![[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
42: @center last updated 24 April 1988
43: @sp 1
44: @center for version 1.21
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
281: function integration.
282:
283: @item
1.1.1.3 ! root 284: Robert Brown implemented the support for Encore 32000 systems.
! 285:
! 286: @item
1.1.1.2 root 287: Michael Kashtan of SRI adapted GNU CC to the Vomit-Making System.
288:
289: @item
290: Alex Crain provided changes for the 3b1.
291:
292: @item
293: Chris Hanson and another person who should remind me of his name
294: assisted in making GNU CC work on HP-UX for the 9000 series 300.
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
452: Produce debugging information in DBX format.
453:
454: Unlike most other C compilers, GNU CC allows you to use @samp{-g} with
455: @samp{-O}. The shortcuts taken by optimized code may occasionally
456: produce surprising results: some variables you declared may not exist
457: at all; flow of control may briefly move where you did not expect it;
458: some statements may not be executed because they compute constant
459: results or their values were already at hand; some statements may
460: execute in different places because they were moved out of loops.
461: Nevertheless it proves possible to debug optimized output. This makes
462: it reasonable to use the optimizer for programs that might have bugs.
463:
464: @item -gg
465: Produce debugging information in GDB's own format. This requires
466: the GNU assembler and linker in order to work.
467:
468: @item -w
469: Inhibit all warning messages.
470:
471: @item -W
472: Print extra warning messages for these events:
473:
474: @itemize @bullet
475: @item
476: An automatic variable is used without first being initialized.
477:
478: These warnings are possible only in optimizing compilation,
479: because they require data flow information that is computed only
480: when optimizing. They occur only for variables that are
481: candidates for register allocation. Therefore, they do not occur
482: for a variable that is declared @code{volatile}, or whose address
483: is taken, or whose size is other than 1, 2, 4 or 8 bytes. Also,
484: they do not occur for structures, unions or arrays, even when
485: they are in registers.
486:
487: Note that there may be no warning about a variable that is used
488: only to compute a value that itself is never used, because such
489: computations may be deleted by the flow analysis pass before the
490: warnings are printed.
491:
492: These warnings are made optional because GNU CC is not smart
493: enough to see all the reasons why the code might be correct
494: despite appearing to have an error. Here is one example of how
495: this can happen:
496:
497: @example
498: @{
499: int x;
500: switch (y)
501: @{
502: case 1: x = 1;
503: break;
504: case 2: x = 4;
505: break;
506: case 3: x = 5;
507: @}
508: foo (x);
509: @}
510: @end example
511:
512: @noindent
513: If the value of @code{y} is always 1, 2 or 3, then @code{x} is
514: always initialized, but GNU CC doesn't know this. Here is
515: another common case:
516:
517: @example
518: @{
519: int save_y;
520: if (change_y) save_y = y, y = new_y;
521: @dots{}
522: if (change_y) y = save_y;
523: @}
524: @end example
525:
526: @noindent
527: This has no bug because @code{x} is used only if it is set.
528:
529: @item
530: A nonvolatile automatic variable might be changed by a call to
531: @code{longjmp}. These warnings as well are possible only in
532: optimizing compilation.
533:
534: The compiler sees only the calls to @code{setjmp}. It cannot know
535: where @code{longjmp} will be called; in fact, a signal handler could
536: call it at any point in the code. As a result, you may get a warning
537: even when there is in fact no problem because @code{longjmp} cannot
538: in fact be called at the place which would cause a problem.
539:
540: @item
541: A function can return either with or without a value. (Falling
542: off the end of the function body is considered returning without
543: a value.) For example, this function would inspire such a
544: warning:
545:
546: @example
547: foo (a)
548: @{
549: if (a > 0)
550: return a;
551: @}
552: @end example
553:
554: Spurious warnings can occur because GNU CC does not realize that
555: certain functions (including @code{abort} and @code{longjmp})
556: will never return.
557: @end itemize
558:
559: In the future, other useful warnings may also be enabled by this
560: option.
561:
562: @item -Wimplicit
563: Warn whenever a function is implicitly declared.
564:
565: @item -Wreturn-type
566: Warn whenever a function is defined with a return-type that defaults
567: to @code{int}. Also warn about any @code{return} statement with no
568: return-value in a function whose return-type is not @code{void}.
569:
570: @item -Wcomment
571: Warn whenever a comment-start sequence @samp{/*} appears in a comment.
572:
1.1.1.3 ! root 573: @item -Wall
! 574: All of the above @samp{-W} options combined.
! 575:
1.1.1.2 root 576: @item -p
577: Generate extra code to write profile information suitable for the
578: analysis program @code{prof}.
579:
580: @item -pg
581: Generate extra code to write profile information suitable for the
582: analysis program @code{gprof}.
583:
1.1.1.3 ! root 584: @item -l@var{library}
! 585: Search a standard list of directories for a library named
! 586: @var{library}, which is actually a file named
! 587: @file{lib@var{library}.a}. The linker uses this file as if it
! 588: had been specified precisely by name.
! 589:
! 590: The directories searched include several standard system directories
! 591: plus any that you specify with @samp{-L}.
! 592:
! 593: Normally the files found this way are library files---archive files
! 594: whose members are object files. The linker handles an archive file by
! 595: through it for members which define symbols that have so far been
! 596: referenced but not defined. But if the file that is found is an
! 597: ordinary object file, it is linked in the usual fashion. The only
! 598: difference between an @samp{-l} option and the full file name of the
! 599: file that is found is syntactic and the fact that several directories
! 600: are searched.
! 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
951: Edit @file{Makefile}. If you are using HPUX, you must make a few
952: changes described in comments at the beginning of the file.
953:
954: @item
955: Choose configuration files.
956:
957: @itemize @bullet
958: @item
959: Make a symbolic link named @file{config.h} to the top-level
960: config file for the machine you are using (@pxref{Config}). This
961: file is responsible for defining information about the host
962: machine. It includes @file{tm.h}.
963:
964: The file's name should be @file{config-@var{machine}.h}. On VMS,
965: use @file{config-vms.h} rather than @file{config-vax.h}. On the
966: HP 9000 series 300, use @file{config-hp9k3.h} rather than
967: @file{config-m68k.h}.@refill
968:
969: If your system does not support symbolic links, you might want to
970: set up @file{config.h} to contain a @samp{#include} command which
971: refers to the appropriate file.
972:
973: @item
974: Make a symbolic link named @file{tm.h} to the machine-description
975: macro file for your machine (its name should be
976: @file{tm-@var{machine}.h}).
977:
978: For the 68000/68020, do not use @file{tm-m68k.h} directly;
979: instead use one of the files @file{tm-sun3.h}, @file{tm-sun2.h},
980: @file{tm-isi68.h}, @file{tm-news800.h} or @file{tm-3b1.h}. Each
981: of those files includes @file{tm-m68k.h} but sets up a few things
982: differently as appropriate to the specific model of
983: machine.@refill
984:
985: There are two files you can use for a 680x0 running HPUX:
986: @file{tm-hp9k320.h} and @file{tm-hp9k320g.h}. Use the former if
987: you are installing GNU CC alone. The latter is for another option
988: where GNU CC together with the GNU assembler, linker, debugger
989: and other utilities are used to replace all of HPUX that deals
990: with compilation. Not all of the pieces of GNU software needed for
991: this mode of operation are as yet in distribution; full instructions
992: will appear here in the future.@refill
993:
994: For the 32000, use @file{tm-sequent.h} if you are using a Sequent
1.1.1.3 ! root 995: machine, or @file{tm-encore.h} for an Encore machine; otherwise,
! 996: perhaps @file{tm-ns32k.h} will work for you.
1.1.1.2 root 997:
1.1.1.3 ! root 998: For the vax, use @file{tm-vax.h} on Unix, or @file{tm-vms.h} on
! 999: VMS.@refill
1.1.1.2 root 1000:
1001: @item
1002: Make a symbolic link named @file{md} to the machine description
1003: pattern file (its name should be @file{@var{machine}.md}).
1004:
1005: @item
1006: Make a symbolic link named @file{aux-output.c} to the output
1007: subroutine file for your machine (its name should be
1008: @file{output-@var{machine}.c}).
1009: @end itemize
1010:
1011: @item
1012: Make sure the Bison parser generator is installed. (This is
1013: unnecessary if the Bison output file @file{parse.tab.c} is more recent
1014: than @file{parse.y} and you do not plan to change @file{parse.y}.)
1015:
1016: Note that if you have an old version of Bison you may get an error
1017: from the line with the @samp{%expect} directive. If so, simply remove
1018: that line from @file{parse.y} and proceed.
1019:
1020: @item
1021: If you are using a Sun, make sure the environment variable
1022: @code{FLOAT_OPTION} is not set. If this option were set to
1023: @code{f68881} when @file{gnulib} is compiled, the resulting code would
1024: demand to be linked with a special startup file and will not link
1025: properly without special pains.
1026:
1027: @item
1028: Build the compiler. Just type @samp{make} in the compiler directory.
1029:
1030: @item
1031: Move the first-stage object files and executables into a subdirectory
1032: with this command:
1033:
1034: @example
1035: make stage1
1036: @end example
1037:
1038: The files are moved into a subdirectory named @file{stage1}.
1039: Once installation is complete, you may wish to delete these files
1040: with @code{rm -r stage1}.
1041:
1042: @item
1043: Recompile the compiler with itself, with this command:
1044:
1045: @example
1046: make CC=stage1/gcc CFLAGS="-g -O -Bstage1/"
1047: @end example
1048:
1049: On a 68000 or 68020 system lacking floating point hardware,
1050: unless you have selected a @file{tm.h} file that expects by default
1051: that there is no such hardware, do this instead:
1052:
1053: @example
1054: make CC=stage1/gcc CFLAGS="-g -O -Bstage1/ -msoft-float"
1055: @end example
1056:
1057: @item
1058: If you wish to test the compiler by compiling it with itself one more
1059: time, do this:
1060:
1061: @example
1062: make stage2
1063: make CC=stage2/gcc CFLAGS="-g -O -Bstage2/"
1064: foreach file (*.o)
1065: cmp $file stage2/$file
1066: end
1067: @end example
1068:
1069: This will notify you if any of these stage 3 object files differs from
1070: those of stage 2. Any difference, no matter how innocuous, indicates
1071: that the stage 2 compiler has compiled GNU CC incorrectly, and is
1072: therefore a potentially serious bug which you should investigate and
1073: report (@pxref{Bugs}).
1074:
1075: @item
1076: Install the compiler driver, the compiler's passes and run-time support.
1077: You can use the following command:
1078:
1079: @example
1080: make install
1081: @end example
1082:
1083: @noindent
1084: This copies the files @file{cc1}, @file{cpp} and @file{gnulib} to
1085: files @file{gcc-cc1}, @file{gcc-cpp} and @file{gcc-gnulib} in
1086: directory @file{/usr/local/lib}, which is where the compiler driver
1087: program looks for them. It also copies the driver program @file{gcc}
1088: into the directory @file{/usr/local}, so that it appears in typical
1089: execution search paths.@refill
1090:
1091: @strong{Warning: the GNU CPP may not work for @file{ioctl.h},
1092: @file{ttychars.h} and other system header files unless the
1093: @samp{-traditional} option is used.} The bug is in the header files:
1094: at least on some machines, they rely on behavior that is incompatible
1095: with ANSI C. This behavior consists of substituting for macro
1096: argument names when they appear inside of character constants. The
1097: @samp{-traditional} option tells GNU CC to behave the way these
1098: headers expect.
1099:
1100: Because of this problem, you might prefer to configure GNU CC to use
1101: the system's own C preprocessor. To do so, make the file
1102: @file{/usr/local/lib/gcc-cpp} a link to @file{/lib/cpp}.
1103:
1104: Alternatively, on Sun systems and 4.3BSD at least, you can correct the
1105: include files by running the shell script @file{fixincludes}. This
1106: installs modified, corrected copies of the files @file{ioctl.h} and
1107: @file{ttychars.h} in a special directory where only GNU CC will
1108: normally look for them.
1109:
1110: The file @file{/usr/include/vaxuba/qvioctl.h} used in the X window
1111: system needs a similar correction.
1112: @end enumerate
1113:
1114: If you cannot install the compiler's passes and run-time support in
1115: @file{/usr/local/lib}, you can alternatively use the @samp{-B} option to
1116: specify a prefix by which they may be found. The compiler concatenates
1117: the prefix with the names @file{cpp}, @file{cc1} and @file{gnulib}.
1118: Thus, you can put the files in a directory @file{/usr/foo/gcc} and
1119: specify @samp{-B/usr/foo/gcc/} when you run GNU CC.
1120:
1121: @node VMS Install,, Installation, Installation
1122: @section Installing GNU CC on VMS
1123:
1.1.1.3 ! root 1124: The VMS version of GNU CC is distributed in an unusual tape format which
! 1125: consists of several tape files. The first is a command file; the second is
! 1126: an executable program which reads Unix tar format; the third is another
! 1127: command file which uses this program to read the remainder of the tape.
! 1128:
! 1129: To load the tape, it suffices to mount it @samp{/foreign} and then do
! 1130: @samp{@@mta0:} to execute the command file at the beginning of the tape.
! 1131:
! 1132: The tape contains executables and object files as well as sources, so no
! 1133: compilation is necessary unless you change the sources. (This is a good
! 1134: thing, since you probably don't have any other C compiler.) If you must
! 1135: recompile, here is how:
1.1.1.2 root 1136:
1137: @enumerate
1138: @item
1139: Copy the file @file{tm-vms.h} to @file{tm.h}, @file{config-vms.h} to
1140: @file{config.h}, @file{vax.md} to @file{md.} and @file{output-vax.c}
1141: to @file{aux-output.c}.@refill
1142:
1143: @item
1144: Type @samp{@@make} to do recompile everything.
1145: @end enumerate
1146:
1.1.1.3 ! root 1147: To install the @samp{GCC} command so you can use the compiler easily, in
! 1148: the same manner as you use the VMS C compiler, you must install the VMS CLD
! 1149: file for GNU CC as follows:
! 1150:
! 1151: @enumerate
! 1152: @item
! 1153: Define the VMS logical names @samp{GNU_CC} and @samp{GNU_CC_INCLUDE}
! 1154: to point to the directories where the GNU CC executables
! 1155: (@samp{gcc-cpp}, @samp{gcc-cc1}, etc.) and the C include files are
! 1156: kept. This should be done with the commands:@refill
! 1157:
! 1158: @example
! 1159: $ assign /super /system disk:[gcc] gnu_cc
! 1160: $ assign /super /system disk:[gcc.include] gnu_cc_include
! 1161: @end example
! 1162:
! 1163: @noindent
! 1164: with the appropriate disk and directory names. These commands can be
! 1165: placed in your system startup file so they will be executed whenever
! 1166: the machine is rebooted.
! 1167:
! 1168: @item
! 1169: Install the @samp{GCC} command with the command line:
! 1170:
! 1171: @example
! 1172: $ set command /table=sys$library:dcltables gnu_cc:gcc
! 1173: @end example
! 1174:
! 1175: @noindent
! 1176: Now you can invoke the compiler with a command like @samp{gcc /verbose
! 1177: file.c}, which is equivalent to the command @samp{gcc -v -c file.c} in
! 1178: Unix.
! 1179: @end enumerate
! 1180:
! 1181: @node Trouble, Incompatibilities, Installation, Top
! 1182: @chapter Trouble in Installation
! 1183:
! 1184: Here are some of the things that have caused trouble for people installing
! 1185: GNU CC.
! 1186:
! 1187: @itemize
! 1188: @item
! 1189: On certain systems, defining certain environment variables such as
! 1190: @samp{CC} can interfere with the functioning of @code{make}.
! 1191: @end itemize
! 1192:
! 1193: @node Incompatibilities, Extensions, Trouble, Top
1.1.1.2 root 1194: @chapter Incompatibilities of GNU CC
1195:
1196: There are several noteworthy incompatibilities between GNU C and most
1197: existing (non-ANSI) versions of C.
1198:
1199: Ultimately our intention is that the @samp{-traditional} option will
1200: eliminate most of these incompatibilities by telling GNU C to behave
1201: like the other C compilers.
1202:
1203: @itemize @bullet
1204: @item
1205: GNU CC normally makes string constants read-only. If several
1206: identical-looking string constants are used, GNU CC stores only one
1207: copy of the string.
1208:
1209: One consequence is that you cannot call @code{mktemp} with a string
1210: constant argument. The function @code{mktemp} always alters the
1211: string its argument points to.
1212:
1213: Another consequence is that @code{sscanf} does not work on some
1214: systems when passed a string constant as its format control string.
1215: This is because @code{sscanf} incorrectly tries to write into the
1216: string constant.
1217:
1218: The best solution to these problems is to change the program to use
1219: @code{char}-array variables with initialization strings for these
1220: purposes instead of string constants. But if this is not possible,
1221: you can use the @samp{-fwritable-strings} flag, which directs GNU CC
1222: to handle string constants the same way most C compilers do.
1223:
1224: @item
1225: GNU CC does not substitute macro arguments when they appear inside of
1226: string constants. For example, the following macro in GNU CC
1227:
1228: @example
1229: #define foo(a) "a"
1230: @end example
1231:
1232: @noindent
1233: will produce output @samp{"a"} regardless of what the argument @var{a} is.
