![[BACK]](/cvsweb/icons/back.gif){kind=icon}
Return to internals.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 internals.texinfo CVS log | Up to [GCC 1.x] / gcc |
1.1 root 1: \input texinfo @c -*-texinfo-*-
2:
3: @settitle Internals of GNU CC
4: @setfilename internals
5:
6: @ifinfo
7: This file documents the internals of the GNU compiler.
8:
1.1.1.2 root 9: Copyright (C) 1988 Free Software Foundation, Inc.
1.1 root 10:
11: Permission is granted to make and distribute verbatim copies of
12: this manual provided the copyright notice and this permission notice
13: are preserved on all copies.
14:
15: @ignore
16: Permission is granted to process this file through Tex and print the
17: results, provided the printed document carries copying permission
18: notice identical to this one except for the removal of this paragraph
19: (this paragraph not being relevant to the printed manual).
20:
21: @end ignore
22: Permission is granted to copy and distribute modified versions of this
23: manual under the conditions for verbatim copying, provided also that the
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,
1.1.1.2 root 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.
1.1 root 33: @end ifinfo
34:
35: @setchapternewpage odd
36:
37: @titlepage
38: @center @titlefont{Internals of GNU CC}
39: @sp 2
40: @center Richard M. Stallman
1.1.1.2 root 41: @sp 3
1.1.1.4 ! root 42: @center last updated 26 June 1988
1.1.1.2 root 43: @sp 1
1.1.1.4 ! root 44: @center for version 1.23
1.1 root 45: @page
46: @vskip 0pt plus 1filll
1.1.1.2 root 47: Copyright @copyright{} 1988 Free Software Foundation, Inc.
1.1 root 48:
49: Permission is granted to make and distribute verbatim copies of
50: this manual provided the copyright notice and this permission notice
51: are preserved on all copies.
52:
53: Permission is granted to copy and distribute modified versions of this
54: manual under the conditions for verbatim copying, provided also that the
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'' may be
62: included in a translation approved by the author instead of in the original
63: English.
64: @end titlepage
65: @page
66:
67: @ifinfo
1.1.1.2 root 68: @node Top, Copying,, (DIR)
69: @ichapter Introduction
1.1 root 70:
1.1.1.2 root 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.
1.1 root 73:
74: @end ifinfo
75: @menu
76: * Copying:: GNU CC General Public License says
77: how you can copy and share GNU CC.
1.1.1.2 root 78: * Contributors:: People who have contributed to GNU CC.
79: * Options:: Command options supported by @samp{gcc}.
1.1 root 80: * Installation:: How to configure, compile and install GNU CC.
1.1.1.3 root 81: * Trouble:: If you have trouble installing GNU CC.
1.1.1.2 root 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).
1.1 root 85: * Portability:: Goals of GNU CC's portability features.
1.1.1.2 root 86: * Interface:: Function-call interface of GNU CC output.
1.1 root 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:
1.1.1.2 root 93: @node Copying, Contributors, Top, Top
1.1 root 94: @unnumbered GNU CC GENERAL PUBLIC LICENSE
1.1.1.2 root 95: @center (Clarified 11 Feb 1988)
1.1 root 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:
1.1.1.2 root 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: Nobuyuki Hikichi of Software Research Associates, Tokyo, contributed
271: the support for the SONY NEWS machine.
272:
273: @item
274: Charles LaBrec contributed the support for the Integrated Solutions
275: 68020 system.
276:
277: @item
278: Michael Tiemann of MCC wrote the description of the National
279: Semiconductor 32000 series cpu, with some contributions from Jan Stein
280: of the Chalmers Computer Club. Tiemann also wrote the code for inline
1.1.1.4 ! root 281: function integration and for the SPARC cpu.
1.1.1.2 root 282:
283: @item
1.1.1.3 root 284: Robert Brown implemented the support for Encore 32000 systems.
285:
286: @item
1.1.1.4 ! root 287: David Kashtan of SRI adapted GNU CC to the Vomit-Making System.
1.1.1.2 root 288:
289: @item
290: Alex Crain provided changes for the 3b1.
291:
292: @item
1.1.1.4 ! root 293: Greg Satz and Chris Hanson assisted in making GNU CC work on HP-UX for
! 294: the 9000 series 300.
1.1.1.2 root 295: @end itemize
296:
297: @node Options, Installation, Contributors, Top
298: @chapter GNU CC Command Options
299:
300: The GNU C compiler uses a command syntax much like the Unix C compiler.
301: The @code{gcc} program accepts options and file names as operands.
302: Multiple single-letter options may @emph{not} be grouped: @samp{-dr} is
303: very different from @samp{-d -r}.
304:
305: When you invoke GNU CC, it normally does preprocessing, compilation,
306: assembly and linking. File names which end in @samp{.c} are taken as C
307: source to be preprocessed and compiled; compiler output files plus any
308: input files with names ending in @samp{.s} are assembled; then the
309: resulting object files, plus any other input files, are linked together to
310: produce an executable.
311:
312: Command options allow you to stop this process at an intermediate stage.
313: For example, the @samp{-c} option says not to run the linker. Then the
314: output consists of object files output by the assembler.
315:
316: Other command options are passed on to one stage. Some options control
317: the preprocessor and others the compiler itself. Yet other options
318: control the assembler and linker; these are not documented here because the
319: GNU assembler and linker are not yet released.
320:
321: Here are the options to control the overall compilation process, including
322: those that say whether to link, whether to assemble, and so on.
323:
324: @table @samp
325: @item -o @var{file}
326: Place output in file @var{file}. This applies regardless to whatever
327: sort of output is being produced, whether it be an executable file,
328: an object file, an assembler file or preprocessed C code.
329:
1.1.1.3 root 330: If @samp{-o} is not specified, the default is to put an executable file
1.1.1.2 root 331: in @file{a.out}, the object file @file{@var{source}.c} in
332: @file{@var{source}.o}, an assembler file in @file{@var{source}.s}, and
333: preprocessed C on standard output.@refill
334:
335: @item -c
336: Compile or assemble the source files, but do not link. Produce object
337: files with names made by replacing @samp{.c} or @samp{.s} with
338: @samp{.o} at the end of the input file names. Do nothing at all for
339: object files specified as input.
340:
341: @item -S
342: Compile into assembler code but do not assemble. The assembler output
343: file name is made by replacing @samp{.c} with @samp{.s} at the end of
344: the input file name. Do nothing at all for assembler source files or
345: object files specified as input.
346:
347: @item -E
348: Run only the C preprocessor. Preprocess all the C source files
349: specified and output the results to standard output.
350:
351: @item -v
352: Compiler driver program prints the commands it executes as it runs
353: the preprocessor, compiler proper, assembler and linker. Some of
354: these are directed to print their own version numbers.
355:
356: @item -B@var{prefix}
357: Compiler driver program tries @var{prefix} as a prefix for each
358: program it tries to run. These programs are @file{cpp}, @file{cc1},
359: @file{as} and @file{ld}.
360:
361: For each subprogram to be run, the compiler driver first tries the
362: @samp{-B} prefix, if any. If that name is not found, or if @samp{-B}
363: was not specified, the driver tries two standard prefixes, which are
364: @file{/usr/lib/gcc-} and @file{/usr/local/lib/gcc-}. If neither of
365: those results in a file name that is found, the unmodified program
366: name is searched for using the directories specified in your
367: @samp{PATH} environment variable.
368:
369: The run-time support file @file{gnulib} is also searched for using
370: the @samp{-B} prefix, if needed. If it is not found there, the two
371: standard prefixes above are tried, and that is all. The file is left
372: out of the link if it is not found by those means. Most of the time,
373: on most machines, you can do without it.
374: @end table
375:
376: These options control the details of C compilation itself.
377:
378: @table @samp
379: @item -ansi
380: Support all ANSI standard C programs.
381:
382: This turns off certain features of GNU C that are incompatible with
383: ANSI C, such as the @code{asm}, @code{inline} and @code{typeof}
384: keywords, and predefined macros such as @code{unix} and @code{vax}
385: that identify the type of system you are using. It also enables the
386: undesirable and rarely used ANSI trigraph feature.
387:
388: The @samp{-ansi} option does not cause non-ANSI programs to be
389: rejected gratuitously. For that, @samp{-pedantic} is required in
390: addition to @samp{-ansi}.
391:
392: The macro @code{__STRICT_ANSI__} is predefined when the @samp{-ansi}
393: option is used. Some header files may notice this macro and refrain
394: from declaring certain functions or defining certain macros that the
395: ANSI standard doesn't call for; this is to avoid interfering with
396: any programs that might use these names for other things.
397:
398: @item -traditional
399: Attempt to support some aspects of traditional C compilers.
400: Specifically:
401:
402: @itemize @bullet
403: @item
404: All @code{extern} declarations take effect globally even if they
405: are written inside of a function definition. This includes implicit
406: declarations of functions.
407:
408: @item
409: The keywords @code{typeof}, @code{inline}, @code{signed}, @code{const}
410: and @code{volatile} are not recognized.@refill
411:
412: @item
413: Comparisons between pointers and integers are always allowed.
414:
415: @item
416: Integer types @code{unsigned short} and @code{unsigned char} promote
417: to @code{unsigned int}.
418:
419: @item
420: In the preprocessor, comments convert to nothing at all, rather than to
421: a space. This allows traditional token concatenation.
422:
423: @item
424: In the preprocessor, single and double quote characters are ignored
425: when scanning macro definitions, so that macro arguments can be replaced
426: even within a string or character constant. Quote characters are also
427: ignored when skipping text inside a failing conditional directive.
428: @end itemize
429:
430: @item -O
431: Optimize. Optimizing compilation takes somewhat more time, and a lot
432: more memory for a large function.
433:
434: Without @samp{-O}, the compiler's goal is to reduce the cost of
435: compilation and to make debugging produce the expected results.
436: Statements are independent: if you stop the program with a breakpoint
437: between statements, you can then assign a new value to any variable or
438: change the program counter to any other statement in the function and
439: get exactly the results you would expect from the source code.
440:
441: Without @samp{-O}, only variables declared @code{register} are
442: allocated in registers. The resulting compiled code is a little worse
443: than produced by PCC without @samp{-O}.
444:
445: With @samp{-O}, the compiler tries to reduce code size and execution
446: time.
447:
448: Some of the @samp{-f} options described below turn specific kinds of
449: optimization on or off.
450:
451: @item -g
1.1.1.4 ! root 452: Produce debugging information in the operating system's native
! 453: format (for DBX or SDB).
1.1.1.2 root 454:
455: Unlike most other C compilers, GNU CC allows you to use @samp{-g} with
456: @samp{-O}. The shortcuts taken by optimized code may occasionally
457: produce surprising results: some variables you declared may not exist
458: at all; flow of control may briefly move where you did not expect it;
459: some statements may not be executed because they compute constant
460: results or their values were already at hand; some statements may
461: execute in different places because they were moved out of loops.
462: Nevertheless it proves possible to debug optimized output. This makes
463: it reasonable to use the optimizer for programs that might have bugs.
464:
465: @item -gg
466: Produce debugging information in GDB's own format. This requires
467: the GNU assembler and linker in order to work.
468:
469: @item -w
470: Inhibit all warning messages.
471:
472: @item -W
473: Print extra warning messages for these events:
474:
475: @itemize @bullet
476: @item
477: An automatic variable is used without first being initialized.
478:
479: These warnings are possible only in optimizing compilation,
480: because they require data flow information that is computed only
481: when optimizing. They occur only for variables that are
482: candidates for register allocation. Therefore, they do not occur
483: for a variable that is declared @code{volatile}, or whose address
484: is taken, or whose size is other than 1, 2, 4 or 8 bytes. Also,
485: they do not occur for structures, unions or arrays, even when
486: they are in registers.
487:
488: Note that there may be no warning about a variable that is used
489: only to compute a value that itself is never used, because such
490: computations may be deleted by the flow analysis pass before the
491: warnings are printed.
492:
493: These warnings are made optional because GNU CC is not smart
494: enough to see all the reasons why the code might be correct
495: despite appearing to have an error. Here is one example of how
496: this can happen:
497:
498: @example
499: @{
500: int x;
501: switch (y)
502: @{
503: case 1: x = 1;
504: break;
505: case 2: x = 4;
506: break;
507: case 3: x = 5;
508: @}
509: foo (x);
510: @}
511: @end example
512:
513: @noindent
514: If the value of @code{y} is always 1, 2 or 3, then @code{x} is
515: always initialized, but GNU CC doesn't know this. Here is
516: another common case:
517:
518: @example
519: @{
520: int save_y;
521: if (change_y) save_y = y, y = new_y;
522: @dots{}
523: if (change_y) y = save_y;
524: @}
525: @end example
526:
527: @noindent
1.1.1.4 ! root 528: This has no bug because @code{save_y} is used only if it is set.
1.1.1.2 root 529:
530: @item
531: A nonvolatile automatic variable might be changed by a call to
532: @code{longjmp}. These warnings as well are possible only in
533: optimizing compilation.
534:
535: The compiler sees only the calls to @code{setjmp}. It cannot know
536: where @code{longjmp} will be called; in fact, a signal handler could
537: call it at any point in the code. As a result, you may get a warning
538: even when there is in fact no problem because @code{longjmp} cannot
539: in fact be called at the place which would cause a problem.
540:
541: @item
542: A function can return either with or without a value. (Falling
543: off the end of the function body is considered returning without
544: a value.) For example, this function would inspire such a
545: warning:
546:
547: @example
548: foo (a)
549: @{
550: if (a > 0)
551: return a;
552: @}
553: @end example
554:
555: Spurious warnings can occur because GNU CC does not realize that
556: certain functions (including @code{abort} and @code{longjmp})
557: will never return.
558: @end itemize
559:
560: In the future, other useful warnings may also be enabled by this
561: option.
562:
563: @item -Wimplicit
564: Warn whenever a function is implicitly declared.
565:
566: @item -Wreturn-type
567: Warn whenever a function is defined with a return-type that defaults
568: to @code{int}. Also warn about any @code{return} statement with no
569: return-value in a function whose return-type is not @code{void}.
570:
571: @item -Wcomment
572: Warn whenever a comment-start sequence @samp{/*} appears in a comment.
573:
1.1.1.3 root 574: @item -Wall
575: All of the above @samp{-W} options combined.
576:
1.1.1.2 root 577: @item -p
578: Generate extra code to write profile information suitable for the
579: analysis program @code{prof}.
580:
581: @item -pg
582: Generate extra code to write profile information suitable for the
583: analysis program @code{gprof}.
584:
1.1.1.3 root 585: @item -l@var{library}
586: Search a standard list of directories for a library named
587: @var{library}, which is actually a file named
588: @file{lib@var{library}.a}. The linker uses this file as if it
589: had been specified precisely by name.
590:
591: The directories searched include several standard system directories
592: plus any that you specify with @samp{-L}.
593:
594: Normally the files found this way are library files---archive files
595: whose members are object files. The linker handles an archive file by
1.1.1.4 ! root 596: scanning through it for members which define symbols that have so far
! 597: been referenced but not defined. But if the file that is found is an
1.1.1.3 root 598: ordinary object file, it is linked in the usual fashion. The only
1.1.1.4 ! root 599: difference between using an @samp{-l} option and specifying a file name
! 600: is that @samp{-l} searches several directories.
1.1.1.3 root 601:
602: @item -L@var{dir}
603: Add directory @var{dir} to the list of directories to be searched
604: for @samp{-l}.
1.1.1.2 root 605:
606: @item -nostdlib
607: Don't use the standard system libraries and startup files when
608: linking. Only the files you specify (plus @file{gnulib}) will be
609: passed to the linker.
610:
611: @item -m@var{machinespec}
612: Machine-dependent option specifying something about the type of target
613: machine. These options are defined by the macro
614: @code{TARGET_SWITCHES} in the machine description. The default for
615: the options is also defined by that macro, which enables you to change
616: the defaults.@refill
617:
618: These are the @samp{-m} options defined in the 68000 machine
619: description:
620:
621: @table @samp
622: @item -m68020
623: Generate output for a 68020 (rather than a 68000). This is the
624: default if you use the unmodified sources.
625:
626: @item -m68000
627: Generate output for a 68000 (rather than a 68020).
628:
629: @item -m68881
630: Generate output containing 68881 instructions for floating point.
631: This is the default if you use the unmodified sources.
632:
633: @item -msoft-float
634: Generate output containing library calls for floating point.
635:
636: @item -mshort
637: Consider type @code{int} to be 16 bits wide, like @code{short int}.
638:
639: @item -mnobitfield
640: Do not use the bit-field instructions. @samp{-m68000} implies
641: @samp{-mnobitfield}.
642:
643: @item -mbitfield
644: Do use the bit-field instructions. @samp{-m68020} implies
645: @samp{-mbitfield}. This is the default if you use the unmodified
646: sources.
647:
648: @item -mrtd
649: Use a different function-calling convention, in which functions
650: that take a fixed number of arguments return with the @code{rtd}
651: instruction, which pops their arguments while returning. This
652: saves one instruction in the caller since there is no need to pop
653: the arguments there.
654:
655: This calling convention is incompatible with the one normally
656: used on Unix, so you cannot use it if you need to call libraries
657: compiled with the Unix compiler.
658:
659: Also, you must provide function prototypes for all functions that
660: take variable numbers of arguments (including @code{printf});
661: otherwise incorrect code will be generated for calls to those
662: functions.
663:
664: In addition, seriously incorrect code will result if you call a
665: function with too many arguments. (Normally, extra arguments are
666: harmlessly ignored.)
667:
668: The @code{rtd} instruction is supported by the 68010 and 68020
669: processors, but not by the 68000.
670: @end table
671:
672: These @samp{-m} options are defined in the Vax machine description:
673:
674: @table @samp
675: @item -munix
676: Do not output certain jump instructions (@code{aobleq} and so on)
677: that the Unix assembler for the Vax cannot handle across long
678: ranges.
679:
680: @item -mgnu
681: Do output those jump instructions, on the assumption that you
682: will assemble with the GNU assembler.
1.1.1.3 root 683:
684: @item -mg
685: Output code for g-format floating point numbers instead of d-format.
1.1.1.2 root 686: @end table
687:
688: @item -f@var{flag}
689: Specify machine-independent flags. These are the flags:
690:
691: @table @samp
692: @item -ffloat-store
693: Do not store floating-point variables in registers. This
694: prevents undesirable excess precision on machines such as the
695: 68000 where the floating registers (of the 68881) keep more
696: precision than a @code{double} is supposed to have.
697:
698: For most programs, the excess precision does only good, but a few
699: programs rely on the precise definition of IEEE floating point.
700: Use @samp{-ffloat-store} for such programs.
701:
702: @item -fno-asm
703: Do not recognize @code{asm}, @code{inline} or @code{typeof} as a
704: keyword. These words may then be used as identifiers.
705:
706: @item -fno-defer-pop
707: Always pop the arguments to each function call as soon as that
708: function returns. Normally the compiler (when optimizing) lets
709: arguments accumulate on the stack for several function calls and
710: pops them all at once.
711:
712: @item -fcombine-regs
713: Allow the combine pass to combine an instruction that copies one
714: register into another. This might or might not produce better
715: code when used in addition to @samp{-O}. I am interested in
716: hearing about the difference this makes.
717:
718: @item -fforce-mem
719: Force memory operands to be copied into registers before doing
720: arithmetic on them. This may produce better code by making all
721: memory references potential common subexpressions. When they are
722: not common subexpressions, instruction combination should
723: eliminate the separate register-load. I am interested in hearing
724: about the difference this makes.
725:
726: @item -fforce-addr
727: Force memory address constants to be copied into registers before
728: doing arithmetic on them. This may produce better code just as
729: @samp{-fforce-mem} may. I am interested in hearing about the
730: difference this makes.
731:
732: @item -fomit-frame-pointer
733: Don't keep the frame pointer in a register for functions that
734: don't need one. This avoids the instructions to save, set up and
735: restore frame pointers; it also makes an extra register available
736: in many functions. @strong{It also makes debugging impossible.}
737:
738: On some machines, such as the Vax, this flag has no effect,
739: because the standard calling sequence automatically handles the
740: frame pointer and nothing is saved by pretending it doesn't
741: exist. The machine-description macro
742: @code{FRAME_POINTER_REQUIRED} controls whether a target machine
743: supports this flag. @xref{Registers}.@refill
744:
745: @item -finline-functions
746: Integrate all simple functions into their callers. The compiler
747: heuristically decides which functions are simple enough to be
748: worth integrating in this way.
749:
750: If all calls to a given function are integrated, and the function
751: is declared @code{static}, then the function is normally not
752: output as assembler code in its own right.
753:
754: @item -fkeep-inline-functions
755: Even if all calls to a given function are integrated, and the
756: function is declared @code{static}, nevertheless output a
757: separate run-time callable version of the function.
758:
759: @item -fwritable-strings
760: Store string constants in the writable data segment and don't
761: uniquize them. This is for compatibility with old programs which
762: assume they can write into string constants. Writing into string
763: constants is a very bad idea; ``constants'' should be constant.
764:
765: @item -fno-function-cse
766: Do not put function addresses in registers; make each instruction
767: that calls a constant function contain the function's address
768: explicitly.
769:
770: This option results in less efficient code, but some strange
771: hacks that alter the assembler output may be confused by the
772: optimizations performed when this option is not used.
773:
774: @item -fvolatile
775: Consider all memory references through pointers to be volatile.
776:
777: @item -funsigned-char
778: Let the type @code{char} be the unsigned, like @code{unsigned
779: char}.
780:
781: Each kind of machine has a default for what @code{char} should
782: be. It is either like @code{unsigned char} by default or like
783: @code{signed char} by default. (Actually, at present, the
784: default is always signed.)
785:
786: The type @code{char} is always a distinct type from either
787: @code{signed char} or @code{unsigned char}, even though its
788: behavior is always just like one of those two.
789:
790: @item -fsigned-char
791: Let the type @code{char} be signed, like @code{signed char}.
792:
793: @item -ffixed-@var{reg}
794: Treat the register named @var{reg} as a fixed register; generated
795: code should never refer to it (except perhaps as a stack pointer,
796: frame pointer or in some other fixed role).
797:
798: @var{reg} must be the name of a register. The register names
799: accepted are machine-specific and are defined in the
800: @code{REGISTER_NAMES} macro in the machine description macro
801: file.
802:
803: @item -fcall-used-@var{reg}
804: Treat the register named @var{reg} as an allocatable register
805: that is clobbered by function calls. It may be allocated for
806: temporaries or variables that do not live across a call.
807: Functions compiled this way will not save and restore the
808: register @var{reg}.
809:
810: Use of this flag for a register that has a fixed pervasive role
811: in the machine's execution model, such as the stack pointer or
812: frame pointer, will produce disastrous results.
813:
814: @item -fcall-saved-@var{reg}
815: Treat the register named @var{reg} as an allocatable register
816: saved by functions. It may be allocated even for temporaries or
817: variables that live across a call. Functions compiled this way
818: will save and restore the register @var{reg} if they use it.
819:
820: Use of this flag for a register that has a fixed pervasive role
821: in the machine's execution model, such as the stack pointer or
822: frame pointer, will produce disastrous results.
823:
824: A different sort of disaster will result from the use of this
825: flag for a register in which function values are may be returned.
826: @end table
827:
828: @item -d@var{letters}
829: Says to make debugging dumps at times specified by @var{letters}.
830: Here are the possible letters:
831:
832: @table @samp
833: @item r
834: Dump after RTL generation.
835: @item j
836: Dump after first jump optimization.
837: @item J
838: Dump after last jump optimization.
839: @item s
840: Dump after CSE (including the jump optimization that sometimes
841: follows CSE).
842: @item L
843: Dump after loop optimization.
844: @item f
845: Dump after flow analysis.
846: @item c
847: Dump after instruction combination.
848: @item l
849: Dump after local register allocation.
850: @item g
851: Dump after global register allocation.
852: @item m
853: Print statistics on memory usage, at the end of the run.
854: @end table
855:
856: @item -pedantic
857: Issue all the warnings demanded by strict ANSI standard C; reject
858: all programs that use forbidden extensions.
859:
860: Valid ANSI standard C programs should compile properly with or without
861: this option (though a rare few will require @samp{-ansi}). However,
862: without this option, certain GNU extensions and traditional C features
863: are supported as well. With this option, they are rejected. There is
864: no reason to @i{use} this option; it exists only to satisfy pedants.
865: @end table
866:
867: These options control the C preprocessor, which is run on each C source
868: file before actual compilation. If you use the @samp{-E} option, nothing
869: is done except C preprocessing. Some of these options make sense only
870: together with @samp{-E} because they request preprocessor output that is
871: not suitable for actual compilation.
872:
873: @table @samp
874: @item -C
875: Tell the preprocessor not to discard comments. Used with the
876: @samp{-E} option.
877:
878: @item -I@var{dir}
879: Search directory @var{dir} for include files.
880:
1.1.1.3 root 881: @item -I-
882: Any directories specified with @samp{-I} options before the @samp{-I-}
883: option are searched only for the case of @samp{#include "@var{file}"};
884: they are not searched for @samp{#include <@var{file}>}.
885:
886: If additional directories are specified with @samp{-I} options after
887: the @samp{-I-}, these directories are searched for all @samp{#include}
888: directives. (Ordinarily @emph{all} @samp{-I} directories are used
889: this way.)
890:
891: In addition, the @samp{-I-} option inhibits the use of the current
892: directory as the first search directory for @samp{#include
893: "@var{file}"}. Therefore, the current directory is searched only if
894: it is requested explicitly with @samp{-I.}. Specifying both
895: @samp{-I-} and @samp{-I.} allows you to control precisely which
896: directories are searched before the current one and which are searched
897: after.
898:
899: @item -nostdinc
900: Do not search the standard system directories for header files. Only
901: the directories you have specified with @samp{-I} options (and the
902: current directory, if appropriate) are searched.
903:
904: Between @samp{-nostdinc} and @samp{-I-}, you can eliminate all
905: directories from the search path except those you specify.
906:
1.1.1.2 root 907: @item -M
908: Tell the preprocessor to output a rule suitable for @code{make}
909: describing the dependencies of each source file. For each source
910: file, the preprocessor outputs one @code{make}-rule whose target is
911: the object file name for that source file and whose dependencies are
912: all the files @samp{#include}d in it. This rule may be a single line
913: or may be continued with @samp{\}-newline if it is long.
914:
915: @samp{-M} implies @samp{-E}.
916:
917: @item -MM
918: Like @samp{-M} but the output mentions only the user-header files
919: included with @samp{#include "@var{file}"}. System header files
920: included with @samp{#include <@var{file}>} are omitted.
921:
922: @samp{-MM} implies @samp{-E}.
923:
924: @item -D@var{macro}
925: Define macro @var{macro} with the empty string as its definition.
926:
927: @item -D@var{macro}=@var{defn}
928: Define macro @var{macro} as @var{defn}.
929:
930: @item -U@var{macro}
931: Undefine macro @var{macro}.
932:
933: @item -T
934: Support ANSI C trigraphs. You don't want to know about this
935: brain-damage. The @samp{-ansi} option also has this effect.
936: @end table
937:
1.1.1.3 root 938: @node Installation, Trouble, Options, Top
1.1.1.2 root 939: @chapter Installing GNU CC
940:
941: Here is the procedure for installing GNU CC on a Unix system.
942: @menu
943: * VMS Install:: See below for installation on VMS.
944: @end menu
945: @iftex
946: (See below for VMS.)
947: @end iftex
948:
949: @enumerate
950: @item
1.1.1.4 ! root 951: Edit @file{Makefile}. If you are using HPUX, or any form of system V,
! 952: you must make a few changes described in comments at the beginning of
! 953: the file.
! 954:
! 955: @item
! 956: On a Sequent system, go to the Berkeley universe.
1.1.1.2 root 957:
958: @item
959: Choose configuration files.
960:
961: @itemize @bullet
962: @item
963: Make a symbolic link named @file{config.h} to the top-level
964: config file for the machine you are using (@pxref{Config}). This
965: file is responsible for defining information about the host
966: machine. It includes @file{tm.h}.
967:
968: The file's name should be @file{config-@var{machine}.h}. On VMS,
969: use @file{config-vms.h} rather than @file{config-vax.h}. On the
970: HP 9000 series 300, use @file{config-hp9k3.h} rather than
1.1.1.4 ! root 971: @file{config-m68k.h}.
1.1.1.2 root 972:
973: If your system does not support symbolic links, you might want to
974: set up @file{config.h} to contain a @samp{#include} command which
975: refers to the appropriate file.
976:
977: @item
978: Make a symbolic link named @file{tm.h} to the machine-description
979: macro file for your machine (its name should be
980: @file{tm-@var{machine}.h}).
981:
982: For the 68000/68020, do not use @file{tm-m68k.h} directly;
983: instead use one of the files @file{tm-sun3.h}, @file{tm-sun2.h},
984: @file{tm-isi68.h}, @file{tm-news800.h} or @file{tm-3b1.h}. Each
985: of those files includes @file{tm-m68k.h} but sets up a few things
986: differently as appropriate to the specific model of
987: machine.@refill
988:
989: There are two files you can use for a 680x0 running HPUX:
990: @file{tm-hp9k320.h} and @file{tm-hp9k320g.h}. Use the former if
991: you are installing GNU CC alone. The latter is for another option
992: where GNU CC together with the GNU assembler, linker, debugger
993: and other utilities are used to replace all of HPUX that deals
994: with compilation. Not all of the pieces of GNU software needed for
995: this mode of operation are as yet in distribution; full instructions
996: will appear here in the future.@refill
997:
1.1.1.4 ! root 998: For the vax, use @file{tm-vax.h} on Unix, or @file{tm-vms.h} on
! 999: VMS.@refill
! 1000:
! 1001: For the sparc, use @file{tm-sparc.h}.
! 1002:
1.1.1.2 root 1003: For the 32000, use @file{tm-sequent.h} if you are using a Sequent
1.1.1.3 root 1004: machine, or @file{tm-encore.h} for an Encore machine; otherwise,
1.1.1.4 ! root 1005: perhaps @file{tm-ns32k.h} will work for you. If you are trying to use
! 1006: GNU CC on GENIX, you may need to get the version of @code{malloc} from
! 1007: GNU Emacs instead of the system library version, and you probably need
! 1008: to cause the following assembler code to be executed in @file{crt0.o}
! 1009: in order to run the GNU CC output:
1.1.1.2 root 1010:
1.1.1.4 ! root 1011: @example
! 1012: lprd sb,$0
! 1013: sprd mod,r0
! 1014: movqd $0,0(r0)
! 1015: @end example
! 1016:
! 1017: Note that Encore systems are supported only under BSD.
1.1.1.2 root 1018:
1019: @item
1020: Make a symbolic link named @file{md} to the machine description
1021: pattern file (its name should be @file{@var{machine}.md}).
1022:
1023: @item
1024: Make a symbolic link named @file{aux-output.c} to the output
1025: subroutine file for your machine (its name should be
1026: @file{output-@var{machine}.c}).
1027: @end itemize
1028:
1029: @item
1030: Make sure the Bison parser generator is installed. (This is
1.1.1.4 ! root 1031: unnecessary if the Bison output files @file{parse.tab.c} and
! 1032: @file{cexp.c} are more recent than @file{parse.y} and @file{cexp.y}
! 1033: and you do not plan to change the @samp{.y} files.)
1.1.1.2 root 1034:
1035: Note that if you have an old version of Bison you may get an error
1036: from the line with the @samp{%expect} directive. If so, simply remove
1037: that line from @file{parse.y} and proceed.
1038:
1039: @item
1040: If you are using a Sun, make sure the environment variable
1041: @code{FLOAT_OPTION} is not set. If this option were set to
1042: @code{f68881} when @file{gnulib} is compiled, the resulting code would
1043: demand to be linked with a special startup file and will not link
1044: properly without special pains.
1045:
1046: @item
1047: Build the compiler. Just type @samp{make} in the compiler directory.
1048:
1049: @item
1050: Move the first-stage object files and executables into a subdirectory
1051: with this command:
1052:
1053: @example
1054: make stage1
1055: @end example
1056:
1057: The files are moved into a subdirectory named @file{stage1}.
1058: Once installation is complete, you may wish to delete these files
1059: with @code{rm -r stage1}.
1060:
1061: @item
1062: Recompile the compiler with itself, with this command:
1063:
1064: @example
1065: make CC=stage1/gcc CFLAGS="-g -O -Bstage1/"
1066: @end example
1067:
1068: On a 68000 or 68020 system lacking floating point hardware,
1069: unless you have selected a @file{tm.h} file that expects by default
1070: that there is no such hardware, do this instead:
1071:
1072: @example
1073: make CC=stage1/gcc CFLAGS="-g -O -Bstage1/ -msoft-float"
1074: @end example
1075:
1076: @item
1077: If you wish to test the compiler by compiling it with itself one more
1078: time, do this:
1079:
1080: @example
1081: make stage2
1082: make CC=stage2/gcc CFLAGS="-g -O -Bstage2/"
1083: foreach file (*.o)
1084: cmp $file stage2/$file
1085: end
1086: @end example
1087:
1088: This will notify you if any of these stage 3 object files differs from
1089: those of stage 2. Any difference, no matter how innocuous, indicates
1090: that the stage 2 compiler has compiled GNU CC incorrectly, and is
1091: therefore a potentially serious bug which you should investigate and
1092: report (@pxref{Bugs}).
1093:
1094: @item
1095: Install the compiler driver, the compiler's passes and run-time support.
1096: You can use the following command:
1097:
1098: @example
1099: make install
1100: @end example
1101:
1102: @noindent
1103: This copies the files @file{cc1}, @file{cpp} and @file{gnulib} to
1104: files @file{gcc-cc1}, @file{gcc-cpp} and @file{gcc-gnulib} in
1105: directory @file{/usr/local/lib}, which is where the compiler driver
1106: program looks for them. It also copies the driver program @file{gcc}
1107: into the directory @file{/usr/local}, so that it appears in typical
1108: execution search paths.@refill
1109:
1110: @strong{Warning: the GNU CPP may not work for @file{ioctl.h},
1111: @file{ttychars.h} and other system header files unless the
1112: @samp{-traditional} option is used.} The bug is in the header files:
1113: at least on some machines, they rely on behavior that is incompatible
1114: with ANSI C. This behavior consists of substituting for macro
1115: argument names when they appear inside of character constants. The
1116: @samp{-traditional} option tells GNU CC to behave the way these
1117: headers expect.
1118:
1119: Because of this problem, you might prefer to configure GNU CC to use
1120: the system's own C preprocessor. To do so, make the file
1121: @file{/usr/local/lib/gcc-cpp} a link to @file{/lib/cpp}.
1122:
1123: Alternatively, on Sun systems and 4.3BSD at least, you can correct the
1124: include files by running the shell script @file{fixincludes}. This
1125: installs modified, corrected copies of the files @file{ioctl.h} and
1126: @file{ttychars.h} in a special directory where only GNU CC will
1127: normally look for them.
1128:
1129: The file @file{/usr/include/vaxuba/qvioctl.h} used in the X window
1130: system needs a similar correction.
1131: @end enumerate
1132:
1133: If you cannot install the compiler's passes and run-time support in
1134: @file{/usr/local/lib}, you can alternatively use the @samp{-B} option to
1135: specify a prefix by which they may be found. The compiler concatenates
1136: the prefix with the names @file{cpp}, @file{cc1} and @file{gnulib}.
1137: Thus, you can put the files in a directory @file{/usr/foo/gcc} and
1138: specify @samp{-B/usr/foo/gcc/} when you run GNU CC.
1139:
1.1.1.4 ! root 1140: Also, you can specify an alternative default directory for these files
! 1141: by setting the Make variable @code{libdir} when you make GNU CC.
! 1142:
1.1.1.2 root 1143: @node VMS Install,, Installation, Installation
1144: @section Installing GNU CC on VMS
1145:
1.1.1.3 root 1146: The VMS version of GNU CC is distributed in an unusual tape format which
1147: consists of several tape files. The first is a command file; the second is
1148: an executable program which reads Unix tar format; the third is another
1149: command file which uses this program to read the remainder of the tape.
1150:
1151: To load the tape, it suffices to mount it @samp{/foreign} and then do
1152: @samp{@@mta0:} to execute the command file at the beginning of the tape.
1153:
1154: The tape contains executables and object files as well as sources, so no
1155: compilation is necessary unless you change the sources. (This is a good
1156: thing, since you probably don't have any other C compiler.) If you must
1157: recompile, here is how:
1.1.1.2 root 1158:
1159: @enumerate
1160: @item
1161: Copy the file @file{tm-vms.h} to @file{tm.h}, @file{config-vms.h} to
1162: @file{config.h}, @file{vax.md} to @file{md.} and @file{output-vax.c}
1163: to @file{aux-output.c}.@refill
1164:
1165: @item
1166: Type @samp{@@make} to do recompile everything.
1167: @end enumerate
1168:
1.1.1.3 root 1169: To install the @samp{GCC} command so you can use the compiler easily, in
1170: the same manner as you use the VMS C compiler, you must install the VMS CLD
1171: file for GNU CC as follows:
1172:
1173: @enumerate
1174: @item
1175: Define the VMS logical names @samp{GNU_CC} and @samp{GNU_CC_INCLUDE}
1176: to point to the directories where the GNU CC executables
1177: (@samp{gcc-cpp}, @samp{gcc-cc1}, etc.) and the C include files are
1178: kept. This should be done with the commands:@refill
1179:
1180: @example
1181: $ assign /super /system disk:[gcc] gnu_cc
1182: $ assign /super /system disk:[gcc.include] gnu_cc_include
1183: @end example
1184:
1185: @noindent
1186: with the appropriate disk and directory names. These commands can be
1187: placed in your system startup file so they will be executed whenever
1188: the machine is rebooted.
1189:
1190: @item
1191: Install the @samp{GCC} command with the command line:
1192:
1193: @example
1194: $ set command /table=sys$library:dcltables gnu_cc:gcc
1195: @end example
1196:
1197: @noindent
1198: Now you can invoke the compiler with a command like @samp{gcc /verbose
1199: file.c}, which is equivalent to the command @samp{gcc -v -c file.c} in
1200: Unix.
1201: @end enumerate
1202:
1203: @node Trouble, Incompatibilities, Installation, Top
1204: @chapter Trouble in Installation
1205:
1206: Here are some of the things that have caused trouble for people installing
1207: GNU CC.
1208:
1.1.1.4 ! root 1209: @itemize @bullet
1.1.1.3 root 1210: @item
1211: On certain systems, defining certain environment variables such as
1212: @samp{CC} can interfere with the functioning of @code{make}.
1.1.1.4 ! root 1213:
! 1214: @item
! 1215: Cross compilation can run into trouble for certain machines because
! 1216: some target machines' assemblers require floating point numbers to be
! 1217: written as @emph{integer} constants in certain contexts.
! 1218:
! 1219: The compiler writes these integer constants by examining the floating
! 1220: point value as an integer and printing that integer, because this is
! 1221: simple to write and independent of the details of the floating point
! 1222: representation. But this does not work if the compiler is running on
! 1223: a different machine with an incompatible floating point format, or
! 1224: even a different byte-ordering.
