In fact, your macro definition may process less than the entire opcode
9900: name, or more than the opcode name; but if you want to process text
9901: that includes @samp{%}-sequences to substitute operands, you must take
9902: care of the substitution yourself. Just be sure to increment
9903: @var{ptr} over whatever text should not be output normally.
9904:
1.1.1.8 root 9905: If you need to look at the operand values, they can be found as the
9906: elements of @code{recog_operand}.
9907:
1.1 root 9908: If the macro definition does nothing, the instruction is output
9909: in the usual way.
9910:
9911: @item FINAL_PRESCAN_INSN (@var{insn}, @var{opvec}, @var{noperands})
9912: If defined, a C statement to be executed just prior to the output of
9913: assembler code for @var{insn}, to modify the extracted operands so
9914: they will be output differently.
9915:
9916: Here the argument @var{opvec} is the vector containing the operands
9917: extracted from @var{insn}, and @var{noperands} is the number of
9918: elements of the vector which contain meaningful data for this insn.
9919: The contents of this vector are what will be used to convert the insn
9920: template into assembler code, so you can change the assembler output
9921: by changing the contents of the vector.
9922:
9923: This macro is useful when various assembler syntaxes share a single
9924: file of instruction patterns; by defining this macro differently, you
9925: can cause a large class of instructions to be output differently (such
9926: as with rearranged operands). Naturally, variations in assembler
9927: syntax affecting individual insn patterns ought to be handled by
9928: writing conditional output routines in those patterns.
9929:
9930: If this macro is not defined, it is equivalent to a null statement.
9931:
9932: @item PRINT_OPERAND (@var{stream}, @var{x}, @var{code})
9933: A C compound statement to output to stdio stream @var{stream} the
9934: assembler syntax for an instruction operand @var{x}. @var{x} is an
9935: RTL expression.
9936:
9937: @var{code} is a value that can be used to specify one of several ways
9938: of printing the operand. It is used when identical operands must be
9939: printed differently depending on the context. @var{code} comes from
9940: the @samp{%} specification that was used to request printing of the
9941: operand. If the specification was just @samp{%@var{digit}} then
9942: @var{code} is 0; if the specification was @samp{%@var{ltr}
9943: @var{digit}} then @var{code} is the ASCII code for @var{ltr}.
9944:
9945: If @var{x} is a register, this macro should print the register's name.
9946: The names can be found in an array @code{reg_names} whose type is
9947: @code{char *[]}. @code{reg_names} is initialized from
9948: @code{REGISTER_NAMES}.
9949:
9950: When the machine description has a specification @samp{%@var{punct}}
9951: (a @samp{%} followed by a punctuation character), this macro is called
9952: with a null pointer for @var{x} and the punctuation character for
9953: @var{code}.
9954:
1.1.1.8 root 9955: @item PRINT_OPERAND_PUNCT_VALID_P (@var{code})
9956: A C expression which evaluates to true if @var{code} is a valid
9957: punctuation character for use in the @code{PRINT_OPERAND} macro. If
9958: @code{PRINT_OPERAND_PUNCT_VALID_P} is not defined, it means that no
9959: punctuation characters (except for the standard one, @samp{%}) are used
9960: in this way.
9961:
1.1 root 9962: @item PRINT_OPERAND_ADDRESS (@var{stream}, @var{x})
9963: A C compound statement to output to stdio stream @var{stream} the
9964: assembler syntax for an instruction operand that is a memory reference
9965: whose address is @var{x}. @var{x} is an RTL expression.
9966:
9967: @item ASM_OPEN_PAREN
9968: @itemx ASM_CLOSE_PAREN
9969: These macros are defined as C string constant, describing the syntax
9970: in the assembler for grouping arithmetic expressions. The following
9971: definitions are correct for most assemblers:
9972:
9973: @example
9974: #define ASM_OPEN_PAREN "("
9975: #define ASM_CLOSE_PAREN ")"
9976: @end example
9977: @end table
9978:
9979: @node Config,, Machine Macros, Top
9980: @chapter The Configuration File
9981:
1.1.1.3 root 9982: The configuration file @file{xm-@var{machine}.h} contains macro definitions
9983: that describe the machine and system on which the compiler is running.
9984: Most of the values in it are actually the same on all machines that GNU CC
9985: runs on, so large parts of all configuration files are identical. But
1.1 root 9986: there are some macros that vary:
9987:
9988: @table @code
9989: @item FAILURE_EXIT_CODE
9990: A C expression for the status code to be returned when the compiler
9991: exits after serious errors.
9992:
9993: @item SUCCESS_EXIT_CODE
9994: A C expression for the status code to be returned when the compiler
9995: exits without serious errors.
9996: @end table
9997:
1.1.1.3 root 9998: In addition, configuration files for system V define @code{bcopy},
9999: @code{bzero} and @code{bcmp} as aliases. Some files define @code{alloca}
10000: as a macro when compiled with GNU CC, in order to take advantage of the
10001: benefit of GNU CC's built-in @code{alloca}.
10002:
1.1 root 10003: @contents
10004: @bye
This archive runs on limited infrastructure. Preserving old code on modern bandwidth. Automated agents are requested to crawl responsibly.