![[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:
9: Copyright (C) 1988 Free Software Foundation, Inc.
10:
11: Permission is granted to make and distribute verbatim copies of
12: this manual provided the copyright notice and this permission notice
13: are preserved on all copies.
14:
15: @ignore
16: Permission is granted to process this file through Tex and print the
17: results, provided the printed document carries copying permission
18: notice identical to this one except for the removal of this paragraph
19: (this paragraph not being relevant to the printed manual).
20:
21: @end ignore
22: Permission is granted to copy and distribute modified versions of this
23: manual under the conditions for verbatim copying, provided also that the
24: section entitled ``GNU CC General Public License'' is included exactly as
25: in the original, and provided that the entire resulting derived work is
26: distributed under the terms of a permission notice identical to this one.
27:
28: Permission is granted to copy and distribute translations of this manual
29: into another language, under the above conditions for modified versions,
30: except that the section entitled ``GNU CC General Public License'' and
31: this permission notice may be included in translations approved by the
32: Free Software Foundation instead of in the original English.
33: @end ifinfo
34:
35: @setchapternewpage odd
36:
37: @titlepage
38: @center @titlefont{Using and Porting GNU CC}
39: @sp 2
40: @center Richard M. Stallman
41: @sp 3
42: @center last updated 3 October 1988
43: @sp 1
44: @center for version 1.29
45: @page
46: @vskip 0pt plus 1filll
47: Copyright @copyright{} 1988 Free Software Foundation, Inc.
48:
49: Permission is granted to make and distribute verbatim copies of
50: this manual provided the copyright notice and this permission notice
51: are preserved on all copies.
52:
53: Permission is granted to copy and distribute modified versions of this
54: manual under the conditions for verbatim copying, provided also that the
55: section entitled ``GNU CC General Public License'' is included exactly as
56: in the original, and provided that the entire resulting derived work is
57: distributed under the terms of a permission notice identical to this one.
58:
59: Permission is granted to copy and distribute translations of this manual
60: into another language, under the above conditions for modified versions,
61: except that the section entitled ``GNU CC General Public License'' and
62: this permission notice may be included in translations approved by the
63: Free Software Foundation instead of in the original English.
64: @end titlepage
65: @page
66:
67: @ifinfo
68: @node Top, Copying,, (DIR)
69: @ichapter Introduction
70:
71: This manual documents how to run, install and port the GNU C compiler, as
72: well as its new features and incompatibilities, and how to report bugs.
73:
74: @end ifinfo
75: @menu
76: * Copying:: GNU CC General Public License says
77: how you can copy and share GNU CC.
78: * Contributors:: People who have contributed to GNU CC.
79: * Options:: Command options supported by @samp{gcc}.
80: * Installation:: How to configure, compile and install GNU CC.
81: * Trouble:: If you have trouble installing GNU CC.
82: * Incompatibilities:: Incompatibilities of GNU CC.
83: * Extensions:: GNU extensions to the C language.
84: * Bugs:: How to report bugs (if you want to get them fixed).
85: * Portability:: Goals of GNU CC's portability features.
86: * Interface:: Function-call interface of GNU CC output.
87: * Passes:: Order of passes, what they do, and what each file is for.
88: * RTL:: The intermediate representation that most passes work on.
89: * Machine Desc:: How to write machine description instruction patterns.
90: * Machine Macros:: How to write the machine description C macros.
91: @end menu
92:
93: @node Copying, Contributors, Top, Top
94: @unnumbered GNU CC GENERAL PUBLIC LICENSE
95: @center (Clarified 11 Feb 1988)
96:
97: The license agreements of most software companies keep you at the
98: mercy of those companies. By contrast, our general public license is
99: intended to give everyone the right to share GNU CC. To make sure that
100: you get the rights we want you to have, we need to make restrictions
101: that forbid anyone to deny you these rights or to ask you to surrender
102: the rights. Hence this license agreement.
103:
104: Specifically, we want to make sure that you have the right to give
105: away copies of GNU CC, that you receive source code or else can get it
106: if you want it, that you can change GNU CC or use pieces of it in new
107: free programs, and that you know you can do these things.
108:
109: To make sure that everyone has such rights, we have to forbid you to
110: deprive anyone else of these rights. For example, if you distribute
111: copies of GNU CC, you must give the recipients all the rights that you
112: have. You must make sure that they, too, receive or can get the
113: source code. And you must tell them their rights.
114:
115: Also, for our own protection, we must make certain that everyone
116: finds out that there is no warranty for GNU CC. If GNU CC is modified by
117: someone else and passed on, we want its recipients to know that what
118: they have is not what we distributed, so that any problems introduced
119: by others will not reflect on our reputation.
120:
121: Therefore we (Richard Stallman and the Free Software Foundation,
122: Inc.) make the following terms which say what you must do to be
123: allowed to distribute or change GNU CC.
124:
125: @unnumberedsec COPYING POLICIES
126:
127: @enumerate
128: @item
129: You may copy and distribute verbatim copies of GNU CC source code as
130: you receive it, in any medium, provided that you conspicuously and
131: appropriately publish on each copy a valid copyright notice
132: ``Copyright @copyright{} 1988 Free Software Foundation, Inc.'' (or
133: with whatever year is appropriate); keep intact the notices on all
134: files that refer to this License Agreement and to the absence of any
135: warranty; and give any other recipients of the GNU CC program a copy
136: of this License Agreement along with the program. You may charge a
137: distribution fee for the physical act of transferring a copy.
138:
139: @item
140: You may modify your copy or copies of GNU CC or any portion of it,
141: and copy and distribute such modifications under the terms of
142: Paragraph 1 above, provided that you also do the following:
143:
144: @itemize @bullet
145: @item
146: cause the modified files to carry prominent notices stating
147: that you changed the files and the date of any change; and
148:
149: @item
150: cause the whole of any work that you distribute or publish, that
151: in whole or in part contains or is a derivative of GNU CC or any
152: part thereof, to be licensed at no charge to all third parties on
153: terms identical to those contained in this License Agreement
154: (except that you may choose to grant more extensive warranty
155: protection to some or all third parties, at your option).
156:
157: @item
158: You may charge a distribution fee for the physical act of
159: transferring a copy, and you may at your option offer warranty
160: protection in exchange for a fee.
161: @end itemize
162:
163: Mere aggregation of another unrelated program with this program (or its
164: derivative) on a volume of a storage or distribution medium does not bring
165: the other program under the scope of these terms.
166:
167: @item
168: You may copy and distribute GNU CC (or a portion or derivative of it,
169: under Paragraph 2) in object code or executable form under the terms
170: of Paragraphs 1 and 2 above provided that you also do one of the
171: following:
172:
173: @itemize @bullet
174: @item
175: accompany it with the complete corresponding machine-readable
176: source code, which must be distributed under the terms of
177: Paragraphs 1 and 2 above; or,
178:
179: @item
180: accompany it with a written offer, valid for at least three
181: years, to give any third party free (except for a nominal
182: shipping charge) a complete machine-readable copy of the
183: corresponding source code, to be distributed under the terms of
184: Paragraphs 1 and 2 above; or,
185:
186: @item
187: accompany it with the information you received as to where the
188: corresponding source code may be obtained. (This alternative is
189: allowed only for noncommercial distribution and only if you
190: received the program in object code or executable form alone.)
191: @end itemize
192:
193: For an executable file, complete source code means all the source code
194: for all modules it contains; but, as a special exception, it need not
195: include source code for modules which are standard libraries that
196: accompany the operating system on which the executable file runs.
197:
198: @item
199: You may not copy, sublicense, distribute or transfer GNU CC except as
200: expressly provided under this License Agreement. Any attempt
201: otherwise to copy, sublicense, distribute or transfer GNU CC is void
202: and your rights to use the program under this License agreement shall
203: be automatically terminated. However, parties who have received
204: computer software programs from you with this License Agreement will
205: not have their licenses terminated so long as such parties remain in
206: full compliance.
207:
208: @item
209: If you wish to incorporate parts of GNU CC into other free programs
210: whose distribution conditions are different, write to the Free Software
211: Foundation at 675 Mass Ave, Cambridge, MA 02139. We have not yet worked
212: out a simple rule that can be stated here, but we will often permit this.
213: We will be guided by the two goals of preserving the free status of all
214: derivatives of our free software and of promoting the sharing and reuse of
215: software.
216: @end enumerate
217:
218: Your comments and suggestions about our licensing policies and our
219: software are welcome! Please contact the Free Software Foundation, Inc.,
220: 675 Mass Ave, Cambridge, MA 02139, or call (617) 876-3296.
221:
222: @unnumberedsec NO WARRANTY
223:
224: BECAUSE GNU CC IS LICENSED FREE OF CHARGE, WE PROVIDE ABSOLUTELY NO
225: WARRANTY, TO THE EXTENT PERMITTED BY APPLICABLE STATE LAW. EXCEPT
226: WHEN OTHERWISE STATED IN WRITING, FREE SOFTWARE FOUNDATION, INC,
227: RICHARD M. STALLMAN AND/OR OTHER PARTIES PROVIDE GNU CC "AS IS" WITHOUT
228: WARRANTY OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT
229: LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
230: A PARTICULAR PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND
231: PERFORMANCE OF GNU CC IS WITH YOU. SHOULD GNU CC PROVE DEFECTIVE, YOU
232: ASSUME THE COST OF ALL NECESSARY SERVICING, REPAIR OR CORRECTION.
233:
234: IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW WILL RICHARD M.
235: STALLMAN, THE FREE SOFTWARE FOUNDATION, INC., AND/OR ANY OTHER PARTY
236: WHO MAY MODIFY AND REDISTRIBUTE GNU CC AS PERMITTED ABOVE, BE LIABLE TO
237: YOU FOR DAMAGES, INCLUDING ANY LOST PROFITS, LOST MONIES, OR OTHER
238: SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE USE OR
239: INABILITY TO USE (INCLUDING BUT NOT LIMITED TO LOSS OF DATA OR DATA
240: BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY THIRD PARTIES OR A
241: FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS) GNU CC, EVEN
242: IF YOU HAVE BEEN ADVISED OF THE POSSIBILITY OF SUCH DAMAGES, OR FOR
243: ANY CLAIM BY ANY OTHER PARTY.
244:
245: @node Contributors, Options, Copying, Top
246: @unnumbered Contributors to GNU CC
247:
248: In addition to Richard Stallman, several people have written parts
249: of GNU CC.
250:
251: @itemize @bullet
252: @item
253: The idea of using RTL and some of the optimization ideas came from the
254: U. of Arizona Portable Optimizer, written by Jack Davidson and
255: Christopher Fraser. See ``Register Allocation and Exhaustive Peephole
256: Optimization'', Software Practice and Experience 14 (9), Sept. 1984,
257: 857-866.
258:
259: @item
260: Paul Rubin wrote most of the preprocessor.
261:
262: @item
263: Leonard Tower wrote parts of the parser, RTL generator, RTL
264: definitions, and of the Vax machine description.
265:
266: @item
267: Ted Lemon wrote parts of the RTL reader and printer.
268:
269: @item
270: Jim Wilson implemented loop strength reduction and some other
271: loop optimizations.
272:
273: @item
274: Nobuyuki Hikichi of Software Research Associates, Tokyo, contributed
275: the support for the SONY NEWS machine.
276:
277: @item
278: Charles LaBrec contributed the support for the Integrated Solutions
279: 68020 system.
280:
281: @item
282: Michael Tiemann of MCC wrote most of the description of the National
283: Semiconductor 32000 series cpu. He also wrote the code for inline
284: function integration and for the SPARC cpu and Motorola 88000 cpu
285: and part of the Sun FPA support.
286:
287: @item
288: Jan Stein of the Chalmers Computer Society provided support for
289: Genix, as well as part of the 32000 machine description.
290:
291: @item
292: Randy Smith finished the Sun FPA support.
293:
294: @item
295: Robert Brown implemented the support for Encore 32000 systems.
296:
297: @item
298: David Kashtan of SRI adapted GNU CC to the Vomit-Making System.
299:
300: @item
301: Alex Crain provided changes for the 3b1.
302:
303: @item
304: Greg Satz and Chris Hanson assisted in making GNU CC work on HP-UX for
305: the 9000 series 300.
306:
307: @item
308: William Schelter did most of the work on the Intel 80386 support.
309: @end itemize
310:
311: @node Options, Installation, Contributors, Top
312: @chapter GNU CC Command Options
313:
314: The GNU C compiler uses a command syntax much like the Unix C compiler.
315: The @code{gcc} program accepts options and file names as operands.
316: Multiple single-letter options may @emph{not} be grouped: @samp{-dr} is
317: very different from @samp{-d -r}.
318:
319: When you invoke GNU CC, it normally does preprocessing, compilation,
320: assembly and linking. File names which end in @samp{.c} are taken as C
321: source to be preprocessed and compiled; compiler output files plus any
322: input files with names ending in @samp{.s} are assembled; then the
323: resulting object files, plus any other input files, are linked together to
324: produce an executable.
325:
326: Command options allow you to stop this process at an intermediate stage.
327: For example, the @samp{-c} option says not to run the linker. Then the
328: output consists of object files output by the assembler.
329:
330: Other command options are passed on to one stage. Some options control
331: the preprocessor and others the compiler itself. Yet other options
332: control the assembler and linker; these are not documented here because the
333: GNU assembler and linker are not yet released.
334:
335: Here are the options to control the overall compilation process, including
336: those that say whether to link, whether to assemble, and so on.
337:
338: @table @samp
339: @item -o @var{file}
340: Place output in file @var{file}. This applies regardless to whatever
341: sort of output is being produced, whether it be an executable file,
342: an object file, an assembler file or preprocessed C code.
343:
344: If @samp{-o} is not specified, the default is to put an executable file
345: in @file{a.out}, the object file @file{@var{source}.c} in
346: @file{@var{source}.o}, an assembler file in @file{@var{source}.s}, and
347: preprocessed C on standard output.@refill
348:
349: @item -c
350: Compile or assemble the source files, but do not link. Produce object
351: files with names made by replacing @samp{.c} or @samp{.s} with
352: @samp{.o} at the end of the input file names. Do nothing at all for
353: object files specified as input.
354:
355: @item -S
356: Compile into assembler code but do not assemble. The assembler output
357: file name is made by replacing @samp{.c} with @samp{.s} at the end of
358: the input file name. Do nothing at all for assembler source files or
359: object files specified as input.
360:
361: @item -E
362: Run only the C preprocessor. Preprocess all the C source files
363: specified and output the results to standard output.
364:
365: @item -v
366: Compiler driver program prints the commands it executes as it runs
367: the preprocessor, compiler proper, assembler and linker. Some of
368: these are directed to print their own version numbers.
369:
370: @item -B@var{prefix}
371: Compiler driver program tries @var{prefix} as a prefix for each
372: program it tries to run. These programs are @file{cpp}, @file{cc1},
373: @file{as} and @file{ld}.
374:
375: For each subprogram to be run, the compiler driver first tries the
376: @samp{-B} prefix, if any. If that name is not found, or if @samp{-B}
377: was not specified, the driver tries two standard prefixes, which are
378: @file{/usr/lib/gcc-} and @file{/usr/local/lib/gcc-}. If neither of
379: those results in a file name that is found, the unmodified program
380: name is searched for using the directories specified in your
381: @samp{PATH} environment variable.
382:
383: The run-time support file @file{gnulib} is also searched for using
384: the @samp{-B} prefix, if needed. If it is not found there, the two
385: standard prefixes above are tried, and that is all. The file is left
386: out of the link if it is not found by those means. Most of the time,
387: on most machines, you can do without it.
388: @end table
389:
390: These options control the details of C compilation itself.
391:
392: @table @samp
393: @item -ansi
394: Support all ANSI standard C programs.
395:
396: This turns off certain features of GNU C that are incompatible with
397: ANSI C, such as the @code{asm}, @code{inline} and @code{typeof}
398: keywords, and predefined macros such as @code{unix} and @code{vax}
399: that identify the type of system you are using. It also enables the
400: undesirable and rarely used ANSI trigraph feature.
401:
402: The @samp{-ansi} option does not cause non-ANSI programs to be
403: rejected gratuitously. For that, @samp{-pedantic} is required in
404: addition to @samp{-ansi}.
405:
406: The macro @code{__STRICT_ANSI__} is predefined when the @samp{-ansi}
407: option is used. Some header files may notice this macro and refrain
408: from declaring certain functions or defining certain macros that the
409: ANSI standard doesn't call for; this is to avoid interfering with
410: any programs that might use these names for other things.
411:
412: @item -traditional
413: Attempt to support some aspects of traditional C compilers.
414: Specifically:
415:
416: @itemize @bullet
417: @item
418: All @code{extern} declarations take effect globally even if they
419: are written inside of a function definition. This includes implicit
420: declarations of functions.
421:
422: @item
423: The keywords @code{typeof}, @code{inline}, @code{signed}, @code{const}
424: and @code{volatile} are not recognized.@refill
425:
426: @item
427: Comparisons between pointers and integers are always allowed.
428:
429: @item
430: Integer types @code{unsigned short} and @code{unsigned char} promote
431: to @code{unsigned int}.
432:
433: @item
434: Out-of-range floating point literals are not an error.
435:
436: @item
437: In the preprocessor, comments convert to nothing at all, rather than
438: to a space. This allows traditional token concatenation.
439:
440: @item
441: In the preprocessor, macro arguments are recognized within string
442: constants in a macro definition (and their values are stringified,
443: though without additional quote marks, when they appear in such a
444: context). The preprocessor always considers a string constant to end
445: at a newline.
446:
447: @item
448: The predefined macro @code{__STDC__} is not defined when you use
449: @samp{-traditional}, but @code{__GNUC__} is (since the GNU extensions
450: which @code{__GNUC__} indicates are not affected by
451: @samp{-traditional}). If you need to write header files that work
452: differently depending on whether @samp{-traditional} is in use, by
453: testing both of these predefined macros you can distinguish four
454: situations: GNU C, traditional GNU C, other ANSI C compilers, and
455: other old C compilers.
456: @end itemize
457:
458: @item -O
459: Optimize. Optimizing compilation takes somewhat more time, and a lot
460: more memory for a large function.
461:
462: Without @samp{-O}, the compiler's goal is to reduce the cost of
463: compilation and to make debugging produce the expected results.
464: Statements are independent: if you stop the program with a breakpoint
465: between statements, you can then assign a new value to any variable or
466: change the program counter to any other statement in the function and
467: get exactly the results you would expect from the source code.
468:
469: Without @samp{-O}, only variables declared @code{register} are
470: allocated in registers. The resulting compiled code is a little worse
471: than produced by PCC without @samp{-O}.
472:
473: With @samp{-O}, the compiler tries to reduce code size and execution
474: time.
475:
476: Some of the @samp{-f} options described below turn specific kinds of
477: optimization on or off.
478:
479: @item -g
480: Produce debugging information in the operating system's native format
481: (for DBX or SDB). GDB also can work with this debugging information.
482:
483: Unlike most other C compilers, GNU CC allows you to use @samp{-g} with
484: @samp{-O}. The shortcuts taken by optimized code may occasionally
485: produce surprising results: some variables you declared may not exist
486: at all; flow of control may briefly move where you did not expect it;
487: some statements may not be executed because they compute constant
488: results or their values were already at hand; some statements may
489: execute in different places because they were moved out of loops.
490: Nevertheless it proves possible to debug optimized output. This makes
491: it reasonable to use the optimizer for programs that might have bugs.
492:
493: @item -gg
494: Produce debugging information in GDB's own format. This requires the
495: GNU assembler and linker in order to work.
496:
497: This feature will probably be eliminated. It was intended to enable
498: GDB to read the symbol table faster, but it doesn't result in enough
499: of a speedup to be worth the larger object files and executables. We
500: are working on other ways of making GDB start faster, which work with
501: DBX format debugging information and could be made to work with SDB
502: format.
503:
504: @item -w
505: Inhibit all warning messages.
506:
507: @item -W
508: Print extra warning messages for these events:
509:
510: @itemize @bullet
511: @item
512: An automatic variable is used without first being initialized.
513:
514: These warnings are possible only in optimizing compilation,
515: because they require data flow information that is computed only
516: when optimizing. They occur only for variables that are
517: candidates for register allocation. Therefore, they do not occur
518: for a variable that is declared @code{volatile}, or whose address
519: is taken, or whose size is other than 1, 2, 4 or 8 bytes. Also,
520: they do not occur for structures, unions or arrays, even when
521: they are in registers.
522:
523: Note that there may be no warning about a variable that is used
524: only to compute a value that itself is never used, because such
525: computations may be deleted by the flow analysis pass before the
526: warnings are printed.
527:
528: These warnings are made optional because GNU CC is not smart
529: enough to see all the reasons why the code might be correct
530: despite appearing to have an error. Here is one example of how
531: this can happen:
532:
533: @example
534: @{
535: int x;
536: switch (y)
537: @{
538: case 1: x = 1;
539: break;
540: case 2: x = 4;
541: break;
542: case 3: x = 5;
543: @}
544: foo (x);
545: @}
546: @end example
547:
548: @noindent
549: If the value of @code{y} is always 1, 2 or 3, then @code{x} is
550: always initialized, but GNU CC doesn't know this. Here is
551: another common case:
552:
553: @example
554: @{
555: int save_y;
556: if (change_y) save_y = y, y = new_y;
557: @dots{}
558: if (change_y) y = save_y;
559: @}
560: @end example
561:
562: @noindent
563: This has no bug because @code{save_y} is used only if it is set.
564:
565: @item
566: A nonvolatile automatic variable might be changed by a call to
567: @code{longjmp}. These warnings as well are possible only in
568: optimizing compilation.
569:
570: The compiler sees only the calls to @code{setjmp}. It cannot know
571: where @code{longjmp} will be called; in fact, a signal handler could
572: call it at any point in the code. As a result, you may get a warning
573: even when there is in fact no problem because @code{longjmp} cannot
574: in fact be called at the place which would cause a problem.
575:
576: @item
577: A function can return either with or without a value. (Falling
578: off the end of the function body is considered returning without
579: a value.) For example, this function would inspire such a
580: warning:
581:
582: @example
583: foo (a)
584: @{
585: if (a > 0)
586: return a;
587: @}
588: @end example
589:
590: Spurious warnings can occur because GNU CC does not realize that
591: certain functions (including @code{abort} and @code{longjmp})
592: will never return.
593: @end itemize
594:
595: In the future, other useful warnings may also be enabled by this
596: option.
597:
598: @item -Wimplicit
599: Warn whenever a function is implicitly declared.
600:
601: @item -Wreturn-type
602: Warn whenever a function is defined with a return-type that defaults
603: to @code{int}. Also warn about any @code{return} statement with no
604: return-value in a function whose return-type is not @code{void}.
605:
606: @item -Wunused
607: Warn whenever a local variable is unused aside from its declaration.
608:
609: @item -Wcomment
610: Warn whenever a comment-start sequence @samp{/*} appears in a comment.
611:
612: @item -Wall
613: All of the above @samp{-W} options combined.
614:
615: @item -Wwrite-strings
616: Give string constants the type @code{const char[@var{length}]} so that
617: copying the address of one into a non-@code{const} @code{char *}
618: pointer will get a warning. These warnings will help you find at
619: compile time code that can try to write into a string constant, but
620: only if you have been very careful about using @code{const} in
621: declarations and prototypes. Otherwise, it will just be a nuisance;
622: this is why we did not make @samp{-Wall} request these warnings.
623:
624: @item -p
625: Generate extra code to write profile information suitable for the
626: analysis program @code{prof}.
627:
628: @item -pg
629: Generate extra code to write profile information suitable for the
630: analysis program @code{gprof}.
631:
632: @item -l@var{library}
633: Search a standard list of directories for a library named
634: @var{library}, which is actually a file named
635: @file{lib@var{library}.a}. The linker uses this file as if it
636: had been specified precisely by name.
637:
638: The directories searched include several standard system directories
639: plus any that you specify with @samp{-L}.
640:
641: Normally the files found this way are library files---archive files
642: whose members are object files. The linker handles an archive file by
643: scanning through it for members which define symbols that have so far
644: been referenced but not defined. But if the file that is found is an
645: ordinary object file, it is linked in the usual fashion. The only
646: difference between using an @samp{-l} option and specifying a file name
647: is that @samp{-l} searches several directories.
648:
649: @item -L@var{dir}
650: Add directory @var{dir} to the list of directories to be searched
651: for @samp{-l}.
652:
653: @item -nostdlib
654: Don't use the standard system libraries and startup files when
655: linking. Only the files you specify (plus @file{gnulib}) will be
656: passed to the linker.
657:
658: @item -m@var{machinespec}
659: Machine-dependent option specifying something about the type of target
660: machine. These options are defined by the macro
661: @code{TARGET_SWITCHES} in the machine description. The default for
662: the options is also defined by that macro, which enables you to change
663: the defaults.@refill
664:
665: These are the @samp{-m} options defined in the 68000 machine
666: description:
667:
668: @table @samp
669: @item -m68020
670: @itemx -mc68020
671: Generate output for a 68020 (rather than a 68000). This is the
672: default if you use the unmodified sources.
673:
674: @item -m68000
675: @item -mc68000
676: Generate output for a 68000 (rather than a 68020).
677:
678: @item -m68881
679: Generate output containing 68881 instructions for floating point.
680: This is the default if you use the unmodified sources.
681:
682: @item -mfpa
683: Generate output containing Sun FPA instructions for floating point.
684:
685: @item -msoft-float
686: Generate output containing library calls for floating point.
687:
688: @item -mshort
689: Consider type @code{int} to be 16 bits wide, like @code{short int}.
690:
691: @item -mnobitfield
692: Do not use the bit-field instructions. @samp{-m68000} implies
693: @samp{-mnobitfield}.
694:
695: @item -mbitfield
696: Do use the bit-field instructions. @samp{-m68020} implies
697: @samp{-mbitfield}. This is the default if you use the unmodified
698: sources.
699:
700: @item -mrtd
701: Use a different function-calling convention, in which functions
702: that take a fixed number of arguments return with the @code{rtd}
703: instruction, which pops their arguments while returning. This
704: saves one instruction in the caller since there is no need to pop
705: the arguments there.
706:
707: This calling convention is incompatible with the one normally
708: used on Unix, so you cannot use it if you need to call libraries
709: compiled with the Unix compiler.
710:
711: Also, you must provide function prototypes for all functions that
712: take variable numbers of arguments (including @code{printf});
713: otherwise incorrect code will be generated for calls to those
714: functions.
715:
716: In addition, seriously incorrect code will result if you call a
717: function with too many arguments. (Normally, extra arguments are
718: harmlessly ignored.)
719:
720: The @code{rtd} instruction is supported by the 68010 and 68020
721: processors, but not by the 68000.
722: @end table
723:
724: These @samp{-m} options are defined in the Vax machine description:
725:
726: @table @samp
727: @item -munix
728: Do not output certain jump instructions (@code{aobleq} and so on)
729: that the Unix assembler for the Vax cannot handle across long
730: ranges.
731:
732: @item -mgnu
733: Do output those jump instructions, on the assumption that you
734: will assemble with the GNU assembler.
735:
736: @item -mg
737: Output code for g-format floating point numbers instead of d-format.
738: @end table
739:
740: @item -f@var{flag}
741: Specify machine-independent flags. These are the flags:
742:
743: @table @samp
744: @item -ffloat-store
745: Do not store floating-point variables in registers. This
746: prevents undesirable excess precision on machines such as the
747: 68000 where the floating registers (of the 68881) keep more
748: precision than a @code{double} is supposed to have.
749:
750: For most programs, the excess precision does only good, but a few
751: programs rely on the precise definition of IEEE floating point.
752: Use @samp{-ffloat-store} for such programs.
753:
754: @item -fno-asm
755: Do not recognize @code{asm}, @code{inline} or @code{typeof} as a
756: keyword. These words may then be used as identifiers.
757:
758: @item -fno-defer-pop
759: Always pop the arguments to each function call as soon as that
760: function returns. Normally the compiler (when optimizing) lets
761: arguments accumulate on the stack for several function calls and
762: pops them all at once.
763:
764: @item -fstrength-reduce
765: Perform the optimizations of loop strength reduction and
766: elimination of iteration variables.
767:
768: @item -fcombine-regs
769: Allow the combine pass to combine an instruction that copies one
770: register into another. This might or might not produce better
771: code when used in addition to @samp{-O}. I am interested in
772: hearing about the difference this makes.
773:
774: @item -fforce-mem
775: Force memory operands to be copied into registers before doing
776: arithmetic on them. This may produce better code by making all
777: memory references potential common subexpressions. When they are
778: not common subexpressions, instruction combination should
779: eliminate the separate register-load. I am interested in hearing
780: about the difference this makes.
781:
782: @item -fforce-addr
783: Force memory address constants to be copied into registers before
784: doing arithmetic on them. This may produce better code just as
785: @samp{-fforce-mem} may. I am interested in hearing about the
786: difference this makes.
787:
788: @item -fomit-frame-pointer
789: Don't keep the frame pointer in a register for functions that
790: don't need one. This avoids the instructions to save, set up and
791: restore frame pointers; it also makes an extra register available
792: in many functions. @strong{It also makes debugging impossible.}
793:
794: On some machines, such as the Vax, this flag has no effect,
795: because the standard calling sequence automatically handles the
796: frame pointer and nothing is saved by pretending it doesn't
797: exist. The machine-description macro
798: @code{FRAME_POINTER_REQUIRED} controls whether a target machine
799: supports this flag. @xref{Registers}.@refill
800:
801: @item -finline-functions
802: Integrate all simple functions into their callers. The compiler
803: heuristically decides which functions are simple enough to be
804: worth integrating in this way.
805:
806: If all calls to a given function are integrated, and the function
807: is declared @code{static}, then the function is normally not
808: output as assembler code in its own right.
809:
810: @item -fkeep-inline-functions
811: Even if all calls to a given function are integrated, and the
812: function is declared @code{static}, nevertheless output a
813: separate run-time callable version of the function.
814:
815: @item -fwritable-strings
816: Store string constants in the writable data segment and don't
817: uniquize them. This is for compatibility with old programs which
818: assume they can write into string constants. Writing into string
819: constants is a very bad idea; ``constants'' should be constant.
820:
821: @item -fno-function-cse
822: Do not put function addresses in registers; make each instruction
823: that calls a constant function contain the function's address
824: explicitly.
825:
826: This option results in less efficient code, but some strange
827: hacks that alter the assembler output may be confused by the
828: optimizations performed when this option is not used.
829:
830: @item -fvolatile
831: Consider all memory references through pointers to be volatile.
832:
833: @item -funsigned-char
834: Let the type @code{char} be the unsigned, like @code{unsigned
835: char}.
836:
837: Each kind of machine has a default for what @code{char} should
838: be. It is either like @code{unsigned char} by default or like
839: @code{signed char} by default. (Actually, at present, the
840: default is always signed.)
841:
842: The type @code{char} is always a distinct type from either
843: @code{signed char} or @code{unsigned char}, even though its
844: behavior is always just like one of those two.
845:
846: @item -fsigned-char
847: Let the type @code{char} be signed, like @code{signed char}.
848:
849: @item -ffixed-@var{reg}
850: Treat the register named @var{reg} as a fixed register; generated
851: code should never refer to it (except perhaps as a stack pointer,
852: frame pointer or in some other fixed role).
853:
854: @var{reg} must be the name of a register. The register names
855: accepted are machine-specific and are defined in the
856: @code{REGISTER_NAMES} macro in the machine description macro
857: file.
858:
859: @item -fcall-used-@var{reg}
860: Treat the register named @var{reg} as an allocatable register
861: that is clobbered by function calls. It may be allocated for
862: temporaries or variables that do not live across a call.
863: Functions compiled this way will not save and restore the
864: register @var{reg}.
865:
866: Use of this flag for a register that has a fixed pervasive role
867: in the machine's execution model, such as the stack pointer or
868: frame pointer, will produce disastrous results.
869:
870: @item -fcall-saved-@var{reg}
871: Treat the register named @var{reg} as an allocatable register
872: saved by functions. It may be allocated even for temporaries or
873: variables that live across a call. Functions compiled this way
874: will save and restore the register @var{reg} if they use it.
875:
876: Use of this flag for a register that has a fixed pervasive role
877: in the machine's execution model, such as the stack pointer or
878: frame pointer, will produce disastrous results.
879:
880: A different sort of disaster will result from the use of this
881: flag for a register in which function values may be returned.
882: @end table
883:
884: @item -d@var{letters}
885: Says to make debugging dumps at times specified by @var{letters}.
886: Here are the possible letters:
887:
888: @table @samp
889: @item r
890: Dump after RTL generation.
891: @item j
892: Dump after first jump optimization.
893: @item J
894: Dump after last jump optimization.
895: @item s
896: Dump after CSE (including the jump optimization that sometimes
897: follows CSE).
898: @item L
899: Dump after loop optimization.
900: @item f
901: Dump after flow analysis.
902: @item c
903: Dump after instruction combination.
904: @item l
905: Dump after local register allocation.
906: @item g
907: Dump after global register allocation.
908: @item m
909: Print statistics on memory usage, at the end of the run.
910: @end table
911:
912: @item -pedantic
913: Issue all the warnings demanded by strict ANSI standard C; reject
914: all programs that use forbidden extensions.
915:
916: Valid ANSI standard C programs should compile properly with or without
917: this option (though a rare few will require @samp{-ansi}). However,
918: without this option, certain GNU extensions and traditional C features
919: are supported as well. With this option, they are rejected. There is
920: no reason to @i{use} this option; it exists only to satisfy pedants.
921: @end table
922:
923: These options control the C preprocessor, which is run on each C source
924: file before actual compilation. If you use the @samp{-E} option, nothing
925: is done except C preprocessing. Some of these options make sense only
926: together with @samp{-E} because they request preprocessor output that is
927: not suitable for actual compilation.
928:
929: @table @samp
930: @item -C
931: Tell the preprocessor not to discard comments. Used with the
932: @samp{-E} option.
933:
934: @item -I@var{dir}
935: Search directory @var{dir} for include files.
936:
937: @item -I-
938: Any directories specified with @samp{-I} options before the @samp{-I-}
939: option are searched only for the case of @samp{#include "@var{file}"};
940: they are not searched for @samp{#include <@var{file}>}.
941:
942: If additional directories are specified with @samp{-I} options after
943: the @samp{-I-}, these directories are searched for all @samp{#include}
944: directives. (Ordinarily @emph{all} @samp{-I} directories are used
945: this way.)
946:
947: In addition, the @samp{-I-} option inhibits the use of the current
948: directory as the first search directory for @samp{#include
949: "@var{file}"}. Therefore, the current directory is searched only if
950: it is requested explicitly with @samp{-I.}. Specifying both
951: @samp{-I-} and @samp{-I.} allows you to control precisely which
952: directories are searched before the current one and which are searched
953: after.
954:
955: @item -nostdinc
956: Do not search the standard system directories for header files. Only
957: the directories you have specified with @samp{-I} options (and the
958: current directory, if appropriate) are searched.
959:
960: Between @samp{-nostdinc} and @samp{-I-}, you can eliminate all
961: directories from the search path except those you specify.
962:
963: @item -M
964: Tell the preprocessor to output a rule suitable for @code{make}
965: describing the dependencies of each source file. For each source
966: file, the preprocessor outputs one @code{make}-rule whose target is
967: the object file name for that source file and whose dependencies are
968: all the files @samp{#include}d in it. This rule may be a single line
969: or may be continued with @samp{\}-newline if it is long.
970:
971: @samp{-M} implies @samp{-E}.
972:
973: @item -MM
974: Like @samp{-M} but the output mentions only the user-header files
975: included with @samp{#include "@var{file}"}. System header files
976: included with @samp{#include <@var{file}>} are omitted.
977:
978: @samp{-MM} implies @samp{-E}.
979:
980: @item -D@var{macro}
981: Define macro @var{macro} with the empty string as its definition.
982:
983: @item -D@var{macro}=@var{defn}
984: Define macro @var{macro} as @var{defn}.
985:
986: @item -U@var{macro}
987: Undefine macro @var{macro}.
988:
989: @item -T
990: Support ANSI C trigraphs. You don't want to know about this
991: brain-damage. The @samp{-ansi} option also has this effect.
992: @end table
993:
994: @node Installation, Trouble, Options, Top
995: @chapter Installing GNU CC
996:
997: Here is the procedure for installing GNU CC on a Unix system.
998: @menu
999: * VMS Install:: See below for installation on VMS.
1000: @end menu
1001: @iftex
1002: (See below for VMS.)
1003: @end iftex
1004:
1005: @enumerate
1006: @item
1007: Edit @file{Makefile}. If you are using HPUX, or any form of system V,
1008: you must make a few changes described in comments at the beginning of
1009: the file.
1010:
1011: @item
1012: On a Sequent system, go to the Berkeley universe.
1013:
1014: @item
1015: Choose configuration files.
1016:
1017: @itemize @bullet
1018: @item
1019: Make a symbolic link named @file{config.h} to the top-level
1020: config file for the machine you are using (@pxref{Config}). This
1021: file is responsible for defining information about the host
1022: machine. It includes @file{tm.h}.
1023:
1024: The file's name should be @file{config-@var{machine}.h}, with these
1025: exceptions:
1026:
1027: @table @file
1028: @item config-vms.h
1029: for vaxen running VMS.
1030: @item config-vaxv.h
1031: for vaxen running system V.
1032: @item config-i386v.h
1033: for Intel 80386's running system V.
1034: @item config-sun4.h
1035: for Suns (model 2, 3 or 4) running @emph{operating system} version 4.
1036: @item config-hp9k3.h
1037: for the HP 9000 series 300.
1038: @item config-gnx.h
1039: for the ns32000 running Genix
1040: @end table
1041:
1042: If your system does not support symbolic links, you might want to
1043: set up @file{config.h} to contain a @samp{#include} command which
1044: refers to the appropriate file.
1045:
1046: @item
1047: Make a symbolic link named @file{tm.h} to the machine-description
1048: macro file for your machine (its name should be
1049: @file{tm-@var{machine}.h}).
1050:
1051: If your system is a 68000, don't use the file @file{tm-m68k.h}
1052: directly. Instead, use one of these files:
1053:
1054: @table @file
1055: @item tm-sun3.h
1056: for Sun 3 machines.
1057: @item tm-sun2.h
1058: for Sun 2 machines.
1059: @item tm-3b1.h
1060: for AT&T 3b1 (aka 7300 Unix PC).
1061: @item tm-isi68.h
1062: for Integrated Solutions systems.
1063: @item tm-news800.h
1064: for SONY News systems.
1065: @item tm-hp9k320.h
1066: for HPUX systems, if you are using GNU CC with the system's
1067: assembler and linker.
1068: @item tm-hp9k320g.h
1069: for HPUX systems, if you are using the GNU assembler, linker and
1070: other utilities. Not all of the pieces of GNU software needed
1071: for this mode of operation are as yet in distribution; full
1072: instructions will appear here in the future.@refill
1073: @end table
1074:
1075: For the vax, use @file{tm-vax.h} on BSD Unix, @file{tm-vaxv.h} on
1076: system V, or @file{tm-vms.h} on VMS.@refill
1077:
1078: For the SPARC (Sun 4), use @file{tm-sparc.h}.
1079:
1080: For the Motorola 88000, use @file{tm-m88k.h}. The support for the
1081: 88000 has a few unfinished spots because there was no way to run the
1082: output. Bugs are suspected in handling of branch-tables and in
1083: the function prologue and epilogue.
1084:
1085: For the 80386, don't use @file{tm-i386.h} directly. Use
1086: @file{tm-i386v.h} if the target machine is running system V,
1087: @file{tm-seq386.h} for a Sequent 386 system, or @file{tm-compaq.h} for
1088: a Compaq.
1089:
1090: For the 32000, use @file{tm-sequent.h} if you are using a Sequent
1091: machine, or @file{tm-encore.h} for an Encore machine, or
1092: @file{tm-gnx.h} if you are using Genix version 3; otherwise, perhaps
1093: @file{tm-ns32k.h} will work for you.
1094:
1095: Note that Genix has bugs in @code{alloca} and @code{malloc}; you must
1096: get the compiled versions of these from GNU Emacs and edit GNU CC's
1097: @file{Makefile} to use them.
1098:
1099: Note that Encore systems are supported only under BSD.
1100:
1101: @item
1102: Make a symbolic link named @file{md} to the machine description
1103: pattern file (its name should be @file{@var{machine}.md}).
1104:
1105: @item
1106: Make a symbolic link named @file{aux-output.c} to the output
1107: subroutine file for your machine (its name should be
1108: @file{output-@var{machine}.c}).
1109: @end itemize
1110:
1111: @item
1112: Make sure the Bison parser generator is installed. (This is
1113: unnecessary if the Bison output files @file{c-parse.tab.c} and
1114: @file{cexp.c} are more recent than @file{c-parse.y} and @file{cexp.y}
1115: and you do not plan to change the @samp{.y} files.)
1116:
1117: Bison versions older that Sept 8, 1988 will produce incorrect output
1118: for @file{c-parse.tab.c}.
1119:
1120: @item
1121: If you are using a Sun, make sure the environment variable
1122: @code{FLOAT_OPTION} is not set. If this option were set to
1123: @code{f68881} when @file{gnulib} is compiled, the resulting code would
1124: demand to be linked with a special startup file and will not link
1125: properly without special pains.
1126:
1127: @item
1128: Build the compiler. Just type @samp{make} in the compiler directory.
1129:
1130: @item
1131: Move the first-stage object files and executables into a subdirectory
1132: with this command:
1133:
1134: @example
1135: make stage1
1136: @end example
1137:
1138: The files are moved into a subdirectory named @file{stage1}.
1139: Once installation is complete, you may wish to delete these files
1140: with @code{rm -r stage1}.