! 1225:
! 1226: It is possible to fix this by writing machine-independent code which
! 1227: understands the floating point representation of the target machine.
! 1228: I am not interested in doing that much work to compensate for bugs
! 1229: in assemblers.
1.1.1.3 root 1230: @end itemize
1231:
1232: @node Incompatibilities, Extensions, Trouble, Top
1.1.1.2 root 1233: @chapter Incompatibilities of GNU CC
1234:
1235: There are several noteworthy incompatibilities between GNU C and most
1236: existing (non-ANSI) versions of C.
1237:
1238: Ultimately our intention is that the @samp{-traditional} option will
1239: eliminate most of these incompatibilities by telling GNU C to behave
1240: like the other C compilers.
1241:
1242: @itemize @bullet
1243: @item
1244: GNU CC normally makes string constants read-only. If several
1245: identical-looking string constants are used, GNU CC stores only one
1246: copy of the string.
1247:
1248: One consequence is that you cannot call @code{mktemp} with a string
1249: constant argument. The function @code{mktemp} always alters the
1250: string its argument points to.
1251:
1252: Another consequence is that @code{sscanf} does not work on some
1253: systems when passed a string constant as its format control string.
1254: This is because @code{sscanf} incorrectly tries to write into the
1255: string constant.
1256:
1257: The best solution to these problems is to change the program to use
1258: @code{char}-array variables with initialization strings for these
1259: purposes instead of string constants. But if this is not possible,
1260: you can use the @samp{-fwritable-strings} flag, which directs GNU CC
1261: to handle string constants the same way most C compilers do.
1262:
1263: @item
1264: GNU CC does not substitute macro arguments when they appear inside of
1265: string constants. For example, the following macro in GNU CC
1266:
1267: @example
1268: #define foo(a) "a"
1269: @end example
1270:
1271: @noindent
1272: will produce output @samp{"a"} regardless of what the argument @var{a} is.
1273:
1274: The @samp{-traditional} option directs GNU CC to handle such cases
1275: (among others) in the old-fashioned (non-ANSI) fashion.
1276:
1277: @item
1278: When you use @code{setjmp} and @code{longjmp}, the only automatic
1279: variables guaranteed to remain valid are those declared
1280: @code{volatile}. This is a consequence of automatic register
1281: allocation. Consider this function:
1282:
1283: @example
1284: jmp_buf j;
1285:
1286: foo ()
1287: @{
1288: int a, b;
1289:
1290: a = fun1 ();
1291: if (setjmp (j))
1292: return a;
1293:
1294: a = fun2 ();
1295: /* @r{@code{longjmp (j)} may be occur in @code{fun3}.} */
1296: return a + fun3 ();
1297: @}
1298: @end example
1299:
1300: Here @code{a} may or may not be restored to its first value when the
1301: @code{longjmp} occurs. If @code{a} is allocated in a register, then
1302: its first value is restored; otherwise, it keeps the last value stored
1303: in it.
1304:
1305: If you use the @samp{-W} option with the @samp{-O} option, you will
1306: get a warning when GNU CC thinks such a problem might be possible.
1307:
1308: @item
1309: Declarations of external variables and functions within a block apply
1310: only to the block containing the declaration. In other words, they
1311: have the same scope as any other declaration in the same place.
1312:
1313: In some other C compilers, a @code{extern} declaration affects all the
1314: rest of the file even if it happens within a block.
1315:
1316: The @samp{-traditional} option directs GNU C to treat all @code{extern}
1317: declarations as global, like traditional compilers.
1318:
1319: @item
1320: In traditional C, you can combine @code{long}, etc., with a typedef name,
1321: as shown here:
1322:
1323: @example
1324: typedef int foo;
1325: typedef long foo bar;
1326: @end example
1327:
1328: In ANSI C, this is not allowed: @code{long} and other type modifiers
1329: require an explicit @code{int}. Because this criterion is expressed
1330: by Bison grammar rules rather than C code, the @samp{-traditional}
1331: flag cannot alter it.
1332:
1333: @item
1334: When compiling functions that return structures or unions, GNU CC
1335: output code uses a method different from that used on most versions of
1336: Unix. As a result, code compiled with GNU CC cannot call a
1337: structure-returning function compiled with PCC, and vice versa.
1338:
1339: The method used by GCC is as follows: a structure or union which is 1,
1340: 2, 4 or 8 bytes long is returned like a scalar. A structure or union
1341: with any other size is stored into an address supplied by the caller
1342: in a special, fixed register.
1343:
1344: PCC usually handles all sizes of structures and unions by returning
1345: the address of a block of static storage containing the value. This
1346: method is not used in GCC because it is slower and nonreentrant.
1347:
1348: On systems where PCC works this way, you may be able to make GCC-compiled
1349: code call such functions that were compiled with PCC by declaring them
1350: to return a pointer to the structure or union instead of the structure
1351: or union itself. For example, instead of this:
1352:
1353: @example
1354: struct foo nextfoo ();
1355: @end example
1356:
1357: @noindent
1358: write this:
1359:
1360: @example
1361: struct foo *nextfoo ();
1362: #define nextfoo *nextfoo
1363: @end example
1364:
1365: @noindent
1.1.1.4 ! root 1366: (Note that this assumes you are using the GNU preprocessor and not
! 1367: @samp{-traditional}, so that the ANSI antirecursion rules for macro
! 1368: expansions are effective.)
1.1.1.2 root 1369: @end itemize
1370:
1371: @node Extensions, Bugs, Incompatibilities, Top
1372: @chapter GNU Extensions to the C Language
1373:
1374: GNU C provides several language features not found in ANSI standard C.
1375: (The @samp{-pedantic} option directs GNU CC to print a warning message if
1376: any of these features is used.) To test for the availability of these
1377: features in conditional compilation, check for a predefined macro
1378: @code{__GNUC__}, which is always defined under GNU CC.
1379:
1380: @menu
1381: * Statement Exprs:: Putting statements and declarations inside expressions.
1382: * Naming Types:: Giving a name to the type of some expression.
1383: * Typeof:: @code{typeof}: referring to the type of an expression.
1384: * Lvalues:: Using @samp{?:}, @samp{,} and casts in lvalues.
1385: * Conditionals:: Omitting the middle operand of a @samp{?:} expression.
1386: * Zero-Length:: Zero-length arrays.
1387: * Variable-Length:: Arrays whose length is computed at run time.
1388: * Subscripting:: Any array can be subscripted, even if not an lvalue.
1389: * Pointer Arith:: Arithmetic on @code{void}-pointers and function pointers.
1390: * Constructors:: Constructor expressions give structures, unions
1391: or arrays as values.
1392: * Dollar Signs:: Dollar sign is allowed in identifiers.
1393: * Alignment:: Inquiring about the alignment of a type or variable.
1394: * Inline:: Defining inline functions (as fast as macros).
1395: * Extended Asm:: Assembler instructions with C expressions as operands.
1396: (With them you can define ``built-in'' functions.)
1397: * Asm Labels:: Specifying the assembler name to use for a C symbol.
1398: @end menu
1399:
1400: @node Statement Exprs, Naming Types, Extensions, Extensions
1401: @section Statements and Declarations inside of Expressions
1402:
1403: A compound statement in parentheses may appear inside an expression in GNU
1404: C. This allows you to declare variables within an expression. For
1405: example:
1406:
1407: @example
1408: (@{ int y = foo (); int z;
1409: if (y > 0) z = y;
1410: else z = - y;
1411: z; @})
1412: @end example
1413:
1414: @noindent
1415: is a valid (though slightly more complex than necessary) expression
1416: for the absolute value of @code{foo ()}.
1417:
1418: This feature is especially useful in making macro definitions ``safe'' (so
1419: that they evaluate each operand exactly once). For example, the
1420: ``maximum'' function is commonly defined as a macro in standard C as
1421: follows:
1422:
1423: @example
1424: #define max(a,b) ((a) > (b) ? (a) : (b))
1425: @end example
1426:
1427: @noindent
1428: But this definition computes either @var{a} or @var{b} twice, with bad
1429: results if the operand has side effects. In GNU C, if you know the
1430: type of the operands (here let's assume @code{int}), you can define
1431: the macro safely as follows:
1432:
1433: @example
1434: #define maxint(a,b) \
1435: (@{int _a = (a), _b = (b); _a > _b ? _a : _b; @})
1436: @end example
1437:
1438: Embedded statements are not allowed in constant expressions, such as
1439: the value of an enumeration constant, the width of a bit field, or
1440: the initial value of a static variable.
1441:
1442: If you don't know the type of the operand, you can still do this, but you
1443: must use @code{typeof} (@pxref{Typeof}) or type naming (@pxref{Naming
1444: Types}).
1445:
1446: @node Naming Types, Typeof, Statement Exprs, Extensions
1447: @section Naming an Expression's Type
1448:
1449: You can give a name to the type of an expression using a @code{typedef}
1450: declaration with an initializer. Here is how to define @var{name} as a
1451: type name for the type of @var{exp}:
1452:
1453: @example
1454: typedef @var{name} = @var{exp};
1455: @end example
1456:
1457: This is useful in conjunction with the statements-within-expressions
1458: feature. Here is how the two together can be used to define a safe
1459: ``maximum'' macro that operates on any arithmetic type:
1460:
1461: @example
1462: #define max(a,b) \
1463: (@{typedef _ta = (a), _tb = (b); \
1464: _ta _a = (a); _tb _b = (b); \
1465: _a > _b ? _a : _b; @})
1466: @end example
1467:
1468: The reason for using names that start with underscores for the local
1469: variables is to avoid conflicts with variable names that occur within the
1470: expressions that are substituted for @code{a} and @code{b}. Eventually we
1471: hope to design a new form of declaration syntax that allows you to declare
1472: variables whose scopes start only after their initializers; this will be a
1473: more reliable way to prevent such conflicts.
1474:
1475: @node Typeof, Lvalues, Naming Types, Extensions
1476: @section Referring to a Type with @code{typeof}
1477:
1478: Another way to refer to the type of an expression is with @code{typeof}.
1479: The syntax of using of this keyword looks like @code{sizeof}, but the
1480: construct acts semantically like a type name defined with @code{typedef}.
1481:
1482: There are two ways of writing the argument to @code{typeof}: with an
1483: expression or with a type. Here is an example with an expression:
1484:
1485: @example
1486: typeof (x[0](1))
1487: @end example
1488:
1489: @noindent
1490: This assumes that @code{x} is an array of functions; the type described
1491: is that of the values of the functions.
1492:
1493: Here is an example with a typename as the argument:
1494:
1495: @example
1496: typeof (int *)
1497: @end example
1498:
1499: @noindent
1500: Here the type described is that of pointers to @code{int}.
1501:
1502: A @code{typeof}-construct can be used anywhere a typedef name could be
1503: used. For example, you can use it in a declaration, in a cast, or inside
1504: of @code{sizeof} or @code{typeof}.
1505:
1506: @itemize @bullet
1507: @item
1508: This declares @code{y} with the type of what @code{x} points to.
1509:
1510: @example
1511: typeof (*x) y;
1512: @end example
1513:
1514: @item
1515: This declares @code{y} as an array of such values.
1516:
1517: @example
1518: typeof (*x) y[4];
1519: @end example
1520:
1521: @item
1522: This declares @code{y} as an array of pointers to characters:
1523:
1524: @example
1525: typeof (typeof (char *)[4]) y;
1526: @end example
1527:
1528: @noindent
1529: It is equivalent to the following traditional C declaration:
1530:
1531: @example
1532: char *y[4];
1533: @end example
1534:
1535: To see the meaning of the declaration using @code{typeof}, and why it
1536: might be a useful way to write, let's rewrite it with these macros:
1537:
1538: @example
1539: #define pointer(T) typeof(T *)
1540: #define array(T, N) typeof(T [N])
1541: @end example
1542:
1543: @noindent
1544: Now the declaration can be rewritten this way:
1545:
1546: @example
1547: array (pointer (char), 4) y;
1548: @end example
1549:
1550: @noindent
1551: Thus, @samp{array (pointer (char), 4)} is the type of arrays of 4
1552: pointers to @code{char}.
1553: @end itemize
1554:
1555: @node Lvalues, Conditionals, Typeof, Extensions
1556: @section Generalized Lvalues
1557:
1558: Compound expressions, conditional expressions and casts are allowed as
1559: lvalues provided their operands are lvalues. This means that you can take
1560: their addresses or store values into them.
1561:
1562: For example, a compound expression can be assigned, provided the last
1563: expression in the sequence is an lvalue. These two expressions are
1564: equivalent:
1565:
1566: @example
1567: (a, b) += 5
1568: a, (b += 5)
1569: @end example
1570:
1571: Similarly, the address of the compound expression can be taken. These two
1572: expressions are equivalent:
1573:
1574: @example
1575: &(a, b)
1576: a, &b
1577: @end example
1578:
1579: A conditional expression is a valid lvalue if its type is not void and the
1580: true and false branches are both valid lvalues. For example, these two
1581: expressions are equivalent:
1582:
1583: @example
1584: (a ? b : c) = 5
1585: (a ? b = 5 : (c = 5))
1586: @end example
1587:
1588: A cast is a valid lvalue if its operand is valid. Taking the address of
1589: the cast is the same as taking the address without a cast, except for the
1590: type of the result. For example, these two expressions are equivalent (but
1591: the second may be valid when the type of @samp{a} does not permit a cast to
1592: @samp{int *}).
1593:
1594: @example
1595: &(int *)a
1596: (int **)&a
1597: @end example
1598:
1599: A simple assignment whose left-hand side is a cast works by converting the
1600: right-hand side first to the specified type, then to the type of the inner
1601: left-hand side expression. After this is stored, the value is converter
1602: back to the specified type to become the value of the assignment. Thus, if
1603: @samp{a} has type @samp{char *}, the following two expressions are
1604: equivalent:
1605:
1606: @example
1607: (int)a = 5
1608: (int)(a = (char *)5)
1609: @end example
1610:
1611: An assignment-with-arithmetic operation such as @samp{+=} applied to a cast
1612: performs the arithmetic using the type resulting from the cast, and then
1613: continues as in the previous case. Therefore, these two expressions are
1614: equivalent:
1615:
1616: @example
1617: (int)a += 5
1618: (int)(a = (char *) ((int)a + 5))
1619: @end example
1620:
1621: @node Conditionals, Zero-Length, Lvalues, Extensions
1622: @section Conditional Expressions with Omitted Middle-Operands
1623:
1624: The middle operand in a conditional expression may be omitted. Then
1625: if the first operand is nonzero, its value is the value of the conditional
1626: expression.
1627:
1628: Therefore, the expression
1629:
1630: @example
1631: x ? : y
1632: @end example
1633:
1634: @noindent
1635: has the value of @code{x} if that is nonzero; otherwise, the value of
1636: @code{y}.
1637:
1638: This example is perfectly equivalent to
1639:
1640: @example
1641: x ? x : y
1642: @end example
1643:
1644: @noindent
1645: In this simple case, the ability to omit the middle operand is not
1646: especially useful. When it becomes useful is when the first operand does,
1647: or may (if it is a macro argument), contain a side effect. Then repeating
1648: the operand in the middle would perform the side effect twice. Omitting
1649: the middle operand uses the value already computed without the undesirable
1650: effects of recomputing it.
1651:
1652: @node Zero-Length, Variable-Length, Conditionals, Extensions
1653: @section Arrays of Length Zero
1654:
1655: Zero-length arrays are allowed in GNU C. They are very useful as the last
1656: element of a structure which is really a header for a variable-length
1657: object:
1658:
1659: @example
1660: struct line @{
1661: int length;
1662: char contents[0];
1663: @};
1664:
1665: @{
1666: struct line *thisline
1667: = (struct line *) malloc (sizeof (struct line) + this_length);
1668: thisline->length = thislength;
1669: @}
1670: @end example
1671:
1672: In standard C, you would have to give @code{contents} a length of 1, which
1673: means either you waste space or complicate the argument to @code{malloc}.
1674:
1675: @node Variable-Length, Subscripting, Zero-Length, Extensions
1676: @section Arrays of Variable Length
1677:
1678: Variable-length automatic arrays are allowed in GNU C. These arrays are
1679: declared like any other automatic arrays, but with a length that is not a
1680: constant expression. The storage is allocated at that time and
1681: deallocated when the brace-level is exited. For example:
1682:
1683: @example
1684: FILE *concat_fopen (char *s1, char *s2, char *mode)
1685: @{
1686: char str[strlen (s1) + strlen (s2) + 1];
1687: strcpy (str, s1);
1688: strcat (str, s2);
1689: return fopen (str, mode);
1690: @}
1691: @end example
1692:
1693: You can also define structure types containing variable-length arrays, and
1694: use them even for arguments or function values, as shown here:
1695:
1696: @example
1697: int foo;
1698:
1699: struct entry
1700: @{
1701: char data[foo];
1702: @};
1703:
1704: struct entry
1705: tester (struct entry arg)
1706: @{
1707: struct entry new;
1708: int i;
1709: for (i = 0; i < foo; i++)
1710: new.data[i] = arg.data[i] + 1;
1711: return new;
1712: @}
1713: @end example
1714:
1715: @noindent
1716: (Eventually there will be a way to say that the size of the array is
1717: another member of the same structure.)
1718:
1719: The length of an array is computed on entry to the brace-level where the
1720: array is declared and is remembered for the scope of the array in case you
1721: access it with @code{sizeof}.
1722:
1723: Jumping or breaking out of the scope of the array name will also deallocate
1724: the storage. Jumping into the scope is not allowed; you will get an error
1725: message for it.
1726:
1727: You can use the function @code{alloca} to get an effect much like
1728: variable-length arrays. The function @code{alloca} is available in
1729: many other C implementations (but not in all). On the other hand,
1730: variable-length arrays are more elegant.
1731:
1732: There are other differences between these two methods. Space allocated
1733: with @code{alloca} exists until the containing @emph{function} returns.
1734: The space for a variable-length array is deallocated as soon as the array
1735: name's scope ends. (If you use both variable-length arrays and
1736: @code{alloca} in the same function, deallocation of a variable-length array
1737: will also deallocate anything more recently allocated with @code{alloca}.)
1738:
1739: @node Subscripting, Pointer Arith, Variable-Length, Extensions
1740: @section Non-Lvalue Arrays May Have Subscripts
1741:
1742: Subscripting is allowed on arrays that are not lvalues, even though the
1743: unary @samp{&} operator is not. For example, this is valid in GNU C though
1744: not valid in other C dialects:
1745:
1746: @example
1747: struct foo @{int a[4];@};
1748:
1749: struct foo f();
1750:
1751: bar (int index)
1752: @{
1753: return f().a[index];
1754: @}
1755: @end example
1756:
1757: @node Pointer Arith, Initializers, Subscripting, Extensions
1758: @section Arithmetic on @code{void}-Pointers and Function Pointers
1759:
1760: In GNU C, addition and subtraction operations are supported on pointers to
1761: @code{void} and on pointers to functions. This is done by treating the
1762: size of a @code{void} or of a function as 1.
1763:
1764: A consequence of this is that @code{sizeof} is also allowed on @code{void}
1765: and on function types, and returns 1.
1766:
1767: @node Initializers, Constructors, Pointer Arith, Extensions
1768: @section Non-Constant Initializers
1769:
1770: The elements of an aggregate initializer are not required to be constant
1771: expressions in GNU C. Here is an example of an initializer with run-time
1772: varying elements:
1773:
1774: @example
1775: foo (float f, float g)
1776: @{
1777: float beat_freqs[2] = @{ f-g, f+g @};
1778: @dots{}
1779: @}
1780: @end example
1781:
1782: @node Constructors, Dollar Signs, Initializers, Extensions
1783: @section Constructor Expressions
1784:
1785: GNU C supports constructor expressions. A constructor looks like a cast
1786: containing an initializer. Its value is an object of the type specified in
1787: the cast, containing the elements specified in the initializer. The type
1788: must be a structure, union or array type.
1789:
1790: Assume that @code{struct foo} and @code{structure} are declared as shown:
1791:
1792: @example
1793: struct foo @{int a; char b[2];@} structure;
1794: @end example
1795:
1796: @noindent
1797: Here is an example of constructing a @samp{struct foo} with a constructor:
1798:
1799: @example
1800: structure = ((struct foo) @{x + y, 'a', 0@});
1801: @end example
1802:
1803: @noindent
1804: This is equivalent to writing the following:
1805:
1806: @example
1807: @{
1808: struct foo temp = @{x + y, 'a', 0@};
1809: structure = temp;
1810: @}
1811: @end example
1812:
1813: You can also construct an array. If all the elements of the constructor
1814: are (made up of) simple constant expressions, suitable for use in
1815: initializers, then the constructor is an lvalue and can be coerced to a
1816: pointer to its first element, as shown here:
1817:
1818: @example
1819: char **foo = (char *[]) @{ "x", "y", "z" @};
1820: @end example
1821:
1822: Array constructors whose elements are not simple constants are not very
1823: useful, because the constructor is not an lvalue. There are only two valid
1824: ways to use it: to subscript it, or initialize an array variable with it.
1825: The former is probably slower than a @code{switch} statement, while the
1826: latter does the same thing an ordinary C initializer would do.
1827:
1828: @example
1829: output = ((int[]) @{ 2, x, 28 @}) [input];
1830: @end example
1831:
1832: @node Dollar Signs, Alignment, Constructors, Extensions
1833: @section Dollar Signs in Identifier Names
1834:
1835: In GNU C, you may use dollar signs in identifier names. This is because
1836: many traditional C implementations allow such identifiers.
1837:
1838: @node Alignment, Inline, Dollar Signs, Extensions
1839: @section Inquiring about the Alignment of a Type or Variable
1840:
1841: The keyword @code{__alignof} allows you to inquire about how an object
1842: is aligned, or the minimum alignment usually required by a type. Its
1843: syntax is just like @code{sizeof}.
1844:
1845: For example, if the target machine requires a @code{double} value to be
1846: aligned on an 8-byte boundary, then @code{__alignof (double)} is 8. This
1847: is true on many RISC machines. On more traditional machine designs,
1848: @code{__alignof (double)} is 4 or even 2.
1849:
1850: Some machines never actually require alignment; they allow reference to any
1851: data type even at an odd addresses. For these machines, @code{__alignof}
1852: reports the @emph{recommended} alignment of a type.
1853:
1854: When the operand of @code{__alignof} is an lvalue rather than a type, the
1855: value is the largest alignment that the lvalue is known to have. It may
1856: have this alignment as a result of its data type, or because it is part of
1857: a structure and inherits alignment from that structure. For example, after
1858: this declaration:
1859:
1860: @example
1861: struct foo @{ int x; char y; @} foo1;
1862: @end example
1863:
1864: @noindent
1865: the value of @code{__alignof (foo1.y)} is probably 2 or 4, the same as
1866: @code{__alignof (int)}, even though the data type of @code{foo1.y} does not
1867: itself demand any alignment.@refill
1868:
1869: @node Inline, Extended Asm, Alignment, Extensions
1870: @section An Inline Function is As Fast As a Macro
1871:
1872: By declaring a function @code{inline}, you can direct GNU CC to integrate
1873: that function's code into the code for its callers. This makes execution
1874: faster by eliminating the function-call overhead; in addition, if any of
1875: the actual argument values are constant, their known values may permit
1876: simplifications at compile time so that not all of the inline function's
1877: code needs to be included.
1878:
1879: To declare a function inline, use the @code{inline} keyword in its
1880: declaration, like this:
1881:
1882: @example
1883: inline int
1884: inc (int *a)
1885: @{
1886: (*a)++;
1887: @}
1888: @end example
1889:
1890: You can also make all ``simple enough'' functions inline with the
1891: option @samp{-finline-functions}. Note that certain usages in a
1892: function definition can make it unsuitable for inline substitution.
1893:
1894: When a function is both inline and @code{static}, if all calls to the
1895: function are integrated into the caller, then the function's own assembler
1896: code is never referenced. In this case, GNU CC does not actually output
1897: assembler code for the function, unless you specify the option
1898: @samp{-fkeep-inline-functions}. Some calls cannot be integrated for
1899: various reasons (in particular, calls that precede the function's
1900: definition cannot be integrated, and neither can recursive calls within the
1901: definition). If there is a nonintegrated call, then the function is
1902: compiled to assembler code as usual.
1903:
1904: When an inline function is not @code{static}, then the compiler must assume
1905: that there may be calls from other source files; since a global symbol can
1906: be defined only once in any program, the function must not be defined in
1907: the other source files, so the calls therein cannot be integrated.
1908: Therefore, a non-@code{static} inline function is always compiled on its
1909: own in the usual fashion.
1910:
1911: @node Extended Asm, Asm Labels, Inline, Extensions
1912: @section Assembler Instructions with C Expression Operands
1913:
1914: In an assembler instruction using @code{asm}, you can now specify the
1915: operands of the instruction using C expressions. This means no more
1916: guessing which registers or memory locations will contain the data you want
1917: to use.
1918:
1919: You must specify an assembler instruction template much like what appears
1920: in a machine description, plus an operand constraint string for each
1921: operand.
1922:
1923: For example, here is how to use the 68881's @code{fsinx} instruction:
1924:
1925: @example
1926: asm ("fsinx %1,%0" : "=f" (result) : "f" (angle));
1927: @end example
1928:
1929: @noindent
1930: Here @code{angle} is the C expression for the input operand while
1931: @code{result} is that of the output operand. Each has @samp{"f"} as its
1932: operand constraint, saying that a floating-point register is required. The
1933: constraints use the same language used in the machine description
1934: (@pxref{Constraints}).
1935:
1936: Each operand is described by an operand-constraint string followed by the C
1937: expression in parentheses. A colon separates the assembler template from
1938: the first output operand, and another separates the last output operand
1939: from the first input, if any. Commas separate output operands and separate
1940: inputs. The number of operands is limited to the maximum number of
1941: operands in any instruction pattern in the machine description.
1942:
1943: Output operand expressions must be lvalues, and there must be at least one
1944: of them. The compiler can check this. The input operands need not be
1945: lvalues, and there need not be any. The compiler cannot check whether the
1946: operands have data types that are reasonable for the instruction being
1947: executed.
1948:
1949: The output operands must be write-only; GNU CC will assume that the values
1950: in these operands before the instruction are dead and need not be
1951: generated. For an operand that is read-write, you must logically split its
1952: function into two separate operands, one input operand and one write-only
1953: output operand. The connection between them is expressed by constraints
1954: which say they need to be in the same location when the instruction
1955: executes. You can use the same C expression for both operands, or
1956: different expressions. For example, here we write the (fictitious)
1957: @samp{combine} instruction with @code{bar} as its read-only source operand
1958: and @code{foo} as its read-write destination:
1959:
1960: @example
1961: asm ("combine %2,%0" : "=r" (foo) : "0" (foo), "g" (bar));
1962: @end example
1963:
1964: @noindent
1965: The constraint @samp{"0"} for operand 1 says that it must occupy the same
1966: location as operand 0. Therefore it is not necessary to substitute operand
1967: 1 into the assembler code output.
1968:
1969: Usually the most convenient way to use these @code{asm} instructions is to
1970: encapsulate them in macros that look like functions. For example,
1971:
1972: @example
1973: #define sin(x) \
1974: (@{ double __value, __arg = (x); \
1975: asm ("fsinx %1,%0": "=f" (__value): "f" (__arg)); \
1976: __value; @})
1977: @end example
1978:
1979: @noindent
1980: Here the variable @code{__arg} is used to make sure that the instruction
1981: operates on a proper @code{double} value, and to accept only those
1982: arguments @code{x} which can convert automatically to a @code{double}.
1983:
1984: Another way to make sure the instruction operates on the correct data type
1985: is to use a cast in the @code{asm}. This is different from using a
1986: variable @code{__arg} in that it converts more different types. For
1987: example, if the desired type were @code{int}, casting the argument to
1988: @code{int} would accept a pointer with no complaint, while assigning the
1989: argument to an @code{int} variable named @code{__arg} would warn about
1990: using a pointer unless the caller explicitly casts it.
1991:
1992: GNU CC assumes for optimization purposes that these instructions have no
1993: side effects except to change the output operands. This does not mean that
1994: instructions with a side effect cannot be used, but you must be careful,
1995: because the compiler may eliminate them if the output operands aren't used,
1996: or move them out of loops, or replace two with one if they constitute a
1997: common subexpression. Also, if your instruction does have a side effect on
1998: a variable that otherwise appears not to change, the old value of the
1999: variable may be reused later if it happens to be found in a register.
2000:
2001: You can prevent an @code{asm} instruction from being deleted, moved or
2002: combined by writing the keyword @code{volatile} after the @code{asm}. For
2003: example:
1.1 root 2004:
1.1.1.2 root 2005: @example
2006: #define set_priority(x) \
2007: asm volatile ("set_priority %1": \
2008: "=m" (*(char *)0): "g" (x))
2009: @end example
1.1 root 2010:
1.1.1.2 root 2011: @noindent
2012: Note that we have supplied an output operand which is not actually used in
2013: the instruction. This is because @code{asm} requires at least one output
2014: operand. This requirement exists for internal implementation reasons and
2015: we might be able to relax it in the future.
2016:
2017: In this case output operand has the additional benefit effect of giving the
2018: appearance of writing in memory. As a result, GNU CC will assume that data
2019: previously fetched from memory must be fetched again if needed again later.
2020: This may be desirable if you have not employed the @code{volatile} keyword
2021: on all the variable declarations that ought to have it.
2022:
2023: @node Asm Labels,,Extended Asm, Extensions
2024: @section Controlling Names Used in Assembler Code
2025:
2026: You can specify the name to be used in the assembler code for a C function
2027: or variable by writing the @code{asm} keyword after the declarator as
2028: follows:
1.1 root 2029:
1.1.1.2 root 2030: @example
2031: int foo asm ("myfoo") = 2;
2032: @end example
1.1 root 2033:
1.1.1.2 root 2034: @noindent
2035: This specifies that the name to be used for the variable @code{foo} in
2036: the assembler code should be @samp{myfoo} rather than the usual
2037: @samp{_foo}.
2038:
2039: On systems where an underscore is normally prepended to the name of a C
2040: function or variable, this feature allows you to define names for the
2041: linker that do not start with an underscore.
2042:
2043: You cannot use @code{asm} in this way in a function @emph{definition}; but
2044: you can get the same effect by writing a declaration for the function
2045: before its definition and putting @code{asm} there, like this:
1.1 root 2046:
1.1.1.2 root 2047: @example
2048: extern func () asm ("FUNC");
1.1 root 2049:
1.1.1.2 root 2050: func (x, y)
2051: int x, y;
2052: @dots{}
2053: @end example
1.1 root 2054:
1.1.1.2 root 2055: It is up to you to make sure that the assembler names you choose do not
2056: conflict with any other assembler symbols. Also, you must not use a
2057: register name; that would produce completely invalid assembler code. GNU
2058: CC does not as yet have the ability to store static variables in registers.
2059: Perhaps that will be added.
2060:
2061: @node Bugs, Portability, Extensions, Top
2062: @chapter Reporting Bugs
2063:
2064: Your bug reports play an essential role in making GNU CC reliable.
2065:
2066: Reporting a bug may help you by bringing a solution to your problem, or it
2067: may not. But in any case the important function of a bug report is to help
2068: the entire community by making the next version of GNU CC work better. Bug
2069: reports are your contribution to the maintenance of GNU CC.
1.1 root 2070:
1.1.1.2 root 2071: In order for a bug report to serve its purpose, you must include the
2072: information that makes for fixing the bug.
1.1 root 2073:
1.1.1.2 root 2074: @menu
2075: * Criteria: Bug Criteria. Have you really found a bug?
2076: * Reporting: Bug Reporting. How to report a bug effectively.
2077: @end menu
1.1 root 2078:
1.1.1.2 root 2079: @node Bug Criteria, Bug Reporting, Bugs, Bugs
2080: @section Have You Found a Bug?
2081:
2082: If you are not sure whether you have found a bug, here are some guidelines:
1.1 root 2083:
2084: @itemize @bullet
2085: @item
1.1.1.2 root 2086: If the compiler gets a fatal signal, for any input whatever, that is a
2087: compiler bug. Reliable compilers never crash.
1.1 root 2088:
2089: @item
1.1.1.2 root 2090: If the compiler produces invalid assembly code, for any input whatever
2091: (except an @code{asm} statement), that is a compiler bug, unless the
2092: compiler reports errors (not just warnings) which would ordinarily
2093: prevent the assembler from being run.
1.1 root 2094:
2095: @item
1.1.1.2 root 2096: If the compiler produces valid assembly code that does not correctly
2097: execute the input source code, that is a compiler bug.
1.1 root 2098:
1.1.1.2 root 2099: However, you must double-check to make sure, because you may have run
2100: into an incompatibility between GNU C and traditional C
2101: (@pxref{Incompatibilities}). These incompatibilities might be considered
1.1.1.3 root 2102: bugs, but they are inescapable consequences of valuable features.
1.1 root 2103:
1.1.1.2 root 2104: Or you may have a program whose behavior is undefined, which happened
2105: by chance to give the desired results with another C compiler.
1.1 root 2106:
1.1.1.2 root 2107: For example, in many nonoptimizing compilers, you can write @samp{x;}
2108: at the end of a function instead of @samp{return x;}, with the same
2109: results. But the value of the function is undefined if @samp{return}
2110: is omitted; it is not a bug when GNU CC produces different results.
1.1 root 2111:
1.1.1.2 root 2112: Problems often result from expressions with two increment operators,
2113: as in @samp{f (*p++, *p++)}. Your previous compiler might have
2114: interpreted that expression the way you intended; GNU CC might
2115: interpret it another way; neither compiler is wrong.
1.1 root 2116:
1.1.1.2 root 2117: After you have localized the error to a single source line, it should
2118: be easy to check for these things. If your program is correct and
2119: well defined, you have found a compiler bug.
1.1 root 2120:
1.1.1.2 root 2121: @item
2122: If the compiler produces an error message for valid input, that is a
2123: compiler bug.
1.1 root 2124:
1.1.1.2 root 2125: Note that the following is not valid input, and the error message for
2126: it is not a bug:
1.1 root 2127:
1.1.1.2 root 2128: @example
2129: int foo (char);
1.1 root 2130:
1.1.1.2 root 2131: int
2132: foo (x)
2133: char x;
2134: @{ @dots{} @}
2135: @end example
1.1 root 2136:
1.1.1.2 root 2137: @noindent
2138: The prototype says to pass a @code{char}, while the definition says to
2139: pass an @code{int} and treat the value as a @code{char}. This is what
2140: the ANSI standard says, and it makes sense.
1.1 root 2141:
1.1.1.2 root 2142: @item
2143: If the compiler does not produce an error message for invalid input,
2144: that is a compiler bug. However, you should note that your idea of
2145: ``invalid input'' might be my idea of ``an extension'' or ``support
2146: for traditional practice''.
1.1 root 2147:
1.1.1.2 root 2148: @item
2149: If you are an experienced user of C compilers, your suggestions
2150: for improvement of GNU CC are welcome in any case.
2151: @end itemize
1.1 root 2152:
1.1.1.2 root 2153: @node Bug Reporting,, Bug Criteria, Bugs
2154: @section How to Report Bugs
1.1 root 2155:
1.1.1.2 root 2156: Send bug reports for GNU C to one of these addresses:
1.1 root 2157:
1.1.1.2 root 2158: @example
2159: bug-gcc@@prep.ai.mit.edu
2160: @{ucbvax|mit-eddie|uunet@}!prep.ai.mit.edu!bug-gcc
2161: @end example
1.1 root 2162:
1.1.1.2 root 2163: As a last resort, snail them to:
1.1 root 2164:
1.1.1.2 root 2165: @example
2166: GNU Compiler Bugs
2167: 545 Tech Sq
2168: Cambridge, MA 02139
2169: @end example
1.1 root 2170:
1.1.1.2 root 2171: The fundamental principle of reporting bugs usefully is this:
2172: @strong{report all the facts}. If you are not sure whether to mention a
2173: fact or leave it out, mention it!
2174:
2175: Often people omit facts because they think they know what causes the
2176: problem and they conclude that some details don't matter. Thus, you might
2177: assume that the name of the variable you use in an example does not matter.
2178: Well, probably it doesn't, but one cannot be sure. Perhaps the bug is a
2179: stray memory reference which happens to fetch from the location where that
2180: name is stored in memory; perhaps, if the name were different, the contents
2181: of that location would fool the compiler into doing the right thing despite
2182: the bug. Play it safe and give an exact example.
1.1 root 2183:
1.1.1.2 root 2184: If you want to enable me to fix the bug, you should include all these
2185: things:
1.1 root 2186:
1.1.1.2 root 2187: @itemize @bullet
2188: @item
2189: The version of GNU CC. You can get this by running it with the
2190: @samp{-v} option.
1.1 root 2191:
1.1.1.2 root 2192: Without this, I won't know whether there is any point in looking for
2193: the bug in the current version of GNU CC.
1.1 root 2194:
1.1.1.2 root 2195: @item
2196: A complete input file that will reproduce the bug. If the bug is in
2197: the C preprocessor, send me a source file and any header files that it
2198: requires. If the bug is in the compiler proper (@file{cc1}), run your
2199: source file through the C preprocessor by doing @samp{gcc -E
2200: @var{sourcefile} > @var{outfile}}, then include the contents of
2201: @var{outfile} in the bug report. (Any @samp{-I}, @samp{-D} or
2202: @samp{-U} options that you used in actual compilation should also be
2203: used when doing this.)
1.1 root 2204:
1.1.1.2 root 2205: A single statement is not enough of an example. In order to compile
2206: it, it must be embedded in a function definition; and the bug might
2207: depend on the details of how this is done.
2208:
2209: Without a real example I can compile, all I can do about your bug
2210: report is wish you luck. It would be futile to try to guess how to
2211: provoke the bug. For example, bugs in register allocation and
2212: reloading frequently depend on every little detail of the function
2213: they happen in.
1.1 root 2214:
2215: @item
1.1.1.2 root 2216: The command arguments you gave GNU CC to compile that example and
2217: observe the bug. For example, did you use @samp{-O}? To guarantee
2218: you won't omit something important, list them all.
2219:
2220: If I were to try to guess the arguments, I would probably guess wrong
2221: and then I would not encounter the bug.
1.1 root 2222:
2223: @item
1.1.1.2 root 2224: The names of the files that you used for @file{tm.h} and @file{md}
2225: when you installed the compiler.
1.1 root 2226:
2227: @item
1.1.1.2 root 2228: The type of machine you are using, and the operating system name and
2229: version number.
1.1 root 2230:
2231: @item
1.1.1.2 root 2232: A description of what behavior you observe that you believe is
2233: incorrect. For example, ``It gets a fatal signal,'' or, ``There is an
2234: incorrect assembler instruction in the output.''
2235:
2236: Of course, if the bug is that the compiler gets a fatal signal, then I
2237: will certainly notice it. But if the bug is incorrect output, I might
2238: not notice unless it is glaringly wrong. I won't study all the
2239: assembler code from a 50-line C program just on the off chance that it
2240: might be wrong.