1234:
1235: The @samp{-traditional} option directs GNU CC to handle such cases
1236: (among others) in the old-fashioned (non-ANSI) fashion.
1237:
1238: @item
1239: When you use @code{setjmp} and @code{longjmp}, the only automatic
1240: variables guaranteed to remain valid are those declared
1241: @code{volatile}. This is a consequence of automatic register
1242: allocation. Consider this function:
1243:
1244: @example
1245: jmp_buf j;
1246:
1247: foo ()
1248: @{
1249: int a, b;
1250:
1251: a = fun1 ();
1252: if (setjmp (j))
1253: return a;
1254:
1255: a = fun2 ();
1256: /* @r{@code{longjmp (j)} may be occur in @code{fun3}.} */
1257: return a + fun3 ();
1258: @}
1259: @end example
1260:
1261: Here @code{a} may or may not be restored to its first value when the
1262: @code{longjmp} occurs. If @code{a} is allocated in a register, then
1263: its first value is restored; otherwise, it keeps the last value stored
1264: in it.
1265:
1266: If you use the @samp{-W} option with the @samp{-O} option, you will
1267: get a warning when GNU CC thinks such a problem might be possible.
1268:
1269: @item
1270: Declarations of external variables and functions within a block apply
1271: only to the block containing the declaration. In other words, they
1272: have the same scope as any other declaration in the same place.
1273:
1274: In some other C compilers, a @code{extern} declaration affects all the
1275: rest of the file even if it happens within a block.
1276:
1277: The @samp{-traditional} option directs GNU C to treat all @code{extern}
1278: declarations as global, like traditional compilers.
1279:
1280: @item
1281: In traditional C, you can combine @code{long}, etc., with a typedef name,
1282: as shown here:
1283:
1284: @example
1285: typedef int foo;
1286: typedef long foo bar;
1287: @end example
1288:
1289: In ANSI C, this is not allowed: @code{long} and other type modifiers
1290: require an explicit @code{int}. Because this criterion is expressed
1291: by Bison grammar rules rather than C code, the @samp{-traditional}
1292: flag cannot alter it.
1293:
1294: @item
1295: When compiling functions that return structures or unions, GNU CC
1296: output code uses a method different from that used on most versions of
1297: Unix. As a result, code compiled with GNU CC cannot call a
1298: structure-returning function compiled with PCC, and vice versa.
1299:
1300: The method used by GCC is as follows: a structure or union which is 1,
1301: 2, 4 or 8 bytes long is returned like a scalar. A structure or union
1302: with any other size is stored into an address supplied by the caller
1303: in a special, fixed register.
1304:
1305: PCC usually handles all sizes of structures and unions by returning
1306: the address of a block of static storage containing the value. This
1307: method is not used in GCC because it is slower and nonreentrant.
1308:
1309: On systems where PCC works this way, you may be able to make GCC-compiled
1310: code call such functions that were compiled with PCC by declaring them
1311: to return a pointer to the structure or union instead of the structure
1312: or union itself. For example, instead of this:
1313:
1314: @example
1315: struct foo nextfoo ();
1316: @end example
1317:
1318: @noindent
1319: write this:
1320:
1321: @example
1322: struct foo *nextfoo ();
1323: #define nextfoo *nextfoo
1324: @end example
1325:
1326: @noindent
1327: (Note that this assumes you are using the GNU preprocessor, so that
1328: the ANSI antirecursion rules for macro expansions are effective.)
1329: @end itemize
1330:
1331: @node Extensions, Bugs, Incompatibilities, Top
1332: @chapter GNU Extensions to the C Language
1333:
1334: GNU C provides several language features not found in ANSI standard C.
1335: (The @samp{-pedantic} option directs GNU CC to print a warning message if
1336: any of these features is used.) To test for the availability of these
1337: features in conditional compilation, check for a predefined macro
1338: @code{__GNUC__}, which is always defined under GNU CC.
1339:
1340: @menu
1341: * Statement Exprs:: Putting statements and declarations inside expressions.
1342: * Naming Types:: Giving a name to the type of some expression.
1343: * Typeof:: @code{typeof}: referring to the type of an expression.
1344: * Lvalues:: Using @samp{?:}, @samp{,} and casts in lvalues.
1345: * Conditionals:: Omitting the middle operand of a @samp{?:} expression.
1346: * Zero-Length:: Zero-length arrays.
1347: * Variable-Length:: Arrays whose length is computed at run time.
1348: * Subscripting:: Any array can be subscripted, even if not an lvalue.
1349: * Pointer Arith:: Arithmetic on @code{void}-pointers and function pointers.
1350: * Constructors:: Constructor expressions give structures, unions
1351: or arrays as values.
1352: * Dollar Signs:: Dollar sign is allowed in identifiers.
1353: * Alignment:: Inquiring about the alignment of a type or variable.
1354: * Inline:: Defining inline functions (as fast as macros).
1355: * Extended Asm:: Assembler instructions with C expressions as operands.
1356: (With them you can define ``built-in'' functions.)
1357: * Asm Labels:: Specifying the assembler name to use for a C symbol.
1358: @end menu
1359:
1360: @node Statement Exprs, Naming Types, Extensions, Extensions
1361: @section Statements and Declarations inside of Expressions
1362:
1363: A compound statement in parentheses may appear inside an expression in GNU
1364: C. This allows you to declare variables within an expression. For
1365: example:
1366:
1367: @example
1368: (@{ int y = foo (); int z;
1369: if (y > 0) z = y;
1370: else z = - y;
1371: z; @})
1372: @end example
1373:
1374: @noindent
1375: is a valid (though slightly more complex than necessary) expression
1376: for the absolute value of @code{foo ()}.
1377:
1378: This feature is especially useful in making macro definitions ``safe'' (so
1379: that they evaluate each operand exactly once). For example, the
1380: ``maximum'' function is commonly defined as a macro in standard C as
1381: follows:
1382:
1383: @example
1384: #define max(a,b) ((a) > (b) ? (a) : (b))
1385: @end example
1386:
1387: @noindent
1388: But this definition computes either @var{a} or @var{b} twice, with bad
1389: results if the operand has side effects. In GNU C, if you know the
1390: type of the operands (here let's assume @code{int}), you can define
1391: the macro safely as follows:
1392:
1393: @example
1394: #define maxint(a,b) \
1395: (@{int _a = (a), _b = (b); _a > _b ? _a : _b; @})
1396: @end example
1397:
1398: Embedded statements are not allowed in constant expressions, such as
1399: the value of an enumeration constant, the width of a bit field, or
1400: the initial value of a static variable.
1401:
1402: If you don't know the type of the operand, you can still do this, but you
1403: must use @code{typeof} (@pxref{Typeof}) or type naming (@pxref{Naming
1404: Types}).
1405:
1406: @node Naming Types, Typeof, Statement Exprs, Extensions
1407: @section Naming an Expression's Type
1408:
1409: You can give a name to the type of an expression using a @code{typedef}
1410: declaration with an initializer. Here is how to define @var{name} as a
1411: type name for the type of @var{exp}:
1412:
1413: @example
1414: typedef @var{name} = @var{exp};
1415: @end example
1416:
1417: This is useful in conjunction with the statements-within-expressions
1418: feature. Here is how the two together can be used to define a safe
1419: ``maximum'' macro that operates on any arithmetic type:
1420:
1421: @example
1422: #define max(a,b) \
1423: (@{typedef _ta = (a), _tb = (b); \
1424: _ta _a = (a); _tb _b = (b); \
1425: _a > _b ? _a : _b; @})
1426: @end example
1427:
1428: The reason for using names that start with underscores for the local
1429: variables is to avoid conflicts with variable names that occur within the
1430: expressions that are substituted for @code{a} and @code{b}. Eventually we
1431: hope to design a new form of declaration syntax that allows you to declare
1432: variables whose scopes start only after their initializers; this will be a
1433: more reliable way to prevent such conflicts.
1434:
1435: @node Typeof, Lvalues, Naming Types, Extensions
1436: @section Referring to a Type with @code{typeof}
1437:
1438: Another way to refer to the type of an expression is with @code{typeof}.
1439: The syntax of using of this keyword looks like @code{sizeof}, but the
1440: construct acts semantically like a type name defined with @code{typedef}.
1441:
1442: There are two ways of writing the argument to @code{typeof}: with an
1443: expression or with a type. Here is an example with an expression:
1444:
1445: @example
1446: typeof (x[0](1))
1447: @end example
1448:
1449: @noindent
1450: This assumes that @code{x} is an array of functions; the type described
1451: is that of the values of the functions.
1452:
1453: Here is an example with a typename as the argument:
1454:
1455: @example
1456: typeof (int *)
1457: @end example
1458:
1459: @noindent
1460: Here the type described is that of pointers to @code{int}.
1461:
1462: A @code{typeof}-construct can be used anywhere a typedef name could be
1463: used. For example, you can use it in a declaration, in a cast, or inside
1464: of @code{sizeof} or @code{typeof}.
1465:
1466: @itemize @bullet
1467: @item
1468: This declares @code{y} with the type of what @code{x} points to.
1469:
1470: @example
1471: typeof (*x) y;
1472: @end example
1473:
1474: @item
1475: This declares @code{y} as an array of such values.
1476:
1477: @example
1478: typeof (*x) y[4];
1479: @end example
1480:
1481: @item
1482: This declares @code{y} as an array of pointers to characters:
1483:
1484: @example
1485: typeof (typeof (char *)[4]) y;
1486: @end example
1487:
1488: @noindent
1489: It is equivalent to the following traditional C declaration:
1490:
1491: @example
1492: char *y[4];
1493: @end example
1494:
1495: To see the meaning of the declaration using @code{typeof}, and why it
1496: might be a useful way to write, let's rewrite it with these macros:
1497:
1498: @example
1499: #define pointer(T) typeof(T *)
1500: #define array(T, N) typeof(T [N])
1501: @end example
1502:
1503: @noindent
1504: Now the declaration can be rewritten this way:
1505:
1506: @example
1507: array (pointer (char), 4) y;
1508: @end example
1509:
1510: @noindent
1511: Thus, @samp{array (pointer (char), 4)} is the type of arrays of 4
1512: pointers to @code{char}.
1513: @end itemize
1514:
1515: @node Lvalues, Conditionals, Typeof, Extensions
1516: @section Generalized Lvalues
1517:
1518: Compound expressions, conditional expressions and casts are allowed as
1519: lvalues provided their operands are lvalues. This means that you can take
1520: their addresses or store values into them.
1521:
1522: For example, a compound expression can be assigned, provided the last
1523: expression in the sequence is an lvalue. These two expressions are
1524: equivalent:
1525:
1526: @example
1527: (a, b) += 5
1528: a, (b += 5)
1529: @end example
1530:
1531: Similarly, the address of the compound expression can be taken. These two
1532: expressions are equivalent:
1533:
1534: @example
1535: &(a, b)
1536: a, &b
1537: @end example
1538:
1539: A conditional expression is a valid lvalue if its type is not void and the
1540: true and false branches are both valid lvalues. For example, these two
1541: expressions are equivalent:
1542:
1543: @example
1544: (a ? b : c) = 5
1545: (a ? b = 5 : (c = 5))
1546: @end example
1547:
1548: A cast is a valid lvalue if its operand is valid. Taking the address of
1549: the cast is the same as taking the address without a cast, except for the
1550: type of the result. For example, these two expressions are equivalent (but
1551: the second may be valid when the type of @samp{a} does not permit a cast to
1552: @samp{int *}).
1553:
1554: @example
1555: &(int *)a
1556: (int **)&a
1557: @end example
1558:
1559: A simple assignment whose left-hand side is a cast works by converting the
1560: right-hand side first to the specified type, then to the type of the inner
1561: left-hand side expression. After this is stored, the value is converter
1562: back to the specified type to become the value of the assignment. Thus, if
1563: @samp{a} has type @samp{char *}, the following two expressions are
1564: equivalent:
1565:
1566: @example
1567: (int)a = 5
1568: (int)(a = (char *)5)
1569: @end example
1570:
1571: An assignment-with-arithmetic operation such as @samp{+=} applied to a cast
1572: performs the arithmetic using the type resulting from the cast, and then
1573: continues as in the previous case. Therefore, these two expressions are
1574: equivalent:
1575:
1576: @example
1577: (int)a += 5
1578: (int)(a = (char *) ((int)a + 5))
1579: @end example
1580:
1581: @node Conditionals, Zero-Length, Lvalues, Extensions
1582: @section Conditional Expressions with Omitted Middle-Operands
1583:
1584: The middle operand in a conditional expression may be omitted. Then
1585: if the first operand is nonzero, its value is the value of the conditional
1586: expression.
1587:
1588: Therefore, the expression
1589:
1590: @example
1591: x ? : y
1592: @end example
1593:
1594: @noindent
1595: has the value of @code{x} if that is nonzero; otherwise, the value of
1596: @code{y}.
1597:
1598: This example is perfectly equivalent to
1599:
1600: @example
1601: x ? x : y
1602: @end example
1603:
1604: @noindent
1605: In this simple case, the ability to omit the middle operand is not
1606: especially useful. When it becomes useful is when the first operand does,
1607: or may (if it is a macro argument), contain a side effect. Then repeating
1608: the operand in the middle would perform the side effect twice. Omitting
1609: the middle operand uses the value already computed without the undesirable
1610: effects of recomputing it.
1611:
1612: @node Zero-Length, Variable-Length, Conditionals, Extensions
1613: @section Arrays of Length Zero
1614:
1615: Zero-length arrays are allowed in GNU C. They are very useful as the last
1616: element of a structure which is really a header for a variable-length
1617: object:
1618:
1619: @example
1620: struct line @{
1621: int length;
1622: char contents[0];
1623: @};
1624:
1625: @{
1626: struct line *thisline
1627: = (struct line *) malloc (sizeof (struct line) + this_length);
1628: thisline->length = thislength;
1629: @}
1630: @end example
1631:
1632: In standard C, you would have to give @code{contents} a length of 1, which
1633: means either you waste space or complicate the argument to @code{malloc}.
1634:
1635: @node Variable-Length, Subscripting, Zero-Length, Extensions
1636: @section Arrays of Variable Length
1637:
1638: Variable-length automatic arrays are allowed in GNU C. These arrays are
1639: declared like any other automatic arrays, but with a length that is not a
1640: constant expression. The storage is allocated at that time and
1641: deallocated when the brace-level is exited. For example:
1642:
1643: @example
1644: FILE *concat_fopen (char *s1, char *s2, char *mode)
1645: @{
1646: char str[strlen (s1) + strlen (s2) + 1];
1647: strcpy (str, s1);
1648: strcat (str, s2);
1649: return fopen (str, mode);
1650: @}
1651: @end example
1652:
1653: You can also define structure types containing variable-length arrays, and
1654: use them even for arguments or function values, as shown here:
1655:
1656: @example
1657: int foo;
1658:
1659: struct entry
1660: @{
1661: char data[foo];
1662: @};
1663:
1664: struct entry
1665: tester (struct entry arg)
1666: @{
1667: struct entry new;
1668: int i;
1669: for (i = 0; i < foo; i++)
1670: new.data[i] = arg.data[i] + 1;
1671: return new;
1672: @}
1673: @end example
1674:
1675: @noindent
1676: (Eventually there will be a way to say that the size of the array is
1677: another member of the same structure.)
1678:
1679: The length of an array is computed on entry to the brace-level where the
1680: array is declared and is remembered for the scope of the array in case you
1681: access it with @code{sizeof}.
1682:
1683: Jumping or breaking out of the scope of the array name will also deallocate
1684: the storage. Jumping into the scope is not allowed; you will get an error
1685: message for it.
1686:
1687: You can use the function @code{alloca} to get an effect much like
1688: variable-length arrays. The function @code{alloca} is available in
1689: many other C implementations (but not in all). On the other hand,
1690: variable-length arrays are more elegant.
1691:
1692: There are other differences between these two methods. Space allocated
1693: with @code{alloca} exists until the containing @emph{function} returns.
1694: The space for a variable-length array is deallocated as soon as the array
1695: name's scope ends. (If you use both variable-length arrays and
1696: @code{alloca} in the same function, deallocation of a variable-length array
1697: will also deallocate anything more recently allocated with @code{alloca}.)
1698:
1699: @node Subscripting, Pointer Arith, Variable-Length, Extensions
1700: @section Non-Lvalue Arrays May Have Subscripts
1701:
1702: Subscripting is allowed on arrays that are not lvalues, even though the
1703: unary @samp{&} operator is not. For example, this is valid in GNU C though
1704: not valid in other C dialects:
1705:
1706: @example
1707: struct foo @{int a[4];@};
1708:
1709: struct foo f();
1710:
1711: bar (int index)
1712: @{
1713: return f().a[index];
1714: @}
1715: @end example
1716:
1717: @node Pointer Arith, Initializers, Subscripting, Extensions
1718: @section Arithmetic on @code{void}-Pointers and Function Pointers
1719:
1720: In GNU C, addition and subtraction operations are supported on pointers to
1721: @code{void} and on pointers to functions. This is done by treating the
1722: size of a @code{void} or of a function as 1.
1723:
1724: A consequence of this is that @code{sizeof} is also allowed on @code{void}
1725: and on function types, and returns 1.