1141:
1142: @item
1143: Recompile the compiler with itself, with this command:
1144:
1145: @example
1146: make CC=stage1/gcc CFLAGS="-g -O -Bstage1/"
1147: @end example
1148:
1149: On a 68000 or 68020 system lacking floating point hardware,
1150: unless you have selected a @file{tm.h} file that expects by default
1151: that there is no such hardware, do this instead:
1152:
1153: @example
1154: make CC=stage1/gcc CFLAGS="-g -O -Bstage1/ -msoft-float"
1155: @end example
1156:
1157: @item
1158: If you wish to test the compiler by compiling it with itself one more
1159: time, do this:
1160:
1161: @example
1162: make stage2
1163: make CC=stage2/gcc CFLAGS="-g -O -Bstage2/"
1164: foreach file (*.o)
1165: cmp $file stage2/$file
1166: end
1167: @end example
1168:
1169: This will notify you if any of these stage 3 object files differs from
1170: those of stage 2. Any difference, no matter how innocuous, indicates
1171: that the stage 2 compiler has compiled GNU CC incorrectly, and is
1172: therefore a potentially serious bug which you should investigate and
1173: report (@pxref{Bugs}).
1174:
1175: Aside from the @samp{-B} option, the options should be the same as
1176: when you made stage 2.
1177:
1178: @item
1179: Install the compiler driver, the compiler's passes and run-time support.
1180: You can use the following command:
1181:
1182: @example
1183: make install
1184: @end example
1185:
1186: @noindent
1187: This copies the files @file{cc1}, @file{cpp} and @file{gnulib} to
1188: files @file{gcc-cc1}, @file{gcc-cpp} and @file{gcc-gnulib} in
1189: directory @file{/usr/local/lib}, which is where the compiler driver
1190: program looks for them. It also copies the driver program @file{gcc}
1191: into the directory @file{/usr/local}, so that it appears in typical
1192: execution search paths.@refill
1193:
1194: @strong{Warning: there is a bug in @code{alloca} in the Sun library.
1195: To avoid this bug, install the binaries of GNU CC that were compiled
1196: by GNU CC. They use @code{alloca} as a built-in function and never
1197: the one in the library.}
1198:
1199: @strong{Warning: the GNU CPP may not work for @file{ioctl.h},
1200: @file{ttychars.h} and other system header files unless the
1201: @samp{-traditional} option is used.} The bug is in the header files:
1202: at least on some machines, they rely on behavior that is incompatible
1203: with ANSI C. This behavior consists of substituting for macro
1204: argument names when they appear inside of character constants. The
1205: @samp{-traditional} option tells GNU CC to behave the way these
1206: headers expect.
1207:
1208: Because of this problem, you might prefer to configure GNU CC to use
1209: the system's own C preprocessor. To do so, make the file
1210: @file{/usr/local/lib/gcc-cpp} a link to @file{/lib/cpp}.
1211:
1212: Alternatively, on Sun systems and 4.3BSD at least, you can correct the
1213: include files by running the shell script @file{fixincludes}. This
1214: installs modified, corrected copies of the files @file{ioctl.h},
1215: @file{ttychars.h} and many others, in a special directory where only
1216: GNU CC will normally look for them.
1217:
1218: See the file @file{fixincludes} for a list of all the files we know to
1219: require correction.
1220: @end enumerate
1221:
1222: If you cannot install the compiler's passes and run-time support in
1223: @file{/usr/local/lib}, you can alternatively use the @samp{-B} option to
1224: specify a prefix by which they may be found. The compiler concatenates
1225: the prefix with the names @file{cpp}, @file{cc1} and @file{gnulib}.
1226: Thus, you can put the files in a directory @file{/usr/foo/gcc} and
1227: specify @samp{-B/usr/foo/gcc/} when you run GNU CC.
1228:
1229: Also, you can specify an alternative default directory for these files
1230: by setting the Make variable @code{libdir} when you make GNU CC.
1231:
1232: @node VMS Install,, Installation, Installation
1233: @section Installing GNU CC on VMS
1234:
1235: The VMS version of GNU CC is distributed in an unusual tape format which
1236: consists of several tape files. The first is a command file; the second is
1237: an executable program which reads Unix tar format; the third is another
1238: command file which uses this program to read the remainder of the tape.
1239:
1240: To load the tape, it suffices to mount it @samp{/foreign} and then do
1241: @samp{@@mta0:} to execute the command file at the beginning of the tape.
1242:
1243: The tape contains executables and object files as well as sources, so no
1244: compilation is necessary unless you change the sources. (This is a good
1245: thing, since you probably don't have any other C compiler.) If you must
1246: recompile, here is how:
1247:
1248: @enumerate
1249: @item
1250: Copy the file @file{tm-vms.h} to @file{tm.h}, @file{config-vms.h} to
1251: @file{config.h}, @file{vax.md} to @file{md.} and @file{output-vax.c}
1252: to @file{aux-output.c}.@refill
1253:
1254: @item
1255: Type @samp{@@make} to do recompile everything.
1256: @end enumerate
1257:
1258: To install the @samp{GCC} command so you can use the compiler easily, in
1259: the same manner as you use the VMS C compiler, you must install the VMS CLD
1260: file for GNU CC as follows:
1261:
1262: @enumerate
1263: @item
1264: Define the VMS logical names @samp{GNU_CC} and @samp{GNU_CC_INCLUDE}
1265: to point to the directories where the GNU CC executables
1266: (@samp{gcc-cpp}, @samp{gcc-cc1}, etc.) and the C include files are
1267: kept. This should be done with the commands:@refill
1268:
1269: @example
1270: $ assign /super /system disk:[gcc] gnu_cc
1271: $ assign /super /system disk:[gcc.include] gnu_cc_include
1272: @end example
1273:
1274: @noindent
1275: with the appropriate disk and directory names. These commands can be
1276: placed in your system startup file so they will be executed whenever
1277: the machine is rebooted.
1278:
1279: @item
1280: Install the @samp{GCC} command with the command line:
1281:
1282: @example
1283: $ set command /table=sys$library:dcltables gnu_cc:gcc
1284: @end example
1285:
1286: @noindent
1287: Now you can invoke the compiler with a command like @samp{gcc /verbose
1288: file.c}, which is equivalent to the command @samp{gcc -v -c file.c} in
1289: Unix.
1290: @end enumerate
1291:
1292: @node Trouble, Incompatibilities, Installation, Top
1293: @chapter Known Causes of Trouble with GNU CC.
1294:
1295: Here are some of the things that have caused trouble for people installing
1296: or using GNU CC.
1297:
1298: @itemize @bullet
1299: @item
1300: On certain systems, defining certain environment variables such as
1301: @samp{CC} can interfere with the functioning of @code{make}.
1302:
1303: @item
1304: Cross compilation can run into trouble for certain machines because
1305: some target machines' assemblers require floating point numbers to be
1306: written as @emph{integer} constants in certain contexts.
1307:
1308: The compiler writes these integer constants by examining the floating
1309: point value as an integer and printing that integer, because this is
1310: simple to write and independent of the details of the floating point
1311: representation. But this does not work if the compiler is running on
1312: a different machine with an incompatible floating point format, or
1313: even a different byte-ordering.
1314:
1315: It is possible to fix this by writing machine-independent code which
1316: understands the floating point representation of the target machine.
1317: I am not interested in doing that much work to compensate for bugs
1318: in assemblers.
1319:
1320: @item
1321: DBX rejects some files produced by GNU CC, though it accepts similar
1322: constructs in output from PCC. Until someone can supply a coherent
1323: description of what is valid DBX input and what is not, there is
1324: nothing I can do about these problems. You are on your own.
1325: @end itemize
1326:
1327: @node Incompatibilities, Extensions, Trouble, Top
1328: @chapter Incompatibilities of GNU CC
1329:
1330: There are several noteworthy incompatibilities between GNU C and most
1331: existing (non-ANSI) versions of C.
1332:
1333: Ultimately our intention is that the @samp{-traditional} option will
1334: eliminate most of these incompatibilities by telling GNU C to behave
1335: like the other C compilers.
1336:
1337: @itemize @bullet
1338: @item
1339: GNU CC normally makes string constants read-only. If several
1340: identical-looking string constants are used, GNU CC stores only one
1341: copy of the string.
1342:
1343: One consequence is that you cannot call @code{mktemp} with a string
1344: constant argument. The function @code{mktemp} always alters the
1345: string its argument points to.
1346:
1347: Another consequence is that @code{sscanf} does not work on some
1348: systems when passed a string constant as its format control string.
1349: This is because @code{sscanf} incorrectly tries to write into the
1350: string constant.
1351:
1352: The best solution to these problems is to change the program to use
1353: @code{char}-array variables with initialization strings for these
1354: purposes instead of string constants. But if this is not possible,
1355: you can use the @samp{-fwritable-strings} flag, which directs GNU CC
1356: to handle string constants the same way most C compilers do.
1357:
1358: @item
1359: GNU CC does not substitute macro arguments when they appear inside of
1360: string constants. For example, the following macro in GNU CC
1361:
1362: @example
1363: #define foo(a) "a"
1364: @end example
1365:
1366: @noindent
1367: will produce output @samp{"a"} regardless of what the argument @var{a} is.
1368:
1369: The @samp{-traditional} option directs GNU CC to handle such cases
1370: (among others) in the old-fashioned (non-ANSI) fashion.
1371:
1372: @item
1373: When you use @code{setjmp} and @code{longjmp}, the only automatic
1374: variables guaranteed to remain valid are those declared
1375: @code{volatile}. This is a consequence of automatic register
1376: allocation. Consider this function:
1377:
1378: @example
1379: jmp_buf j;
1380:
1381: foo ()
1382: @{
1383: int a, b;
1384:
1385: a = fun1 ();
1386: if (setjmp (j))
1387: return a;
1388:
1389: a = fun2 ();
1390: /* @r{@code{longjmp (j)} may be occur in @code{fun3}.} */
1391: return a + fun3 ();
1392: @}
1393: @end example
1394:
1395: Here @code{a} may or may not be restored to its first value when the
1396: @code{longjmp} occurs. If @code{a} is allocated in a register, then
1397: its first value is restored; otherwise, it keeps the last value stored
1398: in it.
1399:
1400: If you use the @samp{-W} option with the @samp{-O} option, you will
1401: get a warning when GNU CC thinks such a problem might be possible.
1402:
1403: @item
1404: Declarations of external variables and functions within a block apply
1405: only to the block containing the declaration. In other words, they
1406: have the same scope as any other declaration in the same place.
1407:
1408: In some other C compilers, a @code{extern} declaration affects all the
1409: rest of the file even if it happens within a block.
1410:
1411: The @samp{-traditional} option directs GNU C to treat all @code{extern}
1412: declarations as global, like traditional compilers.
1413:
1414: @item
1415: In traditional C, you can combine @code{long}, etc., with a typedef name,
1416: as shown here:
1417:
1418: @example
1419: typedef int foo;
1420: typedef long foo bar;
1421: @end example
1422:
1423: In ANSI C, this is not allowed: @code{long} and other type modifiers
1424: require an explicit @code{int}. Because this criterion is expressed
1425: by Bison grammar rules rather than C code, the @samp{-traditional}
1426: flag cannot alter it.
1427:
1428: @item
1429: PCC allows typedef names to be used as function parameters. The
1430: difficulty described immediately above applies here too.
1431:
1432: @item
1433: PCC allows whitespace in the middle of compound assignment operators
1434: such as @samp{+=}. GNU CC, following the ANSI standard, does not
1435: allow this. The difficulty described immediately above applies here
1436: too.
1437:
1438: @item
1439: GNU CC will flag unterminated character constants inside of preprocessor
1440: conditionals that fail. Some programs have English comments enclosed in
1441: conditionals that are guaranteed to fail; if these comments contain
1442: apostrophes, GNU CC will probably report an error. For example,
1443: this code would produce an error:
1444:
1445: @example
1446: #if 0
1447: You can't expect this to work.
1448: #endif
1449: @end example
1450:
1451: The best solution to such a problem is to put the text into an actual
1452: C comment delimited by @samp{/*@dots{}*/}. However,
1453: @samp{-traditional} suppresses these error messages.
1454:
1455: @item
1456: When compiling functions that return @code{float}, PCC converts it to
1457: a double. GNU CC actually returns a @code{float}. If you are concerned
1458: with PCC compatibility, you should declare your functions to return
1459: @code{double}; you might as well say what you mean.
1460:
1461: @item
1462: When compiling functions that return structures or unions, GNU CC
1463: output code uses a method different from that used on most versions of
1464: Unix. As a result, code compiled with GNU CC cannot call a
1465: structure-returning function compiled with PCC, and vice versa.
1466:
1467: The method used by GCC is as follows: a structure or union which is 1,
1468: 2, 4 or 8 bytes long is returned like a scalar. A structure or union
1469: with any other size is stored into an address supplied by the caller
1470: in a special, fixed register.
1471:
1472: PCC usually handles all sizes of structures and unions by returning
1473: the address of a block of static storage containing the value. This
1474: method is not used in GCC because it is slower and nonreentrant.
1475:
1476: On systems where PCC works this way, you may be able to make GCC-compiled
1477: code call such functions that were compiled with PCC by declaring them
1478: to return a pointer to the structure or union instead of the structure
1479: or union itself. For example, instead of this:
1480:
1481: @example
1482: struct foo nextfoo ();
1483: @end example
1484:
1485: @noindent
1486: write this:
1487:
1488: @example
1489: struct foo *nextfoo ();
1490: #define nextfoo *nextfoo
1491: @end example
1492:
1493: @noindent
1494: (Note that this assumes you are using the GNU preprocessor and not
1495: @samp{-traditional}, so that the ANSI antirecursion rules for macro
1496: expansions are effective.)
1497: @end itemize
1498:
1499: @node Extensions, Bugs, Incompatibilities, Top
1500: @chapter GNU Extensions to the C Language
1501:
1502: GNU C provides several language features not found in ANSI standard C.
1503: (The @samp{-pedantic} option directs GNU CC to print a warning message if
1504: any of these features is used.) To test for the availability of these
1505: features in conditional compilation, check for a predefined macro
1506: @code{__GNUC__}, which is always defined under GNU CC.
1507:
1508: @menu
1509: * Statement Exprs:: Putting statements and declarations inside expressions.
1510: * Naming Types:: Giving a name to the type of some expression.
1511: * Typeof:: @code{typeof}: referring to the type of an expression.
1512: * Lvalues:: Using @samp{?:}, @samp{,} and casts in lvalues.
1513: * Conditionals:: Omitting the middle operand of a @samp{?:} expression.
1514: * Zero-Length:: Zero-length arrays.
1515: * Variable-Length:: Arrays whose length is computed at run time.
1516: * Subscripting:: Any array can be subscripted, even if not an lvalue.
1517: * Pointer Arith:: Arithmetic on @code{void}-pointers and function pointers.
1518: * Constructors:: Constructor expressions give structures, unions
1519: or arrays as values.
1520: * Dollar Signs:: Dollar sign is allowed in identifiers.
1521: * Alignment:: Inquiring about the alignment of a type or variable.
1522: * Inline:: Defining inline functions (as fast as macros).
1523: * Extended Asm:: Assembler instructions with C expressions as operands.
1524: (With them you can define ``built-in'' functions.)
1525: * Asm Labels:: Specifying the assembler name to use for a C symbol.
1526: @end menu
1527:
1528: @node Statement Exprs, Naming Types, Extensions, Extensions
1529: @section Statements and Declarations inside of Expressions
1530:
1531: A compound statement in parentheses may appear inside an expression in GNU
1532: C. This allows you to declare variables within an expression. For
1533: example:
1534:
1535: @example
1536: (@{ int y = foo (); int z;
1537: if (y > 0) z = y;
1538: else z = - y;
1539: z; @})
1540: @end example
1541:
1542: @noindent
1543: is a valid (though slightly more complex than necessary) expression
1544: for the absolute value of @code{foo ()}.
1545:
1546: This feature is especially useful in making macro definitions ``safe'' (so
1547: that they evaluate each operand exactly once). For example, the
1548: ``maximum'' function is commonly defined as a macro in standard C as
1549: follows:
1550:
1551: @example
1552: #define max(a,b) ((a) > (b) ? (a) : (b))
1553: @end example
1554:
1555: @noindent
1556: But this definition computes either @var{a} or @var{b} twice, with bad
1557: results if the operand has side effects. In GNU C, if you know the
1558: type of the operands (here let's assume @code{int}), you can define
1559: the macro safely as follows:
1560:
1561: @example
1562: #define maxint(a,b) \
1563: (@{int _a = (a), _b = (b); _a > _b ? _a : _b; @})
1564: @end example
1565:
1566: Embedded statements are not allowed in constant expressions, such as
1567: the value of an enumeration constant, the width of a bit field, or
1568: the initial value of a static variable.
1569:
1570: If you don't know the type of the operand, you can still do this, but you
1571: must use @code{typeof} (@pxref{Typeof}) or type naming (@pxref{Naming
1572: Types}).
1573:
1574: @node Naming Types, Typeof, Statement Exprs, Extensions
1575: @section Naming an Expression's Type
1576:
1577: You can give a name to the type of an expression using a @code{typedef}
1578: declaration with an initializer. Here is how to define @var{name} as a
1579: type name for the type of @var{exp}:
1580:
1581: @example
1582: typedef @var{name} = @var{exp};
1583: @end example
1584:
1585: This is useful in conjunction with the statements-within-expressions
1586: feature. Here is how the two together can be used to define a safe
1587: ``maximum'' macro that operates on any arithmetic type:
1588:
1589: @example
1590: #define max(a,b) \
1591: (@{typedef _ta = (a), _tb = (b); \
1592: _ta _a = (a); _tb _b = (b); \
1593: _a > _b ? _a : _b; @})
1594: @end example
1595:
1596: The reason for using names that start with underscores for the local
1597: variables is to avoid conflicts with variable names that occur within the
1598: expressions that are substituted for @code{a} and @code{b}. Eventually we
1599: hope to design a new form of declaration syntax that allows you to declare
1600: variables whose scopes start only after their initializers; this will be a
1601: more reliable way to prevent such conflicts.
1602:
1603: @node Typeof, Lvalues, Naming Types, Extensions
1604: @section Referring to a Type with @code{typeof}
1605:
1606: Another way to refer to the type of an expression is with @code{typeof}.
1607: The syntax of using of this keyword looks like @code{sizeof}, but the
1608: construct acts semantically like a type name defined with @code{typedef}.
1609:
1610: There are two ways of writing the argument to @code{typeof}: with an
1611: expression or with a type. Here is an example with an expression:
1612:
1613: @example
1614: typeof (x[0](1))
1615: @end example
1616:
1617: @noindent
1618: This assumes that @code{x} is an array of functions; the type described
1619: is that of the values of the functions.
1620:
1621: Here is an example with a typename as the argument:
1622:
1623: @example
1624: typeof (int *)
1625: @end example
1626:
1627: @noindent
1628: Here the type described is that of pointers to @code{int}.
1629:
1630: A @code{typeof}-construct can be used anywhere a typedef name could be
1631: used. For example, you can use it in a declaration, in a cast, or inside
1632: of @code{sizeof} or @code{typeof}.
1633:
1634: @itemize @bullet
1635: @item
1636: This declares @code{y} with the type of what @code{x} points to.
1637:
1638: @example
1639: typeof (*x) y;
1640: @end example
1641:
1642: @item
1643: This declares @code{y} as an array of such values.
1644:
1645: @example
1646: typeof (*x) y[4];
1647: @end example
1648:
1649: @item
1650: This declares @code{y} as an array of pointers to characters:
1651:
1652: @example
1653: typeof (typeof (char *)[4]) y;
1654: @end example
1655:
1656: @noindent
1657: It is equivalent to the following traditional C declaration:
1658:
1659: @example
1660: char *y[4];
1661: @end example
1662:
1663: To see the meaning of the declaration using @code{typeof}, and why it
1664: might be a useful way to write, let's rewrite it with these macros:
1665:
1666: @example
1667: #define pointer(T) typeof(T *)
1668: #define array(T, N) typeof(T [N])
1669: @end example
1670:
1671: @noindent
1672: Now the declaration can be rewritten this way:
1673:
1674: @example
1675: array (pointer (char), 4) y;
1676: @end example
1677:
1678: @noindent
1679: Thus, @samp{array (pointer (char), 4)} is the type of arrays of 4
1680: pointers to @code{char}.
1681: @end itemize
1682:
1683: @node Lvalues, Conditionals, Typeof, Extensions
1684: @section Generalized Lvalues
1685:
1686: Compound expressions, conditional expressions and casts are allowed as
1687: lvalues provided their operands are lvalues. This means that you can take
1688: their addresses or store values into them.
1689:
1690: For example, a compound expression can be assigned, provided the last
1691: expression in the sequence is an lvalue. These two expressions are
1692: equivalent:
1693:
1694: @example
1695: (a, b) += 5
1696: a, (b += 5)
1697: @end example
1698:
1699: Similarly, the address of the compound expression can be taken. These two
1700: expressions are equivalent:
1701:
1702: @example
1703: &(a, b)
1704: a, &b
1705: @end example
1706:
1707: A conditional expression is a valid lvalue if its type is not void and the
1708: true and false branches are both valid lvalues. For example, these two
1709: expressions are equivalent:
1710:
1711: @example
1712: (a ? b : c) = 5
1713: (a ? b = 5 : (c = 5))
1714: @end example
1715:
1716: A cast is a valid lvalue if its operand is valid. Taking the address of
1717: the cast is the same as taking the address without a cast, except for the
1718: type of the result. For example, these two expressions are equivalent (but
1719: the second may be valid when the type of @samp{a} does not permit a cast to
1720: @samp{int *}).
1721:
1722: @example
1723: &(int *)a
1724: (int **)&a
1725: @end example
1726:
1727: A simple assignment whose left-hand side is a cast works by converting the
1728: right-hand side first to the specified type, then to the type of the inner
1729: left-hand side expression. After this is stored, the value is converter
1730: back to the specified type to become the value of the assignment. Thus, if
1731: @samp{a} has type @samp{char *}, the following two expressions are
1732: equivalent:
1733:
1734: @example
1735: (int)a = 5
1736: (int)(a = (char *)5)
1737: @end example
1738:
1739: An assignment-with-arithmetic operation such as @samp{+=} applied to a cast
1740: performs the arithmetic using the type resulting from the cast, and then
1741: continues as in the previous case. Therefore, these two expressions are
1742: equivalent:
1743:
1744: @example
1745: (int)a += 5
1746: (int)(a = (char *) ((int)a + 5))
1747: @end example
1748:
1749: @node Conditionals, Zero-Length, Lvalues, Extensions
1750: @section Conditional Expressions with Omitted Middle-Operands
1751:
1752: The middle operand in a conditional expression may be omitted. Then
1753: if the first operand is nonzero, its value is the value of the conditional
1754: expression.
1755:
1756: Therefore, the expression
1757:
1758: @example
1759: x ? : y
1760: @end example
1761:
1762: @noindent
1763: has the value of @code{x} if that is nonzero; otherwise, the value of
1764: @code{y}.
1765:
1766: This example is perfectly equivalent to
1767:
1768: @example
1769: x ? x : y
1770: @end example
1771:
1772: @noindent
1773: In this simple case, the ability to omit the middle operand is not
1774: especially useful. When it becomes useful is when the first operand does,
1775: or may (if it is a macro argument), contain a side effect. Then repeating
1776: the operand in the middle would perform the side effect twice. Omitting
1777: the middle operand uses the value already computed without the undesirable
1778: effects of recomputing it.
1779:
1780: @node Zero-Length, Variable-Length, Conditionals, Extensions
1781: @section Arrays of Length Zero
1782:
1783: Zero-length arrays are allowed in GNU C. They are very useful as the last
1784: element of a structure which is really a header for a variable-length
1785: object:
1786:
1787: @example
1788: struct line @{
1789: int length;
1790: char contents[0];
1791: @};
1792:
1793: @{
1794: struct line *thisline
1795: = (struct line *) malloc (sizeof (struct line) + this_length);
1796: thisline->length = this_length;
1797: @}
1798: @end example
1799:
1800: In standard C, you would have to give @code{contents} a length of 1, which
1801: means either you waste space or complicate the argument to @code{malloc}.
1802:
1803: @node Variable-Length, Subscripting, Zero-Length, Extensions
1804: @section Arrays of Variable Length
1805:
1806: Variable-length automatic arrays are allowed in GNU C. These arrays are
1807: declared like any other automatic arrays, but with a length that is not a
1808: constant expression. The storage is allocated at that time and
1809: deallocated when the brace-level is exited. For example:
1810:
1811: @example
1812: FILE *concat_fopen (char *s1, char *s2, char *mode)
1813: @{
1814: char str[strlen (s1) + strlen (s2) + 1];
1815: strcpy (str, s1);
1816: strcat (str, s2);
1817: return fopen (str, mode);
1818: @}
1819: @end example
1820:
1821: You can also define structure types containing variable-length arrays, and
1822: use them even for arguments or function values, as shown here:
1823:
1824: @example
1825: int foo;
1826:
1827: struct entry
1828: @{
1829: char data[foo];
1830: @};
1831:
1832: struct entry
1833: tester (struct entry arg)
1834: @{
1835: struct entry new;
1836: int i;
1837: for (i = 0; i < foo; i++)
1838: new.data[i] = arg.data[i] + 1;
1839: return new;
1840: @}
1841: @end example
1842:
1843: @noindent
1844: (Eventually there will be a way to say that the size of the array is
1845: another member of the same structure.)
1846:
1847: The length of an array is computed on entry to the brace-level where the
1848: array is declared and is remembered for the scope of the array in case you
1849: access it with @code{sizeof}.
1850:
1851: Jumping or breaking out of the scope of the array name will also deallocate
1852: the storage. Jumping into the scope is not allowed; you will get an error
1853: message for it.
1854:
1855: You can use the function @code{alloca} to get an effect much like
1856: variable-length arrays. The function @code{alloca} is available in
1857: many other C implementations (but not in all). On the other hand,
1858: variable-length arrays are more elegant.
1859:
1860: There are other differences between these two methods. Space allocated
1861: with @code{alloca} exists until the containing @emph{function} returns.
1862: The space for a variable-length array is deallocated as soon as the array
1863: name's scope ends. (If you use both variable-length arrays and
1864: @code{alloca} in the same function, deallocation of a variable-length array
1865: will also deallocate anything more recently allocated with @code{alloca}.)
1866:
1867: @node Subscripting, Pointer Arith, Variable-Length, Extensions
1868: @section Non-Lvalue Arrays May Have Subscripts
1869:
1870: Subscripting is allowed on arrays that are not lvalues, even though the
1871: unary @samp{&} operator is not. For example, this is valid in GNU C though
1872: not valid in other C dialects:
1873:
1874: @example
1875: struct foo @{int a[4];@};
1876:
1877: struct foo f();
1878:
1879: bar (int index)
1880: @{
1881: return f().a[index];
1882: @}
1883: @end example
1884:
1885: @node Pointer Arith, Initializers, Subscripting, Extensions
1886: @section Arithmetic on @code{void}-Pointers and Function Pointers
1887:
1888: In GNU C, addition and subtraction operations are supported on pointers to
1889: @code{void} and on pointers to functions. This is done by treating the
1890: size of a @code{void} or of a function as 1.
1891:
1892: A consequence of this is that @code{sizeof} is also allowed on @code{void}
1893: and on function types, and returns 1.
1894:
1895: @node Initializers, Constructors, Pointer Arith, Extensions
1896: @section Non-Constant Initializers
1897:
1898: The elements of an aggregate initializer are not required to be constant
1899: expressions in GNU C. Here is an example of an initializer with run-time
1900: varying elements:
1901:
1902: @example
1903: foo (float f, float g)
1904: @{
1905: float beat_freqs[2] = @{ f-g, f+g @};
1906: @dots{}
1907: @}
1908: @end example
1909:
1910: @node Constructors, Dollar Signs, Initializers, Extensions
1911: @section Constructor Expressions
1912:
1913: GNU C supports constructor expressions. A constructor looks like a cast
1914: containing an initializer. Its value is an object of the type specified in
1915: the cast, containing the elements specified in the initializer. The type
1916: must be a structure, union or array type.
1917:
1918: Assume that @code{struct foo} and @code{structure} are declared as shown:
1919:
1920: @example
1921: struct foo @{int a; char b[2];@} structure;
1922: @end example
1923:
1924: @noindent
1925: Here is an example of constructing a @samp{struct foo} with a constructor:
1926:
1927: @example
1928: structure = ((struct foo) @{x + y, 'a', 0@});
1929: @end example
1930:
1931: @noindent
1932: This is equivalent to writing the following:
1933:
1934: @example
1935: @{
1936: struct foo temp = @{x + y, 'a', 0@};
1937: structure = temp;
1938: @}
1939: @end example
1940:
1941: You can also construct an array. If all the elements of the constructor
1942: are (made up of) simple constant expressions, suitable for use in
1943: initializers, then the constructor is an lvalue and can be coerced to a
1944: pointer to its first element, as shown here:
1945:
1946: @example
1947: char **foo = (char *[]) @{ "x", "y", "z" @};
1948: @end example
1949:
1950: Array constructors whose elements are not simple constants are not very
1951: useful, because the constructor is not an lvalue. There are only two valid
1952: ways to use it: to subscript it, or initialize an array variable with it.
1953: The former is probably slower than a @code{switch} statement, while the
1954: latter does the same thing an ordinary C initializer would do.
1955:
1956: @example
1957: output = ((int[]) @{ 2, x, 28 @}) [input];
1958: @end example
1959:
1960: @node Dollar Signs, Alignment, Constructors, Extensions
1961: @section Dollar Signs in Identifier Names
1962:
1963: In GNU C, you may use dollar signs in identifier names. This is because
1964: many traditional C implementations allow such identifiers.
1965:
1966: @node Alignment, Inline, Dollar Signs, Extensions
1967: @section Inquiring about the Alignment of a Type or Variable
1968:
1969: The keyword @code{__alignof} allows you to inquire about how an object
1970: is aligned, or the minimum alignment usually required by a type. Its
1971: syntax is just like @code{sizeof}.
1972:
1973: For example, if the target machine requires a @code{double} value to be
1974: aligned on an 8-byte boundary, then @code{__alignof (double)} is 8. This
1975: is true on many RISC machines. On more traditional machine designs,
1976: @code{__alignof (double)} is 4 or even 2.
1977:
1978: Some machines never actually require alignment; they allow reference to any
1979: data type even at an odd addresses. For these machines, @code{__alignof}
1980: reports the @emph{recommended} alignment of a type.
1981:
1982: When the operand of @code{__alignof} is an lvalue rather than a type, the
1983: value is the largest alignment that the lvalue is known to have. It may
1984: have this alignment as a result of its data type, or because it is part of
1985: a structure and inherits alignment from that structure. For example, after
1986: this declaration:
1987:
1988: @example
1989: struct foo @{ int x; char y; @} foo1;
1990: @end example
1991:
1992: @noindent
1993: the value of @code{__alignof (foo1.y)} is probably 2 or 4, the same as
1994: @code{__alignof (int)}, even though the data type of @code{foo1.y} does not
1995: itself demand any alignment.@refill
1996:
1997: @node Inline, Extended Asm, Alignment, Extensions
1998: @section An Inline Function is As Fast As a Macro
1999:
2000: By declaring a function @code{inline}, you can direct GNU CC to integrate
2001: that function's code into the code for its callers. This makes execution
2002: faster by eliminating the function-call overhead; in addition, if any of
2003: the actual argument values are constant, their known values may permit
2004: simplifications at compile time so that not all of the inline function's
2005: code needs to be included.
2006:
2007: To declare a function inline, use the @code{inline} keyword in its
2008: declaration, like this:
2009:
2010: @example
2011: inline int
2012: inc (int *a)
2013: @{
2014: (*a)++;
2015: @}
2016: @end example
2017:
2018: You can also make all ``simple enough'' functions inline with the
2019: option @samp{-finline-functions}. Note that certain usages in a
2020: function definition can make it unsuitable for inline substitution.
2021:
2022: When a function is both inline and @code{static}, if all calls to the
2023: function are integrated into the caller, then the function's own assembler
2024: code is never referenced. In this case, GNU CC does not actually output
2025: assembler code for the function, unless you specify the option
2026: @samp{-fkeep-inline-functions}. Some calls cannot be integrated for
2027: various reasons (in particular, calls that precede the function's
2028: definition cannot be integrated, and neither can recursive calls within the
2029: definition). If there is a nonintegrated call, then the function is
2030: compiled to assembler code as usual.
2031:
2032: When an inline function is not @code{static}, then the compiler must assume
2033: that there may be calls from other source files; since a global symbol can
2034: be defined only once in any program, the function must not be defined in
2035: the other source files, so the calls therein cannot be integrated.
2036: Therefore, a non-@code{static} inline function is always compiled on its
2037: own in the usual fashion.
2038:
2039: @node Extended Asm, Asm Labels, Inline, Extensions
2040: @section Assembler Instructions with C Expression Operands
2041:
2042: In an assembler instruction using @code{asm}, you can now specify the
2043: operands of the instruction using C expressions. This means no more
2044: guessing which registers or memory locations will contain the data you want
2045: to use.
2046:
2047: You must specify an assembler instruction template much like what appears
2048: in a machine description, plus an operand constraint string for each
2049: operand.
2050:
2051: For example, here is how to use the 68881's @code{fsinx} instruction:
2052:
2053: @example
2054: asm ("fsinx %1,%0" : "=f" (result) : "f" (angle));
2055: @end example
2056:
2057: @noindent
2058: Here @code{angle} is the C expression for the input operand while
2059: @code{result} is that of the output operand. Each has @samp{"f"} as its
2060: operand constraint, saying that a floating-point register is required. The
2061: constraints use the same language used in the machine description
2062: (@pxref{Constraints}).
2063:
2064: Each operand is described by an operand-constraint string followed by the C
2065: expression in parentheses. A colon separates the assembler template from
2066: the first output operand, and another separates the last output operand
2067: from the first input, if any. Commas separate output operands and separate
2068: inputs. The number of operands is limited to the maximum number of
2069: operands in any instruction pattern in the machine description.
2070:
2071: Output operand expressions must be lvalues; the compiler can check this.
2072: The input operands need not be lvalues. The compiler cannot check whether
2073: the operands have data types that are reasonable for the instruction being
2074: executed. It does not parse the assembler instruction template and does
2075: not know what it means, or whether it is valid assembler input. The
2076: extended @code{asm} feature is most often used for machine instructions
2077: that the compiler itself does not know exist.
2078:
2079: If there are no output operands, and there are input operands, then you
2080: should write two colons in a row where the output operands would go.
2081:
2082: The output operands must be write-only; GNU CC will assume that the values
2083: in these operands before the instruction are dead and need not be
2084: generated. For an operand that is read-write, or in which not all bits are
2085: written and the other bits contain useful information, you must logically
2086: split its function into two separate operands, one input operand and one
2087: write-only output operand. The connection between them is expressed by
2088: constraints which say they need to be in the same location when the
2089: instruction executes. You can use the same C expression for both operands,
2090: or different expressions. For example, here we write the (fictitious)
2091: @samp{combine} instruction with @code{bar} as its read-only source operand
2092: and @code{foo} as its read-write destination:
2093:
2094: @example
2095: asm ("combine %2,%0" : "=r" (foo) : "0" (foo), "g" (bar));
2096: @end example
2097:
2098: @noindent
2099: The constraint @samp{"0"} for operand 1 says that it must occupy the same
2100: location as operand 0.
2101:
2102: Only a digit in the constraint can guarantee that one operand will be in
2103: the same place as another. The mere fact that @code{foo} is the value of
2104: both operands is not enough to guarantee that they will be in the same
2105: place in the generated assembler code. The following would not work:
2106:
2107: @example
2108: asm ("combine %2,%0" : "=r" (foo) : "r" (foo), "g" (bar));
2109: @end example
2110:
2111: Various optimizations or reloading could cause operands 0 and 1 to be in
2112: different registers; GNU CC knows no reason not to do so. For example, the
2113: compiler might find a copy of the value of @code{foo} in one register and
2114: use it for operand 1, but generate the output operand 0 in a different
2115: register (copying it afterward to @code{foo}'s own address). Of course,
2116: since the register for operand 1 is not even mentioned in the assembler
2117: code, the result will not work, but GNU CC can't tell that.
2118:
2119: Unless an output operand has the @samp{&} constraint modifier, GNU CC may
2120: allocate it in the same register as an unrelated input operand, on the
2121: assumption that the inputs are consumed before the outputs are produced.
2122: This assumption may be false if the assembler code actually consists of
2123: more than one instruction. In such a case, use @samp{&} for each output
2124: operand that may not overlap an input. @xref{Modifiers}.
2125:
2126: Some instructions clobber specific hard registers. To describe this,
2127: write a third colon after the input operands, followed by the names of
2128: the clobbered hard registers (given as strings). For example, on the vax,
2129:
2130: @example
2131: asm volatile ("movc3 %0,%1,%2"
2132: : /* no outputs */
2133: : "g" (from), "g" (to), "g" (count)
2134: : "r0", "r1", "r2", "r3", "r4", "r5");
2135: @end example
2136:
2137: Usually the most convenient way to use these @code{asm} instructions is to
2138: encapsulate them in macros that look like functions. For example,
2139:
2140: @example
2141: #define sin(x) \
2142: (@{ double __value, __arg = (x); \
2143: asm ("fsinx %1,%0": "=f" (__value): "f" (__arg)); \
2144: __value; @})
2145: @end example
2146:
2147: @noindent
2148: Here the variable @code{__arg} is used to make sure that the instruction
2149: operates on a proper @code{double} value, and to accept only those
2150: arguments @code{x} which can convert automatically to a @code{double}.
2151:
2152: Another way to make sure the instruction operates on the correct data type
2153: is to use a cast in the @code{asm}. This is different from using a
2154: variable @code{__arg} in that it converts more different types. For
2155: example, if the desired type were @code{int}, casting the argument to
2156: @code{int} would accept a pointer with no complaint, while assigning the
2157: argument to an @code{int} variable named @code{__arg} would warn about
2158: using a pointer unless the caller explicitly casts it.
2159:
2160: GNU CC assumes for optimization purposes that these instructions have no
2161: side effects except to change the output operands. This does not mean that
2162: instructions with a side effect cannot be used, but you must be careful,
2163: because the compiler may eliminate them if the output operands aren't used,
2164: or move them out of loops, or replace two with one if they constitute a
2165: common subexpression. Also, if your instruction does have a side effect on
2166: a variable that otherwise appears not to change, the old value of the
2167: variable may be reused later if it happens to be found in a register.
2168:
2169: You can prevent an @code{asm} instruction from being deleted, moved or
2170: combined by writing the keyword @code{volatile} after the @code{asm}. For
2171: example:
2172:
2173: @example
2174: #define set_priority(x) \
2175: asm volatile ("set_priority %0": /* no outputs */ : "g" (x))
2176: @end example
2177:
2178: It is a natural idea to look for a way to give access to the condition
2179: code left by the assembler instruction. However, when we attempted to
2180: implement this, we found no way to make it work reliably. The problem
2181: is that output operands might need reloading, which would result in
2182: additional following ``store'' instructions. On most machines, these
2183: instructions would alter the condition code before there was time to
2184: test it. This problem doesn't arise for ordinary ``test'' and
2185: ``compare'' instructions because they don't have any output operands.
2186:
2187: @node Asm Labels,,Extended Asm, Extensions
2188: @section Controlling Names Used in Assembler Code
2189:
2190: You can specify the name to be used in the assembler code for a C function
2191: or variable by writing the @code{asm} keyword after the declarator as
2192: follows:
2193:
2194: @example
2195: int foo asm ("myfoo") = 2;
2196: @end example
2197:
2198: @noindent
2199: This specifies that the name to be used for the variable @code{foo} in
2200: the assembler code should be @samp{myfoo} rather than the usual
2201: @samp{_foo}.
2202:
2203: On systems where an underscore is normally prepended to the name of a C
2204: function or variable, this feature allows you to define names for the
2205: linker that do not start with an underscore.
2206:
2207: You cannot use @code{asm} in this way in a function @emph{definition}; but
2208: you can get the same effect by writing a declaration for the function
2209: before its definition and putting @code{asm} there, like this:
2210:
2211: @example
2212: extern func () asm ("FUNC");
2213:
2214: func (x, y)
2215: int x, y;
2216: @dots{}
2217: @end example
2218:
2219: It is up to you to make sure that the assembler names you choose do not
2220: conflict with any other assembler symbols. Also, you must not use a
2221: register name; that would produce completely invalid assembler code. GNU
2222: CC does not as yet have the ability to store static variables in registers.
2223: Perhaps that will be added.
2224:
2225: @node Bugs, Portability, Extensions, Top
2226: @chapter Reporting Bugs
2227:
2228: Your bug reports play an essential role in making GNU CC reliable.
2229:
2230: Reporting a bug may help you by bringing a solution to your problem, or it
2231: may not. But in any case the important function of a bug report is to help
2232: the entire community by making the next version of GNU CC work better. Bug
2233: reports are your contribution to the maintenance of GNU CC.
2234:
2235: In order for a bug report to serve its purpose, you must include the
2236: information that makes for fixing the bug.
2237:
2238: @menu
2239: * Criteria: Bug Criteria. Have you really found a bug?
2240: * Reporting: Bug Reporting. How to report a bug effectively.
2241: @end menu
2242:
2243: @node Bug Criteria, Bug Reporting, Bugs, Bugs
2244: @section Have You Found a Bug?
2245:
2246: If you are not sure whether you have found a bug, here are some guidelines:
2247:
2248: @itemize @bullet
2249: @item
2250: If the compiler gets a fatal signal, for any input whatever, that is a
2251: compiler bug. Reliable compilers never crash.
2252:
2253: @item
2254: If the compiler produces invalid assembly code, for any input whatever
2255: (except an @code{asm} statement), that is a compiler bug, unless the
2256: compiler reports errors (not just warnings) which would ordinarily
2257: prevent the assembler from being run.