2241:
2242: Even if the problem you experience is a fatal signal, you should still
2243: say so explicitly. Suppose something strange is going on, such as,
2244: your copy of the compiler is out of synch, or you have encountered a
2245: bug in the C library on your system. (This has happened!) Your copy
2246: might crash and mine would not. If you @i{told} me to expect a crash,
2247: then when mine fails to crash, I would know that the bug was not
2248: happening for me. If you had not told me to expect a crash, then I
2249: would not be able to draw any conclusion from my observations.
2250:
2251: In cases where GNU CC generates incorrect code, if you send me a small
2252: complete sample program I will find the error myself by running the
2253: program under a debugger. If you send me a large example or a part of
2254: a larger program, I cannot do this; you must debug the compiled
2255: program and narrow the problem down to one source line. Tell me which
2256: source line it is, and what you believe is incorrect about the code
2257: generated for that line.
1.1 root 2258:
2259: @item
1.1.1.2 root 2260: If you send me examples of output from GNU CC, please use @samp{-g}
2261: when you make them. The debugging information includes source line
2262: numbers which are essential for correlating the output with the input.
1.1.1.4 ! root 2263:
! 2264: @item
! 2265: If you wish to suggest changes to the GNU CC source, send me context
! 2266: diffs. If you even discuss something in the GNU CC source, refer to
! 2267: it by context, not by line number.
! 2268:
! 2269: The line numbers in my development sources don't match those in your
! 2270: sources. They won't tell me anything.
1.1 root 2271: @end itemize
2272:
1.1.1.2 root 2273: Here are some things that are not necessary:
1.1 root 2274:
1.1.1.2 root 2275: @itemize @bullet
1.1 root 2276: @item
1.1.1.2 root 2277: A description of the envelope of the bug.
1.1 root 2278:
1.1.1.2 root 2279: Often people who encounter a bug spend a lot of time investigating
2280: which changes to the input file will make the bug go away and which
2281: changes will not affect it.
1.1 root 2282:
1.1.1.2 root 2283: This is often time consuming and not very useful, because the way I
2284: will find the bug is by running a single example under the debugger
2285: with breakpoints, not by pure deduction from a series of examples.
1.1 root 2286:
1.1.1.2 root 2287: Of course, it can't hurt if you can find a simpler example that
2288: triggers the same bug. Errors in the output will be easier to spot,
2289: running under the debugger will take less time, etc. An easy way
2290: to simplify an example is to delete all the function definitions
2291: except the one where the bug occurs. Those earlier in the file
2292: may be replaced by external declarations.
2293:
2294: However, simplification is not necessary; if you don't want to do
2295: this, report the bug anyway.
1.1 root 2296:
2297: @item
1.1.1.2 root 2298: A patch for the bug.
1.1 root 2299:
1.1.1.2 root 2300: A patch for the bug does help me if it is a good one. But don't omit
2301: the necessary information, such as the test case, because I might see
2302: problems with your patch and decide to fix the problem another way.
1.1 root 2303:
1.1.1.2 root 2304: Sometimes with a program as complicated as GNU CC it is very hard to
2305: construct an example that will make the program go through a certain
2306: point in the code. If you don't send me the example, I won't be able
2307: to verify that the bug is fixed.
1.1 root 2308:
2309: @item
1.1.1.2 root 2310: A guess about what the bug is or what it depends on.
2311:
2312: Such guesses are usually wrong. Even I can't guess right about such
2313: things without using the debugger to find the facts. They also don't
2314: serve a useful purpose.
2315: @end itemize
1.1 root 2316:
1.1.1.2 root 2317: @node Portability, Interface, Bugs, Top
1.1 root 2318: @chapter GNU CC and Portability
2319:
2320: The main goal of GNU CC was to make a good, fast compiler for machines in
2321: the class that the GNU system aims to run on: 32-bit machines that address
2322: 8-bit bytes and have several general registers. Elegance, theoretical
2323: power and simplicity are only secondary.
2324:
2325: GNU CC gets most of the information about the target machine from a machine
2326: description which gives an algebraic formula for each of the machine's
2327: instructions. This is a very clean way to describe the target. But when
2328: the compiler needs information that is difficult to express in this
2329: fashion, I have not hesitated to define an ad-hoc parameter to the machine
2330: description. The purpose of portability is to reduce the total work needed
2331: on the compiler; it was not of interest for its own sake.
2332:
2333: GNU CC does not contain machine dependent code, but it does contain code
2334: that depends on machine parameters such as endianness (whether the most
2335: significant byte has the highest or lowest address of the bytes in a word)
2336: and the availability of autoincrement addressing. In the RTL-generation
2337: pass, it is often necessary to have multiple strategies for generating code
2338: for a particular kind of syntax tree, strategies that are usable for different
2339: combinations of parameters. Often I have not tried to address all possible
2340: cases, but only the common ones or only the ones that I have encountered.
2341: As a result, a new target may require additional strategies. You will know
2342: if this happens because the compiler will call @code{abort}. Fortunately,
1.1.1.2 root 2343: the new strategies can be added in a machine-independent fashion, and will
2344: affect only the target machines that need them.
2345:
2346: @node Interface, Passes, Portability, Top
2347: @chapter Interfacing to GNU CC Output
2348:
2349: GNU CC is normally configured to use the same function calling convention
2350: normally in use on the target system. This is done with the
2351: machine-description macros described (@pxref{Machine Macros}).
2352:
2353: However, returning of structure and union values is done differently.
2354: As a result, functions compiled with PCC returning such types cannot
2355: be called from code compiled with GNU CC, and vice versa. This usually
2356: does not cause trouble because the Unix library routines don't return
2357: structures and unions.
2358:
2359: Structures and unions that are 1, 2, 4 or 8 bytes long are returned in the
2360: same registers used for @code{int} or @code{double} return values. (GNU CC
2361: typically allocates variables of such types in registers also.) Structures
2362: and unions of other sizes are returned by storing them into an address
2363: passed by the caller in a register. This method is faster than the one
2364: normally used by PCC and is also reentrant. The register used for passing
2365: the address is specified by the machine-description macro
2366: @code{STRUCT_VALUE_REGNUM}.
2367:
2368: GNU CC always passes arguments on the stack. At some point it will be
2369: extended to pass arguments in registers, for machines which use that as
2370: the standard calling convention. This will make it possible to use such
2371: a convention on other machines as well. However, that would render it
2372: completely incompatible with PCC. We will probably do this once we
2373: have a complete GNU system so we can compile the libraries with GNU CC.
2374:
2375: If you use @code{longjmp}, beware of automatic variables. ANSI C says that
2376: automatic variables that are not declared @code{volatile} have undefined
2377: values after a @code{longjmp}. And this is all GNU CC promises to do,
2378: because it is very difficult to restore register variables correctly, and
2379: one of GNU CC's features is that it can put variables in registers without
2380: your asking it to.
2381:
2382: If you want a variable to be unaltered by @code{longjmp}, and you don't
2383: want to write @code{volatile} because old C compilers don't accept it,
2384: just take the address of the variable. If a variable's address is ever
2385: taken, even if just to compute it and ignore it, then the variable cannot
2386: go in a register:
2387:
2388: @example
2389: @{
2390: int careful;
2391: &careful;
2392: @dots{}
2393: @}
2394: @end example
1.1 root 2395:
1.1.1.2 root 2396: Code compiled with GNU CC may call certain library routines. The routines
2397: needed on the Vax and 68000 are in the file @file{gnulib.c}. You must
2398: compile this file with the standard C compiler, not with GNU CC, and then
2399: link it with each program you compile with GNU CC. (In actuality, many
2400: programs will not need it.) The usual function call interface is used
2401: for calling the library routines. Some standard parts of the C library,
2402: such as @code{bcopy}, are also called automatically.
2403:
2404: @node Passes, RTL, Interface, Top
1.1 root 2405: @chapter Passes and Files of the Compiler
2406:
2407: The overall control structure of the compiler is in @file{toplev.c}. This
2408: file is responsible for initialization, decoding arguments, opening and
2409: closing files, and sequencing the passes.
2410:
1.1.1.2 root 2411: The parsing pass is invoked only once, to parse the entire input. The RTL
2412: intermediate code for a function is generated as the function is parsed, a
2413: statement at a time. Each statement is read in as a syntax tree and then
2414: converted to RTL; then the storage for the tree for the statement is
2415: reclaimed. Storage for types (and the expressions for their sizes),
2416: declarations, and a representation of the binding contours and how they nest,
2417: remains until the function is finished being compiled; these are all needed
2418: to output the debugging information.
2419:
2420: Each time the parsing pass reads a complete function definition or
2421: top-level declaration, it calls the function
2422: @code{rest_of_compilation} or @code{rest_of_decl_compilation} in
2423: @file{toplev.c}, which are responsible for all further processing
2424: necessary, ending with output of the assembler language. All other
2425: compiler passes run, in sequence, within @code{rest_of_compilation}.
2426: When that function returns from compiling a function definition, the
2427: storage used for that function definition's compilation is entirely
2428: freed, unless it is an inline function (@pxref{Inline}).
1.1 root 2429:
2430: Here is a list of all the passes of the compiler and their source files.
2431: Also included is a description of where debugging dumps can be requested
1.1.1.2 root 2432: with @samp{-d} options.
1.1 root 2433:
2434: @itemize @bullet
2435: @item
2436: Parsing. This pass reads the entire text of a function definition,
1.1.1.2 root 2437: constructing partial syntax trees. This and RTL generation are no longer
2438: truly separate passes (formerly they were), but it is easier to think
2439: of them as separate.
2440:
2441: The tree representation does not entirely follow C syntax, because it is
2442: intended to support other languages as well.
1.1 root 2443:
1.1.1.2 root 2444: C data type analysis is also done in this pass, and every tree node
2445: that represents an expression has a data type attached. Variables are
2446: represented as declaration nodes.
1.1 root 2447:
1.1.1.2 root 2448: Constant folding and associative-law simplifications are also done
2449: during this pass.
1.1 root 2450:
1.1.1.2 root 2451: The source files for parsing are @file{parse.y}, @file{decl.c},
1.1 root 2452: @file{typecheck.c}, @file{stor-layout.c}, @file{fold-const.c}, and
2453: @file{tree.c}. The last three are intended to be language-independent.
2454: There are also header files @file{parse.h}, @file{c-tree.h},
2455: @file{tree.h} and @file{tree.def}. The last two define the format of
1.1.1.2 root 2456: the tree representation.@refill
1.1 root 2457:
2458: @item
1.1.1.2 root 2459: RTL generation. This is the conversion of syntax tree into RTL code.
2460: It is actually done statement-by-statement during parsing, but for
2461: most purposes it can be thought of as a separate pass.
1.1 root 2462:
2463: This is where the bulk of target-parameter-dependent code is found,
2464: since often it is necessary for strategies to apply only when certain
2465: standard kinds of instructions are available. The purpose of named
2466: instruction patterns is to provide this information to the RTL
2467: generation pass.
2468:
2469: Optimization is done in this pass for @code{if}-conditions that are
2470: comparisons, boolean operations or conditional expressions. Tail
2471: recursion is detected at this time also. Decisions are made about how
2472: best to arrange loops and how to output @code{switch} statements.
2473:
1.1.1.2 root 2474: The source files for RTL generation are @file{stmt.c}, @file{expr.c},
1.1 root 2475: @file{explow.c}, @file{expmed.c}, @file{optabs.c} and @file{emit-rtl.c}.
2476: Also, the file @file{insn-emit.c}, generated from the machine description
2477: by the program @code{genemit}, is used in this pass. The header files
1.1.1.2 root 2478: @file{expr.h} is used for communication within this pass.@refill
1.1 root 2479:
1.1.1.2 root 2480: The header files @file{insn-flags.h} and @file{insn-codes.h},
2481: generated from the machine description by the programs @code{genflags}
2482: and @code{gencodes}, tell this pass which standard names are available
2483: for use and which patterns correspond to them.@refill
1.1 root 2484:
2485: Aside from debugging information output, none of the following passes
1.1.1.2 root 2486: refers to the tree structure representation of the function (only
2487: part of which is saved).
1.1 root 2488:
1.1.1.2 root 2489: The decision of whether the function can and should be expanded inline
2490: in its subsequent callers is made at the end of rtl generation. The
2491: function must meet certain criteria, currently related to the size of
2492: the function and the types and number of parameters it has. Note that
2493: this function may contain loops, recursive calls to itself
2494: (tail-recursive functions can be inlined!), gotos, in short, all
2495: constructs supported by GNU CC.
2496:
2497: The option @samp{-dr} causes a debugging dump of the RTL code after
2498: this pass. This dump file's name is made by appending @samp{.rtl} to
2499: the input file name.
1.1 root 2500:
2501: @item
1.1.1.2 root 2502: Jump optimization. This pass simplifies jumps to the following
2503: instruction, jumps across jumps, and jumps to jumps. It deletes
2504: unreferenced labels and unreachable code, except that unreachable code
2505: that contains a loop is not recognized as unreachable in this pass.
2506: (Such loops are deleted later in the basic block analysis.)
1.1 root 2507:
2508: Jump optimization is performed two or three times. The first time is
1.1.1.2 root 2509: immediately following RTL generation. The second time is after CSE,
2510: but only if CSE says repeated jump optimization is needed. The
2511: last time is right before the final pass. That time, cross-jumping
2512: and deletion of no-op move instructions are done together with the
2513: optimizations described above.
1.1 root 2514:
2515: The source file of this pass is @file{jump.c}.
2516:
1.1.1.2 root 2517: The option @samp{-dj} causes a debugging dump of the RTL code after
2518: this pass is run for the first time. This dump file's name is made by
2519: appending @samp{.jump} to the input file name.
1.1 root 2520:
2521: @item
2522: Register scan. This pass finds the first and last use of each
2523: register, as a guide for common subexpression elimination. Its source
2524: is in @file{regclass.c}.
2525:
2526: @item
2527: Common subexpression elimination. This pass also does constant
2528: propagation. Its source file is @file{cse.c}. If constant
2529: propagation causes conditional jumps to become unconditional or to
1.1.1.2 root 2530: become no-ops, jump optimization is run again when CSE is finished.
1.1 root 2531:
1.1.1.2 root 2532: The option @samp{-ds} causes a debugging dump of the RTL code after
1.1 root 2533: this pass. This dump file's name is made by appending @samp{.cse} to
2534: the input file name.
2535:
2536: @item
2537: Loop optimization. This pass moves constant expressions out of loops.
2538: Its source file is @file{loop.c}.
2539:
1.1.1.2 root 2540: The option @samp{-dL} causes a debugging dump of the RTL code after
1.1 root 2541: this pass. This dump file's name is made by appending @samp{.loop} to
2542: the input file name.
2543:
2544: @item
2545: Stupid register allocation is performed at this point in a
2546: nonoptimizing compilation. It does a little data flow analysis as
2547: well. When stupid register allocation is in use, the next pass
2548: executed is the reloading pass; the others in between are skipped.
1.1.1.2 root 2549: The source file is @file{stupid.c}.
1.1 root 2550:
2551: @item
2552: Data flow analysis (@file{flow.c}). This pass divides the program
2553: into basic blocks (and in the process deletes unreachable loops); then
2554: it computes which pseudo-registers are live at each point in the
2555: program, and makes the first instruction that uses a value point at
2556: the instruction that computed the value.
2557:
2558: This pass also deletes computations whose results are never used, and
2559: combines memory references with add or subtract instructions to make
2560: autoincrement or autodecrement addressing.
2561:
1.1.1.2 root 2562: The option @samp{-df} causes a debugging dump of the RTL code after
1.1 root 2563: this pass. This dump file's name is made by appending @samp{.flow} to
2564: the input file name. If stupid register allocation is in use, this
2565: dump file reflects the full results of such allocation.
2566:
2567: @item
2568: Instruction combination (@file{combine.c}). This pass attempts to
2569: combine groups of two or three instructions that are related by data
2570: flow into single instructions. It combines the RTL expressions for
2571: the instructions by substitution, simplifies the result using algebra,
2572: and then attempts to match the result against the machine description.
2573:
1.1.1.2 root 2574: The option @samp{-dc} causes a debugging dump of the RTL code after
1.1 root 2575: this pass. This dump file's name is made by appending @samp{.combine}
2576: to the input file name.
2577:
2578: @item
2579: Register class preferencing. The RTL code is scanned to find out
1.1.1.2 root 2580: which register class is best for each pseudo register. The source
2581: file is @file{regclass.c}.
1.1 root 2582:
2583: @item
2584: Local register allocation (@file{local-alloc.c}). This pass allocates
2585: hard registers to pseudo registers that are used only within one basic
1.1.1.2 root 2586: block. Because the basic block is linear, it can use fast and
2587: powerful techniques to do a very good job.
1.1 root 2588:
1.1.1.2 root 2589: The option @samp{-dl} causes a debugging dump of the RTL code after
1.1 root 2590: this pass. This dump file's name is made by appending @samp{.lreg} to
2591: the input file name.
2592:
2593: @item
2594: Global register allocation (@file{global-alloc.c}). This pass
2595: allocates hard registers for the remaining pseudo registers (those
2596: whose life spans are not contained in one basic block).
2597:
2598: @item
1.1.1.2 root 2599: Reloading. This pass renumbers pseudo registers with the hardware
2600: registers numbers they were allocated. Pseudo registers that did not
2601: get hard registers are replaced with stack slots. Then it finds
2602: instructions that are invalid because a value has failed to end up in
2603: a register, or has ended up in a register of the wrong kind. It fixes
2604: up these instructions by reloading the problematical values
2605: temporarily into registers. Additional instructions are generated to
2606: do the copying.
1.1 root 2607:
2608: Source files are @file{reload.c} and @file{reload1.c}, plus the header
2609: @file{reload.h} used for communication between them.
2610:
1.1.1.2 root 2611: The option @samp{-dg} causes a debugging dump of the RTL code after
1.1 root 2612: this pass. This dump file's name is made by appending @samp{.greg} to
2613: the input file name.
2614:
2615: @item
1.1.1.2 root 2616: Jump optimization is repeated, this time including cross-jumping
2617: and deletion of no-op move instructions. Machine-specific peephole
2618: optimizations are performed at the same time.
2619:
2620: The option @samp{-dJ} causes a debugging dump of the RTL code after
2621: this pass. This dump file's name is made by appending @samp{.jump2}
2622: to the input file name.
1.1 root 2623:
2624: @item
2625: Final. This pass outputs the assembler code for the function. It is
1.1.1.2 root 2626: also responsible for identifying spurious test and compare
2627: instructions. The function entry and exit sequences are generated
2628: directly as assembler code in this pass; they never exist as RTL.
1.1 root 2629:
2630: The source files are @file{final.c} plus @file{insn-output.c}; the
2631: latter is generated automatically from the machine description by the
2632: tool @file{genoutput}. The header file @file{conditions.h} is used
2633: for communication between these files.
2634:
2635: @item
2636: Debugging information output. This is run after final because it must
2637: output the stack slot offsets for pseudo registers that did not get
2638: hard registers. Source files are @file{dbxout.c} for DBX symbol table
2639: format and @file{symout.c} for GDB's own symbol table format.
2640: @end itemize
2641:
2642: Some additional files are used by all or many passes:
2643:
2644: @itemize @bullet
2645: @item
2646: Every pass uses @file{machmode.def}, which defines the machine modes.
2647:
2648: @item
2649: All the passes that work with RTL use the header files @file{rtl.h}
1.1.1.2 root 2650: and @file{rtl.def}, and subroutines in file @file{rtl.c}. The tools
2651: @code{gen*} also use these files to read and work with the machine
2652: description RTL.
1.1 root 2653:
2654: @item
2655: Several passes refer to the header file @file{insn-config.h} which
2656: contains a few parameters (C macro definitions) generated
2657: automatically from the machine description RTL by the tool
2658: @code{genconfig}.
2659:
2660: @item
2661: Several passes use the instruction recognizer, which consists of
2662: @file{recog.c} and @file{recog.h}, plus the files @file{insn-recog.c}
2663: and @file{insn-extract.c} that are generated automatically from the
1.1.1.2 root 2664: machine description by the tools @file{genrecog} and
2665: @file{genextract}.@refill
1.1 root 2666:
2667: @item
1.1.1.2 root 2668: Several passes use the header files @file{regs.h} which defines the
2669: information recorded about pseudo register usage, and @file{basic-block.h}
1.1 root 2670: which defines the information recorded about basic blocks.
2671:
2672: @item
2673: @file{hard-reg-set.h} defines the type @code{HARD_REG_SET}, a bit-vector
2674: with a bit for each hard register, and some macros to manipulate it.
2675: This type is just @code{int} if the machine has few enough hard registers;
2676: otherwise it is an array of @code{int} and some of the macros expand
2677: into loops.
2678: @end itemize
2679:
2680: @node RTL, Machine Desc, Passes, Top
2681: @chapter RTL Representation
2682:
2683: Most of the work of the compiler is done on an intermediate representation
1.1.1.2 root 2684: called register transfer language. In this language, the instructions to be
1.1 root 2685: output are described, pretty much one by one, in an algebraic form that
2686: describes what the instruction does.
2687:
2688: RTL is inspired by Lisp lists. It has both an internal form, made up of
2689: structures that point at other structures, and a textual form that is used
2690: in the machine description and in printed debugging dumps. The textual
2691: form uses nested parentheses to indicate the pointers in the internal form.
2692:
2693: @menu
2694: * RTL Objects:: Expressions vs vectors vs strings vs integers.
2695: * Accessors:: Macros to access expression operands or vector elts.
1.1.1.2 root 2696: * Flags:: Other flags in an RTL expression.
1.1 root 2697: * Machine Modes:: Describing the size and format of a datum.
2698: * Constants:: Expressions with constant values.
2699: * Regs and Memory:: Expressions representing register contents or memory.
2700: * Arithmetic:: Expressions representing arithmetic on other expressions.
2701: * Comparisons:: Expressions representing comparison of expressions.
2702: * Bit Fields:: Expressions representing bit-fields in memory or reg.
2703: * Conversions:: Extending, truncating, floating or fixing.
2704: * RTL Declarations:: Declaring volatility, constancy, etc.
2705: * Side Effects:: Expressions for storing in registers, etc.
2706: * Incdec:: Embedded side-effects for autoincrement addressing.
1.1.1.2 root 2707: * Assembler:: Representing @code{asm} with operands.
1.1 root 2708: * Insns:: Expression types for entire insns.
1.1.1.2 root 2709: * Calls:: RTL representation of function call insns.
1.1 root 2710: * Sharing:: Some expressions are unique; others *must* be copied.
2711: @end menu
2712:
2713: @node RTL Objects, Accessors, RTL, RTL
2714: @section RTL Object Types
2715:
2716: RTL uses four kinds of objects: expressions, integers, strings and vectors.
1.1.1.2 root 2717: Expressions are the most important ones. An RTL expression (``RTX'', for
2718: short) is a C structure, but it is usually referred to with a pointer; a
2719: type that is given the typedef name @code{rtx}.
1.1 root 2720:
2721: An integer is simply an @code{int}, and a string is a @code{char *}.
1.1.1.2 root 2722: Within RTL code, strings appear only inside @samp{symbol_ref} expressions,
2723: but they appear in other contexts in the RTL expressions that make up
1.1 root 2724: machine descriptions. Their written form uses decimal digits.
2725:
2726: A string is a sequence of characters. In core it is represented as a
1.1.1.2 root 2727: @code{char *} in usual C fashion, and it is written in C syntax as well.
1.1 root 2728: However, strings in RTL may never be null. If you write an empty string in
2729: a machine description, it is represented in core as a null pointer rather
2730: than as a pointer to a null character. In certain contexts, these null
2731: pointers instead of strings are valid.
2732:
2733: A vector contains an arbitrary, specified number of pointers to
2734: expressions. The number of elements in the vector is explicitly present in
2735: the vector. The written form of a vector consists of square brackets
2736: (@samp{[@dots{}]}) surrounding the elements, in sequence and with
2737: whitespace separating them. Vectors of length zero are not created; null
2738: pointers are used instead.
2739:
1.1.1.2 root 2740: Expressions are classified by @dfn{expression codes} (also called RTX
2741: codes). The expression code is a name defined in @file{rtl.def}, which is
2742: also (in upper case) a C enumeration constant. The possible expression
2743: codes and their meanings are machine-independent. The code of an RTX can
2744: be extracted with the macro @code{GET_CODE (@var{x})} and altered with
2745: @code{PUT_CODE (@var{x}, @var{newcode})}.
1.1 root 2746:
2747: The expression code determines how many operands the expression contains,
2748: and what kinds of objects they are. In RTL, unlike Lisp, you cannot tell
2749: by looking at an operand what kind of object it is. Instead, you must know
2750: from its context---from the expression code of the containing expression.
1.1.1.2 root 2751: For example, in an expression of code @samp{subreg}, the first operand is
1.1 root 2752: to be regarded as an expression and the second operand as an integer. In
1.1.1.2 root 2753: an expression of code @samp{plus}, there are two operands, both of which
2754: are to be regarded as expressions. In a @samp{symbol_ref} expression,
1.1 root 2755: there is one operand, which is to be regarded as a string.
2756:
2757: Expressions are written as parentheses containing the name of the
2758: expression type, its flags and machine mode if any, and then the operands
2759: of the expression (separated by spaces).
2760:
1.1.1.2 root 2761: Expression code names in the @samp{md} file are written in lower case,
2762: but when they appear in C code they are written in upper case. In this
2763: manual, they are shown as follows: @samp{const_int}.
2764:
1.1 root 2765: In a few contexts a null pointer is valid where an expression is normally
2766: wanted. The written form of this is @samp{(nil)}.
2767:
1.1.1.2 root 2768: @node Accessors, Flags, RTL Objects, RTL
1.1 root 2769: @section Access to Operands
2770:
2771: For each expression type @file{rtl.def} specifies the number of contained
2772: objects and their kinds, with four possibilities: @samp{e} for expression
2773: (actually a pointer to an expression), @samp{i} for integer, @samp{s} for
2774: string, and @samp{E} for vector of expressions. The sequence of letters
2775: for an expression code is called its @dfn{format}. Thus, the format of
1.1.1.2 root 2776: @samp{subreg} is @samp{ei}.@refill
1.1 root 2777:
2778: Two other format characters are used occasionally: @samp{u} and @samp{0}.
2779: @samp{u} is equivalent to @samp{e} except that it is printed differently in
2780: debugging dumps, and @samp{0} means a slot whose contents do not fit any
2781: normal category. @samp{0} slots are not printed at all in dumps, and are
1.1.1.2 root 2782: often used in special ways by small parts of the compiler.@refill
1.1 root 2783:
2784: There are macros to get the number of operands and the format of an
2785: expression code:
2786:
2787: @table @code
2788: @item GET_RTX_LENGTH (@var{code})
1.1.1.2 root 2789: Number of operands of an RTX of code @var{code}.
1.1 root 2790:
2791: @item GET_RTX_FORMAT (@var{code})
1.1.1.2 root 2792: The format of an RTX of code @var{code}, as a C string.
1.1 root 2793: @end table
2794:
2795: Operands of expressions are accessed using the macros @code{XEXP},
2796: @code{XINT} and @code{XSTR}. Each of these macros takes two arguments: an
1.1.1.2 root 2797: expression-pointer (RTX) and an operand number (counting from zero).
2798: Thus,@refill
1.1 root 2799:
2800: @example
1.1.1.2 root 2801: XEXP (@var{x}, 2)
1.1 root 2802: @end example
2803:
2804: @noindent
2805: accesses operand 2 of expression @var{x}, as an expression.
2806:
2807: @example
1.1.1.2 root 2808: XINT (@var{x}, 2)
1.1 root 2809: @end example
2810:
2811: @noindent
2812: accesses the same operand as an integer. @code{XSTR}, used in the same
2813: fashion, would access it as a string.
2814:
2815: Any operand can be accessed as an integer, as an expression or as a string.
2816: You must choose the correct method of access for the kind of value actually
2817: stored in the operand. You would do this based on the expression code of
2818: the containing expression. That is also how you would know how many
2819: operands there are.
2820:
2821: For example, if @var{x} is a @samp{subreg} expression, you know that it has
1.1.1.2 root 2822: two operands which can be correctly accessed as @code{XEXP (@var{x}, 0)}
2823: and @code{XINT (@var{x}, 1)}. If you did @code{XINT (@var{x}, 0)}, you
2824: would get the address of the expression operand but cast as an integer;
2825: that might occasionally be useful, but it would be cleaner to write
2826: @code{(int) XEXP (@var{x}, 0)}. @code{XEXP (@var{x}, 1)} would also
2827: compile without error, and would return the second, integer operand cast as
2828: an expression pointer, which would probably result in a crash when
2829: accessed. Nothing stops you from writing @code{XEXP (@var{x}, 28)} either,
2830: but this will access memory past the end of the expression with
2831: unpredictable results.@refill
1.1 root 2832:
2833: Access to operands which are vectors is more complicated. You can use the
2834: macro @code{XVEC} to get the vector-pointer itself, or the macros
2835: @code{XVECEXP} and @code{XVECLEN} to access the elements and length of a
2836: vector.
2837:
2838: @table @code
2839: @item XVEC (@var{exp}, @var{idx})
2840: Access the vector-pointer which is operand number @var{idx} in @var{exp}.
2841:
2842: @item XVECLEN (@var{exp}, @var{idx})
2843: Access the length (number of elements) in the vector which is
2844: in operand number @var{idx} in @var{exp}. This value is an @code{int}.
2845:
1.1.1.2 root 2846: @item XVECEXP (@var{exp}, @var{idx}, @var{eltnum})
1.1 root 2847: Access element number @var{eltnum} in the vector which is
1.1.1.2 root 2848: in operand number @var{idx} in @var{exp}. This value is an RTX.
1.1 root 2849:
2850: It is up to you to make sure that @var{eltnum} is not negative
2851: and is less than @code{XVECLEN (@var{exp}, @var{idx})}.
2852: @end table
2853:
2854: All the macros defined in this section expand into lvalues and therefore
2855: can be used to assign the operands, lengths and vector elements as well as
2856: to access them.
2857:
1.1.1.2 root 2858: @node Flags, Machine Modes, Accessors, RTL
2859: @section Flags in an RTL Expression
2860:
2861: RTL expressions contain several flags (one-bit bit-fields) that are used
2862: in certain types of expression.
2863:
2864: @table @code
2865: @item used
2866: This flag is used only momentarily, at the end of RTL generation for a
2867: function, to count the number of times an expression appears in insns.
2868: Expressions that appear more than once are copied, according to the
2869: rules for shared structure (@pxref{Sharing}).
2870:
2871: @item volatil
2872: This flag is used in @samp{mem} and @samp{reg} expressions and in insns.
2873: In RTL dump files, it is printed as @samp{/v}.
2874:
2875: In a @samp{mem} expression, it is 1 if the memory reference is volatile.
2876: Volatile memory references may not be deleted, reordered or combined.
2877:
2878: In a @samp{reg} expression, it is 1 if the value is a user-level variable.
2879: 0 indicates an internal compiler temporary.
2880:
2881: In an insn, 1 means the insn has been deleted.
2882:
2883: @item in_struct
2884: This flag is used in @samp{mem} expressions. It is 1 if the memory
2885: datum referred to is all or part of a structure or array; 0 if it is (or
2886: might be) a scalar variable. A reference through a C pointer has 0
2887: because the pointer might point to a scalar variable.
2888:
2889: This information allows the compiler to determine something about possible
2890: cases of aliasing.
2891:
2892: In an RTL dump, this flag is represented as @samp{/s}.
2893:
2894: @item unchanging
2895: This flag is used in @samp{reg} and @samp{mem} expressions. 1 means
2896: that the value of the expression never changes (at least within the
2897: current function).
2898:
2899: In an RTL dump, this flag is represented as @samp{/u}.
2900: @end table
2901:
2902: @node Machine Modes, Constants, Flags, RTL
1.1 root 2903: @section Machine Modes
2904:
2905: A machine mode describes a size of data object and the representation used
2906: for it. In the C code, machine modes are represented by an enumeration
1.1.1.2 root 2907: type, @code{enum machine_mode}, defined in @file{machmode.def}. Each RTL
2908: expression has room for a machine mode and so do certain kinds of tree
2909: expressions (declarations and types, to be precise).
1.1 root 2910:
2911: In debugging dumps and machine descriptions, the machine mode of an RTL
2912: expression is written after the expression code with a colon to separate
2913: them. The letters @samp{mode} which appear at the end of each machine mode
2914: name are omitted. For example, @code{(reg:SI 38)} is a @samp{reg}
2915: expression with machine mode @code{SImode}. If the mode is
2916: @code{VOIDmode}, it is not written at all.
2917:
2918: Here is a table of machine modes.
2919:
2920: @table @code
2921: @item QImode
2922: ``Quarter-Integer'' mode represents a single byte treated as an integer.
2923:
2924: @item HImode
2925: ``Half-Integer'' mode represents a two-byte integer.
2926:
2927: @item SImode
2928: ``Single Integer'' mode represents a four-byte integer.
2929:
2930: @item DImode
2931: ``Double Integer'' mode represents an eight-byte integer.
2932:
2933: @item TImode
2934: ``Tetra Integer'' (?) mode represents a sixteen-byte integer.
2935:
2936: @item SFmode
2937: ``Single Floating'' mode represents a single-precision (four byte) floating
2938: point number.
2939:
2940: @item DFmode
2941: ``Double Floating'' mode represents a double-precision (eight byte) floating
2942: point number.
2943:
2944: @item TFmode
2945: ``Tetra Floating'' mode represents a quadruple-precision (sixteen byte)
2946: floating point number.
2947:
2948: @item BLKmode
2949: ``Block'' mode represents values that are aggregates to which none of
1.1.1.2 root 2950: the other modes apply. In RTL, only memory references can have this mode,
1.1 root 2951: and only if they appear in string-move or vector instructions. On machines
2952: which have no such instructions, @code{BLKmode} will not appear in RTL.
2953:
2954: @item VOIDmode
2955: Void mode means the absence of a mode or an unspecified mode.
1.1.1.2 root 2956: For example, RTL expressions of code @samp{const_int} have mode
1.1 root 2957: @code{VOIDmode} because they can be taken to have whatever mode the context
2958: requires. In debugging dumps of RTL, @code{VOIDmode} is expressed by
2959: the absence of any mode.
2960:
2961: @item EPmode
2962: ``Entry Pointer'' mode is intended to be used for function variables in
2963: Pascal and other block structured languages. Such values contain
2964: both a function address and a static chain pointer for access to
2965: automatic variables of outer levels. This mode is only partially
2966: implemented since C does not use it.
2967:
2968: @item CSImode@r{, @dots{}}
2969: ``Complex Single Integer'' mode stands for a complex number represented
2970: as a pair of @code{SImode} integers. Any of the integer and floating modes
2971: may have @samp{C} prefixed to its name to obtain a complex number mode.
2972: For example, there are @code{CQImode}, @code{CSFmode}, and @code{CDFmode}.
2973: Since C does not support complex numbers, these machine modes are only
2974: partially implemented.
2975:
2976: @item BImode
2977: This is the machine mode of a bit-field in a structure. It is used
2978: only in the syntax tree, never in RTL, and in the syntax tree it appears
2979: only in declaration nodes. In C, it appears only in @code{FIELD_DECL}
2980: nodes for structure fields defined with a bit size.
2981: @end table
2982:
2983: The machine description defines @code{Pmode} as a C macro which expands
2984: into the machine mode used for addresses. Normally this is @code{SImode}.
2985:
2986: The only modes which a machine description @i{must} support are
2987: @code{QImode}, @code{SImode}, @code{SFmode} and @code{DFmode}. The
2988: compiler will attempt to use @code{DImode} for two-word structures and
2989: unions, but it would not be hard to program it to avoid this. Likewise,
2990: you can arrange for the C type @code{short int} to avoid using
2991: @code{HImode}. In the long term it would be desirable to make the set of
2992: available machine modes machine-dependent and eliminate all assumptions
2993: about specific machine modes or their uses from the machine-independent
2994: code of the compiler.
2995:
2996: Here are some C macros that relate to machine modes:
2997:
2998: @table @code
2999: @item GET_MODE (@var{x})
1.1.1.2 root 3000: Returns the machine mode of the RTX @var{x}.
1.1 root 3001:
3002: @item PUT_MODE (@var{x}, @var{newmode})
1.1.1.2 root 3003: Alters the machine mode of the RTX @var{x} to be @var{newmode}.
1.1 root 3004:
3005: @item GET_MODE_SIZE (@var{m})
3006: Returns the size in bytes of a datum of mode @var{m}.
3007:
3008: @item GET_MODE_BITSIZE (@var{m})
3009: Returns the size in bits of a datum of mode @var{m}.
3010:
3011: @item GET_MODE_UNIT_SIZE (@var{m})
3012: Returns the size in bits of the subunits of a datum of mode @var{m}.
3013: This is the same as @code{GET_MODE_SIZE} except in the case of
1.1.1.2 root 3014: complex modes and @code{EPmode}. For them, the unit size is the
1.1 root 3015: size of the real or imaginary part, or the size of the function
3016: pointer or the context pointer.
3017: @end table
3018:
3019: @node Constants, Regs and Memory, Machine Modes, RTL
3020: @section Constant Expression Types
3021:
3022: The simplest RTL expressions are those that represent constant values.
3023:
3024: @table @code
3025: @item (const_int @var{i})
3026: This type of expression represents the integer value @var{i}. @var{i}
3027: is customarily accessed with the macro @code{INTVAL} as in
1.1.1.2 root 3028: @code{INTVAL (@var{exp})}, which is equivalent to @code{XINT (@var{exp}, 0)}.
1.1 root 3029:
3030: There is only one expression object for the integer value zero;
3031: it is the value of the variable @code{const0_rtx}. Likewise, the
3032: only expression for integer value one is found in @code{const1_rtx}.
1.1.1.2 root 3033: Any attempt to create an expression of code @samp{const_int} and
1.1 root 3034: value zero or one will return @code{const0_rtx} or @code{const1_rtx}
3035: as appropriate.
3036:
3037: @item (const_double:@var{m} @var{i0} @var{i1})
3038: Represents a floating point constant value of mode @var{m}. The two
3039: integers @var{i0} and @var{i1} together contain the bits of a
3040: @code{double} value. To convert them to a @code{double}, do
3041:
3042: @example
1.1.1.2 root 3043: union @{ double d; int i[2];@} u;
1.1 root 3044: u.i[0] = XINT (x, 0);
3045: u.i[1] = XINT (x, 1);
3046: @end example
3047:
3048: @noindent
3049: and then refer to @code{u.d}. The value of the constant is
3050: represented as a double in this fashion even if the value represented
3051: is single-precision.
3052:
1.1.1.2 root 3053: The global variables @code{dconst0_rtx} and @code{fconst0_rtx} hold
3054: @samp{const_double} expressions with value 0, in modes @code{DFmode} and
3055: @code{SFmode}, respectively.