1726:
1727: @node Initializers, Constructors, Pointer Arith, Extensions
1728: @section Non-Constant Initializers
1729:
1730: The elements of an aggregate initializer are not required to be constant
1731: expressions in GNU C. Here is an example of an initializer with run-time
1732: varying elements:
1733:
1734: @example
1735: foo (float f, float g)
1736: @{
1737: float beat_freqs[2] = @{ f-g, f+g @};
1738: @dots{}
1739: @}
1740: @end example
1741:
1742: @node Constructors, Dollar Signs, Initializers, Extensions
1743: @section Constructor Expressions
1744:
1745: GNU C supports constructor expressions. A constructor looks like a cast
1746: containing an initializer. Its value is an object of the type specified in
1747: the cast, containing the elements specified in the initializer. The type
1748: must be a structure, union or array type.
1749:
1750: Assume that @code{struct foo} and @code{structure} are declared as shown:
1751:
1752: @example
1753: struct foo @{int a; char b[2];@} structure;
1754: @end example
1755:
1756: @noindent
1757: Here is an example of constructing a @samp{struct foo} with a constructor:
1758:
1759: @example
1760: structure = ((struct foo) @{x + y, 'a', 0@});
1761: @end example
1762:
1763: @noindent
1764: This is equivalent to writing the following:
1765:
1766: @example
1767: @{
1768: struct foo temp = @{x + y, 'a', 0@};
1769: structure = temp;
1770: @}
1771: @end example
1772:
1773: You can also construct an array. If all the elements of the constructor
1774: are (made up of) simple constant expressions, suitable for use in
1775: initializers, then the constructor is an lvalue and can be coerced to a
1776: pointer to its first element, as shown here:
1777:
1778: @example
1779: char **foo = (char *[]) @{ "x", "y", "z" @};
1780: @end example
1781:
1782: Array constructors whose elements are not simple constants are not very
1783: useful, because the constructor is not an lvalue. There are only two valid
1784: ways to use it: to subscript it, or initialize an array variable with it.
1785: The former is probably slower than a @code{switch} statement, while the
1786: latter does the same thing an ordinary C initializer would do.
1787:
1788: @example
1789: output = ((int[]) @{ 2, x, 28 @}) [input];
1790: @end example
1791:
1792: @node Dollar Signs, Alignment, Constructors, Extensions
1793: @section Dollar Signs in Identifier Names
1794:
1795: In GNU C, you may use dollar signs in identifier names. This is because
1796: many traditional C implementations allow such identifiers.
1797:
1798: @node Alignment, Inline, Dollar Signs, Extensions
1799: @section Inquiring about the Alignment of a Type or Variable
1800:
1801: The keyword @code{__alignof} allows you to inquire about how an object
1802: is aligned, or the minimum alignment usually required by a type. Its
1803: syntax is just like @code{sizeof}.
1804:
1805: For example, if the target machine requires a @code{double} value to be
1806: aligned on an 8-byte boundary, then @code{__alignof (double)} is 8. This
1807: is true on many RISC machines. On more traditional machine designs,
1808: @code{__alignof (double)} is 4 or even 2.
1809:
1810: Some machines never actually require alignment; they allow reference to any
1811: data type even at an odd addresses. For these machines, @code{__alignof}
1812: reports the @emph{recommended} alignment of a type.
1813:
1814: When the operand of @code{__alignof} is an lvalue rather than a type, the
1815: value is the largest alignment that the lvalue is known to have. It may
1816: have this alignment as a result of its data type, or because it is part of
1817: a structure and inherits alignment from that structure. For example, after
1818: this declaration:
1819:
1820: @example
1821: struct foo @{ int x; char y; @} foo1;
1822: @end example
1823:
1824: @noindent
1825: the value of @code{__alignof (foo1.y)} is probably 2 or 4, the same as
1826: @code{__alignof (int)}, even though the data type of @code{foo1.y} does not
1827: itself demand any alignment.@refill
1828:
1829: @node Inline, Extended Asm, Alignment, Extensions
1830: @section An Inline Function is As Fast As a Macro
1831:
1832: By declaring a function @code{inline}, you can direct GNU CC to integrate
1833: that function's code into the code for its callers. This makes execution
1834: faster by eliminating the function-call overhead; in addition, if any of
1835: the actual argument values are constant, their known values may permit
1836: simplifications at compile time so that not all of the inline function's
1837: code needs to be included.
1838:
1839: To declare a function inline, use the @code{inline} keyword in its
1840: declaration, like this:
1841:
1842: @example
1843: inline int
1844: inc (int *a)
1845: @{
1846: (*a)++;
1847: @}
1848: @end example
1849:
1850: You can also make all ``simple enough'' functions inline with the
1851: option @samp{-finline-functions}. Note that certain usages in a
1852: function definition can make it unsuitable for inline substitution.
1853:
1854: When a function is both inline and @code{static}, if all calls to the
1855: function are integrated into the caller, then the function's own assembler
1856: code is never referenced. In this case, GNU CC does not actually output
1857: assembler code for the function, unless you specify the option
1858: @samp{-fkeep-inline-functions}. Some calls cannot be integrated for
1859: various reasons (in particular, calls that precede the function's
1860: definition cannot be integrated, and neither can recursive calls within the
1861: definition). If there is a nonintegrated call, then the function is
1862: compiled to assembler code as usual.
1863:
1864: When an inline function is not @code{static}, then the compiler must assume
1865: that there may be calls from other source files; since a global symbol can
1866: be defined only once in any program, the function must not be defined in
1867: the other source files, so the calls therein cannot be integrated.
1868: Therefore, a non-@code{static} inline function is always compiled on its
1869: own in the usual fashion.
1870:
1871: @node Extended Asm, Asm Labels, Inline, Extensions
1872: @section Assembler Instructions with C Expression Operands
1873:
1874: In an assembler instruction using @code{asm}, you can now specify the
1875: operands of the instruction using C expressions. This means no more
1876: guessing which registers or memory locations will contain the data you want
1877: to use.
1878:
1879: You must specify an assembler instruction template much like what appears
1880: in a machine description, plus an operand constraint string for each
1881: operand.
1882:
1883: For example, here is how to use the 68881's @code{fsinx} instruction:
1884:
1885: @example
1886: asm ("fsinx %1,%0" : "=f" (result) : "f" (angle));
1887: @end example
1888:
1889: @noindent
1890: Here @code{angle} is the C expression for the input operand while
1891: @code{result} is that of the output operand. Each has @samp{"f"} as its
1892: operand constraint, saying that a floating-point register is required. The
1893: constraints use the same language used in the machine description
1894: (@pxref{Constraints}).
1895:
1896: Each operand is described by an operand-constraint string followed by the C
1897: expression in parentheses. A colon separates the assembler template from
1898: the first output operand, and another separates the last output operand
1899: from the first input, if any. Commas separate output operands and separate
1900: inputs. The number of operands is limited to the maximum number of
1901: operands in any instruction pattern in the machine description.
1902:
1903: Output operand expressions must be lvalues, and there must be at least one
1904: of them. The compiler can check this. The input operands need not be
1905: lvalues, and there need not be any. The compiler cannot check whether the
1906: operands have data types that are reasonable for the instruction being
1907: executed.
1908:
1909: The output operands must be write-only; GNU CC will assume that the values
1910: in these operands before the instruction are dead and need not be
1911: generated. For an operand that is read-write, you must logically split its
1912: function into two separate operands, one input operand and one write-only
1913: output operand. The connection between them is expressed by constraints
1914: which say they need to be in the same location when the instruction
1915: executes. You can use the same C expression for both operands, or
1916: different expressions. For example, here we write the (fictitious)
1917: @samp{combine} instruction with @code{bar} as its read-only source operand
1918: and @code{foo} as its read-write destination:
1919:
1920: @example
1921: asm ("combine %2,%0" : "=r" (foo) : "0" (foo), "g" (bar));
1922: @end example
1923:
1924: @noindent
1925: The constraint @samp{"0"} for operand 1 says that it must occupy the same
1926: location as operand 0. Therefore it is not necessary to substitute operand
1927: 1 into the assembler code output.
1928:
1929: Usually the most convenient way to use these @code{asm} instructions is to
1930: encapsulate them in macros that look like functions. For example,
1931:
1932: @example
1933: #define sin(x) \
1934: (@{ double __value, __arg = (x); \
1935: asm ("fsinx %1,%0": "=f" (__value): "f" (__arg)); \
1936: __value; @})
1937: @end example
1938:
1939: @noindent
1940: Here the variable @code{__arg} is used to make sure that the instruction
1941: operates on a proper @code{double} value, and to accept only those
1942: arguments @code{x} which can convert automatically to a @code{double}.
1943:
1944: Another way to make sure the instruction operates on the correct data type
1945: is to use a cast in the @code{asm}. This is different from using a
1946: variable @code{__arg} in that it converts more different types. For
1947: example, if the desired type were @code{int}, casting the argument to
1948: @code{int} would accept a pointer with no complaint, while assigning the
1949: argument to an @code{int} variable named @code{__arg} would warn about
1950: using a pointer unless the caller explicitly casts it.
1951:
1952: GNU CC assumes for optimization purposes that these instructions have no
1953: side effects except to change the output operands. This does not mean that
1954: instructions with a side effect cannot be used, but you must be careful,
1955: because the compiler may eliminate them if the output operands aren't used,
1956: or move them out of loops, or replace two with one if they constitute a
1957: common subexpression. Also, if your instruction does have a side effect on
1958: a variable that otherwise appears not to change, the old value of the
1959: variable may be reused later if it happens to be found in a register.
1960:
1961: You can prevent an @code{asm} instruction from being deleted, moved or
1962: combined by writing the keyword @code{volatile} after the @code{asm}. For
1963: example:
1.1 root 1964:
1.1.1.2 root 1965: @example
1966: #define set_priority(x) \
1967: asm volatile ("set_priority %1": \
1968: "=m" (*(char *)0): "g" (x))
1969: @end example
1.1 root 1970:
1.1.1.2 root 1971: @noindent
1972: Note that we have supplied an output operand which is not actually used in
1973: the instruction. This is because @code{asm} requires at least one output
1974: operand. This requirement exists for internal implementation reasons and
1975: we might be able to relax it in the future.
1976:
1977: In this case output operand has the additional benefit effect of giving the
1978: appearance of writing in memory. As a result, GNU CC will assume that data
1979: previously fetched from memory must be fetched again if needed again later.
1980: This may be desirable if you have not employed the @code{volatile} keyword
1981: on all the variable declarations that ought to have it.
1982:
1983: @node Asm Labels,,Extended Asm, Extensions
1984: @section Controlling Names Used in Assembler Code
1985:
1986: You can specify the name to be used in the assembler code for a C function
1987: or variable by writing the @code{asm} keyword after the declarator as
1988: follows:
1.1 root 1989:
1.1.1.2 root 1990: @example
1991: int foo asm ("myfoo") = 2;
1992: @end example
1.1 root 1993:
1.1.1.2 root 1994: @noindent
1995: This specifies that the name to be used for the variable @code{foo} in
1996: the assembler code should be @samp{myfoo} rather than the usual
1997: @samp{_foo}.
1998:
1999: On systems where an underscore is normally prepended to the name of a C
2000: function or variable, this feature allows you to define names for the
2001: linker that do not start with an underscore.
2002:
2003: You cannot use @code{asm} in this way in a function @emph{definition}; but
2004: you can get the same effect by writing a declaration for the function
2005: before its definition and putting @code{asm} there, like this:
1.1 root 2006:
1.1.1.2 root 2007: @example
2008: extern func () asm ("FUNC");
1.1 root 2009:
1.1.1.2 root 2010: func (x, y)
2011: int x, y;
2012: @dots{}
2013: @end example
1.1 root 2014:
1.1.1.2 root 2015: It is up to you to make sure that the assembler names you choose do not
2016: conflict with any other assembler symbols. Also, you must not use a
2017: register name; that would produce completely invalid assembler code. GNU
2018: CC does not as yet have the ability to store static variables in registers.
2019: Perhaps that will be added.
2020:
2021: @node Bugs, Portability, Extensions, Top
2022: @chapter Reporting Bugs
2023:
2024: Your bug reports play an essential role in making GNU CC reliable.
2025:
2026: Reporting a bug may help you by bringing a solution to your problem, or it
2027: may not. But in any case the important function of a bug report is to help
2028: the entire community by making the next version of GNU CC work better. Bug
2029: reports are your contribution to the maintenance of GNU CC.
1.1 root 2030:
1.1.1.2 root 2031: In order for a bug report to serve its purpose, you must include the
2032: information that makes for fixing the bug.
1.1 root 2033:
1.1.1.2 root 2034: @menu
2035: * Criteria: Bug Criteria. Have you really found a bug?
2036: * Reporting: Bug Reporting. How to report a bug effectively.
2037: @end menu
1.1 root 2038:
1.1.1.2 root 2039: @node Bug Criteria, Bug Reporting, Bugs, Bugs
2040: @section Have You Found a Bug?
2041:
2042: If you are not sure whether you have found a bug, here are some guidelines:
1.1 root 2043:
2044: @itemize @bullet
2045: @item
1.1.1.2 root 2046: If the compiler gets a fatal signal, for any input whatever, that is a
2047: compiler bug. Reliable compilers never crash.
1.1 root 2048:
2049: @item
1.1.1.2 root 2050: If the compiler produces invalid assembly code, for any input whatever
2051: (except an @code{asm} statement), that is a compiler bug, unless the
2052: compiler reports errors (not just warnings) which would ordinarily
2053: prevent the assembler from being run.
1.1 root 2054:
2055: @item
1.1.1.2 root 2056: If the compiler produces valid assembly code that does not correctly
2057: execute the input source code, that is a compiler bug.
1.1 root 2058:
1.1.1.2 root 2059: However, you must double-check to make sure, because you may have run
2060: into an incompatibility between GNU C and traditional C
2061: (@pxref{Incompatibilities}). These incompatibilities might be considered
1.1.1.3 ! root 2062: bugs, but they are inescapable consequences of valuable features.
1.1 root 2063:
1.1.1.2 root 2064: Or you may have a program whose behavior is undefined, which happened
2065: by chance to give the desired results with another C compiler.
1.1 root 2066:
1.1.1.2 root 2067: For example, in many nonoptimizing compilers, you can write @samp{x;}
2068: at the end of a function instead of @samp{return x;}, with the same
2069: results. But the value of the function is undefined if @samp{return}
2070: is omitted; it is not a bug when GNU CC produces different results.
1.1 root 2071:
1.1.1.2 root 2072: Problems often result from expressions with two increment operators,
2073: as in @samp{f (*p++, *p++)}. Your previous compiler might have
2074: interpreted that expression the way you intended; GNU CC might
2075: interpret it another way; neither compiler is wrong.
1.1 root 2076:
1.1.1.2 root 2077: After you have localized the error to a single source line, it should
2078: be easy to check for these things. If your program is correct and
2079: well defined, you have found a compiler bug.
1.1 root 2080:
1.1.1.2 root 2081: @item
2082: If the compiler produces an error message for valid input, that is a
2083: compiler bug.
1.1 root 2084:
1.1.1.2 root 2085: Note that the following is not valid input, and the error message for
2086: it is not a bug:
1.1 root 2087:
1.1.1.2 root 2088: @example
2089: int foo (char);
1.1 root 2090:
1.1.1.2 root 2091: int
2092: foo (x)
2093: char x;
2094: @{ @dots{} @}
2095: @end example
1.1 root 2096:
1.1.1.2 root 2097: @noindent
2098: The prototype says to pass a @code{char}, while the definition says to
2099: pass an @code{int} and treat the value as a @code{char}. This is what
2100: the ANSI standard says, and it makes sense.
1.1 root 2101:
1.1.1.2 root 2102: @item
2103: If the compiler does not produce an error message for invalid input,
2104: that is a compiler bug. However, you should note that your idea of
2105: ``invalid input'' might be my idea of ``an extension'' or ``support
2106: for traditional practice''.
1.1 root 2107:
1.1.1.2 root 2108: @item
2109: If you are an experienced user of C compilers, your suggestions
2110: for improvement of GNU CC are welcome in any case.
2111: @end itemize
1.1 root 2112:
1.1.1.2 root 2113: @node Bug Reporting,, Bug Criteria, Bugs
2114: @section How to Report Bugs
1.1 root 2115:
1.1.1.2 root 2116: Send bug reports for GNU C to one of these addresses:
1.1 root 2117:
1.1.1.2 root 2118: @example
2119: bug-gcc@@prep.ai.mit.edu
2120: @{ucbvax|mit-eddie|uunet@}!prep.ai.mit.edu!bug-gcc
2121: @end example
1.1 root 2122:
1.1.1.2 root 2123: As a last resort, snail them to:
1.1 root 2124:
1.1.1.2 root 2125: @example
2126: GNU Compiler Bugs
2127: 545 Tech Sq
2128: Cambridge, MA 02139
2129: @end example
1.1 root 2130:
1.1.1.2 root 2131: The fundamental principle of reporting bugs usefully is this:
2132: @strong{report all the facts}. If you are not sure whether to mention a
2133: fact or leave it out, mention it!
2134:
2135: Often people omit facts because they think they know what causes the
2136: problem and they conclude that some details don't matter. Thus, you might
2137: assume that the name of the variable you use in an example does not matter.