2258:
2259: @item
2260: If the compiler produces valid assembly code that does not correctly
2261: execute the input source code, that is a compiler bug.
2262:
2263: However, you must double-check to make sure, because you may have run
2264: into an incompatibility between GNU C and traditional C
2265: (@pxref{Incompatibilities}). These incompatibilities might be considered
2266: bugs, but they are inescapable consequences of valuable features.
2267:
2268: Or you may have a program whose behavior is undefined, which happened
2269: by chance to give the desired results with another C compiler.
2270:
2271: For example, in many nonoptimizing compilers, you can write @samp{x;}
2272: at the end of a function instead of @samp{return x;}, with the same
2273: results. But the value of the function is undefined if @samp{return}
2274: is omitted; it is not a bug when GNU CC produces different results.
2275:
2276: Problems often result from expressions with two increment operators,
2277: as in @samp{f (*p++, *p++)}. Your previous compiler might have
2278: interpreted that expression the way you intended; GNU CC might
2279: interpret it another way; neither compiler is wrong.
2280:
2281: After you have localized the error to a single source line, it should
2282: be easy to check for these things. If your program is correct and
2283: well defined, you have found a compiler bug.
2284:
2285: @item
2286: If the compiler produces an error message for valid input, that is a
2287: compiler bug.
2288:
2289: Note that the following is not valid input, and the error message for
2290: it is not a bug:
2291:
2292: @example
2293: int foo (char);
2294:
2295: int
2296: foo (x)
2297: char x;
2298: @{ @dots{} @}
2299: @end example
2300:
2301: @noindent
2302: The prototype says to pass a @code{char}, while the definition says to
2303: pass an @code{int} and treat the value as a @code{char}. This is what
2304: the ANSI standard says, and it makes sense.
2305:
2306: @item
2307: If the compiler does not produce an error message for invalid input,
2308: that is a compiler bug. However, you should note that your idea of
2309: ``invalid input'' might be my idea of ``an extension'' or ``support
2310: for traditional practice''.
2311:
2312: @item
2313: If you are an experienced user of C compilers, your suggestions
2314: for improvement of GNU CC are welcome in any case.
2315: @end itemize
2316:
2317: @node Bug Reporting,, Bug Criteria, Bugs
2318: @section How to Report Bugs
2319:
2320: Send bug reports for GNU C to one of these addresses:
2321:
2322: @example
2323: bug-gcc@@prep.ai.mit.edu
2324: @{ucbvax|mit-eddie|uunet@}!prep.ai.mit.edu!bug-gcc
2325: @end example
2326:
2327: As a last resort, snail them to:
2328:
2329: @example
2330: GNU Compiler Bugs
2331: 545 Tech Sq
2332: Cambridge, MA 02139
2333: @end example
2334:
2335: The fundamental principle of reporting bugs usefully is this:
2336: @strong{report all the facts}. If you are not sure whether to mention a
2337: fact or leave it out, mention it!
2338:
2339: Often people omit facts because they think they know what causes the
2340: problem and they conclude that some details don't matter. Thus, you might
2341: assume that the name of the variable you use in an example does not matter.
2342: Well, probably it doesn't, but one cannot be sure. Perhaps the bug is a
2343: stray memory reference which happens to fetch from the location where that
2344: name is stored in memory; perhaps, if the name were different, the contents
2345: of that location would fool the compiler into doing the right thing despite
2346: the bug. Play it safe and give an exact example.
2347:
2348: If you want to enable me to fix the bug, you should include all these
2349: things:
2350:
2351: @itemize @bullet
2352: @item
2353: The version of GNU CC. You can get this by running it with the
2354: @samp{-v} option.
2355:
2356: Without this, I won't know whether there is any point in looking for
2357: the bug in the current version of GNU CC.
2358:
2359: @item
2360: A complete input file that will reproduce the bug. If the bug is in
2361: the C preprocessor, send me a source file and any header files that it
2362: requires. If the bug is in the compiler proper (@file{cc1}), run your
2363: source file through the C preprocessor by doing @samp{gcc -E
2364: @var{sourcefile} > @var{outfile}}, then include the contents of
2365: @var{outfile} in the bug report. (Any @samp{-I}, @samp{-D} or
2366: @samp{-U} options that you used in actual compilation should also be
2367: used when doing this.)
2368:
2369: A single statement is not enough of an example. In order to compile
2370: it, it must be embedded in a function definition; and the bug might
2371: depend on the details of how this is done.
2372:
2373: Without a real example I can compile, all I can do about your bug
2374: report is wish you luck. It would be futile to try to guess how to
2375: provoke the bug. For example, bugs in register allocation and
2376: reloading frequently depend on every little detail of the function
2377: they happen in.
2378:
2379: @item
2380: The command arguments you gave GNU CC to compile that example and
2381: observe the bug. For example, did you use @samp{-O}? To guarantee
2382: you won't omit something important, list them all.
2383:
2384: If I were to try to guess the arguments, I would probably guess wrong
2385: and then I would not encounter the bug.
2386:
2387: @item
2388: The names of the files that you used for @file{tm.h} and @file{md}
2389: when you installed the compiler.
2390:
2391: @item
2392: The type of machine you are using, and the operating system name and
2393: version number.
2394:
2395: @item
2396: A description of what behavior you observe that you believe is
2397: incorrect. For example, ``It gets a fatal signal,'' or, ``There is an
2398: incorrect assembler instruction in the output.''
2399:
2400: Of course, if the bug is that the compiler gets a fatal signal, then I
2401: will certainly notice it. But if the bug is incorrect output, I might
2402: not notice unless it is glaringly wrong. I won't study all the
2403: assembler code from a 50-line C program just on the off chance that it
2404: might be wrong.
2405:
2406: Even if the problem you experience is a fatal signal, you should still
2407: say so explicitly. Suppose something strange is going on, such as,
2408: your copy of the compiler is out of synch, or you have encountered a
2409: bug in the C library on your system. (This has happened!) Your copy
2410: might crash and mine would not. If you @i{told} me to expect a crash,
2411: then when mine fails to crash, I would know that the bug was not
2412: happening for me. If you had not told me to expect a crash, then I
2413: would not be able to draw any conclusion from my observations.
2414:
2415: In cases where GNU CC generates incorrect code, if you send me a small
2416: complete sample program I will find the error myself by running the
2417: program under a debugger. If you send me a large example or a part of
2418: a larger program, I cannot do this; you must debug the compiled
2419: program and narrow the problem down to one source line. Tell me which
2420: source line it is, and what you believe is incorrect about the code
2421: generated for that line.
2422:
2423: @item
2424: If you send me examples of output from GNU CC, please use @samp{-g}
2425: when you make them. The debugging information includes source line
2426: numbers which are essential for correlating the output with the input.
2427:
2428: @item
2429: If you wish to suggest changes to the GNU CC source, send me context
2430: diffs. If you even discuss something in the GNU CC source, refer to
2431: it by context, not by line number.
2432:
2433: The line numbers in my development sources don't match those in your
2434: sources. Your line numbers would convey no useful information to me.
2435:
2436: @item
2437: Additional information from a debugger might enable me to find
2438: a problem on a machine which I do not have available myself.
2439: However, you need to think when you collect this information if
2440: you want it to have any chance of being useful.
2441:
2442: For example, many people send just a backtrace, but that is never
2443: useful by itself. A simple backtrace with arguments conveys little
2444: about GNU CC because the compiler is largely data-driven; the same
2445: functions are called over and over for different RTL insns, doing
2446: different things depending on the details of the insn.
2447:
2448: Most of the arguments listed in the backtrace are useless because they
2449: are pointers to RTL list structure. The numeric values of the
2450: pointers, which the debugger prints in the backtrace, have no
2451: significance whatever; all that matters is the contents of the objects
2452: they point to (and most of the contents are other such pointers).
2453:
2454: In addition, most compiler passes consist of one or more loops that
2455: scan the RTL insn sequence. The most vital piece of information about
2456: such a loop--which insn it has reached--is usually in a local variable,
2457: not in an argument.
2458:
2459: What you need to provide in addition to a backtrace are the values of
2460: the local variables for several stack frames up. When a local
2461: variable or an argument is an RTX, first print its value and then use
2462: the GDB command @code{pr} to print the RTL expression that it points
2463: to. (If GDB doesn't run on your machine, use your debugger to call
2464: the function @code{debug_rtx} with the RTX as an argument.) In
2465: general, whenever a variable is a pointer, its value is no use
2466: without the data it points to.
2467:
2468: In addition, include a debugging dump from just before the pass
2469: in which the crash happens. Most bugs involve a series of insns,
2470: not just one.
2471: @end itemize
2472:
2473: Here are some things that are not necessary:
2474:
2475: @itemize @bullet
2476: @item
2477: A description of the envelope of the bug.
2478:
2479: Often people who encounter a bug spend a lot of time investigating
2480: which changes to the input file will make the bug go away and which
2481: changes will not affect it.
2482:
2483: This is often time consuming and not very useful, because the way I
2484: will find the bug is by running a single example under the debugger
2485: with breakpoints, not by pure deduction from a series of examples.
2486:
2487: Of course, if you can find a simpler example to report @emph{instead}
2488: of the original one, that is a convenience for me. Errors in the
2489: output will be easier to spot, running under the debugger will take
2490: less time, etc. Most GNU CC bugs involve just one function, so the
2491: most straightforward way to simplify an example is to delete all the
2492: function definitions except the one where the bug occurs. Those
2493: earlier in the file may be replaced by external declarations if the
2494: crucial function depends on them.
2495:
2496: However, simplification is not vital; if you don't want to do this,
2497: report the bug anyway.
2498:
2499: @item
2500: A patch for the bug.
2501:
2502: A patch for the bug does help me if it is a good one. But don't omit
2503: the necessary information, such as the test case, because I might see
2504: problems with your patch and decide to fix the problem another way.
2505:
2506: Sometimes with a program as complicated as GNU CC it is very hard to
2507: construct an example that will make the program follow a certain path
2508: through the code. If you don't send me the example, I won't be able
2509: to construct one, so I won't be able to verify that the bug is fixed.
2510:
2511: @item
2512: A guess about what the bug is or what it depends on.
2513:
2514: Such guesses are usually wrong. Even I can't guess right about such
2515: things without using the debugger to find the facts.
2516: @end itemize
2517:
2518: @node Portability, Interface, Bugs, Top
2519: @chapter GNU CC and Portability
2520:
2521: The main goal of GNU CC was to make a good, fast compiler for machines in
2522: the class that the GNU system aims to run on: 32-bit machines that address
2523: 8-bit bytes and have several general registers. Elegance, theoretical
2524: power and simplicity are only secondary.
2525:
2526: GNU CC gets most of the information about the target machine from a machine
2527: description which gives an algebraic formula for each of the machine's
2528: instructions. This is a very clean way to describe the target. But when
2529: the compiler needs information that is difficult to express in this
2530: fashion, I have not hesitated to define an ad-hoc parameter to the machine
2531: description. The purpose of portability is to reduce the total work needed
2532: on the compiler; it was not of interest for its own sake.
2533:
2534: GNU CC does not contain machine dependent code, but it does contain code
2535: that depends on machine parameters such as endianness (whether the most
2536: significant byte has the highest or lowest address of the bytes in a word)
2537: and the availability of autoincrement addressing. In the RTL-generation
2538: pass, it is often necessary to have multiple strategies for generating code
2539: for a particular kind of syntax tree, strategies that are usable for different
2540: combinations of parameters. Often I have not tried to address all possible
2541: cases, but only the common ones or only the ones that I have encountered.
2542: As a result, a new target may require additional strategies. You will know
2543: if this happens because the compiler will call @code{abort}. Fortunately,
2544: the new strategies can be added in a machine-independent fashion, and will
2545: affect only the target machines that need them.
2546:
2547: @node Interface, Passes, Portability, Top
2548: @chapter Interfacing to GNU CC Output
2549:
2550: GNU CC is normally configured to use the same function calling convention
2551: normally in use on the target system. This is done with the
2552: machine-description macros described (@pxref{Machine Macros}).
2553:
2554: However, returning of structure and union values is done differently on
2555: some target machines. As a result, functions compiled with PCC
2556: returning such types cannot be called from code compiled with GNU CC,
2557: and vice versa. This does not cause trouble often because few Unix
2558: library routines return structures or unions.
2559:
2560: GNU CC code returns structures and unions that are 1, 2, 4 or 8 bytes
2561: long in the same registers used for @code{int} or @code{double} return
2562: values. (GNU CC typically allocates variables of such types in
2563: registers also.) Structures and unions of other sizes are returned by
2564: storing them into an address passed by the caller (usually in a
2565: register). The machine-description macros @code{STRUCT_VALUE} and
2566: @code{STRUCT_INCOMING_VALUE} tell GNU CC where to pass this address.
2567:
2568: By contrast, PCC on most target machines returns structures and unions
2569: of any size by copying the data into an area of static storage, and then
2570: returning the address of that storage as if it were a pointer value.
2571: The caller must copy the data from that memory area to the place where
2572: the value is wanted. This is slower than the method used by GNU CC, and
2573: fails to be reentrant.
2574:
2575: On some target machines, such as RISC machines and the 80386, the
2576: standard system convention is to pass to the subroutine the address of
2577: where to return the value. On these machines, GNU CC has been
2578: configured to be compatible with the standard compiler, when this method
2579: is used. It may not be compatible for structures of 1, 2, 4 or 8 bytes.
2580:
2581: GNU CC uses the system's standard convention for passing arguments. On
2582: some machines, the first few arguments are passed in registers; in
2583: others, all are passed on the stack. It would be possible to use
2584: registers for argument passing on any machine, and this would probably
2585: result in a significant speedup. But the result would be complete
2586: incompatibility with code that follows the standard convention. So this
2587: change is practical only if you are switching to GNU CC as the sole C
2588: compiler for the system. We may implement register argument passing on
2589: certain machines once we have a complete GNU system so that we can
2590: compile the libraries with GNU CC.
2591:
2592: If you use @code{longjmp}, beware of automatic variables. ANSI C says that
2593: automatic variables that are not declared @code{volatile} have undefined
2594: values after a @code{longjmp}. And this is all GNU CC promises to do,
2595: because it is very difficult to restore register variables correctly, and
2596: one of GNU CC's features is that it can put variables in registers without
2597: your asking it to.
2598:
2599: If you want a variable to be unaltered by @code{longjmp}, and you don't
2600: want to write @code{volatile} because old C compilers don't accept it,
2601: just take the address of the variable. If a variable's address is ever
2602: taken, even if just to compute it and ignore it, then the variable cannot
2603: go in a register:
2604:
2605: @example
2606: @{
2607: int careful;
2608: &careful;
2609: @dots{}
2610: @}
2611: @end example
2612:
2613: Code compiled with GNU CC may call certain library routines. Most of
2614: them handle arithmetic for which there are no instructions. This
2615: includes multiply and divide on some machines, and floating point
2616: operations on any machine for which floating point support is disabled
2617: with @samp{-msoft-float}. Some standard parts of the C library, such as
2618: @code{bcopy} or @code{memcpy}, are also called automatically. The usual
2619: function call interface is used for calling the library routines.
2620:
2621: These library routines should be defined in the library @file{gnulib},
2622: which GNU CC automatically searches whenever it links a program. On
2623: machines that have multiply and divide instructions, if hardware
2624: floating point is in use, normally @file{gnulib} is not needed, but it
2625: is searched just in case.
2626:
2627: Each arithmetic function is defined in @file{gnulib.c} to use the
2628: corresponding C arithmetic operator. As long as the file is compiled
2629: with another C compiler, which supports all the C arithmetic operators,
2630: this file will work portably. However, @file{gnulib.c} does not work if
2631: compiled with GNU CC, because each arithmetic function would compile
2632: into a call to itself!
2633:
2634: @node Passes, RTL, Interface, Top
2635: @chapter Passes and Files of the Compiler
2636:
2637: The overall control structure of the compiler is in @file{toplev.c}. This
2638: file is responsible for initialization, decoding arguments, opening and
2639: closing files, and sequencing the passes.
2640:
2641: The parsing pass is invoked only once, to parse the entire input. The RTL
2642: intermediate code for a function is generated as the function is parsed, a
2643: statement at a time. Each statement is read in as a syntax tree and then
2644: converted to RTL; then the storage for the tree for the statement is
2645: reclaimed. Storage for types (and the expressions for their sizes),
2646: declarations, and a representation of the binding contours and how they nest,
2647: remains until the function is finished being compiled; these are all needed
2648: to output the debugging information.
2649:
2650: Each time the parsing pass reads a complete function definition or
2651: top-level declaration, it calls the function
2652: @code{rest_of_compilation} or @code{rest_of_decl_compilation} in
2653: @file{toplev.c}, which are responsible for all further processing
2654: necessary, ending with output of the assembler language. All other
2655: compiler passes run, in sequence, within @code{rest_of_compilation}.
2656: When that function returns from compiling a function definition, the
2657: storage used for that function definition's compilation is entirely
2658: freed, unless it is an inline function (@pxref{Inline}).
2659:
2660: Here is a list of all the passes of the compiler and their source files.
2661: Also included is a description of where debugging dumps can be requested
2662: with @samp{-d} options.
2663:
2664: @itemize @bullet
2665: @item
2666: Parsing. This pass reads the entire text of a function definition,
2667: constructing partial syntax trees. This and RTL generation are no longer
2668: truly separate passes (formerly they were), but it is easier to think
2669: of them as separate.
2670:
2671: The tree representation does not entirely follow C syntax, because it is
2672: intended to support other languages as well.
2673:
2674: C data type analysis is also done in this pass, and every tree node
2675: that represents an expression has a data type attached. Variables are
2676: represented as declaration nodes.
2677:
2678: Constant folding and associative-law simplifications are also done
2679: during this pass.
2680:
2681: The source files for parsing are @file{c-parse.y}, @file{c-decl.c},
2682: @file{c-typeck.c}, @file{c-convert.c}, @file{stor-layout.c},
2683: @file{fold-const.c}, and @file{tree.c}. The last three files are
2684: intended to be language-independent. There are also header files
2685: @file{c-parse.h}, @file{c-tree.h}, @file{tree.h} and @file{tree.def}.
2686: The last two define the format of the tree representation.@refill
2687:
2688: @item
2689: RTL generation. This is the conversion of syntax tree into RTL code.
2690: It is actually done statement-by-statement during parsing, but for
2691: most purposes it can be thought of as a separate pass.
2692:
2693: This is where the bulk of target-parameter-dependent code is found,
2694: since often it is necessary for strategies to apply only when certain
2695: standard kinds of instructions are available. The purpose of named
2696: instruction patterns is to provide this information to the RTL
2697: generation pass.
2698:
2699: Optimization is done in this pass for @code{if}-conditions that are
2700: comparisons, boolean operations or conditional expressions. Tail
2701: recursion is detected at this time also. Decisions are made about how
2702: best to arrange loops and how to output @code{switch} statements.
2703:
2704: The source files for RTL generation are @file{stmt.c}, @file{expr.c},
2705: @file{explow.c}, @file{expmed.c}, @file{optabs.c} and @file{emit-rtl.c}.
2706: Also, the file @file{insn-emit.c}, generated from the machine description
2707: by the program @code{genemit}, is used in this pass. The header files
2708: @file{expr.h} is used for communication within this pass.@refill
2709:
2710: The header files @file{insn-flags.h} and @file{insn-codes.h},
2711: generated from the machine description by the programs @code{genflags}
2712: and @code{gencodes}, tell this pass which standard names are available
2713: for use and which patterns correspond to them.@refill
2714:
2715: Aside from debugging information output, none of the following passes
2716: refers to the tree structure representation of the function (only
2717: part of which is saved).
2718:
2719: The decision of whether the function can and should be expanded inline
2720: in its subsequent callers is made at the end of rtl generation. The
2721: function must meet certain criteria, currently related to the size of
2722: the function and the types and number of parameters it has. Note that
2723: this function may contain loops, recursive calls to itself
2724: (tail-recursive functions can be inlined!), gotos, in short, all
2725: constructs supported by GNU CC.
2726:
2727: The option @samp{-dr} causes a debugging dump of the RTL code after
2728: this pass. This dump file's name is made by appending @samp{.rtl} to
2729: the input file name.
2730:
2731: @item
2732: Jump optimization. This pass simplifies jumps to the following
2733: instruction, jumps across jumps, and jumps to jumps. It deletes
2734: unreferenced labels and unreachable code, except that unreachable code
2735: that contains a loop is not recognized as unreachable in this pass.
2736: (Such loops are deleted later in the basic block analysis.)
2737:
2738: Jump optimization is performed two or three times. The first time is
2739: immediately following RTL generation. The second time is after CSE,
2740: but only if CSE says repeated jump optimization is needed. The
2741: last time is right before the final pass. That time, cross-jumping
2742: and deletion of no-op move instructions are done together with the
2743: optimizations described above.
2744:
2745: The source file of this pass is @file{jump.c}.
2746:
2747: The option @samp{-dj} causes a debugging dump of the RTL code after
2748: this pass is run for the first time. This dump file's name is made by
2749: appending @samp{.jump} to the input file name.
2750:
2751: @item
2752: Register scan. This pass finds the first and last use of each
2753: register, as a guide for common subexpression elimination. Its source
2754: is in @file{regclass.c}.
2755:
2756: @item
2757: Common subexpression elimination. This pass also does constant
2758: propagation. Its source file is @file{cse.c}. If constant
2759: propagation causes conditional jumps to become unconditional or to
2760: become no-ops, jump optimization is run again when CSE is finished.
2761:
2762: The option @samp{-ds} causes a debugging dump of the RTL code after
2763: this pass. This dump file's name is made by appending @samp{.cse} to
2764: the input file name.
2765:
2766: @item
2767: Loop optimization. This pass moves constant expressions out of loops.
2768: Its source file is @file{loop.c}.
2769:
2770: The option @samp{-dL} causes a debugging dump of the RTL code after
2771: this pass. This dump file's name is made by appending @samp{.loop} to
2772: the input file name.
2773:
2774: @item
2775: Stupid register allocation is performed at this point in a
2776: nonoptimizing compilation. It does a little data flow analysis as
2777: well. When stupid register allocation is in use, the next pass
2778: executed is the reloading pass; the others in between are skipped.
2779: The source file is @file{stupid.c}.
2780:
2781: @item
2782: Data flow analysis (@file{flow.c}). This pass divides the program
2783: into basic blocks (and in the process deletes unreachable loops); then
2784: it computes which pseudo-registers are live at each point in the
2785: program, and makes the first instruction that uses a value point at
2786: the instruction that computed the value.
2787:
2788: This pass also deletes computations whose results are never used, and
2789: combines memory references with add or subtract instructions to make
2790: autoincrement or autodecrement addressing.
2791:
2792: The option @samp{-df} causes a debugging dump of the RTL code after
2793: this pass. This dump file's name is made by appending @samp{.flow} to
2794: the input file name. If stupid register allocation is in use, this
2795: dump file reflects the full results of such allocation.
2796:
2797: @item
2798: Instruction combination (@file{combine.c}). This pass attempts to
2799: combine groups of two or three instructions that are related by data
2800: flow into single instructions. It combines the RTL expressions for
2801: the instructions by substitution, simplifies the result using algebra,
2802: and then attempts to match the result against the machine description.
2803:
2804: The option @samp{-dc} causes a debugging dump of the RTL code after
2805: this pass. This dump file's name is made by appending @samp{.combine}
2806: to the input file name.
2807:
2808: @item
2809: Register class preferencing. The RTL code is scanned to find out
2810: which register class is best for each pseudo register. The source
2811: file is @file{regclass.c}.
2812:
2813: @item
2814: Local register allocation (@file{local-alloc.c}). This pass allocates
2815: hard registers to pseudo registers that are used only within one basic
2816: block. Because the basic block is linear, it can use fast and
2817: powerful techniques to do a very good job.
2818:
2819: The option @samp{-dl} causes a debugging dump of the RTL code after
2820: this pass. This dump file's name is made by appending @samp{.lreg} to
2821: the input file name.
2822:
2823: @item
2824: Global register allocation (@file{global-alloc.c}). This pass
2825: allocates hard registers for the remaining pseudo registers (those
2826: whose life spans are not contained in one basic block).
2827:
2828: @item
2829: Reloading. This pass renumbers pseudo registers with the hardware
2830: registers numbers they were allocated. Pseudo registers that did not
2831: get hard registers are replaced with stack slots. Then it finds
2832: instructions that are invalid because a value has failed to end up in
2833: a register, or has ended up in a register of the wrong kind. It fixes
2834: up these instructions by reloading the problematical values
2835: temporarily into registers. Additional instructions are generated to
2836: do the copying.
2837:
2838: Source files are @file{reload.c} and @file{reload1.c}, plus the header
2839: @file{reload.h} used for communication between them.
2840:
2841: The option @samp{-dg} causes a debugging dump of the RTL code after
2842: this pass. This dump file's name is made by appending @samp{.greg} to
2843: the input file name.
2844:
2845: @item
2846: Jump optimization is repeated, this time including cross-jumping
2847: and deletion of no-op move instructions. Machine-specific peephole
2848: optimizations are performed at the same time.
2849:
2850: The option @samp{-dJ} causes a debugging dump of the RTL code after
2851: this pass. This dump file's name is made by appending @samp{.jump2}
2852: to the input file name.
2853:
2854: @item
2855: Final. This pass outputs the assembler code for the function. It is
2856: also responsible for identifying spurious test and compare
2857: instructions. The function entry and exit sequences are generated
2858: directly as assembler code in this pass; they never exist as RTL.
2859:
2860: The source files are @file{final.c} plus @file{insn-output.c}; the
2861: latter is generated automatically from the machine description by the
2862: tool @file{genoutput}. The header file @file{conditions.h} is used
2863: for communication between these files.
2864:
2865: @item
2866: Debugging information output. This is run after final because it must
2867: output the stack slot offsets for pseudo registers that did not get
2868: hard registers. Source files are @file{dbxout.c} for DBX symbol table
2869: format and @file{symout.c} for GDB's own symbol table format.
2870: @end itemize
2871:
2872: Some additional files are used by all or many passes:
2873:
2874: @itemize @bullet
2875: @item
2876: Every pass uses @file{machmode.def}, which defines the machine modes.
2877:
2878: @item
2879: All the passes that work with RTL use the header files @file{rtl.h}
2880: and @file{rtl.def}, and subroutines in file @file{rtl.c}. The tools
2881: @code{gen*} also use these files to read and work with the machine
2882: description RTL.
2883:
2884: @item
2885: Several passes refer to the header file @file{insn-config.h} which
2886: contains a few parameters (C macro definitions) generated
2887: automatically from the machine description RTL by the tool
2888: @code{genconfig}.
2889:
2890: @item
2891: Several passes use the instruction recognizer, which consists of
2892: @file{recog.c} and @file{recog.h}, plus the files @file{insn-recog.c}
2893: and @file{insn-extract.c} that are generated automatically from the
2894: machine description by the tools @file{genrecog} and
2895: @file{genextract}.@refill
2896:
2897: @item
2898: Several passes use the header files @file{regs.h} which defines the
2899: information recorded about pseudo register usage, and @file{basic-block.h}
2900: which defines the information recorded about basic blocks.
2901:
2902: @item
2903: @file{hard-reg-set.h} defines the type @code{HARD_REG_SET}, a bit-vector
2904: with a bit for each hard register, and some macros to manipulate it.
2905: This type is just @code{int} if the machine has few enough hard registers;
2906: otherwise it is an array of @code{int} and some of the macros expand
2907: into loops.
2908: @end itemize
2909:
2910: @node RTL, Machine Desc, Passes, Top
2911: @chapter RTL Representation
2912:
2913: Most of the work of the compiler is done on an intermediate representation
2914: called register transfer language. In this language, the instructions to be
2915: output are described, pretty much one by one, in an algebraic form that
2916: describes what the instruction does.
2917:
2918: RTL is inspired by Lisp lists. It has both an internal form, made up of
2919: structures that point at other structures, and a textual form that is used
2920: in the machine description and in printed debugging dumps. The textual
2921: form uses nested parentheses to indicate the pointers in the internal form.
2922:
2923: @menu
2924: * RTL Objects:: Expressions vs vectors vs strings vs integers.
2925: * Accessors:: Macros to access expression operands or vector elts.
2926: * Flags:: Other flags in an RTL expression.
2927: * Machine Modes:: Describing the size and format of a datum.
2928: * Constants:: Expressions with constant values.
2929: * Regs and Memory:: Expressions representing register contents or memory.
2930: * Arithmetic:: Expressions representing arithmetic on other expressions.
2931: * Comparisons:: Expressions representing comparison of expressions.
2932: * Bit Fields:: Expressions representing bit-fields in memory or reg.
2933: * Conversions:: Extending, truncating, floating or fixing.
2934: * RTL Declarations:: Declaring volatility, constancy, etc.
2935: * Side Effects:: Expressions for storing in registers, etc.
2936: * Incdec:: Embedded side-effects for autoincrement addressing.
2937: * Assembler:: Representing @code{asm} with operands.
2938: * Insns:: Expression types for entire insns.
2939: * Calls:: RTL representation of function call insns.
2940: * Sharing:: Some expressions are unique; others *must* be copied.
2941: @end menu
2942:
2943: @node RTL Objects, Accessors, RTL, RTL
2944: @section RTL Object Types
2945:
2946: RTL uses four kinds of objects: expressions, integers, strings and vectors.
2947: Expressions are the most important ones. An RTL expression (``RTX'', for
2948: short) is a C structure, but it is usually referred to with a pointer; a
2949: type that is given the typedef name @code{rtx}.
2950:
2951: An integer is simply an @code{int}, and a string is a @code{char *}.
2952: Within RTL code, strings appear only inside @samp{symbol_ref} expressions,
2953: but they appear in other contexts in the RTL expressions that make up
2954: machine descriptions. Their written form uses decimal digits.
2955:
2956: A string is a sequence of characters. In core it is represented as a
2957: @code{char *} in usual C fashion, and it is written in C syntax as well.
2958: However, strings in RTL may never be null. If you write an empty string in
2959: a machine description, it is represented in core as a null pointer rather
2960: than as a pointer to a null character. In certain contexts, these null
2961: pointers instead of strings are valid.
2962:
2963: A vector contains an arbitrary, specified number of pointers to
2964: expressions. The number of elements in the vector is explicitly present in
2965: the vector. The written form of a vector consists of square brackets
2966: (@samp{[@dots{}]}) surrounding the elements, in sequence and with
2967: whitespace separating them. Vectors of length zero are not created; null
2968: pointers are used instead.
2969:
2970: Expressions are classified by @dfn{expression codes} (also called RTX
2971: codes). The expression code is a name defined in @file{rtl.def}, which is
2972: also (in upper case) a C enumeration constant. The possible expression
2973: codes and their meanings are machine-independent. The code of an RTX can
2974: be extracted with the macro @code{GET_CODE (@var{x})} and altered with
2975: @code{PUT_CODE (@var{x}, @var{newcode})}.
2976:
2977: The expression code determines how many operands the expression contains,
2978: and what kinds of objects they are. In RTL, unlike Lisp, you cannot tell
2979: by looking at an operand what kind of object it is. Instead, you must know
2980: from its context---from the expression code of the containing expression.
2981: For example, in an expression of code @samp{subreg}, the first operand is
2982: to be regarded as an expression and the second operand as an integer. In
2983: an expression of code @samp{plus}, there are two operands, both of which
2984: are to be regarded as expressions. In a @samp{symbol_ref} expression,
2985: there is one operand, which is to be regarded as a string.
2986:
2987: Expressions are written as parentheses containing the name of the
2988: expression type, its flags and machine mode if any, and then the operands
2989: of the expression (separated by spaces).
2990:
2991: Expression code names in the @samp{md} file are written in lower case,
2992: but when they appear in C code they are written in upper case. In this
2993: manual, they are shown as follows: @samp{const_int}.
2994:
2995: In a few contexts a null pointer is valid where an expression is normally
2996: wanted. The written form of this is @samp{(nil)}.
2997:
2998: @node Accessors, Flags, RTL Objects, RTL
2999: @section Access to Operands
3000:
3001: For each expression type @file{rtl.def} specifies the number of contained
3002: objects and their kinds, with four possibilities: @samp{e} for expression
3003: (actually a pointer to an expression), @samp{i} for integer, @samp{s} for
3004: string, and @samp{E} for vector of expressions. The sequence of letters
3005: for an expression code is called its @dfn{format}. Thus, the format of
3006: @samp{subreg} is @samp{ei}.@refill
3007:
3008: Two other format characters are used occasionally: @samp{u} and @samp{0}.
3009: @samp{u} is equivalent to @samp{e} except that it is printed differently in
3010: debugging dumps, and @samp{0} means a slot whose contents do not fit any
3011: normal category. @samp{0} slots are not printed at all in dumps, and are
3012: often used in special ways by small parts of the compiler.@refill
3013:
3014: There are macros to get the number of operands and the format of an
3015: expression code:
3016:
3017: @table @code
3018: @item GET_RTX_LENGTH (@var{code})
3019: Number of operands of an RTX of code @var{code}.
3020:
3021: @item GET_RTX_FORMAT (@var{code})
3022: The format of an RTX of code @var{code}, as a C string.
3023: @end table
3024:
3025: Operands of expressions are accessed using the macros @code{XEXP},
3026: @code{XINT} and @code{XSTR}. Each of these macros takes two arguments: an
3027: expression-pointer (RTX) and an operand number (counting from zero).
3028: Thus,@refill
3029:
3030: @example
3031: XEXP (@var{x}, 2)
3032: @end example
3033:
3034: @noindent
3035: accesses operand 2 of expression @var{x}, as an expression.
3036:
3037: @example
3038: XINT (@var{x}, 2)
3039: @end example
3040:
3041: @noindent
3042: accesses the same operand as an integer. @code{XSTR}, used in the same
3043: fashion, would access it as a string.
3044:
3045: Any operand can be accessed as an integer, as an expression or as a string.
3046: You must choose the correct method of access for the kind of value actually
3047: stored in the operand. You would do this based on the expression code of
3048: the containing expression. That is also how you would know how many
3049: operands there are.
3050:
3051: For example, if @var{x} is a @samp{subreg} expression, you know that it has
3052: two operands which can be correctly accessed as @code{XEXP (@var{x}, 0)}
3053: and @code{XINT (@var{x}, 1)}. If you did @code{XINT (@var{x}, 0)}, you
3054: would get the address of the expression operand but cast as an integer;
3055: that might occasionally be useful, but it would be cleaner to write
3056: @code{(int) XEXP (@var{x}, 0)}. @code{XEXP (@var{x}, 1)} would also
3057: compile without error, and would return the second, integer operand cast as
3058: an expression pointer, which would probably result in a crash when
3059: accessed. Nothing stops you from writing @code{XEXP (@var{x}, 28)} either,
3060: but this will access memory past the end of the expression with
3061: unpredictable results.@refill
3062:
3063: Access to operands which are vectors is more complicated. You can use the
3064: macro @code{XVEC} to get the vector-pointer itself, or the macros
3065: @code{XVECEXP} and @code{XVECLEN} to access the elements and length of a
3066: vector.
3067:
3068: @table @code
3069: @item XVEC (@var{exp}, @var{idx})
3070: Access the vector-pointer which is operand number @var{idx} in @var{exp}.
3071:
3072: @item XVECLEN (@var{exp}, @var{idx})
3073: Access the length (number of elements) in the vector which is
3074: in operand number @var{idx} in @var{exp}. This value is an @code{int}.
3075:
3076: @item XVECEXP (@var{exp}, @var{idx}, @var{eltnum})
3077: Access element number @var{eltnum} in the vector which is
3078: in operand number @var{idx} in @var{exp}. This value is an RTX.
3079:
3080: It is up to you to make sure that @var{eltnum} is not negative
3081: and is less than @code{XVECLEN (@var{exp}, @var{idx})}.
3082: @end table
3083:
3084: All the macros defined in this section expand into lvalues and therefore
3085: can be used to assign the operands, lengths and vector elements as well as
3086: to access them.
3087:
3088: @node Flags, Machine Modes, Accessors, RTL
3089: @section Flags in an RTL Expression
3090:
3091: RTL expressions contain several flags (one-bit bit-fields) that are used
3092: in certain types of expression. Most often they are accessed with the
3093: following macros:
3094:
3095: @table @code
3096: @item MEM_VOLATILE_P (@var{x})
3097: In @samp{mem} expressions, nonzero for volatile memory references.
3098: Stored in the @code{volatil} field and printed as @samp{/v}.
3099:
3100: @item MEM_IN_STRUCT_P (@var{x})
3101: In @samp{mem} expressions, nonzero for reference to an entire
3102: structure, union or array, or to a component of one. Zero for
3103: references to a scalar variable or through a pointer to a scalar.
3104: Stored in the @code{in_struct} field and printed as @samp{/s}.
3105:
3106: @item REG_USER_VAR_P (@var{x})
3107: In a @samp{reg}, nonzero if it corresponds to a variable present in
3108: the user's source code. Zero for temporaries generated internally by
3109: the compiler. Stored in the @code{volatil} field and printed as
3110: @samp{/v}.
3111:
3112: @item REG_FUNCTION_VALUE_P (@var{x})
3113: Nonzero in a @samp{reg} if it is the place in which this function's
3114: value is going to be returned. (This happens only in a hard
3115: register.) Stored in the @code{integrated} field and printed as
3116: @samp{/i}.
3117:
3118: The same hard register may be used also for collecting the values of
3119: functions called by this one, but @code{REG_FUNCTION_VALUE_P} is zero
3120: in this kind of use.
3121:
3122: @item RTX_UNCHANGING_P (@var{x})
3123: Nonzero in a @samp{reg} or @samp{mem} if the value is not changed
3124: explicitly by the current function. (If it is a memory reference then
3125: it may be changed by other functions or by aliasing.) Stored in the
3126: @code{unchanging} field and printed as @samp{/u}.
3127:
3128: @item RTX_INTEGRATED_P (@var{insn})
3129: Nonzero in an insn if it resulted from an in-line function call.
3130: Stored in the @code{integrated} field and printed as @samp{/i}. This
3131: may be deleted; nothing currently depends on it.
3132:
3133: @item INSN_DELETED_P (@var{insn})
3134: In an insn, nonzero if the insn has been deleted. Stored in the
3135: @code{volatil} field and printed as @samp{/v}.
3136:
3137: @item CONSTANT_POOL_ADDRESS_P (@var{x})
3138: Nonzero in a @samp{symbol_ref} if it refers to part of the current
3139: function's ``constants pool''. These are addresses close to the
3140: beginning of the function, and GNU CC assumes they can be addressed
3141: directly (perhaps with the help of base registers). Stored in the
3142: @code{unchanging} field and printed as @samp{/u}.
3143: @end table
3144:
3145: These are the fields which the above macros refer to:
3146:
3147: @table @code
3148: @item used
3149: This flag is used only momentarily, at the end of RTL generation for a
3150: function, to count the number of times an expression appears in insns.
3151: Expressions that appear more than once are copied, according to the
3152: rules for shared structure (@pxref{Sharing}).
3153:
3154: @item volatil
3155: This flag is used in @samp{mem} and @samp{reg} expressions and in insns.
3156: In RTL dump files, it is printed as @samp{/v}.
3157:
3158: In a @samp{mem} expression, it is 1 if the memory reference is volatile.
3159: Volatile memory references may not be deleted, reordered or combined.
3160:
3161: In a @samp{reg} expression, it is 1 if the value is a user-level variable.
3162: 0 indicates an internal compiler temporary.
3163:
3164: In an insn, 1 means the insn has been deleted.
3165:
3166: @item in_struct
3167: This flag is used in @samp{mem} expressions. It is 1 if the memory
3168: datum referred to is all or part of a structure or array; 0 if it is (or
3169: might be) a scalar variable. A reference through a C pointer has 0
3170: because the pointer might point to a scalar variable.
3171:
3172: This information allows the compiler to determine something about possible
3173: cases of aliasing.
3174:
3175: In an RTL dump, this flag is represented as @samp{/s}.
3176:
3177: @item unchanging
3178: This flag is used in @samp{reg} and @samp{mem} expressions. 1 means
3179: that the value of the expression never changes (at least within the
3180: current function).
3181:
3182: In an RTL dump, this flag is represented as @samp{/u}.
3183:
3184: @item integrated
3185: In some kinds of expressions, including insns, this flag means the
3186: rtl was produced by procedure integration.
3187:
3188: In a @samp{reg} expression, this flag indicates the register
3189: containing the value to be returned by the current function. On
3190: machines that pass parameters in registers, the same register number
3191: may be used for parameters as well, but this flag is not set on such
3192: uses.
3193: @end table
3194:
3195: @node Machine Modes, Constants, Flags, RTL
3196: @section Machine Modes
3197:
3198: A machine mode describes a size of data object and the representation used
3199: for it. In the C code, machine modes are represented by an enumeration
3200: type, @code{enum machine_mode}, defined in @file{machmode.def}. Each RTL
3201: expression has room for a machine mode and so do certain kinds of tree
3202: expressions (declarations and types, to be precise).
3203:
3204: In debugging dumps and machine descriptions, the machine mode of an RTL
3205: expression is written after the expression code with a colon to separate
3206: them. The letters @samp{mode} which appear at the end of each machine mode
3207: name are omitted. For example, @code{(reg:SI 38)} is a @samp{reg}
3208: expression with machine mode @code{SImode}. If the mode is
3209: @code{VOIDmode}, it is not written at all.
3210:
3211: Here is a table of machine modes.
3212:
3213: @table @code
3214: @item QImode
3215: ``Quarter-Integer'' mode represents a single byte treated as an integer.
3216:
3217: @item HImode
3218: ``Half-Integer'' mode represents a two-byte integer.
3219:
3220: @item SImode
3221: ``Single Integer'' mode represents a four-byte integer.