1.1 root 3056:
3057: @item (symbol_ref @var{symbol})
3058: Represents the value of an assembler label for data. @var{symbol} is
3059: a string that describes the name of the assembler label. If it starts
3060: with a @samp{*}, the label is the rest of @var{symbol} not including
3061: the @samp{*}. Otherwise, the label is @var{symbol}, prefixed with
3062: @samp{_}.
3063:
3064: @item (label_ref @var{label})
3065: Represents the value of an assembler label for code. It contains one
1.1.1.2 root 3066: operand, an expression, which must be a @samp{code_label} that appears
1.1 root 3067: in the instruction sequence to identify the place where the label
3068: should go.
3069:
3070: The reason for using a distinct expression type for code label
3071: references is so that jump optimization can distinguish them.
3072:
3073: @item (const @var{exp})
3074: Represents a constant that is the result of an assembly-time
3075: arithmetic computation. The operand, @var{exp}, is an expression that
3076: contains only constants (@samp{const_int}, @samp{symbol_ref} and
3077: @samp{label_ref} expressions) combined with @samp{plus} and
3078: @samp{minus}. However, not all combinations are valid, since the
3079: assembler cannot do arbitrary arithmetic on relocatable symbols.
3080: @end table
3081:
3082: @node Regs and Memory, Arithmetic, Constants, RTL
3083: @section Registers and Memory
3084:
3085: Here are the RTL expression types for describing access to machine
3086: registers and to main memory.
3087:
3088: @table @code
3089: @item (reg:@var{m} @var{n})
3090: For small values of the integer @var{n} (less than
3091: @code{FIRST_PSEUDO_REGISTER}), this stands for a reference to machine
3092: register number @var{n}: a @dfn{hard register}. For larger values of
3093: @var{n}, it stands for a temporary value or @dfn{pseudo register}.
3094: The compiler's strategy is to generate code assuming an unlimited
3095: number of such pseudo registers, and later convert them into hard
3096: registers or into memory references.
3097:
3098: The symbol @code{FIRST_PSEUDO_REGISTER} is defined by the machine
3099: description, since the number of hard registers on the machine is an
3100: invariant characteristic of the machine. Note, however, that not
3101: all of the machine registers must be general registers. All the
3102: machine registers that can be used for storage of data are given
3103: hard register numbers, even those that can be used only in certain
3104: instructions or can hold only certain types of data.
3105:
1.1.1.2 root 3106: Each pseudo register number used in a function's RTL code is
1.1 root 3107: represented by a unique @samp{reg} expression.
3108:
3109: @var{m} is the machine mode of the reference. It is necessary because
3110: machines can generally refer to each register in more than one mode.
3111: For example, a register may contain a full word but there may be
3112: instructions to refer to it as a half word or as a single byte, as
3113: well as instructions to refer to it as a floating point number of
3114: various precisions.
3115:
3116: Even for a register that the machine can access in only one mode,
3117: the mode must always be specified.
3118:
3119: A hard register may be accessed in various modes throughout one
3120: function, but each pseudo register is given a natural mode
3121: and is accessed only in that mode. When it is necessary to describe
3122: an access to a pseudo register using a nonnatural mode, a @samp{subreg}
3123: expression is used.
3124:
3125: A @samp{reg} expression with a machine mode that specifies more than
3126: one word of data may actually stand for several consecutive registers.
3127: If in addition the register number specifies a hardware register, then
3128: it actually represents several consecutive hardware registers starting
3129: with the specified one.
3130:
3131: Such multi-word hardware register @samp{reg} expressions may not be live
3132: across the boundary of a basic block. The lifetime analysis pass does not
3133: know how to record properly that several consecutive registers are
3134: actually live there, and therefore register allocation would be confused.
3135: The CSE pass must go out of its way to make sure the situation does
3136: not arise.
3137:
3138: @item (subreg:@var{m} @var{reg} @var{wordnum})
3139: @samp{subreg} expressions are used to refer to a register in a machine
3140: mode other than its natural one, or to refer to one register of
3141: a multi-word @samp{reg} that actually refers to several registers.
3142:
3143: Each pseudo-register has a natural mode. If it is necessary to
3144: operate on it in a different mode---for example, to perform a fullword
3145: move instruction on a pseudo-register that contains a single byte---
3146: the pseudo-register must be enclosed in a @samp{subreg}. In such
3147: a case, @var{wordnum} is zero.
3148:
3149: The other use of @samp{subreg} is to extract the individual registers
3150: of a multi-register value. Machine modes such as @code{DImode} and
3151: @code{EPmode} indicate values longer than a word, values which usually
3152: require two consecutive registers. To access one of the registers,
3153: use a @samp{subreg} with mode @code{SImode} and a @var{wordnum} that
3154: says which register.
3155:
3156: The compilation parameter @code{WORDS_BIG_ENDIAN}, if defined, says
3157: that word number zero is the most significant part; otherwise, it is
3158: the least significant part.
3159:
3160: Note that it is not valid to access a @code{DFmode} value in @code{SFmode}
3161: using a @samp{subreg}. On some machines the most significant part of a
3162: @code{DFmode} value does not have the same format as a single-precision
3163: floating value.
3164:
3165: @item (cc0)
3166: This refers to the machine's condition code register. It has no
3167: operands and may not have a machine mode. It may be validly used in
3168: only two contexts: as the destination of an assignment (in test and
3169: compare instructions) and in comparison operators comparing against
1.1.1.2 root 3170: zero (@samp{const_int} with value zero; that is to say,
3171: @code{const0_rtx}).
1.1 root 3172:
1.1.1.2 root 3173: There is only one expression object of code @samp{cc0}; it is the
1.1 root 3174: value of the variable @code{cc0_rtx}. Any attempt to create an
1.1.1.2 root 3175: expression of code @samp{cc0} will return @code{cc0_rtx}.
1.1 root 3176:
1.1.1.2 root 3177: One special thing about the condition code register is that
3178: instructions can set it implicitly. On many machines, nearly all
3179: instructions set the condition code based on the value that they
3180: compute or store. It is not necessary to record these actions
3181: explicitly in the RTL because the machine description includes a
3182: prescription for recognizing the instructions that do so (by means of
3183: the macro @code{NOTICE_UPDATE_CC}). Only instructions whose sole
3184: purpose is to set the condition code, and instructions that use the
3185: condition code, need mention @code{(cc0)}.
1.1 root 3186:
3187: @item (pc)
3188: This represents the machine's program counter. It has no operands and
3189: may not have a machine mode. @code{(pc)} may be validly used only in
3190: certain specific contexts in jump instructions.
3191:
1.1.1.2 root 3192: There is only one expression object of code @samp{pc}; it is the value
3193: of the variable @code{pc_rtx}. Any attempt to create an expression of
3194: code @samp{pc} will return @code{pc_rtx}.
1.1 root 3195:
1.1.1.2 root 3196: All instructions that do not jump alter the program counter implicitly
3197: by incrementing it, but there is no need to mention this in the RTL.
1.1 root 3198:
3199: @item (mem:@var{m} @var{addr})
1.1.1.2 root 3200: This RTX represents a reference to main memory at an address
3201: represented by the expression @var{addr}. @var{m} specifies how large
3202: a unit of memory is accessed.
1.1 root 3203: @end table
3204:
3205: @node Arithmetic, Comparisons, Regs and Memory, RTL
3206: @section RTL Expressions for Arithmetic
3207:
3208: @table @code
3209: @item (plus:@var{m} @var{x} @var{y})
3210: Represents the sum of the values represented by @var{x} and @var{y}
3211: carried out in machine mode @var{m}. This is valid only if
3212: @var{x} and @var{y} both are valid for mode @var{m}.
3213:
3214: @item (minus:@var{m} @var{x} @var{y})
3215: Like @samp{plus} but represents subtraction.
3216:
3217: @item (minus @var{x} @var{y})
3218: Represents the result of subtracting @var{y} from @var{x}
3219: for purposes of comparison. The absence of a machine mode
3220: in the @samp{minus} expression indicates that the result is
3221: computed without overflow, as if with infinite precision.
3222:
3223: Of course, machines can't really subtract with infinite precision.
3224: However, they can pretend to do so when only the sign of the
3225: result will be used, which is the case when the result is stored
1.1.1.2 root 3226: in @code{(cc0)}. And that is the only way this kind of expression
1.1 root 3227: may validly be used: as a value to be stored in the condition codes.
3228:
3229: @item (neg:@var{m} @var{x})
3230: Represents the negation (subtraction from zero) of the value
3231: represented by @var{x}, carried out in mode @var{m}. @var{x} must be
3232: valid for mode @var{m}.
3233:
3234: @item (mult:@var{m} @var{x} @var{y})
3235: Represents the signed product of the values represented by @var{x} and
3236: @var{y} carried out in machine mode @var{m}. If
3237: @var{x} and @var{y} are both valid for mode @var{m}, this is ordinary
1.1.1.2 root 3238: size-preserving multiplication. Alternatively, both @var{x} and @var{y}
1.1 root 3239: may be valid for a different, narrower mode. This represents the
3240: kind of multiplication that generates a product wider than the operands.
3241: Widening multiplication and same-size multiplication are completely
3242: distinct and supported by different machine instructions; machines may
1.1.1.2 root 3243: support one but not the other.@refill
1.1 root 3244:
3245: @samp{mult} may be used for floating point division as well.
3246: Then @var{m} is a floating point machine mode.
3247:
3248: @item (umult:@var{m} @var{x} @var{y})
3249: Like @samp{mult} but represents unsigned multiplication. It may be
3250: used in both same-size and widening forms, like @samp{mult}.
1.1.1.2 root 3251: @samp{umult} is used only for fixed-point multiplication.
1.1 root 3252:
3253: @item (div:@var{m} @var{x} @var{y})
3254: Represents the quotient in signed division of @var{x} by @var{y},
3255: carried out in machine mode @var{m}. If @var{m} is a floating-point
3256: mode, it represents the exact quotient; otherwise, the integerized
3257: quotient. If @var{x} and @var{y} are both valid for mode @var{m},
3258: this is ordinary size-preserving division. Some machines have
3259: division instructions in which the operands and quotient widths are
3260: not all the same; such instructions are represented by @samp{div}
3261: expressions in which the machine modes are not all the same.
3262:
3263: @item (udiv:@var{m} @var{x} @var{y})
3264: Like @samp{div} but represents unsigned division.
3265:
3266: @item (mod:@var{m} @var{x} @var{y})
3267: @itemx (umod:@var{m} @var{x} @var{y})
3268: Like @samp{div} and @samp{udiv} but represent the remainder instead of
3269: the quotient.
3270:
3271: @item (not:@var{m} @var{x})
3272: Represents the bitwise complement of the value represented by @var{x},
3273: carried out in mode @var{m}, which must be a fixed-point machine mode.
3274: @var{x} must be valid for mode @var{m}, which must be a fixed-point mode.
3275:
3276: @item (and:@var{m} @var{x} @var{y})
3277: Represents the bitwise logical-and of the values represented by
3278: @var{x} and @var{y}, carried out in machine mode @var{m}. This is
3279: valid only if @var{x} and @var{y} both are valid for mode @var{m},
3280: which must be a fixed-point mode.
3281:
3282: @item (ior:@var{m} @var{x} @var{y})
3283: Represents the bitwise inclusive-or of the values represented by
3284: @var{x} and @var{y}, carried out in machine mode @var{m}. This is
3285: valid only if @var{x} and @var{y} both are valid for mode @var{m},
3286: which must be a fixed-point mode.
3287:
3288: @item (xor:@var{m} @var{x} @var{y})
3289: Represents the bitwise exclusive-or of the values represented by
3290: @var{x} and @var{y}, carried out in machine mode @var{m}. This is
3291: valid only if @var{x} and @var{y} both are valid for mode @var{m},
3292: which must be a fixed-point mode.
3293:
3294: @item (lshift:@var{m} @var{x} @var{c})
3295: Represents the result of logically shifting @var{x} left by @var{c}
3296: places. @var{x} must be valid for the mode @var{m}, a fixed-point
3297: machine mode. @var{c} must be valid for a fixed-point mode;
3298: which mode is determined by the mode called for in the machine
3299: description entry for the left-shift instruction. For example,
3300: on the Vax, the mode of @var{c} is @code{QImode} regardless of @var{m}.
3301:
3302: On some machines, negative values of @var{c} may be meaningful; this
1.1.1.2 root 3303: is why logical left shift and arithmetic left shift are distinguished.
1.1 root 3304: For example, Vaxes have no right-shift instructions, and right shifts
3305: are represented as left-shift instructions whose counts happen
3306: to be negative constants or else computed (in a previous instruction)
3307: by negation.
3308:
3309: @item (ashift:@var{m} @var{x} @var{c})
3310: Like @samp{lshift} but for arithmetic left shift.
3311:
3312: @item (lshiftrt:@var{m} @var{x} @var{c})
3313: @itemx (ashiftrt:@var{m} @var{x} @var{c})
3314: Like @samp{lshift} and @samp{ashift} but for right shift.
3315:
3316: @item (rotate:@var{m} @var{x} @var{c})
3317: @itemx (rotatert:@var{m} @var{x} @var{c})
3318: Similar but represent left and right rotate.
3319:
3320: @item (abs:@var{m} @var{x})
3321: Represents the absolute value of @var{x}, computed in mode @var{m}.
3322: @var{x} must be valid for @var{m}.
3323:
3324: @item (sqrt:@var{m} @var{x})
3325: Represents the square root of @var{x}, computed in mode @var{m}.
3326: @var{x} must be valid for @var{m}. Most often @var{m} will be
3327: a floating point mode.
1.1.1.2 root 3328:
3329: @item (ffs:@var{m} @var{x})
3330: Represents the one plus the index of the least significant 1-bit in
3331: @var{x}, represented as an integer of mode @var{m}. (The value is
3332: zero if @var{x} is zero.) The mode of @var{x} need not be @var{m};
3333: depending on the target machine, various mode combinations may be
3334: valid.
1.1 root 3335: @end table
3336:
3337: @node Comparisons, Bit Fields, Arithmetic, RTL
3338: @section Comparison Operations
3339:
3340: Comparison operators test a relation on two operands and are considered to
3341: represent the value 1 if the relation holds, or zero if it does not. The
3342: mode of the comparison is determined by the operands; they must both be
3343: valid for a common machine mode. A comparison with both operands constant
3344: would be invalid as the machine mode could not be deduced from it, but such
1.1.1.2 root 3345: a comparison should never exist in RTL due to constant folding.
1.1 root 3346:
3347: Inequality comparisons come in two flavors, signed and unsigned. Thus,
1.1.1.2 root 3348: there are distinct expression codes @samp{gt} and @samp{gtu} for signed and
1.1 root 3349: unsigned greater-than. These can produce different results for the same
3350: pair of integer values: for example, 1 is signed greater-than -1 but not
3351: unsigned greater-than, because -1 when regarded as unsigned is actually
1.1.1.2 root 3352: @code{0xffffffff} which is greater than 1.
1.1 root 3353:
3354: The signed comparisons are also used for floating point values. Floating
3355: point comparisons are distinguished by the machine modes of the operands.
3356:
3357: The comparison operators may be used to compare the condition codes
1.1.1.2 root 3358: @code{(cc0)} against zero, as in @code{(eq (cc0) (const_int 0))}. Such a
3359: construct actually refers to the result of the preceding instruction in
3360: which the condition codes were set. The above example stands for 1 if the
3361: condition codes were set to say ``zero'' or ``equal'', 0 otherwise.
3362: Although the same comparison operators are used for this as may be used in
3363: other contexts on actual data, no confusion can result since the machine
3364: description would never allow both kinds of uses in the same context.
1.1 root 3365:
3366: @table @code
3367: @item (eq @var{x} @var{y})
3368: 1 if the values represented by @var{x} and @var{y} are equal,
3369: otherwise 0.
3370:
3371: @item (ne @var{x} @var{y})
3372: 1 if the values represented by @var{x} and @var{y} are not equal,
3373: otherwise 0.
3374:
3375: @item (gt @var{x} @var{y})
3376: 1 if the @var{x} is greater than @var{y}. If they are fixed-point,
3377: the comparison is done in a signed sense.
3378:
3379: @item (gtu @var{x} @var{y})
3380: Like @samp{gt} but does unsigned comparison, on fixed-point numbers only.
3381:
3382: @item (lt @var{x} @var{y})
3383: @item (ltu @var{x} @var{y})
3384: Like @samp{gt} and @samp{gtu} but test for ``less than''.
3385:
3386: @item (ge @var{x} @var{y})
3387: @item (geu @var{x} @var{y})
3388: Like @samp{gt} and @samp{gtu} but test for ``greater than or equal''.
3389:
3390: @item (le @var{x} @var{y})
3391: @item (leu @var{x} @var{y})
3392: Like @samp{gt} and @samp{gtu} but test for ``less than or equal''.
3393:
3394: @item (if_then_else @var{cond} @var{then} @var{else})
3395: This is not a comparison operation but is listed here because it is
3396: always used in conjunction with a comparison operation. To be
3397: precise, @var{cond} is a comparison expression. This expression
3398: represents a choice, according to @var{cond}, between the value
3399: represented by @var{then} and the one represented by @var{else}.
3400:
3401: On most machines, @samp{if_then_else} expressions are valid only
3402: to express conditional jumps.
3403: @end table
3404:
3405: @node Bit Fields, Conversions, Comparisons, RTL
3406: @section Bit-fields
3407:
3408: Special expression codes exist to represent bit-field instructions.
1.1.1.2 root 3409: These types of expressions are lvalues in RTL; they may appear
1.1 root 3410: on the left side of a assignment, indicating insertion of a value
3411: into the specified bit field.
3412:
3413: @table @code
3414: @item (sign_extract:SI @var{loc} @var{size} @var{pos})
3415: This represents a reference to a sign-extended bit-field contained or
3416: starting in @var{loc} (a memory or register reference). The bit field
3417: is @var{size} bits wide and starts at bit @var{pos}. The compilation
1.1.1.2 root 3418: option @code{BITS_BIG_ENDIAN} says which end of the memory unit
1.1 root 3419: @var{pos} counts from.
3420:
3421: Which machine modes are valid for @var{loc} depends on the machine,
3422: but typically @var{loc} should be a single byte when in memory
3423: or a full word in a register.
3424:
1.1.1.2 root 3425: @item (zero_extract:SI @var{loc} @var{size} @var{pos})
1.1 root 3426: Like @samp{sign_extract} but refers to an unsigned or zero-extended
3427: bit field. The same sequence of bits are extracted, but they
3428: are filled to an entire word with zeros instead of by sign-extension.
3429: @end table
3430:
3431: @node Conversions, RTL Declarations, Bit Fields, RTL
3432: @section Conversions
3433:
3434: All conversions between machine modes must be represented by
3435: explicit conversion operations. For example, an expression
1.1.1.2 root 3436: which is the sum of a byte and a full word cannot be written as
1.1 root 3437: @code{(plus:SI (reg:QI 34) (reg:SI 80))} because the @samp{plus}
3438: operation requires two operands of the same machine mode.
3439: Therefore, the byte-sized operand is enclosed in a conversion
3440: operation, as in
3441:
3442: @example
3443: (plus:SI (sign_extend:SI (reg:QI 34)) (reg:SI 80))
3444: @end example
3445:
3446: The conversion operation is not a mere placeholder, because there
3447: may be more than one way of converting from a given starting mode
3448: to the desired final mode. The conversion operation code says how
3449: to do it.
3450:
3451: @table @code
3452: @item (sign_extend:@var{m} @var{x})
3453: Represents the result of sign-extending the value @var{x}
3454: to machine mode @var{m}. @var{m} must be a fixed-point mode
3455: and @var{x} a fixed-point value of a mode narrower than @var{m}.
3456:
3457: @item (zero_extend:@var{m} @var{x})
3458: Represents the result of zero-extending the value @var{x}
3459: to machine mode @var{m}. @var{m} must be a fixed-point mode
3460: and @var{x} a fixed-point value of a mode narrower than @var{m}.
3461:
3462: @item (float_extend:@var{m} @var{x})
3463: Represents the result of extending the value @var{x}
3464: to machine mode @var{m}. @var{m} must be a floating point mode
3465: and @var{x} a floating point value of a mode narrower than @var{m}.
3466:
3467: @item (truncate:@var{m} @var{x})
3468: Represents the result of truncating the value @var{x}
3469: to machine mode @var{m}. @var{m} must be a fixed-point mode
3470: and @var{x} a fixed-point value of a mode wider than @var{m}.
3471:
3472: @item (float_truncate:@var{m} @var{x})
3473: Represents the result of truncating the value @var{x}
3474: to machine mode @var{m}. @var{m} must be a floating point mode
3475: and @var{x} a floating point value of a mode wider than @var{m}.
3476:
3477: @item (float:@var{m} @var{x})
1.1.1.2 root 3478: Represents the result of converting fixed point value @var{x},
3479: regarded as signed, to floating point mode @var{m}.
3480:
3481: @item (unsigned_float:@var{m} @var{x})
3482: Represents the result of converting fixed point value @var{x},
3483: regarded as unsigned, to floating point mode @var{m}.
1.1 root 3484:
3485: @item (fix:@var{m} @var{x})
1.1.1.2 root 3486: When @var{m} is a fixed point mode, represents the result of
3487: converting floating point value @var{x} to mode @var{m}, regarded as
3488: signed. How rounding is done is not specified, so this operation may
3489: be used validly in compiling C code only for integer-valued operands.
3490:
3491: @item (unsigned_fix:@var{m} @var{x})
3492: Represents the result of converting floating point value @var{x} to
3493: fixed point mode @var{m}, regarded as unsigned. How rounding is done
3494: is not specified.
1.1 root 3495:
1.1.1.2 root 3496: @item (fix:@var{m} @var{x})
3497: When @var{m} is a floating point mode, represents the result of
3498: converting floating point value @var{x} (valid for mode @var{m}) to an
3499: integer, still represented in floating point mode @var{m}, by rounding
3500: towards zero.
1.1 root 3501: @end table
3502:
3503: @node RTL Declarations, Side Effects, Conversions, RTL
3504: @section Declarations
3505:
3506: Declaration expression codes do not represent arithmetic operations
3507: but rather state assertions about their operands.
3508:
3509: @table @code
3510: @item (strict_low_part (subreg:@var{m} (reg:@var{n} @var{r}) 0))
3511: This expression code is used in only one context: operand 0 of a
3512: @samp{set} expression. In addition, the operand of this expression
3513: must be a @samp{subreg} expression.
3514:
3515: The presence of @samp{strict_low_part} says that the part of the
1.1.1.2 root 3516: register which is meaningful in mode @var{n}, but is not part of
3517: mode @var{m}, is not to be altered. Normally, an assignment to such
1.1 root 3518: a subreg is allowed to have undefined effects on the rest of the
3519: register when @var{m} is less than a word.
3520: @end table
3521:
3522: @node Side Effects, Incdec, RTL Declarations, RTL
3523: @section Side Effect Expressions
3524:
3525: The expression codes described so far represent values, not actions.
3526: But machine instructions never produce values; they are meaningful
3527: only for their side effects on the state of the machine. Special
3528: expression codes are used to represent side effects.
3529:
3530: The body of an instruction is always one of these side effect codes;
3531: the codes described above, which represent values, appear only as
3532: the operands of these.
3533:
3534: @table @code
3535: @item (set @var{lval} @var{x})
3536: Represents the action of storing the value of @var{x} into the place
3537: represented by @var{lval}. @var{lval} must be an expression
3538: representing a place that can be stored in: @samp{reg} (or
3539: @samp{subreg} or @samp{strict_low_part}), @samp{mem}, @samp{pc} or
1.1.1.2 root 3540: @samp{cc0}.@refill
1.1 root 3541:
3542: If @var{lval} is a @samp{reg}, @samp{subreg} or @samp{mem}, it has a
1.1.1.2 root 3543: machine mode; then @var{x} must be valid for that mode.@refill
1.1 root 3544:
3545: If @var{lval} is a @samp{reg} whose machine mode is less than the full
3546: width of the register, then it means that the part of the register
3547: specified by the machine mode is given the specified value and the
3548: rest of the register receives an undefined value. Likewise, if
3549: @var{lval} is a @samp{subreg} whose machine mode is narrower than
3550: @code{SImode}, the rest of the register can be changed in an undefined way.
3551:
3552: If @var{lval} is a @samp{strict_low_part} of a @samp{subreg}, then the
3553: part of the register specified by the machine mode of the
3554: @samp{subreg} is given the value @var{x} and the rest of the register
1.1.1.2 root 3555: is not changed.@refill
1.1 root 3556:
3557: If @var{lval} is @code{(cc0)}, it has no machine mode, and @var{x} may
1.1.1.2 root 3558: have any mode. This represents a ``test'' or ``compare'' instruction.@refill
1.1 root 3559:
3560: If @var{lval} is @code{(pc)}, we have a jump instruction, and the
3561: possibilities for @var{x} are very limited. It may be a
3562: @samp{label_ref} expression (unconditional jump). It may be an
3563: @samp{if_then_else} (conditional jump), in which case either the
3564: second or the third operand must be @code{(pc)} (for the case which
3565: does not jump) and the other of the two must be a @samp{label_ref}
3566: (for the case which does jump). @var{x} may also be a @samp{mem} or
3567: @code{(plus:SI (pc) @var{y})}, where @var{y} may be a @samp{reg} or a
3568: @samp{mem}; these unusual patterns are used to represent jumps through
1.1.1.2 root 3569: branch tables.@refill
1.1 root 3570:
3571: @item (return)
1.1.1.2 root 3572: Represents a return from the current function, on machines where this
3573: can be done with one instruction, such as Vaxes. On machines where a
3574: multi-instruction ``epilogue'' must be executed in order to return
3575: from the function, returning is done by jumping to a label which
3576: precedes the epilogue, and the @samp{return} expression code is never
3577: used.
1.1 root 3578:
3579: @item (call @var{function} @var{nargs})
3580: Represents a function call. @var{function} is a @samp{mem} expression
3581: whose address is the address of the function to be called. @var{nargs}
3582: is an expression representing the number of words of argument.
3583:
3584: Each machine has a standard machine mode which @var{function} must
1.1.1.2 root 3585: have. The machine description defines macro @code{FUNCTION_MODE} to
1.1 root 3586: expand into the requisite mode name. The purpose of this mode is to
3587: specify what kind of addressing is allowed, on machines where the
3588: allowed kinds of addressing depend on the machine mode being
3589: addressed.
3590:
3591: @item (clobber @var{x})
3592: Represents the storing or possible storing of an unpredictable,
3593: undescribed value into @var{x}, which must be a @samp{reg} or
3594: @samp{mem} expression.
3595:
3596: One place this is used is in string instructions that store standard
3597: values into particular hard registers. It may not be worth the
1.1.1.2 root 3598: trouble to describe the values that are stored, but it is essential to
3599: inform the compiler that the registers will be altered, lest it
1.1 root 3600: attempt to keep data in them across the string instruction.
3601:
3602: @var{x} may also be null---a null C pointer, no expression at all.
3603: Such a @code{(clobber (null))} expression means that all memory
3604: locations must be presumed clobbered.
3605:
3606: Note that the machine description classifies certain hard registers as
3607: ``call-clobbered''. All function call instructions are assumed by
3608: default to clobber these registers, so there is no need to use
3609: @samp{clobber} expressions to indicate this fact. Also, each function
3610: call is assumed to have the potential to alter any memory location.
3611:
3612: @item (use @var{x})
1.1.1.2 root 3613: Represents the use of the value of @var{x}. It indicates that the
3614: value in @var{x} at this point in the program is needed, even though
3615: it may not be apparent why this is so. Therefore, the compiler will
3616: not attempt to delete instructions whose only effect is to store a
3617: value in @var{x}. @var{x} must be a @samp{reg} expression.
1.1 root 3618:
3619: @item (parallel [@var{x0} @var{x1} @dots{}])
3620: Represents several side effects performed in parallel. The square
3621: brackets stand for a vector; the operand of @samp{parallel} is a
3622: vector of expressions. @var{x0}, @var{x1} and so on are individual
3623: side effects---expressions of code @samp{set}, @samp{call},
1.1.1.2 root 3624: @samp{return}, @samp{clobber} or @samp{use}.@refill
1.1 root 3625:
1.1.1.2 root 3626: ``In parallel'' means that first all the values used in the individual
3627: side-effects are computed, and second all the actual side-effects are
3628: performed. For example,
1.1 root 3629:
3630: @example
3631: (parallel [(set (reg:SI 1) (mem:SI (reg:SI 1)))
3632: (set (mem:SI (reg:SI 1)) (reg:SI 1))])
3633: @end example
3634:
3635: @noindent
3636: says unambiguously that the values of hard register 1 and the memory
3637: location addressed by it are interchanged. In both places where
3638: @code{(reg:SI 1)} appears as a memory address it refers to the value
1.1.1.2 root 3639: in register 1 @emph{before} the execution of the instruction.
3640:
3641: @item (sequence [@var{insns} @dots{}])
3642: Represents a sequence of insns. Each of the @var{insns} that appears
3643: in the vector is suitable for appearing in the chain of insns, so it
3644: must be an @samp{insn}, @samp{jump_insn}, @samp{call_insn},
3645: @samp{code_label}, @samp{barrier} or @samp{note}.
3646:
3647: A @samp{sequence} RTX never appears in an actual insn. It represents
3648: the sequence of insns that result from a @samp{define_expand}
3649: @emph{before} those insns are passed to @code{emit_insn} to insert
3650: them in the chain of insns. When actually inserted, the individual
3651: sub-insns are separated out and the @samp{sequence} is forgotten.
1.1 root 3652: @end table
3653:
1.1.1.2 root 3654: Three expression codes appear in place of a side effect, as the body of an
3655: insn, though strictly speaking they do not describe side effects as such:
1.1 root 3656:
3657: @table @code
3658: @item (asm_input @var{s})
3659: Represents literal assembler code as described by the string @var{s}.
3660:
3661: @item (addr_vec:@var{m} [@var{lr0} @var{lr1} @dots{}])
1.1.1.2 root 3662: Represents a table of jump addresses. The vector elements @var{lr0},
3663: etc., are @samp{label_ref} expressions. The mode @var{m} specifies
3664: how much space is given to each address; normally @var{m} would be
1.1 root 3665: @code{Pmode}.
3666:
3667: @item (addr_diff_vec:@var{m} @var{base} [@var{lr0} @var{lr1} @dots{}])
3668: Represents a table of jump addresses expressed as offsets from
1.1.1.2 root 3669: @var{base}. The vector elements @var{lr0}, etc., are @samp{label_ref}
3670: expressions and so is @var{base}. The mode @var{m} specifies how much
3671: space is given to each address-difference.@refill
1.1 root 3672: @end table
3673:
1.1.1.2 root 3674: @node Incdec, Assembler, Side Effects, RTL
1.1 root 3675: @section Embedded Side-Effects on Addresses
3676:
3677: Four special side-effect expression codes appear as memory addresses.
3678:
3679: @table @code
3680: @item (pre_dec:@var{m} @var{x})
3681: Represents the side effect of decrementing @var{x} by a standard
3682: amount and represents also the value that @var{x} has after being
3683: decremented. @var{x} must be a @samp{reg} or @samp{mem}, but most
3684: machines allow only a @samp{reg}. @var{m} must be the machine mode
1.1.1.2 root 3685: for pointers on the machine in use. The amount @var{x} is decremented
1.1 root 3686: by is the length in bytes of the machine mode of the containing memory
3687: reference of which this expression serves as the address. Here is an
1.1.1.2 root 3688: example of its use:@refill
1.1 root 3689:
3690: @example
3691: (mem:DF (pre_dec:SI (reg:SI 39)))
3692: @end example
3693:
3694: @noindent
3695: This says to decrement pseudo register 39 by the length of a @code{DFmode}
3696: value and use the result to address a @code{DFmode} value.
3697:
3698: @item (pre_inc:@var{m} @var{x})
3699: Similar, but specifies incrementing @var{x} instead of decrementing it.
3700:
3701: @item (post_dec:@var{m} @var{x})
3702: Represents the same side effect as @samp{pre_decrement} but a different
3703: value. The value represented here is the value @var{x} has @i{before}
3704: being decremented.
3705:
3706: @item (post_inc:@var{m} @var{x})
3707: Similar, but specifies incrementing @var{x} instead of decrementing it.
3708: @end table
3709:
3710: These embedded side effect expressions must be used with care. Instruction
3711: patterns may not use them. Until the @samp{flow} pass of the compiler,
3712: they may occur only to represent pushes onto the stack. The @samp{flow}
3713: pass finds cases where registers are incremented or decremented in one
3714: instruction and used as an address shortly before or after; these cases are
3715: then transformed to use pre- or post-increment or -decrement.
3716:
3717: Explicit popping of the stack could be represented with these embedded
3718: side effect operators, but that would not be safe; the instruction
3719: combination pass could move the popping past pushes, thus changing
3720: the meaning of the code.
3721:
3722: An instruction that can be represented with an embedded side effect
3723: could also be represented using @samp{parallel} containing an additional
3724: @samp{set} to describe how the address register is altered. This is not
3725: done because machines that allow these operations at all typically
3726: allow them wherever a memory address is called for. Describing them as
3727: additional parallel stores would require doubling the number of entries
3728: in the machine description.
3729:
1.1.1.2 root 3730: @node Assembler, Insns, IncDec, RTL
3731: @section Assembler Instructions as Expressions
3732:
3733: The RTX code @samp{asm_operands} represents a value produced by a
3734: user-specified assembler instruction. It is used to represent
3735: an @code{asm} statement with arguments. An @code{asm} statement with
3736: a single output operand, like this:
3737:
3738: @example
3739: asm ("foo %1,%2,%0" : "a" (outputvar) : "g" (x + y), "di" (*z));
3740: @end example
3741:
3742: @noindent
3743: is represented using a single @samp{asm_operands} RTX which represents
3744: the value that is stored in @code{outputvar}:
3745:
3746: @example
3747: (set @var{rtx-for-outputvar}
3748: (asm_operands "foo %1,%2,%0" "a" 0
3749: [@var{rtx-for-addition-result} @var{rtx-for-*z}]
3750: [(asm_input:@var{m1} "g")
3751: (asm_input:@var{m2} "di")]))
3752: @end example
3753:
3754: @noindent
3755: Here the operands of the @samp{asm_operands} RTX are the assembler
3756: template string, the output-operand's constraint, the index-number of the
3757: output operand among the output operands specified, a vector of input
3758: operand RTX's, and a vector of input-operand modes and constraints. The
3759: mode @var{m1} is the mode of the sum @code{x+y}; @var{m2} is that of
3760: @code{*z}.
3761:
3762: When an @code{asm} statement has multiple output values, its insn has
3763: several such @samp{set} RTX's inside of a @samp{parallel}. Each @samp{set}
3764: contains a @samp{asm_operands}; all of these share the same assembler
3765: template and vectors, but each contains the constraint for the respective
3766: output operand. They are also distinguished by the output-operand index
3767: number, which is 0, 1, @dots{} for successive output operands.
3768:
3769: @node Insns, Calls, Assembler, RTL
1.1 root 3770: @section Insns
3771:
3772: The RTL representation of the code for a function is a doubly-linked
3773: chain of objects called @dfn{insns}. Insns are expressions with
3774: special codes that are used for no other purpose. Some insns are
3775: actual instructions; others represent dispatch tables for @code{switch}
3776: statements; others represent labels to jump to or various sorts of
1.1.1.2 root 3777: declarative information.
1.1 root 3778:
1.1.1.2 root 3779: In addition to its own specific data, each insn must have a unique id-number
1.1 root 3780: that distinguishes it from all other insns in the current function, and
3781: chain pointers to the preceding and following insns. These three fields
3782: occupy the same position in every insn, independent of the expression code
3783: of the insn. They could be accessed with @code{XEXP} and @code{XINT},
3784: but instead three special macros are always used:
3785:
3786: @table @code
3787: @item INSN_UID (@var{i})
3788: Accesses the unique id of insn @var{i}.
3789:
3790: @item PREV_INSN (@var{i})
3791: Accesses the chain pointer to the insn preceding @var{i}.
3792: If @var{i} is the first insn, this is a null pointer.
3793:
3794: @item NEXT_INSN (@var{i})
3795: Accesses the chain pointer to the insn following @var{i}.
3796: If @var{i} is the last insn, this is a null pointer.
3797: @end table
3798:
3799: The @code{NEXT_INSN} and @code{PREV_INSN} pointers must always
3800: correspond: if @var{i} is not the first insn,
3801:
3802: @example
3803: NEXT_INSN (PREV_INSN (@var{insn})) == @var{insn}
3804: @end example
3805:
3806: @noindent
3807: is always true.
3808:
3809: Every insn has one of the following six expression codes:
3810:
1.1.1.2 root 3811: @table @samp
1.1 root 3812: @item insn
3813: The expression code @samp{insn} is used for instructions that do not jump
3814: and do not do function calls. Insns with code @samp{insn} have four
3815: additional fields beyond the three mandatory ones listed above.
3816: These four are described in a table below.
3817:
3818: @item jump_insn
3819: The expression code @samp{jump_insn} is used for instructions that may jump
3820: (or, more generally, may contain @samp{label_ref} expressions).
3821: @samp{jump_insn} insns have the same extra fields as @samp{insn} insns,
3822: accessed in the same way.
3823:
3824: @item call_insn
3825: The expression code @samp{call_insn} is used for instructions that may do
3826: function calls. It is important to distinguish these instructions because
3827: they imply that certain registers and memory locations may be altered
3828: unpredictably.
3829:
3830: @samp{call_insn} insns have the same extra fields as @samp{insn} insns,
3831: accessed in the same way.
3832:
3833: @item code_label
3834: A @samp{code_label} insn represents a label that a jump insn can jump to.
3835: It contains one special field of data in addition to the three standard ones.
3836: It is used to hold the @dfn{label number}, a number that identifies this
3837: label uniquely among all the labels in the compilation (not just in the
3838: current function). Ultimately, the label is represented in the assembler
3839: output as an assembler label @samp{L@var{n}} where @var{n} is the label number.
3840:
3841: @item barrier
3842: Barriers are placed in the instruction stream after unconditional
3843: jump instructions to indicate that the jumps are unconditional.
3844: They contain no information beyond the three standard fields.
3845:
3846: @item note
3847: @samp{note} insns are used to represent additional debugging and
1.1.1.2 root 3848: declarative information. They contain two nonstandard fields, an
1.1 root 3849: integer which is accessed with the macro @code{NOTE_LINE_NUMBER} and a
3850: string accessed with @code{NOTE_SOURCE_FILE}.
3851:
3852: If @code{NOTE_LINE_NUMBER} is positive, the note represents the
3853: position of a source line and @code{NOTE_SOURCE_FILE} is the source file name
3854: that the line came from. These notes control generation of line
3855: number data in the assembler output.
3856:
3857: Otherwise, @code{NOTE_LINE_NUMBER} is not really a line number but a
3858: code with one of the following values (and @code{NOTE_SOURCE_FILE}
3859: must contain a null pointer):
3860:
3861: @table @code
3862: @item NOTE_INSN_DELETED
3863: Such a note is completely ignorable. Some passes of the compiler
3864: delete insns by altering them into notes of this kind.