2138: Well, probably it doesn't, but one cannot be sure. Perhaps the bug is a
2139: stray memory reference which happens to fetch from the location where that
2140: name is stored in memory; perhaps, if the name were different, the contents
2141: of that location would fool the compiler into doing the right thing despite
2142: the bug. Play it safe and give an exact example.
1.1 root 2143:
1.1.1.2 root 2144: If you want to enable me to fix the bug, you should include all these
2145: things:
1.1 root 2146:
1.1.1.2 root 2147: @itemize @bullet
2148: @item
2149: The version of GNU CC. You can get this by running it with the
2150: @samp{-v} option.
1.1 root 2151:
1.1.1.2 root 2152: Without this, I won't know whether there is any point in looking for
2153: the bug in the current version of GNU CC.
1.1 root 2154:
1.1.1.2 root 2155: @item
2156: A complete input file that will reproduce the bug. If the bug is in
2157: the C preprocessor, send me a source file and any header files that it
2158: requires. If the bug is in the compiler proper (@file{cc1}), run your
2159: source file through the C preprocessor by doing @samp{gcc -E
2160: @var{sourcefile} > @var{outfile}}, then include the contents of
2161: @var{outfile} in the bug report. (Any @samp{-I}, @samp{-D} or
2162: @samp{-U} options that you used in actual compilation should also be
2163: used when doing this.)
1.1 root 2164:
1.1.1.2 root 2165: A single statement is not enough of an example. In order to compile
2166: it, it must be embedded in a function definition; and the bug might
2167: depend on the details of how this is done.
2168:
2169: Without a real example I can compile, all I can do about your bug
2170: report is wish you luck. It would be futile to try to guess how to
2171: provoke the bug. For example, bugs in register allocation and
2172: reloading frequently depend on every little detail of the function
2173: they happen in.
1.1 root 2174:
2175: @item
1.1.1.2 root 2176: The command arguments you gave GNU CC to compile that example and
2177: observe the bug. For example, did you use @samp{-O}? To guarantee
2178: you won't omit something important, list them all.
2179:
2180: If I were to try to guess the arguments, I would probably guess wrong
2181: and then I would not encounter the bug.
1.1 root 2182:
2183: @item
1.1.1.2 root 2184: The names of the files that you used for @file{tm.h} and @file{md}
2185: when you installed the compiler.
1.1 root 2186:
2187: @item
1.1.1.2 root 2188: The type of machine you are using, and the operating system name and
2189: version number.
1.1 root 2190:
2191: @item
1.1.1.2 root 2192: A description of what behavior you observe that you believe is
2193: incorrect. For example, ``It gets a fatal signal,'' or, ``There is an
2194: incorrect assembler instruction in the output.''
2195:
2196: Of course, if the bug is that the compiler gets a fatal signal, then I
2197: will certainly notice it. But if the bug is incorrect output, I might
2198: not notice unless it is glaringly wrong. I won't study all the
2199: assembler code from a 50-line C program just on the off chance that it
2200: might be wrong.
2201:
2202: Even if the problem you experience is a fatal signal, you should still
2203: say so explicitly. Suppose something strange is going on, such as,
2204: your copy of the compiler is out of synch, or you have encountered a
2205: bug in the C library on your system. (This has happened!) Your copy
2206: might crash and mine would not. If you @i{told} me to expect a crash,
2207: then when mine fails to crash, I would know that the bug was not
2208: happening for me. If you had not told me to expect a crash, then I
2209: would not be able to draw any conclusion from my observations.
2210:
2211: In cases where GNU CC generates incorrect code, if you send me a small
2212: complete sample program I will find the error myself by running the
2213: program under a debugger. If you send me a large example or a part of
2214: a larger program, I cannot do this; you must debug the compiled
2215: program and narrow the problem down to one source line. Tell me which
2216: source line it is, and what you believe is incorrect about the code
2217: generated for that line.
1.1 root 2218:
2219: @item
1.1.1.2 root 2220: If you send me examples of output from GNU CC, please use @samp{-g}
2221: when you make them. The debugging information includes source line
2222: numbers which are essential for correlating the output with the input.
1.1 root 2223: @end itemize
2224:
1.1.1.2 root 2225: Here are some things that are not necessary:
1.1 root 2226:
1.1.1.2 root 2227: @itemize @bullet
1.1 root 2228: @item
1.1.1.2 root 2229: A description of the envelope of the bug.
1.1 root 2230:
1.1.1.2 root 2231: Often people who encounter a bug spend a lot of time investigating
2232: which changes to the input file will make the bug go away and which
2233: changes will not affect it.
1.1 root 2234:
1.1.1.2 root 2235: This is often time consuming and not very useful, because the way I
2236: will find the bug is by running a single example under the debugger
2237: with breakpoints, not by pure deduction from a series of examples.
1.1 root 2238:
1.1.1.2 root 2239: Of course, it can't hurt if you can find a simpler example that
2240: triggers the same bug. Errors in the output will be easier to spot,
2241: running under the debugger will take less time, etc. An easy way
2242: to simplify an example is to delete all the function definitions
2243: except the one where the bug occurs. Those earlier in the file
2244: may be replaced by external declarations.
2245:
2246: However, simplification is not necessary; if you don't want to do
2247: this, report the bug anyway.
1.1 root 2248:
2249: @item
1.1.1.2 root 2250: A patch for the bug.
1.1 root 2251:
1.1.1.2 root 2252: A patch for the bug does help me if it is a good one. But don't omit
2253: the necessary information, such as the test case, because I might see
2254: problems with your patch and decide to fix the problem another way.
1.1 root 2255:
1.1.1.2 root 2256: Sometimes with a program as complicated as GNU CC it is very hard to
2257: construct an example that will make the program go through a certain
2258: point in the code. If you don't send me the example, I won't be able
2259: to verify that the bug is fixed.
1.1 root 2260:
2261: @item
1.1.1.2 root 2262: A guess about what the bug is or what it depends on.
2263:
2264: Such guesses are usually wrong. Even I can't guess right about such
2265: things without using the debugger to find the facts. They also don't
2266: serve a useful purpose.
2267: @end itemize
1.1 root 2268:
1.1.1.2 root 2269: @node Portability, Interface, Bugs, Top
1.1 root 2270: @chapter GNU CC and Portability
2271:
2272: The main goal of GNU CC was to make a good, fast compiler for machines in
2273: the class that the GNU system aims to run on: 32-bit machines that address
2274: 8-bit bytes and have several general registers. Elegance, theoretical
2275: power and simplicity are only secondary.
2276:
2277: GNU CC gets most of the information about the target machine from a machine
2278: description which gives an algebraic formula for each of the machine's
2279: instructions. This is a very clean way to describe the target. But when
2280: the compiler needs information that is difficult to express in this
2281: fashion, I have not hesitated to define an ad-hoc parameter to the machine
2282: description. The purpose of portability is to reduce the total work needed
2283: on the compiler; it was not of interest for its own sake.
2284:
2285: GNU CC does not contain machine dependent code, but it does contain code
2286: that depends on machine parameters such as endianness (whether the most
2287: significant byte has the highest or lowest address of the bytes in a word)
2288: and the availability of autoincrement addressing. In the RTL-generation
2289: pass, it is often necessary to have multiple strategies for generating code
2290: for a particular kind of syntax tree, strategies that are usable for different
2291: combinations of parameters. Often I have not tried to address all possible
2292: cases, but only the common ones or only the ones that I have encountered.
2293: As a result, a new target may require additional strategies. You will know
2294: if this happens because the compiler will call @code{abort}. Fortunately,
1.1.1.2 root 2295: the new strategies can be added in a machine-independent fashion, and will
2296: affect only the target machines that need them.
2297:
2298: @node Interface, Passes, Portability, Top
2299: @chapter Interfacing to GNU CC Output
2300:
2301: GNU CC is normally configured to use the same function calling convention
2302: normally in use on the target system. This is done with the
2303: machine-description macros described (@pxref{Machine Macros}).
2304:
2305: However, returning of structure and union values is done differently.
2306: As a result, functions compiled with PCC returning such types cannot
2307: be called from code compiled with GNU CC, and vice versa. This usually
2308: does not cause trouble because the Unix library routines don't return
2309: structures and unions.
2310:
2311: Structures and unions that are 1, 2, 4 or 8 bytes long are returned in the
2312: same registers used for @code{int} or @code{double} return values. (GNU CC
2313: typically allocates variables of such types in registers also.) Structures
2314: and unions of other sizes are returned by storing them into an address
2315: passed by the caller in a register. This method is faster than the one
2316: normally used by PCC and is also reentrant. The register used for passing
2317: the address is specified by the machine-description macro
2318: @code{STRUCT_VALUE_REGNUM}.
2319:
2320: GNU CC always passes arguments on the stack. At some point it will be
2321: extended to pass arguments in registers, for machines which use that as
2322: the standard calling convention. This will make it possible to use such
2323: a convention on other machines as well. However, that would render it
2324: completely incompatible with PCC. We will probably do this once we
2325: have a complete GNU system so we can compile the libraries with GNU CC.
2326:
2327: If you use @code{longjmp}, beware of automatic variables. ANSI C says that
2328: automatic variables that are not declared @code{volatile} have undefined
2329: values after a @code{longjmp}. And this is all GNU CC promises to do,
2330: because it is very difficult to restore register variables correctly, and
2331: one of GNU CC's features is that it can put variables in registers without
2332: your asking it to.
2333:
2334: If you want a variable to be unaltered by @code{longjmp}, and you don't
2335: want to write @code{volatile} because old C compilers don't accept it,
2336: just take the address of the variable. If a variable's address is ever
2337: taken, even if just to compute it and ignore it, then the variable cannot
2338: go in a register:
2339:
2340: @example
2341: @{
2342: int careful;
2343: &careful;
2344: @dots{}
2345: @}
2346: @end example
1.1 root 2347:
1.1.1.2 root 2348: Code compiled with GNU CC may call certain library routines. The routines
2349: needed on the Vax and 68000 are in the file @file{gnulib.c}. You must
2350: compile this file with the standard C compiler, not with GNU CC, and then
2351: link it with each program you compile with GNU CC. (In actuality, many
2352: programs will not need it.) The usual function call interface is used
2353: for calling the library routines. Some standard parts of the C library,
2354: such as @code{bcopy}, are also called automatically.
2355:
2356: @node Passes, RTL, Interface, Top
1.1 root 2357: @chapter Passes and Files of the Compiler
2358:
2359: The overall control structure of the compiler is in @file{toplev.c}. This
2360: file is responsible for initialization, decoding arguments, opening and
2361: closing files, and sequencing the passes.
2362:
1.1.1.2 root 2363: The parsing pass is invoked only once, to parse the entire input. The RTL
2364: intermediate code for a function is generated as the function is parsed, a
2365: statement at a time. Each statement is read in as a syntax tree and then
2366: converted to RTL; then the storage for the tree for the statement is
2367: reclaimed. Storage for types (and the expressions for their sizes),
2368: declarations, and a representation of the binding contours and how they nest,
2369: remains until the function is finished being compiled; these are all needed
2370: to output the debugging information.
2371:
2372: Each time the parsing pass reads a complete function definition or
2373: top-level declaration, it calls the function
2374: @code{rest_of_compilation} or @code{rest_of_decl_compilation} in
2375: @file{toplev.c}, which are responsible for all further processing
2376: necessary, ending with output of the assembler language. All other
2377: compiler passes run, in sequence, within @code{rest_of_compilation}.
2378: When that function returns from compiling a function definition, the
2379: storage used for that function definition's compilation is entirely
2380: freed, unless it is an inline function (@pxref{Inline}).
1.1 root 2381:
2382: Here is a list of all the passes of the compiler and their source files.
2383: Also included is a description of where debugging dumps can be requested
1.1.1.2 root 2384: with @samp{-d} options.
1.1 root 2385:
2386: @itemize @bullet
2387: @item
2388: Parsing. This pass reads the entire text of a function definition,
1.1.1.2 root 2389: constructing partial syntax trees. This and RTL generation are no longer
2390: truly separate passes (formerly they were), but it is easier to think
2391: of them as separate.
2392:
2393: The tree representation does not entirely follow C syntax, because it is
2394: intended to support other languages as well.
1.1 root 2395:
1.1.1.2 root 2396: C data type analysis is also done in this pass, and every tree node
2397: that represents an expression has a data type attached. Variables are
2398: represented as declaration nodes.
1.1 root 2399:
1.1.1.2 root 2400: Constant folding and associative-law simplifications are also done
2401: during this pass.
1.1 root 2402:
1.1.1.2 root 2403: The source files for parsing are @file{parse.y}, @file{decl.c},
1.1 root 2404: @file{typecheck.c}, @file{stor-layout.c}, @file{fold-const.c}, and
2405: @file{tree.c}. The last three are intended to be language-independent.
2406: There are also header files @file{parse.h}, @file{c-tree.h},
2407: @file{tree.h} and @file{tree.def}. The last two define the format of
1.1.1.2 root 2408: the tree representation.@refill
1.1 root 2409:
2410: @item
1.1.1.2 root 2411: RTL generation. This is the conversion of syntax tree into RTL code.
2412: It is actually done statement-by-statement during parsing, but for
2413: most purposes it can be thought of as a separate pass.
1.1 root 2414:
2415: This is where the bulk of target-parameter-dependent code is found,
2416: since often it is necessary for strategies to apply only when certain
2417: standard kinds of instructions are available. The purpose of named
2418: instruction patterns is to provide this information to the RTL
2419: generation pass.
2420:
2421: Optimization is done in this pass for @code{if}-conditions that are
2422: comparisons, boolean operations or conditional expressions. Tail
2423: recursion is detected at this time also. Decisions are made about how
2424: best to arrange loops and how to output @code{switch} statements.
2425:
1.1.1.2 root 2426: The source files for RTL generation are @file{stmt.c}, @file{expr.c},
1.1 root 2427: @file{explow.c}, @file{expmed.c}, @file{optabs.c} and @file{emit-rtl.c}.
2428: Also, the file @file{insn-emit.c}, generated from the machine description
2429: by the program @code{genemit}, is used in this pass. The header files
1.1.1.2 root 2430: @file{expr.h} is used for communication within this pass.@refill
1.1 root 2431:
1.1.1.2 root 2432: The header files @file{insn-flags.h} and @file{insn-codes.h},
2433: generated from the machine description by the programs @code{genflags}
2434: and @code{gencodes}, tell this pass which standard names are available
2435: for use and which patterns correspond to them.@refill
1.1 root 2436:
2437: Aside from debugging information output, none of the following passes
1.1.1.2 root 2438: refers to the tree structure representation of the function (only
2439: part of which is saved).
1.1 root 2440:
1.1.1.2 root 2441: The decision of whether the function can and should be expanded inline
2442: in its subsequent callers is made at the end of rtl generation. The
2443: function must meet certain criteria, currently related to the size of
2444: the function and the types and number of parameters it has. Note that
2445: this function may contain loops, recursive calls to itself
2446: (tail-recursive functions can be inlined!), gotos, in short, all
2447: constructs supported by GNU CC.
2448:
2449: The option @samp{-dr} causes a debugging dump of the RTL code after
2450: this pass. This dump file's name is made by appending @samp{.rtl} to
2451: the input file name.
1.1 root 2452:
2453: @item
1.1.1.2 root 2454: Jump optimization. This pass simplifies jumps to the following
2455: instruction, jumps across jumps, and jumps to jumps. It deletes
2456: unreferenced labels and unreachable code, except that unreachable code
2457: that contains a loop is not recognized as unreachable in this pass.
2458: (Such loops are deleted later in the basic block analysis.)
1.1 root 2459:
2460: Jump optimization is performed two or three times. The first time is
1.1.1.2 root 2461: immediately following RTL generation. The second time is after CSE,
2462: but only if CSE says repeated jump optimization is needed. The
2463: last time is right before the final pass. That time, cross-jumping
2464: and deletion of no-op move instructions are done together with the
2465: optimizations described above.
1.1 root 2466:
2467: The source file of this pass is @file{jump.c}.
2468:
1.1.1.2 root 2469: The option @samp{-dj} causes a debugging dump of the RTL code after
2470: this pass is run for the first time. This dump file's name is made by
2471: appending @samp{.jump} to the input file name.
1.1 root 2472:
2473: @item
2474: Register scan. This pass finds the first and last use of each
2475: register, as a guide for common subexpression elimination. Its source
2476: is in @file{regclass.c}.
2477:
2478: @item
2479: Common subexpression elimination. This pass also does constant
2480: propagation. Its source file is @file{cse.c}. If constant
2481: propagation causes conditional jumps to become unconditional or to
1.1.1.2 root 2482: become no-ops, jump optimization is run again when CSE is finished.
1.1 root 2483:
1.1.1.2 root 2484: The option @samp{-ds} causes a debugging dump of the RTL code after
1.1 root 2485: this pass. This dump file's name is made by appending @samp{.cse} to
2486: the input file name.
2487:
2488: @item
2489: Loop optimization. This pass moves constant expressions out of loops.
2490: Its source file is @file{loop.c}.
2491:
1.1.1.2 root 2492: The option @samp{-dL} causes a debugging dump of the RTL code after
1.1 root 2493: this pass. This dump file's name is made by appending @samp{.loop} to
2494: the input file name.