3222:
3223: @item DImode
3224: ``Double Integer'' mode represents an eight-byte integer.
3225:
3226: @item TImode
3227: ``Tetra Integer'' (?) mode represents a sixteen-byte integer.
3228:
3229: @item SFmode
3230: ``Single Floating'' mode represents a single-precision (four byte) floating
3231: point number.
3232:
3233: @item DFmode
3234: ``Double Floating'' mode represents a double-precision (eight byte) floating
3235: point number.
3236:
3237: @item TFmode
3238: ``Tetra Floating'' mode represents a quadruple-precision (sixteen byte)
3239: floating point number.
3240:
3241: @item BLKmode
3242: ``Block'' mode represents values that are aggregates to which none of
3243: the other modes apply. In RTL, only memory references can have this mode,
3244: and only if they appear in string-move or vector instructions. On machines
3245: which have no such instructions, @code{BLKmode} will not appear in RTL.
3246:
3247: @item VOIDmode
3248: Void mode means the absence of a mode or an unspecified mode.
3249: For example, RTL expressions of code @samp{const_int} have mode
3250: @code{VOIDmode} because they can be taken to have whatever mode the context
3251: requires. In debugging dumps of RTL, @code{VOIDmode} is expressed by
3252: the absence of any mode.
3253:
3254: @item EPmode
3255: ``Entry Pointer'' mode is intended to be used for function variables in
3256: Pascal and other block structured languages. Such values contain
3257: both a function address and a static chain pointer for access to
3258: automatic variables of outer levels. This mode is only partially
3259: implemented since C does not use it.
3260:
3261: @item CSImode@r{, @dots{}}
3262: ``Complex Single Integer'' mode stands for a complex number represented
3263: as a pair of @code{SImode} integers. Any of the integer and floating modes
3264: may have @samp{C} prefixed to its name to obtain a complex number mode.
3265: For example, there are @code{CQImode}, @code{CSFmode}, and @code{CDFmode}.
3266: Since C does not support complex numbers, these machine modes are only
3267: partially implemented.
3268:
3269: @item BImode
3270: This is the machine mode of a bit-field in a structure. It is used
3271: only in the syntax tree, never in RTL, and in the syntax tree it appears
3272: only in declaration nodes. In C, it appears only in @code{FIELD_DECL}
3273: nodes for structure fields defined with a bit size.
3274: @end table
3275:
3276: The machine description defines @code{Pmode} as a C macro which expands
3277: into the machine mode used for addresses. Normally this is @code{SImode}.
3278:
3279: The only modes which a machine description @i{must} support are
3280: @code{QImode}, @code{SImode}, @code{SFmode} and @code{DFmode}. The
3281: compiler will attempt to use @code{DImode} for two-word structures and
3282: unions, but it would not be hard to program it to avoid this. Likewise,
3283: you can arrange for the C type @code{short int} to avoid using
3284: @code{HImode}. In the long term it would be desirable to make the set of
3285: available machine modes machine-dependent and eliminate all assumptions
3286: about specific machine modes or their uses from the machine-independent
3287: code of the compiler.
3288:
3289: Here are some C macros that relate to machine modes:
3290:
3291: @table @code
3292: @item GET_MODE (@var{x})
3293: Returns the machine mode of the RTX @var{x}.
3294:
3295: @item PUT_MODE (@var{x}, @var{newmode})
3296: Alters the machine mode of the RTX @var{x} to be @var{newmode}.
3297:
3298: @item GET_MODE_SIZE (@var{m})
3299: Returns the size in bytes of a datum of mode @var{m}.
3300:
3301: @item GET_MODE_BITSIZE (@var{m})
3302: Returns the size in bits of a datum of mode @var{m}.
3303:
3304: @item GET_MODE_UNIT_SIZE (@var{m})
3305: Returns the size in bits of the subunits of a datum of mode @var{m}.
3306: This is the same as @code{GET_MODE_SIZE} except in the case of
3307: complex modes and @code{EPmode}. For them, the unit size is the
3308: size of the real or imaginary part, or the size of the function
3309: pointer or the context pointer.
3310: @end table
3311:
3312: @node Constants, Regs and Memory, Machine Modes, RTL
3313: @section Constant Expression Types
3314:
3315: The simplest RTL expressions are those that represent constant values.
3316:
3317: @table @code
3318: @item (const_int @var{i})
3319: This type of expression represents the integer value @var{i}. @var{i}
3320: is customarily accessed with the macro @code{INTVAL} as in
3321: @code{INTVAL (@var{exp})}, which is equivalent to @code{XINT (@var{exp}, 0)}.
3322:
3323: There is only one expression object for the integer value zero;
3324: it is the value of the variable @code{const0_rtx}. Likewise, the
3325: only expression for integer value one is found in @code{const1_rtx}.
3326: Any attempt to create an expression of code @samp{const_int} and
3327: value zero or one will return @code{const0_rtx} or @code{const1_rtx}
3328: as appropriate.
3329:
3330: @item (const_double:@var{m} @var{i0} @var{i1})
3331: Represents a 64-bit constant or mode @var{m}. All floating point
3332: constants are represented in this way, and so are 64-bit @code{DImode}
3333: integer constants.
3334:
3335: The two integers @var{i0} and @var{i1} together contain the bits of
3336: the value. If the constant is floating point (either single or double
3337: precision), then they represent a @code{double}. To convert them to a
3338: @code{double}, do
3339:
3340: @example
3341: union @{ double d; int i[2];@} u;
3342: u.i[0] = XINT (x, 0);
3343: u.i[1] = XINT (x, 1);
3344: @end example
3345:
3346: @noindent
3347: and then refer to @code{u.d}.
3348:
3349: The global variables @code{dconst0_rtx} and @code{fconst0_rtx} hold
3350: @samp{const_double} expressions with value 0, in modes @code{DFmode} and
3351: @code{SFmode}, respectively.
3352:
3353: @item (symbol_ref @var{symbol})
3354: Represents the value of an assembler label for data. @var{symbol} is
3355: a string that describes the name of the assembler label. If it starts
3356: with a @samp{*}, the label is the rest of @var{symbol} not including
3357: the @samp{*}. Otherwise, the label is @var{symbol}, prefixed with
3358: @samp{_}.
3359:
3360: @item (label_ref @var{label})
3361: Represents the value of an assembler label for code. It contains one
3362: operand, an expression, which must be a @samp{code_label} that appears
3363: in the instruction sequence to identify the place where the label
3364: should go.
3365:
3366: The reason for using a distinct expression type for code label
3367: references is so that jump optimization can distinguish them.
3368:
3369: @item (const @var{exp})
3370: Represents a constant that is the result of an assembly-time
3371: arithmetic computation. The operand, @var{exp}, is an expression that
3372: contains only constants (@samp{const_int}, @samp{symbol_ref} and
3373: @samp{label_ref} expressions) combined with @samp{plus} and
3374: @samp{minus}. However, not all combinations are valid, since the
3375: assembler cannot do arbitrary arithmetic on relocatable symbols.
3376: @end table
3377:
3378: @node Regs and Memory, Arithmetic, Constants, RTL
3379: @section Registers and Memory
3380:
3381: Here are the RTL expression types for describing access to machine
3382: registers and to main memory.
3383:
3384: @table @code
3385: @item (reg:@var{m} @var{n})
3386: For small values of the integer @var{n} (less than
3387: @code{FIRST_PSEUDO_REGISTER}), this stands for a reference to machine
3388: register number @var{n}: a @dfn{hard register}. For larger values of
3389: @var{n}, it stands for a temporary value or @dfn{pseudo register}.
3390: The compiler's strategy is to generate code assuming an unlimited
3391: number of such pseudo registers, and later convert them into hard
3392: registers or into memory references.
3393:
3394: The symbol @code{FIRST_PSEUDO_REGISTER} is defined by the machine
3395: description, since the number of hard registers on the machine is an
3396: invariant characteristic of the machine. Note, however, that not
3397: all of the machine registers must be general registers. All the
3398: machine registers that can be used for storage of data are given
3399: hard register numbers, even those that can be used only in certain
3400: instructions or can hold only certain types of data.
3401:
3402: Each pseudo register number used in a function's RTL code is
3403: represented by a unique @samp{reg} expression.
3404:
3405: @var{m} is the machine mode of the reference. It is necessary because
3406: machines can generally refer to each register in more than one mode.
3407: For example, a register may contain a full word but there may be
3408: instructions to refer to it as a half word or as a single byte, as
3409: well as instructions to refer to it as a floating point number of
3410: various precisions.
3411:
3412: Even for a register that the machine can access in only one mode,
3413: the mode must always be specified.
3414:
3415: A hard register may be accessed in various modes throughout one
3416: function, but each pseudo register is given a natural mode
3417: and is accessed only in that mode. When it is necessary to describe
3418: an access to a pseudo register using a nonnatural mode, a @samp{subreg}
3419: expression is used.
3420:
3421: A @samp{reg} expression with a machine mode that specifies more than
3422: one word of data may actually stand for several consecutive registers.
3423: If in addition the register number specifies a hardware register, then
3424: it actually represents several consecutive hardware registers starting
3425: with the specified one.
3426:
3427: Such multi-word hardware register @samp{reg} expressions may not be live
3428: across the boundary of a basic block. The lifetime analysis pass does not
3429: know how to record properly that several consecutive registers are
3430: actually live there, and therefore register allocation would be confused.
3431: The CSE pass must go out of its way to make sure the situation does
3432: not arise.
3433:
3434: @item (subreg:@var{m} @var{reg} @var{wordnum})
3435: @samp{subreg} expressions are used to refer to a register in a machine
3436: mode other than its natural one, or to refer to one register of
3437: a multi-word @samp{reg} that actually refers to several registers.
3438:
3439: Each pseudo-register has a natural mode. If it is necessary to
3440: operate on it in a different mode---for example, to perform a fullword
3441: move instruction on a pseudo-register that contains a single byte---
3442: the pseudo-register must be enclosed in a @samp{subreg}. In such
3443: a case, @var{wordnum} is zero.
3444:
3445: The other use of @samp{subreg} is to extract the individual registers
3446: of a multi-register value. Machine modes such as @code{DImode} and
3447: @code{EPmode} indicate values longer than a word, values which usually
3448: require two consecutive registers. To access one of the registers,
3449: use a @samp{subreg} with mode @code{SImode} and a @var{wordnum} that
3450: says which register.
3451:
3452: The compilation parameter @code{WORDS_BIG_ENDIAN}, if defined, says
3453: that word number zero is the most significant part; otherwise, it is
3454: the least significant part.
3455:
3456: Between the combiner pass and the reload pass, it is possible to have
3457: a @samp{subreg} which contains a @samp{mem} instead of a @samp{reg} as
3458: its first operand. The reload pass eliminates these cases by
3459: reloading the @samp{mem} into a suitable register.
3460:
3461: Note that it is not valid to access a @code{DFmode} value in @code{SFmode}
3462: using a @samp{subreg}. On some machines the most significant part of a
3463: @code{DFmode} value does not have the same format as a single-precision
3464: floating value.
3465:
3466: @item (cc0)
3467: This refers to the machine's condition code register. It has no
3468: operands and may not have a machine mode. It may be validly used in
3469: only two contexts: as the destination of an assignment (in test and
3470: compare instructions) and in comparison operators comparing against
3471: zero (@samp{const_int} with value zero; that is to say,
3472: @code{const0_rtx}).
3473:
3474: There is only one expression object of code @samp{cc0}; it is the
3475: value of the variable @code{cc0_rtx}. Any attempt to create an
3476: expression of code @samp{cc0} will return @code{cc0_rtx}.
3477:
3478: One special thing about the condition code register is that
3479: instructions can set it implicitly. On many machines, nearly all
3480: instructions set the condition code based on the value that they
3481: compute or store. It is not necessary to record these actions
3482: explicitly in the RTL because the machine description includes a
3483: prescription for recognizing the instructions that do so (by means of
3484: the macro @code{NOTICE_UPDATE_CC}). Only instructions whose sole
3485: purpose is to set the condition code, and instructions that use the
3486: condition code, need mention @code{(cc0)}.
3487:
3488: @item (pc)
3489: This represents the machine's program counter. It has no operands and
3490: may not have a machine mode. @code{(pc)} may be validly used only in
3491: certain specific contexts in jump instructions.
3492:
3493: There is only one expression object of code @samp{pc}; it is the value
3494: of the variable @code{pc_rtx}. Any attempt to create an expression of
3495: code @samp{pc} will return @code{pc_rtx}.
3496:
3497: All instructions that do not jump alter the program counter implicitly
3498: by incrementing it, but there is no need to mention this in the RTL.
3499:
3500: @item (mem:@var{m} @var{addr})
3501: This RTX represents a reference to main memory at an address
3502: represented by the expression @var{addr}. @var{m} specifies how large
3503: a unit of memory is accessed.
3504: @end table
3505:
3506: @node Arithmetic, Comparisons, Regs and Memory, RTL
3507: @section RTL Expressions for Arithmetic
3508:
3509: @table @code
3510: @item (plus:@var{m} @var{x} @var{y})
3511: Represents the sum of the values represented by @var{x} and @var{y}
3512: carried out in machine mode @var{m}. This is valid only if
3513: @var{x} and @var{y} both are valid for mode @var{m}.
3514:
3515: @item (minus:@var{m} @var{x} @var{y})
3516: Like @samp{plus} but represents subtraction.
3517:
3518: @item (minus @var{x} @var{y})
3519: Represents the result of subtracting @var{y} from @var{x}
3520: for purposes of comparison. The absence of a machine mode
3521: in the @samp{minus} expression indicates that the result is
3522: computed without overflow, as if with infinite precision.
3523:
3524: Of course, machines can't really subtract with infinite precision.
3525: However, they can pretend to do so when only the sign of the
3526: result will be used, which is the case when the result is stored
3527: in @code{(cc0)}. And that is the only way this kind of expression
3528: may validly be used: as a value to be stored in the condition codes.
3529:
3530: @item (neg:@var{m} @var{x})
3531: Represents the negation (subtraction from zero) of the value
3532: represented by @var{x}, carried out in mode @var{m}. @var{x} must be
3533: valid for mode @var{m}.
3534:
3535: @item (mult:@var{m} @var{x} @var{y})
3536: Represents the signed product of the values represented by @var{x} and
3537: @var{y} carried out in machine mode @var{m}. If
3538: @var{x} and @var{y} are both valid for mode @var{m}, this is ordinary
3539: size-preserving multiplication. Alternatively, both @var{x} and @var{y}
3540: may be valid for a different, narrower mode. This represents the
3541: kind of multiplication that generates a product wider than the operands.
3542: Widening multiplication and same-size multiplication are completely
3543: distinct and supported by different machine instructions; machines may
3544: support one but not the other.@refill
3545:
3546: @samp{mult} may be used for floating point division as well.
3547: Then @var{m} is a floating point machine mode.
3548:
3549: @item (umult:@var{m} @var{x} @var{y})
3550: Like @samp{mult} but represents unsigned multiplication. It may be
3551: used in both same-size and widening forms, like @samp{mult}.
3552: @samp{umult} is used only for fixed-point multiplication.
3553:
3554: @item (div:@var{m} @var{x} @var{y})
3555: Represents the quotient in signed division of @var{x} by @var{y},
3556: carried out in machine mode @var{m}. If @var{m} is a floating-point
3557: mode, it represents the exact quotient; otherwise, the integerized
3558: quotient. If @var{x} and @var{y} are both valid for mode @var{m},
3559: this is ordinary size-preserving division. Some machines have
3560: division instructions in which the operands and quotient widths are
3561: not all the same; such instructions are represented by @samp{div}
3562: expressions in which the machine modes are not all the same.
3563:
3564: @item (udiv:@var{m} @var{x} @var{y})
3565: Like @samp{div} but represents unsigned division.
3566:
3567: @item (mod:@var{m} @var{x} @var{y})
3568: @itemx (umod:@var{m} @var{x} @var{y})
3569: Like @samp{div} and @samp{udiv} but represent the remainder instead of
3570: the quotient.
3571:
3572: @item (not:@var{m} @var{x})
3573: Represents the bitwise complement of the value represented by @var{x},
3574: carried out in mode @var{m}, which must be a fixed-point machine mode.
3575: @var{x} must be valid for mode @var{m}, which must be a fixed-point mode.
3576:
3577: @item (and:@var{m} @var{x} @var{y})
3578: Represents the bitwise logical-and of the values represented by
3579: @var{x} and @var{y}, carried out in machine mode @var{m}. This is
3580: valid only if @var{x} and @var{y} both are valid for mode @var{m},
3581: which must be a fixed-point mode.
3582:
3583: @item (ior:@var{m} @var{x} @var{y})
3584: Represents the bitwise inclusive-or of the values represented by
3585: @var{x} and @var{y}, carried out in machine mode @var{m}. This is
3586: valid only if @var{x} and @var{y} both are valid for mode @var{m},
3587: which must be a fixed-point mode.
3588:
3589: @item (xor:@var{m} @var{x} @var{y})
3590: Represents the bitwise exclusive-or of the values represented by
3591: @var{x} and @var{y}, carried out in machine mode @var{m}. This is
3592: valid only if @var{x} and @var{y} both are valid for mode @var{m},
3593: which must be a fixed-point mode.
3594:
3595: @item (lshift:@var{m} @var{x} @var{c})
3596: Represents the result of logically shifting @var{x} left by @var{c}
3597: places. @var{x} must be valid for the mode @var{m}, a fixed-point
3598: machine mode. @var{c} must be valid for a fixed-point mode;
3599: which mode is determined by the mode called for in the machine
3600: description entry for the left-shift instruction. For example,
3601: on the Vax, the mode of @var{c} is @code{QImode} regardless of @var{m}.
3602:
3603: On some machines, negative values of @var{c} may be meaningful; this
3604: is why logical left shift and arithmetic left shift are distinguished.
3605: For example, Vaxes have no right-shift instructions, and right shifts
3606: are represented as left-shift instructions whose counts happen
3607: to be negative constants or else computed (in a previous instruction)
3608: by negation.
3609:
3610: @item (ashift:@var{m} @var{x} @var{c})
3611: Like @samp{lshift} but for arithmetic left shift.
3612:
3613: @item (lshiftrt:@var{m} @var{x} @var{c})
3614: @itemx (ashiftrt:@var{m} @var{x} @var{c})
3615: Like @samp{lshift} and @samp{ashift} but for right shift.
3616:
3617: @item (rotate:@var{m} @var{x} @var{c})
3618: @itemx (rotatert:@var{m} @var{x} @var{c})
3619: Similar but represent left and right rotate.
3620:
3621: @item (abs:@var{m} @var{x})
3622: Represents the absolute value of @var{x}, computed in mode @var{m}.
3623: @var{x} must be valid for @var{m}.
3624:
3625: @item (sqrt:@var{m} @var{x})
3626: Represents the square root of @var{x}, computed in mode @var{m}.
3627: @var{x} must be valid for @var{m}. Most often @var{m} will be
3628: a floating point mode.
3629:
3630: @item (ffs:@var{m} @var{x})
3631: Represents the one plus the index of the least significant 1-bit in
3632: @var{x}, represented as an integer of mode @var{m}. (The value is
3633: zero if @var{x} is zero.) The mode of @var{x} need not be @var{m};
3634: depending on the target machine, various mode combinations may be
3635: valid.
3636: @end table
3637:
3638: @node Comparisons, Bit Fields, Arithmetic, RTL
3639: @section Comparison Operations
3640:
3641: Comparison operators test a relation on two operands and are considered to
3642: represent the value 1 if the relation holds, or zero if it does not. The
3643: mode of the comparison is determined by the operands; they must both be
3644: valid for a common machine mode. A comparison with both operands constant
3645: would be invalid as the machine mode could not be deduced from it, but such
3646: a comparison should never exist in RTL due to constant folding.
3647:
3648: Inequality comparisons come in two flavors, signed and unsigned. Thus,
3649: there are distinct expression codes @samp{gt} and @samp{gtu} for signed and
3650: unsigned greater-than. These can produce different results for the same
3651: pair of integer values: for example, 1 is signed greater-than -1 but not
3652: unsigned greater-than, because -1 when regarded as unsigned is actually
3653: @code{0xffffffff} which is greater than 1.
3654:
3655: The signed comparisons are also used for floating point values. Floating
3656: point comparisons are distinguished by the machine modes of the operands.
3657:
3658: The comparison operators may be used to compare the condition codes
3659: @code{(cc0)} against zero, as in @code{(eq (cc0) (const_int 0))}. Such a
3660: construct actually refers to the result of the preceding instruction in
3661: which the condition codes were set. The above example stands for 1 if the
3662: condition codes were set to say ``zero'' or ``equal'', 0 otherwise.
3663: Although the same comparison operators are used for this as may be used in
3664: other contexts on actual data, no confusion can result since the machine
3665: description would never allow both kinds of uses in the same context.
3666:
3667: @table @code
3668: @item (eq @var{x} @var{y})
3669: 1 if the values represented by @var{x} and @var{y} are equal,
3670: otherwise 0.
3671:
3672: @item (ne @var{x} @var{y})
3673: 1 if the values represented by @var{x} and @var{y} are not equal,
3674: otherwise 0.
3675:
3676: @item (gt @var{x} @var{y})
3677: 1 if the @var{x} is greater than @var{y}. If they are fixed-point,
3678: the comparison is done in a signed sense.
3679:
3680: @item (gtu @var{x} @var{y})
3681: Like @samp{gt} but does unsigned comparison, on fixed-point numbers only.
3682:
3683: @item (lt @var{x} @var{y})
3684: @item (ltu @var{x} @var{y})
3685: Like @samp{gt} and @samp{gtu} but test for ``less than''.
3686:
3687: @item (ge @var{x} @var{y})
3688: @item (geu @var{x} @var{y})
3689: Like @samp{gt} and @samp{gtu} but test for ``greater than or equal''.
3690:
3691: @item (le @var{x} @var{y})
3692: @item (leu @var{x} @var{y})
3693: Like @samp{gt} and @samp{gtu} but test for ``less than or equal''.
3694:
3695: @item (if_then_else @var{cond} @var{then} @var{else})
3696: This is not a comparison operation but is listed here because it is
3697: always used in conjunction with a comparison operation. To be
3698: precise, @var{cond} is a comparison expression. This expression
3699: represents a choice, according to @var{cond}, between the value
3700: represented by @var{then} and the one represented by @var{else}.
3701:
3702: On most machines, @samp{if_then_else} expressions are valid only
3703: to express conditional jumps.
3704: @end table
3705:
3706: @node Bit Fields, Conversions, Comparisons, RTL
3707: @section Bit-fields
3708:
3709: Special expression codes exist to represent bit-field instructions.
3710: These types of expressions are lvalues in RTL; they may appear
3711: on the left side of a assignment, indicating insertion of a value
3712: into the specified bit field.
3713:
3714: @table @code
3715: @item (sign_extract:SI @var{loc} @var{size} @var{pos})
3716: This represents a reference to a sign-extended bit-field contained or
3717: starting in @var{loc} (a memory or register reference). The bit field
3718: is @var{size} bits wide and starts at bit @var{pos}. The compilation
3719: option @code{BITS_BIG_ENDIAN} says which end of the memory unit
3720: @var{pos} counts from.
3721:
3722: Which machine modes are valid for @var{loc} depends on the machine,
3723: but typically @var{loc} should be a single byte when in memory
3724: or a full word in a register.
3725:
3726: @item (zero_extract:SI @var{loc} @var{size} @var{pos})
3727: Like @samp{sign_extract} but refers to an unsigned or zero-extended
3728: bit field. The same sequence of bits are extracted, but they
3729: are filled to an entire word with zeros instead of by sign-extension.
3730: @end table
3731:
3732: @node Conversions, RTL Declarations, Bit Fields, RTL
3733: @section Conversions
3734:
3735: All conversions between machine modes must be represented by
3736: explicit conversion operations. For example, an expression
3737: which is the sum of a byte and a full word cannot be written as
3738: @code{(plus:SI (reg:QI 34) (reg:SI 80))} because the @samp{plus}
3739: operation requires two operands of the same machine mode.
3740: Therefore, the byte-sized operand is enclosed in a conversion
3741: operation, as in
3742:
3743: @example
3744: (plus:SI (sign_extend:SI (reg:QI 34)) (reg:SI 80))
3745: @end example
3746:
3747: The conversion operation is not a mere placeholder, because there
3748: may be more than one way of converting from a given starting mode
3749: to the desired final mode. The conversion operation code says how
3750: to do it.
3751:
3752: @table @code
3753: @item (sign_extend:@var{m} @var{x})
3754: Represents the result of sign-extending the value @var{x}
3755: to machine mode @var{m}. @var{m} must be a fixed-point mode
3756: and @var{x} a fixed-point value of a mode narrower than @var{m}.
3757:
3758: @item (zero_extend:@var{m} @var{x})
3759: Represents the result of zero-extending the value @var{x}
3760: to machine mode @var{m}. @var{m} must be a fixed-point mode
3761: and @var{x} a fixed-point value of a mode narrower than @var{m}.
3762:
3763: @item (float_extend:@var{m} @var{x})
3764: Represents the result of extending the value @var{x}
3765: to machine mode @var{m}. @var{m} must be a floating point mode
3766: and @var{x} a floating point value of a mode narrower than @var{m}.
3767:
3768: @item (truncate:@var{m} @var{x})
3769: Represents the result of truncating the value @var{x}
3770: to machine mode @var{m}. @var{m} must be a fixed-point mode
3771: and @var{x} a fixed-point value of a mode wider than @var{m}.
3772:
3773: @item (float_truncate:@var{m} @var{x})
3774: Represents the result of truncating the value @var{x}
3775: to machine mode @var{m}. @var{m} must be a floating point mode
3776: and @var{x} a floating point value of a mode wider than @var{m}.
3777:
3778: @item (float:@var{m} @var{x})
3779: Represents the result of converting fixed point value @var{x},
3780: regarded as signed, to floating point mode @var{m}.
3781:
3782: @item (unsigned_float:@var{m} @var{x})
3783: Represents the result of converting fixed point value @var{x},
3784: regarded as unsigned, to floating point mode @var{m}.
3785:
3786: @item (fix:@var{m} @var{x})
3787: When @var{m} is a fixed point mode, represents the result of
3788: converting floating point value @var{x} to mode @var{m}, regarded as
3789: signed. How rounding is done is not specified, so this operation may
3790: be used validly in compiling C code only for integer-valued operands.
3791:
3792: @item (unsigned_fix:@var{m} @var{x})
3793: Represents the result of converting floating point value @var{x} to
3794: fixed point mode @var{m}, regarded as unsigned. How rounding is done
3795: is not specified.
3796:
3797: @item (fix:@var{m} @var{x})
3798: When @var{m} is a floating point mode, represents the result of
3799: converting floating point value @var{x} (valid for mode @var{m}) to an
3800: integer, still represented in floating point mode @var{m}, by rounding
3801: towards zero.
3802: @end table
3803:
3804: @node RTL Declarations, Side Effects, Conversions, RTL
3805: @section Declarations
3806:
3807: Declaration expression codes do not represent arithmetic operations
3808: but rather state assertions about their operands.
3809:
3810: @table @code
3811: @item (strict_low_part (subreg:@var{m} (reg:@var{n} @var{r}) 0))
3812: This expression code is used in only one context: operand 0 of a
3813: @samp{set} expression. In addition, the operand of this expression
3814: must be a @samp{subreg} expression.
3815:
3816: The presence of @samp{strict_low_part} says that the part of the
3817: register which is meaningful in mode @var{n}, but is not part of
3818: mode @var{m}, is not to be altered. Normally, an assignment to such
3819: a subreg is allowed to have undefined effects on the rest of the
3820: register when @var{m} is less than a word.
3821: @end table
3822:
3823: @node Side Effects, Incdec, RTL Declarations, RTL
3824: @section Side Effect Expressions
3825:
3826: The expression codes described so far represent values, not actions.
3827: But machine instructions never produce values; they are meaningful
3828: only for their side effects on the state of the machine. Special
3829: expression codes are used to represent side effects.
3830:
3831: The body of an instruction is always one of these side effect codes;
3832: the codes described above, which represent values, appear only as
3833: the operands of these.
3834:
3835: @table @code
3836: @item (set @var{lval} @var{x})
3837: Represents the action of storing the value of @var{x} into the place
3838: represented by @var{lval}. @var{lval} must be an expression
3839: representing a place that can be stored in: @samp{reg} (or
3840: @samp{subreg} or @samp{strict_low_part}), @samp{mem}, @samp{pc} or
3841: @samp{cc0}.@refill
3842:
3843: If @var{lval} is a @samp{reg}, @samp{subreg} or @samp{mem}, it has a
3844: machine mode; then @var{x} must be valid for that mode.@refill
3845:
3846: If @var{lval} is a @samp{reg} whose machine mode is less than the full
3847: width of the register, then it means that the part of the register
3848: specified by the machine mode is given the specified value and the
3849: rest of the register receives an undefined value. Likewise, if
3850: @var{lval} is a @samp{subreg} whose machine mode is narrower than
3851: @code{SImode}, the rest of the register can be changed in an undefined way.
3852:
3853: If @var{lval} is a @samp{strict_low_part} of a @samp{subreg}, then the
3854: part of the register specified by the machine mode of the
3855: @samp{subreg} is given the value @var{x} and the rest of the register
3856: is not changed.@refill
3857:
3858: If @var{lval} is @code{(cc0)}, it has no machine mode, and @var{x} may
3859: have any mode. This represents a ``test'' or ``compare'' instruction.@refill
3860:
3861: If @var{lval} is @code{(pc)}, we have a jump instruction, and the
3862: possibilities for @var{x} are very limited. It may be a
3863: @samp{label_ref} expression (unconditional jump). It may be an
3864: @samp{if_then_else} (conditional jump), in which case either the
3865: second or the third operand must be @code{(pc)} (for the case which
3866: does not jump) and the other of the two must be a @samp{label_ref}
3867: (for the case which does jump). @var{x} may also be a @samp{mem} or
3868: @code{(plus:SI (pc) @var{y})}, where @var{y} may be a @samp{reg} or a
3869: @samp{mem}; these unusual patterns are used to represent jumps through
3870: branch tables.@refill
3871:
3872: @item (return)
3873: Represents a return from the current function, on machines where this
3874: can be done with one instruction, such as Vaxes. On machines where a
3875: multi-instruction ``epilogue'' must be executed in order to return
3876: from the function, returning is done by jumping to a label which
3877: precedes the epilogue, and the @samp{return} expression code is never
3878: used.
3879:
3880: @item (call @var{function} @var{nargs})
3881: Represents a function call. @var{function} is a @samp{mem} expression
3882: whose address is the address of the function to be called.
3883: @var{nargs} is an expression which can be used for two purposes: on
3884: some machines it represents the number of bytes of stack argument; on
3885: others, it represents the number of argument registers.
3886:
3887: Each machine has a standard machine mode which @var{function} must
3888: have. The machine description defines macro @code{FUNCTION_MODE} to
3889: expand into the requisite mode name. The purpose of this mode is to
3890: specify what kind of addressing is allowed, on machines where the
3891: allowed kinds of addressing depend on the machine mode being
3892: addressed.
3893:
3894: @item (clobber @var{x})
3895: Represents the storing or possible storing of an unpredictable,
3896: undescribed value into @var{x}, which must be a @samp{reg} or
3897: @samp{mem} expression.
3898:
3899: One place this is used is in string instructions that store standard
3900: values into particular hard registers. It may not be worth the
3901: trouble to describe the values that are stored, but it is essential to
3902: inform the compiler that the registers will be altered, lest it
3903: attempt to keep data in them across the string instruction.
3904:
3905: @var{x} may also be null---a null C pointer, no expression at all.
3906: Such a @code{(clobber (null))} expression means that all memory
3907: locations must be presumed clobbered.
3908:
3909: Note that the machine description classifies certain hard registers as
3910: ``call-clobbered''. All function call instructions are assumed by
3911: default to clobber these registers, so there is no need to use
3912: @samp{clobber} expressions to indicate this fact. Also, each function
3913: call is assumed to have the potential to alter any memory location.
3914:
3915: @item (use @var{x})
3916: Represents the use of the value of @var{x}. It indicates that the
3917: value in @var{x} at this point in the program is needed, even though
3918: it may not be apparent why this is so. Therefore, the compiler will
3919: not attempt to delete instructions whose only effect is to store a
3920: value in @var{x}. @var{x} must be a @samp{reg} expression.
3921:
3922: @item (parallel [@var{x0} @var{x1} @dots{}])
3923: Represents several side effects performed in parallel. The square
3924: brackets stand for a vector; the operand of @samp{parallel} is a
3925: vector of expressions. @var{x0}, @var{x1} and so on are individual
3926: side effects---expressions of code @samp{set}, @samp{call},
3927: @samp{return}, @samp{clobber} or @samp{use}.@refill
3928:
3929: ``In parallel'' means that first all the values used in the individual
3930: side-effects are computed, and second all the actual side-effects are
3931: performed. For example,
3932:
3933: @example
3934: (parallel [(set (reg:SI 1) (mem:SI (reg:SI 1)))
3935: (set (mem:SI (reg:SI 1)) (reg:SI 1))])
3936: @end example
3937:
3938: @noindent
3939: says unambiguously that the values of hard register 1 and the memory
3940: location addressed by it are interchanged. In both places where
3941: @code{(reg:SI 1)} appears as a memory address it refers to the value
3942: in register 1 @emph{before} the execution of the instruction.
3943:
3944: Peephole optimization, which takes place in the last jump-optimization
3945: pass, can produce insns whose patterns consist of a @samp{parallel}
3946: whose elements are the operands needed to output the resulting
3947: assembler code--often @samp{reg}, @samp{mem} or constant expressions.
3948: This would not be well-formed RTL at any other stage in compilation,
3949: but it is ok then because no further optimization remains to be done.
3950: However, the definition of the macro @code{NOTICE_UPDATE_CC} may need
3951: to deal with such insns.
3952:
3953: @item (sequence [@var{insns} @dots{}])
3954: Represents a sequence of insns. Each of the @var{insns} that appears
3955: in the vector is suitable for appearing in the chain of insns, so it
3956: must be an @samp{insn}, @samp{jump_insn}, @samp{call_insn},
3957: @samp{code_label}, @samp{barrier} or @samp{note}.
3958:
3959: A @samp{sequence} RTX never appears in an actual insn. It represents
3960: the sequence of insns that result from a @samp{define_expand}
3961: @emph{before} those insns are passed to @code{emit_insn} to insert
3962: them in the chain of insns. When actually inserted, the individual
3963: sub-insns are separated out and the @samp{sequence} is forgotten.
3964: @end table
3965:
3966: Three expression codes appear in place of a side effect, as the body of an
3967: insn, though strictly speaking they do not describe side effects as such:
3968:
3969: @table @code
3970: @item (asm_input @var{s})
3971: Represents literal assembler code as described by the string @var{s}.
3972:
3973: @item (addr_vec:@var{m} [@var{lr0} @var{lr1} @dots{}])
3974: Represents a table of jump addresses. The vector elements @var{lr0},
3975: etc., are @samp{label_ref} expressions. The mode @var{m} specifies
3976: how much space is given to each address; normally @var{m} would be
3977: @code{Pmode}.
3978:
3979: @item (addr_diff_vec:@var{m} @var{base} [@var{lr0} @var{lr1} @dots{}])
3980: Represents a table of jump addresses expressed as offsets from
3981: @var{base}. The vector elements @var{lr0}, etc., are @samp{label_ref}
3982: expressions and so is @var{base}. The mode @var{m} specifies how much
3983: space is given to each address-difference.@refill
3984: @end table
3985:
3986: @node Incdec, Assembler, Side Effects, RTL
3987: @section Embedded Side-Effects on Addresses
3988:
3989: Four special side-effect expression codes appear as memory addresses.
3990:
3991: @table @code
3992: @item (pre_dec:@var{m} @var{x})
3993: Represents the side effect of decrementing @var{x} by a standard
3994: amount and represents also the value that @var{x} has after being
3995: decremented. @var{x} must be a @samp{reg} or @samp{mem}, but most
3996: machines allow only a @samp{reg}. @var{m} must be the machine mode
3997: for pointers on the machine in use. The amount @var{x} is decremented
3998: by is the length in bytes of the machine mode of the containing memory
3999: reference of which this expression serves as the address. Here is an
4000: example of its use:@refill
4001:
4002: @example
4003: (mem:DF (pre_dec:SI (reg:SI 39)))
4004: @end example
4005:
4006: @noindent
4007: This says to decrement pseudo register 39 by the length of a @code{DFmode}
4008: value and use the result to address a @code{DFmode} value.
4009:
4010: @item (pre_inc:@var{m} @var{x})
4011: Similar, but specifies incrementing @var{x} instead of decrementing it.
4012:
4013: @item (post_dec:@var{m} @var{x})
4014: Represents the same side effect as @samp{pre_decrement} but a different
4015: value. The value represented here is the value @var{x} has @i{before}
4016: being decremented.
4017:
4018: @item (post_inc:@var{m} @var{x})
4019: Similar, but specifies incrementing @var{x} instead of decrementing it.
4020: @end table
4021:
4022: These embedded side effect expressions must be used with care. Instruction
4023: patterns may not use them. Until the @samp{flow} pass of the compiler,
4024: they may occur only to represent pushes onto the stack. The @samp{flow}
4025: pass finds cases where registers are incremented or decremented in one
4026: instruction and used as an address shortly before or after; these cases are
4027: then transformed to use pre- or post-increment or -decrement.
4028:
4029: Explicit popping of the stack could be represented with these embedded
4030: side effect operators, but that would not be safe; the instruction
4031: combination pass could move the popping past pushes, thus changing
4032: the meaning of the code.
4033:
4034: An instruction that can be represented with an embedded side effect
4035: could also be represented using @samp{parallel} containing an additional
4036: @samp{set} to describe how the address register is altered. This is not
4037: done because machines that allow these operations at all typically
4038: allow them wherever a memory address is called for. Describing them as
4039: additional parallel stores would require doubling the number of entries
4040: in the machine description.
4041:
4042: @node Assembler, Insns, IncDec, RTL
4043: @section Assembler Instructions as Expressions
4044:
4045: The RTX code @samp{asm_operands} represents a value produced by a
4046: user-specified assembler instruction. It is used to represent
4047: an @code{asm} statement with arguments. An @code{asm} statement with
4048: a single output operand, like this:
4049:
4050: @example
4051: asm ("foo %1,%2,%0" : "a" (outputvar) : "g" (x + y), "di" (*z));
4052: @end example
4053:
4054: @noindent
4055: is represented using a single @samp{asm_operands} RTX which represents
4056: the value that is stored in @code{outputvar}:
4057:
4058: @example
4059: (set @var{rtx-for-outputvar}
4060: (asm_operands "foo %1,%2,%0" "a" 0
4061: [@var{rtx-for-addition-result} @var{rtx-for-*z}]
4062: [(asm_input:@var{m1} "g")
4063: (asm_input:@var{m2} "di")]))
4064: @end example
4065:
4066: @noindent
4067: Here the operands of the @samp{asm_operands} RTX are the assembler
4068: template string, the output-operand's constraint, the index-number of the
4069: output operand among the output operands specified, a vector of input
4070: operand RTX's, and a vector of input-operand modes and constraints. The
4071: mode @var{m1} is the mode of the sum @code{x+y}; @var{m2} is that of
4072: @code{*z}.
4073:
4074: When an @code{asm} statement has multiple output values, its insn has
4075: several such @samp{set} RTX's inside of a @samp{parallel}. Each @samp{set}
4076: contains a @samp{asm_operands}; all of these share the same assembler
4077: template and vectors, but each contains the constraint for the respective
4078: output operand. They are also distinguished by the output-operand index
4079: number, which is 0, 1, @dots{} for successive output operands.
4080:
4081: @node Insns, Calls, Assembler, RTL
4082: @section Insns
4083:
4084: The RTL representation of the code for a function is a doubly-linked
4085: chain of objects called @dfn{insns}. Insns are expressions with
4086: special codes that are used for no other purpose. Some insns are
4087: actual instructions; others represent dispatch tables for @code{switch}
4088: statements; others represent labels to jump to or various sorts of
4089: declarative information.
4090:
4091: In addition to its own specific data, each insn must have a unique id-number
4092: that distinguishes it from all other insns in the current function, and
4093: chain pointers to the preceding and following insns. These three fields
4094: occupy the same position in every insn, independent of the expression code
4095: of the insn. They could be accessed with @code{XEXP} and @code{XINT},
4096: but instead three special macros are always used:
4097:
4098: @table @code
4099: @item INSN_UID (@var{i})
4100: Accesses the unique id of insn @var{i}.
4101:
4102: @item PREV_INSN (@var{i})
4103: Accesses the chain pointer to the insn preceding @var{i}.
4104: If @var{i} is the first insn, this is a null pointer.
4105:
4106: @item NEXT_INSN (@var{i})
4107: Accesses the chain pointer to the insn following @var{i}.
4108: If @var{i} is the last insn, this is a null pointer.
4109: @end table
4110:
4111: The @code{NEXT_INSN} and @code{PREV_INSN} pointers must always
4112: correspond: if @var{i} is not the first insn,
4113:
4114: @example
4115: NEXT_INSN (PREV_INSN (@var{insn})) == @var{insn}
4116: @end example
4117:
4118: @noindent
4119: is always true.
4120:
4121: Every insn has one of the following six expression codes:
4122:
4123: @table @samp
4124: @item insn
4125: The expression code @samp{insn} is used for instructions that do not jump
4126: and do not do function calls. Insns with code @samp{insn} have four
4127: additional fields beyond the three mandatory ones listed above.
4128: These four are described in a table below.
4129:
4130: @item jump_insn
4131: The expression code @samp{jump_insn} is used for instructions that may jump
4132: (or, more generally, may contain @samp{label_ref} expressions).