3865:
3866: @item NOTE_INSN_BLOCK_BEG
3867: @itemx NOTE_INSN_BLOCK_END
3868: These types of notes indicate the position of the beginning and end
3869: of a level of scoping of variable names. They control the output
3870: of debugging information.
3871:
3872: @item NOTE_INSN_LOOP_BEG
3873: @itemx NOTE_INSN_LOOP_END
3874: These types of notes indicate the position of the beginning and end
3875: of a @code{while} or @code{for} loop. They enable the loop optimizer
3876: to find loops quickly.
3877: @end table
3878: @end table
3879:
3880: Here is a table of the extra fields of @samp{insn}, @samp{jump_insn}
3881: and @samp{call_insn} insns:
3882:
3883: @table @code
3884: @item PATTERN (@var{i})
3885: An expression for the side effect performed by this insn.
3886:
3887: @item REG_NOTES (@var{i})
3888: A list (chain of @samp{expr_list} expressions) giving information
3889: about the usage of registers in this insn. This list is set up by the
1.1.1.2 root 3890: flow analysis pass; it is a null pointer until then.
1.1 root 3891:
3892: @item LOG_LINKS (@var{i})
3893: A list (chain of @samp{insn_list} expressions) of previous ``related''
3894: insns: insns which store into registers values that are used for the
3895: first time in this insn. (An additional constraint is that neither a
3896: jump nor a label may come between the related insns). This list is
1.1.1.2 root 3897: set up by the flow analysis pass; it is a null pointer until then.
1.1 root 3898:
3899: @item INSN_CODE (@var{i})
3900: An integer that says which pattern in the machine description matches
3901: this insn, or -1 if the matching has not yet been attempted.
3902:
3903: Such matching is never attempted and this field is not used on an insn
3904: whose pattern consists of a single @samp{use}, @samp{clobber},
3905: @samp{asm}, @samp{addr_vec} or @samp{addr_diff_vec} expression.
3906: @end table
3907:
3908: The @code{LOG_LINKS} field of an insn is a chain of @samp{insn_list}
3909: expressions. Each of these has two operands: the first is an insn,
3910: and the second is another @samp{insn_list} expression (the next one in
3911: the chain). The last @samp{insn_list} in the chain has a null pointer
3912: as second operand. The significant thing about the chain is which
1.1.1.2 root 3913: insns appear in it (as first operands of @samp{insn_list}
1.1 root 3914: expressions). Their order is not significant.
3915:
3916: The @code{REG_NOTES} field of an insn is a similar chain but of
1.1.1.2 root 3917: @samp{expr_list} expressions instead of @samp{insn_list}. There are four
3918: kinds of register notes, which are distinguished by the machine mode of the
3919: @samp{expr_list}, which a register note is really understood as being an
3920: @code{enum reg_note}. The first operand @var{op} of the @samp{expr_list}
3921: is data whose meaning depends on the kind of note. Here are the four
3922: kinds:
1.1 root 3923:
3924: @table @code
3925: @item REG_DEAD
1.1.1.2 root 3926: The register @var{op} dies in this insn; that is to say, altering the
3927: value immediately after this insn would not affect the future behavior
3928: of the program.
1.1 root 3929:
3930: @item REG_INC
1.1.1.2 root 3931: The register @var{op} is incremented (or decremented; at this level
1.1 root 3932: there is no distinction) by an embedded side effect inside this insn.
1.1.1.2 root 3933: This means it appears in a @code{POST_INC}, @code{PRE_INC},
3934: @code{POST_DEC} or @code{PRE_DEC} RTX.
1.1 root 3935:
1.1.1.2 root 3936: @item REG_EQUIV
3937: The register that is set by this insn will be equal to @var{op} at run
3938: time, and could validly be replaced in all its occurrences by
3939: @var{op}. (``Validly'' here refers to the data flow of the program;
3940: simple replacement may make some insns invalid.)
3941:
3942: The value which the insn explicitly copies into the register may look
3943: different from @var{op}, but they will be equal at run time.
3944:
3945: For example, when a constant is loaded into a register that is never
3946: assigned any other value, this kind of note is used.
3947:
3948: When a parameter is copied into a pseudo-register at entry to a function,
3949: a note of this kind records that the register is equivalent to the stack
3950: slot where the parameter was passed. Although in this case the register
3951: may be set by other insns, it is still valid to replace the register
3952: by the stack slot throughout the function.
3953:
3954: @item REG_EQUAL
3955: The register that is set by this insn will be equal to @var{op} at run
3956: time at the end of this insn (but not necessarily elsewhere in the
3957: function).
3958:
3959: The RTX @var{op} is typically an arithmetic expression. For example,
3960: when a sequence of insns such as a library call is used to perform an
3961: arithmetic operation, this kind of note is attached to the insn that
3962: produces or copies the final value. It tells the CSE pass how to
3963: think of that value.
3964:
3965: @item REG_RETVAL
3966: This insn copies the value of a library call, and @var{op} is the
3967: first insn that was generated to set up the arguments for the library
3968: call.
3969:
3970: Flow analysis uses this note to delete all of a library call whose
3971: result is dead.
1.1 root 3972:
3973: @item REG_WAS_0
1.1.1.2 root 3974: The register @var{op} contained zero before this insn. You can rely
1.1 root 3975: on this note if it is present; its absence implies nothing.
3976: @end table
3977:
3978: (The only difference between the expression codes @samp{insn_list} and
3979: @samp{expr_list} is that the first operand of an @samp{insn_list} is
3980: assumed to be an insn and is printed in debugging dumps as the insn's
3981: unique id; the first operand of an @samp{expr_list} is printed in the
3982: ordinary way as an expression.)
3983:
1.1.1.2 root 3984: @node Calls, Sharing, Insns, RTL
3985: @section RTL Representation of Function-Call Insns
3986:
3987: Insns that call subroutines have the RTL expression code @samp{call_insn}.
3988: These insns must satisfy special rules, and their bodies must use a special
3989: RTL expression code, @samp{call}.
3990:
3991: A @samp{call} expression has two operands, as follows:
3992:
3993: @example
3994: (call @var{nbytes} (mem:@var{fm} @var{addr}))
3995: @end example
3996:
3997: @noindent
3998: Here @var{nbytes} is an operand that represents the number of bytes of
3999: argument data being passed to the subroutine, @var{fm} is a machine mode
4000: (which must equal as the definition of the @code{FUNCTION_MODE} macro in
4001: the machine description) and @var{addr} represents the address of the
4002: subroutine.
4003:
4004: For a subroutine that returns no value, the @samp{call} RTX as shown above
4005: is the entire body of the insn.
4006:
4007: For a subroutine that returns a value whose mode is not @code{BLKmode},
4008: the value is returned in a hard register. If this register's number is
4009: @var{r}, then the body of the call insn looks like this:
4010:
4011: @example
4012: (set (reg:@var{m} @var{r})
4013: (call @var{nbytes} (mem:@var{fm} @var{addr})))
4014: @end example
4015:
4016: @noindent
4017: This RTL expression makes it clear (to the optimizer passes) that the
4018: appropriate register receives a useful value in this insn.
4019:
4020: Immediately after RTL generation, if the value of the subroutine is
4021: actually used, this call insn is always followed closely by an insn which
1.1.1.3 root 4022: refers to the register @var{r}. This remains true through all the
4023: optimizer passes until cross jumping occurs.
4024:
4025: The following insn has one of two forms. Either it copies the value into a
4026: pseudo-register, like this:
1.1.1.2 root 4027:
4028: @example
4029: (set (reg:@var{m} @var{p}) (reg:@var{m} @var{r}))
4030: @end example
4031:
4032: @noindent
4033: or (in the case where the calling function will simply return whatever
4034: value the call produced, and no operation is needed to do this):
4035:
4036: @example
4037: (use (reg:@var{m} @var{r}))
4038: @end example
4039:
4040: @noindent
1.1.1.3 root 4041: Between the call insn and this following insn there may intervene only a
1.1.1.2 root 4042: stack-adjustment insn (and perhaps some @samp{note} insns).
4043:
4044: When a subroutine returns a @code{BLKmode} value, it is handled by
4045: passing to the subroutine the address of a place to store the value.
4046: So the call insn itself does not ``return'' any value, and it has the
4047: same RTL form as a call that returns nothing.
4048:
4049: @node Sharing,, Calls, RTL
1.1 root 4050: @section Structure Sharing Assumptions
4051:
4052: The compiler assumes that certain kinds of RTL expressions are unique;
4053: there do not exist two distinct objects representing the same value.
4054: In other cases, it makes an opposite assumption: that no RTL expression
4055: object of a certain kind appears in more than one place in the
4056: containing structure.
4057:
4058: These assumptions refer to a single function; except for the RTL
4059: objects that describe global variables and external functions,
4060: no RTL objects are common to two functions.
4061:
4062: @itemize @bullet
4063: @item
4064: Each pseudo-register has only a single @samp{reg} object to represent it,
4065: and therefore only a single machine mode.
4066:
4067: @item
4068: For any symbolic label, there is only one @samp{symbol_ref} object
4069: referring to it.
4070:
4071: @item
4072: There is only one @samp{const_int} expression with value zero,
4073: and only one with value one.
4074:
4075: @item
4076: There is only one @samp{pc} expression.
4077:
4078: @item
4079: There is only one @samp{cc0} expression.
4080:
4081: @item
4082: There is only one @samp{const_double} expression with mode
4083: @code{SFmode} and value zero, and only one with mode @code{DFmode} and
4084: value zero.
4085:
4086: @item
1.1.1.2 root 4087: No @samp{label_ref} appears in more than one place in the RTL
4088: structure; in other words, it is safe to do a tree-walk of all the
4089: insns in the function and assume that each time a @samp{label_ref} is
4090: seen it is distinct from all others that are seen.
1.1 root 4091:
4092: @item
1.1.1.2 root 4093: Only one @samp{mem} object is normally created for each static
4094: variable or stack slot, so these objects are frequently shared in all
4095: the places they appear. However, separate but equal objects for these
4096: variables are occasionally made.
4097:
4098: @item
4099: No RTL object appears in more than one place in the RTL structure
4100: except as described above. Many passes of the compiler rely on this
4101: by assuming that they can modify RTL objects in place without unwanted
4102: side-effects on other insns.
4103:
4104: @item
4105: During initial RTL generation, shared structure is freely introduced.
4106: After all the RTL for a function has been generated, all shared
4107: structure is copied by @code{unshare_all_rtl} in @file{emit-rtl.c},
4108: after which the above rules are guaranteed to be followed.
4109:
4110: @item
4111: During the combiner pass, shared structure with an insn can exist
4112: temporarily. However, the shared structure is copied before the
4113: combiner is finished with the insn. This is done by
4114: @code{copy_substitutions} in @samp{combine.c}.
1.1 root 4115: @end itemize
4116:
4117: @node Machine Desc, Machine Macros, RTL, Top
4118: @chapter Machine Descriptions
4119:
4120: A machine description has two parts: a file of instruction patterns
4121: (@file{.md} file) and a C header file of macro definitions.
4122:
4123: The @file{.md} file for a target machine contains a pattern for each
4124: instruction that the target machine supports (or at least each instruction
4125: that is worth telling the compiler about). It may also contain comments.
4126: A semicolon causes the rest of the line to be a comment, unless the semicolon
4127: is inside a quoted string.
4128:
4129: See the next chapter for information on the C header file.
4130:
4131: @menu
4132: * Patterns:: How to write instruction patterns.
1.1.1.2 root 4133: * Example:: An explained example of a @samp{define_insn} pattern.
4134: * RTL Template:: The RTL template defines what insns match a pattern.
4135: * Output Template:: The output template says how to make assembler code
4136: from such an insn.
4137: * Output Statement:: For more generality, write C code to output
4138: the assembler code.
1.1 root 4139: * Constraints:: When not all operands are general operands.
4140: * Standard Names:: Names mark patterns to use for code generation.
1.1.1.2 root 4141: * Pattern Ordering:: When the order of patterns makes a difference.
1.1 root 4142: * Dependent Patterns:: Having one pattern may make you need another.
1.1.1.2 root 4143: * Jump Patterns:: Special considerations for patterns for jump insns.
4144: * Peephole Definitions::Defining machine-specific peephole optimizations.
4145: * Expander Definitions::Generating a sequence of several RTL insns
4146: for a standard operation.
1.1 root 4147: @end menu
4148:
4149: @node Patterns, Example, Machine Desc, Machine Desc
1.1.1.2 root 4150: @section Everything about Instruction Patterns
1.1 root 4151:
4152: Each instruction pattern contains an incomplete RTL expression, with pieces
4153: to be filled in later, operand constraints that restrict how the pieces can
4154: be filled in, and an output pattern or C code to generate the assembler
4155: output, all wrapped up in a @samp{define_insn} expression.
4156:
1.1.1.2 root 4157: A @samp{define_insn} is an RTL expression containing four operands:
1.1 root 4158:
4159: @enumerate
4160: @item
4161: An optional name. The presence of a name indicate that this instruction
4162: pattern can perform a certain standard job for the RTL-generation
4163: pass of the compiler. This pass knows certain names and will use
4164: the instruction patterns with those names, if the names are defined
4165: in the machine description.
4166:
4167: The absence of a name is indicated by writing an empty string
4168: where the name should go. Nameless instruction patterns are never
4169: used for generating RTL code, but they may permit several simpler insns
4170: to be combined later on.
4171:
4172: Names that are not thus known and used in RTL-generation have no
4173: effect; they are equivalent to no name at all.
4174:
4175: @item
1.1.1.2 root 4176: The @dfn{RTL template} (@pxref{RTL Template}) is a vector of
4177: incomplete RTL expressions which show what the instruction should look
4178: like. It is incomplete because it may contain @samp{match_operand}
4179: and @samp{match_dup} expressions that stand for operands of the
1.1 root 4180: instruction.
4181:
4182: If the vector has only one element, that element is what the
4183: instruction should look like. If the vector has multiple elements,
4184: then the instruction looks like a @samp{parallel} expression
4185: containing that many elements as described.
4186:
4187: @item
4188: A condition. This is a string which contains a C expression that is
4189: the final test to decide whether an insn body matches this pattern.
4190:
4191: For a named pattern, the condition (if present) may not depend on
4192: the data in the insn being matched, but only the target-machine-type
4193: flags. The compiler needs to test these conditions during
4194: initialization in order to learn exactly which named instructions are
4195: available in a particular run.
4196:
4197: For nameless patterns, the condition is applied only when matching an
4198: individual insn, and only after the insn has matched the pattern's
4199: recognition template. The insn's operands may be found in the vector
4200: @code{operands}.
4201:
4202: @item
1.1.1.2 root 4203: The @dfn{output template}: a string that says how to output matching
4204: insns as assembler code. @samp{%} in this string specifies where
4205: to substitute the value of an operand. @xref{Output Template}.
4206:
4207: When simple substitution isn't general enough, you can specify a piece
4208: of C code to compute the output. @xref{Output Statement}.
4209: @end enumerate
4210:
4211: @node Example, RTL Template, Patterns, Machine Desc
4212: @section Example of @samp{define_insn}
4213:
4214: Here is an actual example of an instruction pattern, for the 68000/68020.
4215:
4216: @example
4217: (define_insn "tstsi"
4218: [(set (cc0)
4219: (match_operand:SI 0 "general_operand" "rm"))]
4220: ""
4221: "*
4222: @{ if (TARGET_68020 || ! ADDRESS_REG_P (operands[0]))
4223: return \"tstl %0\";
4224: return \"cmpl #0,%0\"; @}")
4225: @end example
1.1 root 4226:
1.1.1.2 root 4227: This is an instruction that sets the condition codes based on the value of
4228: a general operand. It has no condition, so any insn whose RTL description
4229: has the form shown may be handled according to this pattern. The name
4230: @samp{tstsi} means ``test a @code{SImode} value'' and tells the RTL generation
4231: pass that, when it is necessary to test such a value, an insn to do so
4232: can be constructed using this pattern.
1.1 root 4233:
1.1.1.2 root 4234: The output control string is a piece of C code which chooses which
4235: output template to return based on the kind of operand and the specific
4236: type of CPU for which code is being generated.
1.1 root 4237:
1.1.1.2 root 4238: @samp{"rm"} is an operand constraint. Its meaning is explained below.
1.1 root 4239:
1.1.1.2 root 4240: @node RTL Template, Output Template, Example, Machine Desc
4241: @section RTL Template for Generating and Recognizing Insns
1.1 root 4242:
1.1.1.2 root 4243: The RTL template is used to define which insns match the particular pattern
4244: and how to find their operands. For named patterns, the RTL template also
4245: says how to construct an insn from specified operands.
4246:
4247: Construction involves substituting specified operands into a copy of the
4248: template. Matching involves determining the values that serve as the
4249: operands in the insn being matched. Both of these activities are
4250: controlled by special expression types that direct matching and
4251: substitution of the operands.
1.1 root 4252:
4253: @table @code
4254: @item (match_operand:@var{m} @var{n} @var{testfn} @var{constraint})
4255: This expression is a placeholder for operand number @var{n} of
4256: the insn. When constructing an insn, operand number @var{n}
4257: will be substituted at this point. When matching an insn, whatever
4258: appears at this position in the insn will be taken as operand
4259: number @var{n}; but it must satisfy @var{testfn} or this instruction
4260: pattern will not match at all.
4261:
4262: Operand numbers must be chosen consecutively counting from zero in
4263: each instruction pattern. There may be only one @samp{match_operand}
4264: expression in the pattern for each expression number, and they must
4265: appear in order of increasing expression number.
4266:
4267: @var{testfn} is a string that is the name of a C function that accepts
4268: two arguments, a machine mode and an expression. During matching,
4269: the function will be called with @var{m} as the mode argument
4270: and the putative operand as the other argument. If it returns zero,
4271: this instruction pattern fails to match. @var{testfn} may be
4272: an empty string; then it means no test is to be done on the operand.
4273:
4274: Most often, @var{testfn} is @code{"general_operand"}. It checks
4275: that the putative operand is either a constant, a register or a
4276: memory reference, and that it is valid for mode @var{m}.
4277:
1.1.1.2 root 4278: For an operand that must be a register, @var{testfn} should be
4279: @code{"register_operand"}. This prevents GNU CC from creating insns
4280: that have memory references in these operands, insns which would only
4281: have to be taken apart in the reload pass.
4282:
4283: For an operand that must be a constant, either @var{testfn} should be
4284: @code{"immediate_operand"}, or the instruction pattern's extra condition
4285: should check for constants, or both.
4286:
4287: @var{constraint} is explained later (@pxref{Constraints}).
1.1 root 4288:
4289: @item (match_dup @var{n})
4290: This expression is also a placeholder for operand number @var{n}.
4291: It is used when the operand needs to appear more than once in the
4292: insn.
4293:
4294: In construction, @samp{match_dup} behaves exactly like
1.1.1.2 root 4295: @samp{match_operand}: the operand is substituted into the insn being
1.1 root 4296: constructed. But in matching, @samp{match_dup} behaves differently.
4297: It assumes that operand number @var{n} has already been determined by
1.1.1.2 root 4298: a @samp{match_operand} appearing earlier in the recognition template,
1.1 root 4299: and it matches only an identical-looking expression.
4300:
4301: @item (address (match_operand:@var{m} @var{n} "address_operand" ""))
4302: This complex of expressions is a placeholder for an operand number
4303: @var{n} in a ``load address'' instruction: an operand which specifies
4304: a memory location in the usual way, but for which the actual operand
4305: value used is the address of the location, not the contents of the
4306: location.
4307:
4308: @samp{address} expressions never appear in RTL code, only in machine
4309: descriptions. And they are used only in machine descriptions that do
4310: not use the operand constraint feature. When operand constraints are
4311: in use, the letter @samp{p} in the constraint serves this purpose.
4312:
4313: @var{m} is the machine mode of the @emph{memory location being
4314: addressed}, not the machine mode of the address itself. That mode is
4315: always the same on a given target machine (it is @code{Pmode}, which
4316: normally is @code{SImode}), so there is no point in mentioning it;
4317: thus, no machine mode is written in the @samp{address} expression. If
4318: some day support is added for machines in which addresses of different
4319: kinds of objects appear differently or are used differently (such as
4320: the PDP-10), different formats would perhaps need different machine
4321: modes and these modes might be written in the @samp{address}
4322: expression.
4323: @end table
4324:
1.1.1.2 root 4325: @node Output Template, Output Statement, RTL Template, Machine Desc
4326: @section Output Templates and Operand Substitution
1.1 root 4327:
1.1.1.2 root 4328: The @dfn{output template} is a string which specifies how to output
4329: the assembler code for an instruction pattern. Most of the template
4330: is a fixed string which is output literally. The character @samp{%}
4331: is used to specify where to substitute an operand; it can also be
4332: used to identify places different variants of the assembler require
4333: different syntax.
4334:
4335: In the simplest case, a @samp{%} followed by a digit @var{n} says to output
4336: operand @var{n} at that point in the string.
4337:
4338: @samp{%} followed by a letter and a digit says to output an operand in an
4339: alternate fashion. Four letters have standard, built-in meanings described
4340: below. The machine description macro @code{PRINT_OPERAND} can define
4341: additional letters with nonstandard meanings.
4342:
4343: @samp{%c@var{digit}} can be used to substitute an operand that is a
4344: constant value without the syntax that normally indicates an immediate
4345: operand.
4346:
4347: @samp{%n@var{digit}} is like @samp{%c@var{digit}} except that the value of
4348: the constant is negated before printing.
4349:
4350: @samp{%a@var{digit}} can be used to substitute an operand as if it were a
4351: memory reference, with the actual operand treated as the address. This may
4352: be useful when outputting a ``load address'' instruction, because often the
4353: assembler syntax for such an instruction requires you to write the operand
4354: as if it were a memory reference.
4355:
4356: @samp{%l@var{digit}} is used to substitute a @code{label_ref} into a jump
4357: instruction.
4358:
4359: @samp{%} followed by a punctuation character specifies a substitution that
4360: does not use an operand. Only one case is standard: @samp{%%} outputs a
4361: @samp{%} into the assembler code. Other nonstandard cases can be
4362: defined in the @code{PRINT_OPERAND} macro.
4363:
4364: The template may generate multiple assembler instructions. Write the text
4365: for the instructions, with @samp{\;} between them.
4366:
4367: When the RTL contains two operand which are required by constraint to match
4368: each other, the output template must refer only to the lower-numbered operand.
4369: Matching operands are not always identical, and the rest of the compiler
4370: arranges to put the proper RTL expression for printing into the lower-numbered
4371: operand.
4372:
4373: One use of nonstandard letters or punctuation following @samp{%} is to
4374: distinguish between different assembler languages for the same machine; for
4375: example, Motorola syntax versus MIT syntax for the 68000. Motorola syntax
4376: requires periods in most opcode names, while MIT syntax does not. For
4377: example, the opcode @samp{movel} in MIT syntax is @samp{move.l} in Motorola
4378: syntax. The same file of patterns is used for both kinds of output syntax,
4379: but the character sequence @samp{%.} is used in each place where Motorola
4380: syntax wants a period. The @code{PRINT_OPERAND} macro for Motorola syntax
4381: defines the sequence to output a period; the macro for MIT syntax defines
4382: it to do nothing.
4383:
4384: @node Output Statement, Constraints, Output Template, Machine Desc
4385: @section C Statements for Generating Assembler Output
4386:
4387: Often a single fixed template string cannot produce correct and efficient
4388: assembler code for all the cases that are recognized by a single
4389: instruction pattern. For example, the opcodes may depend on the kinds of
4390: operands; or some unfortunate combinations of operands may require extra
4391: machine instructions.
4392:
4393: If the output control string starts with a @samp{*}, then it is not an
4394: output template but rather a piece of C program that should compute a
4395: template. It should execute a @code{return} statement to return the
4396: template-string you want. Most such templates use C string literals, which
4397: require doublequote characters to delimit them. To include these
4398: doublequote characters in the string, prefix each one with @samp{\}.
4399:
4400: The operands may be found in the array @code{operands}, whose C data type
4401: is @code{rtx []}.
4402:
4403: It is possible to output an assembler instruction and then go on to output
4404: or compute more of them, using the subroutine @code{output_asm_insn}. This
4405: receives two arguments: a template-string and a vector of operands. The
4406: vector may be @code{operands}, or it may be another array of @code{rtx}
4407: that you declare locally and initialize yourself.
4408:
4409: When an insn pattern has multiple alternatives in its constraints, often
4410: the appearance of the assembler code determined mostly by which alternative
4411: was matched. When this is so, the C code can test the variable
4412: @code{which_alternative}, which is the ordinal number of the alternative
4413: that was actually satisfied (0 for the first, 1 for the second alternative,
4414: etc.).
4415:
4416: For example, suppose there are two opcodes for storing zero, @samp{clrreg}
4417: for registers and @samp{clrmem} for memory locations. Here is how
4418: a pattern could use @code{which_alternative} to choose between them:
1.1 root 4419:
4420: @example
1.1.1.2 root 4421: (define_insn ""
4422: [(set (match_operand:SI 0 "general_operand" "r,m")
4423: (const_int 0))]
1.1 root 4424: ""
4425: "*
1.1.1.2 root 4426: return (which_alternative == 0
4427: ? \"clrreg %0\" : \"clrmem %0\");
4428: ")
1.1 root 4429: @end example
4430:
1.1.1.2 root 4431: @node Constraints, Standard Names, Output Statement, Machine Desc
1.1 root 4432: @section Operand Constraints
4433:
4434: Each @samp{match_operand} in an instruction pattern can specify a
4435: constraint for the type of operands allowed. Constraints can say whether
4436: an operand may be in a register, and which kinds of register; whether the
4437: operand can be a memory reference, and which kinds of address; whether the
4438: operand may be an immediate constant, and which possible values it may
4439: have. Constraints can also require two operands to match.
4440:
4441: @menu
4442: * Simple Constraints:: Basic use of constraints.
1.1.1.2 root 4443: * Multi-Alternative:: When an insn has two alternative constraint-patterns.
1.1 root 4444: * Class Preferences:: Constraints guide which hard register to put things in.
4445: * Modifiers:: More precise control over effects of constraints.
4446: * No Constraints:: Describing a clean machine without constraints.
4447: @end menu
4448:
4449: @node Simple Constraints, Multi-Alternative, Constraints, Constraints
4450: @subsection Simple Constraints
4451:
4452: The simplest kind of constraint is a string full of letters, each of
4453: which describes one kind of operand that is permitted. Here are
4454: the letters that are allowed:
4455:
1.1.1.2 root 4456: @table @asis
4457: @item @samp{m}
1.1 root 4458: A memory operand is allowed, with any kind of address that the machine
4459: supports in general.
4460:
1.1.1.2 root 4461: @item @samp{o}
4462: A memory operand is allowed, but only if the address is
4463: @dfn{offsetable}. This means that adding a small integer (actually,
4464: the width in bytes of the operand, as determined by its machine mode)
4465: may be added to the address and the result is also a valid memory
4466: address.
4467:
4468: For example, an address which is constant is offsetable; so is an
4469: address that is the sum of a register and a constant (as long as a
4470: slightly larger constant is also within the range of address-offsets
4471: supported by the machine); but an autoincrement or autodecrement
4472: address is not offsetable. More complicated indirect/indexed
4473: addresses may or may not be offsetable depending on the other
4474: addressing modes that the machine supports.
4475:
4476: Note that in an output operand which can be matched by another
4477: operand, the constraint letter @samp{o} is valid only when accompanied
4478: by both @samp{<} (if the target machine has predecrement addressing)
4479: and @samp{>} (if the target machine has preincrement addressing).
1.1 root 4480:
1.1.1.2 root 4481: @item @samp{<}
1.1 root 4482: A memory operand with autodecrement addressing (either predecrement or
4483: postdecrement) is allowed.
4484:
1.1.1.2 root 4485: @item @samp{>}
1.1 root 4486: A memory operand with autoincrement addressing (either preincrement or
4487: postincrement) is allowed.
4488:
1.1.1.2 root 4489: @item @samp{r}
4490: A register operand is allowed provided that it is in a general
4491: register.
1.1 root 4492:
1.1.1.2 root 4493: @item @samp{d}, @samp{a}, @samp{f}, @dots{}
1.1 root 4494: Other letters can be defined in machine-dependent fashion to stand for
4495: particular classes of registers. @samp{d}, @samp{a} and @samp{f} are
1.1.1.2 root 4496: defined on the 68000/68020 to stand for data, address and floating
4497: point registers.
1.1 root 4498:
1.1.1.2 root 4499: @item @samp{i}
1.1 root 4500: An immediate integer operand (one with constant value) is allowed.
1.1.1.2 root 4501: This includes symbolic constants whose values will be known only at
4502: assembly time.
1.1 root 4503:
1.1.1.2 root 4504: @item @samp{n}
4505: An immediate integer operand with a known numeric value is allowed.
4506: Many systems cannot support assembly-time constants for operands less
4507: than a word wide. Constraints for these operands should use @samp{n}
4508: rather than @samp{i}.
4509:
4510: @item @samp{I}, @samp{J}, @samp{K}, @dots{}
4511: Other letters in the range @samp{I} through @samp{M} may be defined in
4512: a machine-dependent fashion to permit immediate integer operands with
4513: explicit integer values in specified ranges. For example, on the
4514: 68000, @samp{I} is defined to stand for the range of values 1 to 8.
4515: This is the range permitted as a shift count in the shift
4516: instructions.
1.1 root 4517:
1.1.1.2 root 4518: @item @samp{F}
1.1 root 4519: An immediate floating operand (expression code @samp{const_double}) is
4520: allowed.
4521:
1.1.1.2 root 4522: @item @samp{G}, @samp{H}
1.1 root 4523: @samp{G} and @samp{H} may be defined in a machine-dependent fashion to
4524: permit immediate floating operands in particular ranges of values.
4525:
1.1.1.2 root 4526: @item @samp{s}
1.1 root 4527: An immediate integer operand whose value is not an explicit integer is
1.1.1.2 root 4528: allowed.
4529:
4530: This might appear strange; if an insn allows a constant operand with a
4531: value not known at compile time, it certainly must allow any known
1.1 root 4532: value. So why use @samp{s} instead of @samp{i}? Sometimes it allows
1.1.1.2 root 4533: better code to be generated.
4534:
4535: For example, on the 68000 in a fullword instruction it is possible to
4536: use an immediate operand; but if the immediate value is between -32
4537: and 31, better code results from loading the value into a register and
4538: using the register. This is because the load into the register can be
4539: done with a @samp{moveq} instruction. We arrange for this to happen
4540: by defining the letter @samp{K} to mean ``any integer outside the
4541: range -32 to 31'', and then specifying @samp{Ks} in the operand
1.1 root 4542: constraints.
4543:
1.1.1.2 root 4544: @item @samp{g}
1.1 root 4545: Any register, memory or immediate integer operand is allowed, except for
4546: registers that are not general registers.
4547:
1.1.1.2 root 4548: @item @samp{@var{n}} (a digit)
4549: An operand that matches operand number @var{n} is allowed.
1.1 root 4550: If a digit is used together with letters, the digit should come last.
4551:
1.1.1.2 root 4552: This is called a @dfn{matching constraint} and what it really means is
4553: that the assembler has only a single operand that fills two roles
4554: considered separate in the RTL insn. For example, an add insn has two
4555: input operands and one output operand in the RTL, but on most machines
4556: an add instruction really has only two operands, one of them an
4557: input-output operand.
4558:
4559: Matching constraints work only in circumstances like that add insn.
4560: More precisely, the matching constraint must appear in an input-only
4561: operand and the operand that it matches must be an output-only operand
4562: with a lower number.
4563:
4564: For operands to match in a particular case usually means that they
4565: are identical-looking RTL expressions. But in a few special cases
4566: specific kinds of dissimilarity are allowed. For example, @code{*x}
4567: as an input operand will match @code{*x++} as an output operand.
4568: For proper results in such cases, the output template should always
4569: use the output-operand's number when printing the operand.
4570:
4571: @item @samp{p}
1.1 root 4572: An operand that is a valid memory address is allowed. This is
4573: for ``load address'' and ``push address'' instructions.
4574:
4575: If @samp{p} is used in the constraint, the test-function in the
4576: @samp{match_operand} must be @code{address_operand}.
4577: @end table
4578:
4579: In order to have valid assembler code, each operand must satisfy
4580: its constraint. But a failure to do so does not prevent the pattern
4581: from applying to an insn. Instead, it directs the compiler to modify
1.1.1.2 root 4582: the code so that the constraint will be satisfied. Usually this is
1.1 root 4583: done by copying an operand into a register.
4584:
4585: Contrast, therefore, the two instruction patterns that follow:
4586:
4587: @example
4588: (define_insn ""
4589: [(set (match_operand:SI 0 "general_operand" "r")
4590: (plus:SI (match_dup 0)
4591: (match_operand:SI 1 "general_operand" "r")))]
4592: ""
4593: "@dots{}")
4594: @end example
4595:
4596: @noindent
4597: which has two operands, one of which must appear in two places, and
4598:
4599: @example
4600: (define_insn ""
4601: [(set (match_operand:SI 0 "general_operand" "r")
4602: (plus:SI (match_operand:SI 1 "general_operand" "0")
4603: (match_operand:SI 2 "general_operand" "r")))]
4604: ""
4605: "@dots{}")
4606: @end example
4607:
4608: @noindent
4609: which has three operands, two of which are required by a constraint to be
4610: identical. If we are considering an insn of the form
4611:
4612: @example
4613: (insn @var{n} @var{prev} @var{next}
4614: (set (reg:SI 3)
4615: (plus:SI (reg:SI 6) (reg:SI 109)))
4616: @dots{})
4617: @end example
4618:
4619: @noindent
4620: the first pattern would not apply at all, because this insn does not
4621: contain two identical subexpressions in the right place. The pattern would
4622: say, ``That does not look like an add instruction; try other patterns.''
4623: The second pattern would say, ``Yes, that's an add instruction, but there
4624: is something wrong with it.'' It would direct the reload pass of the
4625: compiler to generate additional insns to make the constraint true. The
4626: results might look like this:
4627:
4628: @example
4629: (insn @var{n2} @var{prev} @var{n}
4630: (set (reg:SI 3) (reg:SI 6))
4631: @dots{})
4632:
4633: (insn @var{n} @var{n2} @var{next}
4634: (set (reg:SI 3)
4635: (plus:SI (reg:SI 3) (reg:SI 109)))
4636: @dots{})
4637: @end example
4638:
4639: Because insns that don't fit the constraints are fixed up by loading
4640: operands into registers, every instruction pattern's constraints must
4641: permit the case where all the operands are in registers. It need not
4642: permit all classes of registers; the compiler knows how to copy registers
4643: into other registers of the proper class in order to make an instruction
4644: valid. But if no registers are permitted, the compiler will be stymied: it
4645: does not know how to save a register in memory in order to make an
4646: instruction valid. Instruction patterns that reject registers can be
4647: made valid by attaching a condition-expression that refuses to match
4648: an insn at all if the crucial operand is a register.
4649:
4650: @node Multi-Alternative, Class Preferences, Simple Constraints, Constraints
4651: @subsection Multiple Alternative Constraints
4652:
4653: Sometimes a single instruction has multiple alternative sets of possible
4654: operands. For example, on the 68000, a logical-or instruction can combine
4655: register or an immediate value into memory, or it can combine any kind of
4656: operand into a register; but it cannot combine one memory location into
4657: another.
4658:
4659: These constraints are represented as multiple alternatives. An alternative
4660: can be described by a series of letters for each operand. The overall
4661: constraint for an operand is made from the letters for this operand
4662: from the first alternative, a comma, the letters for this operand from
4663: the second alternative, a comma, and so on until the last alternative.
4664: Here is how it is done for fullword logical-or on the 68000:
4665:
4666: @example
4667: (define_insn "iorsi3"
4668: [(set (match_operand:SI 0 "general_operand" "=%m,d")
1.1.1.2 root 4669: (ior:SI (match_operand:SI 1 "general_operand" "0,0")
4670: (match_operand:SI 2 "general_operand" "dKs,dmKs")))]
1.1 root 4671: @dots{})
4672: @end example
4673:
4674: The first alternative has @samp{m} (memory) for operand 0, @samp{0} for
4675: operand 1 (meaning it must match operand 0), and @samp{dKs} for operand 2.
4676: The second alternative has @samp{d} (data register) for operand 0, @samp{0}
4677: for operand 1, and @samp{dmKs} for operand 2. The @samp{=} and @samp{%} in
4678: the constraint for operand 0 are not part of any alternative; their meaning
4679: is explained in the next section.
4680:
4681: If all the operands fit any one alternative, the instruction is valid.
4682: Otherwise, for each alternative, the compiler counts how many instructions
4683: must be added to copy the operands so that that alternative applies.
4684: The alternative requiring the least copying is chosen. If two alternatives
4685: need the same amount of copying, the one that comes first is chosen.
4686: These choices can be altered with the @samp{?} and @samp{!} characters:
4687:
4688: @table @samp
4689: @item ?
4690: Disparage slightly the alternative that the @samp{?} appears in,
4691: as a choice when no alternative applies exactly. The compiler regards
4692: this alternative as one unit more costly for each @samp{?} that appears
4693: in it.
4694:
4695: @item !
4696: Disparage severely the alternative that the @samp{!} appears in.
4697: When operands must be copied into registers, the compiler will
4698: never choose this alternative as the one to strive for.
4699: @end table
4700:
1.1.1.2 root 4701: When an insn pattern has multiple alternatives in its constraints,
4702: often the appearance of the assembler code determined mostly by which
4703: alternative was matched. When this is so, the C code for writing the
4704: assembler code can use the variable @code{which_alternative}, which is
4705: the ordinal number of the alternative that was actually satisfied
4706: (0 for the first, 1 for the second alternative, etc.). For example:
4707:
4708: @example
4709: (define_insn ""
4710: [(set (match_operand:SI 0 "general_operand" "r,m")
4711: (const_int 0))]
4712: ""
4713: "*
4714: return (which_alternative == 0
4715: ? \"clrreg %0\" : \"clrmem %0\");
4716: ")
4717: @end example
4718:
1.1 root 4719: @node Class Preferences, Modifiers, Multi-Alternative, Constraints
4720: @subsection Register Class Preferences
4721:
4722: The operand constraints have another function: they enable the compiler
4723: to decide which kind of hardware register a pseudo register is best
4724: allocated to. The compiler examines the constraints that apply to the
4725: insns that use the pseudo register, looking for the machine-dependent
4726: letters such as @samp{d} and @samp{a} that specify classes of registers.
4727: The pseudo register is put in whichever class gets the most ``votes''.
4728: The constraint letters @samp{g} and @samp{r} also vote: they vote in
4729: favor of a general register. The machine description says which registers
4730: are considered general.
4731:
4732: Of course, on some machines all registers are equivalent, and no register
4733: classes are defined. Then none of this complexity is relevant.