2495:
2496: @item
2497: Stupid register allocation is performed at this point in a
2498: nonoptimizing compilation. It does a little data flow analysis as
2499: well. When stupid register allocation is in use, the next pass
2500: executed is the reloading pass; the others in between are skipped.
1.1.1.2 root 2501: The source file is @file{stupid.c}.
1.1 root 2502:
2503: @item
2504: Data flow analysis (@file{flow.c}). This pass divides the program
2505: into basic blocks (and in the process deletes unreachable loops); then
2506: it computes which pseudo-registers are live at each point in the
2507: program, and makes the first instruction that uses a value point at
2508: the instruction that computed the value.
2509:
2510: This pass also deletes computations whose results are never used, and
2511: combines memory references with add or subtract instructions to make
2512: autoincrement or autodecrement addressing.
2513:
1.1.1.2 root 2514: The option @samp{-df} causes a debugging dump of the RTL code after
1.1 root 2515: this pass. This dump file's name is made by appending @samp{.flow} to
2516: the input file name. If stupid register allocation is in use, this
2517: dump file reflects the full results of such allocation.
2518:
2519: @item
2520: Instruction combination (@file{combine.c}). This pass attempts to
2521: combine groups of two or three instructions that are related by data
2522: flow into single instructions. It combines the RTL expressions for
2523: the instructions by substitution, simplifies the result using algebra,
2524: and then attempts to match the result against the machine description.
2525:
1.1.1.2 root 2526: The option @samp{-dc} causes a debugging dump of the RTL code after
1.1 root 2527: this pass. This dump file's name is made by appending @samp{.combine}
2528: to the input file name.
2529:
2530: @item
2531: Register class preferencing. The RTL code is scanned to find out
1.1.1.2 root 2532: which register class is best for each pseudo register. The source
2533: file is @file{regclass.c}.
1.1 root 2534:
2535: @item
2536: Local register allocation (@file{local-alloc.c}). This pass allocates
2537: hard registers to pseudo registers that are used only within one basic
1.1.1.2 root 2538: block. Because the basic block is linear, it can use fast and
2539: powerful techniques to do a very good job.
1.1 root 2540:
1.1.1.2 root 2541: The option @samp{-dl} causes a debugging dump of the RTL code after
1.1 root 2542: this pass. This dump file's name is made by appending @samp{.lreg} to
2543: the input file name.
2544:
2545: @item
2546: Global register allocation (@file{global-alloc.c}). This pass
2547: allocates hard registers for the remaining pseudo registers (those
2548: whose life spans are not contained in one basic block).
2549:
2550: @item
1.1.1.2 root 2551: Reloading. This pass renumbers pseudo registers with the hardware
2552: registers numbers they were allocated. Pseudo registers that did not
2553: get hard registers are replaced with stack slots. Then it finds
2554: instructions that are invalid because a value has failed to end up in
2555: a register, or has ended up in a register of the wrong kind. It fixes
2556: up these instructions by reloading the problematical values
2557: temporarily into registers. Additional instructions are generated to
2558: do the copying.
1.1 root 2559:
2560: Source files are @file{reload.c} and @file{reload1.c}, plus the header
2561: @file{reload.h} used for communication between them.
2562:
1.1.1.2 root 2563: The option @samp{-dg} causes a debugging dump of the RTL code after
1.1 root 2564: this pass. This dump file's name is made by appending @samp{.greg} to
2565: the input file name.
2566:
2567: @item
1.1.1.2 root 2568: Jump optimization is repeated, this time including cross-jumping
2569: and deletion of no-op move instructions. Machine-specific peephole
2570: optimizations are performed at the same time.
2571:
2572: The option @samp{-dJ} causes a debugging dump of the RTL code after
2573: this pass. This dump file's name is made by appending @samp{.jump2}
2574: to the input file name.
1.1 root 2575:
2576: @item
2577: Final. This pass outputs the assembler code for the function. It is
1.1.1.2 root 2578: also responsible for identifying spurious test and compare
2579: instructions. The function entry and exit sequences are generated
2580: directly as assembler code in this pass; they never exist as RTL.
1.1 root 2581:
2582: The source files are @file{final.c} plus @file{insn-output.c}; the
2583: latter is generated automatically from the machine description by the
2584: tool @file{genoutput}. The header file @file{conditions.h} is used
2585: for communication between these files.
2586:
2587: @item
2588: Debugging information output. This is run after final because it must
2589: output the stack slot offsets for pseudo registers that did not get
2590: hard registers. Source files are @file{dbxout.c} for DBX symbol table
2591: format and @file{symout.c} for GDB's own symbol table format.
2592: @end itemize
2593:
2594: Some additional files are used by all or many passes:
2595:
2596: @itemize @bullet
2597: @item
2598: Every pass uses @file{machmode.def}, which defines the machine modes.
2599:
2600: @item
2601: All the passes that work with RTL use the header files @file{rtl.h}
1.1.1.2 root 2602: and @file{rtl.def}, and subroutines in file @file{rtl.c}. The tools
2603: @code{gen*} also use these files to read and work with the machine
2604: description RTL.
1.1 root 2605:
2606: @item
2607: Several passes refer to the header file @file{insn-config.h} which
2608: contains a few parameters (C macro definitions) generated
2609: automatically from the machine description RTL by the tool
2610: @code{genconfig}.
2611:
2612: @item
2613: Several passes use the instruction recognizer, which consists of
2614: @file{recog.c} and @file{recog.h}, plus the files @file{insn-recog.c}
2615: and @file{insn-extract.c} that are generated automatically from the
1.1.1.2 root 2616: machine description by the tools @file{genrecog} and
2617: @file{genextract}.@refill
1.1 root 2618:
2619: @item
1.1.1.2 root 2620: Several passes use the header files @file{regs.h} which defines the
2621: information recorded about pseudo register usage, and @file{basic-block.h}
1.1 root 2622: which defines the information recorded about basic blocks.
2623:
2624: @item
2625: @file{hard-reg-set.h} defines the type @code{HARD_REG_SET}, a bit-vector
2626: with a bit for each hard register, and some macros to manipulate it.
2627: This type is just @code{int} if the machine has few enough hard registers;
2628: otherwise it is an array of @code{int} and some of the macros expand
2629: into loops.
2630: @end itemize
2631:
2632: @node RTL, Machine Desc, Passes, Top
2633: @chapter RTL Representation
2634:
2635: Most of the work of the compiler is done on an intermediate representation
1.1.1.2 root 2636: called register transfer language. In this language, the instructions to be
1.1 root 2637: output are described, pretty much one by one, in an algebraic form that
2638: describes what the instruction does.
2639:
2640: RTL is inspired by Lisp lists. It has both an internal form, made up of
2641: structures that point at other structures, and a textual form that is used
2642: in the machine description and in printed debugging dumps. The textual
2643: form uses nested parentheses to indicate the pointers in the internal form.
2644:
2645: @menu
2646: * RTL Objects:: Expressions vs vectors vs strings vs integers.
2647: * Accessors:: Macros to access expression operands or vector elts.
1.1.1.2 root 2648: * Flags:: Other flags in an RTL expression.
1.1 root 2649: * Machine Modes:: Describing the size and format of a datum.
2650: * Constants:: Expressions with constant values.
2651: * Regs and Memory:: Expressions representing register contents or memory.
2652: * Arithmetic:: Expressions representing arithmetic on other expressions.
2653: * Comparisons:: Expressions representing comparison of expressions.
2654: * Bit Fields:: Expressions representing bit-fields in memory or reg.
2655: * Conversions:: Extending, truncating, floating or fixing.
2656: * RTL Declarations:: Declaring volatility, constancy, etc.
2657: * Side Effects:: Expressions for storing in registers, etc.
2658: * Incdec:: Embedded side-effects for autoincrement addressing.
1.1.1.2 root 2659: * Assembler:: Representing @code{asm} with operands.
1.1 root 2660: * Insns:: Expression types for entire insns.
1.1.1.2 root 2661: * Calls:: RTL representation of function call insns.
1.1 root 2662: * Sharing:: Some expressions are unique; others *must* be copied.
2663: @end menu
2664:
2665: @node RTL Objects, Accessors, RTL, RTL
2666: @section RTL Object Types
2667:
2668: RTL uses four kinds of objects: expressions, integers, strings and vectors.
1.1.1.2 root 2669: Expressions are the most important ones. An RTL expression (``RTX'', for
2670: short) is a C structure, but it is usually referred to with a pointer; a
2671: type that is given the typedef name @code{rtx}.
1.1 root 2672:
2673: An integer is simply an @code{int}, and a string is a @code{char *}.
1.1.1.2 root 2674: Within RTL code, strings appear only inside @samp{symbol_ref} expressions,
2675: but they appear in other contexts in the RTL expressions that make up
1.1 root 2676: machine descriptions. Their written form uses decimal digits.
2677:
2678: A string is a sequence of characters. In core it is represented as a
1.1.1.2 root 2679: @code{char *} in usual C fashion, and it is written in C syntax as well.
1.1 root 2680: However, strings in RTL may never be null. If you write an empty string in
2681: a machine description, it is represented in core as a null pointer rather
2682: than as a pointer to a null character. In certain contexts, these null
2683: pointers instead of strings are valid.
2684:
2685: A vector contains an arbitrary, specified number of pointers to
2686: expressions. The number of elements in the vector is explicitly present in
2687: the vector. The written form of a vector consists of square brackets
2688: (@samp{[@dots{}]}) surrounding the elements, in sequence and with
2689: whitespace separating them. Vectors of length zero are not created; null
2690: pointers are used instead.
2691:
1.1.1.2 root 2692: Expressions are classified by @dfn{expression codes} (also called RTX
2693: codes). The expression code is a name defined in @file{rtl.def}, which is
2694: also (in upper case) a C enumeration constant. The possible expression
2695: codes and their meanings are machine-independent. The code of an RTX can
2696: be extracted with the macro @code{GET_CODE (@var{x})} and altered with
2697: @code{PUT_CODE (@var{x}, @var{newcode})}.
1.1 root 2698:
2699: The expression code determines how many operands the expression contains,
2700: and what kinds of objects they are. In RTL, unlike Lisp, you cannot tell
2701: by looking at an operand what kind of object it is. Instead, you must know
2702: from its context---from the expression code of the containing expression.
1.1.1.2 root 2703: For example, in an expression of code @samp{subreg}, the first operand is
1.1 root 2704: to be regarded as an expression and the second operand as an integer. In
1.1.1.2 root 2705: an expression of code @samp{plus}, there are two operands, both of which
2706: are to be regarded as expressions. In a @samp{symbol_ref} expression,
1.1 root 2707: there is one operand, which is to be regarded as a string.
2708:
2709: Expressions are written as parentheses containing the name of the
2710: expression type, its flags and machine mode if any, and then the operands
2711: of the expression (separated by spaces).
2712:
1.1.1.2 root 2713: Expression code names in the @samp{md} file are written in lower case,
2714: but when they appear in C code they are written in upper case. In this
2715: manual, they are shown as follows: @samp{const_int}.
2716:
1.1 root 2717: In a few contexts a null pointer is valid where an expression is normally
2718: wanted. The written form of this is @samp{(nil)}.
2719:
1.1.1.2 root 2720: @node Accessors, Flags, RTL Objects, RTL
1.1 root 2721: @section Access to Operands
2722:
2723: For each expression type @file{rtl.def} specifies the number of contained
2724: objects and their kinds, with four possibilities: @samp{e} for expression
2725: (actually a pointer to an expression), @samp{i} for integer, @samp{s} for
2726: string, and @samp{E} for vector of expressions. The sequence of letters
2727: for an expression code is called its @dfn{format}. Thus, the format of
1.1.1.2 root 2728: @samp{subreg} is @samp{ei}.@refill
1.1 root 2729:
2730: Two other format characters are used occasionally: @samp{u} and @samp{0}.
2731: @samp{u} is equivalent to @samp{e} except that it is printed differently in
2732: debugging dumps, and @samp{0} means a slot whose contents do not fit any
2733: normal category. @samp{0} slots are not printed at all in dumps, and are
1.1.1.2 root 2734: often used in special ways by small parts of the compiler.@refill
1.1 root 2735:
2736: There are macros to get the number of operands and the format of an
2737: expression code:
2738:
2739: @table @code
2740: @item GET_RTX_LENGTH (@var{code})
1.1.1.2 root 2741: Number of operands of an RTX of code @var{code}.
1.1 root 2742:
2743: @item GET_RTX_FORMAT (@var{code})
1.1.1.2 root 2744: The format of an RTX of code @var{code}, as a C string.
1.1 root 2745: @end table
2746:
2747: Operands of expressions are accessed using the macros @code{XEXP},
2748: @code{XINT} and @code{XSTR}. Each of these macros takes two arguments: an
1.1.1.2 root 2749: expression-pointer (RTX) and an operand number (counting from zero).
2750: Thus,@refill
1.1 root 2751:
2752: @example
1.1.1.2 root 2753: XEXP (@var{x}, 2)
1.1 root 2754: @end example
2755:
2756: @noindent
2757: accesses operand 2 of expression @var{x}, as an expression.
2758:
2759: @example
1.1.1.2 root 2760: XINT (@var{x}, 2)
1.1 root 2761: @end example
2762:
2763: @noindent
2764: accesses the same operand as an integer. @code{XSTR}, used in the same
2765: fashion, would access it as a string.
2766:
2767: Any operand can be accessed as an integer, as an expression or as a string.
2768: You must choose the correct method of access for the kind of value actually
2769: stored in the operand. You would do this based on the expression code of
2770: the containing expression. That is also how you would know how many
2771: operands there are.
2772:
2773: For example, if @var{x} is a @samp{subreg} expression, you know that it has
1.1.1.2 root 2774: two operands which can be correctly accessed as @code{XEXP (@var{x}, 0)}
2775: and @code{XINT (@var{x}, 1)}. If you did @code{XINT (@var{x}, 0)}, you
2776: would get the address of the expression operand but cast as an integer;
2777: that might occasionally be useful, but it would be cleaner to write
2778: @code{(int) XEXP (@var{x}, 0)}. @code{XEXP (@var{x}, 1)} would also
2779: compile without error, and would return the second, integer operand cast as
2780: an expression pointer, which would probably result in a crash when
2781: accessed. Nothing stops you from writing @code{XEXP (@var{x}, 28)} either,
2782: but this will access memory past the end of the expression with
2783: unpredictable results.@refill
1.1 root 2784:
2785: Access to operands which are vectors is more complicated. You can use the
2786: macro @code{XVEC} to get the vector-pointer itself, or the macros
2787: @code{XVECEXP} and @code{XVECLEN} to access the elements and length of a
2788: vector.
2789:
2790: @table @code
2791: @item XVEC (@var{exp}, @var{idx})
2792: Access the vector-pointer which is operand number @var{idx} in @var{exp}.
2793:
2794: @item XVECLEN (@var{exp}, @var{idx})
2795: Access the length (number of elements) in the vector which is
2796: in operand number @var{idx} in @var{exp}. This value is an @code{int}.
2797:
1.1.1.2 root 2798: @item XVECEXP (@var{exp}, @var{idx}, @var{eltnum})
1.1 root 2799: Access element number @var{eltnum} in the vector which is
1.1.1.2 root 2800: in operand number @var{idx} in @var{exp}. This value is an RTX.
1.1 root 2801:
2802: It is up to you to make sure that @var{eltnum} is not negative
2803: and is less than @code{XVECLEN (@var{exp}, @var{idx})}.
2804: @end table
2805:
2806: All the macros defined in this section expand into lvalues and therefore
2807: can be used to assign the operands, lengths and vector elements as well as
2808: to access them.
2809:
1.1.1.2 root 2810: @node Flags, Machine Modes, Accessors, RTL
2811: @section Flags in an RTL Expression
2812:
2813: RTL expressions contain several flags (one-bit bit-fields) that are used
2814: in certain types of expression.
2815:
2816: @table @code
2817: @item used
2818: This flag is used only momentarily, at the end of RTL generation for a
2819: function, to count the number of times an expression appears in insns.
2820: Expressions that appear more than once are copied, according to the
2821: rules for shared structure (@pxref{Sharing}).
2822:
2823: @item volatil
2824: This flag is used in @samp{mem} and @samp{reg} expressions and in insns.
2825: In RTL dump files, it is printed as @samp{/v}.
2826:
2827: In a @samp{mem} expression, it is 1 if the memory reference is volatile.
2828: Volatile memory references may not be deleted, reordered or combined.
2829:
2830: In a @samp{reg} expression, it is 1 if the value is a user-level variable.
2831: 0 indicates an internal compiler temporary.
2832:
2833: In an insn, 1 means the insn has been deleted.
2834:
2835: @item in_struct
2836: This flag is used in @samp{mem} expressions. It is 1 if the memory
2837: datum referred to is all or part of a structure or array; 0 if it is (or
2838: might be) a scalar variable. A reference through a C pointer has 0
2839: because the pointer might point to a scalar variable.
2840:
2841: This information allows the compiler to determine something about possible
2842: cases of aliasing.
2843:
2844: In an RTL dump, this flag is represented as @samp{/s}.