4133: @samp{jump_insn} insns have the same extra fields as @samp{insn} insns,
4134: accessed in the same way.
4135:
4136: @item call_insn
4137: The expression code @samp{call_insn} is used for instructions that may do
4138: function calls. It is important to distinguish these instructions because
4139: they imply that certain registers and memory locations may be altered
4140: unpredictably.
4141:
4142: @samp{call_insn} insns have the same extra fields as @samp{insn} insns,
4143: accessed in the same way.
4144:
4145: @item code_label
4146: A @samp{code_label} insn represents a label that a jump insn can jump to.
4147: It contains one special field of data in addition to the three standard ones.
4148: It is used to hold the @dfn{label number}, a number that identifies this
4149: label uniquely among all the labels in the compilation (not just in the
4150: current function). Ultimately, the label is represented in the assembler
4151: output as an assembler label @samp{L@var{n}} where @var{n} is the label number.
4152:
4153: @item barrier
4154: Barriers are placed in the instruction stream after unconditional
4155: jump instructions to indicate that the jumps are unconditional.
4156: They contain no information beyond the three standard fields.
4157:
4158: @item note
4159: @samp{note} insns are used to represent additional debugging and
4160: declarative information. They contain two nonstandard fields, an
4161: integer which is accessed with the macro @code{NOTE_LINE_NUMBER} and a
4162: string accessed with @code{NOTE_SOURCE_FILE}.
4163:
4164: If @code{NOTE_LINE_NUMBER} is positive, the note represents the
4165: position of a source line and @code{NOTE_SOURCE_FILE} is the source file name
4166: that the line came from. These notes control generation of line
4167: number data in the assembler output.
4168:
4169: Otherwise, @code{NOTE_LINE_NUMBER} is not really a line number but a
4170: code with one of the following values (and @code{NOTE_SOURCE_FILE}
4171: must contain a null pointer):
4172:
4173: @table @code
4174: @item NOTE_INSN_DELETED
4175: Such a note is completely ignorable. Some passes of the compiler
4176: delete insns by altering them into notes of this kind.
4177:
4178: @item NOTE_INSN_BLOCK_BEG
4179: @itemx NOTE_INSN_BLOCK_END
4180: These types of notes indicate the position of the beginning and end
4181: of a level of scoping of variable names. They control the output
4182: of debugging information.
4183:
4184: @item NOTE_INSN_LOOP_BEG
4185: @itemx NOTE_INSN_LOOP_END
4186: These types of notes indicate the position of the beginning and end
4187: of a @code{while} or @code{for} loop. They enable the loop optimizer
4188: to find loops quickly.
4189: @end table
4190: @end table
4191:
4192: Here is a table of the extra fields of @samp{insn}, @samp{jump_insn}
4193: and @samp{call_insn} insns:
4194:
4195: @table @code
4196: @item PATTERN (@var{i})
4197: An expression for the side effect performed by this insn.
4198:
4199: @item REG_NOTES (@var{i})
4200: A list (chain of @samp{expr_list} expressions) giving information
4201: about the usage of registers in this insn. This list is set up by the
4202: flow analysis pass; it is a null pointer until then.
4203:
4204: @item LOG_LINKS (@var{i})
4205: A list (chain of @samp{insn_list} expressions) of previous ``related''
4206: insns: insns which store into registers values that are used for the
4207: first time in this insn. (An additional constraint is that neither a
4208: jump nor a label may come between the related insns). This list is
4209: set up by the flow analysis pass; it is a null pointer until then.
4210:
4211: @item INSN_CODE (@var{i})
4212: An integer that says which pattern in the machine description matches
4213: this insn, or -1 if the matching has not yet been attempted.
4214:
4215: Such matching is never attempted and this field is not used on an insn
4216: whose pattern consists of a single @samp{use}, @samp{clobber},
4217: @samp{asm}, @samp{addr_vec} or @samp{addr_diff_vec} expression.
4218: @end table
4219:
4220: The @code{LOG_LINKS} field of an insn is a chain of @samp{insn_list}
4221: expressions. Each of these has two operands: the first is an insn,
4222: and the second is another @samp{insn_list} expression (the next one in
4223: the chain). The last @samp{insn_list} in the chain has a null pointer
4224: as second operand. The significant thing about the chain is which
4225: insns appear in it (as first operands of @samp{insn_list}
4226: expressions). Their order is not significant.
4227:
4228: The @code{REG_NOTES} field of an insn is a similar chain but of
4229: @samp{expr_list} expressions instead of @samp{insn_list}. There are four
4230: kinds of register notes, which are distinguished by the machine mode of the
4231: @samp{expr_list}, which a register note is really understood as being an
4232: @code{enum reg_note}. The first operand @var{op} of the @samp{expr_list}
4233: is data whose meaning depends on the kind of note. Here are the four
4234: kinds:
4235:
4236: @table @code
4237: @item REG_DEAD
4238: The register @var{op} dies in this insn; that is to say, altering the
4239: value immediately after this insn would not affect the future behavior
4240: of the program.
4241:
4242: @item REG_INC
4243: The register @var{op} is incremented (or decremented; at this level
4244: there is no distinction) by an embedded side effect inside this insn.
4245: This means it appears in a @code{POST_INC}, @code{PRE_INC},
4246: @code{POST_DEC} or @code{PRE_DEC} RTX.
4247:
4248: @item REG_EQUIV
4249: The register that is set by this insn will be equal to @var{op} at run
4250: time, and could validly be replaced in all its occurrences by
4251: @var{op}. (``Validly'' here refers to the data flow of the program;
4252: simple replacement may make some insns invalid.)
4253:
4254: The value which the insn explicitly copies into the register may look
4255: different from @var{op}, but they will be equal at run time.
4256:
4257: For example, when a constant is loaded into a register that is never
4258: assigned any other value, this kind of note is used.
4259:
4260: When a parameter is copied into a pseudo-register at entry to a function,
4261: a note of this kind records that the register is equivalent to the stack
4262: slot where the parameter was passed. Although in this case the register
4263: may be set by other insns, it is still valid to replace the register
4264: by the stack slot throughout the function.
4265:
4266: @item REG_EQUAL
4267: The register that is set by this insn will be equal to @var{op} at run
4268: time at the end of this insn (but not necessarily elsewhere in the
4269: function).
4270:
4271: The RTX @var{op} is typically an arithmetic expression. For example,
4272: when a sequence of insns such as a library call is used to perform an
4273: arithmetic operation, this kind of note is attached to the insn that
4274: produces or copies the final value. It tells the CSE pass how to
4275: think of that value.
4276:
4277: @item REG_RETVAL
4278: This insn copies the value of a library call, and @var{op} is the
4279: first insn that was generated to set up the arguments for the library
4280: call.
4281:
4282: Flow analysis uses this note to delete all of a library call whose
4283: result is dead.
4284:
4285: @item REG_WAS_0
4286: The register @var{op} contained zero before this insn. You can rely
4287: on this note if it is present; its absence implies nothing.
4288:
4289: @item REG_LIBCALL
4290: This is the inverse of @code{REG_RETVAL}: it is placed on the first
4291: insn of a library call, and it points to the last one.
4292:
4293: Loop optimization uses this note to move an entire library call out
4294: of a loop when its value is constant.
4295:
4296: @item REG_NONNEG
4297: The register @var{op} is known to have nonnegative value when this
4298: insn is reached.
4299: @end table
4300:
4301: (The only difference between the expression codes @samp{insn_list} and
4302: @samp{expr_list} is that the first operand of an @samp{insn_list} is
4303: assumed to be an insn and is printed in debugging dumps as the insn's
4304: unique id; the first operand of an @samp{expr_list} is printed in the
4305: ordinary way as an expression.)
4306:
4307: @node Calls, Sharing, Insns, RTL
4308: @section RTL Representation of Function-Call Insns
4309:
4310: Insns that call subroutines have the RTL expression code @samp{call_insn}.
4311: These insns must satisfy special rules, and their bodies must use a special
4312: RTL expression code, @samp{call}.
4313:
4314: A @samp{call} expression has two operands, as follows:
4315:
4316: @example
4317: (call @var{nbytes} (mem:@var{fm} @var{addr}))
4318: @end example
4319:
4320: @noindent
4321: Here @var{nbytes} is an operand that represents the number of bytes of
4322: argument data being passed to the subroutine, @var{fm} is a machine mode
4323: (which must equal as the definition of the @code{FUNCTION_MODE} macro in
4324: the machine description) and @var{addr} represents the address of the
4325: subroutine.
4326:
4327: For a subroutine that returns no value, the @samp{call} RTX as shown above
4328: is the entire body of the insn.
4329:
4330: For a subroutine that returns a value whose mode is not @code{BLKmode},
4331: the value is returned in a hard register. If this register's number is
4332: @var{r}, then the body of the call insn looks like this:
4333:
4334: @example
4335: (set (reg:@var{m} @var{r})
4336: (call @var{nbytes} (mem:@var{fm} @var{addr})))
4337: @end example
4338:
4339: @noindent
4340: This RTL expression makes it clear (to the optimizer passes) that the
4341: appropriate register receives a useful value in this insn.
4342:
4343: Immediately after RTL generation, if the value of the subroutine is
4344: actually used, this call insn is always followed closely by an insn which
4345: refers to the register @var{r}. This remains true through all the
4346: optimizer passes until cross jumping occurs.
4347:
4348: The following insn has one of two forms. Either it copies the value into a
4349: pseudo-register, like this:
4350:
4351: @example
4352: (set (reg:@var{m} @var{p}) (reg:@var{m} @var{r}))
4353: @end example
4354:
4355: @noindent
4356: or (in the case where the calling function will simply return whatever
4357: value the call produced, and no operation is needed to do this):
4358:
4359: @example
4360: (use (reg:@var{m} @var{r}))
4361: @end example
4362:
4363: @noindent
4364: Between the call insn and this following insn there may intervene only a
4365: stack-adjustment insn (and perhaps some @samp{note} insns).
4366:
4367: When a subroutine returns a @code{BLKmode} value, it is handled by
4368: passing to the subroutine the address of a place to store the value.
4369: So the call insn itself does not ``return'' any value, and it has the
4370: same RTL form as a call that returns nothing.
4371:
4372: @node Sharing,, Calls, RTL
4373: @section Structure Sharing Assumptions
4374:
4375: The compiler assumes that certain kinds of RTL expressions are unique;
4376: there do not exist two distinct objects representing the same value.
4377: In other cases, it makes an opposite assumption: that no RTL expression
4378: object of a certain kind appears in more than one place in the
4379: containing structure.
4380:
4381: These assumptions refer to a single function; except for the RTL
4382: objects that describe global variables and external functions,
4383: no RTL objects are common to two functions.
4384:
4385: @itemize @bullet
4386: @item
4387: Each pseudo-register has only a single @samp{reg} object to represent it,
4388: and therefore only a single machine mode.
4389:
4390: @item
4391: For any symbolic label, there is only one @samp{symbol_ref} object
4392: referring to it.
4393:
4394: @item
4395: There is only one @samp{const_int} expression with value zero,
4396: and only one with value one.
4397:
4398: @item
4399: There is only one @samp{pc} expression.
4400:
4401: @item
4402: There is only one @samp{cc0} expression.
4403:
4404: @item
4405: There is only one @samp{const_double} expression with mode
4406: @code{SFmode} and value zero, and only one with mode @code{DFmode} and
4407: value zero.
4408:
4409: @item
4410: No @samp{label_ref} appears in more than one place in the RTL
4411: structure; in other words, it is safe to do a tree-walk of all the
4412: insns in the function and assume that each time a @samp{label_ref} is
4413: seen it is distinct from all others that are seen.
4414:
4415: @item
4416: Only one @samp{mem} object is normally created for each static
4417: variable or stack slot, so these objects are frequently shared in all
4418: the places they appear. However, separate but equal objects for these
4419: variables are occasionally made.
4420:
4421: @item
4422: No RTL object appears in more than one place in the RTL structure
4423: except as described above. Many passes of the compiler rely on this
4424: by assuming that they can modify RTL objects in place without unwanted
4425: side-effects on other insns.
4426:
4427: @item
4428: During initial RTL generation, shared structure is freely introduced.
4429: After all the RTL for a function has been generated, all shared
4430: structure is copied by @code{unshare_all_rtl} in @file{emit-rtl.c},
4431: after which the above rules are guaranteed to be followed.
4432:
4433: @item
4434: During the combiner pass, shared structure with an insn can exist
4435: temporarily. However, the shared structure is copied before the
4436: combiner is finished with the insn. This is done by
4437: @code{copy_substitutions} in @samp{combine.c}.
4438: @end itemize
4439:
4440: @node Machine Desc, Machine Macros, RTL, Top
4441: @chapter Machine Descriptions
4442:
4443: A machine description has two parts: a file of instruction patterns
4444: (@file{.md} file) and a C header file of macro definitions.
4445:
4446: The @file{.md} file for a target machine contains a pattern for each
4447: instruction that the target machine supports (or at least each instruction
4448: that is worth telling the compiler about). It may also contain comments.
4449: A semicolon causes the rest of the line to be a comment, unless the semicolon
4450: is inside a quoted string.
4451:
4452: See the next chapter for information on the C header file.
4453:
4454: @menu
4455: * Patterns:: How to write instruction patterns.
4456: * Example:: An explained example of a @samp{define_insn} pattern.
4457: * RTL Template:: The RTL template defines what insns match a pattern.
4458: * Output Template:: The output template says how to make assembler code
4459: from such an insn.
4460: * Output Statement:: For more generality, write C code to output
4461: the assembler code.
4462: * Constraints:: When not all operands are general operands.
4463: * Standard Names:: Names mark patterns to use for code generation.
4464: * Pattern Ordering:: When the order of patterns makes a difference.
4465: * Dependent Patterns:: Having one pattern may make you need another.
4466: * Jump Patterns:: Special considerations for patterns for jump insns.
4467: * Peephole Definitions::Defining machine-specific peephole optimizations.
4468: * Expander Definitions::Generating a sequence of several RTL insns
4469: for a standard operation.
4470: @end menu
4471:
4472: @node Patterns, Example, Machine Desc, Machine Desc
4473: @section Everything about Instruction Patterns
4474:
4475: Each instruction pattern contains an incomplete RTL expression, with pieces
4476: to be filled in later, operand constraints that restrict how the pieces can
4477: be filled in, and an output pattern or C code to generate the assembler
4478: output, all wrapped up in a @samp{define_insn} expression.
4479:
4480: A @samp{define_insn} is an RTL expression containing four or five operands:
4481:
4482: @enumerate
4483: @item
4484: An optional name. The presence of a name indicate that this instruction
4485: pattern can perform a certain standard job for the RTL-generation
4486: pass of the compiler. This pass knows certain names and will use
4487: the instruction patterns with those names, if the names are defined
4488: in the machine description.
4489:
4490: The absence of a name is indicated by writing an empty string
4491: where the name should go. Nameless instruction patterns are never
4492: used for generating RTL code, but they may permit several simpler insns
4493: to be combined later on.
4494:
4495: Names that are not thus known and used in RTL-generation have no
4496: effect; they are equivalent to no name at all.
4497:
4498: @item
4499: The @dfn{RTL template} (@pxref{RTL Template}) is a vector of
4500: incomplete RTL expressions which show what the instruction should look
4501: like. It is incomplete because it may contain @samp{match_operand}
4502: and @samp{match_dup} expressions that stand for operands of the
4503: instruction.
4504:
4505: If the vector has only one element, that element is what the
4506: instruction should look like. If the vector has multiple elements,
4507: then the instruction looks like a @samp{parallel} expression
4508: containing that many elements as described.
4509:
4510: @item
4511: A condition. This is a string which contains a C expression that is
4512: the final test to decide whether an insn body matches this pattern.
4513:
4514: For a named pattern, the condition (if present) may not depend on
4515: the data in the insn being matched, but only the target-machine-type
4516: flags. The compiler needs to test these conditions during
4517: initialization in order to learn exactly which named instructions are
4518: available in a particular run.
4519:
4520: For nameless patterns, the condition is applied only when matching an
4521: individual insn, and only after the insn has matched the pattern's
4522: recognition template. The insn's operands may be found in the vector
4523: @code{operands}.
4524:
4525: @item
4526: The @dfn{output template}: a string that says how to output matching
4527: insns as assembler code. @samp{%} in this string specifies where
4528: to substitute the value of an operand. @xref{Output Template}.
4529:
4530: When simple substitution isn't general enough, you can specify a piece
4531: of C code to compute the output. @xref{Output Statement}.
4532:
4533: @item
4534: Optionally, some @dfn{machine-specific information}. The meaning
4535: of this information is defined only by an individual machine description;
4536: typically it might say whether this insn alters the condition codes,
4537: or how many bytes of output it generates.
4538:
4539: This operand is written as a string containing a C initializer
4540: (complete with braces) for the structure type @code{INSN_MACHINE_INFO},
4541: whose definition is up to you (@pxref{Misc}).
4542: @end enumerate
4543:
4544: @node Example, RTL Template, Patterns, Machine Desc
4545: @section Example of @samp{define_insn}
4546:
4547: Here is an actual example of an instruction pattern, for the 68000/68020.
4548:
4549: @example
4550: (define_insn "tstsi"
4551: [(set (cc0)
4552: (match_operand:SI 0 "general_operand" "rm"))]
4553: ""
4554: "*
4555: @{ if (TARGET_68020 || ! ADDRESS_REG_P (operands[0]))
4556: return \"tstl %0\";
4557: return \"cmpl #0,%0\"; @}")
4558: @end example
4559:
4560: This is an instruction that sets the condition codes based on the value of
4561: a general operand. It has no condition, so any insn whose RTL description
4562: has the form shown may be handled according to this pattern. The name
4563: @samp{tstsi} means ``test a @code{SImode} value'' and tells the RTL generation
4564: pass that, when it is necessary to test such a value, an insn to do so
4565: can be constructed using this pattern.
4566:
4567: The output control string is a piece of C code which chooses which
4568: output template to return based on the kind of operand and the specific
4569: type of CPU for which code is being generated.
4570:
4571: @samp{"rm"} is an operand constraint. Its meaning is explained below.
4572:
4573: @node RTL Template, Output Template, Example, Machine Desc
4574: @section RTL Template for Generating and Recognizing Insns
4575:
4576: The RTL template is used to define which insns match the particular pattern
4577: and how to find their operands. For named patterns, the RTL template also
4578: says how to construct an insn from specified operands.
4579:
4580: Construction involves substituting specified operands into a copy of the
4581: template. Matching involves determining the values that serve as the
4582: operands in the insn being matched. Both of these activities are
4583: controlled by special expression types that direct matching and
4584: substitution of the operands.
4585:
4586: @table @code
4587: @item (match_operand:@var{m} @var{n} @var{testfn} @var{constraint})
4588: This expression is a placeholder for operand number @var{n} of
4589: the insn. When constructing an insn, operand number @var{n}
4590: will be substituted at this point. When matching an insn, whatever
4591: appears at this position in the insn will be taken as operand
4592: number @var{n}; but it must satisfy @var{testfn} or this instruction
4593: pattern will not match at all.
4594:
4595: Operand numbers must be chosen consecutively counting from zero in
4596: each instruction pattern. There may be only one @samp{match_operand}
4597: expression in the pattern for each operand number. Usually operands
4598: are numbered in the order of appearance in @samp{match_operand}
4599: expressions.
4600:
4601: @var{testfn} is a string that is the name of a C function that accepts
4602: two arguments, a machine mode and an expression. During matching,
4603: the function will be called with @var{m} as the mode argument
4604: and the putative operand as the other argument. If it returns zero,
4605: this instruction pattern fails to match. @var{testfn} may be
4606: an empty string; then it means no test is to be done on the operand.
4607:
4608: @var{constraint} is explained later (@pxref{Constraints}).
4609:
4610: Most often, @var{testfn} is @code{"general_operand"}. It checks
4611: that the putative operand is either a constant, a register or a
4612: memory reference, and that it is valid for mode @var{m}.
4613:
4614: For an operand that must be a register, @var{testfn} should be
4615: @code{"register_operand"}. It would be valid to use
4616: @code{"general_operand"}, since the reload pass would copy any
4617: non-register operands through registers, but this would make GNU CC do
4618: extra work, and it would prevent the register allocator from doing the
4619: best possible job.
4620:
4621: For an operand that must be a constant, either @var{testfn} should be
4622: @code{"immediate_operand"}, or the instruction pattern's extra
4623: condition should check for constants, or both. You cannot expect the
4624: constraints to do this work! If the constraints allow only constants,
4625: but the predicate allows something else, the compiler will crash when
4626: that case arises.
4627:
4628: @item (match_dup @var{n})
4629: This expression is also a placeholder for operand number @var{n}.
4630: It is used when the operand needs to appear more than once in the
4631: insn.
4632:
4633: In construction, @samp{match_dup} behaves exactly like
4634: @samp{match_operand}: the operand is substituted into the insn being
4635: constructed. But in matching, @samp{match_dup} behaves differently.
4636: It assumes that operand number @var{n} has already been determined by
4637: a @samp{match_operand} appearing earlier in the recognition template,
4638: and it matches only an identical-looking expression.
4639:
4640: @item (address (match_operand:@var{m} @var{n} "address_operand" ""))
4641: This complex of expressions is a placeholder for an operand number
4642: @var{n} in a ``load address'' instruction: an operand which specifies
4643: a memory location in the usual way, but for which the actual operand
4644: value used is the address of the location, not the contents of the
4645: location.
4646:
4647: @samp{address} expressions never appear in RTL code, only in machine
4648: descriptions. And they are used only in machine descriptions that do
4649: not use the operand constraint feature. When operand constraints are
4650: in use, the letter @samp{p} in the constraint serves this purpose.
4651:
4652: @var{m} is the machine mode of the @emph{memory location being
4653: addressed}, not the machine mode of the address itself. That mode is
4654: always the same on a given target machine (it is @code{Pmode}, which
4655: normally is @code{SImode}), so there is no point in mentioning it;
4656: thus, no machine mode is written in the @samp{address} expression. If
4657: some day support is added for machines in which addresses of different
4658: kinds of objects appear differently or are used differently (such as
4659: the PDP-10), different formats would perhaps need different machine
4660: modes and these modes might be written in the @samp{address}
4661: expression.
4662: @end table
4663:
4664: @node Output Template, Output Statement, RTL Template, Machine Desc
4665: @section Output Templates and Operand Substitution
4666:
4667: The @dfn{output template} is a string which specifies how to output
4668: the assembler code for an instruction pattern. Most of the template
4669: is a fixed string which is output literally. The character @samp{%}
4670: is used to specify where to substitute an operand; it can also be
4671: used to identify places different variants of the assembler require
4672: different syntax.
4673:
4674: In the simplest case, a @samp{%} followed by a digit @var{n} says to output
4675: operand @var{n} at that point in the string.
4676:
4677: @samp{%} followed by a letter and a digit says to output an operand in an
4678: alternate fashion. Four letters have standard, built-in meanings described
4679: below. The machine description macro @code{PRINT_OPERAND} can define
4680: additional letters with nonstandard meanings.
4681:
4682: @samp{%c@var{digit}} can be used to substitute an operand that is a
4683: constant value without the syntax that normally indicates an immediate
4684: operand.
4685:
4686: @samp{%n@var{digit}} is like @samp{%c@var{digit}} except that the value of
4687: the constant is negated before printing.
4688:
4689: @samp{%a@var{digit}} can be used to substitute an operand as if it were a
4690: memory reference, with the actual operand treated as the address. This may
4691: be useful when outputting a ``load address'' instruction, because often the
4692: assembler syntax for such an instruction requires you to write the operand
4693: as if it were a memory reference.
4694:
4695: @samp{%l@var{digit}} is used to substitute a @code{label_ref} into a jump
4696: instruction.
4697:
4698: @samp{%} followed by a punctuation character specifies a substitution that
4699: does not use an operand. Only one case is standard: @samp{%%} outputs a
4700: @samp{%} into the assembler code. Other nonstandard cases can be
4701: defined in the @code{PRINT_OPERAND} macro.
4702:
4703: The template may generate multiple assembler instructions. Write the text
4704: for the instructions, with @samp{\;} between them.
4705:
4706: When the RTL contains two operand which are required by constraint to match
4707: each other, the output template must refer only to the lower-numbered operand.
4708: Matching operands are not always identical, and the rest of the compiler
4709: arranges to put the proper RTL expression for printing into the lower-numbered
4710: operand.
4711:
4712: One use of nonstandard letters or punctuation following @samp{%} is to
4713: distinguish between different assembler languages for the same machine; for
4714: example, Motorola syntax versus MIT syntax for the 68000. Motorola syntax
4715: requires periods in most opcode names, while MIT syntax does not. For
4716: example, the opcode @samp{movel} in MIT syntax is @samp{move.l} in Motorola
4717: syntax. The same file of patterns is used for both kinds of output syntax,
4718: but the character sequence @samp{%.} is used in each place where Motorola
4719: syntax wants a period. The @code{PRINT_OPERAND} macro for Motorola syntax
4720: defines the sequence to output a period; the macro for MIT syntax defines
4721: it to do nothing.
4722:
4723: @node Output Statement, Constraints, Output Template, Machine Desc
4724: @section C Statements for Generating Assembler Output
4725:
4726: Often a single fixed template string cannot produce correct and efficient
4727: assembler code for all the cases that are recognized by a single
4728: instruction pattern. For example, the opcodes may depend on the kinds of
4729: operands; or some unfortunate combinations of operands may require extra
4730: machine instructions.
4731:
4732: If the output control string starts with a @samp{*}, then it is not an
4733: output template but rather a piece of C program that should compute a
4734: template. It should execute a @code{return} statement to return the
4735: template-string you want. Most such templates use C string literals, which
4736: require doublequote characters to delimit them. To include these
4737: doublequote characters in the string, prefix each one with @samp{\}.
4738:
4739: The operands may be found in the array @code{operands}, whose C data type
4740: is @code{rtx []}.
4741:
4742: It is possible to output an assembler instruction and then go on to output
4743: or compute more of them, using the subroutine @code{output_asm_insn}. This
4744: receives two arguments: a template-string and a vector of operands. The
4745: vector may be @code{operands}, or it may be another array of @code{rtx}
4746: that you declare locally and initialize yourself.
4747:
4748: When an insn pattern has multiple alternatives in its constraints, often
4749: the appearance of the assembler code determined mostly by which alternative
4750: was matched. When this is so, the C code can test the variable
4751: @code{which_alternative}, which is the ordinal number of the alternative
4752: that was actually satisfied (0 for the first, 1 for the second alternative,
4753: etc.).
4754:
4755: For example, suppose there are two opcodes for storing zero, @samp{clrreg}
4756: for registers and @samp{clrmem} for memory locations. Here is how
4757: a pattern could use @code{which_alternative} to choose between them:
4758:
4759: @example
4760: (define_insn ""
4761: [(set (match_operand:SI 0 "general_operand" "r,m")
4762: (const_int 0))]
4763: ""
4764: "*
4765: return (which_alternative == 0
4766: ? \"clrreg %0\" : \"clrmem %0\");
4767: ")
4768: @end example
4769:
4770: @node Constraints, Standard Names, Output Statement, Machine Desc
4771: @section Operand Constraints
4772:
4773: Each @samp{match_operand} in an instruction pattern can specify a
4774: constraint for the type of operands allowed. Constraints can say whether
4775: an operand may be in a register, and which kinds of register; whether the
4776: operand can be a memory reference, and which kinds of address; whether the
4777: operand may be an immediate constant, and which possible values it may
4778: have. Constraints can also require two operands to match.
4779:
4780: @menu
4781: * Simple Constraints:: Basic use of constraints.
4782: * Multi-Alternative:: When an insn has two alternative constraint-patterns.
4783: * Class Preferences:: Constraints guide which hard register to put things in.
4784: * Modifiers:: More precise control over effects of constraints.
4785: * No Constraints:: Describing a clean machine without constraints.
4786: @end menu
4787:
4788: @node Simple Constraints, Multi-Alternative, Constraints, Constraints
4789: @subsection Simple Constraints
4790:
4791: The simplest kind of constraint is a string full of letters, each of
4792: which describes one kind of operand that is permitted. Here are
4793: the letters that are allowed:
4794:
4795: @table @asis
4796: @item @samp{m}
4797: A memory operand is allowed, with any kind of address that the machine
4798: supports in general.
4799:
4800: @item @samp{o}
4801: A memory operand is allowed, but only if the address is
4802: @dfn{offsetable}. This means that adding a small integer (actually,
4803: the width in bytes of the operand, as determined by its machine mode)
4804: may be added to the address and the result is also a valid memory
4805: address.
4806:
4807: For example, an address which is constant is offsetable; so is an
4808: address that is the sum of a register and a constant (as long as a
4809: slightly larger constant is also within the range of address-offsets
4810: supported by the machine); but an autoincrement or autodecrement
4811: address is not offsetable. More complicated indirect/indexed
4812: addresses may or may not be offsetable depending on the other
4813: addressing modes that the machine supports.
4814:
4815: Note that in an output operand which can be matched by another
4816: operand, the constraint letter @samp{o} is valid only when accompanied
4817: by both @samp{<} (if the target machine has predecrement addressing)
4818: and @samp{>} (if the target machine has preincrement addressing).
4819:
4820: When the constraint letter @samp{o} is used, the reload pass may
4821: generate instructions which copy a nonoffsetable address into an index
4822: register. The idea is that the register can be used as a replacement
4823: offsetable address. But this method requires that there be patterns
4824: to copy any kind of address into a register. Auto-increment
4825: and auto-decrement addresses are an exception; there need not be an
4826: instruction that can copy such an address into a register, because
4827: reload handles these cases specially.
4828:
4829: Most older machine designs have ``load address'' instructions which do
4830: just what is needed here. Some RISC machines do not advertise such
4831: instructions, but the possible addresses on these machines are very
4832: limited, so it is easy to fake them.
4833:
4834: @item @samp{<}
4835: A memory operand with autodecrement addressing (either predecrement or
4836: postdecrement) is allowed.
4837:
4838: @item @samp{>}
4839: A memory operand with autoincrement addressing (either preincrement or
4840: postincrement) is allowed.
4841:
4842: @item @samp{r}
4843: A register operand is allowed provided that it is in a general
4844: register.
4845:
4846: @item @samp{d}, @samp{a}, @samp{f}, @dots{}
4847: Other letters can be defined in machine-dependent fashion to stand for
4848: particular classes of registers. @samp{d}, @samp{a} and @samp{f} are
4849: defined on the 68000/68020 to stand for data, address and floating
4850: point registers.
4851:
4852: @item @samp{i}
4853: An immediate integer operand (one with constant value) is allowed.
4854: This includes symbolic constants whose values will be known only at
4855: assembly time.
4856:
4857: @item @samp{n}
4858: An immediate integer operand with a known numeric value is allowed.
4859: Many systems cannot support assembly-time constants for operands less
4860: than a word wide. Constraints for these operands should use @samp{n}
4861: rather than @samp{i}.
4862:
4863: @item @samp{I}, @samp{J}, @samp{K}, @dots{}
4864: Other letters in the range @samp{I} through @samp{M} may be defined in
4865: a machine-dependent fashion to permit immediate integer operands with
4866: explicit integer values in specified ranges. For example, on the
4867: 68000, @samp{I} is defined to stand for the range of values 1 to 8.
4868: This is the range permitted as a shift count in the shift
4869: instructions.
4870:
4871: @item @samp{F}
4872: An immediate floating operand (expression code @samp{const_double}) is
4873: allowed.
4874:
4875: @item @samp{G}, @samp{H}
4876: @samp{G} and @samp{H} may be defined in a machine-dependent fashion to
4877: permit immediate floating operands in particular ranges of values.
4878:
4879: @item @samp{s}
4880: An immediate integer operand whose value is not an explicit integer is
4881: allowed.
4882:
4883: This might appear strange; if an insn allows a constant operand with a
4884: value not known at compile time, it certainly must allow any known
4885: value. So why use @samp{s} instead of @samp{i}? Sometimes it allows
4886: better code to be generated.
4887:
4888: For example, on the 68000 in a fullword instruction it is possible to
4889: use an immediate operand; but if the immediate value is between -32
4890: and 31, better code results from loading the value into a register and
4891: using the register. This is because the load into the register can be
4892: done with a @samp{moveq} instruction. We arrange for this to happen
4893: by defining the letter @samp{K} to mean ``any integer outside the
4894: range -32 to 31'', and then specifying @samp{Ks} in the operand
4895: constraints.
4896:
4897: @item @samp{g}
4898: Any register, memory or immediate integer operand is allowed, except for
4899: registers that are not general registers.
4900:
4901: @item @samp{@var{n}} (a digit)
4902: An operand that matches operand number @var{n} is allowed.
4903: If a digit is used together with letters, the digit should come last.
4904:
4905: This is called a @dfn{matching constraint} and what it really means is
4906: that the assembler has only a single operand that fills two roles
4907: considered separate in the RTL insn. For example, an add insn has two
4908: input operands and one output operand in the RTL, but on most machines
4909: an add instruction really has only two operands, one of them an
4910: input-output operand.
4911:
4912: Matching constraints work only in circumstances like that add insn.
4913: More precisely, the matching constraint must appear in an input-only
4914: operand and the operand that it matches must be an output-only operand
4915: with a lower number.
4916:
4917: For operands to match in a particular case usually means that they
4918: are identical-looking RTL expressions. But in a few special cases
4919: specific kinds of dissimilarity are allowed. For example, @code{*x}
4920: as an input operand will match @code{*x++} as an output operand.
4921: For proper results in such cases, the output template should always
4922: use the output-operand's number when printing the operand.
4923:
4924: @item @samp{p}
4925: An operand that is a valid memory address is allowed. This is
4926: for ``load address'' and ``push address'' instructions.
4927:
4928: If @samp{p} is used in the constraint, the test-function in the
4929: @samp{match_operand} must be @code{address_operand}.
4930: @end table
4931:
4932: In order to have valid assembler code, each operand must satisfy
4933: its constraint. But a failure to do so does not prevent the pattern
4934: from applying to an insn. Instead, it directs the compiler to modify
4935: the code so that the constraint will be satisfied. Usually this is
4936: done by copying an operand into a register.
4937:
4938: Contrast, therefore, the two instruction patterns that follow:
4939:
4940: @example
4941: (define_insn ""
4942: [(set (match_operand:SI 0 "general_operand" "r")
4943: (plus:SI (match_dup 0)
4944: (match_operand:SI 1 "general_operand" "r")))]
4945: ""
4946: "@dots{}")
4947: @end example
4948:
4949: @noindent
4950: which has two operands, one of which must appear in two places, and
4951:
4952: @example
4953: (define_insn ""
4954: [(set (match_operand:SI 0 "general_operand" "r")
4955: (plus:SI (match_operand:SI 1 "general_operand" "0")
4956: (match_operand:SI 2 "general_operand" "r")))]
4957: ""
4958: "@dots{}")
4959: @end example
4960:
4961: @noindent
4962: which has three operands, two of which are required by a constraint to be
4963: identical. If we are considering an insn of the form
4964:
4965: @example
4966: (insn @var{n} @var{prev} @var{next}
4967: (set (reg:SI 3)
4968: (plus:SI (reg:SI 6) (reg:SI 109)))
4969: @dots{})
4970: @end example
4971:
4972: @noindent
4973: the first pattern would not apply at all, because this insn does not
4974: contain two identical subexpressions in the right place. The pattern would
4975: say, ``That does not look like an add instruction; try other patterns.''
4976: The second pattern would say, ``Yes, that's an add instruction, but there
4977: is something wrong with it.'' It would direct the reload pass of the
4978: compiler to generate additional insns to make the constraint true. The
4979: results might look like this:
4980:
4981: @example
4982: (insn @var{n2} @var{prev} @var{n}
4983: (set (reg:SI 3) (reg:SI 6))
4984: @dots{})
4985:
4986: (insn @var{n} @var{n2} @var{next}
4987: (set (reg:SI 3)
4988: (plus:SI (reg:SI 3) (reg:SI 109)))
4989: @dots{})
4990: @end example
4991:
4992: It is up to you to make sure that each operand, in each pattern, has
4993: constraints that can handle any RTL expression that could be present for
4994: that operand. (When multiple alternatives are in use, each pattern must,
4995: for each possible combination of operand expressions, have at least one
4996: alternative which can handle that combination of operands.) The
4997: constraints don't need to @emph{allow} any possible operand---when this is
4998: the case, they do not constrain---but they must at least point the way to
4999: reloading any possible operand so that it will fit.
5000:
5001: @itemize @bullet
5002: @item
5003: If the constraint accepts whatever operands the predicate permits,
5004: there is no problem: reloading is never necessary for this operand.
5005:
5006: For example, an operand whose constraints permit everything except
5007: registers is safe provided its predicate rejects registers.
5008:
5009: An operand whose predicate accepts only constant values is safe
5010: provided its constraints include the letter @samp{i}. If any possible
5011: constant value is accepted, then nothing less than @samp{i} will do;
5012: if the predicate is more selective, than the constraints may also be
5013: more selective.
5014:
5015: @item
5016: Any operand expression can be reloaded by copying it into a register.
5017: So if an operand's constraints allow some kind of register, it is
5018: certain to be safe. It need not permit all classes of registers; the
5019: compiler knows how to copy a register into another register of the
5020: proper class in order to make an instruction valid.
5021:
5022: @item
5023: A nonoffsetable memory reference can be reloaded by copying the
5024: address into a register. So if the constraint uses the letter
5025: @samp{o}, all memory references are taken care of.
5026:
5027: @item
5028: A constant operand can be reloaded by storing it in memory; it then
5029: becomes an offsetable memory reference. So if the constraint uses the
5030: letters @samp{o} or @samp{m}, constant operands are not a problem.
5031: @end itemize
5032:
5033: If the operand's predicate can recognize registers, but the constraint does
5034: not permit them, it can make the compiler crash. When this operand happens
5035: to be a register, the reload pass will be stymied, because it does not know
5036: how to copy a register temporarily into memory.
5037:
5038: @node Multi-Alternative, Class Preferences, Simple Constraints, Constraints
5039: @subsection Multiple Alternative Constraints
5040:
5041: Sometimes a single instruction has multiple alternative sets of possible
5042: operands. For example, on the 68000, a logical-or instruction can combine
5043: register or an immediate value into memory, or it can combine any kind of
5044: operand into a register; but it cannot combine one memory location into
5045: another.
5046:
5047: These constraints are represented as multiple alternatives. An alternative
5048: can be described by a series of letters for each operand. The overall
5049: constraint for an operand is made from the letters for this operand
5050: from the first alternative, a comma, the letters for this operand from
5051: the second alternative, a comma, and so on until the last alternative.
5052: Here is how it is done for fullword logical-or on the 68000:
5053:
5054: @example
5055: (define_insn "iorsi3"
5056: [(set (match_operand:SI 0 "general_operand" "=%m,d")
5057: (ior:SI (match_operand:SI 1 "general_operand" "0,0")
5058: (match_operand:SI 2 "general_operand" "dKs,dmKs")))]
5059: @dots{})
5060: @end example
5061:
5062: The first alternative has @samp{m} (memory) for operand 0, @samp{0} for
5063: operand 1 (meaning it must match operand 0), and @samp{dKs} for operand 2.
5064: The second alternative has @samp{d} (data register) for operand 0, @samp{0}
5065: for operand 1, and @samp{dmKs} for operand 2. The @samp{=} and @samp{%} in
5066: the constraint for operand 0 are not part of any alternative; their meaning
5067: is explained in the next section.
5068:
5069: If all the operands fit any one alternative, the instruction is valid.
5070: Otherwise, for each alternative, the compiler counts how many instructions
5071: must be added to copy the operands so that that alternative applies.
5072: The alternative requiring the least copying is chosen. If two alternatives
5073: need the same amount of copying, the one that comes first is chosen.
5074: These choices can be altered with the @samp{?} and @samp{!} characters:
5075:
5076: @table @samp
5077: @item ?
5078: Disparage slightly the alternative that the @samp{?} appears in,
5079: as a choice when no alternative applies exactly. The compiler regards
5080: this alternative as one unit more costly for each @samp{?} that appears
5081: in it.
5082:
5083: @item !
5084: Disparage severely the alternative that the @samp{!} appears in.
5085: When operands must be copied into registers, the compiler will
5086: never choose this alternative as the one to strive for.
5087: @end table
5088:
5089: When an insn pattern has multiple alternatives in its constraints,
5090: often the appearance of the assembler code determined mostly by which
5091: alternative was matched. When this is so, the C code for writing the
5092: assembler code can use the variable @code{which_alternative}, which is
5093: the ordinal number of the alternative that was actually satisfied
5094: (0 for the first, 1 for the second alternative, etc.). For example:
5095:
5096: @example
5097: (define_insn ""
5098: [(set (match_operand:SI 0 "general_operand" "r,m")
5099: (const_int 0))]
5100: ""
5101: "*
5102: return (which_alternative == 0
5103: ? \"clrreg %0\" : \"clrmem %0\");
5104: ")
5105: @end example
5106:
5107: @node Class Preferences, Modifiers, Multi-Alternative, Constraints
5108: @subsection Register Class Preferences
5109:
5110: The operand constraints have another function: they enable the compiler
5111: to decide which kind of hardware register a pseudo register is best
5112: allocated to. The compiler examines the constraints that apply to the
5113: insns that use the pseudo register, looking for the machine-dependent
5114: letters such as @samp{d} and @samp{a} that specify classes of registers.
5115: The pseudo register is put in whichever class gets the most ``votes''.
5116: The constraint letters @samp{g} and @samp{r} also vote: they vote in
5117: favor of a general register. The machine description says which registers
5118: are considered general.
5119:
5120: Of course, on some machines all registers are equivalent, and no register
5121: classes are defined. Then none of this complexity is relevant.