4734:
4735: @node Modifiers, No Constraints, Class Preferences, Constraints
4736: @subsection Constraint Modifier Characters
4737:
4738: @table @samp
4739: @item =
1.1.1.2 root 4740: Means that this operand is write-only for this instruction: the previous
4741: value is discarded and replaced by output data.
1.1 root 4742:
4743: @item +
4744: Means that this operand is both read and written by the instruction.
4745:
4746: When the compiler fixes up the operands to satisfy the constraints,
4747: it needs to know which operands are inputs to the instruction and
4748: which are outputs from it. @samp{=} identifies an output; @samp{+}
4749: identifies an operand that is both input and output; all other operands
4750: are assumed to be input only.
4751:
1.1.1.2 root 4752: @item &
4753: Means (in a particular alternative) that this operand is written
4754: before the instruction is finished using the input operands.
4755: Therefore, this operand may not lie in a register that is used as an
4756: input operand or as part of any memory address.
4757:
4758: @samp{&} applies only to the alternative in which it is written. In
4759: constraints with multiple alternatives, sometimes one alternative
4760: requires @samp{&} while others do not. See, for example, the
4761: @samp{movdf} insn of the 68000.
4762:
4763: @samp{&} does not obviate the need to write @samp{=}.
4764:
1.1 root 4765: @item %
1.1.1.2 root 4766: Declares the instruction to be commutative for this operand and the
4767: following operand. This means that the compiler may interchange the
4768: two operands if that is the cheapest way to make all operands fit the
4769: constraints. This is often used in patterns for addition instructions
4770: that really have only two operands: the result must go in one of the
4771: arguments. Here for example, is how the 68000 halfword-add
4772: instruction is defined:
4773:
4774: @example
4775: (define_insn "addhi3"
4776: [(set (match_operand:HI 0 "general_operand" "=m,r")
4777: (plus:HI (match_operand:HI 1 "general_operand" "%0,0")
4778: (match_operand:HI 2 "general_operand" "di,g")))]
4779: @dots{})
4780: @end example
4781:
4782: Note that in previous versions of GNU CC the @samp{%} constraint
4783: modifier always applied to operands 1 and 2 regardless of which
4784: operand it was written in. The usual custom was to write it in
4785: operand 0. Now it must be in operand 1 if the operands to be
4786: exchanged are 1 and 2.
1.1 root 4787:
4788: @item #
1.1.1.2 root 4789: Says that all following characters, up to the next comma, are to be
4790: ignored as a constraint. They are significant only for choosing
4791: register preferences.
1.1 root 4792:
4793: @item *
4794: Says that the following character should be ignored when choosing
1.1.1.2 root 4795: register preferences. @samp{*} has no effect on the meaning of the
4796: constraint as a constraint.
4797:
4798: Here is an example: the 68000 has an instruction to sign-extend a
4799: halfword in a data register, and can also sign-extend a value by
4800: copying it into an address register. While either kind of register is
4801: acceptable, the constraints on an address-register destination are
4802: less strict, so it is best if register allocation makes an address
4803: register its goal. Therefore, @samp{*} is used so that the @samp{d}
4804: constraint letter (for data register) is ignored when computing
4805: register preferences.
4806:
4807: @example
4808: (define_insn "extendhisi2"
4809: [(set (match_operand:SI 0 "general_operand" "=*d,a")
4810: (sign_extend:SI
4811: (match_operand:HI 1 "general_operand" "0,g")))]
4812: @dots{})
4813: @end example
1.1 root 4814: @end table
4815:
4816: @node No Constraints,, Modifiers, Constraints
4817: @subsection Not Using Constraints
4818:
4819: Some machines are so clean that operand constraints are not required. For
4820: example, on the Vax, an operand valid in one context is valid in any other
1.1.1.2 root 4821: context. On such a machine, every operand constraint would be @samp{g},
1.1 root 4822: excepting only operands of ``load address'' instructions which are
4823: written as if they referred to a memory location's contents but actual
1.1.1.2 root 4824: refer to its address. They would have constraint @samp{p}.
1.1 root 4825:
1.1.1.2 root 4826: For such machines, instead of writing @samp{g} and @samp{p} for all
1.1 root 4827: the constraints, you can choose to write a description with empty constraints.
4828: Then you write @samp{""} for the constraint in every @samp{match_operand}.
4829: Address operands are identified by writing an @samp{address} expression
4830: around the @samp{match_operand}, not by their constraints.
4831:
4832: When the machine description has just empty constraints, certain parts
4833: of compilation are skipped, making the compiler faster.
4834:
1.1.1.2 root 4835: @node Standard Names, Pattern Ordering, Constraints, Machine Desc
4836: @section Standard Names for Patterns Used in Generation
1.1 root 4837:
4838: Here is a table of the instruction names that are meaningful in the RTL
4839: generation pass of the compiler. Giving one of these names to an
4840: instruction pattern tells the RTL generation pass that it can use the
4841: pattern in to accomplish a certain task.
4842:
1.1.1.2 root 4843: @table @asis
4844: @item @samp{mov@var{m}}
1.1 root 4845: Here @var{m} is a two-letter machine mode name, in lower case. This
4846: instruction pattern moves data with that machine mode from operand 1 to
4847: operand 0. For example, @samp{movsi} moves full-word data.
4848:
4849: If operand 0 is a @samp{subreg} with mode @var{m} of a register whose
4850: natural mode is wider than @var{m}, the effect of this instruction is
4851: to store the specified value in the part of the register that corresponds
4852: to mode @var{m}. The effect on the rest of the register is undefined.
4853:
1.1.1.3 root 4854: This class of patterns is special in several ways. First of all, each
4855: of these names @emph{must} be defined, because there is no other way
4856: to copy a datum from one place to another.
4857:
4858: Second, these patterns are not used solely in the RTL generation pass.
4859: Even the reload pass can generate move insns to copy values from stack
4860: slots into temporary registers. When it does so, one of the operands
4861: is a hard register and the other is an operand that can have a reload.
4862:
4863: Therefore, when given such a pair of operands, the pattern must
4864: generate RTL which needs no temporary registers---no registers other
4865: than the operands. For example, if you support the pattern with a
4866: @code{define_expand}, then in such a case you mustn't call
4867: @code{force_reg} or any other such function which might generate new
4868: pseudo registers.
4869:
4870: This requirement exists even for subword modes on a RISC machine where
4871: fetching those modes from memory normally requires several insns and
4872: some temporary registers. Look in @file{spur.md} to see how the
4873: requirement is satisfied.
4874:
4875: The variety of operands that have reloads depends on the rest of the
4876: machine description, but typically on a RISC machine these can only be
4877: pseudo registers that did not get hard registers, while on other
4878: machines explicit memory references will get optional reloads.
4879:
1.1.1.2 root 4880: @item @samp{movstrict@var{m}}
1.1 root 4881: Like @samp{mov@var{m}} except that if operand 0 is a @samp{subreg}
4882: with mode @var{m} of a register whose natural mode is wider,
4883: the @samp{movstrict@var{m}} instruction is guaranteed not to alter
4884: any of the register except the part which belongs to mode @var{m}.
4885:
1.1.1.2 root 4886: @item @samp{add@var{m}3}
1.1 root 4887: Add operand 2 and operand 1, storing the result in operand 0. All operands
4888: must have mode @var{m}. This can be used even on two-address machines, by
4889: means of constraints requiring operands 1 and 0 to be the same location.
4890:
1.1.1.2 root 4891: @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}
1.1 root 4892: Similar, for other arithmetic operations.
4893:
1.1.1.2 root 4894: @item @samp{andcb@var{m}3}
1.1 root 4895: Bitwise logical-and operand 1 with the complement of operand 2
4896: and store the result in operand 0.
4897:
1.1.1.2 root 4898: @item @samp{mulhisi3}
1.1 root 4899: Multiply operands 1 and 2, which have mode @code{HImode}, and store
4900: a @code{SImode} product in operand 0.
4901:
1.1.1.2 root 4902: @item @samp{mulqihi3}, @samp{mulsidi3}
1.1 root 4903: Similar widening-multiplication instructions of other widths.
4904:
1.1.1.2 root 4905: @item @samp{umulqihi3}, @samp{umulhisi3}, @samp{umulsidi3}
1.1 root 4906: Similar widening-multiplication instructions that do unsigned
4907: multiplication.
4908:
1.1.1.2 root 4909: @item @samp{divmod@var{m}4}
1.1 root 4910: Signed division that produces both a quotient and a remainder.
4911: Operand 1 is divided by operand 2 to produce a quotient stored
4912: in operand 0 and a remainder stored in operand 3.
4913:
1.1.1.2 root 4914: @item @samp{udivmod@var{m}4}
1.1 root 4915: Similar, but does unsigned division.
4916:
1.1.1.2 root 4917: @item @samp{divmod@var{m}@var{n}4}
1.1 root 4918: Like @samp{divmod@var{m}4} except that only the dividend has mode
4919: @var{m}; the divisor, quotient and remainder have mode @var{n}.
4920: For example, the Vax has a @samp{divmoddisi4} instruction
4921: (but it is omitted from the machine description, because it
4922: is so slow that it is faster to compute remainders by the
4923: circumlocution that the compiler will use if this instruction is
4924: not available).
4925:
1.1.1.2 root 4926: @item @samp{ashl@var{m}3}
1.1 root 4927: Arithmetic-shift operand 1 left by a number of bits specified by
4928: operand 2, and store the result in operand 0. Operand 2 has
4929: mode @code{SImode}, not mode @var{m}.
4930:
1.1.1.2 root 4931: @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}
1.1 root 4932: Other shift and rotate instructions.
4933:
1.1.1.2 root 4934: Logical and arithmetic left shift are the same. Machines that do not
4935: allow negative shift counts often have only one instruction for
4936: shifting left. On such machines, you should define a pattern named
4937: @samp{ashl@var{m}3} and leave @samp{lshl@var{m}3} undefined.
4938:
4939: @item @samp{neg@var{m}2}
1.1 root 4940: Negate operand 1 and store the result in operand 0.
4941:
1.1.1.2 root 4942: @item @samp{abs@var{m}2}
1.1 root 4943: Store the absolute value of operand 1 into operand 0.
4944:
1.1.1.2 root 4945: @item @samp{sqrt@var{m}2}
1.1 root 4946: Store the square root of operand 1 into operand 0.
4947:
1.1.1.2 root 4948: @item @samp{ffs@var{m}2}
4949: Store into operand 0 one plus the index of the least significant 1-bit
4950: of operand 1. If operand 1 is zero, store zero. @var{m} is the mode
4951: of operand 0; operand 1's mode is specified by the instruction
4952: pattern, and the compiler will convert the operand to that mode before
4953: generating the instruction.
4954:
4955: @item @samp{one_cmpl@var{m}2}
1.1 root 4956: Store the bitwise-complement of operand 1 into operand 0.
4957:
1.1.1.2 root 4958: @item @samp{cmp@var{m}}
1.1 root 4959: Compare operand 0 and operand 1, and set the condition codes.
1.1.1.2 root 4960: The RTL pattern should look like this:
1.1 root 4961:
1.1.1.2 root 4962: @example
4963: (set (cc0) (minus (match_operand:@var{m} 0 @dots{})
4964: (match_operand:@var{m} 1 @dots{})))
4965: @end example
4966:
4967: Each such definition in the machine description, for integer mode
4968: @var{m}, must have a corresponding @samp{tst@var{m}} pattern, because
4969: optimization can simplify the compare into a test when operand 1 is
4970: zero.
4971:
4972: @item @samp{tst@var{m}}
1.1 root 4973: Compare operand 0 against zero, and set the condition codes.
1.1.1.2 root 4974: The RTL pattern should look like this:
1.1 root 4975:
1.1.1.2 root 4976: @example
4977: (set (cc0) (match_operand:@var{m} 0 @dots{}))
4978: @end example
4979:
4980: @item @samp{movstr@var{m}}
1.1 root 4981: Block move instruction. The addresses of the destination and source
4982: strings are the first two operands, and both are in mode @code{Pmode}.
4983: The number of bytes to move is the third operand, in mode @var{m}.
4984:
1.1.1.2 root 4985: @item @samp{cmpstr@var{m}}
1.1 root 4986: Block compare instruction, with operands like @samp{movstr@var{m}}
4987: except that the two memory blocks are compared byte by byte
4988: in lexicographic order. The effect of the instruction is to set
4989: the condition codes.
4990:
1.1.1.2 root 4991: @item @samp{float@var{m}@var{n}2}
1.1 root 4992: Convert operand 1 (valid for fixed point mode @var{m}) to floating
4993: point mode @var{n} and store in operand 0 (which has mode @var{n}).
4994:
1.1.1.2 root 4995: @item @samp{fix@var{m}@var{n}2}
4996: Convert operand 1 (valid for floating point mode @var{m}) to fixed
4997: point mode @var{n} as a signed number and store in operand 0 (which
4998: has mode @var{n}). This instruction's result is defined only when
4999: the value of operand 1 is an integer.
5000:
5001: @item @samp{fixuns@var{m}@var{n}2}
5002: Convert operand 1 (valid for floating point mode @var{m}) to fixed
5003: point mode @var{n} as an unsigned number and store in operand 0 (which
5004: has mode @var{n}). This instruction's result is defined only when the
5005: value of operand 1 is an integer.
5006:
5007: @item @samp{ftrunc@var{m}2}
5008: Convert operand 1 (valid for floating point mode @var{m}) to an
5009: integer value, still represented in floating point mode @var{m}, and
5010: store it in operand 0 (valid for floating point mode @var{m}).
5011:
5012: @item @samp{fix_trunc@var{m}@var{n}2}
5013: Like @samp{fix@var{m}@var{n}2} but works for any floating point value
5014: of mode @var{m} by converting the value to an integer.
5015:
5016: @item @samp{fixuns_trunc@var{m}@var{n}2}
5017: Like @samp{fixuns@var{m}@var{n}2} but works for any floating point
5018: value of mode @var{m} by converting the value to an integer.
5019:
5020: @item @samp{trunc@var{m}@var{n}}
1.1 root 5021: Truncate operand 1 (valid for mode @var{m}) to mode @var{n} and
5022: store in operand 0 (which has mode @var{n}). Both modes must be fixed
5023: point or both floating point.
5024:
1.1.1.2 root 5025: @item @samp{extend@var{m}@var{n}}
1.1 root 5026: Sign-extend operand 1 (valid for mode @var{m}) to mode @var{n} and
5027: store in operand 0 (which has mode @var{n}). Both modes must be fixed
5028: point or both floating point.
5029:
1.1.1.2 root 5030: @item @samp{zero_extend@var{m}@var{n}}
1.1 root 5031: Zero-extend operand 1 (valid for mode @var{m}) to mode @var{n} and
5032: store in operand 0 (which has mode @var{n}). Both modes must be fixed
5033: point.
5034:
1.1.1.2 root 5035: @item @samp{extv}
1.1 root 5036: Extract a bit-field from operand 1 (a register or memory operand),
5037: where operand 2 specifies the width in bits and operand 3 the starting
5038: bit, and store it in operand 0. Operand 0 must have @code{Simode}.
5039: Operand 1 may have mode @code{QImode} or @code{SImode}; often
5040: @code{SImode} is allowed only for registers. Operands 2 and 3 must be
5041: valid for @code{SImode}.
5042:
5043: The RTL generation pass generates this instruction only with constants
5044: for operands 2 and 3.
5045:
5046: The bit-field value is sign-extended to a full word integer
5047: before it is stored in operand 0.
5048:
1.1.1.2 root 5049: @item @samp{extzv}
1.1 root 5050: Like @samp{extv} except that the bit-field value is zero-extended.
5051:
1.1.1.2 root 5052: @item @samp{insv}
1.1 root 5053: Store operand 3 (which must be valid for @code{SImode}) into a
5054: bit-field in operand 0, where operand 1 specifies the width in bits
5055: and operand 2 the starting bit. Operand 0 may have mode @code{QImode}
5056: or @code{SImode}; often @code{SImode} is allowed only for registers.
5057: Operands 1 and 2 must be valid for @code{SImode}.
5058:
5059: The RTL generation pass generates this instruction only with constants
5060: for operands 1 and 2.
5061:
1.1.1.2 root 5062: @item @samp{s@var{cond}}
5063: Store zero or nonzero in the operand according to the condition codes.
5064: Value stored is nonzero iff the condition @var{cond} is true.
5065: @var{cond} is the name of a comparison operation expression code, such
1.1 root 5066: as @samp{eq}, @samp{lt} or @samp{leu}.
5067:
1.1.1.2 root 5068: You specify the mode that the operand must have when you write the
5069: @code{match_operand} expression. The compiler automatically sees
5070: which mode you have used and supplies an operand of that mode.
5071:
5072: The value stored for a true condition must have 1 as its low bit.
5073: Otherwise the instruction is not suitable and must be omitted from the
5074: machine description. You must tell the compiler exactly which value
5075: is stored by defining the macro @code{STORE_FLAG_VALUE}.
5076:
5077: @item @samp{b@var{cond}}
1.1 root 5078: Conditional branch instruction. Operand 0 is a @samp{label_ref}
5079: that refers to the label to jump to. Jump if the condition codes
5080: meet condition @var{cond}.
5081:
1.1.1.2 root 5082: @item @samp{call}
5083: Subroutine call instruction. Operand 1 is the number of bytes of
5084: arguments pushed (in mode @code{SImode}), and operand 0 is the
5085: function to call. Operand 0 should be a @samp{mem} RTX whose address
5086: is the address of the function.
1.1 root 5087:
1.1.1.2 root 5088: @item @samp{return}
1.1 root 5089: Subroutine return instruction. This instruction pattern name should be
5090: defined only if a single instruction can do all the work of returning
5091: from a function.
5092:
1.1.1.3 root 5093: @item @samp{casesi}
5094: Instruction to jump through a dispatch table, including bounds checking.
5095: This instruction takes five operands:
5096:
5097: @enumerate
5098: @item
5099: The index to dispatch on, which has mode @code{SImode}.
5100:
5101: @item
5102: The lower bound for indices in the table, an integer constant.
5103:
5104: @item
5105: The upper bound for indices in the table, an integer constant.
5106:
5107: @item
5108: A label to jump to if the index has a value outside the bounds.
5109: (If the machine-description macro @code{CASE_DROPS_THROUGH} is defined,
5110: then an out-of-bounds index drops through to the code following
5111: the jump table instead of jumping to this label. In that case,
5112: this label is not actually used by the @samp{casesi} instruction,
5113: but it is always provided as an operand.)
5114:
5115: @item
5116: A label that precedes the table itself.
5117: @end enumerate
5118:
5119: The table is a @samp{addr_vec} or @samp{addr_diff_vec} inside of a
5120: @samp{jump_insn}. The number of elements in the table is one plus the
5121: difference between the upper bound and the lower bound.
5122:
1.1.1.2 root 5123: @item @samp{tablejump}
1.1.1.3 root 5124: Instruction to jump to a variable address. This is a low-level
5125: capability which can be used to implement a dispatch table when there
5126: is no @samp{casesi} pattern.
5127:
5128: This pattern requires two operands: the address or offset, and a label
5129: which should immediately precede the jump table. If the macro
5130: @code{CASE_VECTOR_PC_RELATIVE} is defined then the first operand is an
5131: absolute address to jump to; otherwise, it is an offset which counts
5132: from the address of the table.
5133:
5134: The @samp{tablejump} insn is always the last insn before the jump
5135: table it uses. Its assembler code normally has no need to use the
5136: second operand, but you should incorporate it in the RTL pattern so
5137: that the jump optimizer will not delete the table as unreachable code.
1.1 root 5138: @end table
5139:
1.1.1.2 root 5140: @node Pattern Ordering, Dependent Patterns, Standard Names, Machine Desc
5141: @section When the Order of Patterns Matters
5142:
5143: Sometimes an insn can match more than one instruction pattern. Then the
5144: pattern that appears first in the machine description is the one used.
5145: Therefore, more specific patterns (patterns that will match fewer things)
5146: and faster instructions (those that will produce better code when they
5147: do match) should usually go first in the description.
5148:
5149: In some cases the effect of ordering the patterns can be used to hide
5150: a pattern when it is not valid. For example, the 68000 has an
5151: instruction for converting a fullword to floating point and another
5152: for converting a byte to floating point. An instruction converting
5153: an integer to floating point could match either one. We put the
5154: pattern to convert the fullword first to make sure that one will
5155: be used rather than the other. (Otherwise a large integer might
5156: be generated as a single-byte immediate quantity, which would not work.)
5157: Instead of using this pattern ordering it would be possible to make the
5158: pattern for convert-a-byte smart enough to deal properly with any
5159: constant value.
5160:
5161: @node Dependent Patterns, Jump Patterns, Pattern Ordering, Machine Desc
5162: @section Interdependence of Patterns
1.1 root 5163:
5164: Every machine description must have a named pattern for each of the
5165: conditional branch names @samp{b@var{cond}}. The recognition template
5166: must always have the form
5167:
5168: @example
5169: (set (pc)
5170: (if_then_else (@var{cond} (cc0) (const_int 0))
5171: (label_ref (match_operand 0 "" ""))
5172: (pc)))
5173: @end example
5174:
5175: @noindent
5176: In addition, every machine description must have an anonymous pattern
5177: for each of the possible reverse-conditional branches. These patterns
5178: look like
5179:
5180: @example
5181: (set (pc)
5182: (if_then_else (@var{cond} (cc0) (const_int 0))
5183: (pc)
5184: (label_ref (match_operand 0 "" ""))))
5185: @end example
5186:
5187: @noindent
5188: They are necessary because jump optimization can turn direct-conditional
5189: branches into reverse-conditional branches.
5190:
5191: The compiler does more with RTL than just create it from patterns
5192: and recognize the patterns: it can perform arithmetic expression codes
5193: when constant values for their operands can be determined. As a result,
5194: sometimes having one pattern can require other patterns. For example, the
5195: Vax has no `and' instruction, but it has `and not' instructions. Here
5196: is the definition of one of them:
5197:
5198: @example
5199: (define_insn "andcbsi2"
5200: [(set (match_operand:SI 0 "general_operand" "")
5201: (and:SI (match_dup 0)
5202: (not:SI (match_operand:SI
5203: 1 "general_operand" ""))))]
5204: ""
5205: "bicl2 %1,%0")
5206: @end example
5207:
5208: @noindent
5209: If operand 1 is an explicit integer constant, an instruction constructed
1.1.1.2 root 5210: using that pattern can be simplified into an `and' like this:
1.1 root 5211:
5212: @example
5213: (set (reg:SI 41)
5214: (and:SI (reg:SI 41)
5215: (const_int 0xffff7fff)))
5216: @end example
5217:
5218: @noindent
5219: (where the integer constant is the one's complement of what
5220: appeared in the original instruction).
5221:
5222: To avoid a fatal error, the compiler must have a pattern that recognizes
5223: such an instruction. Here is what is used:
5224:
5225: @example
5226: (define_insn ""
5227: [(set (match_operand:SI 0 "general_operand" "")
5228: (and:SI (match_dup 0)
5229: (match_operand:SI 1 "general_operand" "")))]
5230: "GET_CODE (operands[1]) == CONST_INT"
5231: "*
1.1.1.2 root 5232: @{ operands[1]
1.1 root 5233: = gen_rtx (CONST_INT, VOIDmode, ~INTVAL (operands[1]));
5234: return \"bicl2 %1,%0\";
1.1.1.2 root 5235: @}")
1.1 root 5236: @end example
5237:
5238: @noindent
5239: Whereas a pattern to match a general `and' instruction is impossible to
5240: support on the Vax, this pattern is possible because it matches only a
5241: constant second argument: a special case that can be output as an `and not'
5242: instruction.
5243:
1.1.1.2 root 5244: A ``compare'' instruction whose RTL looks like this:
5245:
5246: @example
5247: (set (cc0) (minus @var{operand} (const_int 0)))
5248: @end example
5249:
5250: @noindent
5251: may be simplified by optimization into a ``test'' like this:
5252:
5253: @example
5254: (set (cc0) @var{operand})
5255: @end example
5256:
5257: @noindent
5258: So in the machine description, each ``compare'' pattern for an integer
5259: mode must have a corresponding ``test'' pattern that will match the
5260: result of such simplification.
5261:
5262: In some cases machines support instructions identical except for the
5263: machine mode of one or more operands. For example, there may be
5264: ``sign-extend halfword'' and ``sign-extend byte'' instructions whose
5265: patterns are
5266:
5267: @example
5268: (set (match_operand:SI 0 @dots{})
5269: (extend:SI (match_operand:HI 1 @dots{})))
5270:
5271: (set (match_operand:SI 0 @dots{})
5272: (extend:SI (match_operand:QI 1 @dots{})))
5273: @end example
5274:
5275: @noindent
5276: Constant integers do not specify a machine mode, so an instruction to
5277: extend a constant value could match either pattern. The pattern it
5278: actually will match is the one that appears first in the file. For correct
5279: results, this must be the one for the widest possible mode (@code{HImode},
5280: here). If the pattern matches the @code{QImode} instruction, the results
5281: will be incorrect if the constant value does not actually fit that mode.
5282:
5283: Such instructions to extend constants are rarely generated because they are
5284: optimized away, but they do occasionally happen in nonoptimized
5285: compilations.
5286:
5287: @node Jump Patterns, Peephole Definitions, Dependent Patterns, Machine Desc
5288: @section Defining Jump Instruction Patterns
5289:
5290: GNU CC assumes that the machine has a condition code. A comparison insn
5291: sets the condition code, recording the results of both signed and unsigned
5292: comparison of the given operands. A separate branch insn tests the
5293: condition code and branches or not according its value. The branch insns
5294: come in distinct signed and unsigned flavors. Many common machines, such
5295: as the Vax, the 68000 and the 32000, work this way.
5296:
5297: Some machines have distinct signed and unsigned compare instructions, and
5298: only one set of conditional branch instructions. The easiest way to handle
5299: these machines is to treat them just like the others until the final stage
5300: where assembly code is written. At this time, when outputting code for the
5301: compare instruction, peek ahead at the following branch using
5302: @code{NEXT_INSN (insn)}. (The variable @code{insn} refers to the insn
5303: being output, in the output-writing code in an instruction pattern.) If
5304: the RTL says that is an unsigned branch, output an unsigned compare;
5305: otherwise output a signed compare. When the branch itself is output, you
5306: can treat signed and unsigned branches identically.
5307:
5308: The reason you can do this is that GNU CC always generates a pair of
5309: consecutive RTL insns, one to set the condition code and one to test it,
5310: and keeps the pair inviolate until the end.
5311:
5312: To go with this technique, you must define the machine-description macro
5313: @code{NOTICE_UPDATE_CC} to do @code{CC_STATUS_INIT}; in other words, no
5314: compare instruction is superfluous.
5315:
5316: Some machines have compare-and-branch instructions and no condition code.
5317: A similar technique works for them. When it is time to ``output'' a
5318: compare instruction, record its operands in two static variables. When
5319: outputting the branch-on-condition-code instruction that follows, actually
5320: output a compare-and-branch instruction that uses the remembered operands.
5321:
5322: It also works to define patterns for compare-and-branch instructions.
5323: In optimizing compilation, the pair of compare and branch instructions
5324: will be combined accoprding to these patterns. But this does not happen
5325: if optimization is not requested. So you must use one of the solutions
5326: above in addition to any special patterns you define.
5327:
5328: @node Peephole Definitions, Expander Definitions, Jump Patterns, Machine Desc
5329: @section Defining Machine-Specific Peephole Optimizers
5330:
5331: In addition to instruction patterns the @file{md} file may contain
5332: definitions of machine-specific peephole optimizations.
5333:
5334: The combiner does not notice certain peephole optimizations when the data
5335: flow in the program does not suggest that it should try them. For example,
5336: sometimes two consecutive insns related in purpose can be combined even
5337: though the second one does not appear to use a register computed in the
5338: first one. A machine-specific peephole optimizer can detect such
5339: opportunities.
5340:
5341: A definition looks like this:
5342:
5343: @example
5344: (define_peephole
5345: [@var{insn-pattern-1}
5346: @var{insn-pattern-2}
5347: @dots{}]
5348: "@var{condition}"
5349: "@var{template}")
5350: @end example
5351:
5352: In this skeleton, @var{insn-pattern-1} and so on are patterns to match
5353: consecutive instructions. The optimization applies to a sequence of
5354: instructions when @var{insn-pattern-1} matches the first one,
5355: @var{insn-pattern-2} matches the next, and so on.@refill
5356:
5357: @var{insn-pattern-1} and so on look @emph{almost} like the second operand
5358: of @code{define_insn}. There is one important difference: this pattern is
5359: an RTX, not a vector. If the @code{define_insn} pattern would be a vector
5360: of one element, the @var{insn-pattern} should be just that element, no
5361: vector. If the @code{define_insn} pattern would have multiple elements
5362: then the @var{insn-pattern} must place the vector inside an explicit
5363: @code{parallel} RTX.@refill
5364:
5365: The operands of the instructions are matched with @code{match_operands} and
5366: @code{match_dup}, as usual). What is not usual is that the operand numbers
5367: apply to all the instruction patterns in the definition. So, you can check
5368: for identical operands in two instructions by using @code{match_operand}
5369: in one instruction and @code{match_dup} in the other.
5370:
5371: The operand constraints used in @code{match_operand} patterns do not have
5372: any direct effect on the applicability of the optimization, but they will
5373: be validated afterward, so write constraints that are sure to fit whenever
5374: the optimization is applied. It is safe to use @code{"g"} for each
5375: operand.
5376:
5377: Once a sequence of instructions matches the patterns, the @var{condition}
5378: is checked. This is a C expression which makes the final decision whether
5379: to perform the optimization (do so if the expression is nonzero). If
5380: @var{condition} is omitted (in other words, the string is empty) then the
5381: optimization is applied to every sequence of instructions that matches the
5382: patterns.
5383:
5384: The defined peephole optimizations are applied after register allocation is
5385: complete. Therefore, the optimizer can check which operands have ended up
5386: in which kinds of registers, just by looking at the operands.
5387:
5388: The way to refer to the operands in @var{condition} is to write
5389: @code{operands[@var{i}]} for operand number @var{i} (as matched by
5390: @code{(match_operand @var{i} @dots{})}). Use the variable @code{insn} to
5391: refer to the last of the insns being matched; use @code{PREV_INSN} to find
5392: the preceding insns (but be careful to skip over any @samp{note} insns that
5393: intervene).@refill
5394:
5395: When optimizing computations with intermediate results, you can use
5396: @var{condition} to match only when the intermediate results are not used
5397: elsewhere. Use the C expression @code{dead_or_set_p (@var{insn},
5398: @var{op})}, where @var{insn} is the insn in which you expect the value to
5399: be used for the last time (from the value of @code{insn}, together with use
5400: of @code{PREV_INSN}), and @var{op} is the intermediate value (from
5401: @code{operands[@var{i}]}).@refill
5402:
5403: Applying the optimization means replacing the sequence of instructions with
5404: one new instruction. The @var{template} controls ultimate output of
5405: assembler code for this combined instruction. It works exactly like the
5406: template of a @code{define_insn}. Operand numbers in this template are the
5407: same ones used in matching the original sequence of instructions.
5408:
5409: The result of a defined peephole optimizer does not need to match any of
5410: the instruction patterns, and it does not have an opportunity to match
5411: them. The peephole optimizer definition itself serves as the instruction
5412: pattern to control how the instruction is output.
5413:
5414: Defined peephole optimizers are run in the last jump optimization pass, so
5415: the instructions they produce are never combined or rearranged
5416: automatically in any way.
5417:
5418: Here is an example, taken from the 68000 machine description:
5419:
5420: @example
5421: (define_peephole
5422: [(set (reg:SI 15) (plus:SI (reg:SI 15) (const_int 4)))
5423: (set (match_operand:DF 0 "register_operand" "f")
5424: (match_operand:DF 1 "register_operand" "ad"))]
5425: "FP_REG_P (operands[0]) && ! FP_REG_P (operands[1])"
5426: "*
5427: @{
5428: rtx xoperands[2];
5429: xoperands[1] = gen_rtx (REG, SImode, REGNO (operands[1]) + 1);
5430: #ifdef MOTOROLA
5431: output_asm_insn (\"move.l %1,(sp)\", xoperands);
5432: output_asm_insn (\"move.l %1,-(sp)\", operands);
5433: return \"fmove.d (sp)+,%0\";
5434: #else
5435: output_asm_insn (\"movel %1,sp@@\", xoperands);
5436: output_asm_insn (\"movel %1,sp@@-\", operands);
5437: return \"fmoved sp@@+,%0\";
5438: #endif
5439: @}
5440: ")
5441: @end example
5442:
5443: The effect of this optimization is to change
5444:
5445: @example
5446: jbsr _foobar
5447: addql #4,sp
5448: movel d1,sp@@-
5449: movel d0,sp@@-
5450: fmoved sp@@+,fp0
5451: @end example
5452:
5453: @noindent
5454: into
5455:
5456: @example
5457: jbsr _foobar
5458: movel d1,sp@@
5459: movel d0,sp@@-
5460: fmoved sp@@+,fp0
5461: @end example
5462:
5463: @node Expander Definitions,, Peephole Definitions, Machine Desc
5464: @section Defining RTL Sequences for Code Generation
5465:
5466: On some target machines, some standard pattern names for RTL generation
5467: cannot be handled with single insn, but a sequence of RTL insns can
5468: represent them. For these target machines, you can write a
5469: @samp{define_expand} to specify how to generate the sequence of RTL.
5470:
5471: A @samp{define_expand} is an RTL expression that looks almost like a
5472: @samp{define_insn}; but, unlike the latter, a @samp{define_expand} is used
5473: only for RTL generation and it can produce more than one RTL insn.
5474:
5475: A @samp{define_expand} RTX has four operands:
5476:
5477: @itemize @bullet
5478: @item
5479: The name. Each @samp{define_expand} must have a name, since the only
5480: use for it is to refer to it by name.
5481:
5482: @item
5483: The RTL template. This is just like the RTL template for a
5484: @samp{define_peephole} in that it is a vector of RTL expressions
5485: each being one insn.
5486:
5487: @item
5488: The condition, a string containing a C expression. This expression is
5489: used to express how the availability of this pattern depends on
5490: subclasses of target machine, selected by command-line options when
5491: GNU CC is run. This is just like the condition of a
5492: @samp{define_insn} that has a standard name.
5493:
5494: @item
5495: The preparation statements, a string containing zero or more C
5496: statements which are to be executed before RTL code is generated from
5497: the RTL template.
5498:
5499: Usually these statements prepare temporary registers for use as
5500: internal operands in the RTL template, but they can also generate RTL
5501: insns directly by calling routines such as @samp{emit_insn}, etc.
5502: Any such insns precede the ones that come from the RTL template.
5503: @end itemize
5504:
5505: The RTL template, in addition to controlling generation of RTL insns,
5506: also describes the operands that need to be specified when this pattern
5507: is used. In particular, it gives a predicate for each operand.
5508:
5509: A true operand, which need to be specified in order to generate RTL from
5510: the pattern, should be described with a @samp{match_operand} in its first
5511: occurrence in the RTL template. This enters information on the operand's
5512: predicate into the tables that record such things. GNU CC uses the
5513: information to preload the operand into a register if that is required for
5514: valid RTL code. If the operand is referred to more than once, subsequent
5515: references should use @samp{match_dup}.
5516:
5517: The RTL template may also refer to internal ``operands'' which are
5518: temporary registers or labels used only within the sequence made by the
5519: @samp{define_expand}. Internal operands are substituted into the RTL
5520: template with @samp{match_dup}, never with @samp{match_operand}. The
5521: values of the internal operands are not passed in as arguments by the
5522: compiler when it requests use of this pattern. Instead, they are computed
5523: within the pattern, in the preparation statements. These statements
5524: compute the values and store them into the appropriate elements of
5525: @code{operands} so that @samp{match_dup} can find them.
5526:
5527: There are two special macros defined for use in the preparation statements:
5528: @code{DONE} and @code{FAIL}. Use them with a following semicolon,
5529: as a statement.
5530:
5531: @table @code
5532: @item DONE
5533: Use the @code{DONE} macro to end RTL generation for the pattern. The
5534: only RTL insns resulting from the pattern on this occasion will be
5535: those already emitted by explicit calls to @code{emit_insn} within the
5536: preparation statements; the RTL template will not be generated.
5537:
5538: @item FAIL
5539: Make the pattern fail on this occasion. When a pattern fails, it means
5540: that the pattern was not truly available. The calling routines in the
5541: compiler will try other strategies for code generation using other patterns.
5542:
5543: Failure is currently supported only for binary operations (addition,
5544: multiplication, shifting, etc.).
5545:
5546: Do not emit any insns explicitly with @code{emit_insn} before failing.
5547: @end table
5548:
5549: Here is an example, the definition of left-shift for the SPUR chip:
5550:
5551: @example
5552: (define_expand "ashlsi3"
5553: [(set (match_operand:SI 0 "register_operand" "")
5554: (ashift:SI
5555: (match_operand:SI 1 "register_operand" "")
5556: (match_operand:SI 2 "nonmemory_operand" "")))]
5557: ""
5558: "
5559: @{
5560: if (GET_CODE (operands[2]) != CONST_INT
5561: || (unsigned) INTVAL (operands[2]) > 3)
5562: FAIL;
5563: @}")
5564: @end example
5565:
5566: @noindent
5567: This example uses @samp{define_expand} so that it can generate an RTL insn
5568: for shifting when the shift-count is in the supported range of 0 to 3 but
5569: fail in other cases where machine insns aren't available. When it fails,
5570: the compiler tries another strategy using different patterns (such as, a
5571: library call).
5572:
5573: If the compiler were able to handle nontrivial condition-strings in
5574: patterns with names, then there would be possible to use a
5575: @samp{define_insn} in that case. Here is another case (zero-extension on
5576: the 68000) which makes more use of the power of @samp{define_expand}:
5577:
5578: @example
5579: (define_expand "zero_extendhisi2"
5580: [(set (match_operand:SI 0 "general_operand" "")
5581: (const_int 0))
5582: (set (strict_low_part
5583: (subreg:HI
5584: (match_operand:SI 0 "general_operand" "")
5585: 0))
5586: (match_operand:HI 1 "general_operand" ""))]
5587: ""
5588: "operands[1] = make_safe_from (operands[1], operands[0]);")
5589: @end example
5590:
5591: @noindent
5592: Here two RTL insns are generated, one to clear the entire output operand
5593: and the other to copy the input operand into its low half. This sequence
5594: is incorrect if the input operand refers to [the old value of] the output
5595: operand, so the preparation statement makes sure this isn't so. The
5596: function @code{make_safe_from} copies the @code{operands[1]} into a
5597: temporary register if it refers to @code{operands[0]}. It does this
5598: by emitting another RTL insn.
5599:
5600: Finally, a third example shows the use of an internal operand.