2845:
2846: @item unchanging
2847: This flag is used in @samp{reg} and @samp{mem} expressions. 1 means
2848: that the value of the expression never changes (at least within the
2849: current function).
2850:
2851: In an RTL dump, this flag is represented as @samp{/u}.
2852: @end table
2853:
2854: @node Machine Modes, Constants, Flags, RTL
1.1 root 2855: @section Machine Modes
2856:
2857: A machine mode describes a size of data object and the representation used
2858: for it. In the C code, machine modes are represented by an enumeration
1.1.1.2 root 2859: type, @code{enum machine_mode}, defined in @file{machmode.def}. Each RTL
2860: expression has room for a machine mode and so do certain kinds of tree
2861: expressions (declarations and types, to be precise).
1.1 root 2862:
2863: In debugging dumps and machine descriptions, the machine mode of an RTL
2864: expression is written after the expression code with a colon to separate
2865: them. The letters @samp{mode} which appear at the end of each machine mode
2866: name are omitted. For example, @code{(reg:SI 38)} is a @samp{reg}
2867: expression with machine mode @code{SImode}. If the mode is
2868: @code{VOIDmode}, it is not written at all.
2869:
2870: Here is a table of machine modes.
2871:
2872: @table @code
2873: @item QImode
2874: ``Quarter-Integer'' mode represents a single byte treated as an integer.
2875:
2876: @item HImode
2877: ``Half-Integer'' mode represents a two-byte integer.
2878:
2879: @item SImode
2880: ``Single Integer'' mode represents a four-byte integer.
2881:
2882: @item DImode
2883: ``Double Integer'' mode represents an eight-byte integer.
2884:
2885: @item TImode
2886: ``Tetra Integer'' (?) mode represents a sixteen-byte integer.
2887:
2888: @item SFmode
2889: ``Single Floating'' mode represents a single-precision (four byte) floating
2890: point number.
2891:
2892: @item DFmode
2893: ``Double Floating'' mode represents a double-precision (eight byte) floating
2894: point number.
2895:
2896: @item TFmode
2897: ``Tetra Floating'' mode represents a quadruple-precision (sixteen byte)
2898: floating point number.
2899:
2900: @item BLKmode
2901: ``Block'' mode represents values that are aggregates to which none of
1.1.1.2 root 2902: the other modes apply. In RTL, only memory references can have this mode,
1.1 root 2903: and only if they appear in string-move or vector instructions. On machines
2904: which have no such instructions, @code{BLKmode} will not appear in RTL.
2905:
2906: @item VOIDmode
2907: Void mode means the absence of a mode or an unspecified mode.
1.1.1.2 root 2908: For example, RTL expressions of code @samp{const_int} have mode
1.1 root 2909: @code{VOIDmode} because they can be taken to have whatever mode the context
2910: requires. In debugging dumps of RTL, @code{VOIDmode} is expressed by
2911: the absence of any mode.
2912:
2913: @item EPmode
2914: ``Entry Pointer'' mode is intended to be used for function variables in
2915: Pascal and other block structured languages. Such values contain
2916: both a function address and a static chain pointer for access to
2917: automatic variables of outer levels. This mode is only partially
2918: implemented since C does not use it.
2919:
2920: @item CSImode@r{, @dots{}}
2921: ``Complex Single Integer'' mode stands for a complex number represented
2922: as a pair of @code{SImode} integers. Any of the integer and floating modes
2923: may have @samp{C} prefixed to its name to obtain a complex number mode.
2924: For example, there are @code{CQImode}, @code{CSFmode}, and @code{CDFmode}.
2925: Since C does not support complex numbers, these machine modes are only
2926: partially implemented.
2927:
2928: @item BImode
2929: This is the machine mode of a bit-field in a structure. It is used
2930: only in the syntax tree, never in RTL, and in the syntax tree it appears
2931: only in declaration nodes. In C, it appears only in @code{FIELD_DECL}
2932: nodes for structure fields defined with a bit size.
2933: @end table
2934:
2935: The machine description defines @code{Pmode} as a C macro which expands
2936: into the machine mode used for addresses. Normally this is @code{SImode}.
2937:
2938: The only modes which a machine description @i{must} support are
2939: @code{QImode}, @code{SImode}, @code{SFmode} and @code{DFmode}. The
2940: compiler will attempt to use @code{DImode} for two-word structures and
2941: unions, but it would not be hard to program it to avoid this. Likewise,
2942: you can arrange for the C type @code{short int} to avoid using
2943: @code{HImode}. In the long term it would be desirable to make the set of
2944: available machine modes machine-dependent and eliminate all assumptions
2945: about specific machine modes or their uses from the machine-independent
2946: code of the compiler.
2947:
2948: Here are some C macros that relate to machine modes:
2949:
2950: @table @code
2951: @item GET_MODE (@var{x})
1.1.1.2 root 2952: Returns the machine mode of the RTX @var{x}.
1.1 root 2953:
2954: @item PUT_MODE (@var{x}, @var{newmode})
1.1.1.2 root 2955: Alters the machine mode of the RTX @var{x} to be @var{newmode}.
1.1 root 2956:
2957: @item GET_MODE_SIZE (@var{m})
2958: Returns the size in bytes of a datum of mode @var{m}.
2959:
2960: @item GET_MODE_BITSIZE (@var{m})
2961: Returns the size in bits of a datum of mode @var{m}.
2962:
2963: @item GET_MODE_UNIT_SIZE (@var{m})
2964: Returns the size in bits of the subunits of a datum of mode @var{m}.
2965: This is the same as @code{GET_MODE_SIZE} except in the case of
1.1.1.2 root 2966: complex modes and @code{EPmode}. For them, the unit size is the
1.1 root 2967: size of the real or imaginary part, or the size of the function
2968: pointer or the context pointer.
2969: @end table
2970:
2971: @node Constants, Regs and Memory, Machine Modes, RTL
2972: @section Constant Expression Types
2973:
2974: The simplest RTL expressions are those that represent constant values.
2975:
2976: @table @code
2977: @item (const_int @var{i})
2978: This type of expression represents the integer value @var{i}. @var{i}
2979: is customarily accessed with the macro @code{INTVAL} as in
1.1.1.2 root 2980: @code{INTVAL (@var{exp})}, which is equivalent to @code{XINT (@var{exp}, 0)}.
1.1 root 2981:
2982: There is only one expression object for the integer value zero;
2983: it is the value of the variable @code{const0_rtx}. Likewise, the
2984: only expression for integer value one is found in @code{const1_rtx}.
1.1.1.2 root 2985: Any attempt to create an expression of code @samp{const_int} and
1.1 root 2986: value zero or one will return @code{const0_rtx} or @code{const1_rtx}
2987: as appropriate.
2988:
2989: @item (const_double:@var{m} @var{i0} @var{i1})
2990: Represents a floating point constant value of mode @var{m}. The two
2991: integers @var{i0} and @var{i1} together contain the bits of a
2992: @code{double} value. To convert them to a @code{double}, do
2993:
2994: @example
1.1.1.2 root 2995: union @{ double d; int i[2];@} u;
1.1 root 2996: u.i[0] = XINT (x, 0);
2997: u.i[1] = XINT (x, 1);
2998: @end example
2999:
3000: @noindent
3001: and then refer to @code{u.d}. The value of the constant is
3002: represented as a double in this fashion even if the value represented
3003: is single-precision.
3004:
1.1.1.2 root 3005: The global variables @code{dconst0_rtx} and @code{fconst0_rtx} hold
3006: @samp{const_double} expressions with value 0, in modes @code{DFmode} and
3007: @code{SFmode}, respectively.
1.1 root 3008:
3009: @item (symbol_ref @var{symbol})
3010: Represents the value of an assembler label for data. @var{symbol} is
3011: a string that describes the name of the assembler label. If it starts
3012: with a @samp{*}, the label is the rest of @var{symbol} not including
3013: the @samp{*}. Otherwise, the label is @var{symbol}, prefixed with
3014: @samp{_}.
3015:
3016: @item (label_ref @var{label})
3017: Represents the value of an assembler label for code. It contains one
1.1.1.2 root 3018: operand, an expression, which must be a @samp{code_label} that appears
1.1 root 3019: in the instruction sequence to identify the place where the label
3020: should go.
3021:
3022: The reason for using a distinct expression type for code label
3023: references is so that jump optimization can distinguish them.
3024:
3025: @item (const @var{exp})
3026: Represents a constant that is the result of an assembly-time
3027: arithmetic computation. The operand, @var{exp}, is an expression that
3028: contains only constants (@samp{const_int}, @samp{symbol_ref} and
3029: @samp{label_ref} expressions) combined with @samp{plus} and
3030: @samp{minus}. However, not all combinations are valid, since the
3031: assembler cannot do arbitrary arithmetic on relocatable symbols.
3032: @end table
3033:
3034: @node Regs and Memory, Arithmetic, Constants, RTL
3035: @section Registers and Memory
3036:
3037: Here are the RTL expression types for describing access to machine
3038: registers and to main memory.
3039:
3040: @table @code
3041: @item (reg:@var{m} @var{n})
3042: For small values of the integer @var{n} (less than
3043: @code{FIRST_PSEUDO_REGISTER}), this stands for a reference to machine
3044: register number @var{n}: a @dfn{hard register}. For larger values of
3045: @var{n}, it stands for a temporary value or @dfn{pseudo register}.
3046: The compiler's strategy is to generate code assuming an unlimited
3047: number of such pseudo registers, and later convert them into hard
3048: registers or into memory references.
3049:
3050: The symbol @code{FIRST_PSEUDO_REGISTER} is defined by the machine
3051: description, since the number of hard registers on the machine is an
3052: invariant characteristic of the machine. Note, however, that not
3053: all of the machine registers must be general registers. All the
3054: machine registers that can be used for storage of data are given
3055: hard register numbers, even those that can be used only in certain
3056: instructions or can hold only certain types of data.
3057:
1.1.1.2 root 3058: Each pseudo register number used in a function's RTL code is
1.1 root 3059: represented by a unique @samp{reg} expression.
3060:
3061: @var{m} is the machine mode of the reference. It is necessary because
3062: machines can generally refer to each register in more than one mode.
3063: For example, a register may contain a full word but there may be
3064: instructions to refer to it as a half word or as a single byte, as
3065: well as instructions to refer to it as a floating point number of
3066: various precisions.
3067:
3068: Even for a register that the machine can access in only one mode,
3069: the mode must always be specified.
3070:
3071: A hard register may be accessed in various modes throughout one
3072: function, but each pseudo register is given a natural mode
3073: and is accessed only in that mode. When it is necessary to describe
3074: an access to a pseudo register using a nonnatural mode, a @samp{subreg}
3075: expression is used.
3076:
3077: A @samp{reg} expression with a machine mode that specifies more than
3078: one word of data may actually stand for several consecutive registers.
3079: If in addition the register number specifies a hardware register, then
3080: it actually represents several consecutive hardware registers starting
3081: with the specified one.
3082:
3083: Such multi-word hardware register @samp{reg} expressions may not be live
3084: across the boundary of a basic block. The lifetime analysis pass does not
3085: know how to record properly that several consecutive registers are
3086: actually live there, and therefore register allocation would be confused.
3087: The CSE pass must go out of its way to make sure the situation does
3088: not arise.
3089:
3090: @item (subreg:@var{m} @var{reg} @var{wordnum})
3091: @samp{subreg} expressions are used to refer to a register in a machine
3092: mode other than its natural one, or to refer to one register of
3093: a multi-word @samp{reg} that actually refers to several registers.
3094:
3095: Each pseudo-register has a natural mode. If it is necessary to
3096: operate on it in a different mode---for example, to perform a fullword
3097: move instruction on a pseudo-register that contains a single byte---
3098: the pseudo-register must be enclosed in a @samp{subreg}. In such
3099: a case, @var{wordnum} is zero.
3100:
3101: The other use of @samp{subreg} is to extract the individual registers
3102: of a multi-register value. Machine modes such as @code{DImode} and
3103: @code{EPmode} indicate values longer than a word, values which usually
3104: require two consecutive registers. To access one of the registers,
3105: use a @samp{subreg} with mode @code{SImode} and a @var{wordnum} that
3106: says which register.
3107:
3108: The compilation parameter @code{WORDS_BIG_ENDIAN}, if defined, says
3109: that word number zero is the most significant part; otherwise, it is
3110: the least significant part.
3111:
3112: Note that it is not valid to access a @code{DFmode} value in @code{SFmode}
3113: using a @samp{subreg}. On some machines the most significant part of a
3114: @code{DFmode} value does not have the same format as a single-precision
3115: floating value.
3116:
3117: @item (cc0)
3118: This refers to the machine's condition code register. It has no
3119: operands and may not have a machine mode. It may be validly used in
3120: only two contexts: as the destination of an assignment (in test and
3121: compare instructions) and in comparison operators comparing against
1.1.1.2 root 3122: zero (@samp{const_int} with value zero; that is to say,
3123: @code{const0_rtx}).
1.1 root 3124:
1.1.1.2 root 3125: There is only one expression object of code @samp{cc0}; it is the
1.1 root 3126: value of the variable @code{cc0_rtx}. Any attempt to create an
1.1.1.2 root 3127: expression of code @samp{cc0} will return @code{cc0_rtx}.
1.1 root 3128:
1.1.1.2 root 3129: One special thing about the condition code register is that
3130: instructions can set it implicitly. On many machines, nearly all
3131: instructions set the condition code based on the value that they
3132: compute or store. It is not necessary to record these actions
3133: explicitly in the RTL because the machine description includes a
3134: prescription for recognizing the instructions that do so (by means of
3135: the macro @code{NOTICE_UPDATE_CC}). Only instructions whose sole
3136: purpose is to set the condition code, and instructions that use the
3137: condition code, need mention @code{(cc0)}.
1.1 root 3138:
3139: @item (pc)
3140: This represents the machine's program counter. It has no operands and
3141: may not have a machine mode. @code{(pc)} may be validly used only in
3142: certain specific contexts in jump instructions.
3143:
1.1.1.2 root 3144: There is only one expression object of code @samp{pc}; it is the value
3145: of the variable @code{pc_rtx}. Any attempt to create an expression of
3146: code @samp{pc} will return @code{pc_rtx}.
1.1 root 3147:
1.1.1.2 root 3148: All instructions that do not jump alter the program counter implicitly
3149: by incrementing it, but there is no need to mention this in the RTL.
1.1 root 3150:
3151: @item (mem:@var{m} @var{addr})
1.1.1.2 root 3152: This RTX represents a reference to main memory at an address
3153: represented by the expression @var{addr}. @var{m} specifies how large
3154: a unit of memory is accessed.
1.1 root 3155: @end table
3156:
3157: @node Arithmetic, Comparisons, Regs and Memory, RTL
3158: @section RTL Expressions for Arithmetic
3159:
3160: @table @code
3161: @item (plus:@var{m} @var{x} @var{y})
3162: Represents the sum of the values represented by @var{x} and @var{y}
3163: carried out in machine mode @var{m}. This is valid only if
3164: @var{x} and @var{y} both are valid for mode @var{m}.
3165:
3166: @item (minus:@var{m} @var{x} @var{y})
3167: Like @samp{plus} but represents subtraction.
3168:
3169: @item (minus @var{x} @var{y})
3170: Represents the result of subtracting @var{y} from @var{x}
3171: for purposes of comparison. The absence of a machine mode
3172: in the @samp{minus} expression indicates that the result is
3173: computed without overflow, as if with infinite precision.
3174:
3175: Of course, machines can't really subtract with infinite precision.
3176: However, they can pretend to do so when only the sign of the
3177: result will be used, which is the case when the result is stored
1.1.1.2 root 3178: in @code{(cc0)}. And that is the only way this kind of expression
1.1 root 3179: may validly be used: as a value to be stored in the condition codes.
3180:
3181: @item (neg:@var{m} @var{x})
3182: Represents the negation (subtraction from zero) of the value
3183: represented by @var{x}, carried out in mode @var{m}. @var{x} must be
3184: valid for mode @var{m}.
3185:
3186: @item (mult:@var{m} @var{x} @var{y})
3187: Represents the signed product of the values represented by @var{x} and
3188: @var{y} carried out in machine mode @var{m}. If
3189: @var{x} and @var{y} are both valid for mode @var{m}, this is ordinary
1.1.1.2 root 3190: size-preserving multiplication. Alternatively, both @var{x} and @var{y}
1.1 root 3191: may be valid for a different, narrower mode. This represents the
3192: kind of multiplication that generates a product wider than the operands.
3193: Widening multiplication and same-size multiplication are completely
3194: distinct and supported by different machine instructions; machines may
1.1.1.2 root 3195: support one but not the other.@refill
1.1 root 3196:
3197: @samp{mult} may be used for floating point division as well.
3198: Then @var{m} is a floating point machine mode.
3199:
3200: @item (umult:@var{m} @var{x} @var{y})
3201: Like @samp{mult} but represents unsigned multiplication. It may be
3202: used in both same-size and widening forms, like @samp{mult}.
1.1.1.2 root 3203: @samp{umult} is used only for fixed-point multiplication.