5122:
5123: @node Modifiers, No Constraints, Class Preferences, Constraints
5124: @subsection Constraint Modifier Characters
5125:
5126: @table @samp
5127: @item =
5128: Means that this operand is write-only for this instruction: the previous
5129: value is discarded and replaced by output data.
5130:
5131: @item +
5132: Means that this operand is both read and written by the instruction.
5133:
5134: When the compiler fixes up the operands to satisfy the constraints,
5135: it needs to know which operands are inputs to the instruction and
5136: which are outputs from it. @samp{=} identifies an output; @samp{+}
5137: identifies an operand that is both input and output; all other operands
5138: are assumed to be input only.
5139:
5140: @item &
5141: Means (in a particular alternative) that this operand is written
5142: before the instruction is finished using the input operands.
5143: Therefore, this operand may not lie in a register that is used as an
5144: input operand or as part of any memory address.
5145:
5146: @samp{&} applies only to the alternative in which it is written. In
5147: constraints with multiple alternatives, sometimes one alternative
5148: requires @samp{&} while others do not. See, for example, the
5149: @samp{movdf} insn of the 68000.
5150:
5151: @samp{&} does not obviate the need to write @samp{=}.
5152:
5153: @item %
5154: Declares the instruction to be commutative for this operand and the
5155: following operand. This means that the compiler may interchange the
5156: two operands if that is the cheapest way to make all operands fit the
5157: constraints. This is often used in patterns for addition instructions
5158: that really have only two operands: the result must go in one of the
5159: arguments. Here for example, is how the 68000 halfword-add
5160: instruction is defined:
5161:
5162: @example
5163: (define_insn "addhi3"
5164: [(set (match_operand:HI 0 "general_operand" "=m,r")
5165: (plus:HI (match_operand:HI 1 "general_operand" "%0,0")
5166: (match_operand:HI 2 "general_operand" "di,g")))]
5167: @dots{})
5168: @end example
5169:
5170: Note that in previous versions of GNU CC the @samp{%} constraint
5171: modifier always applied to operands 1 and 2 regardless of which
5172: operand it was written in. The usual custom was to write it in
5173: operand 0. Now it must be in operand 1 if the operands to be
5174: exchanged are 1 and 2.
5175:
5176: @item #
5177: Says that all following characters, up to the next comma, are to be
5178: ignored as a constraint. They are significant only for choosing
5179: register preferences.
5180:
5181: @item *
5182: Says that the following character should be ignored when choosing
5183: register preferences. @samp{*} has no effect on the meaning of the
5184: constraint as a constraint.
5185:
5186: Here is an example: the 68000 has an instruction to sign-extend a
5187: halfword in a data register, and can also sign-extend a value by
5188: copying it into an address register. While either kind of register is
5189: acceptable, the constraints on an address-register destination are
5190: less strict, so it is best if register allocation makes an address
5191: register its goal. Therefore, @samp{*} is used so that the @samp{d}
5192: constraint letter (for data register) is ignored when computing
5193: register preferences.
5194:
5195: @example
5196: (define_insn "extendhisi2"
5197: [(set (match_operand:SI 0 "general_operand" "=*d,a")
5198: (sign_extend:SI
5199: (match_operand:HI 1 "general_operand" "0,g")))]
5200: @dots{})
5201: @end example
5202: @end table
5203:
5204: @node No Constraints,, Modifiers, Constraints
5205: @subsection Not Using Constraints
5206:
5207: Some machines are so clean that operand constraints are not required. For
5208: example, on the Vax, an operand valid in one context is valid in any other
5209: context. On such a machine, every operand constraint would be @samp{g},
5210: excepting only operands of ``load address'' instructions which are
5211: written as if they referred to a memory location's contents but actual
5212: refer to its address. They would have constraint @samp{p}.
5213:
5214: For such machines, instead of writing @samp{g} and @samp{p} for all
5215: the constraints, you can choose to write a description with empty constraints.
5216: Then you write @samp{""} for the constraint in every @samp{match_operand}.
5217: Address operands are identified by writing an @samp{address} expression
5218: around the @samp{match_operand}, not by their constraints.
5219:
5220: When the machine description has just empty constraints, certain parts
5221: of compilation are skipped, making the compiler faster.
5222:
5223: @node Standard Names, Pattern Ordering, Constraints, Machine Desc
5224: @section Standard Names for Patterns Used in Generation
5225:
5226: Here is a table of the instruction names that are meaningful in the RTL
5227: generation pass of the compiler. Giving one of these names to an
5228: instruction pattern tells the RTL generation pass that it can use the
5229: pattern in to accomplish a certain task.
5230:
5231: @table @asis
5232: @item @samp{mov@var{m}}
5233: Here @var{m} is a two-letter machine mode name, in lower case. This
5234: instruction pattern moves data with that machine mode from operand 1 to
5235: operand 0. For example, @samp{movsi} moves full-word data.
5236:
5237: If operand 0 is a @samp{subreg} with mode @var{m} of a register whose
5238: natural mode is wider than @var{m}, the effect of this instruction is
5239: to store the specified value in the part of the register that corresponds
5240: to mode @var{m}. The effect on the rest of the register is undefined.
5241:
5242: This class of patterns is special in several ways. First of all, each
5243: of these names @emph{must} be defined, because there is no other way
5244: to copy a datum from one place to another.
5245:
5246: Second, these patterns are not used solely in the RTL generation pass.
5247: Even the reload pass can generate move insns to copy values from stack
5248: slots into temporary registers. When it does so, one of the operands
5249: is a hard register and the other is an operand that can have a reload.
5250:
5251: Therefore, when given such a pair of operands, the pattern must
5252: generate RTL which needs no temporary registers---no registers other
5253: than the operands. For example, if you support the pattern with a
5254: @code{define_expand}, then in such a case you mustn't call
5255: @code{force_reg} or any other such function which might generate new
5256: pseudo registers.
5257:
5258: This requirement exists even for subword modes on a RISC machine where
5259: fetching those modes from memory normally requires several insns and
5260: some temporary registers. Look in @file{spur.md} to see how the
5261: requirement is satisfied.
5262:
5263: The variety of operands that have reloads depends on the rest of the
5264: machine description, but typically on a RISC machine these can only be
5265: pseudo registers that did not get hard registers, while on other
5266: machines explicit memory references will get optional reloads.
5267:
5268: In addition, the constraints must allow any hard register to be moved
5269: to any other hard register (provided that @code{HARD_REGNO_MODE_OK}
5270: permits mode @var{m} in each of the registers).
5271:
5272: @item @samp{movstrict@var{m}}
5273: Like @samp{mov@var{m}} except that if operand 0 is a @samp{subreg}
5274: with mode @var{m} of a register whose natural mode is wider,
5275: the @samp{movstrict@var{m}} instruction is guaranteed not to alter
5276: any of the register except the part which belongs to mode @var{m}.
5277:
5278: @item @samp{add@var{m}3}
5279: Add operand 2 and operand 1, storing the result in operand 0. All operands
5280: must have mode @var{m}. This can be used even on two-address machines, by
5281: means of constraints requiring operands 1 and 0 to be the same location.
5282:
5283: @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}
5284: Similar, for other arithmetic operations.
5285:
5286: There are special considerations for register classes for logical-and
5287: instructions, affecting also the macro @code{PREFERRED_RELOAD_CLASS}.
5288: They apply not only to the patterns with these standard names, but to
5289: any patterns that will match such an instruction. @xref{Register
5290: Classes}.
5291:
5292: @item @samp{mulhisi3}
5293: Multiply operands 1 and 2, which have mode @code{HImode}, and store
5294: a @code{SImode} product in operand 0.
5295:
5296: @item @samp{mulqihi3}, @samp{mulsidi3}
5297: Similar widening-multiplication instructions of other widths.
5298:
5299: @item @samp{umulqihi3}, @samp{umulhisi3}, @samp{umulsidi3}
5300: Similar widening-multiplication instructions that do unsigned
5301: multiplication.
5302:
5303: @item @samp{divmod@var{m}4}
5304: Signed division that produces both a quotient and a remainder.
5305: Operand 1 is divided by operand 2 to produce a quotient stored
5306: in operand 0 and a remainder stored in operand 3.
5307:
5308: @item @samp{udivmod@var{m}4}
5309: Similar, but does unsigned division.
5310:
5311: @item @samp{divmod@var{m}@var{n}4}
5312: Like @samp{divmod@var{m}4} except that only the dividend has mode
5313: @var{m}; the divisor, quotient and remainder have mode @var{n}.
5314: For example, the Vax has a @samp{divmoddisi4} instruction
5315: (but it is omitted from the machine description, because it
5316: is so slow that it is faster to compute remainders by the
5317: circumlocution that the compiler will use if this instruction is
5318: not available).
5319:
5320: @item @samp{ashl@var{m}3}
5321: Arithmetic-shift operand 1 left by a number of bits specified by
5322: operand 2, and store the result in operand 0. Operand 2 has
5323: mode @code{SImode}, not mode @var{m}.
5324:
5325: @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}
5326: Other shift and rotate instructions.
5327:
5328: Logical and arithmetic left shift are the same. Machines that do not
5329: allow negative shift counts often have only one instruction for
5330: shifting left. On such machines, you should define a pattern named
5331: @samp{ashl@var{m}3} and leave @samp{lshl@var{m}3} undefined.
5332:
5333: There are special considerations for register classes for shift
5334: instructions, affecting also the macro @code{PREFERRED_RELOAD_CLASS}.
5335: They apply not only to the patterns with these standard names, but to
5336: any patterns that will match such an instruction. @xref{Register
5337: Classes}.
5338:
5339: @item @samp{neg@var{m}2}
5340: Negate operand 1 and store the result in operand 0.
5341:
5342: @item @samp{abs@var{m}2}
5343: Store the absolute value of operand 1 into operand 0.
5344:
5345: @item @samp{sqrt@var{m}2}
5346: Store the square root of operand 1 into operand 0.
5347:
5348: @item @samp{ffs@var{m}2}
5349: Store into operand 0 one plus the index of the least significant 1-bit
5350: of operand 1. If operand 1 is zero, store zero. @var{m} is the mode
5351: of operand 0; operand 1's mode is specified by the instruction
5352: pattern, and the compiler will convert the operand to that mode before
5353: generating the instruction.
5354:
5355: @item @samp{one_cmpl@var{m}2}
5356: Store the bitwise-complement of operand 1 into operand 0.
5357:
5358: @item @samp{cmp@var{m}}
5359: Compare operand 0 and operand 1, and set the condition codes.
5360: The RTL pattern should look like this:
5361:
5362: @example
5363: (set (cc0) (minus (match_operand:@var{m} 0 @dots{})
5364: (match_operand:@var{m} 1 @dots{})))
5365: @end example
5366:
5367: Each such definition in the machine description, for integer mode
5368: @var{m}, must have a corresponding @samp{tst@var{m}} pattern, because
5369: optimization can simplify the compare into a test when operand 1 is
5370: zero.
5371:
5372: @item @samp{tst@var{m}}
5373: Compare operand 0 against zero, and set the condition codes.
5374: The RTL pattern should look like this:
5375:
5376: @example
5377: (set (cc0) (match_operand:@var{m} 0 @dots{}))
5378: @end example
5379:
5380: @item @samp{movstr@var{m}}
5381: Block move instruction. The addresses of the destination and source
5382: strings are the first two operands, and both are in mode @code{Pmode}.
5383: The number of bytes to move is the third operand, in mode @var{m}.
5384:
5385: @item @samp{cmpstr@var{m}}
5386: Block compare instruction, with operands like @samp{movstr@var{m}}
5387: except that the two memory blocks are compared byte by byte
5388: in lexicographic order. The effect of the instruction is to set
5389: the condition codes.
5390:
5391: @item @samp{float@var{m}@var{n}2}
5392: Convert operand 1 (valid for fixed point mode @var{m}) to floating
5393: point mode @var{n} and store in operand 0 (which has mode @var{n}).
5394:
5395: @item @samp{fix@var{m}@var{n}2}
5396: Convert operand 1 (valid for floating point mode @var{m}) to fixed
5397: point mode @var{n} as a signed number and store in operand 0 (which
5398: has mode @var{n}). This instruction's result is defined only when
5399: the value of operand 1 is an integer.
5400:
5401: @item @samp{fixuns@var{m}@var{n}2}
5402: Convert operand 1 (valid for floating point mode @var{m}) to fixed
5403: point mode @var{n} as an unsigned number and store in operand 0 (which
5404: has mode @var{n}). This instruction's result is defined only when the
5405: value of operand 1 is an integer.
5406:
5407: @item @samp{ftrunc@var{m}2}
5408: Convert operand 1 (valid for floating point mode @var{m}) to an
5409: integer value, still represented in floating point mode @var{m}, and
5410: store it in operand 0 (valid for floating point mode @var{m}).
5411:
5412: @item @samp{fix_trunc@var{m}@var{n}2}
5413: Like @samp{fix@var{m}@var{n}2} but works for any floating point value
5414: of mode @var{m} by converting the value to an integer.
5415:
5416: @item @samp{fixuns_trunc@var{m}@var{n}2}
5417: Like @samp{fixuns@var{m}@var{n}2} but works for any floating point
5418: value of mode @var{m} by converting the value to an integer.
5419:
5420: @item @samp{trunc@var{m}@var{n}}
5421: Truncate operand 1 (valid for mode @var{m}) to mode @var{n} and
5422: store in operand 0 (which has mode @var{n}). Both modes must be fixed
5423: point or both floating point.
5424:
5425: @item @samp{extend@var{m}@var{n}}
5426: Sign-extend operand 1 (valid for mode @var{m}) to mode @var{n} and
5427: store in operand 0 (which has mode @var{n}). Both modes must be fixed
5428: point or both floating point.
5429:
5430: @item @samp{zero_extend@var{m}@var{n}}
5431: Zero-extend operand 1 (valid for mode @var{m}) to mode @var{n} and
5432: store in operand 0 (which has mode @var{n}). Both modes must be fixed
5433: point.
5434:
5435: @item @samp{extv}
5436: Extract a bit-field from operand 1 (a register or memory operand),
5437: where operand 2 specifies the width in bits and operand 3 the starting
5438: bit, and store it in operand 0. Operand 0 must have @code{Simode}.
5439: Operand 1 may have mode @code{QImode} or @code{SImode}; often
5440: @code{SImode} is allowed only for registers. Operands 2 and 3 must be
5441: valid for @code{SImode}.
5442:
5443: The RTL generation pass generates this instruction only with constants
5444: for operands 2 and 3.
5445:
5446: The bit-field value is sign-extended to a full word integer
5447: before it is stored in operand 0.
5448:
5449: @item @samp{extzv}
5450: Like @samp{extv} except that the bit-field value is zero-extended.
5451:
5452: @item @samp{insv}
5453: Store operand 3 (which must be valid for @code{SImode}) into a
5454: bit-field in operand 0, where operand 1 specifies the width in bits
5455: and operand 2 the starting bit. Operand 0 may have mode @code{QImode}
5456: or @code{SImode}; often @code{SImode} is allowed only for registers.
5457: Operands 1 and 2 must be valid for @code{SImode}.
5458:
5459: The RTL generation pass generates this instruction only with constants
5460: for operands 1 and 2.
5461:
5462: @item @samp{s@var{cond}}
5463: Store zero or nonzero in the operand according to the condition codes.
5464: Value stored is nonzero iff the condition @var{cond} is true.
5465: @var{cond} is the name of a comparison operation expression code, such
5466: as @samp{eq}, @samp{lt} or @samp{leu}.
5467:
5468: You specify the mode that the operand must have when you write the
5469: @code{match_operand} expression. The compiler automatically sees
5470: which mode you have used and supplies an operand of that mode.
5471:
5472: The value stored for a true condition must have 1 as its low bit.
5473: Otherwise the instruction is not suitable and must be omitted from the
5474: machine description. You must tell the compiler exactly which value
5475: is stored by defining the macro @code{STORE_FLAG_VALUE}.
5476:
5477: @item @samp{b@var{cond}}
5478: Conditional branch instruction. Operand 0 is a @samp{label_ref}
5479: that refers to the label to jump to. Jump if the condition codes
5480: meet condition @var{cond}.
5481:
5482: @item @samp{call}
5483: Subroutine call instruction returning no value. Operand 0 is the
5484: function to call; operand 1 is the number of bytes of arguments pushed
5485: (in mode @code{SImode}, except it is normally a @samp{const_int});
5486: operand 2 is the number of registers used as operands.
5487:
5488: On most machines, operand 2 is not actually stored into the RTL
5489: pattern. It is supplied for the sake of some RISC machines which need
5490: to put this information into the assembler code; they can put it in
5491: the RTL instead of operand 1.
5492:
5493: Operand 0 should be a @samp{mem} RTX whose address is the address of
5494: the function.
5495:
5496: @item @samp{call_value}
5497: Subroutine call instruction returning a value. Operand 0 is the hard
5498: register in which the value is returned. There are three more
5499: operands, the same as the three operands of the @samp{call}
5500: instruction (but with numbers increased by one).
5501:
5502: Subroutines that return @code{BLKmode} objects use the @samp{call}
5503: insn.
5504:
5505: @item @samp{return}
5506: Subroutine return instruction. This instruction pattern name should be
5507: defined only if a single instruction can do all the work of returning
5508: from a function.
5509:
5510: @item @samp{casesi}
5511: Instruction to jump through a dispatch table, including bounds checking.
5512: This instruction takes five operands:
5513:
5514: @enumerate
5515: @item
5516: The index to dispatch on, which has mode @code{SImode}.
5517:
5518: @item
5519: The lower bound for indices in the table, an integer constant.
5520:
5521: @item
5522: The upper bound for indices in the table, an integer constant.
5523:
5524: @item
5525: A label to jump to if the index has a value outside the bounds.
5526: (If the machine-description macro @code{CASE_DROPS_THROUGH} is defined,
5527: then an out-of-bounds index drops through to the code following
5528: the jump table instead of jumping to this label. In that case,
5529: this label is not actually used by the @samp{casesi} instruction,
5530: but it is always provided as an operand.)
5531:
5532: @item
5533: A label that precedes the table itself.
5534: @end enumerate
5535:
5536: The table is a @samp{addr_vec} or @samp{addr_diff_vec} inside of a
5537: @samp{jump_insn}. The number of elements in the table is one plus the
5538: difference between the upper bound and the lower bound.
5539:
5540: @item @samp{tablejump}
5541: Instruction to jump to a variable address. This is a low-level
5542: capability which can be used to implement a dispatch table when there
5543: is no @samp{casesi} pattern.
5544:
5545: This pattern requires two operands: the address or offset, and a label
5546: which should immediately precede the jump table. If the macro
5547: @code{CASE_VECTOR_PC_RELATIVE} is defined then the first operand is an
5548: absolute address to jump to; otherwise, it is an offset which counts
5549: from the address of the table.
5550:
5551: The @samp{tablejump} insn is always the last insn before the jump
5552: table it uses. Its assembler code normally has no need to use the
5553: second operand, but you should incorporate it in the RTL pattern so
5554: that the jump optimizer will not delete the table as unreachable code.
5555: @end table
5556:
5557: @node Pattern Ordering, Dependent Patterns, Standard Names, Machine Desc
5558: @section When the Order of Patterns Matters
5559:
5560: Sometimes an insn can match more than one instruction pattern. Then the
5561: pattern that appears first in the machine description is the one used.
5562: Therefore, more specific patterns (patterns that will match fewer things)
5563: and faster instructions (those that will produce better code when they
5564: do match) should usually go first in the description.
5565:
5566: In some cases the effect of ordering the patterns can be used to hide
5567: a pattern when it is not valid. For example, the 68000 has an
5568: instruction for converting a fullword to floating point and another
5569: for converting a byte to floating point. An instruction converting
5570: an integer to floating point could match either one. We put the
5571: pattern to convert the fullword first to make sure that one will
5572: be used rather than the other. (Otherwise a large integer might
5573: be generated as a single-byte immediate quantity, which would not work.)
5574: Instead of using this pattern ordering it would be possible to make the
5575: pattern for convert-a-byte smart enough to deal properly with any
5576: constant value.
5577:
5578: @node Dependent Patterns, Jump Patterns, Pattern Ordering, Machine Desc
5579: @section Interdependence of Patterns
5580:
5581: Every machine description must have a named pattern for each of the
5582: conditional branch names @samp{b@var{cond}}. The recognition template
5583: must always have the form
5584:
5585: @example
5586: (set (pc)
5587: (if_then_else (@var{cond} (cc0) (const_int 0))
5588: (label_ref (match_operand 0 "" ""))
5589: (pc)))
5590: @end example
5591:
5592: @noindent
5593: In addition, every machine description must have an anonymous pattern
5594: for each of the possible reverse-conditional branches. These patterns
5595: look like
5596:
5597: @example
5598: (set (pc)
5599: (if_then_else (@var{cond} (cc0) (const_int 0))
5600: (pc)
5601: (label_ref (match_operand 0 "" ""))))
5602: @end example
5603:
5604: @noindent
5605: They are necessary because jump optimization can turn direct-conditional
5606: branches into reverse-conditional branches.
5607:
5608: The compiler does more with RTL than just create it from patterns
5609: and recognize the patterns: it can perform arithmetic expression codes
5610: when constant values for their operands can be determined. As a result,
5611: sometimes having one pattern can require other patterns. For example, the
5612: Vax has no `and' instruction, but it has `and not' instructions. Here
5613: is the definition of one of them:
5614:
5615: @example
5616: (define_insn "andcbsi2"
5617: [(set (match_operand:SI 0 "general_operand" "")
5618: (and:SI (match_dup 0)
5619: (not:SI (match_operand:SI
5620: 1 "general_operand" ""))))]
5621: ""
5622: "bicl2 %1,%0")
5623: @end example
5624:
5625: @noindent
5626: If operand 1 is an explicit integer constant, an instruction constructed
5627: using that pattern can be simplified into an `and' like this:
5628:
5629: @example
5630: (set (reg:SI 41)
5631: (and:SI (reg:SI 41)
5632: (const_int 0xffff7fff)))
5633: @end example
5634:
5635: @noindent
5636: (where the integer constant is the one's complement of what
5637: appeared in the original instruction).
5638:
5639: To avoid a fatal error, the compiler must have a pattern that recognizes
5640: such an instruction. Here is what is used:
5641:
5642: @example
5643: (define_insn ""
5644: [(set (match_operand:SI 0 "general_operand" "")
5645: (and:SI (match_dup 0)
5646: (match_operand:SI 1 "general_operand" "")))]
5647: "GET_CODE (operands[1]) == CONST_INT"
5648: "*
5649: @{ operands[1]
5650: = gen_rtx (CONST_INT, VOIDmode, ~INTVAL (operands[1]));
5651: return \"bicl2 %1,%0\";
5652: @}")
5653: @end example
5654:
5655: @noindent
5656: Whereas a pattern to match a general `and' instruction is impossible to
5657: support on the Vax, this pattern is possible because it matches only a
5658: constant second argument: a special case that can be output as an `and not'
5659: instruction.
5660:
5661: A ``compare'' instruction whose RTL looks like this:
5662:
5663: @example
5664: (set (cc0) (minus @var{operand} (const_int 0)))
5665: @end example
5666:
5667: @noindent
5668: may be simplified by optimization into a ``test'' like this:
5669:
5670: @example
5671: (set (cc0) @var{operand})
5672: @end example
5673:
5674: @noindent
5675: So in the machine description, each ``compare'' pattern for an integer
5676: mode must have a corresponding ``test'' pattern that will match the
5677: result of such simplification.
5678:
5679: In some cases machines support instructions identical except for the
5680: machine mode of one or more operands. For example, there may be
5681: ``sign-extend halfword'' and ``sign-extend byte'' instructions whose
5682: patterns are
5683:
5684: @example
5685: (set (match_operand:SI 0 @dots{})
5686: (extend:SI (match_operand:HI 1 @dots{})))
5687:
5688: (set (match_operand:SI 0 @dots{})
5689: (extend:SI (match_operand:QI 1 @dots{})))
5690: @end example
5691:
5692: @noindent
5693: Constant integers do not specify a machine mode, so an instruction to
5694: extend a constant value could match either pattern. The pattern it
5695: actually will match is the one that appears first in the file. For correct
5696: results, this must be the one for the widest possible mode (@code{HImode},
5697: here). If the pattern matches the @code{QImode} instruction, the results
5698: will be incorrect if the constant value does not actually fit that mode.
5699:
5700: Such instructions to extend constants are rarely generated because they are
5701: optimized away, but they do occasionally happen in nonoptimized
5702: compilations.
5703:
5704: When an instruction has the constraint letter @samp{o}, the reload
5705: pass may generate instructions which copy a nonoffsetable address into
5706: an index register. The idea is that the register can be used as a
5707: replacement offsetable address. In order for these generated
5708: instructions to work, there must be patterns to copy any kind of valid
5709: address into a register.
5710:
5711: Most older machine designs have ``load address'' instructions which do
5712: just what is needed here. Some RISC machines do not advertise such
5713: instructions, but the possible addresses on these machines are very
5714: limited, so it is easy to fake them.
5715:
5716: Auto-increment and auto-decrement addresses are an exception; there
5717: need not be an instruction that can copy such an address into a
5718: register, because reload handles these cases in a different manner.
5719:
5720: @node Jump Patterns, Peephole Definitions, Dependent Patterns, Machine Desc
5721: @section Defining Jump Instruction Patterns
5722:
5723: GNU CC assumes that the machine has a condition code. A comparison insn
5724: sets the condition code, recording the results of both signed and unsigned
5725: comparison of the given operands. A separate branch insn tests the
5726: condition code and branches or not according its value. The branch insns
5727: come in distinct signed and unsigned flavors. Many common machines, such
5728: as the Vax, the 68000 and the 32000, work this way.
5729:
5730: Some machines have distinct signed and unsigned compare instructions, and
5731: only one set of conditional branch instructions. The easiest way to handle
5732: these machines is to treat them just like the others until the final stage
5733: where assembly code is written. At this time, when outputting code for the
5734: compare instruction, peek ahead at the following branch using
5735: @code{NEXT_INSN (insn)}. (The variable @code{insn} refers to the insn
5736: being output, in the output-writing code in an instruction pattern.) If
5737: the RTL says that is an unsigned branch, output an unsigned compare;
5738: otherwise output a signed compare. When the branch itself is output, you
5739: can treat signed and unsigned branches identically.
5740:
5741: The reason you can do this is that GNU CC always generates a pair of
5742: consecutive RTL insns, one to set the condition code and one to test it,
5743: and keeps the pair inviolate until the end.
5744:
5745: To go with this technique, you must define the machine-description macro
5746: @code{NOTICE_UPDATE_CC} to do @code{CC_STATUS_INIT}; in other words, no
5747: compare instruction is superfluous.
5748:
5749: Some machines have compare-and-branch instructions and no condition code.
5750: A similar technique works for them. When it is time to ``output'' a
5751: compare instruction, record its operands in two static variables. When
5752: outputting the branch-on-condition-code instruction that follows, actually
5753: output a compare-and-branch instruction that uses the remembered operands.
5754:
5755: It also works to define patterns for compare-and-branch instructions.
5756: In optimizing compilation, the pair of compare and branch instructions
5757: will be combined accoprding to these patterns. But this does not happen
5758: if optimization is not requested. So you must use one of the solutions
5759: above in addition to any special patterns you define.
5760:
5761: @node Peephole Definitions, Expander Definitions, Jump Patterns, Machine Desc
5762: @section Defining Machine-Specific Peephole Optimizers
5763:
5764: In addition to instruction patterns the @file{md} file may contain
5765: definitions of machine-specific peephole optimizations.
5766:
5767: The combiner does not notice certain peephole optimizations when the data
5768: flow in the program does not suggest that it should try them. For example,
5769: sometimes two consecutive insns related in purpose can be combined even
5770: though the second one does not appear to use a register computed in the
5771: first one. A machine-specific peephole optimizer can detect such
5772: opportunities.
5773:
5774: A definition looks like this:
5775:
5776: @example
5777: (define_peephole
5778: [@var{insn-pattern-1}
5779: @var{insn-pattern-2}
5780: @dots{}]
5781: "@var{condition}"
5782: "@var{template}"
5783: "@var{machine-specific info}")
5784: @end example
5785:
5786: @noindent
5787: The last string operand may be omitted if you are not using any
5788: machine-specific information in this machine description. If present,
5789: it must obey the same rules as in a @samp{define_insn}.
5790:
5791: In this skeleton, @var{insn-pattern-1} and so on are patterns to match
5792: consecutive instructions. The optimization applies to a sequence of
5793: instructions when @var{insn-pattern-1} matches the first one,
5794: @var{insn-pattern-2} matches the next, and so on.@refill
5795:
5796: @var{insn-pattern-1} and so on look @emph{almost} like the second operand
5797: of @code{define_insn}. There is one important difference: this pattern is
5798: an RTX, not a vector. If the @code{define_insn} pattern would be a vector
5799: of one element, the @var{insn-pattern} should be just that element, no
5800: vector. If the @code{define_insn} pattern would have multiple elements
5801: then the @var{insn-pattern} must place the vector inside an explicit
5802: @code{parallel} RTX.@refill
5803:
5804: The operands of the instructions are matched with @code{match_operands} and
5805: @code{match_dup}, as usual). What is not usual is that the operand numbers
5806: apply to all the instruction patterns in the definition. So, you can check
5807: for identical operands in two instructions by using @code{match_operand}
5808: in one instruction and @code{match_dup} in the other.
5809:
5810: The operand constraints used in @code{match_operand} patterns do not have
5811: any direct effect on the applicability of the optimization, but they will
5812: be validated afterward, so write constraints that are sure to fit whenever
5813: the optimization is applied. It is safe to use @code{"g"} for each
5814: operand.
5815:
5816: Once a sequence of instructions matches the patterns, the @var{condition}
5817: is checked. This is a C expression which makes the final decision whether
5818: to perform the optimization (do so if the expression is nonzero). If
5819: @var{condition} is omitted (in other words, the string is empty) then the
5820: optimization is applied to every sequence of instructions that matches the
5821: patterns.
5822:
5823: The defined peephole optimizations are applied after register allocation is
5824: complete. Therefore, the optimizer can check which operands have ended up
5825: in which kinds of registers, just by looking at the operands.
5826:
5827: The way to refer to the operands in @var{condition} is to write
5828: @code{operands[@var{i}]} for operand number @var{i} (as matched by
5829: @code{(match_operand @var{i} @dots{})}). Use the variable @code{insn} to
5830: refer to the last of the insns being matched; use @code{PREV_INSN} to find
5831: the preceding insns (but be careful to skip over any @samp{note} insns that
5832: intervene).@refill
5833:
5834: When optimizing computations with intermediate results, you can use
5835: @var{condition} to match only when the intermediate results are not used
5836: elsewhere. Use the C expression @code{dead_or_set_p (@var{insn},
5837: @var{op})}, where @var{insn} is the insn in which you expect the value to
5838: be used for the last time (from the value of @code{insn}, together with use
5839: of @code{PREV_INSN}), and @var{op} is the intermediate value (from
5840: @code{operands[@var{i}]}).@refill
5841:
5842: Applying the optimization means replacing the sequence of instructions with
5843: one new instruction. The @var{template} controls ultimate output of
5844: assembler code for this combined instruction. It works exactly like the
5845: template of a @code{define_insn}. Operand numbers in this template are the
5846: same ones used in matching the original sequence of instructions.
5847:
5848: The result of a defined peephole optimizer does not need to match any of
5849: the instruction patterns, and it does not have an opportunity to match
5850: them. The peephole optimizer definition itself serves as the instruction
5851: pattern to control how the instruction is output.
5852:
5853: Defined peephole optimizers are run in the last jump optimization pass, so
5854: the instructions they produce are never combined or rearranged
5855: automatically in any way.
5856:
5857: Here is an example, taken from the 68000 machine description:
5858:
5859: @example
5860: (define_peephole
5861: [(set (reg:SI 15) (plus:SI (reg:SI 15) (const_int 4)))
5862: (set (match_operand:DF 0 "register_operand" "f")
5863: (match_operand:DF 1 "register_operand" "ad"))]
5864: "FP_REG_P (operands[0]) && ! FP_REG_P (operands[1])"
5865: "*
5866: @{
5867: rtx xoperands[2];
5868: xoperands[1] = gen_rtx (REG, SImode, REGNO (operands[1]) + 1);
5869: #ifdef MOTOROLA
5870: output_asm_insn (\"move.l %1,(sp)\", xoperands);
5871: output_asm_insn (\"move.l %1,-(sp)\", operands);
5872: return \"fmove.d (sp)+,%0\";
5873: #else
5874: output_asm_insn (\"movel %1,sp@@\", xoperands);
5875: output_asm_insn (\"movel %1,sp@@-\", operands);
5876: return \"fmoved sp@@+,%0\";
5877: #endif
5878: @}
5879: ")
5880: @end example
5881:
5882: The effect of this optimization is to change
5883:
5884: @example
5885: jbsr _foobar
5886: addql #4,sp
5887: movel d1,sp@@-
5888: movel d0,sp@@-
5889: fmoved sp@@+,fp0
5890: @end example
5891:
5892: @noindent
5893: into
5894:
5895: @example
5896: jbsr _foobar
5897: movel d1,sp@@
5898: movel d0,sp@@-
5899: fmoved sp@@+,fp0
5900: @end example
5901:
5902: @node Expander Definitions,, Peephole Definitions, Machine Desc
5903: @section Defining RTL Sequences for Code Generation
5904:
5905: On some target machines, some standard pattern names for RTL generation
5906: cannot be handled with single insn, but a sequence of RTL insns can
5907: represent them. For these target machines, you can write a
5908: @samp{define_expand} to specify how to generate the sequence of RTL.
5909:
5910: A @samp{define_expand} is an RTL expression that looks almost like a
5911: @samp{define_insn}; but, unlike the latter, a @samp{define_expand} is used
5912: only for RTL generation and it can produce more than one RTL insn.
5913:
5914: A @samp{define_expand} RTX has four operands:
5915:
5916: @itemize @bullet
5917: @item
5918: The name. Each @samp{define_expand} must have a name, since the only
5919: use for it is to refer to it by name.
5920:
5921: @item
5922: The RTL template. This is just like the RTL template for a
5923: @samp{define_peephole} in that it is a vector of RTL expressions
5924: each being one insn.
5925:
5926: @item
5927: The condition, a string containing a C expression. This expression is
5928: used to express how the availability of this pattern depends on
5929: subclasses of target machine, selected by command-line options when
5930: GNU CC is run. This is just like the condition of a
5931: @samp{define_insn} that has a standard name.
5932:
5933: @item
5934: The preparation statements, a string containing zero or more C
5935: statements which are to be executed before RTL code is generated from
5936: the RTL template.
5937:
5938: Usually these statements prepare temporary registers for use as
5939: internal operands in the RTL template, but they can also generate RTL
5940: insns directly by calling routines such as @samp{emit_insn}, etc.
5941: Any such insns precede the ones that come from the RTL template.
5942: @end itemize
5943:
5944: The RTL template, in addition to controlling generation of RTL insns,
5945: also describes the operands that need to be specified when this pattern
5946: is used. In particular, it gives a predicate for each operand.
5947:
5948: A true operand, which need to be specified in order to generate RTL from
5949: the pattern, should be described with a @samp{match_operand} in its first
5950: occurrence in the RTL template. This enters information on the operand's
5951: predicate into the tables that record such things. GNU CC uses the
5952: information to preload the operand into a register if that is required for
5953: valid RTL code. If the operand is referred to more than once, subsequent
5954: references should use @samp{match_dup}.
5955:
5956: The RTL template may also refer to internal ``operands'' which are
5957: temporary registers or labels used only within the sequence made by the
5958: @samp{define_expand}. Internal operands are substituted into the RTL
5959: template with @samp{match_dup}, never with @samp{match_operand}. The
5960: values of the internal operands are not passed in as arguments by the
5961: compiler when it requests use of this pattern. Instead, they are computed
5962: within the pattern, in the preparation statements. These statements
5963: compute the values and store them into the appropriate elements of
5964: @code{operands} so that @samp{match_dup} can find them.
5965:
5966: There are two special macros defined for use in the preparation statements:
5967: @code{DONE} and @code{FAIL}. Use them with a following semicolon,
5968: as a statement.
5969:
5970: @table @code
5971: @item DONE
5972: Use the @code{DONE} macro to end RTL generation for the pattern. The
5973: only RTL insns resulting from the pattern on this occasion will be
5974: those already emitted by explicit calls to @code{emit_insn} within the
5975: preparation statements; the RTL template will not be generated.
5976:
5977: @item FAIL
5978: Make the pattern fail on this occasion. When a pattern fails, it means
5979: that the pattern was not truly available. The calling routines in the
5980: compiler will try other strategies for code generation using other patterns.
5981:
5982: Failure is currently supported only for binary operations (addition,
5983: multiplication, shifting, etc.).
5984:
5985: Do not emit any insns explicitly with @code{emit_insn} before failing.
5986: @end table
5987:
5988: Here is an example, the definition of left-shift for the SPUR chip:
5989:
5990: @example
5991: (define_expand "ashlsi3"
5992: [(set (match_operand:SI 0 "register_operand" "")
5993: (ashift:SI
5994: (match_operand:SI 1 "register_operand" "")
5995: (match_operand:SI 2 "nonmemory_operand" "")))]
5996: ""
5997: "
5998: @{
5999: if (GET_CODE (operands[2]) != CONST_INT
6000: || (unsigned) INTVAL (operands[2]) > 3)
6001: FAIL;
6002: @}")
6003: @end example
6004:
6005: @noindent
6006: This example uses @samp{define_expand} so that it can generate an RTL insn
6007: for shifting when the shift-count is in the supported range of 0 to 3 but
6008: fail in other cases where machine insns aren't available. When it fails,
6009: the compiler tries another strategy using different patterns (such as, a
6010: library call).
6011:
6012: If the compiler were able to handle nontrivial condition-strings in
6013: patterns with names, then there would be possible to use a
6014: @samp{define_insn} in that case. Here is another case (zero-extension on
6015: the 68000) which makes more use of the power of @samp{define_expand}:
6016:
6017: @example
6018: (define_expand "zero_extendhisi2"
6019: [(set (match_operand:SI 0 "general_operand" "")
6020: (const_int 0))
6021: (set (strict_low_part
6022: (subreg:HI
6023: (match_operand:SI 0 "general_operand" "")
6024: 0))
6025: (match_operand:HI 1 "general_operand" ""))]
6026: ""
6027: "operands[1] = make_safe_from (operands[1], operands[0]);")
6028: @end example
6029:
6030: @noindent
6031: Here two RTL insns are generated, one to clear the entire output operand
6032: and the other to copy the input operand into its low half. This sequence
6033: is incorrect if the input operand refers to [the old value of] the output
6034: operand, so the preparation statement makes sure this isn't so. The
6035: function @code{make_safe_from} copies the @code{operands[1]} into a
6036: temporary register if it refers to @code{operands[0]}. It does this
6037: by emitting another RTL insn.
6038:
6039: Finally, a third example shows the use of an internal operand.
6040: Zero-extension on the SPUR chip is done by @samp{and}-ing the result
6041: against a halfword mask. But this mask cannot be represented by a
6042: @samp{const_int} because the constant value is too large to be legitimate
6043: on this machine. So it must be copied into a register with
6044: @code{force_reg} and then the register used in the @samp{and}.
6045:
6046: @example
6047: (define_expand "zero_extendhisi2"
6048: [(set (match_operand:SI 0 "register_operand" "")
6049: (and:SI (subreg:SI
6050: (match_operand:HI 1 "register_operand" "")
6051: 0)
6052: (match_dup 2)))]
6053: ""
6054: "operands[2]
6055: = force_reg (SImode, gen_rtx (CONST_INT,
6056: VOIDmode, 65535)); ")
6057: @end example
6058:
6059: @node Machine Macros, Config, Machine Desc, Top
6060: @chapter Machine Description Macros
6061:
6062: The other half of the machine description is a C header file conventionally
6063: given the name @file{tm-@var{machine}.h}. The file @file{tm.h} should be a
6064: link to it. The header file @file{config.h} includes @file{tm.h} and most
6065: compiler source files include @file{config.h}.
6066:
6067: @menu
6068: * Run-time Target:: Defining -m options like -m68000 and -m68020.
6069: * Storage Layout:: Defining sizes and alignments of data types.
6070: * Registers:: Naming and describing the hardware registers.
6071: * Register Classes:: Defining the classes of hardware registers.
6072: * Stack Layout:: Defining which way the stack grows and by how much.
6073: * Library Names:: Specifying names of subroutines to call automatically.
6074: * Addressing Modes:: Defining addressing modes valid for memory operands.
6075: * Condition Code:: Defining how insns update the condition code.
6076: * Assembler Format:: Defining how to write insns and pseudo-ops to output.
6077: * Misc:: Everything else.
6078: @end menu
6079:
6080: @node Run-time Target, Storage Layout, Machine Macros, Machine Macros
6081: @section Run-time Target Specification
6082:
6083: @table @code
6084: @item CPP_PREDEFINES
6085: Define this to be a string constant containing @samp{-D} options to
6086: define the predefined macros that identify this machine and system.
6087: These macros will be predefined unless the @samp{-ansi} option is
6088: specified.
6089:
6090: For example, on the Sun, one can use the value
6091:
6092: @example
6093: "-Dmc68000 -Dsun -Dunix"
6094: @end example
6095:
6096: @item CPP_SPEC
6097: A C string constant that tells the GNU CC driver program options to
6098: pass to CPP. It can also specify how to translate options you
6099: give to GNU CC into options for GNU CC to pass to the CPP.
6100:
6101: Do not define this macro if it does not need to do anything.
6102:
6103: @item CC1_SPEC
6104: A C string constant that tells the GNU CC driver program options to
6105: pass to CC1. It can also specify how to translate options you
6106: give to GNU CC into options for GNU CC to pass to the CC1.