5601: Zero-extension on the SPUR chip is done by @samp{and}-ing the result
5602: against a halfword mask. But this mask cannot be represented by a
5603: @samp{const_int} because the constant value is too large to be legitimate
5604: on this machine. So it must be copied into a register with
5605: @code{force_reg} and then the register used in the @samp{and}.
5606:
5607: @example
5608: (define_expand "zero_extendhisi2"
5609: [(set (match_operand:SI 0 "register_operand" "")
5610: (and:SI (subreg:SI
5611: (match_operand:HI 1 "register_operand" "")
5612: 0)
5613: (match_dup 2)))]
5614: ""
5615: "operands[2]
5616: = force_reg (SImode, gen_rtx (CONST_INT,
5617: VOIDmode, 65535)); ")
5618: @end example
5619:
5620: @node Machine Macros, Config, Machine Desc, Top
1.1 root 5621: @chapter Machine Description Macros
5622:
5623: The other half of the machine description is a C header file conventionally
5624: given the name @file{tm-@var{machine}.h}. The file @file{tm.h} should be a
5625: link to it. The header file @file{config.h} includes @file{tm.h} and most
5626: compiler source files include @file{config.h}.
5627:
5628: @menu
1.1.1.2 root 5629: * Run-time Target:: Defining -m options like -m68000 and -m68020.
1.1 root 5630: * Storage Layout:: Defining sizes and alignments of data types.
5631: * Registers:: Naming and describing the hardware registers.
5632: * Register Classes:: Defining the classes of hardware registers.
5633: * Stack Layout:: Defining which way the stack grows and by how much.
1.1.1.2 root 5634: * Library Names:: Specifying names of subroutines to call automatically.
1.1 root 5635: * Addressing Modes:: Defining addressing modes valid for memory operands.
5636: * Condition Code:: Defining how insns update the condition code.
5637: * Assembler Format:: Defining how to write insns and pseudo-ops to output.
5638: * Misc:: Everything else.
5639: @end menu
5640:
5641: @node Run-time Target, Storage Layout, Machine Macros, Machine Macros
5642: @section Run-time Target Specification
5643:
5644: @table @code
5645: @item CPP_PREDEFINES
1.1.1.2 root 5646: Define this to be a string constant containing @samp{-D} options
1.1 root 5647: to define the predefined macros that identify this machine and system.
5648:
5649: For example, on the Sun, one can use the value
5650:
5651: @example
1.1.1.2 root 5652: "-Dmc68000 -Dsun -Dunix"
1.1 root 5653: @end example
5654:
5655: @item extern int target_flags;
5656: This declaration should be present.
5657:
5658: @item TARGET_@dots{}
5659: This series of macros is to allow compiler command arguments to
5660: enable or disable the use of optional features of the target machine.
5661: For example, one machine description serves both the 68000 and
5662: the 68020; a command argument tells the compiler whether it should
5663: use 68020-only instructions or not. This command argument works
5664: by means of a macro @code{TARGET_68020} that tests a bit in
5665: @code{target_flags}.
5666:
5667: Define a macro @code{TARGET_@var{featurename}} for each such option.
5668: Its definition should test a bit in @code{target_flags}; for example:
5669:
5670: @example
5671: #define TARGET_68020 (target_flags & 1)
5672: @end example
5673:
5674: One place where these macros are used is in the condition-expressions
5675: of instruction patterns. Note how @code{TARGET_68020} appears
1.1.1.2 root 5676: frequently in the 68000 machine description file, @file{m68k.md}.
1.1 root 5677: Another place they are used is in the definitions of the other
5678: macros in the @file{tm-@var{machine}.h} file.
5679:
5680: @item TARGET_SWITCHES
1.1.1.2 root 5681: This macro defines names of command options to set and clear
1.1 root 5682: bits in @code{target_flags}. Its definition is an initializer
1.1.1.2 root 5683: with a subgrouping for each command option.
1.1 root 5684:
1.1.1.2 root 5685: Each subgrouping contains a string constant, that defines the option
1.1 root 5686: name, and a number, which contains the bits to set in
5687: @code{target_flags}. A negative number says to clear bits instead;
1.1.1.2 root 5688: the negative of the number is which bits to clear. The actual option
1.1 root 5689: name is made by appending @samp{-m} to the specified name.
5690:
5691: One of the subgroupings should have a null string. The number in
5692: this grouping is the default value for @code{target_flags}. Any
1.1.1.2 root 5693: target options act starting with that value.
1.1 root 5694:
5695: Here is an example which defines @samp{-m68000} and @samp{-m68020}
5696: with opposite meanings, and picks the latter as the default:
5697:
5698: @example
5699: #define TARGET_SWITCHES \
5700: @{ @{ "68020", 1@}, \
5701: @{ "68000", -1@}, \
5702: @{ "", 1@}@}
5703: @end example
5704: @end table
5705:
1.1.1.2 root 5706: Sometimes certain combinations of command options do not make sense on a
5707: particular target machine. You can define a macro @code{OVERRIDE_OPTIONS}
5708: to take account of this. This macro, if defined, is executed once
5709: just after all the command options have been parsed.
5710:
1.1 root 5711: @node Storage Layout, Registers, Run-time Target, Machine Macros
5712: @section Storage Layout
5713:
1.1.1.2 root 5714: Note that the definitions of the macros in this table which are sizes or
5715: alignments measured in bits do not need to be constant. They can be C
5716: expressions that refer to static variables, such as the @code{target_flags}.
5717: @xref{Run-time Target}.
5718:
1.1 root 5719: @table @code
5720: @item BITS_BIG_ENDIAN
5721: Define this macro if the most significant bit in a byte has the lowest
5722: number. This means that bit-field instructions count from the most
5723: significant bit. If the machine has no bit-field instructions, this
5724: macro is irrelevant.
5725:
5726: @item BYTES_BIG_ENDIAN
5727: Define this macro if the most significant byte in a word has the
5728: lowest number.
5729:
5730: @item WORDS_BIG_ENDIAN
1.1.1.2 root 5731: Define this macro if, in a multiword object, the most significant
1.1 root 5732: word has the lowest number.
5733:
5734: @item BITS_PER_UNIT
5735: Number of bits in an addressable storage unit (byte); normally 8.
5736:
5737: @item BITS_PER_WORD
5738: Number of bits in a word; normally 32.
5739:
5740: @item UNITS_PER_WORD
5741: Number of storage units in a word; normally 4.
5742:
5743: @item POINTER_SIZE
5744: Width of a pointer, in bits.
5745:
5746: @item PARM_BOUNDARY
1.1.1.2 root 5747: Alignment required for function parameters on the stack, in bits.
5748:
5749: @item STACK_BOUNDARY
5750: Define this macro if you wish to preserve a certain alignment for
5751: the stack pointer at all times. The definition is a C expression
5752: for the desired alignment (measured in bits).
1.1 root 5753:
5754: @item FUNCTION_BOUNDARY
5755: Alignment required for a function entry point, in bits.
5756:
5757: @item BIGGEST_ALIGNMENT
1.1.1.2 root 5758: Biggest alignment that any data type can require on this machine, in bits.
5759:
5760: @item EMPTY_FIELD_ALIGNMENT
5761: Alignment in bits to be given to a structure bit field that follows an
5762: empty field such as @code{int : 0;}.
5763:
5764: @item STRUCTURE_SIZE_BOUNDARY
5765: Number of bits which any structure or union's size must be a multiple of.
5766: Each structure or union's size is rounded up to a multiple of this.
5767:
5768: If you do not define this macro, the default is the same as
5769: @code{BITS_PER_UNIT}.
1.1 root 5770:
5771: @item STRICT_ALIGNMENT
5772: Define this if instructions will fail to work if given data not
5773: on the nominal alignment. If instructions will merely go slower
5774: in that case, do not define this macro.
1.1.1.4 ! root 5775:
! 5776: @item CHECK_FLOAT_VALUE (@var{mode}, @var{value})
! 5777: A C statement to validate the value @var{value} (or type
! 5778: @code{double}) for mode @var{mode}. This means that you check whether
! 5779: @var{value} fits within the possible range of values for mode
! 5780: @var{mode} on this target machine. The mode @var{mode} is always
! 5781: @code{SFmode} or @code{DFmode}.
! 5782:
! 5783: If @var{value} is not valid, you should call @code{error} to print an
! 5784: error message and then assign some valid value to @var{value}.
! 5785: Allowing an invalid value to go through the compiler can produce
! 5786: incorrect assembler code which may even cause Unix assemblers to
! 5787: crash.
! 5788:
! 5789: This macro need not be defined if there is no work for it to do.
1.1 root 5790: @end table
5791:
5792: @node Registers, Register Classes, Storage Layout, Machine Macros
5793: @section Register Usage
5794:
5795: @table @code
5796: @item FIRST_PSEUDO_REGISTER
5797: Number of hardware registers known to the compiler. They receive
5798: numbers 0 through @code{FIRST_PSEUDO_REGISTER-1}; thus, the first
1.1.1.2 root 5799: pseudo register's number really is assigned the number
1.1 root 5800: @code{FIRST_PSEUDO_REGISTER}.
5801:
5802: @item FIXED_REGISTERS
5803: An initializer that says which registers are used for fixed purposes
5804: all throughout the compiled code and are therefore not available for
1.1.1.2 root 5805: general allocation. These would include the stack pointer, the frame
1.1 root 5806: pointer, the program counter on machines where that is considered one
5807: of the addressable registers, and any other numbered register with a
5808: standard use.
5809:
5810: This information is expressed as a sequence of numbers, separated by
5811: commas and surrounded by braces. The @var{n}th number is 1 if
1.1.1.2 root 5812: register @var{n} is fixed, 0 otherwise.
5813:
5814: The table initialized from this macro, and the table initialized by
5815: the following one, may be overridden at run time either automatically,
5816: by the actions of the macro @code{CONDITIONAL_REGISTER_USAGE}, or by
5817: the user with the command options @samp{-ffixed-@var{reg}},
5818: @samp{-fcall-used-@var{reg}} and @samp{-fcall-saved-@var{reg}}.
1.1 root 5819:
5820: @item CALL_USED_REGISTERS
5821: Like @code{FIXED_REGISTERS} but has 1 for each register that is
5822: clobbered (in general) by function calls as well as for fixed
5823: registers. This macro therefore identifies the registers that are not
5824: available for general allocation of values that must live across
5825: function calls.
5826:
1.1.1.2 root 5827: If a register has 0 in @code{CALL_USED_REGISTERS}, the compiler
1.1 root 5828: automatically saves it on function entry and restores it on function
5829: exit, if the register is used within the function.
5830:
1.1.1.2 root 5831: @item CONDITIONAL_REGISTER_USAGE
5832: Zero or more C statements that may conditionally modify two variables
5833: @code{fixed_regs} and @code{call_used_regs} (both of type @code{char
5834: []}) after they have been initialized from the two preceding macros.
5835:
5836: This is necessary in case the fixed or call-clobbered registers depend
5837: on target flags.
5838:
5839: You need not define this macro if it has no work to do.
5840:
1.1.1.4 ! root 5841: @item OVERLAPPING_REGNO_P (@var{regno})
! 5842: If defined, this is a C expression whose value is @var{regno} is
! 5843: nonzero if hard register number @var{regno} is an overlapping
! 5844: register. This means a hard register which overlaps a hard register
! 5845: with a different number. (Such overlap is undesirable, but
! 5846: occasionally it allows a machine to be supported which otherwise could
! 5847: not be.) This macro must return nonzero for @emph{all} the registers
! 5848: which overlap each other. GNU CC can use an overlapping register only
! 5849: in certain limited ways. It can be used for allocation within a basic
! 5850: block, and may be spilled for reloading; that is all.
! 5851:
! 5852: If this macro is not defined, it means that none of the hard registers
! 5853: overlap each other. This is the usual situation.
! 5854:
! 5855: @item INSN_CLOBBERS_REGNO_P (@var{insn}, @var{regno})
! 5856: If defined, this is a C expression whose value should be nonzero if
! 5857: the insn @var{insn} has the effect of mysteriously clobbering the
! 5858: contents of hard register number @var{regno}. By ``mysterious'' we
! 5859: mean that the insn's RTL expression doesn't describe such an effect.
! 5860:
! 5861: If this macro is not defined, it means that no insn clobbers registers
! 5862: mysteriously. This is the usual situation; all else being equal,
! 5863: it is best for the RTL expression to show all the activity.
! 5864:
! 5865: @item PRESERVE_DEATH_INFO_REGNO_P (@var{regno})
! 5866: If defined, this is a C expression whose value is nonzero if accurate
! 5867: @code{REG_DEAD} notes are needed for hard register number @var{regno}
! 5868: at the time of outputting the assembler code. When this is so, a few
! 5869: optimizations that take place after register allocation and could
! 5870: invalidate the death notes are not done when this register is
! 5871: involved.
! 5872:
! 5873: You would arrange to preserve death info for a register when some
! 5874: of the code in the machine description which is executed to write
! 5875: the assembler code looks at the the death notes. This is
! 5876: necessary only when the actual hardware feature which GNU CC
! 5877: thinks of as a register is not actually a register of the usual sort.
! 5878: (It might, for example, be a hardware stack.)
! 5879:
! 5880: If this macro is not defined, it means that no death notes need to be
! 5881: preserved. This is the usual situation.
! 5882:
1.1 root 5883: @item HARD_REGNO_REGS (@var{regno}, @var{mode})
5884: A C expression for the number of consecutive hard registers, starting
5885: at register number @var{regno}, required to hold a value of mode
5886: @var{mode}.
5887:
5888: On a machine where all registers are exactly one word, a suitable
5889: definition of this macro is
5890:
5891: @example
5892: #define HARD_REGNO_NREGS(REGNO, MODE) \
5893: ((GET_MODE_SIZE (MODE) + UNITS_PER_WORD - 1) \
5894: / UNITS_PER_WORD))
5895: @end example
5896:
5897: @item HARD_REGNO_MODE_OK (@var{regno}, @var{mode})
5898: A C expression that is nonzero if it is permissible to store a value
5899: of mode @var{mode} in hard register number @var{regno} (or in several
5900: registers starting with that one). For a machine where all registers
5901: are equivalent, a suitable definition is
5902:
5903: @example
5904: #define HARD_REGNO_MODE_OK(REGNO, MODE) 1
5905: @end example
5906:
5907: It is not necessary for this macro to check for fixed register numbers
5908: because the allocation mechanism considers them to be always occupied.
5909:
1.1.1.2 root 5910: Many machines have special registers for floating point arithmetic.
5911: Often people assume that floating point machine modes are allowed only
5912: in floating point registers. This is not true. Any registers that
5913: can hold integers can safely @emph{hold} a floating point machine
5914: mode, whether or not floating arithmetic can be done on it in those
5915: registers.
5916:
5917: The true significance of special floating registers is rather than
5918: non-floating-point machine modes @emph{may not} go in those registers.
5919: This is true if the floating registers normalize any value stored in
5920: them, because storing a non-floating value there would garble it. If
5921: the floating registers do not automatically normalize, if you can
5922: store any bit pattern in one and retrieve it unchanged without a trap,
5923: then any machine mode may go in a floating register and this macro
5924: should say so.
5925:
5926: Sometimes there are floating registers that are especially slow to
5927: access, so that it is better to store a value in a stack frame than in
5928: such a register if floating point arithmetic is not being done. As long
5929: as the floating registers are not in class @code{GENERAL_REGS}, they
5930: will not be used unless some insn's constraint asks for one.
5931:
5932: It is obligatory to support floating point `move' instructions into
5933: and out of general registers, because unions and structures (which
5934: have modes @samp{SImode} or @samp{DImode}) can be in those registers
5935: and they may have floating point members.
5936:
1.1 root 5937: @item MODES_TIEABLE_P (@var{mode1}, @var{mode2})
5938: A C expression that is nonzero if it is desirable to choose register
5939: allocation so as to avoid move instructions between a value of mode
5940: @var{mode1} and a value of mode @var{mode2}.
5941:
5942: If @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode1})} and
5943: @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode2})} are ever different
5944: for any @var{r}, then @code{MODES_TIEABLE_P (@var{mode1},
5945: @var{mode2})} must be zero.
5946:
5947: @item PC_REGNUM
5948: If the program counter has a register number, define this as that
5949: register number. Otherwise, do not define it.
5950:
5951: @item STACK_POINTER_REGNUM
5952: The register number of the stack pointer register, which must also be
5953: a fixed register according to @code{FIXED_REGISTERS}. On many
5954: machines, the hardware determines which register this is.
5955:
5956: @item FRAME_POINTER_REGNUM
5957: The register number of the frame pointer register, which is used to
1.1.1.2 root 5958: access automatic variables in the stack frame. On some machines, the
1.1 root 5959: hardware determines which register this is. On other machines, you
5960: can choose any register you wish for this purpose.
5961:
1.1.1.2 root 5962: @item FRAME_POINTER_REQUIRED
5963: A C expression which is nonzero if a function must have and use a
5964: frame pointer. This expression is evaluated in the reload pass, in
5965: the function @code{reload}, and it can in principle examine the
5966: current function and decide according to the facts, but on most
5967: machines the constant 0 or the constant 1 suffices. Use 0 when the
5968: machine allows code to be generated with no frame pointer, and doing
5969: so saves some time or space. Use 1 when there is no possible
5970: advantage to avoiding a frame pointer.
5971:
5972: In certain cases, the compiler does not know how to do without a frame
5973: pointer. The compiler recognizes those cases and automatically gives
5974: the function a frame pointer regardless of what
5975: @code{FRAME_POINTER_REQUIRED} says. You don't need to worry about
5976: them.@refill
5977:
5978: In a function that does not require a frame pointer, the frame pointer
1.1.1.4 ! root 5979: register can be allocated for ordinary usage, unless you mark it as a
! 5980: fixed register. See @code{FIXED_REGISTERS} for more information.
1.1.1.2 root 5981:
1.1 root 5982: @item ARG_POINTER_REGNUM
5983: The register number of the arg pointer register, which is used to
5984: access the function's argument list. On some machines, this is the
5985: same as the frame pointer register. On some machines, the hardware
5986: determines which register this is. On other machines, you can choose
1.1.1.4 ! root 5987: any register you wish for this purpose. If this is not the same
! 5988: register as the frame pointer register, then you must mark it as a
1.1 root 5989: fixed register according to @code{FIXED_REGISTERS}.
5990:
5991: @item STATIC_CHAIN_REGNUM
5992: The register number used for passing a function's static chain
5993: pointer. This is needed for languages such as Pascal and Algol where
5994: functions defined within other functions can access the local
5995: variables of the outer functions; it is not currently used because C
5996: does not provide this feature.
5997:
5998: The static chain register need not be a fixed register.
5999:
6000: @item STRUCT_VALUE_REGNUM
1.1.1.2 root 6001: When a function's value's mode is @code{BLKmode}, the value is not
6002: returned according to @code{FUNCTION_VALUE}. Instead, the caller
6003: passes the address of a block of memory in which the value should be
6004: stored. @code{STRUCT_VALUE_REGNUM} is the register in which this
6005: address is passed.
1.1 root 6006: @end table
6007:
6008: @node Register Classes, Stack Layout, Registers, Machine Macros
6009: @section Register Classes
6010:
6011: On many machines, the numbered registers are not all equivalent.
6012: For example, certain registers may not be allowed for indexed addressing;
6013: certain registers may not be allowed in some instructions. These machine
6014: restrictions are described to the compiler using @dfn{register classes}.
6015:
6016: You define a number of register classes, giving each one a name and saying
6017: which of the registers belong to it. Then you can specify register classes
6018: that are allowed as operands to particular instruction patterns.
6019:
6020: In general, each register will belong to several classes. In fact, one
6021: class must be named @code{ALL_REGS} and contain all the registers. Another
6022: class must be named @code{NO_REGS} and contain no registers. Often the
6023: union of two classes will be another class; however, this is not required.
6024:
6025: One of the classes must be named @code{GENERAL_REGS}. There is nothing
6026: terribly special about the name, but the operand constraint letters
6027: @samp{r} and @samp{g} specify this class. If @code{GENERAL_REGS} is
6028: the same as @code{ALL_REGS}, just define it as a macro which expands
6029: to @code{ALL_REGS}.
6030:
6031: The way classes other than @code{GENERAL_REGS} are specified in operand
6032: constraints is through machine-dependent operand constraint letters.
6033: You can define such letters to correspond to various classes, then use
6034: them in operand constraints.
6035:
1.1.1.2 root 6036: You should define a class for the union of two classes whenever some
6037: instruction allows both classes. For example, if an instruction allows
6038: either a floating-point (coprocessor) register or a general register for a
6039: certain operand, you should define a class @code{FLOAT_OR_GENERAL_REGS}
6040: which includes both of them. Otherwise you will get suboptimal code.
6041:
1.1 root 6042: You must also specify certain redundant information about the register
6043: classes: for each class, which classes contain it and which ones are
6044: contained in it; for each pair of classes, the largest class contained
6045: in their union.
6046:
6047: @table @code
6048: @item enum reg_class
6049: An enumeral type that must be defined with all the register class names
6050: as enumeral values. @code{NO_REGS} must be first. @code{ALL_REGS}
6051: must be the last register class, followed by one more enumeral value,
6052: @code{LIM_REG_CLASSES}, which is not a register class but rather
6053: tells how many classes there are.
6054:
6055: Each register class has a number, which is the value of casting
6056: the class name to type @code{int}. The number serves as an index
6057: in many of the tables described below.
6058:
6059: @item REG_CLASS_NAMES
6060: An initializer containing the names of the register classes as C string
6061: constants. These names are used in writing some of the debugging dumps.
6062:
6063: @item REG_CLASS_CONTENTS
6064: An initializer containing the contents of the register classes, as integers
6065: which are bit masks. The @var{n}th integer specifies the contents of class
6066: @var{n}. The way the integer @var{mask} is interpreted is that
6067: register @var{r} is in the class if @code{@var{mask} & (1 << @var{r})} is 1.
6068:
6069: When the machine has more than 32 registers, an integer does not suffice.
6070: Then the integers are replaced by sub-initializers, braced groupings containing
6071: several integers. Each sub-initializer must be suitable as an initializer
6072: for the type @code{HARD_REG_SET} which is defined in @file{hard-reg-set.h}.
6073:
6074: @item REGNO_REG_CLASS (@var{regno})
6075: A C expression whose value is a register class containing hard register
6076: @var{regno}. In general there is more that one such class; choose a class
6077: which is @dfn{minimal}, meaning that no smaller class also contains the
6078: register.
6079:
6080: @item INDEX_REG_CLASS
6081: A macro whose definition is the name of the class to which a valid index
6082: register must belong.
6083:
6084: @item REG_CLASS_FROM_LETTER (@var{char})
6085: A C expression which defines the machine-dependent operand constraint
6086: letters for register classes. If @var{char} is such a letter, the value
6087: should be the register class corresponding to it. Otherwise, the value
6088: should be @code{NO_REGS}.
6089:
1.1.1.2 root 6090: @item REGNO_OK_FOR_BASE_P (@var{num})
6091: A C expression which is nonzero if register number @var{num} is
6092: suitable for use as a base register in operand addresses. It may be
6093: either a suitable hard register or a pseudo register that has been
6094: allocated such a hard register.
6095:
6096: @item REGNO_OK_FOR_INDEX_P (@var{num})
6097: A C expression which is nonzero if register number @var{num} is
6098: suitable for use as an index register in operand addresses. It may be
6099: either a suitable hard register or a pseudo register that has been
6100: allocated such a hard register.
6101:
6102: The difference between an index register and a base register is that
6103: the index register may be scaled. If an address involves the sum of
6104: two registers, neither one of them scaled, then either one may be
6105: labeled the ``base'' and the other the ``index''; but whichever
6106: labeling is used must fit the machine's constraints of which registers
6107: may serve in each capacity. The compiler will try both labelings,
6108: looking for one that is valid, and reload one or both registers only
6109: if neither labeling works.
1.1 root 6110:
6111: @item PREFERRED_RELOAD_CLASS (@var{x}, @var{class})
6112: A C expression that places additional restrictions on the register class
6113: to use when it is necessary to copy value @var{x} into a register in class
6114: @var{class}. The value is a register class; perhaps @var{class}, or perhaps
6115: another, smaller class. @var{class} is always safe as a value. In fact,
6116: the definition
6117:
6118: @example
6119: #define PREFERRED_RELOAD_CLASS(X,CLASS) CLASS
6120: @end example
6121:
6122: @noindent
6123: is always safe. However, sometimes returning a more restrictive class
6124: makes better code. For example, on the 68000, when @var{x} is an
6125: integer constant that is in range for a @samp{moveq} instruction,
6126: the value of this macro is always @code{DATA_REGS} as long as
6127: @var{class} includes the data registers. Requiring a data register
6128: guarantees that a @samp{moveq} will be used.
1.1.1.2 root 6129:
6130: @item CLASS_MAX_NREGS (@var{class}, @var{mode})
6131: A C expression for the maximum number of consecutive registers
6132: of class @var{class} needed to hold a value of mode @var{mode}.
6133:
6134: This is closely related to the macro @code{HARD_REGNO_NREGS}.
6135: In fact, the value of the macro @code{CLASS_MAX_NREGS (@var{class}, @var{mode})}
6136: should be the maximum value of @code{HARD_REGNO_NREGS (@var{regno}, @var{mode})}
6137: for all @var{regno} values in the class @var{class}.
6138:
6139: This macro helps control the handling of multiple-word values
6140: in the reload pass.
1.1 root 6141: @end table
6142:
1.1.1.2 root 6143: Two other special macros describe which constants fit which constraint
6144: letters.
1.1 root 6145:
6146: @table @code
6147: @item CONST_OK_FOR_LETTER_P (@var{value}, @var{c})
6148: A C expression that defines the machine-dependent operand constraint letters
6149: that specify particular ranges of integer values. If @var{c} is one
6150: of those letters, the expression should check that @var{value}, an integer,
6151: is in the appropriate range and return 1 if so, 0 otherwise. If @var{c} is
6152: not one of those letters, the value should be 0 regardless of @var{value}.
6153:
6154: @item CONST_DOUBLE_OK_FOR_LETTER_P (@var{value}, @var{c})
6155: A C expression that defines the machine-dependent operand constraint
6156: letters that specify particular ranges of floating values. If @var{c} is
1.1.1.2 root 6157: one of those letters, the expression should check that @var{value}, an RTX
1.1 root 6158: of code @samp{const_double}, is in the appropriate range and return 1 if
6159: so, 0 otherwise. If @var{c} is not one of those letters, the value should
6160: be 0 regardless of @var{value}.
6161: @end table
6162:
1.1.1.2 root 6163: @node Stack Layout, Library Names, Register Classes, Machine Macros
1.1 root 6164: @section Describing Stack Layout
6165:
6166: @table @code
6167: @item STACK_GROWS_DOWNWARD
6168: Define this macro if pushing a word onto the stack moves the stack
1.1.1.2 root 6169: pointer to a smaller address.
6170:
6171: When we say, ``define this macro if @dots{},'' it means that the
6172: compiler checks this macro only with @code{#ifdef} so the precise
6173: definition used does not matter.
1.1 root 6174:
6175: @item FRAME_GROWS_DOWNWARD
6176: Define this macro if the addresses of local variable slots are at negative
6177: offsets from the frame pointer.
6178:
6179: @item STARTING_FRAME_OFFSET
6180: Offset from the frame pointer to the first local variable slot to be allocated.
6181:
6182: If @code{FRAME_GROWS_DOWNWARD}, the next slot's offset is found by
6183: subtracting the length of the first slot from @code{STARTING_FRAME_OFFSET}.
6184: Otherwise, it is found by adding the length of the first slot to
6185: the value @code{STARTING_FRAME_OFFSET}.
6186:
6187: @item PUSH_ROUNDING (@var{npushed})
6188: A C expression that is the number of bytes actually pushed onto the
6189: stack when an instruction attempts to push @var{npushed} bytes.
6190:
1.1.1.2 root 6191: If the target machine does not have a push instruction, do not define
6192: this macro. That directs GNU CC to use an alternate strategy: to
6193: allocate the entire argument block and then store the arguments into
6194: it.
6195:
1.1 root 6196: On some machines, the definition
6197:
6198: @example
6199: #define PUSH_ROUNDING(BYTES) (BYTES)
6200: @end example
6201:
6202: @noindent
6203: will suffice. But on other machines, instructions that appear
6204: to push one byte actually push two bytes in an attempt to maintain
6205: alignment. Then the definition should be
6206:
6207: @example
6208: #define PUSH_ROUNDING(BYTES) (((BYTES) + 1) & ~1)
6209: @end example
6210:
6211: @item FIRST_PARM_OFFSET
6212: Offset from the argument pointer register to the first argument's address.
6213:
1.1.1.2 root 6214: @item RETURN_POPS_ARGS (@var{funtype})
6215: A C expression that should be 1 if a function pops its own arguments
6216: on returning, or 0 if the function pops no arguments and the caller
6217: must therefore pop them all after the function returns.
6218:
6219: @var{funtype} is a C variable whose value is a tree node that
6220: describes the function in question. Normally it is a node of type
6221: @code{FUNCTION_TYPE} that describes the data type of the function.
6222: From this it is possible to obtain the data types of the value and
6223: arguments (if known).
6224:
6225: When a call to a library function is being considered, @var{funtype}
6226: will contain an identifier node for the library function. Thus, if
6227: you need to distinguish among various library functions, you can do so
6228: by their names. Note that ``library function'' in this context means
6229: a function used to perform arithmetic, whose name is known specially
6230: in the compiler and was not mentioned in the C code being compiled.
6231:
6232: On the Vax, all functions always pop their arguments, so the
6233: definition of this macro is 1. On the 68000, using the standard
6234: calling convention, no functions pop their arguments, so the value of
6235: the macro is always 0 in this case. But an alternative calling
6236: convention is available in which functions that take a fixed number of
6237: arguments pop them but other functions (such as @code{printf}) pop
6238: nothing (the caller pops all). When this convention is in use,
6239: @var{funtype} is examined to determine whether a function takes a
6240: fixed number of arguments.
6241:
6242: @item FUNCTION_VALUE (@var{valtype}, @var{func})
6243: A C expression to create an RTX representing the place where a
6244: function returns a value of data type @var{valtype}. @var{valtype} is
6245: a tree node representing a data type. Write @code{TYPE_MODE
6246: (@var{valtype})} to get the machine mode used to represent that type.
6247: On many machines, only the mode is relevant. (Actually, on most
6248: machines, scalar values are returned in the same place regardless of
6249: mode).@refill
6250:
6251: If the precise function being called is known, @var{func} is a tree
6252: node (@code{FUNCTION_DECL}) for it; otherwise, @var{func} is a null
6253: pointer. This makes it possible to use a different value-returning
6254: convention for specific functions when all their calls are
6255: known.@refill
6256:
6257: @item FUNCTION_OUTGOING_VALUE (@var{valtype}, @var{func})
6258: Define this macro if the target machine has ``register windows''
6259: so that the register in which a function returns its value is not
6260: the same as the one in which the caller sees the value.
6261:
6262: For such machines, @code{FUNCTION_VALUE} computes the register in
6263: which the caller will see the value, and
6264: @code{FUNCTION_OUTGOING_VALUE} should be defined in a similar fashion
6265: to tell the function where to put the value.@refill
6266:
6267: If @code{FUNCTION_OUTGOING_VALUE} is not defined,
6268: @code{FUNCTION_VALUE} serves both purposes.@refill
6269:
6270: @item LIBCALL_VALUE (@var{mode})
6271: A C expression to create an RTX representing the place where a library
6272: function returns a value of mode @var{mode}. If the precise function
6273: being called is known, @var{func} is a tree node
6274: (@code{FUNCTION_DECL}) for it; otherwise, @var{func} is a null
6275: pointer. This makes it possible to use a different value-returning
6276: convention for specific functions when all their calls are
6277: known.@refill
6278:
6279: Note that ``library function'' in this context means a compiler
6280: support routine, used to perform arithmetic, whose name is known
6281: specially by the compiler and was not mentioned in the C code being
6282: compiled.
6283:
6284: @item FUNCTION_VALUE_REGNO_P (@var{regno})
6285: A C expression that is nonzero if @var{regno} is the number of a hard
6286: register in which function values are sometimes returned.
6287:
6288: A register whose use for returning values is limited to serving as the
6289: second of a pair (for a value of type @code{double}, say) need not be
6290: recognized by this macro. So for most machines, this definition
6291: suffices:
6292:
6293: @example
6294: #define FUNCTION_VALUE_REGNO_P(N) ((N) == 0)
6295: @end example
6296:
6297: @item FUNCTION_ARG (@var{cum}, @var{mode}, @var{type}, @var{named})
6298: A C expression that controls whether a function argument is passed
6299: in a register, and which register.
6300:
6301: The arguments are @var{cum}, which summarizes all the previous
6302: arguments; @var{mode}, the machine mode of the argument; @var{type},
6303: the data type of the argument as a tree node or 0 if that is not known
6304: (which happens for C support library functions); and @var{named},
6305: which is 1 for an ordinary argument and 0 for nameless arguments that
6306: correspond to @samp{...} in the called function's prototype.
6307:
6308: The value of the expression should either be a @samp{reg} RTX for the
6309: hard register in which to pass the argument, or zero to pass the
6310: argument on the stack.
6311:
6312: For the Vax and 68000, where normally all arguments are pushed, zero
6313: suffices as a definition.
6314:
6315: @item FUNCTION_INCOMING_ARG (@var{cum}, @var{mode}, @var{type}, @var{named})
6316: Define this macro if the target machine has ``register windows'', so
6317: that the register in which a function sees an arguments is not
6318: necessarily the same as the one in which the caller passed the
6319: argument.
6320:
6321: For such machines, @code{FUNCTION_ARG} computes the register in which
6322: the caller passes the value, and @code{FUNCTION_INCOMING_ARG} should
6323: be defined in a similar fashion to tell the function being called
6324: where the arguments will arrive.
6325:
6326: If @code{FUNCTION_INCOMING_ARG} is not defined, @code{FUNCTION_ARG}
6327: serves both purposes.@refill
6328:
6329: @item FUNCTION_ARG_PARTIAL_NREGS (@var{cum}, @var{mode}, @var{type}, @var{named})
6330: A C expression for the number of words, at the beginning of an
6331: argument, must be put in registers. The value must be zero for
6332: arguments that are passed entirely in registers or that are entirely
6333: pushed on the stack.
6334:
6335: On some machines, certain arguments must be passed partially in
6336: registers and partially in memory. On these machines, typically the
6337: first @var{n} words of arguments are passed in registers, and the rest
6338: on the stack. If a multi-word argument (a @code{double} or a
6339: structure) crosses that boundary, its first few words must be passed
6340: in registers and the rest must be pushed. This macro tells the
6341: compiler when this occurs, and how many of the words should go in
6342: registers.
6343:
6344: @code{FUNCTION_ARG} for these arguments should return the first
6345: register to be used by the caller for this argument; likewise
6346: @code{FUNCTION_INCOMING_ARG}, for the called function.
6347:
6348: @item CUMULATIVE_ARGS
6349: A C type for declaring a variable that is used as the first argument
6350: of @code{FUNCTION_ARG} and other related values. For some target
6351: machines, the type @code{int} suffices and can hold the number of
6352: bytes of argument so far.
6353:
6354: @item INIT_CUMULATIVE_ARGS (@var{cum})
6355: A C statement (sans semicolon) for initializing the variable @var{cum}
6356: for the state at the beginning of the argument list. The variable has
6357: type @code{CUMULATIVE_ARGS}.
6358:
6359: @item FUNCTION_ARG_ADVANCE (@var{cum}, @var{mode}, @var{type}, @var{named})
6360: Update the summarizer variable @var{cum} to advance past an argument
6361: in the argument list. The values @var{mode}, @var{type} and
6362: @var{named} describe that argument. Once this is done, the variable
6363: @var{cum} is suitable for analyzing the @emph{following} argument
6364: with @code{FUNCTION_ARG}, etc.@refill
6365:
6366: @item FUNCTION_ARG_REGNO_P (@var{regno})
6367: A C expression that is nonzero if @var{regno} is the number of a hard
6368: register in which function arguments are sometimes passed. This does
6369: @emph{not} include implicit arguments such as the static chain and
6370: the structure-value address. On many machines, no registers can be
6371: used for this purpose since all function arguments are pushed on the
6372: stack.
1.1 root 6373:
6374: @item FUNCTION_PROLOGUE (@var{file}, @var{size})
6375: A C compound statement that outputs the assembler code for entry to a
6376: function. The prologue is responsible for setting up the stack frame,
6377: initializing the frame pointer register, saving registers that must be
1.1.1.2 root 6378: saved, and allocating @var{size} additional bytes of storage for the
6379: local variables. @var{size} is an integer. @var{file} is a stdio
6380: stream to which the assembler code should be output.
1.1 root 6381:
6382: The label for the beginning of the function need not be output by this
6383: macro. That has already been done when the macro is run.
6384:
6385: To determine which registers to save, the macro can refer to the array
1.1.1.2 root 6386: @code{regs_ever_live}: element @var{r} is nonzero if hard register
6387: @var{r} is used anywhere within the function. This implies the
6388: function prologue should save register @var{r}, but not if it is one
6389: of the call-used registers.
6390:
6391: On machines where functions may or may not have frame-pointers, the
6392: function entry code must vary accordingly; it must set up the frame
6393: pointer if one is wanted, and not otherwise. To determine whether a
6394: frame pointer is in wanted, the macro can refer to the variable
6395: @code{frame_pointer_needed}. The variable's value will be 1 at run
6396: time in a function that needs a frame pointer.
6397:
6398: @item FUNCTION_PROFILER (@var{file}, @var{labelno})
6399: A C statement or compound statement to output to @var{file} some
6400: assembler code to call the profiling subroutine @code{mcount}.
6401: Before calling, the assembler code must load the address of a
6402: counter variable into a register where @code{mcount} expects to
6403: find the address. The name of this variable is @samp{LP} followed
6404: by the number @var{labelno}, so you would generate the name using
6405: @samp{LP%d} in a @code{fprintf}.
6406:
6407: The details of how the address should be passed to @code{mcount} are
6408: determined by your operating system environment, not by GNU CC. To
6409: figure them out, compile a small program for profiling using the
6410: system's installed C compiler and look at the assembler code that
6411: results.
6412:
6413: @item EXIT_IGNORES_STACK
6414: Define this macro as a C expression that is nonzero if the return
6415: instruction or the function epilogue ignores the value of the stack
6416: pointer; in other words, if it is safe to delete an instruction to
6417: adjust the stack pointer before a return from the function.
6418:
6419: Note that this macro's value is relevant only for for which frame
6420: pointers are maintained. It is never possible to delete a final stack
6421: adjustment in a function that has no frame pointer, and the compiler
6422: knows this regardless of @code{EXIT_IGNORES_STACK}.
1.1 root 6423:
6424: @item FUNCTION_EPILOGUE (@var{file}, @var{size})
6425: A C compound statement that outputs the assembler code for exit from a
6426: function. The epilogue is responsible for restoring the saved
6427: registers and stack pointer to their values when the function was
6428: called, and returning control to the caller. This macro takes the
6429: same arguments as the macro @code{FUNCTION_PROLOGUE}, and the
6430: registers to restore are determined from @code{regs_ever_live} and
6431: @code{CALL_USED_REGISTERS} in the same way.