1.1 root 3204:
3205: @item (div:@var{m} @var{x} @var{y})
3206: Represents the quotient in signed division of @var{x} by @var{y},
3207: carried out in machine mode @var{m}. If @var{m} is a floating-point
3208: mode, it represents the exact quotient; otherwise, the integerized
3209: quotient. If @var{x} and @var{y} are both valid for mode @var{m},
3210: this is ordinary size-preserving division. Some machines have
3211: division instructions in which the operands and quotient widths are
3212: not all the same; such instructions are represented by @samp{div}
3213: expressions in which the machine modes are not all the same.
3214:
3215: @item (udiv:@var{m} @var{x} @var{y})
3216: Like @samp{div} but represents unsigned division.
3217:
3218: @item (mod:@var{m} @var{x} @var{y})
3219: @itemx (umod:@var{m} @var{x} @var{y})
3220: Like @samp{div} and @samp{udiv} but represent the remainder instead of
3221: the quotient.
3222:
3223: @item (not:@var{m} @var{x})
3224: Represents the bitwise complement of the value represented by @var{x},
3225: carried out in mode @var{m}, which must be a fixed-point machine mode.
3226: @var{x} must be valid for mode @var{m}, which must be a fixed-point mode.
3227:
3228: @item (and:@var{m} @var{x} @var{y})
3229: Represents the bitwise logical-and of the values represented by
3230: @var{x} and @var{y}, carried out in machine mode @var{m}. This is
3231: valid only if @var{x} and @var{y} both are valid for mode @var{m},
3232: which must be a fixed-point mode.
3233:
3234: @item (ior:@var{m} @var{x} @var{y})
3235: Represents the bitwise inclusive-or of the values represented by
3236: @var{x} and @var{y}, carried out in machine mode @var{m}. This is
3237: valid only if @var{x} and @var{y} both are valid for mode @var{m},
3238: which must be a fixed-point mode.
3239:
3240: @item (xor:@var{m} @var{x} @var{y})
3241: Represents the bitwise exclusive-or of the values represented by
3242: @var{x} and @var{y}, carried out in machine mode @var{m}. This is
3243: valid only if @var{x} and @var{y} both are valid for mode @var{m},
3244: which must be a fixed-point mode.
3245:
3246: @item (lshift:@var{m} @var{x} @var{c})
3247: Represents the result of logically shifting @var{x} left by @var{c}
3248: places. @var{x} must be valid for the mode @var{m}, a fixed-point
3249: machine mode. @var{c} must be valid for a fixed-point mode;
3250: which mode is determined by the mode called for in the machine
3251: description entry for the left-shift instruction. For example,
3252: on the Vax, the mode of @var{c} is @code{QImode} regardless of @var{m}.
3253:
3254: On some machines, negative values of @var{c} may be meaningful; this
1.1.1.2 root 3255: is why logical left shift and arithmetic left shift are distinguished.
1.1 root 3256: For example, Vaxes have no right-shift instructions, and right shifts
3257: are represented as left-shift instructions whose counts happen
3258: to be negative constants or else computed (in a previous instruction)
3259: by negation.
3260:
3261: @item (ashift:@var{m} @var{x} @var{c})
3262: Like @samp{lshift} but for arithmetic left shift.
3263:
3264: @item (lshiftrt:@var{m} @var{x} @var{c})
3265: @itemx (ashiftrt:@var{m} @var{x} @var{c})
3266: Like @samp{lshift} and @samp{ashift} but for right shift.
3267:
3268: @item (rotate:@var{m} @var{x} @var{c})
3269: @itemx (rotatert:@var{m} @var{x} @var{c})
3270: Similar but represent left and right rotate.
3271:
3272: @item (abs:@var{m} @var{x})
3273: Represents the absolute value of @var{x}, computed in mode @var{m}.
3274: @var{x} must be valid for @var{m}.
3275:
3276: @item (sqrt:@var{m} @var{x})
3277: Represents the square root of @var{x}, computed in mode @var{m}.
3278: @var{x} must be valid for @var{m}. Most often @var{m} will be
3279: a floating point mode.
1.1.1.2 root 3280:
3281: @item (ffs:@var{m} @var{x})
3282: Represents the one plus the index of the least significant 1-bit in
3283: @var{x}, represented as an integer of mode @var{m}. (The value is
3284: zero if @var{x} is zero.) The mode of @var{x} need not be @var{m};
3285: depending on the target machine, various mode combinations may be
3286: valid.
1.1 root 3287: @end table
3288:
3289: @node Comparisons, Bit Fields, Arithmetic, RTL
3290: @section Comparison Operations
3291:
3292: Comparison operators test a relation on two operands and are considered to
3293: represent the value 1 if the relation holds, or zero if it does not. The
3294: mode of the comparison is determined by the operands; they must both be
3295: valid for a common machine mode. A comparison with both operands constant
3296: would be invalid as the machine mode could not be deduced from it, but such
1.1.1.2 root 3297: a comparison should never exist in RTL due to constant folding.
1.1 root 3298:
3299: Inequality comparisons come in two flavors, signed and unsigned. Thus,
1.1.1.2 root 3300: there are distinct expression codes @samp{gt} and @samp{gtu} for signed and
1.1 root 3301: unsigned greater-than. These can produce different results for the same
3302: pair of integer values: for example, 1 is signed greater-than -1 but not
3303: unsigned greater-than, because -1 when regarded as unsigned is actually
1.1.1.2 root 3304: @code{0xffffffff} which is greater than 1.
1.1 root 3305:
3306: The signed comparisons are also used for floating point values. Floating
3307: point comparisons are distinguished by the machine modes of the operands.
3308:
3309: The comparison operators may be used to compare the condition codes
1.1.1.2 root 3310: @code{(cc0)} against zero, as in @code{(eq (cc0) (const_int 0))}. Such a
3311: construct actually refers to the result of the preceding instruction in
3312: which the condition codes were set. The above example stands for 1 if the
3313: condition codes were set to say ``zero'' or ``equal'', 0 otherwise.
3314: Although the same comparison operators are used for this as may be used in
3315: other contexts on actual data, no confusion can result since the machine
3316: description would never allow both kinds of uses in the same context.
1.1 root 3317:
3318: @table @code
3319: @item (eq @var{x} @var{y})
3320: 1 if the values represented by @var{x} and @var{y} are equal,
3321: otherwise 0.
3322:
3323: @item (ne @var{x} @var{y})
3324: 1 if the values represented by @var{x} and @var{y} are not equal,
3325: otherwise 0.
3326:
3327: @item (gt @var{x} @var{y})
3328: 1 if the @var{x} is greater than @var{y}. If they are fixed-point,
3329: the comparison is done in a signed sense.
3330:
3331: @item (gtu @var{x} @var{y})
3332: Like @samp{gt} but does unsigned comparison, on fixed-point numbers only.
3333:
3334: @item (lt @var{x} @var{y})
3335: @item (ltu @var{x} @var{y})
3336: Like @samp{gt} and @samp{gtu} but test for ``less than''.
3337:
3338: @item (ge @var{x} @var{y})
3339: @item (geu @var{x} @var{y})
3340: Like @samp{gt} and @samp{gtu} but test for ``greater than or equal''.
3341:
3342: @item (le @var{x} @var{y})
3343: @item (leu @var{x} @var{y})
3344: Like @samp{gt} and @samp{gtu} but test for ``less than or equal''.
3345:
3346: @item (if_then_else @var{cond} @var{then} @var{else})
3347: This is not a comparison operation but is listed here because it is
3348: always used in conjunction with a comparison operation. To be
3349: precise, @var{cond} is a comparison expression. This expression
3350: represents a choice, according to @var{cond}, between the value
3351: represented by @var{then} and the one represented by @var{else}.
3352:
3353: On most machines, @samp{if_then_else} expressions are valid only
3354: to express conditional jumps.
3355: @end table
3356:
3357: @node Bit Fields, Conversions, Comparisons, RTL
3358: @section Bit-fields
3359:
3360: Special expression codes exist to represent bit-field instructions.
1.1.1.2 root 3361: These types of expressions are lvalues in RTL; they may appear
1.1 root 3362: on the left side of a assignment, indicating insertion of a value
3363: into the specified bit field.
3364:
3365: @table @code
3366: @item (sign_extract:SI @var{loc} @var{size} @var{pos})
3367: This represents a reference to a sign-extended bit-field contained or
3368: starting in @var{loc} (a memory or register reference). The bit field
3369: is @var{size} bits wide and starts at bit @var{pos}. The compilation
1.1.1.2 root 3370: option @code{BITS_BIG_ENDIAN} says which end of the memory unit
1.1 root 3371: @var{pos} counts from.
3372:
3373: Which machine modes are valid for @var{loc} depends on the machine,
3374: but typically @var{loc} should be a single byte when in memory
3375: or a full word in a register.
3376:
1.1.1.2 root 3377: @item (zero_extract:SI @var{loc} @var{size} @var{pos})
1.1 root 3378: Like @samp{sign_extract} but refers to an unsigned or zero-extended
3379: bit field. The same sequence of bits are extracted, but they
3380: are filled to an entire word with zeros instead of by sign-extension.
3381: @end table
3382:
3383: @node Conversions, RTL Declarations, Bit Fields, RTL
3384: @section Conversions
3385:
3386: All conversions between machine modes must be represented by
3387: explicit conversion operations. For example, an expression
1.1.1.2 root 3388: which is the sum of a byte and a full word cannot be written as
1.1 root 3389: @code{(plus:SI (reg:QI 34) (reg:SI 80))} because the @samp{plus}
3390: operation requires two operands of the same machine mode.
3391: Therefore, the byte-sized operand is enclosed in a conversion
3392: operation, as in
3393:
3394: @example
3395: (plus:SI (sign_extend:SI (reg:QI 34)) (reg:SI 80))
3396: @end example
3397:
3398: The conversion operation is not a mere placeholder, because there
3399: may be more than one way of converting from a given starting mode
3400: to the desired final mode. The conversion operation code says how
3401: to do it.
3402:
3403: @table @code
3404: @item (sign_extend:@var{m} @var{x})
3405: Represents the result of sign-extending the value @var{x}
3406: to machine mode @var{m}. @var{m} must be a fixed-point mode
3407: and @var{x} a fixed-point value of a mode narrower than @var{m}.
3408:
3409: @item (zero_extend:@var{m} @var{x})
3410: Represents the result of zero-extending the value @var{x}
3411: to machine mode @var{m}. @var{m} must be a fixed-point mode
3412: and @var{x} a fixed-point value of a mode narrower than @var{m}.
3413:
3414: @item (float_extend:@var{m} @var{x})
3415: Represents the result of extending the value @var{x}
3416: to machine mode @var{m}. @var{m} must be a floating point mode
3417: and @var{x} a floating point value of a mode narrower than @var{m}.
3418:
3419: @item (truncate:@var{m} @var{x})
3420: Represents the result of truncating the value @var{x}
3421: to machine mode @var{m}. @var{m} must be a fixed-point mode
3422: and @var{x} a fixed-point value of a mode wider than @var{m}.
3423:
3424: @item (float_truncate:@var{m} @var{x})
3425: Represents the result of truncating the value @var{x}
3426: to machine mode @var{m}. @var{m} must be a floating point mode
3427: and @var{x} a floating point value of a mode wider than @var{m}.
3428:
3429: @item (float:@var{m} @var{x})
1.1.1.2 root 3430: Represents the result of converting fixed point value @var{x},
3431: regarded as signed, to floating point mode @var{m}.
3432:
3433: @item (unsigned_float:@var{m} @var{x})
3434: Represents the result of converting fixed point value @var{x},
3435: regarded as unsigned, to floating point mode @var{m}.
1.1 root 3436:
3437: @item (fix:@var{m} @var{x})
1.1.1.2 root 3438: When @var{m} is a fixed point mode, represents the result of
3439: converting floating point value @var{x} to mode @var{m}, regarded as
3440: signed. How rounding is done is not specified, so this operation may
3441: be used validly in compiling C code only for integer-valued operands.
3442:
3443: @item (unsigned_fix:@var{m} @var{x})
3444: Represents the result of converting floating point value @var{x} to
3445: fixed point mode @var{m}, regarded as unsigned. How rounding is done
3446: is not specified.
1.1 root 3447:
1.1.1.2 root 3448: @item (fix:@var{m} @var{x})
3449: When @var{m} is a floating point mode, represents the result of
3450: converting floating point value @var{x} (valid for mode @var{m}) to an
3451: integer, still represented in floating point mode @var{m}, by rounding
3452: towards zero.
1.1 root 3453: @end table
3454:
3455: @node RTL Declarations, Side Effects, Conversions, RTL
3456: @section Declarations
3457:
3458: Declaration expression codes do not represent arithmetic operations
3459: but rather state assertions about their operands.
3460:
3461: @table @code
3462: @item (strict_low_part (subreg:@var{m} (reg:@var{n} @var{r}) 0))
3463: This expression code is used in only one context: operand 0 of a
3464: @samp{set} expression. In addition, the operand of this expression
3465: must be a @samp{subreg} expression.
3466:
3467: The presence of @samp{strict_low_part} says that the part of the
1.1.1.2 root 3468: register which is meaningful in mode @var{n}, but is not part of
3469: mode @var{m}, is not to be altered. Normally, an assignment to such
1.1 root 3470: a subreg is allowed to have undefined effects on the rest of the
3471: register when @var{m} is less than a word.
3472: @end table
3473:
3474: @node Side Effects, Incdec, RTL Declarations, RTL
3475: @section Side Effect Expressions
3476:
3477: The expression codes described so far represent values, not actions.
3478: But machine instructions never produce values; they are meaningful
3479: only for their side effects on the state of the machine. Special
3480: expression codes are used to represent side effects.
3481:
3482: The body of an instruction is always one of these side effect codes;
3483: the codes described above, which represent values, appear only as
3484: the operands of these.
3485:
3486: @table @code
3487: @item (set @var{lval} @var{x})
3488: Represents the action of storing the value of @var{x} into the place
3489: represented by @var{lval}. @var{lval} must be an expression
3490: representing a place that can be stored in: @samp{reg} (or
3491: @samp{subreg} or @samp{strict_low_part}), @samp{mem}, @samp{pc} or
1.1.1.2 root 3492: @samp{cc0}.@refill
1.1 root 3493:
3494: If @var{lval} is a @samp{reg}, @samp{subreg} or @samp{mem}, it has a
1.1.1.2 root 3495: machine mode; then @var{x} must be valid for that mode.@refill
1.1 root 3496:
3497: If @var{lval} is a @samp{reg} whose machine mode is less than the full
3498: width of the register, then it means that the part of the register
3499: specified by the machine mode is given the specified value and the
3500: rest of the register receives an undefined value. Likewise, if
3501: @var{lval} is a @samp{subreg} whose machine mode is narrower than
3502: @code{SImode}, the rest of the register can be changed in an undefined way.
3503:
3504: If @var{lval} is a @samp{strict_low_part} of a @samp{subreg}, then the
3505: part of the register specified by the machine mode of the
3506: @samp{subreg} is given the value @var{x} and the rest of the register
1.1.1.2 root 3507: is not changed.@refill
1.1 root 3508:
3509: If @var{lval} is @code{(cc0)}, it has no machine mode, and @var{x} may
1.1.1.2 root 3510: have any mode. This represents a ``test'' or ``compare'' instruction.@refill
1.1 root 3511:
3512: If @var{lval} is @code{(pc)}, we have a jump instruction, and the
3513: possibilities for @var{x} are very limited. It may be a
3514: @samp{label_ref} expression (unconditional jump). It may be an
3515: @samp{if_then_else} (conditional jump), in which case either the
3516: second or the third operand must be @code{(pc)} (for the case which
3517: does not jump) and the other of the two must be a @samp{label_ref}
3518: (for the case which does jump). @var{x} may also be a @samp{mem} or
3519: @code{(plus:SI (pc) @var{y})}, where @var{y} may be a @samp{reg} or a
3520: @samp{mem}; these unusual patterns are used to represent jumps through
1.1.1.2 root 3521: branch tables.@refill
1.1 root 3522:
3523: @item (return)
1.1.1.2 root 3524: Represents a return from the current function, on machines where this
3525: can be done with one instruction, such as Vaxes. On machines where a
3526: multi-instruction ``epilogue'' must be executed in order to return
3527: from the function, returning is done by jumping to a label which
3528: precedes the epilogue, and the @samp{return} expression code is never
3529: used.
1.1 root 3530:
3531: @item (call @var{function} @var{nargs})
3532: Represents a function call. @var{function} is a @samp{mem} expression
3533: whose address is the address of the function to be called. @var{nargs}
3534: is an expression representing the number of words of argument.
3535:
3536: Each machine has a standard machine mode which @var{function} must
1.1.1.2 root 3537: have. The machine description defines macro @code{FUNCTION_MODE} to
1.1 root 3538: expand into the requisite mode name. The purpose of this mode is to
3539: specify what kind of addressing is allowed, on machines where the
3540: allowed kinds of addressing depend on the machine mode being
3541: addressed.
3542:
3543: @item (clobber @var{x})
3544: Represents the storing or possible storing of an unpredictable,
3545: undescribed value into @var{x}, which must be a @samp{reg} or
3546: @samp{mem} expression.
3547:
3548: One place this is used is in string instructions that store standard
3549: values into particular hard registers. It may not be worth the
1.1.1.2 root 3550: trouble to describe the values that are stored, but it is essential to
3551: inform the compiler that the registers will be altered, lest it
1.1 root 3552: attempt to keep data in them across the string instruction.