6107:
6108: Do not define this macro if it does not need to do anything.
6109:
6110: @item extern int target_flags;
6111: This declaration should be present.
6112:
6113: @item TARGET_@dots{}
6114: This series of macros is to allow compiler command arguments to
6115: enable or disable the use of optional features of the target machine.
6116: For example, one machine description serves both the 68000 and
6117: the 68020; a command argument tells the compiler whether it should
6118: use 68020-only instructions or not. This command argument works
6119: by means of a macro @code{TARGET_68020} that tests a bit in
6120: @code{target_flags}.
6121:
6122: Define a macro @code{TARGET_@var{featurename}} for each such option.
6123: Its definition should test a bit in @code{target_flags}; for example:
6124:
6125: @example
6126: #define TARGET_68020 (target_flags & 1)
6127: @end example
6128:
6129: One place where these macros are used is in the condition-expressions
6130: of instruction patterns. Note how @code{TARGET_68020} appears
6131: frequently in the 68000 machine description file, @file{m68k.md}.
6132: Another place they are used is in the definitions of the other
6133: macros in the @file{tm-@var{machine}.h} file.
6134:
6135: @item TARGET_SWITCHES
6136: This macro defines names of command options to set and clear
6137: bits in @code{target_flags}. Its definition is an initializer
6138: with a subgrouping for each command option.
6139:
6140: Each subgrouping contains a string constant, that defines the option
6141: name, and a number, which contains the bits to set in
6142: @code{target_flags}. A negative number says to clear bits instead;
6143: the negative of the number is which bits to clear. The actual option
6144: name is made by appending @samp{-m} to the specified name.
6145:
6146: One of the subgroupings should have a null string. The number in
6147: this grouping is the default value for @code{target_flags}. Any
6148: target options act starting with that value.
6149:
6150: Here is an example which defines @samp{-m68000} and @samp{-m68020}
6151: with opposite meanings, and picks the latter as the default:
6152:
6153: @example
6154: #define TARGET_SWITCHES \
6155: @{ @{ "68020", 1@}, \
6156: @{ "68000", -1@}, \
6157: @{ "", 1@}@}
6158: @end example
6159:
6160: @item OVERRIDE_OPTIONS
6161: Sometimes certain combinations of command options do not make sense on
6162: a particular target machine. You can define a macro
6163: @code{OVERRIDE_OPTIONS} to take account of this. This macro, if
6164: defined, is executed once just after all the command options have been
6165: parsed.
6166: @end table
6167:
6168: @node Storage Layout, Registers, Run-time Target, Machine Macros
6169: @section Storage Layout
6170:
6171: Note that the definitions of the macros in this table which are sizes or
6172: alignments measured in bits do not need to be constant. They can be C
6173: expressions that refer to static variables, such as the @code{target_flags}.
6174: @xref{Run-time Target}.
6175:
6176: @table @code
6177: @item BITS_BIG_ENDIAN
6178: Define this macro if the most significant bit in a byte has the lowest
6179: number. This means that bit-field instructions count from the most
6180: significant bit. If the machine has no bit-field instructions, this
6181: macro is irrelevant.
6182:
6183: @item BYTES_BIG_ENDIAN
6184: Define this macro if the most significant byte in a word has the
6185: lowest number.
6186:
6187: @item WORDS_BIG_ENDIAN
6188: Define this macro if, in a multiword object, the most significant
6189: word has the lowest number.
6190:
6191: @item BITS_PER_UNIT
6192: Number of bits in an addressable storage unit (byte); normally 8.
6193:
6194: @item BITS_PER_WORD
6195: Number of bits in a word; normally 32.
6196:
6197: @item UNITS_PER_WORD
6198: Number of storage units in a word; normally 4.
6199:
6200: @item POINTER_SIZE
6201: Width of a pointer, in bits.
6202:
6203: @item POINTER_BOUNDARY
6204: Alignment required for pointers stored in memory, in bits.
6205:
6206: @item PARM_BOUNDARY
6207: Alignment required for function parameters on the stack, in bits.
6208:
6209: @item STACK_BOUNDARY
6210: Define this macro if you wish to preserve a certain alignment for
6211: the stack pointer at all times. The definition is a C expression
6212: for the desired alignment (measured in bits).
6213:
6214: @item FUNCTION_BOUNDARY
6215: Alignment required for a function entry point, in bits.
6216:
6217: @item BIGGEST_ALIGNMENT
6218: Biggest alignment that any data type can require on this machine, in bits.
6219:
6220: @item EMPTY_FIELD_BOUNDARY
6221: Alignment in bits to be given to a structure bit field that follows an
6222: empty field such as @code{int : 0;}.
6223:
6224: @item STRUCTURE_SIZE_BOUNDARY
6225: Number of bits which any structure or union's size must be a multiple of.
6226: Each structure or union's size is rounded up to a multiple of this.
6227:
6228: If you do not define this macro, the default is the same as
6229: @code{BITS_PER_UNIT}.
6230:
6231: @item STRICT_ALIGNMENT
6232: Define this if instructions will fail to work if given data not
6233: on the nominal alignment. If instructions will merely go slower
6234: in that case, do not define this macro.
6235:
6236: @item PCC_BITFIELD_TYPE_MATTERS
6237: Define this if you wish to imitate a certain bizarre behavior pattern
6238: of some instances of PCC: a bit field whose declared type is
6239: @code{int} has the same effect on the size and alignment of a
6240: structure as an actual @code{int} would have.
6241:
6242: Just what effect that is in GNU CC depends on other parameters, but on
6243: most machines it would force the structure's alignment and size to a
6244: multiple of 32 or @code{BIGGEST_ALIGNMENT} bits.
6245:
6246: @item CHECK_FLOAT_VALUE (@var{mode}, @var{value})
6247: A C statement to validate the value @var{value} (or type
6248: @code{double}) for mode @var{mode}. This means that you check whether
6249: @var{value} fits within the possible range of values for mode
6250: @var{mode} on this target machine. The mode @var{mode} is always
6251: @code{SFmode} or @code{DFmode}.
6252:
6253: If @var{value} is not valid, you should call @code{error} to print an
6254: error message and then assign some valid value to @var{value}.
6255: Allowing an invalid value to go through the compiler can produce
6256: incorrect assembler code which may even cause Unix assemblers to
6257: crash.
6258:
6259: This macro need not be defined if there is no work for it to do.
6260: @end table
6261:
6262: @node Registers, Register Classes, Storage Layout, Machine Macros
6263: @section Register Usage
6264:
6265: @table @code
6266: @item FIRST_PSEUDO_REGISTER
6267: Number of hardware registers known to the compiler. They receive
6268: numbers 0 through @code{FIRST_PSEUDO_REGISTER-1}; thus, the first
6269: pseudo register's number really is assigned the number
6270: @code{FIRST_PSEUDO_REGISTER}.
6271:
6272: @item FIXED_REGISTERS
6273: An initializer that says which registers are used for fixed purposes
6274: all throughout the compiled code and are therefore not available for
6275: general allocation. These would include the stack pointer, the frame
6276: pointer (except on machines where that can be used as a general
6277: register when no frame pointer is needed), the program counter on
6278: machines where that is considered one of the addressable registers,
6279: and any other numbered register with a standard use.
6280:
6281: This information is expressed as a sequence of numbers, separated by
6282: commas and surrounded by braces. The @var{n}th number is 1 if
6283: register @var{n} is fixed, 0 otherwise.
6284:
6285: The table initialized from this macro, and the table initialized by
6286: the following one, may be overridden at run time either automatically,
6287: by the actions of the macro @code{CONDITIONAL_REGISTER_USAGE}, or by
6288: the user with the command options @samp{-ffixed-@var{reg}},
6289: @samp{-fcall-used-@var{reg}} and @samp{-fcall-saved-@var{reg}}.
6290:
6291: @item CALL_USED_REGISTERS
6292: Like @code{FIXED_REGISTERS} but has 1 for each register that is
6293: clobbered (in general) by function calls as well as for fixed
6294: registers. This macro therefore identifies the registers that are not
6295: available for general allocation of values that must live across
6296: function calls.
6297:
6298: If a register has 0 in @code{CALL_USED_REGISTERS}, the compiler
6299: automatically saves it on function entry and restores it on function
6300: exit, if the register is used within the function.
6301:
6302: @item CONDITIONAL_REGISTER_USAGE
6303: Zero or more C statements that may conditionally modify two variables
6304: @code{fixed_regs} and @code{call_used_regs} (both of type @code{char
6305: []}) after they have been initialized from the two preceding macros.
6306:
6307: This is necessary in case the fixed or call-clobbered registers depend
6308: on target flags.
6309:
6310: You need not define this macro if it has no work to do.
6311:
6312: If the usage of an entire class of registers depends on the target
6313: flags, you may indicate this to gcc by using this macro to modify
6314: @code{fixed_regs} and @code{call_used_regs} to 1 for each of the
6315: registers in the classes which should not be used by gcc. Also define
6316: the macro @code{REG_CLASS_FROM_LETTER} to return @code{NO_REGS} if it
6317: is called with a letter for a class that shouldn't be used.
6318:
6319: (However, if this class is not included in @code{GENERAL_REGS} and all
6320: of the insn patterns whose constraints permit this class are
6321: controlled by target switches, then GCC will automatically avoid using
6322: these registers when the target switches are opposed to them.)
6323:
6324: @item OVERLAPPING_REGNO_P (@var{regno})
6325: If defined, this is a C expression whose value is @var{regno} is
6326: nonzero if hard register number @var{regno} is an overlapping
6327: register. This means a hard register which overlaps a hard register
6328: with a different number. (Such overlap is undesirable, but
6329: occasionally it allows a machine to be supported which otherwise could
6330: not be.) This macro must return nonzero for @emph{all} the registers
6331: which overlap each other. GNU CC can use an overlapping register only
6332: in certain limited ways. It can be used for allocation within a basic
6333: block, and may be spilled for reloading; that is all.
6334:
6335: If this macro is not defined, it means that none of the hard registers
6336: overlap each other. This is the usual situation.
6337:
6338: @item INSN_CLOBBERS_REGNO_P (@var{insn}, @var{regno})
6339: If defined, this is a C expression whose value should be nonzero if
6340: the insn @var{insn} has the effect of mysteriously clobbering the
6341: contents of hard register number @var{regno}. By ``mysterious'' we
6342: mean that the insn's RTL expression doesn't describe such an effect.
6343:
6344: If this macro is not defined, it means that no insn clobbers registers
6345: mysteriously. This is the usual situation; all else being equal,
6346: it is best for the RTL expression to show all the activity.
6347:
6348: @item PRESERVE_DEATH_INFO_REGNO_P (@var{regno})
6349: If defined, this is a C expression whose value is nonzero if accurate
6350: @code{REG_DEAD} notes are needed for hard register number @var{regno}
6351: at the time of outputting the assembler code. When this is so, a few
6352: optimizations that take place after register allocation and could
6353: invalidate the death notes are not done when this register is
6354: involved.
6355:
6356: You would arrange to preserve death info for a register when some
6357: of the code in the machine description which is executed to write
6358: the assembler code looks at the the death notes. This is
6359: necessary only when the actual hardware feature which GNU CC
6360: thinks of as a register is not actually a register of the usual sort.
6361: (It might, for example, be a hardware stack.)
6362:
6363: If this macro is not defined, it means that no death notes need to be
6364: preserved. This is the usual situation.
6365:
6366: @item HARD_REGNO_REGS (@var{regno}, @var{mode})
6367: A C expression for the number of consecutive hard registers, starting
6368: at register number @var{regno}, required to hold a value of mode
6369: @var{mode}.
6370:
6371: On a machine where all registers are exactly one word, a suitable
6372: definition of this macro is
6373:
6374: @example
6375: #define HARD_REGNO_NREGS(REGNO, MODE) \
6376: ((GET_MODE_SIZE (MODE) + UNITS_PER_WORD - 1) \
6377: / UNITS_PER_WORD))
6378: @end example
6379:
6380: @item HARD_REGNO_MODE_OK (@var{regno}, @var{mode})
6381: A C expression that is nonzero if it is permissible to store a value
6382: of mode @var{mode} in hard register number @var{regno} (or in several
6383: registers starting with that one). For a machine where all registers
6384: are equivalent, a suitable definition is
6385:
6386: @example
6387: #define HARD_REGNO_MODE_OK(REGNO, MODE) 1
6388: @end example
6389:
6390: It is not necessary for this macro to check for fixed register numbers
6391: because the allocation mechanism considers them to be always occupied.
6392:
6393: Many machines have special registers for floating point arithmetic.
6394: Often people assume that floating point machine modes are allowed only
6395: in floating point registers. This is not true. Any registers that
6396: can hold integers can safely @emph{hold} a floating point machine
6397: mode, whether or not floating arithmetic can be done on it in those
6398: registers.
6399:
6400: The true significance of special floating registers is rather than
6401: non-floating-point machine modes @emph{may not} go in those registers.
6402: This is true if the floating registers normalize any value stored in
6403: them, because storing a non-floating value there would garble it. If
6404: the floating registers do not automatically normalize, if you can
6405: store any bit pattern in one and retrieve it unchanged without a trap,
6406: then any machine mode may go in a floating register and this macro
6407: should say so.
6408:
6409: Sometimes there are floating registers that are especially slow to
6410: access, so that it is better to store a value in a stack frame than in
6411: such a register if floating point arithmetic is not being done. As long
6412: as the floating registers are not in class @code{GENERAL_REGS}, they
6413: will not be used unless some insn's constraint asks for one.
6414:
6415: It is obligatory to support floating point `move' instructions into
6416: and out of any registers that can hold fixed point values, because
6417: unions and structures (which have modes @samp{SImode} or
6418: @samp{DImode}) can be in those registers and they may have floating
6419: point members.
6420:
6421: There may also be a need to support fixed point `move' instructions in
6422: and out of floating point registers. Unfortunately, I have forgotten
6423: why this was so, and I don't know whether it is still true. If
6424: @code{HARD_REGNO_MODE_OK} rejects fixed point values in floating point
6425: registers, then the constraints of the fixed point `move' instructions
6426: must be designed to avoid ever trying to reload into a floating point
6427: register.
6428:
6429: @item MODES_TIEABLE_P (@var{mode1}, @var{mode2})
6430: A C expression that is nonzero if it is desirable to choose register
6431: allocation so as to avoid move instructions between a value of mode
6432: @var{mode1} and a value of mode @var{mode2}.
6433:
6434: If @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode1})} and
6435: @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode2})} are ever different
6436: for any @var{r}, then @code{MODES_TIEABLE_P (@var{mode1},
6437: @var{mode2})} must be zero.
6438:
6439: @item PC_REGNUM
6440: If the program counter has a register number, define this as that
6441: register number. Otherwise, do not define it.
6442:
6443: @item STACK_POINTER_REGNUM
6444: The register number of the stack pointer register, which must also be
6445: a fixed register according to @code{FIXED_REGISTERS}. On many
6446: machines, the hardware determines which register this is.
6447:
6448: @item FRAME_POINTER_REGNUM
6449: The register number of the frame pointer register, which is used to
6450: access automatic variables in the stack frame. On some machines, the
6451: hardware determines which register this is. On other machines, you
6452: can choose any register you wish for this purpose.
6453:
6454: @item FRAME_POINTER_REQUIRED
6455: A C expression which is nonzero if a function must have and use a
6456: frame pointer. This expression is evaluated in the reload pass, in
6457: the function @code{reload}, and it can in principle examine the
6458: current function and decide according to the facts, but on most
6459: machines the constant 0 or the constant 1 suffices. Use 0 when the
6460: machine allows code to be generated with no frame pointer, and doing
6461: so saves some time or space. Use 1 when there is no possible
6462: advantage to avoiding a frame pointer.
6463:
6464: In certain cases, the compiler does not know how to do without a frame
6465: pointer. The compiler recognizes those cases and automatically gives
6466: the function a frame pointer regardless of what
6467: @code{FRAME_POINTER_REQUIRED} says. You don't need to worry about
6468: them.@refill
6469:
6470: In a function that does not require a frame pointer, the frame pointer
6471: register can be allocated for ordinary usage, unless you mark it as a
6472: fixed register. See @code{FIXED_REGISTERS} for more information.
6473:
6474: @item ARG_POINTER_REGNUM
6475: The register number of the arg pointer register, which is used to
6476: access the function's argument list. On some machines, this is the
6477: same as the frame pointer register. On some machines, the hardware
6478: determines which register this is. On other machines, you can choose
6479: any register you wish for this purpose. If this is not the same
6480: register as the frame pointer register, then you must mark it as a
6481: fixed register according to @code{FIXED_REGISTERS}.
6482:
6483: @item STATIC_CHAIN_REGNUM
6484: The register number used for passing a function's static chain
6485: pointer. This is needed for languages such as Pascal and Algol where
6486: functions defined within other functions can access the local
6487: variables of the outer functions; it is not currently used because C
6488: does not provide this feature, but you must define the macro.
6489:
6490: The static chain register need not be a fixed register.
6491:
6492: @item STRUCT_VALUE_REGNUM
6493: When a function's value's mode is @code{BLKmode}, the value is not
6494: returned according to @code{FUNCTION_VALUE}. Instead, the caller
6495: passes the address of a block of memory in which the value should be
6496: stored.
6497:
6498: If this value is passed in a register, then @code{STRUCT_VALUE_REGNUM}
6499: should be the number of that register.
6500:
6501: @item STRUCT_VALUE
6502: If the structure value address is not passed in a register, define
6503: @code{STRUCT_VALUE} as an expression returning an RTX for the place
6504: where the address is passed. If it returns a @samp{mem} RTX, the
6505: address is passed as an ``invisible'' first argument.
6506:
6507: @item STRUCT_VALUE_INCOMING_REGNUM
6508: On some architectures the place where the structure value address
6509: is found by the called function is not the same place that the
6510: caller put it. This can be due to register windows, or it could
6511: be because the function prologue moves it to a different place.
6512:
6513: If the incoming location of the structure value address is in a
6514: register, define this macro as the register number.
6515:
6516: @item STRUCT_VALUE_INCOMING
6517: If the incoming location is not a register, define
6518: @code{STRUCT_VALUE_INCOMING} as an expression for an RTX for where the
6519: called function should find the value. If it should find the value on
6520: the stack, define this to create a @samp{mem} which refers to the
6521: frame pointer. If the value is a @samp{mem}, the compiler assumes it
6522: is for an invisible first argument, and leaves space for it when
6523: finding the first real argument.
6524:
6525: @item REG_ALLOC_ORDER
6526: If defined, an initializer for a vector of integers, containing the
6527: numbers of hard registers in the order in which the GNU CC should
6528: prefer to use them (from most preferred to least).
6529:
6530: If this macro is not defined, registers are used lowest numbered first
6531: (all else being equal).
6532:
6533: One use of this macro is on the 360, where the highest numbered
6534: registers must always be saved and the save-multiple-registers
6535: instruction supports only sequences of consecutive registers. This
6536: macro is defined to cause the highest numbered allocatable registers
6537: to be used first.
6538: @end table
6539:
6540: @node Register Classes, Stack Layout, Registers, Machine Macros
6541: @section Register Classes
6542:
6543: On many machines, the numbered registers are not all equivalent.
6544: For example, certain registers may not be allowed for indexed addressing;
6545: certain registers may not be allowed in some instructions. These machine
6546: restrictions are described to the compiler using @dfn{register classes}.
6547:
6548: You define a number of register classes, giving each one a name and saying
6549: which of the registers belong to it. Then you can specify register classes
6550: that are allowed as operands to particular instruction patterns.
6551:
6552: In general, each register will belong to several classes. In fact, one
6553: class must be named @code{ALL_REGS} and contain all the registers. Another
6554: class must be named @code{NO_REGS} and contain no registers. Often the
6555: union of two classes will be another class; however, this is not required.
6556:
6557: One of the classes must be named @code{GENERAL_REGS}. There is nothing
6558: terribly special about the name, but the operand constraint letters
6559: @samp{r} and @samp{g} specify this class. If @code{GENERAL_REGS} is
6560: the same as @code{ALL_REGS}, just define it as a macro which expands
6561: to @code{ALL_REGS}.
6562:
6563: The way classes other than @code{GENERAL_REGS} are specified in operand
6564: constraints is through machine-dependent operand constraint letters.
6565: You can define such letters to correspond to various classes, then use
6566: them in operand constraints.
6567:
6568: You should define a class for the union of two classes whenever some
6569: instruction allows both classes. For example, if an instruction allows
6570: either a floating-point (coprocessor) register or a general register for a
6571: certain operand, you should define a class @code{FLOAT_OR_GENERAL_REGS}
6572: which includes both of them. Otherwise you will get suboptimal code.
6573:
6574: You must also specify certain redundant information about the register
6575: classes: for each class, which classes contain it and which ones are
6576: contained in it; for each pair of classes, the largest class contained
6577: in their union.
6578:
6579: Register classes used for input-operands of bitwise-and or shift
6580: instructions have a special requirement: each such class must have, for
6581: each fixed-point machine mode, a subclass whose registers can transfer that
6582: mode to or from memory. For example, on some machines, the operations for
6583: single-byte values (@code{QImode}) are limited to certain registers. When
6584: this is so, each register class that is used in a bitwise-and or shift
6585: instruction must have a subclass consisting of registers from which
6586: single-byte values can be loaded or stored. This is so that
6587: @code{PREFERRED_RELOAD_CLASS} can always have a possible value to return.
6588:
6589: @table @code
6590: @item enum reg_class
6591: An enumeral type that must be defined with all the register class names
6592: as enumeral values. @code{NO_REGS} must be first. @code{ALL_REGS}
6593: must be the last register class, followed by one more enumeral value,
6594: @code{LIM_REG_CLASSES}, which is not a register class but rather
6595: tells how many classes there are.
6596:
6597: Each register class has a number, which is the value of casting
6598: the class name to type @code{int}. The number serves as an index
6599: in many of the tables described below.
6600:
6601: @item N_REG_CLASSES
6602: The number of distinct register classes, defined as follows:
6603:
6604: @example
6605: #define N_REG_CLASSES (int) LIM_REG_CLASSES
6606: @end example
6607:
6608: @item REG_CLASS_NAMES
6609: An initializer containing the names of the register classes as C string
6610: constants. These names are used in writing some of the debugging dumps.
6611:
6612: @item REG_CLASS_CONTENTS
6613: An initializer containing the contents of the register classes, as integers
6614: which are bit masks. The @var{n}th integer specifies the contents of class
6615: @var{n}. The way the integer @var{mask} is interpreted is that
6616: register @var{r} is in the class if @code{@var{mask} & (1 << @var{r})} is 1.
6617:
6618: When the machine has more than 32 registers, an integer does not suffice.
6619: Then the integers are replaced by sub-initializers, braced groupings containing
6620: several integers. Each sub-initializer must be suitable as an initializer
6621: for the type @code{HARD_REG_SET} which is defined in @file{hard-reg-set.h}.
6622:
6623: @item REGNO_REG_CLASS (@var{regno})
6624: A C expression whose value is a register class containing hard register
6625: @var{regno}. In general there is more that one such class; choose a class
6626: which is @dfn{minimal}, meaning that no smaller class also contains the
6627: register.
6628:
6629: @item BASE_REG_CLASS
6630: A macro whose definition is the name of the class to which a valid
6631: base register must belong. A base register is one used in an address
6632: which is the register value plus a displacement.
6633:
6634: @item INDEX_REG_CLASS
6635: A macro whose definition is the name of the class to which a valid
6636: index register must belong. An index register is one used in an
6637: address where its value is either multiplied by a scale factor or
6638: added to another register (as well as added to a displacement).
6639:
6640: @item REG_CLASS_FROM_LETTER (@var{char})
6641: A C expression which defines the machine-dependent operand constraint
6642: letters for register classes. If @var{char} is such a letter, the
6643: value should be the register class corresponding to it. Otherwise,
6644: the value should be @code{NO_REGS}.
6645:
6646: @item REGNO_OK_FOR_BASE_P (@var{num})
6647: A C expression which is nonzero if register number @var{num} is
6648: suitable for use as a base register in operand addresses. It may be
6649: either a suitable hard register or a pseudo register that has been
6650: allocated such a hard register.
6651:
6652: @item REGNO_OK_FOR_INDEX_P (@var{num})
6653: A C expression which is nonzero if register number @var{num} is
6654: suitable for use as an index register in operand addresses. It may be
6655: either a suitable hard register or a pseudo register that has been
6656: allocated such a hard register.
6657:
6658: The difference between an index register and a base register is that
6659: the index register may be scaled. If an address involves the sum of
6660: two registers, neither one of them scaled, then either one may be
6661: labeled the ``base'' and the other the ``index''; but whichever
6662: labeling is used must fit the machine's constraints of which registers
6663: may serve in each capacity. The compiler will try both labelings,
6664: looking for one that is valid, and will reload one or both registers
6665: only if neither labeling works.
6666:
6667: @item PREFERRED_RELOAD_CLASS (@var{x}, @var{class})
6668: A C expression that places additional restrictions on the register class
6669: to use when it is necessary to copy value @var{x} into a register in class
6670: @var{class}. The value is a register class; perhaps @var{class}, or perhaps
6671: another, smaller class. On many machines, the definition
6672:
6673: @example
6674: #define PREFERRED_RELOAD_CLASS(X,CLASS) CLASS
6675: @end example
6676:
6677: @noindent
6678: is safe.
6679:
6680: Sometimes returning a more restrictive class makes better code. For
6681: example, on the 68000, when @var{x} is an integer constant that is in range
6682: for a @samp{moveq} instruction, the value of this macro is always
6683: @code{DATA_REGS} as long as @var{class} includes the data registers.
6684: Requiring a data register guarantees that a @samp{moveq} will be used.
6685:
6686: If @var{x} is a @samp{const_double}, by returning @code{NO_REGS}
6687: you can force @var{x} into a memory constant. This is useful on
6688: certain machines where immediate floating values cannot be loaded into
6689: certain kinds of registers.
6690:
6691: In a shift instruction or a bitwise-and instruction, the mode of @var{x},
6692: the value being reloaded, may not be the same as the mode of the
6693: instruction's operand. (They will both be fixed-point modes, however.) In
6694: such a case, @var{class} may not be a safe value to return. @var{class} is
6695: certainly valid for the instruction, but it may not be valid for reloading
6696: @var{x}. This problem can occur on machines such as the 68000 and 80386
6697: where some registers can handle full-word values but cannot handle
6698: single-byte values.
6699:
6700: On such machines, this macro must examine the mode of @var{x} and return a
6701: subclass of @var{class} which can handle loads and stores of that mode. On
6702: the 68000, where address registers cannot handle @code{QImode}, if @var{x}
6703: has @code{QImode} then you must return @code{DATA_REGS}. If @var{class} is
6704: @code{ADDR_REGS}, then there is no correct value to return; but the shift
6705: and bitwise-and instructions don't use @code{ADDR_REGS}, so this fatal case
6706: never arises.
6707:
6708: @item CLASS_MAX_NREGS (@var{class}, @var{mode})
6709: A C expression for the maximum number of consecutive registers
6710: of class @var{class} needed to hold a value of mode @var{mode}.
6711:
6712: This is closely related to the macro @code{HARD_REGNO_NREGS}.
6713: In fact, the value of the macro @code{CLASS_MAX_NREGS (@var{class}, @var{mode})}
6714: should be the maximum value of @code{HARD_REGNO_NREGS (@var{regno}, @var{mode})}
6715: for all @var{regno} values in the class @var{class}.
6716:
6717: This macro helps control the handling of multiple-word values
6718: in the reload pass.
6719: @end table
6720:
6721: Two other special macros describe which constants fit which constraint
6722: letters.
6723:
6724: @table @code
6725: @item CONST_OK_FOR_LETTER_P (@var{value}, @var{c})
6726: A C expression that defines the machine-dependent operand constraint letters
6727: that specify particular ranges of integer values. If @var{c} is one
6728: of those letters, the expression should check that @var{value}, an integer,
6729: is in the appropriate range and return 1 if so, 0 otherwise. If @var{c} is
6730: not one of those letters, the value should be 0 regardless of @var{value}.
6731:
6732: @item CONST_DOUBLE_OK_FOR_LETTER_P (@var{value}, @var{c})
6733: A C expression that defines the machine-dependent operand constraint
6734: letters that specify particular ranges of floating values. If @var{c} is
6735: one of those letters, the expression should check that @var{value}, an RTX
6736: of code @samp{const_double}, is in the appropriate range and return 1 if
6737: so, 0 otherwise. If @var{c} is not one of those letters, the value should
6738: be 0 regardless of @var{value}.
6739: @end table
6740:
6741: @node Stack Layout, Library Names, Register Classes, Machine Macros
6742: @section Describing Stack Layout
6743:
6744: @table @code
6745: @item STACK_GROWS_DOWNWARD
6746: Define this macro if pushing a word onto the stack moves the stack
6747: pointer to a smaller address.
6748:
6749: When we say, ``define this macro if @dots{},'' it means that the
6750: compiler checks this macro only with @code{#ifdef} so the precise
6751: definition used does not matter.
6752:
6753: @item FRAME_GROWS_DOWNWARD
6754: Define this macro if the addresses of local variable slots are at negative
6755: offsets from the frame pointer.
6756:
6757: @item STARTING_FRAME_OFFSET
6758: Offset from the frame pointer to the first local variable slot to be allocated.
6759:
6760: If @code{FRAME_GROWS_DOWNWARD}, the next slot's offset is found by
6761: subtracting the length of the first slot from @code{STARTING_FRAME_OFFSET}.
6762: Otherwise, it is found by adding the length of the first slot to
6763: the value @code{STARTING_FRAME_OFFSET}.
6764:
6765: @item PUSH_ROUNDING (@var{npushed})
6766: A C expression that is the number of bytes actually pushed onto the
6767: stack when an instruction attempts to push @var{npushed} bytes.
6768:
6769: If the target machine does not have a push instruction, do not define
6770: this macro. That directs GNU CC to use an alternate strategy: to
6771: allocate the entire argument block and then store the arguments into
6772: it.
6773:
6774: On some machines, the definition
6775:
6776: @example
6777: #define PUSH_ROUNDING(BYTES) (BYTES)
6778: @end example
6779:
6780: @noindent
6781: will suffice. But on other machines, instructions that appear
6782: to push one byte actually push two bytes in an attempt to maintain
6783: alignment. Then the definition should be
6784:
6785: @example
6786: #define PUSH_ROUNDING(BYTES) (((BYTES) + 1) & ~1)
6787: @end example
6788:
6789: @item FIRST_PARM_OFFSET (@var{fundecl})
6790: Offset from the argument pointer register to the first argument's
6791: address. On some machines it may depend on the data type of the
6792: function. (In the next version of GNU CC, the argument will be
6793: changed to the function data type rather than its declaration.)
6794:
6795: @item FIRST_PARM_CALLER_OFFSET (@var{fundecl})
6796: Define this macro on machines where register parameters have shadow
6797: locations on the stack, at addresses below the nominal parameter.
6798: This matters because certain arguments cannot be passed on the stack.
6799: On these machines, such arguments must be stored into the shadow
6800: locations.
6801:
6802: This macro should expand into a C expression whose value is the offset
6803: of the first parameter's shadow location from the nominal stack
6804: pointer value. (That value is itself computed by adding the value of
6805: @code{STACK_POINTER_OFFSET} to the stack pointer register.)
6806:
6807: @item RETURN_POPS_ARGS (@var{funtype})
6808: A C expression that should be 1 if a function pops its own arguments
6809: on returning, or 0 if the function pops no arguments and the caller
6810: must therefore pop them all after the function returns.
6811:
6812: @var{funtype} is a C variable whose value is a tree node that
6813: describes the function in question. Normally it is a node of type
6814: @code{FUNCTION_TYPE} that describes the data type of the function.
6815: From this it is possible to obtain the data types of the value and
6816: arguments (if known).
6817:
6818: When a call to a library function is being considered, @var{funtype}
6819: will contain an identifier node for the library function. Thus, if
6820: you need to distinguish among various library functions, you can do so
6821: by their names. Note that ``library function'' in this context means
6822: a function used to perform arithmetic, whose name is known specially
6823: in the compiler and was not mentioned in the C code being compiled.
6824:
6825: On the Vax, all functions always pop their arguments, so the
6826: definition of this macro is 1. On the 68000, using the standard
6827: calling convention, no functions pop their arguments, so the value of
6828: the macro is always 0 in this case. But an alternative calling
6829: convention is available in which functions that take a fixed number of
6830: arguments pop them but other functions (such as @code{printf}) pop
6831: nothing (the caller pops all). When this convention is in use,
6832: @var{funtype} is examined to determine whether a function takes a
6833: fixed number of arguments.
6834:
6835: @item FUNCTION_VALUE (@var{valtype}, @var{func})
6836: A C expression to create an RTX representing the place where a
6837: function returns a value of data type @var{valtype}. @var{valtype} is
6838: a tree node representing a data type. Write @code{TYPE_MODE
6839: (@var{valtype})} to get the machine mode used to represent that type.
6840: On many machines, only the mode is relevant. (Actually, on most
6841: machines, scalar values are returned in the same place regardless of
6842: mode).@refill
6843:
6844: If the precise function being called is known, @var{func} is a tree
6845: node (@code{FUNCTION_DECL}) for it; otherwise, @var{func} is a null
6846: pointer. This makes it possible to use a different value-returning
6847: convention for specific functions when all their calls are
6848: known.@refill
6849:
6850: @item FUNCTION_OUTGOING_VALUE (@var{valtype}, @var{func})
6851: Define this macro if the target machine has ``register windows''
6852: so that the register in which a function returns its value is not
6853: the same as the one in which the caller sees the value.
6854:
6855: For such machines, @code{FUNCTION_VALUE} computes the register in
6856: which the caller will see the value, and
6857: @code{FUNCTION_OUTGOING_VALUE} should be defined in a similar fashion
6858: to tell the function where to put the value.@refill
6859:
6860: If @code{FUNCTION_OUTGOING_VALUE} is not defined,
6861: @code{FUNCTION_VALUE} serves both purposes.@refill
6862:
6863: @item LIBCALL_VALUE (@var{mode})
6864: A C expression to create an RTX representing the place where a library
6865: function returns a value of mode @var{mode}. If the precise function
6866: being called is known, @var{func} is a tree node
6867: (@code{FUNCTION_DECL}) for it; otherwise, @var{func} is a null
6868: pointer. This makes it possible to use a different value-returning
6869: convention for specific functions when all their calls are
6870: known.@refill
6871:
6872: Note that ``library function'' in this context means a compiler
6873: support routine, used to perform arithmetic, whose name is known
6874: specially by the compiler and was not mentioned in the C code being
6875: compiled.
6876:
6877: @item FUNCTION_VALUE_REGNO_P (@var{regno})
6878: A C expression that is nonzero if @var{regno} is the number of a hard
6879: register in which the values of called function may come back.
6880:
6881: A register whose use for returning values is limited to serving as the
6882: second of a pair (for a value of type @code{double}, say) need not be
6883: recognized by this macro. So for most machines, this definition
6884: suffices:
6885:
6886: @example
6887: #define FUNCTION_VALUE_REGNO_P(N) ((N) == 0)
6888: @end example
6889:
6890: If the machine has register windows, so that the caller and the called
6891: function use different registers for the return value, this macro
6892: should recognize only the caller's register numbers.
6893:
6894: @item FUNCTION_ARG (@var{cum}, @var{mode}, @var{type}, @var{named})
6895: A C expression that controls whether a function argument is passed
6896: in a register, and which register.
6897:
6898: The arguments are @var{cum}, which summarizes all the previous
6899: arguments; @var{mode}, the machine mode of the argument; @var{type},
6900: the data type of the argument as a tree node or 0 if that is not known
6901: (which happens for C support library functions); and @var{named},
6902: which is 1 for an ordinary argument and 0 for nameless arguments that
6903: correspond to @samp{...} in the called function's prototype.
6904:
6905: The value of the expression should either be a @samp{reg} RTX for the
6906: hard register in which to pass the argument, or zero to pass the
6907: argument on the stack.
6908:
6909: For the Vax and 68000, where normally all arguments are pushed, zero
6910: suffices as a definition.
6911:
6912: @item FUNCTION_INCOMING_ARG (@var{cum}, @var{mode}, @var{type}, @var{named})
6913: Define this macro if the target machine has ``register windows'', so
6914: that the register in which a function sees an arguments is not
6915: necessarily the same as the one in which the caller passed the
6916: argument.
6917:
6918: For such machines, @code{FUNCTION_ARG} computes the register in which
6919: the caller passes the value, and @code{FUNCTION_INCOMING_ARG} should
6920: be defined in a similar fashion to tell the function being called
6921: where the arguments will arrive.
6922:
6923: If @code{FUNCTION_INCOMING_ARG} is not defined, @code{FUNCTION_ARG}
6924: serves both purposes.@refill
6925:
6926: @item FUNCTION_ARG_PARTIAL_NREGS (@var{cum}, @var{mode}, @var{type}, @var{named})
6927: A C expression for the number of words, at the beginning of an
6928: argument, must be put in registers. The value must be zero for
6929: arguments that are passed entirely in registers or that are entirely
6930: pushed on the stack.
6931:
6932: On some machines, certain arguments must be passed partially in
6933: registers and partially in memory. On these machines, typically the
6934: first @var{n} words of arguments are passed in registers, and the rest
6935: on the stack. If a multi-word argument (a @code{double} or a
6936: structure) crosses that boundary, its first few words must be passed
6937: in registers and the rest must be pushed. This macro tells the
6938: compiler when this occurs, and how many of the words should go in
6939: registers.
6940:
6941: @code{FUNCTION_ARG} for these arguments should return the first
6942: register to be used by the caller for this argument; likewise
6943: @code{FUNCTION_INCOMING_ARG}, for the called function.
6944:
6945: @item CUMULATIVE_ARGS
6946: A C type for declaring a variable that is used as the first argument
6947: of @code{FUNCTION_ARG} and other related values. For some target
6948: machines, the type @code{int} suffices and can hold the number of
6949: bytes of argument so far.
6950:
6951: @item INIT_CUMULATIVE_ARGS (@var{cum}, @var{fntype})
6952: A C statement (sans semicolon) for initializing the variable @var{cum}
6953: for the state at the beginning of the argument list. The variable has
6954: type @code{CUMULATIVE_ARGS}. The value of @var{fntype} is the tree node
6955: for the data type of the function which will receive the args, or 0
6956: if the args are to a compiler support library function.
6957:
6958: @item FUNCTION_ARG_ADVANCE (@var{cum}, @var{mode}, @var{type}, @var{named})
6959: Update the summarizer variable @var{cum} to advance past an argument
6960: in the argument list. The values @var{mode}, @var{type} and
6961: @var{named} describe that argument. Once this is done, the variable
6962: @var{cum} is suitable for analyzing the @emph{following} argument
6963: with @code{FUNCTION_ARG}, etc.@refill
6964:
6965: @item FUNCTION_ARG_REGNO_P (@var{regno})
6966: A C expression that is nonzero if @var{regno} is the number of a hard
6967: register in which function arguments are sometimes passed. This does
6968: @emph{not} include implicit arguments such as the static chain and
6969: the structure-value address. On many machines, no registers can be
6970: used for this purpose since all function arguments are pushed on the
6971: stack.
6972:
6973: @item FUNCTION_ARG_PADDING (@var{mode}, @var{size})
6974: If defined, a C expression which determines whether, and in which direction,
6975: to pad out an argument with extra space. The value should be of type
6976: @code{enum direction}: either @code{upward} to pad above the argument,
6977: @code{downward} to pad below, or @code{none} to inhibit padding.
6978:
6979: The argument @var{size} is an RTX which describes the size of the
6980: argument, in bytes. It should be used only if @var{mode} is
6981: @code{BLKmode}. Otherwise, @var{size} is 0.
6982:
6983: This macro does not control the @emph{amount} of padding; that is
6984: always just enough to reach the next multiple of @code{PARM_BOUNDARY}.
6985:
6986: This macro has a default definition which is right for most systems.
6987: For little-endian machines, the default is to pad upward. For
6988: big-endian machines, the default is to pad downward for an argument of
6989: constant size shorter than an @code{int}, and upward otherwise.
6990:
6991: @item FUNCTION_PROLOGUE (@var{file}, @var{size})
6992: A C compound statement that outputs the assembler code for entry to a
6993: function. The prologue is responsible for setting up the stack frame,
6994: initializing the frame pointer register, saving registers that must be
6995: saved, and allocating @var{size} additional bytes of storage for the
6996: local variables. @var{size} is an integer. @var{file} is a stdio
6997: stream to which the assembler code should be output.
6998:
6999: The label for the beginning of the function need not be output by this
7000: macro. That has already been done when the macro is run.
7001:
7002: To determine which registers to save, the macro can refer to the array
7003: @code{regs_ever_live}: element @var{r} is nonzero if hard register
7004: @var{r} is used anywhere within the function. This implies the
7005: function prologue should save register @var{r}, but not if it is one
7006: of the call-used registers.
7007:
7008: On machines where functions may or may not have frame-pointers, the
7009: function entry code must vary accordingly; it must set up the frame
7010: pointer if one is wanted, and not otherwise. To determine whether a
7011: frame pointer is in wanted, the macro can refer to the variable
7012: @code{frame_pointer_needed}. The variable's value will be 1 at run
7013: time in a function that needs a frame pointer.
7014:
7015: @item FUNCTION_PROFILER (@var{file}, @var{labelno})
7016: A C statement or compound statement to output to @var{file} some
7017: assembler code to call the profiling subroutine @code{mcount}.
7018: Before calling, the assembler code must load the address of a
7019: counter variable into a register where @code{mcount} expects to
7020: find the address. The name of this variable is @samp{LP} followed
7021: by the number @var{labelno}, so you would generate the name using
7022: @samp{LP%d} in a @code{fprintf}.