6432:
1.1.1.2 root 6433: On some machines, there is a single instruction that does all the work
6434: of returning from the function. On these machines, give that
6435: instruction the name @samp{return} and do not define the macro
6436: @code{FUNCTION_EPILOGUE} at all.
6437:
6438: On machines where functions may or may not have frame-pointers, the
6439: function exit code must vary accordingly. Sometimes the code for
6440: these two cases is completely different. To determine whether a frame
6441: pointer is in wanted, the macro can refer to the variable
6442: @code{frame_pointer_needed}. The variable's value will be 1 at run
6443: time in a function that needs a frame pointer.
6444:
6445: On some machines, some functions pop their arguments on exit while
6446: others leave that for the caller to do. For example, the 68020 when
6447: given @samp{-mrtd} pops arguments in functions that take a fixed
6448: number of arguments.
6449:
6450: Your definition of the macro @code{RETURN_POPS_ARGS} decides which
6451: functions pop their own arguments. @code{FUNCTION_EPILOGUE} needs to
6452: know what was decided. The variable @code{current_function_pops_args}
6453: is nonzero if the function should pop its own arguments. If so, use
6454: the variable @code{current_function_args_size} as the number of bytes
6455: to pop.
6456:
6457: @item FIX_FRAME_POINTER_ADDRESS (@var{addr}, @var{depth})
6458: A C compound statement to alter a memory address that uses the frame
6459: pointer register so that it uses the stack pointer register instead.
6460: This must be done in the instructions that load parameter values into
6461: registers, when the reload pass determines that a frame pointer is not
6462: necessary for the function. @var{addr} will be a C variable name, and
6463: the updated address should be stored in that variable. @var{depth}
6464: will be the current depth of stack temporaries (number of bytes of
6465: arguments currently pushed). The change in offset between a
6466: frame-pointer-relative address and a stack-pointer-relative address
6467: must include @var{depth}.
6468:
6469: Even if your machine description specifies there will always be a
6470: frame pointer in the frame pointer register, you must still define
6471: @code{FIX_FRAME_POINTER_ADDRESS}, but the definition will never be
6472: executed at run time, so it may be empty.
6473: @end table
6474:
6475: @node Library Names, Addressing Modes, Stack Layout, Machine Macros
6476: @section Library Subroutine Names
6477:
6478: @table @code
6479: @item UDIVSI3_LIBCALL
6480: A C string constant giving the name of the function to call for
6481: division of a full-word by a full-word. If you do not define this
6482: macro, the default name is used, which is @code{_udivsi3}, a function
6483: defined in @file{gnulib}.
6484:
6485: @item UMODSI3_LIBCALL
6486: A C string constant giving the name of the function to call for the
6487: remainder in division of a full-word by a full-word. If you do not
6488: define this macro, the default name is used, which is @code{_umodsi3},
6489: a function defined in @file{gnulib}.
6490:
6491: @item TARGET_MEM_FUNCTIONS
6492: Define this macro if GNU CC should generate calls to the System V
6493: (and ANSI C) library functions @code{memcpy} and @code{memset}
6494: rather than the BSD functions @code{bcopy} and @code{bzero}.
1.1 root 6495: @end table
6496:
1.1.1.2 root 6497: @node Addressing Modes, Misc, Library Names, Machine Macros
1.1 root 6498: @section Addressing Modes
6499:
6500: @table @code
6501: @item HAVE_POST_INCREMENT
6502: Define this macro if the machine supports post-increment addressing.
6503:
6504: @item HAVE_PRE_INCREMENT
6505: @itemx HAVE_POST_DECREMENT
6506: @itemx HAVE_PRE_DECREMENT
6507: Similar for other kinds of addressing.
6508:
6509: @item CONSTANT_ADDRESS_P (@var{x})
1.1.1.2 root 6510: A C expression that is 1 if the RTX @var{x} is a constant whose value
1.1 root 6511: is an integer. This includes integers whose values are not explicitly
1.1.1.2 root 6512: known, such as @samp{symbol_ref} and @samp{label_ref} expressions and
6513: @samp{const} arithmetic expressions.
6514:
6515: On most machines, this can be defined as @code{CONSTANT_P (@var{x})},
6516: but a few machines are more restrictive in which constant addresses
6517: are supported.
1.1 root 6518:
6519: @item MAX_REGS_PER_ADDRESS
6520: A number, the maximum number of registers that can appear in a valid
6521: memory address.
6522:
6523: @item GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{label})
6524: A C compound statement with a conditional @code{goto @var{label};}
1.1.1.2 root 6525: executed if @var{x} (an RTX) is a legitimate memory address on the
6526: target machine for a memory operand of mode @var{mode}.
1.1 root 6527:
6528: It usually pays to define several simpler macros to serve as
1.1.1.2 root 6529: subroutines for this one. Otherwise it may be too complicated to
6530: understand.
6531:
6532: This macro must exist in two variants: a strict variant and a
6533: non-strict one. The strict variant is used in the reload pass. It
6534: must be defined so that any pseudo-register that has not been
6535: allocated a hard register is considered a memory reference. In
6536: contexts where some kind of register is required, a pseudo-register
6537: with no hard register must be rejected.
6538:
6539: The non-strict variant is used in other passes. It must be defined to
6540: accept all pseudo-registers in every context where some kind of
6541: register is required.
6542:
6543: Compiler source files that want to use the strict variant of this
6544: macro define the macro @code{REG_OK_STRICT}. You should use an
6545: @code{#ifdef REG_OK_STRICT} conditional to define the strict variant
6546: in that case and the non-strict variant otherwise.
6547:
6548: Typically among the subroutines used to define
6549: @code{GO_IF_LEGITIMATE_ADDRESS} are subroutines to check for
6550: acceptable registers for various purposes (one for base registers, one
6551: for index registers, and so on). Then only these subroutine macros
6552: need have two variants; the higher levels of macros may be the same
6553: whether strict or not.@refill
1.1 root 6554:
6555: @item LEGITIMIZE_ADDRESS (@var{x}, @var{oldx}, @var{mode}, @var{win})
6556: A C compound statement that attempts to replace @var{x} with a valid
1.1.1.2 root 6557: memory address for an operand of mode @var{mode}. @var{win} will be a
6558: C statement label elsewhere in the code; the macro definition may use
1.1 root 6559:
6560: @example
6561: GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{win});
6562: @end example
6563:
6564: @noindent
6565: to avoid further processing if the address has become legitimate.
6566:
6567: @var{x} will always be the result of a call to @code{break_out_memory_refs},
6568: and @var{oldx} will be the operand that was given to that function to produce
6569: @var{x}.
6570:
1.1.1.2 root 6571: The code generated by this macro should not alter the substructure of
6572: @var{x}. If it transforms @var{x} into a more legitimate form, it
6573: should assign @var{x} (which will always be a C variable) a new value.
6574:
6575: It is not necessary for this macro to come up with a legitimate
6576: address. The compiler has standard ways of doing so in all cases. In
6577: fact, it is safe for this macro to do nothing. But often a
6578: machine-dependent strategy can generate better code.
6579:
6580: @item GO_IF_MODE_DEPENDENT_ADDRESS (@var{addr}, @var{label})
6581: A C statement or compound statement with a conditional @code{goto
6582: @var{label};} executed if memory address @var{x} (an RTX) can have
6583: different meanings depending on the machine mode of the memory
6584: reference it is used for.
6585:
6586: Autoincrement and autodecrement addresses typically have mode-dependent
6587: effects because the amount of the increment or decrement is the size
6588: of the operand being addressed. Some machines have other mode-dependent
6589: addresses. Many RISC machines have no mode-dependent addresses.
6590:
6591: You may assume that @var{addr} is a valid address for the machine.
6592:
6593: @item LEGITIMATE_CONSTANT_P (@var{x})
6594: A C expression that is nonzero if @var{x} is a legitimate constant for
6595: an immediate operand on the target machine. You can assume that
6596: either @var{x} is a @samp{const_double} or it satisfies
6597: @code{CONSTANT_P}, so you need not check these things. In fact,
6598: @samp{1} is a suitable definition for this macro on machines where any
6599: @samp{const_double} is valid and anything @code{CONSTANT_P} is valid.@refill
1.1 root 6600: @end table
6601:
6602: @node Misc, Condition Code, Addressing Modes, Machine Macros
6603: @section Miscellaneous Parameters
6604:
6605: @table @code
6606: @item CASE_VECTOR_MODE
1.1.1.2 root 6607: An alias for a machine mode name. This is the machine mode that
6608: elements of a jump-table should have.
1.1 root 6609:
6610: @item CASE_VECTOR_PC_RELATIVE
6611: Define this macro if jump-tables should contain relative addresses.
6612:
1.1.1.2 root 6613: @item CASE_DROPS_THROUGH
6614: Define this if control falls through a @code{case} insn when the index
6615: value is out of range. This means the specified default-label is
6616: actually ignored by the @code{case} insn proper.
6617:
1.1 root 6618: @item IMPLICIT_FIX_EXPR
6619: An alias for a tree code that should be used by default for conversion
1.1.1.2 root 6620: of floating point values to fixed point. Normally,
6621: @code{FIX_ROUND_EXPR} is used.@refill
6622:
6623: @item FIXUNS_TRUNC_LIKE_FIX_TRUNC
6624: Define this macro if the same instructions that convert a floating
6625: point number to a signed fixed point number also convert validly to an
6626: unsigned one.
1.1 root 6627:
6628: @item EASY_DIV_EXPR
1.1.1.2 root 6629: An alias for a tree code that is the easiest kind of division to
6630: compile code for in the general case. It may be
6631: @code{TRUNC_DIV_EXPR}, @code{FLOOR_DIV_EXPR}, @code{CEIL_DIV_EXPR} or
6632: @code{ROUND_DIV_EXPR}. These four division operators differ in how
6633: they round the result to an integer. @code{EASY_DIV_EXPR} is used
6634: when it is permissible to use any of those kinds of division and the
6635: choice should be made on the basis of efficiency.@refill
6636:
6637: @item DEFAULT_SIGNED_CHAR
6638: An expression whose value is 1 or 0, according to whether the type
6639: @code{char} should be signed or unsigned by default. The user can
6640: always override this default with the options @samp{-fsigned-char}
6641: and @samp{-funsigned-char}.
6642:
6643: @item SCCS_DIRECTIVE
6644: Define this if the preprocessor should ignore @code{#sccs} directives
1.1.1.4 ! root 6645: and print no error message.
! 6646:
! 6647: @item IDENT_DIRECTIVE
! 6648: Define this if the preprocessor should ignore @code{#ident} directives
! 6649: and print no error message.
1.1 root 6650:
6651: @item MOVE_MAX
6652: The maximum number of bytes that a single instruction can move quickly
6653: from memory to memory.
6654:
1.1.1.2 root 6655: @item INT_TYPE_SIZE
6656: A C expression for the size in bits of the type @code{int} on the
6657: target machine.
6658:
6659: @item SLOW_BYTE_ACCESS
6660: Define this macro as a C expression which is nonzero if accessing less
6661: than a word of memory (i.e. a @code{char} or a @code{short}) is slow
6662: (requires more than one instruction).
6663:
1.1 root 6664: @item SLOW_ZERO_EXTEND
1.1.1.2 root 6665: Define this macro if zero-extension (of a @code{char} or @code{short}
6666: to an @code{int}) can be done faster if the destination is a register
6667: that is known to be zero.
6668:
6669: If you define this macro, you must have instruction patterns that
6670: recognize RTL structures like this:
6671:
6672: @example
6673: (set (strict-low-part (subreg:QI (reg:SI @dots{}) 0)) @dots{})
6674: @end example
6675:
6676: @noindent
6677: and likewise for @code{HImode}.
1.1 root 6678:
6679: @item SHIFT_COUNT_TRUNCATED
6680: Define this macro if shift instructions ignore all but the lowest few
6681: bits of the shift count. It implies that a sign-extend or zero-extend
6682: instruction for the shift count can be omitted.
6683:
1.1.1.2 root 6684: @item TRULY_NOOP_TRUNCATION (@var{outprec}, @var{inprec})
1.1 root 6685: A C expression which is nonzero if on this machine it is safe to
1.1.1.2 root 6686: ``convert'' an integer of @var{inprec} bits to one of @var{outprec}
6687: bits (where @var{outprec} is smaller than @var{inprec}) by merely
6688: operating on it as if it had only @var{outprec} bits.
1.1 root 6689:
6690: On many machines, this expression can be 1.
6691:
1.1.1.2 root 6692: @item NO_FUNCTION_CSE
6693: Define this macro if it is as good or better to call a constant
6694: function address than to call an address kept in a register.
6695:
6696: @item STORE_FLAG_VALUE
6697: A C expression for the value stored by a store-flag instruction
6698: (@code{s@var{cond}}) when the condition is true. This is usually 1 or
6699: -1; it is required to be an odd number.
6700:
6701: Do not define @code{STORE_FLAG_VALUE} if the machine has no store-flag
6702: instructions.
6703:
1.1 root 6704: @item Pmode
1.1.1.2 root 6705: An alias for the machine mode for pointers. Normally the definition
6706: can be
1.1 root 6707:
6708: @example
6709: #define Pmode SImode
6710: @end example
6711:
6712: @item FUNCTION_MODE
1.1.1.2 root 6713: An alias for the machine mode used for memory references to functions
6714: being called, in @samp{call} RTL expressions. On most machines this
6715: should be @code{QImode}.
1.1 root 6716:
6717: @item CONST_COST (@var{x}, @var{code})
1.1.1.2 root 6718: A part of a C @code{switch} statement that describes the relative
6719: costs of constant RTL expressions. It must contain @code{case} labels
6720: for expression codes @samp{const_int}, @samp{const}, @samp{symbol_ref}, @samp{label_ref}
6721: and @samp{const_double}. Each case must ultimately reach a
6722: @code{return} statement to return the relative cost of the use of that
1.1 root 6723: kind of constant value in an expression. The cost may depend on the
6724: precise value of the constant, which is available for examination in
6725: @var{x}.
6726:
1.1.1.2 root 6727: @var{code} is the expression code---redundant, since it can be
6728: obtained with @code{GET_CODE (@var{x})}.
6729:
6730: @item DOLLARS_IN_IDENTIFIERS
1.1.1.4 ! root 6731: Define this to be nonzero if the character @samp{$} should be allowed
! 6732: by default in identifier names.
1.1 root 6733: @end table
6734:
6735: @node Condition Code, Assembler Format, Misc, Machine Macros
6736: @section Condition Code Information
6737:
6738: The file @file{conditions.h} defines a variable @code{cc_status} to
6739: describe how the condition code was computed (in case the interpretation of
6740: the condition code depends on the instruction that it was set by). This
6741: variable contains the RTL expressions on which the condition code is
6742: currently based, and several standard flags.
6743:
6744: Sometimes additional machine-specific flags must be defined in the machine
6745: description header file. It can also add additional machine-specific
6746: information by defining @code{CC_STATUS_MDEP}.
6747:
6748: @table @code
6749: @item CC_STATUS_MDEP
1.1.1.2 root 6750: C code for a data type which is used for declaring the @code{mdep}
6751: component of @code{cc_status}. It defaults to @code{int}.
1.1 root 6752:
6753: @item CC_STATUS_MDEP_INIT
1.1.1.2 root 6754: A C expression for the initial value of the @code{mdep} field. It
6755: defaults to 0.
1.1 root 6756:
6757: @item NOTICE_UPDATE_CC (@var{exp})
6758: A C compound statement to set the components of @code{cc_status}
1.1.1.2 root 6759: appropriately for an insn whose body is @var{exp}. It is this macro's
6760: responsibility to recognize insns that set the condition code as a
6761: byproduct of other activity as well as those that explicitly set
6762: @code{(cc0)}.
6763:
6764: If there are insn that do not set the condition code but do alter
6765: other machine registers, this macro must check to see whether they
6766: invalidate the expressions that the condition code is recorded as
6767: reflecting. For example, on the 68000, insns that store in address
6768: registers do not set the condition code, which means that usually
6769: @code{NOTICE_UPDATE_CC} can leave @code{cc_status} unaltered for such
6770: insns. But suppose that the previous insn set the condition code
6771: based on location @samp{a4@@(102)} and the current insn stores a new
6772: value in @samp{a4}. Although the condition code is not changed by
6773: this, it will no longer be true that it reflects the contents of
6774: @samp{a4@@(102)}. Therefore, @code{NOTICE_UPDATE_CC} must alter
1.1 root 6775: @code{cc_status} in this case to say that nothing is known about the
6776: condition code value.
6777: @end table
6778:
6779: @node Assembler Format,, Condition Code, Machine Macros
6780: @section Output of Assembler Code
6781:
6782: @table @code
1.1.1.2 root 6783: @item ASM_SPEC
6784: A C string constant that tells the GNU CC driver program options to
6785: pass to the assembler. It can also specify how to translate options
6786: you give to GNU CC into options for GNU CC to pass to the assembler.
6787: See the file @file{tm-sun3.h} for an example of this.
6788:
6789: Do not define this macro if it does not need to do anything.
6790:
6791: @item LINK_SPEC
6792: A C string constant that tells the GNU CC driver program options to
6793: pass to the linker. It can also specify how to translate options you
6794: give to GNU CC into options for GNU CC to pass to the linker.
6795:
6796: Do not define this macro if it does not need to do anything.
6797:
1.1.1.4 ! root 6798: @item ASM_FILE_START (@var{stream})
! 6799: A C expression which outputs to the stdio stream @var{stream}
! 6800: some appropriate text to go at the start of an assembler file.
! 6801:
! 6802: Normally this macro is defined to output a line containing
! 6803: @samp{#NO_APP}, which is a comment that has no effect on most
! 6804: assemblers but tells the GNU assembler that it can save time by not
! 6805: checking for certain assembler constructs.
! 6806:
! 6807: On systems that use SDB, it is necessary to output certain commands;
! 6808: see @file{tm-attasm.h}.
1.1.1.2 root 6809:
6810: @item ASM_APP_ON
6811: A C string constant for text to be output before each @code{asm}
6812: statement or group of consecutive ones. Normally this is
6813: @code{"#APP"}, which is a comment that has no effect on most
6814: assemblers but tells the GNU assembler that it must check the lines
6815: that follow for all valid assembler constructs.
6816:
6817: @item ASM_APP_OFF
6818: A C string constant for text to be output after each @code{asm}
6819: statement or group of consecutive ones. Normally this is
6820: @code{"#NO_APP"}, which tells the GNU assembler to resume making the
6821: time-saving assumptions that are valid for ordinary compiler output.
6822:
1.1 root 6823: @item TEXT_SECTION_ASM_OP
6824: A C string constant for the assembler operation that should precede
6825: instructions and read-only data. Normally @code{".text"} is right.
6826:
6827: @item DATA_SECTION_ASM_OP
1.1.1.2 root 6828: A C string constant for the assembler operation to identify the
6829: following data as writable initialized data. Normally @code{".data"}
6830: is right.
1.1 root 6831:
6832: @item REGISTER_NAMES
1.1.1.2 root 6833: A C initializer containing the assembler's names for the machine
6834: registers, each one as a C string constant. This is what translates
6835: register numbers in the compiler into assembler language.
1.1 root 6836:
6837: @item DBX_REGISTER_NUMBER (@var{regno})
1.1.1.2 root 6838: A C expression that returns the DBX register number for the compiler
6839: register number @var{regno}. In simple cases, the value of this
6840: expression may be @var{regno} itself. But sometimes there are some
6841: registers that the compiler knows about and DBX does not, or vice
6842: versa. In such cases, some register may need to have one number in
6843: the compiler and another for DBX.
6844:
1.1.1.4 ! root 6845: @item DBX_DEBUGGING_INFO
! 6846: Define this macro if GNU CC should produce debugging output for DBX
! 6847: in response to the @samp{-g} option.
! 6848:
! 6849: @item SDB_DEBUGGING_INFO
! 6850: Define this macro if GNU CC should produce debugging output for SDB
! 6851: in response to the @samp{-g} option.
! 6852:
1.1.1.2 root 6853: @item DBX_NO_XREFS
6854: Define this macro if DBX on your system does not support the construct
6855: @samp{xs@var{tagname}}. On some systems, this construct is used to
6856: describe a forward reference to a structure named @var{tagname}.
6857: On other systems, this construct is not supported at all.
6858:
6859: @item DBX_CONTIN_LENGTH
6860: A symbol name in DBX-format debugging information is normally
6861: continued (split into two separate @code{.stabs} directives) when it
6862: exceeds a certain length (by default, 80 characters). On some
6863: operating systems, DBX requires this splitting; on others, splitting
6864: must not be done. You can inhibit splitting by defining this macro
6865: with the value zero. You can override the default splitting-length by
6866: defining this macro as an expression for the length you desire.
6867:
6868: @item DBX_CONTIN_CHAR
6869: Normally continuation is indicated by adding a @samp{\} character to
6870: the end of a @code{.stabs} string when a continuation follows. To use
6871: a different character instead, define this macro as a character
6872: constant for the character you want to use. Do not define this macro
6873: if backslash is correct for your system.
6874:
1.1.1.4 ! root 6875: @item ASM_OUTPUT_LABEL (@var{stream}, @var{name})
1.1.1.2 root 6876: A C statement (sans semicolon) to output to the stdio stream
1.1.1.4 ! root 6877: @var{stream} the assembler definition of a label named @var{name}. Use
! 6878: the expression @code{assemble_name (@var{stream}, @var{name})} to output
1.1.1.2 root 6879: the name itself; before and after that, output the additional
6880: assembler syntax for defining the name, and a newline.
6881:
1.1.1.4 ! root 6882: @item ASM_DECLARE_FUNCTION_NAME (@var{stream}, @var{name})
1.1.1.2 root 6883: A C statement (sans semicolon) to output to the stdio stream
1.1.1.4 ! root 6884: @var{stream} any text necessary for declaring the name of a function
1.1.1.2 root 6885: which is being defined. This macro is responsible for outputting
6886: the label definition (perhaps using @code{ASM_OUTPUT_LABEL}).
6887:
6888: If this macro is not defined, then the function name is defined in the
6889: usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}).
6890:
1.1.1.4 ! root 6891: @item ASM_GLOBALIZE_LABEL (@var{stream}, @var{name})
1.1.1.2 root 6892: A C statement (sans semicolon) to output to the stdio stream
1.1.1.4 ! root 6893: @var{stream} some commands that will make the label @var{name} global;
1.1.1.2 root 6894: that is, available for reference from other files. Use the expression
1.1.1.4 ! root 6895: @code{assemble_name (@var{stream}, @var{name})} to output the name
1.1.1.2 root 6896: itself; before and after that, output the additional assembler syntax
6897: for making that name global, and a newline.
6898:
1.1.1.4 ! root 6899: @item ASM_OUTPUT_EXTERNAL (@var{stream}, @var{name})
1.1.1.2 root 6900: A C statement (sans semicolon) to output to the stdio stream
1.1.1.4 ! root 6901: @var{stream} any text necessary for declaring the name of an external
1.1.1.2 root 6902: symbol which is referenced in this compilation but not defined.
6903:
6904: This macro need not be defined if it does not need to output anything.
6905: The GNU assembler and most Unix assemblers don't require anything.
6906:
1.1.1.4 ! root 6907: @item ASM_OUTPUT_LABELREF (@var{stream}, @var{name})
! 6908: A C statement to output to the stdio stream @var{stream} a reference in
1.1.1.2 root 6909: assembler syntax to a label named @var{name}. The character @samp{_}
6910: should be added to the front of the name, if that is customary on your
6911: operating system, as it is in most Berkeley Unix systems. This macro
6912: is used in @code{assemble_name}.
6913:
1.1.1.4 ! root 6914: @item ASM_OUTPUT_INTERNAL_LABEL (@var{stream}, @var{prefix}, @var{num})
! 6915: A C statement to output to the stdio stream @var{stream} a label whose
1.1.1.2 root 6916: name is made from the string @var{prefix} and the number @var{num}.
6917: These labels are used for internal purposes, and there is no reason
6918: for them to appear in the symbol table of the object file. On many
6919: systems, the letter @samp{L} at the beginning of a label has this
6920: effect. The usual definition of this macro is as follows:
6921:
6922: @example
1.1.1.4 ! root 6923: fprintf (@var{stream}, "L%s%d:\n", @var{prefix}, @var{num})
1.1.1.2 root 6924: @end example
6925:
1.1.1.4 ! root 6926: @item ASM_OUTPUT_CASE_LABEL (@var{stream}, @var{prefix}, @var{num}, @var{table})
1.1.1.2 root 6927: Define this if the label before a jump-table needs to be output
6928: specially. The first three arguments are the same as for
6929: @code{ASM_OUTPUT_INTERNAL_LABEL}; the fourth argument is the
6930: jump-table which follows (a @samp{jump_insn} containing an
6931: @samp{addr_vec} or @samp{addr_diff_vec}).
6932:
6933: This feature is used on system V to output a @code{swbeg} statement
6934: for the table.
6935:
6936: If this macro is not defined, these labels are output with
6937: @code{ASM_OUTPUT_INTERNAL_LABEL}.
6938:
1.1.1.4 ! root 6939: @item ASM_OUTPUT_CASE_END (@var{stream}, @var{num}, @var{table})
! 6940: Define this if something special must be output at the end of a jump-table.
! 6941: The definition should be a C statement to be executed after the assembler
! 6942: code for the table is written. It should write the appropriate code to
! 6943: stdio stream @var{stream}. The argument @var{table} is the jump-table
! 6944: insn, and @var{num} is the label-number of the preceding label.
! 6945:
! 6946: If this macro is not defined, nothing special is output at the end of
! 6947: the jump-table.
! 6948:
1.1.1.2 root 6949: @item ASM_FORMAT_PRIVATE_NAME (@var{outvar}, @var{name}, @var{number})
6950: A C expression to assign to @var{outvar} (which is a variable of type
6951: @code{char *}) a newly allocated string made from the string
6952: @var{name} and the number @var{number}, with some suitable punctuation
6953: added. Use @code{alloca} to get space for the string.
6954:
6955: This string will be used as the argument to @code{ASM_OUTPUT_LABELREF}
6956: to produce an assembler label for an internal static variable whose
6957: name is @var{name}. Therefore, the string must be such as to result
6958: in valid assembler code. The argument @var{number} is different each
6959: time this macro is executed; it prevents conflicts between
6960: similarly-named internal static variables in different scopes.
6961:
6962: Ideally this string should not be a valid C identifier, to prevent any
6963: conflict with the user's own symbols. Most assemblers allow periods
6964: or percent signs in assembler symbols; putting at least one of these
6965: between the name and the number will suffice.
6966:
1.1.1.4 ! root 6967: @item ASM_OUTPUT_ADDR_DIFF_ELT (@var{stream}, @var{value}, @var{rel})
1.1.1.2 root 6968: This macro should be provided on machines where the addresses
6969: in a dispatch table are relative to the table's own address.
6970:
6971: The definition should be a C statement to output to the stdio stream
1.1.1.4 ! root 6972: @var{stream} an assembler pseudo-instruction to generate a difference
1.1.1.2 root 6973: between two labels. @var{value} and @var{rel} are the numbers of two
6974: internal labels. The definitions of these labels are output using
6975: @code{ASM_OUTPUT_INTERNAL_LABEL}, and they must be printed in the same
6976: way here. For example,
6977:
6978: @example
1.1.1.4 ! root 6979: fprintf (@var{stream}, "\t.word L%d-L%d\n",
1.1.1.2 root 6980: @var{value}, @var{rel})
6981: @end example
6982:
1.1.1.4 ! root 6983: @item ASM_OUTPUT_ADDR_VEC_ELT (@var{stream}, @var{value})
1.1.1.2 root 6984: This macro should be provided on machines where the addresses
6985: in a dispatch table are absolute.
6986:
6987: The definition should be a C statement to output to the stdio stream
1.1.1.4 ! root 6988: @var{stream} an assembler pseudo-instruction to generate a reference to
1.1.1.2 root 6989: a label. @var{value} is the number of an internal label whose
6990: definition is output using @code{ASM_OUTPUT_INTERNAL_LABEL}.
6991: For example,
6992:
6993: @example
1.1.1.4 ! root 6994: fprintf (@var{stream}, "\t.word L%d\n", @var{value})
1.1.1.2 root 6995: @end example
1.1 root 6996:
1.1.1.4 ! root 6997: @item ASM_OUTPUT_DOUBLE (@var{stream}, @var{value})
! 6998: A C statement to output to the stdio stream @var{stream} an assembler
1.1 root 6999: instruction to assemble a @code{double} constant whose value is
1.1.1.2 root 7000: @var{value}. @var{value} will be a C expression of type
7001: @code{double}.
1.1 root 7002:
1.1.1.4 ! root 7003: @item ASM_OUTPUT_FLOAT (@var{stream}, @var{value})
! 7004: A C statement to output to the stdio stream @var{stream} an assembler
1.1.1.2 root 7005: instruction to assemble a @code{float} constant whose value is
7006: @var{value}. @var{value} will be a C expression of type @code{float}.
7007:
1.1.1.4 ! root 7008: @item ASM_OUTPUT_INT (@var{stream}, @var{exp})
! 7009: @itemx ASM_OUTPUT_SHORT (@var{stream}, @var{exp})
! 7010: @itemx ASM_OUTPUT_CHAR (@var{stream}, @var{exp})
! 7011: A C statement to output to the stdio stream @var{stream} an assembler
1.1.1.2 root 7012: instruction to assemble a @code{int}, @code{short} or @code{char}
7013: constant whose value is @var{value}. The argument @var{exp} will be
7014: an RTL expression which represents a constant value. Use
7015: @samp{output_addr_const (@var{exp})} to output this value as an
7016: assembler expression.@refill
7017:
1.1.1.4 ! root 7018: @item ASM_OUTPUT_BYTE (@var{stream}, @var{value})
! 7019: A C statement to output to the stdio stream @var{stream} an assembler
1.1.1.2 root 7020: instruction to assemble a single byte containing the number @var{value}.
7021:
1.1.1.4 ! root 7022: @item ASM_OUTPUT_ASCII (@var{stream}, @var{ptr}, @var{len})
! 7023: A C statement to output to the stdio stream @var{stream} an assembler
1.1.1.2 root 7024: instruction to assemble a string constant containing the @var{len}
7025: bytes at @var{ptr}. @var{ptr} will be a C expression of type
7026: @code{char *} and @var{len} a C expression of type @code{int}.
7027:
7028: If the assembler has a @code{.ascii} pseudo-op as found in the
7029: Berkeley Unix assembler, do not define the macro
7030: @code{ASM_OUTPUT_ASCII}.
1.1 root 7031:
1.1.1.4 ! root 7032: @item ASM_OUTPUT_SKIP (@var{stream}, @var{nbytes})
! 7033: A C statement to output to the stdio stream @var{stream} an assembler
1.1 root 7034: instruction to advance the location counter by @var{nbytes} bytes.
7035: @var{nbytes} will be a C expression of type @code{int}.
7036:
1.1.1.4 ! root 7037: @item ASM_OUTPUT_ALIGN (@var{stream}, @var{power})
! 7038: A C statement to output to the stdio stream @var{stream} an assembler
1.1 root 7039: instruction to advance the location counter to a multiple of 2 to the
7040: @var{power} bytes. @var{power} will be a C expression of type @code{int}.
7041:
1.1.1.4 ! root 7042: @item ASM_OUTPUT_COMMON (@var{stream}, @var{name}, @var{size})
1.1.1.2 root 7043: A C statement (sans semicolon) to output to the stdio stream
1.1.1.4 ! root 7044: @var{stream} the assembler definition of a common-label named @var{name}
1.1.1.2 root 7045: whose size is @var{size} bytes. Use the expression
1.1.1.4 ! root 7046: @code{assemble_name (@var{stream}, @var{name})} to output the name
1.1.1.2 root 7047: itself; before and after that, output the additional assembler syntax
7048: for defining the name, and a newline.
7049:
7050: This macro controls how the assembler definitions of uninitialized
7051: global variables are output.
7052:
1.1.1.4 ! root 7053: @item ASM_OUTPUT_LOCAL (@var{stream}, @var{name}, @var{size})
1.1.1.2 root 7054: A C statement (sans semicolon) to output to the stdio stream
1.1.1.4 ! root 7055: @var{stream} the assembler definition of a local-common-label named
1.1.1.2 root 7056: @var{name} whose size is @var{size} bytes. Use the expression
1.1.1.4 ! root 7057: @code{assemble_name (@var{stream}, @var{name})} to output the name
1.1.1.2 root 7058: itself; before and after that, output the additional assembler syntax
7059: for defining the name, and a newline.
7060:
7061: This macro controls how the assembler definitions of uninitialized
7062: static variables are output.
1.1 root 7063:
1.1.1.4 ! root 7064: @item ASM_OUTPUT_SOURCE_LINE (@var{stream}, @var{line})
! 7065: A C statment to output DBX or SDB debugging information before code
! 7066: for line number @var{line} of the current source file to the
! 7067: stdio stream @var{stream}.
! 7068:
! 7069: This macro need not be defined if the standard form of debugging
! 7070: information for the debugger in use is appropriate.
! 7071:
1.1 root 7072: @item TARGET_BELL
1.1.1.2 root 7073: A C constant expression for the integer value for escape sequence
7074: @samp{\a}.
1.1 root 7075:
7076: @item TARGET_BS
7077: @itemx TARGET_TAB
7078: @itemx TARGET_NEWLINE
7079: C constant expressions for the integer values for escape sequences
7080: @samp{\b}, @samp{\t} and @samp{\n}.
7081:
7082: @item TARGET_VT
7083: @itemx TARGET_FF
7084: @itemx TARGET_CR
7085: C constant expressions for the integer values for escape sequences
7086: @samp{\v}, @samp{\f} and @samp{\r}.
7087:
1.1.1.4 ! root 7088: @item ASM_OUTPUT_OPCODE (@var{stream}, @var{ptr})
1.1.1.2 root 7089: Define this macro if you are using an unusual assembler that
7090: requires different names for the machine instructions.
7091:
7092: The definition is a C statement or statements which output an
1.1.1.4 ! root 7093: assembler instruction opcode to the stdio stream @var{stream}. The
1.1.1.2 root 7094: macro-operand @var{ptr} is a variable of type @code{char *} which
7095: points to the opcode name in its ``internal'' form---the form that is
7096: written in the machine description. The definition should output the
1.1.1.4 ! root 7097: opcode name to @var{stream}, performing any translation you desire, and
1.1.1.2 root 7098: increment the variable @var{ptr} to point at the end of the opcode
7099: so that it will not be output twice.
7100:
7101: In fact, your macro definition may process less than the entire opcode
7102: name, or more than the opcode name; but if you want to process text
7103: that includes @samp{%}-sequences to substitute operands, you must take
7104: care of the substitution yourself. Just be sure to increment
7105: @var{ptr} over whatever text should not be output normally.
7106:
7107: If the macro definition does nothing, the instruction is output
7108: in the usual way.
7109:
1.1.1.4 ! root 7110: @item FINAL_PRESCAN_INSN (@var{insn}, @var{opvec}, @var{noperands})
! 7111: If defined, a C statement to be executed just prior to the output of
! 7112: assembler code for @var{insn}, to modify the extracted operands so
! 7113: they will be output differently.
! 7114:
! 7115: Here the argument @var{opvec} is the vector containing the operands
! 7116: extracted from @var{insn}, and @var{noperands} is the number of
! 7117: elements of the vector which contain meaningful data for this insn.
! 7118: The contents of this vector are what will be used to convert the insn
! 7119: template into assembler code, so you can change the assembler output
! 7120: by changing the contents of the vector.
! 7121:
! 7122: This macro is useful when various assembler syntaxes share a single
! 7123: file of instruction patterns; by defining this macro differently, you
! 7124: can cause a large class of instructions to be output differently (such
! 7125: as with rearranged operands). Naturally, variations in assembler
! 7126: syntax affecting individual insn patterns ought to be handled by
! 7127: writing conditional output routines in those patterns.
! 7128:
! 7129: If this macro is not defined, it is equivalent to a null statement.
! 7130:
! 7131: @item PRINT_OPERAND (@var{stream}, @var{x}, @var{code})
! 7132: A C compound statement to output to stdio stream @var{stream} the
1.1.1.2 root 7133: assembler syntax for an instruction operand @var{x}. @var{x} is an
7134: RTL expression.
7135:
7136: @var{code} is a value that can be used to specify one of several ways
7137: of printing the operand. It is used when identical operands must be
7138: printed differently depending on the context. @var{code} comes from
7139: the @samp{%} specification that was used to request printing of the
7140: operand. If the specification was just @samp{%@var{digit}} then
7141: @var{code} is 0; if the specification was @samp{%@var{ltr}
7142: @var{digit}} then @var{code} is the ASCII code for @var{ltr}.
7143:
7144: If @var{x} is a register, this macro should print the register's name.
7145: The names can be found in an array @code{reg_names} whose type is
7146: @code{char *[]}. @code{reg_names} is initialized from
7147: @code{REGISTER_NAMES}.
7148:
7149: When the machine description has a specification @samp{%@var{punct}}
7150: (a @samp{%} followed by a punctuation character), this macro is called
7151: with a null pointer for @var{x} and the punctuation character for
7152: @var{code}.
1.1 root 7153:
1.1.1.4 ! root 7154: @item PRINT_OPERAND_ADDRESS (@var{stream}, @var{x})
! 7155: A C compound statement to output to stdio stream @var{stream} the
1.1.1.2 root 7156: assembler syntax for an instruction operand that is a memory reference
7157: whose address is @var{x}. @var{x} is an RTL expression.
7158:
7159: @item ASM_OPEN_PAREN
7160: @itemx ASM_CLOSE_PAREN
7161: These macros are defined as C string constant, describing the syntax
7162: in the assembler for grouping arithmetic expressions. The following
7163: definitions are correct for most assemblers:
7164:
7165: @example
7166: #define ASM_OPEN_PAREN "("
7167: #define ASM_CLOSE_PAREN ")"
7168: @end example
7169: @end table
7170:
7171: @node Config,, Machine Macros, Top
7172: @chapter The Configuration File
7173:
7174: The configuration file @file{config-@var{machine}.h} contains macro
7175: definitions that describe the machine and system on which the compiler is
7176: running. Most of the values in it are actually the same on all machines
7177: that GNU CC runs on, so most all configuration files are identical. But
7178: there are some macros that vary:
7179:
7180: @table @code
7181: @item FAILURE_EXIT_CODE
7182: A C expression for the status code to be returned when the compiler
7183: exits after serious errors.
7184:
7185: @item SUCCESS_EXIT_CODE
7186: A C expression for the status code to be returned when the compiler
7187: exits without serious errors.
1.1 root 7188: @end table
7189:
7190: @contents
7191: @bye
This archive runs on limited infrastructure. Preserving old code on modern bandwidth. Automated agents are requested to crawl responsibly.