7023:
7024: The details of how the address should be passed to @code{mcount} are
7025: determined by your operating system environment, not by GNU CC. To
7026: figure them out, compile a small program for profiling using the
7027: system's installed C compiler and look at the assembler code that
7028: results.
7029:
7030: @item EXIT_IGNORES_STACK
7031: Define this macro as a C expression that is nonzero if the return
7032: instruction or the function epilogue ignores the value of the stack
7033: pointer; in other words, if it is safe to delete an instruction to
7034: adjust the stack pointer before a return from the function.
7035:
7036: Note that this macro's value is relevant only for for which frame
7037: pointers are maintained. It is never possible to delete a final stack
7038: adjustment in a function that has no frame pointer, and the compiler
7039: knows this regardless of @code{EXIT_IGNORES_STACK}.
7040:
7041: @item FUNCTION_EPILOGUE (@var{file}, @var{size})
7042: A C compound statement that outputs the assembler code for exit from a
7043: function. The epilogue is responsible for restoring the saved
7044: registers and stack pointer to their values when the function was
7045: called, and returning control to the caller. This macro takes the
7046: same arguments as the macro @code{FUNCTION_PROLOGUE}, and the
7047: registers to restore are determined from @code{regs_ever_live} and
7048: @code{CALL_USED_REGISTERS} in the same way.
7049:
7050: On some machines, there is a single instruction that does all the work
7051: of returning from the function. On these machines, give that
7052: instruction the name @samp{return} and do not define the macro
7053: @code{FUNCTION_EPILOGUE} at all.
7054:
7055: Do not define a pattern named @samp{return} if you want the
7056: @code{FUNCTION_EPILOGUE} to be used. If you want the target switches
7057: to control whether return instructions or epilogues are used, define a
7058: @samp{return} pattern with a validity condition that tests the target
7059: switches appropriately. If the @samp{return} pattern's validity
7060: condition is false, epilogues will be used.
7061:
7062: On machines where functions may or may not have frame-pointers, the
7063: function exit code must vary accordingly. Sometimes the code for
7064: these two cases is completely different. To determine whether a frame
7065: pointer is in wanted, the macro can refer to the variable
7066: @code{frame_pointer_needed}. The variable's value will be 1 at run
7067: time in a function that needs a frame pointer.
7068:
7069: On some machines, some functions pop their arguments on exit while
7070: others leave that for the caller to do. For example, the 68020 when
7071: given @samp{-mrtd} pops arguments in functions that take a fixed
7072: number of arguments.
7073:
7074: Your definition of the macro @code{RETURN_POPS_ARGS} decides which
7075: functions pop their own arguments. @code{FUNCTION_EPILOGUE} needs to
7076: know what was decided. The variable @code{current_function_pops_args}
7077: is nonzero if the function should pop its own arguments. If so, use
7078: the variable @code{current_function_args_size} as the number of bytes
7079: to pop.
7080:
7081: @item FIX_FRAME_POINTER_ADDRESS (@var{addr}, @var{depth})
7082: A C compound statement to alter a memory address that uses the frame
7083: pointer register so that it uses the stack pointer register instead.
7084: This must be done in the instructions that load parameter values into
7085: registers, when the reload pass determines that a frame pointer is not
7086: necessary for the function. @var{addr} will be a C variable name, and
7087: the updated address should be stored in that variable. @var{depth}
7088: will be the current depth of stack temporaries (number of bytes of
7089: arguments currently pushed). The change in offset between a
7090: frame-pointer-relative address and a stack-pointer-relative address
7091: must include @var{depth}.
7092:
7093: Even if your machine description specifies there will always be a
7094: frame pointer in the frame pointer register, you must still define
7095: @code{FIX_FRAME_POINTER_ADDRESS}, but the definition will never be
7096: executed at run time, so it may be empty.
7097: @end table
7098:
7099: @node Library Names, Addressing Modes, Stack Layout, Machine Macros
7100: @section Library Subroutine Names
7101:
7102: @table @code
7103: @item UDIVSI3_LIBCALL
7104: A C string constant giving the name of the function to call for
7105: division of a full-word by a full-word. If you do not define this
7106: macro, the default name is used, which is @code{_udivsi3}, a function
7107: defined in @file{gnulib}.
7108:
7109: @item UMODSI3_LIBCALL
7110: A C string constant giving the name of the function to call for the
7111: remainder in division of a full-word by a full-word. If you do not
7112: define this macro, the default name is used, which is @code{_umodsi3},
7113: a function defined in @file{gnulib}.
7114:
7115: @item TARGET_MEM_FUNCTIONS
7116: Define this macro if GNU CC should generate calls to the System V
7117: (and ANSI C) library functions @code{memcpy} and @code{memset}
7118: rather than the BSD functions @code{bcopy} and @code{bzero}.
7119: @end table
7120:
7121: @node Addressing Modes, Misc, Library Names, Machine Macros
7122: @section Addressing Modes
7123:
7124: @table @code
7125: @item HAVE_POST_INCREMENT
7126: Define this macro if the machine supports post-increment addressing.
7127:
7128: @item HAVE_PRE_INCREMENT
7129: @itemx HAVE_POST_DECREMENT
7130: @itemx HAVE_PRE_DECREMENT
7131: Similar for other kinds of addressing.
7132:
7133: @item CONSTANT_ADDRESS_P (@var{x})
7134: A C expression that is 1 if the RTX @var{x} is a constant whose value
7135: is an integer. This includes integers whose values are not explicitly
7136: known, such as @samp{symbol_ref} and @samp{label_ref} expressions and
7137: @samp{const} arithmetic expressions.
7138:
7139: On most machines, this can be defined as @code{CONSTANT_P (@var{x})},
7140: but a few machines are more restrictive in which constant addresses
7141: are supported.
7142:
7143: @item MAX_REGS_PER_ADDRESS
7144: A number, the maximum number of registers that can appear in a valid
7145: memory address.
7146:
7147: @item GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{label})
7148: A C compound statement with a conditional @code{goto @var{label};}
7149: executed if @var{x} (an RTX) is a legitimate memory address on the
7150: target machine for a memory operand of mode @var{mode}.
7151:
7152: It usually pays to define several simpler macros to serve as
7153: subroutines for this one. Otherwise it may be too complicated to
7154: understand.
7155:
7156: This macro must exist in two variants: a strict variant and a
7157: non-strict one. The strict variant is used in the reload pass. It
7158: must be defined so that any pseudo-register that has not been
7159: allocated a hard register is considered a memory reference. In
7160: contexts where some kind of register is required, a pseudo-register
7161: with no hard register must be rejected.
7162:
7163: The non-strict variant is used in other passes. It must be defined to
7164: accept all pseudo-registers in every context where some kind of
7165: register is required.
7166:
7167: Compiler source files that want to use the strict variant of this
7168: macro define the macro @code{REG_OK_STRICT}. You should use an
7169: @code{#ifdef REG_OK_STRICT} conditional to define the strict variant
7170: in that case and the non-strict variant otherwise.
7171:
7172: Typically among the subroutines used to define
7173: @code{GO_IF_LEGITIMATE_ADDRESS} are subroutines to check for
7174: acceptable registers for various purposes (one for base registers, one
7175: for index registers, and so on). Then only these subroutine macros
7176: need have two variants; the higher levels of macros may be the same
7177: whether strict or not.@refill
7178:
7179: @item REG_OK_FOR_BASE_P (@var{x})
7180: A C expression that is nonzero if @var{x} (asumed to be a @code{reg}
7181: RTX) is valid for use as a base register. For hard registers, it
7182: should always accept those which the hardware permits and reject the
7183: others. Whether the macro accepts or rejects pseudo registers must be
7184: controlled by @code{REG_OK_STRICT} as described above. This usually
7185: requires two variant definitions, of which @code{REG_OK_STRICT}
7186: controls the one actually used.
7187:
7188: @item REG_OK_FOR_INDEX_P (@var{x})
7189: A C expression that is nonzero if @var{x} (asumed to be a @code{reg}
7190: RTX) is valid for use as an index register.
7191:
7192: The difference between an index register and a base register is that
7193: the index register may be scaled. If an address involves the sum of
7194: two registers, neither one of them scaled, then either one may be
7195: labeled the ``base'' and the other the ``index''; but whichever
7196: labeling is used must fit the machine's constraints of which registers
7197: may serve in each capacity. The compiler will try both labelings,
7198: looking for one that is valid, and will reload one or both registers
7199: only if neither labeling works.
7200:
7201: @item LEGITIMIZE_ADDRESS (@var{x}, @var{oldx}, @var{mode}, @var{win})
7202: A C compound statement that attempts to replace @var{x} with a valid
7203: memory address for an operand of mode @var{mode}. @var{win} will be a
7204: C statement label elsewhere in the code; the macro definition may use
7205:
7206: @example
7207: GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{win});
7208: @end example
7209:
7210: @noindent
7211: to avoid further processing if the address has become legitimate.
7212:
7213: @var{x} will always be the result of a call to @code{break_out_memory_refs},
7214: and @var{oldx} will be the operand that was given to that function to produce
7215: @var{x}.
7216:
7217: The code generated by this macro should not alter the substructure of
7218: @var{x}. If it transforms @var{x} into a more legitimate form, it
7219: should assign @var{x} (which will always be a C variable) a new value.
7220:
7221: It is not necessary for this macro to come up with a legitimate
7222: address. The compiler has standard ways of doing so in all cases. In
7223: fact, it is safe for this macro to do nothing. But often a
7224: machine-dependent strategy can generate better code.
7225:
7226: @item GO_IF_MODE_DEPENDENT_ADDRESS (@var{addr}, @var{label})
7227: A C statement or compound statement with a conditional @code{goto
7228: @var{label};} executed if memory address @var{x} (an RTX) can have
7229: different meanings depending on the machine mode of the memory
7230: reference it is used for.
7231:
7232: Autoincrement and autodecrement addresses typically have mode-dependent
7233: effects because the amount of the increment or decrement is the size
7234: of the operand being addressed. Some machines have other mode-dependent
7235: addresses. Many RISC machines have no mode-dependent addresses.
7236:
7237: You may assume that @var{addr} is a valid address for the machine.
7238:
7239: @item LEGITIMATE_CONSTANT_P (@var{x})
7240: A C expression that is nonzero if @var{x} is a legitimate constant for
7241: an immediate operand on the target machine. You can assume that
7242: either @var{x} is a @samp{const_double} or it satisfies
7243: @code{CONSTANT_P}, so you need not check these things. In fact,
7244: @samp{1} is a suitable definition for this macro on machines where any
7245: @samp{const_double} is valid and anything @code{CONSTANT_P} is valid.@refill
7246: @end table
7247:
7248: @node Misc, Condition Code, Addressing Modes, Machine Macros
7249: @section Miscellaneous Parameters
7250:
7251: @table @code
7252: @item CASE_VECTOR_MODE
7253: An alias for a machine mode name. This is the machine mode that
7254: elements of a jump-table should have.
7255:
7256: @item CASE_VECTOR_PC_RELATIVE
7257: Define this macro if jump-tables should contain relative addresses.
7258:
7259: @item CASE_DROPS_THROUGH
7260: Define this if control falls through a @code{case} insn when the index
7261: value is out of range. This means the specified default-label is
7262: actually ignored by the @code{case} insn proper.
7263:
7264: @item IMPLICIT_FIX_EXPR
7265: An alias for a tree code that should be used by default for conversion
7266: of floating point values to fixed point. Normally,
7267: @code{FIX_ROUND_EXPR} is used.@refill
7268:
7269: @item FIXUNS_TRUNC_LIKE_FIX_TRUNC
7270: Define this macro if the same instructions that convert a floating
7271: point number to a signed fixed point number also convert validly to an
7272: unsigned one.
7273:
7274: @item EASY_DIV_EXPR
7275: An alias for a tree code that is the easiest kind of division to
7276: compile code for in the general case. It may be
7277: @code{TRUNC_DIV_EXPR}, @code{FLOOR_DIV_EXPR}, @code{CEIL_DIV_EXPR} or
7278: @code{ROUND_DIV_EXPR}. These four division operators differ in how
7279: they round the result to an integer. @code{EASY_DIV_EXPR} is used
7280: when it is permissible to use any of those kinds of division and the
7281: choice should be made on the basis of efficiency.@refill
7282:
7283: @item DEFAULT_SIGNED_CHAR
7284: An expression whose value is 1 or 0, according to whether the type
7285: @code{char} should be signed or unsigned by default. The user can
7286: always override this default with the options @samp{-fsigned-char}
7287: and @samp{-funsigned-char}.
7288:
7289: @item SCCS_DIRECTIVE
7290: Define this if the preprocessor should ignore @code{#sccs} directives
7291: and print no error message.
7292:
7293: @item IDENT_DIRECTIVE
7294: Define this if the preprocessor should ignore @code{#ident} directives
7295: and print no error message.
7296:
7297: @item MOVE_MAX
7298: The maximum number of bytes that a single instruction can move quickly
7299: from memory to memory.
7300:
7301: @item INT_TYPE_SIZE
7302: A C expression for the size in bits of the type @code{int} on the
7303: target machine.
7304:
7305: @item SLOW_BYTE_ACCESS
7306: Define this macro as a C expression which is nonzero if accessing less
7307: than a word of memory (i.e. a @code{char} or a @code{short}) is slow
7308: (requires more than one instruction).
7309:
7310: @item SLOW_ZERO_EXTEND
7311: Define this macro if zero-extension (of a @code{char} or @code{short}
7312: to an @code{int}) can be done faster if the destination is a register
7313: that is known to be zero.
7314:
7315: If you define this macro, you must have instruction patterns that
7316: recognize RTL structures like this:
7317:
7318: @example
7319: (set (strict-low-part (subreg:QI (reg:SI @dots{}) 0)) @dots{})
7320: @end example
7321:
7322: @noindent
7323: and likewise for @code{HImode}.
7324:
7325: @item SHIFT_COUNT_TRUNCATED
7326: Define this macro if shift instructions ignore all but the lowest few
7327: bits of the shift count. It implies that a sign-extend or zero-extend
7328: instruction for the shift count can be omitted.
7329:
7330: @item TRULY_NOOP_TRUNCATION (@var{outprec}, @var{inprec})
7331: A C expression which is nonzero if on this machine it is safe to
7332: ``convert'' an integer of @var{inprec} bits to one of @var{outprec}
7333: bits (where @var{outprec} is smaller than @var{inprec}) by merely
7334: operating on it as if it had only @var{outprec} bits.
7335:
7336: On many machines, this expression can be 1.
7337:
7338: @item NO_FUNCTION_CSE
7339: Define this macro if it is as good or better to call a constant
7340: function address than to call an address kept in a register.
7341:
7342: @item PROMOTE_PROTOTYPES
7343: Define this macro if an argument declared as @code{char} or
7344: @code{short} in a prototype should actually be passed as an
7345: @code{int}. In addition to avoiding errors in certain cases of
7346: mismatch, it also makes for better code on certain machines.
7347:
7348: @item STORE_FLAG_VALUE
7349: A C expression for the value stored by a store-flag instruction
7350: (@code{s@var{cond}}) when the condition is true. This is usually 1 or
7351: -1; it is required to be an odd number.
7352:
7353: Do not define @code{STORE_FLAG_VALUE} if the machine has no store-flag
7354: instructions.
7355:
7356: @item Pmode
7357: An alias for the machine mode for pointers. Normally the definition
7358: can be
7359:
7360: @example
7361: #define Pmode SImode
7362: @end example
7363:
7364: @item FUNCTION_MODE
7365: An alias for the machine mode used for memory references to functions
7366: being called, in @samp{call} RTL expressions. On most machines this
7367: should be @code{QImode}.
7368:
7369: @item INSN_MACHINE_INFO
7370: This macro should expand into a C structure type to use for the
7371: machine-dependent info field specified with the optional last argument
7372: in @samp{define_insn} and @samp{define_peephole} patterns. For example,
7373: it might expand into @samp{struct machine_info}; then it would be up
7374: to you to define this structure in the @file{tm.h} file.
7375:
7376: You do not need to define this macro if you do not write the optional
7377: last argument in any of the patterns in the machine description.
7378:
7379: @item CONST_COSTS (@var{x}, @var{code})
7380: A part of a C @code{switch} statement that describes the relative
7381: costs of constant RTL expressions. It must contain @code{case} labels
7382: for expression codes @samp{const_int}, @samp{const}, @samp{symbol_ref}, @samp{label_ref}
7383: and @samp{const_double}. Each case must ultimately reach a
7384: @code{return} statement to return the relative cost of the use of that
7385: kind of constant value in an expression. The cost may depend on the
7386: precise value of the constant, which is available for examination in
7387: @var{x}.
7388:
7389: @var{code} is the expression code---redundant, since it can be
7390: obtained with @code{GET_CODE (@var{x})}.
7391:
7392: @item DOLLARS_IN_IDENTIFIERS
7393: Define this to be nonzero if the character @samp{$} should be allowed
7394: by default in identifier names.
7395: @end table
7396:
7397: @node Condition Code, Assembler Format, Misc, Machine Macros
7398: @section Condition Code Information
7399:
7400: The file @file{conditions.h} defines a variable @code{cc_status} to
7401: describe how the condition code was computed (in case the interpretation of
7402: the condition code depends on the instruction that it was set by). This
7403: variable contains the RTL expressions on which the condition code is
7404: currently based, and several standard flags.
7405:
7406: Sometimes additional machine-specific flags must be defined in the machine
7407: description header file. It can also add additional machine-specific
7408: information by defining @code{CC_STATUS_MDEP}.
7409:
7410: @table @code
7411: @item CC_STATUS_MDEP
7412: C code for a data type which is used for declaring the @code{mdep}
7413: component of @code{cc_status}. It defaults to @code{int}.
7414:
7415: @item CC_STATUS_MDEP_INIT
7416: A C expression for the initial value of the @code{mdep} field. It
7417: defaults to 0.
7418:
7419: @item NOTICE_UPDATE_CC (@var{exp}, @var{insn})
7420: A C compound statement to set the components of @code{cc_status}
7421: appropriately for an insn @var{insn} whose body is @var{exp}. It is
7422: this macro's responsibility to recognize insns that set the condition
7423: code as a byproduct of other activity as well as those that explicitly
7424: set @code{(cc0)}.
7425:
7426: If there are insn that do not set the condition code but do alter
7427: other machine registers, this macro must check to see whether they
7428: invalidate the expressions that the condition code is recorded as
7429: reflecting. For example, on the 68000, insns that store in address
7430: registers do not set the condition code, which means that usually
7431: @code{NOTICE_UPDATE_CC} can leave @code{cc_status} unaltered for such
7432: insns. But suppose that the previous insn set the condition code
7433: based on location @samp{a4@@(102)} and the current insn stores a new
7434: value in @samp{a4}. Although the condition code is not changed by
7435: this, it will no longer be true that it reflects the contents of
7436: @samp{a4@@(102)}. Therefore, @code{NOTICE_UPDATE_CC} must alter
7437: @code{cc_status} in this case to say that nothing is known about the
7438: condition code value.
7439:
7440: The definition of @code{NOTICE_UPDATE_CC} must be prepared to deal
7441: with the results of peephole optimization: insns whose patterns are
7442: @samp{parallel} RTXs containing various @samp{reg}, @samp{mem} or
7443: constants which are just the operands. The RTL structure of these
7444: insns is not sufficient to indicate what the insns actually do. What
7445: @code{NOTICE_UPDATE_CC} should do when it sees one is just to run
7446: @code{CC_STATUS_INIT}.
7447: @end table
7448:
7449: @node Assembler Format,, Condition Code, Machine Macros
7450: @section Output of Assembler Code
7451:
7452: @table @code
7453: @item ASM_SPEC
7454: A C string constant that tells the GNU CC driver program options to
7455: pass to the assembler. It can also specify how to translate options
7456: you give to GNU CC into options for GNU CC to pass to the assembler.
7457: See the file @file{tm-sun3.h} for an example of this.
7458:
7459: Do not define this macro if it does not need to do anything.
7460:
7461: @item LINK_SPEC
7462: A C string constant that tells the GNU CC driver program options to
7463: pass to the linker. It can also specify how to translate options you
7464: give to GNU CC into options for GNU CC to pass to the linker.
7465:
7466: Do not define this macro if it does not need to do anything.
7467:
7468: @item LIB_SPEC
7469: Another C string constant used much like @code{LINK_SPEC}. The difference
7470: between the two is that @code{LIBS_SPEC} is used at the end of the
7471: command given to the linker.
7472:
7473: If this macro is not defined, a default is provided that
7474: loads the standard C library from the usual place. See @file{gcc.c}.
7475:
7476: @item STARTFILE_SPEC
7477: Another C string constant used much like @code{LINK_SPEC}. The
7478: difference between the two is that @code{STARTFILE_SPEC} is used at
7479: the very beginning of the command given to the linker.
7480:
7481: If this macro is not defined, a default is provided that loads the
7482: standard C startup file from the usual place. See @file{gcc.c}.
7483:
7484: @item ASM_FILE_START (@var{stream})
7485: A C expression which outputs to the stdio stream @var{stream}
7486: some appropriate text to go at the start of an assembler file.
7487:
7488: Normally this macro is defined to output a line containing
7489: @samp{#NO_APP}, which is a comment that has no effect on most
7490: assemblers but tells the GNU assembler that it can save time by not
7491: checking for certain assembler constructs.
7492:
7493: On systems that use SDB, it is necessary to output certain commands;
7494: see @file{tm-attasm.h}.
7495:
7496: @item ASM_APP_ON
7497: A C string constant for text to be output before each @code{asm}
7498: statement or group of consecutive ones. Normally this is
7499: @code{"#APP"}, which is a comment that has no effect on most
7500: assemblers but tells the GNU assembler that it must check the lines
7501: that follow for all valid assembler constructs.
7502:
7503: @item ASM_APP_OFF
7504: A C string constant for text to be output after each @code{asm}
7505: statement or group of consecutive ones. Normally this is
7506: @code{"#NO_APP"}, which tells the GNU assembler to resume making the
7507: time-saving assumptions that are valid for ordinary compiler output.
7508:
7509: @item TEXT_SECTION_ASM_OP
7510: A C string constant for the assembler operation that should precede
7511: instructions and read-only data. Normally @code{".text"} is right.
7512:
7513: @item DATA_SECTION_ASM_OP
7514: A C string constant for the assembler operation to identify the
7515: following data as writable initialized data. Normally @code{".data"}
7516: is right.
7517:
7518: @item REGISTER_NAMES
7519: A C initializer containing the assembler's names for the machine
7520: registers, each one as a C string constant. This is what translates
7521: register numbers in the compiler into assembler language.
7522:
7523: @item DBX_REGISTER_NUMBER (@var{regno})
7524: A C expression that returns the DBX register number for the compiler
7525: register number @var{regno}. In simple cases, the value of this
7526: expression may be @var{regno} itself. But sometimes there are some
7527: registers that the compiler knows about and DBX does not, or vice
7528: versa. In such cases, some register may need to have one number in
7529: the compiler and another for DBX.
7530:
7531: @item DBX_DEBUGGING_INFO
7532: Define this macro if GNU CC should produce debugging output for DBX
7533: in response to the @samp{-g} option.
7534:
7535: @item SDB_DEBUGGING_INFO
7536: Define this macro if GNU CC should produce debugging output for SDB
7537: in response to the @samp{-g} option.
7538:
7539: @item PUT_SDB_@var{op}
7540: Define these macros to override the assembler syntax for the special
7541: SDB assembler directives. See @file{sdbout.c} for a list of these
7542: macros and their arguments. If the standard syntax is used, you need
7543: not define them yourself.
7544:
7545: @item SDB_GENERATE_FAKE
7546: Define this macro to override the usual method of constructing a dummy
7547: name for anonymous structure and union types. See @file{sdbout.c} for
7548: more infomation.
7549:
7550: @item DBX_NO_XREFS
7551: Define this macro if DBX on your system does not support the construct
7552: @samp{xs@var{tagname}}. On some systems, this construct is used to
7553: describe a forward reference to a structure named @var{tagname}.
7554: On other systems, this construct is not supported at all.
7555:
7556: @item DBX_CONTIN_LENGTH
7557: A symbol name in DBX-format debugging information is normally
7558: continued (split into two separate @code{.stabs} directives) when it
7559: exceeds a certain length (by default, 80 characters). On some
7560: operating systems, DBX requires this splitting; on others, splitting
7561: must not be done. You can inhibit splitting by defining this macro
7562: with the value zero. You can override the default splitting-length by
7563: defining this macro as an expression for the length you desire.
7564:
7565: @item DBX_CONTIN_CHAR
7566: Normally continuation is indicated by adding a @samp{\} character to
7567: the end of a @code{.stabs} string when a continuation follows. To use
7568: a different character instead, define this macro as a character
7569: constant for the character you want to use. Do not define this macro
7570: if backslash is correct for your system.
7571:
7572: @item ASM_OUTPUT_LABEL (@var{stream}, @var{name})
7573: A C statement (sans semicolon) to output to the stdio stream
7574: @var{stream} the assembler definition of a label named @var{name}. Use
7575: the expression @code{assemble_name (@var{stream}, @var{name})} to output
7576: the name itself; before and after that, output the additional
7577: assembler syntax for defining the name, and a newline.
7578:
7579: @item ASM_DECLARE_FUNCTION_NAME (@var{stream}, @var{name}, @var{decl})
7580: A C statement (sans semicolon) to output to the stdio stream
7581: @var{stream} any text necessary for declaring the name @var{name} of a
7582: function which is being defined. This macro is responsible for
7583: outputting the label definition (perhaps using
7584: @code{ASM_OUTPUT_LABEL}). The argument @var{decl} is the
7585: @code{FUNCTION_DECL} tree node representing the function.
7586:
7587: If this macro is not defined, then the function name is defined in the
7588: usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}).
7589:
7590: @item ASM_GLOBALIZE_LABEL (@var{stream}, @var{name})
7591: A C statement (sans semicolon) to output to the stdio stream
7592: @var{stream} some commands that will make the label @var{name} global;
7593: that is, available for reference from other files. Use the expression
7594: @code{assemble_name (@var{stream}, @var{name})} to output the name
7595: itself; before and after that, output the additional assembler syntax
7596: for making that name global, and a newline.
7597:
7598: @item ASM_OUTPUT_EXTERNAL (@var{stream}, @var{name}, @var{decl})
7599: A C statement (sans semicolon) to output to the stdio stream
7600: @var{stream} any text necessary for declaring the name of an external
7601: symbol named @var{name} which is referenced in this compilation but
7602: not defined. The value of @var{decl} is the tree node for the
7603: declaration.
7604:
7605: This macro need not be defined if it does not need to output anything.
7606: The GNU assembler and most Unix assemblers don't require anything.
7607:
7608: @item ASM_OUTPUT_LABELREF (@var{stream}, @var{name})
7609: A C statement to output to the stdio stream @var{stream} a reference in
7610: assembler syntax to a label named @var{name}. The character @samp{_}
7611: should be added to the front of the name, if that is customary on your
7612: operating system, as it is in most Berkeley Unix systems. This macro
7613: is used in @code{assemble_name}.
7614:
7615: @item ASM_GENERATE_INTERNAL_LABEL (@var{string}, @var{prefix}, @var{num})
7616: A C statement to store into the string @var{string} a label whose
7617: name is made from the string @var{prefix} and the number @var{num}.
7618:
7619: This string, when output subsequently by @code{ASM_OUTPUT_LABELREF},
7620: should produce the same output that @code{ASM_OUTPUT_INTERNAL_LABEL}
7621: would produce with the same @var{prefix} and @var{num}.
7622:
7623: @item ASM_OUTPUT_INTERNAL_LABEL (@var{stream}, @var{prefix}, @var{num})
7624: A C statement to output to the stdio stream @var{stream} a label whose
7625: name is made from the string @var{prefix} and the number @var{num}.
7626: These labels are used for internal purposes, and there is no reason
7627: for them to appear in the symbol table of the object file. On many
7628: systems, the letter @samp{L} at the beginning of a label has this
7629: effect. The usual definition of this macro is as follows:
7630:
7631: @example
7632: fprintf (@var{stream}, "L%s%d:\n", @var{prefix}, @var{num})
7633: @end example
7634:
7635: @item ASM_OUTPUT_CASE_LABEL (@var{stream}, @var{prefix}, @var{num}, @var{table})
7636: Define this if the label before a jump-table needs to be output
7637: specially. The first three arguments are the same as for
7638: @code{ASM_OUTPUT_INTERNAL_LABEL}; the fourth argument is the
7639: jump-table which follows (a @samp{jump_insn} containing an
7640: @samp{addr_vec} or @samp{addr_diff_vec}).
7641:
7642: This feature is used on system V to output a @code{swbeg} statement
7643: for the table.
7644:
7645: If this macro is not defined, these labels are output with
7646: @code{ASM_OUTPUT_INTERNAL_LABEL}.
7647:
7648: @item ASM_OUTPUT_CASE_END (@var{stream}, @var{num}, @var{table})
7649: Define this if something special must be output at the end of a jump-table.
7650: The definition should be a C statement to be executed after the assembler
7651: code for the table is written. It should write the appropriate code to
7652: stdio stream @var{stream}. The argument @var{table} is the jump-table
7653: insn, and @var{num} is the label-number of the preceding label.
7654:
7655: If this macro is not defined, nothing special is output at the end of
7656: the jump-table.
7657:
7658: @item ASM_FORMAT_PRIVATE_NAME (@var{outvar}, @var{name}, @var{number})
7659: A C expression to assign to @var{outvar} (which is a variable of type
7660: @code{char *}) a newly allocated string made from the string
7661: @var{name} and the number @var{number}, with some suitable punctuation
7662: added. Use @code{alloca} to get space for the string.
7663:
7664: This string will be used as the argument to @code{ASM_OUTPUT_LABELREF}
7665: to produce an assembler label for an internal static variable whose
7666: name is @var{name}. Therefore, the string must be such as to result
7667: in valid assembler code. The argument @var{number} is different each
7668: time this macro is executed; it prevents conflicts between
7669: similarly-named internal static variables in different scopes.
7670:
7671: Ideally this string should not be a valid C identifier, to prevent any
7672: conflict with the user's own symbols. Most assemblers allow periods
7673: or percent signs in assembler symbols; putting at least one of these
7674: between the name and the number will suffice.
7675:
7676: @item ASM_OUTPUT_REG_PUSH (@var{stream}, @var{regno})
7677: A C expression to output to @var{stream} some assembler code
7678: which will push hard register number @var{regno} onto the stack.
7679: The code need not be optimal, since this macro is used only when
7680: profiling.
7681:
7682: @item ASM_OUTPUT_REG_POP (@var{stream}, @var{regno})
7683: A C expression to output to @var{stream} some assembler code
7684: which will pop hard register number @var{regno} off of the stack.
7685: The code need not be optimal, since this macro is used only when
7686: profiling.
7687:
7688: @item ASM_OUTPUT_ADDR_DIFF_ELT (@var{stream}, @var{value}, @var{rel})
7689: This macro should be provided on machines where the addresses
7690: in a dispatch table are relative to the table's own address.
7691:
7692: The definition should be a C statement to output to the stdio stream
7693: @var{stream} an assembler pseudo-instruction to generate a difference
7694: between two labels. @var{value} and @var{rel} are the numbers of two
7695: internal labels. The definitions of these labels are output using
7696: @code{ASM_OUTPUT_INTERNAL_LABEL}, and they must be printed in the same
7697: way here. For example,
7698:
7699: @example
7700: fprintf (@var{stream}, "\t.word L%d-L%d\n",
7701: @var{value}, @var{rel})
7702: @end example
7703:
7704: @item ASM_OUTPUT_ADDR_VEC_ELT (@var{stream}, @var{value})
7705: This macro should be provided on machines where the addresses
7706: in a dispatch table are absolute.
7707:
7708: The definition should be a C statement to output to the stdio stream
7709: @var{stream} an assembler pseudo-instruction to generate a reference to
7710: a label. @var{value} is the number of an internal label whose
7711: definition is output using @code{ASM_OUTPUT_INTERNAL_LABEL}.
7712: For example,
7713:
7714: @example
7715: fprintf (@var{stream}, "\t.word L%d\n", @var{value})
7716: @end example
7717:
7718: @item ASM_OUTPUT_DOUBLE (@var{stream}, @var{value})
7719: A C statement to output to the stdio stream @var{stream} an assembler
7720: instruction to assemble a @code{double} constant whose value is
7721: @var{value}. @var{value} will be a C expression of type
7722: @code{double}.
7723:
7724: @item ASM_OUTPUT_FLOAT (@var{stream}, @var{value})
7725: A C statement to output to the stdio stream @var{stream} an assembler
7726: instruction to assemble a @code{float} constant whose value is
7727: @var{value}. @var{value} will be a C expression of type @code{float}.
7728:
7729: @item ASM_OUTPUT_INT (@var{stream}, @var{exp})
7730: @itemx ASM_OUTPUT_SHORT (@var{stream}, @var{exp})
7731: @itemx ASM_OUTPUT_CHAR (@var{stream}, @var{exp})
7732: A C statement to output to the stdio stream @var{stream} an assembler
7733: instruction to assemble a @code{int}, @code{short} or @code{char}
7734: constant whose value is @var{value}. The argument @var{exp} will be
7735: an RTL expression which represents a constant value. Use
7736: @samp{output_addr_const (@var{exp})} to output this value as an
7737: assembler expression.@refill
7738:
7739: @item ASM_OUTPUT_BYTE (@var{stream}, @var{value})
7740: A C statement to output to the stdio stream @var{stream} an assembler
7741: instruction to assemble a single byte containing the number @var{value}.
7742:
7743: @item ASM_OUTPUT_ASCII (@var{stream}, @var{ptr}, @var{len})
7744: A C statement to output to the stdio stream @var{stream} an assembler
7745: instruction to assemble a string constant containing the @var{len}
7746: bytes at @var{ptr}. @var{ptr} will be a C expression of type
7747: @code{char *} and @var{len} a C expression of type @code{int}.
7748:
7749: If the assembler has a @code{.ascii} pseudo-op as found in the
7750: Berkeley Unix assembler, do not define the macro
7751: @code{ASM_OUTPUT_ASCII}.
7752:
7753: @item ASM_OUTPUT_SKIP (@var{stream}, @var{nbytes})
7754: A C statement to output to the stdio stream @var{stream} an assembler
7755: instruction to advance the location counter by @var{nbytes} bytes.
7756: @var{nbytes} will be a C expression of type @code{int}.
7757:
7758: @item ASM_OUTPUT_ALIGN (@var{stream}, @var{power})
7759: A C statement to output to the stdio stream @var{stream} an assembler
7760: instruction to advance the location counter to a multiple of 2 to the
7761: @var{power} bytes. @var{power} will be a C expression of type @code{int}.
7762:
7763: @item ASM_OUTPUT_COMMON (@var{stream}, @var{name}, @var{size})
7764: A C statement (sans semicolon) to output to the stdio stream
7765: @var{stream} the assembler definition of a common-label named @var{name}
7766: whose size is @var{size} bytes. Use the expression
7767: @code{assemble_name (@var{stream}, @var{name})} to output the name
7768: itself; before and after that, output the additional assembler syntax
7769: for defining the name, and a newline.
7770:
7771: This macro controls how the assembler definitions of uninitialized
7772: global variables are output.
7773:
7774: @item ASM_OUTPUT_LOCAL (@var{stream}, @var{name}, @var{size})
7775: A C statement (sans semicolon) to output to the stdio stream
7776: @var{stream} the assembler definition of a local-common-label named
7777: @var{name} whose size is @var{size} bytes. Use the expression
7778: @code{assemble_name (@var{stream}, @var{name})} to output the name
7779: itself; before and after that, output the additional assembler syntax
7780: for defining the name, and a newline.
7781:
7782: This macro controls how the assembler definitions of uninitialized
7783: static variables are output.
7784:
7785: @item ASM_OUTPUT_SOURCE_LINE (@var{stream}, @var{line})
7786: A C statment to output DBX or SDB debugging information before code
7787: for line number @var{line} of the current source file to the
7788: stdio stream @var{stream}.
7789:
7790: This macro need not be defined if the standard form of debugging
7791: information for the debugger in use is appropriate.
7792:
7793: @item ASM_OUTPUT_IDENT (@var{stream}, @var{string})
7794: A C statement to output something to the assembler file to handle a
7795: @samp{#ident} directive containing the text @var{string}. If this
7796: macro is not defined, the assembler code @samp{.ident "@var{string}"}
7797: will be output by default.
7798:
7799: This macro is significant only if @code{IDENT_DIRECTIVE} is defined.
7800:
7801: @item TARGET_BELL
7802: A C constant expression for the integer value for escape sequence
7803: @samp{\a}.
7804:
7805: @item TARGET_BS
7806: @itemx TARGET_TAB
7807: @itemx TARGET_NEWLINE
7808: C constant expressions for the integer values for escape sequences
7809: @samp{\b}, @samp{\t} and @samp{\n}.
7810:
7811: @item TARGET_VT
7812: @itemx TARGET_FF
7813: @itemx TARGET_CR
7814: C constant expressions for the integer values for escape sequences
7815: @samp{\v}, @samp{\f} and @samp{\r}.
7816:
7817: @item ASM_OUTPUT_OPCODE (@var{stream}, @var{ptr})
7818: Define this macro if you are using an unusual assembler that
7819: requires different names for the machine instructions.
7820:
7821: The definition is a C statement or statements which output an
7822: assembler instruction opcode to the stdio stream @var{stream}. The
7823: macro-operand @var{ptr} is a variable of type @code{char *} which
7824: points to the opcode name in its ``internal'' form---the form that is
7825: written in the machine description. The definition should output the
7826: opcode name to @var{stream}, performing any translation you desire, and
7827: increment the variable @var{ptr} to point at the end of the opcode
7828: so that it will not be output twice.
7829:
7830: In fact, your macro definition may process less than the entire opcode
7831: name, or more than the opcode name; but if you want to process text
7832: that includes @samp{%}-sequences to substitute operands, you must take
7833: care of the substitution yourself. Just be sure to increment
7834: @var{ptr} over whatever text should not be output normally.
7835:
7836: If the macro definition does nothing, the instruction is output
7837: in the usual way.
7838:
7839: @item FINAL_PRESCAN_INSN (@var{insn}, @var{opvec}, @var{noperands})
7840: If defined, a C statement to be executed just prior to the output of
7841: assembler code for @var{insn}, to modify the extracted operands so
7842: they will be output differently.
7843:
7844: Here the argument @var{opvec} is the vector containing the operands
7845: extracted from @var{insn}, and @var{noperands} is the number of
7846: elements of the vector which contain meaningful data for this insn.
7847: The contents of this vector are what will be used to convert the insn
7848: template into assembler code, so you can change the assembler output
7849: by changing the contents of the vector.
7850:
7851: This macro is useful when various assembler syntaxes share a single
7852: file of instruction patterns; by defining this macro differently, you
7853: can cause a large class of instructions to be output differently (such
7854: as with rearranged operands). Naturally, variations in assembler
7855: syntax affecting individual insn patterns ought to be handled by
7856: writing conditional output routines in those patterns.
7857:
7858: If this macro is not defined, it is equivalent to a null statement.
7859:
7860: @item PRINT_OPERAND (@var{stream}, @var{x}, @var{code})
7861: A C compound statement to output to stdio stream @var{stream} the
7862: assembler syntax for an instruction operand @var{x}. @var{x} is an
7863: RTL expression.
7864:
7865: @var{code} is a value that can be used to specify one of several ways
7866: of printing the operand. It is used when identical operands must be
7867: printed differently depending on the context. @var{code} comes from
7868: the @samp{%} specification that was used to request printing of the
7869: operand. If the specification was just @samp{%@var{digit}} then
7870: @var{code} is 0; if the specification was @samp{%@var{ltr}
7871: @var{digit}} then @var{code} is the ASCII code for @var{ltr}.
7872:
7873: If @var{x} is a register, this macro should print the register's name.
7874: The names can be found in an array @code{reg_names} whose type is
7875: @code{char *[]}. @code{reg_names} is initialized from
7876: @code{REGISTER_NAMES}.
7877:
7878: When the machine description has a specification @samp{%@var{punct}}
7879: (a @samp{%} followed by a punctuation character), this macro is called
7880: with a null pointer for @var{x} and the punctuation character for
7881: @var{code}.
7882:
7883: @item PRINT_OPERAND_ADDRESS (@var{stream}, @var{x})
7884: A C compound statement to output to stdio stream @var{stream} the
7885: assembler syntax for an instruction operand that is a memory reference
7886: whose address is @var{x}. @var{x} is an RTL expression.
7887:
7888: @item ASM_OPEN_PAREN
7889: @itemx ASM_CLOSE_PAREN
7890: These macros are defined as C string constant, describing the syntax
7891: in the assembler for grouping arithmetic expressions. The following
7892: definitions are correct for most assemblers:
7893:
7894: @example
7895: #define ASM_OPEN_PAREN "("
7896: #define ASM_CLOSE_PAREN ")"
7897: @end example
7898: @end table
7899:
7900: @node Config,, Machine Macros, Top
7901: @chapter The Configuration File
7902:
7903: The configuration file @file{config-@var{machine}.h} contains macro
7904: definitions that describe the machine and system on which the compiler is
7905: running. Most of the values in it are actually the same on all machines
7906: that GNU CC runs on, so most all configuration files are identical. But
7907: there are some macros that vary:
7908:
7909: @table @code
7910: @item FAILURE_EXIT_CODE
7911: A C expression for the status code to be returned when the compiler
7912: exits after serious errors.
7913:
7914: @item SUCCESS_EXIT_CODE
7915: A C expression for the status code to be returned when the compiler
7916: exits without serious errors.
7917: @end table
7918:
7919: @contents
7920: @bye
This archive runs on limited infrastructure. Preserving old code on modern bandwidth. Automated agents are requested to crawl responsibly.