3553:
3554: @var{x} may also be null---a null C pointer, no expression at all.
3555: Such a @code{(clobber (null))} expression means that all memory
3556: locations must be presumed clobbered.
3557:
3558: Note that the machine description classifies certain hard registers as
3559: ``call-clobbered''. All function call instructions are assumed by
3560: default to clobber these registers, so there is no need to use
3561: @samp{clobber} expressions to indicate this fact. Also, each function
3562: call is assumed to have the potential to alter any memory location.
3563:
3564: @item (use @var{x})
1.1.1.2 root 3565: Represents the use of the value of @var{x}. It indicates that the
3566: value in @var{x} at this point in the program is needed, even though
3567: it may not be apparent why this is so. Therefore, the compiler will
3568: not attempt to delete instructions whose only effect is to store a
3569: value in @var{x}. @var{x} must be a @samp{reg} expression.
1.1 root 3570:
3571: @item (parallel [@var{x0} @var{x1} @dots{}])
3572: Represents several side effects performed in parallel. The square
3573: brackets stand for a vector; the operand of @samp{parallel} is a
3574: vector of expressions. @var{x0}, @var{x1} and so on are individual
3575: side effects---expressions of code @samp{set}, @samp{call},
1.1.1.2 root 3576: @samp{return}, @samp{clobber} or @samp{use}.@refill
1.1 root 3577:
1.1.1.2 root 3578: ``In parallel'' means that first all the values used in the individual
3579: side-effects are computed, and second all the actual side-effects are
3580: performed. For example,
1.1 root 3581:
3582: @example
3583: (parallel [(set (reg:SI 1) (mem:SI (reg:SI 1)))
3584: (set (mem:SI (reg:SI 1)) (reg:SI 1))])
3585: @end example
3586:
3587: @noindent
3588: says unambiguously that the values of hard register 1 and the memory
3589: location addressed by it are interchanged. In both places where
3590: @code{(reg:SI 1)} appears as a memory address it refers to the value
1.1.1.2 root 3591: in register 1 @emph{before} the execution of the instruction.
3592:
3593: @item (sequence [@var{insns} @dots{}])
3594: Represents a sequence of insns. Each of the @var{insns} that appears
3595: in the vector is suitable for appearing in the chain of insns, so it
3596: must be an @samp{insn}, @samp{jump_insn}, @samp{call_insn},
3597: @samp{code_label}, @samp{barrier} or @samp{note}.
3598:
3599: A @samp{sequence} RTX never appears in an actual insn. It represents
3600: the sequence of insns that result from a @samp{define_expand}
3601: @emph{before} those insns are passed to @code{emit_insn} to insert
3602: them in the chain of insns. When actually inserted, the individual
3603: sub-insns are separated out and the @samp{sequence} is forgotten.
1.1 root 3604: @end table
3605:
1.1.1.2 root 3606: Three expression codes appear in place of a side effect, as the body of an
3607: insn, though strictly speaking they do not describe side effects as such:
1.1 root 3608:
3609: @table @code
3610: @item (asm_input @var{s})
3611: Represents literal assembler code as described by the string @var{s}.
3612:
3613: @item (addr_vec:@var{m} [@var{lr0} @var{lr1} @dots{}])
1.1.1.2 root 3614: Represents a table of jump addresses. The vector elements @var{lr0},
3615: etc., are @samp{label_ref} expressions. The mode @var{m} specifies
3616: how much space is given to each address; normally @var{m} would be
1.1 root 3617: @code{Pmode}.
3618:
3619: @item (addr_diff_vec:@var{m} @var{base} [@var{lr0} @var{lr1} @dots{}])
3620: Represents a table of jump addresses expressed as offsets from
1.1.1.2 root 3621: @var{base}. The vector elements @var{lr0}, etc., are @samp{label_ref}
3622: expressions and so is @var{base}. The mode @var{m} specifies how much
3623: space is given to each address-difference.@refill
1.1 root 3624: @end table
3625:
1.1.1.2 root 3626: @node Incdec, Assembler, Side Effects, RTL
1.1 root 3627: @section Embedded Side-Effects on Addresses
3628:
3629: Four special side-effect expression codes appear as memory addresses.
3630:
3631: @table @code
3632: @item (pre_dec:@var{m} @var{x})
3633: Represents the side effect of decrementing @var{x} by a standard
3634: amount and represents also the value that @var{x} has after being
3635: decremented. @var{x} must be a @samp{reg} or @samp{mem}, but most
3636: machines allow only a @samp{reg}. @var{m} must be the machine mode
1.1.1.2 root 3637: for pointers on the machine in use. The amount @var{x} is decremented
1.1 root 3638: by is the length in bytes of the machine mode of the containing memory
3639: reference of which this expression serves as the address. Here is an
1.1.1.2 root 3640: example of its use:@refill
1.1 root 3641:
3642: @example
3643: (mem:DF (pre_dec:SI (reg:SI 39)))
3644: @end example
3645:
3646: @noindent
3647: This says to decrement pseudo register 39 by the length of a @code{DFmode}
3648: value and use the result to address a @code{DFmode} value.
3649:
3650: @item (pre_inc:@var{m} @var{x})
3651: Similar, but specifies incrementing @var{x} instead of decrementing it.
3652:
3653: @item (post_dec:@var{m} @var{x})
3654: Represents the same side effect as @samp{pre_decrement} but a different
3655: value. The value represented here is the value @var{x} has @i{before}
3656: being decremented.
3657:
3658: @item (post_inc:@var{m} @var{x})
3659: Similar, but specifies incrementing @var{x} instead of decrementing it.
3660: @end table
3661:
3662: These embedded side effect expressions must be used with care. Instruction
3663: patterns may not use them. Until the @samp{flow} pass of the compiler,
3664: they may occur only to represent pushes onto the stack. The @samp{flow}
3665: pass finds cases where registers are incremented or decremented in one
3666: instruction and used as an address shortly before or after; these cases are
3667: then transformed to use pre- or post-increment or -decrement.
3668:
3669: Explicit popping of the stack could be represented with these embedded
3670: side effect operators, but that would not be safe; the instruction
3671: combination pass could move the popping past pushes, thus changing
3672: the meaning of the code.
3673:
3674: An instruction that can be represented with an embedded side effect
3675: could also be represented using @samp{parallel} containing an additional
3676: @samp{set} to describe how the address register is altered. This is not
3677: done because machines that allow these operations at all typically
3678: allow them wherever a memory address is called for. Describing them as
3679: additional parallel stores would require doubling the number of entries
3680: in the machine description.
3681:
1.1.1.2 root 3682: @node Assembler, Insns, IncDec, RTL
3683: @section Assembler Instructions as Expressions
3684:
3685: The RTX code @samp{asm_operands} represents a value produced by a
3686: user-specified assembler instruction. It is used to represent
3687: an @code{asm} statement with arguments. An @code{asm} statement with
3688: a single output operand, like this:
3689:
3690: @example
3691: asm ("foo %1,%2,%0" : "a" (outputvar) : "g" (x + y), "di" (*z));
3692: @end example
3693:
3694: @noindent
3695: is represented using a single @samp{asm_operands} RTX which represents
3696: the value that is stored in @code{outputvar}:
3697:
3698: @example
3699: (set @var{rtx-for-outputvar}
3700: (asm_operands "foo %1,%2,%0" "a" 0
3701: [@var{rtx-for-addition-result} @var{rtx-for-*z}]
3702: [(asm_input:@var{m1} "g")
3703: (asm_input:@var{m2} "di")]))
3704: @end example
3705:
3706: @noindent
3707: Here the operands of the @samp{asm_operands} RTX are the assembler
3708: template string, the output-operand's constraint, the index-number of the
3709: output operand among the output operands specified, a vector of input
3710: operand RTX's, and a vector of input-operand modes and constraints. The
3711: mode @var{m1} is the mode of the sum @code{x+y}; @var{m2} is that of
3712: @code{*z}.
3713:
3714: When an @code{asm} statement has multiple output values, its insn has
3715: several such @samp{set} RTX's inside of a @samp{parallel}. Each @samp{set}
3716: contains a @samp{asm_operands}; all of these share the same assembler
3717: template and vectors, but each contains the constraint for the respective
3718: output operand. They are also distinguished by the output-operand index
3719: number, which is 0, 1, @dots{} for successive output operands.
3720:
3721: @node Insns, Calls, Assembler, RTL
1.1 root 3722: @section Insns
3723:
3724: The RTL representation of the code for a function is a doubly-linked
3725: chain of objects called @dfn{insns}. Insns are expressions with
3726: special codes that are used for no other purpose. Some insns are
3727: actual instructions; others represent dispatch tables for @code{switch}
3728: statements; others represent labels to jump to or various sorts of
1.1.1.2 root 3729: declarative information.
1.1 root 3730:
1.1.1.2 root 3731: In addition to its own specific data, each insn must have a unique id-number
1.1 root 3732: that distinguishes it from all other insns in the current function, and
3733: chain pointers to the preceding and following insns. These three fields
3734: occupy the same position in every insn, independent of the expression code
3735: of the insn. They could be accessed with @code{XEXP} and @code{XINT},
3736: but instead three special macros are always used:
3737:
3738: @table @code
3739: @item INSN_UID (@var{i})
3740: Accesses the unique id of insn @var{i}.
3741:
3742: @item PREV_INSN (@var{i})
3743: Accesses the chain pointer to the insn preceding @var{i}.
3744: If @var{i} is the first insn, this is a null pointer.
3745:
3746: @item NEXT_INSN (@var{i})
3747: Accesses the chain pointer to the insn following @var{i}.
3748: If @var{i} is the last insn, this is a null pointer.
3749: @end table
3750:
3751: The @code{NEXT_INSN} and @code{PREV_INSN} pointers must always
3752: correspond: if @var{i} is not the first insn,
3753:
3754: @example
3755: NEXT_INSN (PREV_INSN (@var{insn})) == @var{insn}
3756: @end example
3757:
3758: @noindent
3759: is always true.
3760:
3761: Every insn has one of the following six expression codes:
3762:
1.1.1.2 root 3763: @table @samp
1.1 root 3764: @item insn
3765: The expression code @samp{insn} is used for instructions that do not jump
3766: and do not do function calls. Insns with code @samp{insn} have four
3767: additional fields beyond the three mandatory ones listed above.
3768: These four are described in a table below.
3769:
3770: @item jump_insn
3771: The expression code @samp{jump_insn} is used for instructions that may jump
3772: (or, more generally, may contain @samp{label_ref} expressions).
3773: @samp{jump_insn} insns have the same extra fields as @samp{insn} insns,
3774: accessed in the same way.
3775:
3776: @item call_insn
3777: The expression code @samp{call_insn} is used for instructions that may do
3778: function calls. It is important to distinguish these instructions because
3779: they imply that certain registers and memory locations may be altered
3780: unpredictably.
3781:
3782: @samp{call_insn} insns have the same extra fields as @samp{insn} insns,
3783: accessed in the same way.
3784:
3785: @item code_label
3786: A @samp{code_label} insn represents a label that a jump insn can jump to.
3787: It contains one special field of data in addition to the three standard ones.
3788: It is used to hold the @dfn{label number}, a number that identifies this
3789: label uniquely among all the labels in the compilation (not just in the
3790: current function). Ultimately, the label is represented in the assembler
3791: output as an assembler label @samp{L@var{n}} where @var{n} is the label number.
3792:
3793: @item barrier
3794: Barriers are placed in the instruction stream after unconditional
3795: jump instructions to indicate that the jumps are unconditional.
3796: They contain no information beyond the three standard fields.
3797:
3798: @item note
3799: @samp{note} insns are used to represent additional debugging and
1.1.1.2 root 3800: declarative information. They contain two nonstandard fields, an
1.1 root 3801: integer which is accessed with the macro @code{NOTE_LINE_NUMBER} and a
3802: string accessed with @code{NOTE_SOURCE_FILE}.
3803:
3804: If @code{NOTE_LINE_NUMBER} is positive, the note represents the
3805: position of a source line and @code{NOTE_SOURCE_FILE} is the source file name
3806: that the line came from. These notes control generation of line
3807: number data in the assembler output.
3808:
3809: Otherwise, @code{NOTE_LINE_NUMBER} is not really a line number but a
3810: code with one of the following values (and @code{NOTE_SOURCE_FILE}
3811: must contain a null pointer):
3812:
3813: @table @code
3814: @item NOTE_INSN_DELETED
3815: Such a note is completely ignorable. Some passes of the compiler
3816: delete insns by altering them into notes of this kind.
3817:
3818: @item NOTE_INSN_BLOCK_BEG
3819: @itemx NOTE_INSN_BLOCK_END
3820: These types of notes indicate the position of the beginning and end
3821: of a level of scoping of variable names. They control the output
3822: of debugging information.
3823:
3824: @item NOTE_INSN_LOOP_BEG
3825: @itemx NOTE_INSN_LOOP_END
3826: These types of notes indicate the position of the beginning and end
3827: of a @code{while} or @code{for} loop. They enable the loop optimizer
3828: to find loops quickly.
3829: @end table
3830: @end table
3831:
3832: Here is a table of the extra fields of @samp{insn}, @samp{jump_insn}
3833: and @samp{call_insn} insns:
3834:
3835: @table @code
3836: @item PATTERN (@var{i})
3837: An expression for the side effect performed by this insn.
3838:
3839: @item REG_NOTES (@var{i})
3840: A list (chain of @samp{expr_list} expressions) giving information
3841: about the usage of registers in this insn. This list is set up by the
1.1.1.2 root 3842: flow analysis pass; it is a null pointer until then.
1.1 root 3843:
3844: @item LOG_LINKS (@var{i})
3845: A list (chain of @samp{insn_list} expressions) of previous ``related''
3846: insns: insns which store into registers values that are used for the
3847: first time in this insn. (An additional constraint is that neither a
3848: jump nor a label may come between the related insns). This list is
1.1.1.2 root 3849: set up by the flow analysis pass; it is a null pointer until then.
1.1 root 3850:
3851: @item INSN_CODE (@var{i})
3852: An integer that says which pattern in the machine description matches
3853: this insn, or -1 if the matching has not yet been attempted.
3854:
3855: Such matching is never attempted and this field is not used on an insn
3856: whose pattern consists of a single @samp{use}, @samp{clobber},
3857: @samp{asm}, @samp{addr_vec} or @samp{addr_diff_vec} expression.
3858: @end table
3859:
3860: The @code{LOG_LINKS} field of an insn is a chain of @samp{insn_list}
3861: expressions. Each of these has two operands: the first is an insn,
3862: and the second is another @samp{insn_list} expression (the next one in
3863: the chain). The last @samp{insn_list} in the chain has a null pointer
3864: as second operand. The significant thing about the chain is which
1.1.1.2 root 3865: insns appear in it (as first operands of @samp{insn_list}
1.1 root 3866: expressions). Their order is not significant.
3867:
3868: The @code{REG_NOTES} field of an insn is a similar chain but of
1.1.1.2 root 3869: @samp{expr_list} expressions instead of @samp{insn_list}. There are four
3870: kinds of register notes, which are distinguished by the machine mode of the
3871: @samp{expr_list}, which a register note is really understood as being an
3872: @code{enum reg_note}. The first operand @var{op} of the @samp{expr_list}
3873: is data whose meaning depends on the kind of note. Here are the four
3874: kinds:
1.1 root 3875:
3876: @table @code
3877: @item REG_DEAD
1.1.1.2 root 3878: The register @var{op} dies in this insn; that is to say, altering the
3879: value immediately after this insn would not affect the future behavior
3880: of the program.
1.1 root 3881:
3882: @item REG_INC
1.1.1.2 root 3883: The register @var{op} is incremented (or decremented; at this level
1.1 root 3884: there is no distinction) by an embedded side effect inside this insn.
1.1.1.2 root 3885: This means it appears in a @code{POST_INC}, @code{PRE_INC},
3886: @code{POST_DEC} or @code{PRE_DEC} RTX.
1.1 root 3887:
1.1.1.2 root 3888: @item REG_EQUIV
3889: The register that is set by this insn will be equal to @var{op} at run
3890: time, and could validly be replaced in all its occurrences by
3891: @var{op}. (``Validly'' here refers to the data flow of the program;
3892: simple replacement may make some insns invalid.)
3893:
3894: The value which the insn explicitly copies into the register may look
3895: different from @var{op}, but they will be equal at run time.
3896:
3897: For example, when a constant is loaded into a register that is never
3898: assigned any other value, this kind of note is used.
3899:
3900: When a parameter is copied into a pseudo-register at entry to a function,
3901: a note of this kind records that the register is equivalent to the stack
3902: slot where the parameter was passed. Although in this case the register
3903: may be set by other insns, it is still valid to replace the register
3904: by the stack slot throughout the function.
3905:
3906: @item REG_EQUAL
3907: The register that is set by this insn will be equal to @var{op} at run
3908: time at the end of this insn (but not necessarily elsewhere in the
3909: function).
3910:
3911: The RTX @var{op} is typically an arithmetic expression. For example,
3912: when a sequence of insns such as a library call is used to perform an
3913: arithmetic operation, this kind of note is attached to the insn that
3914: produces or copies the final value. It tells the CSE pass how to
3915: think of that value.
3916:
3917: @item REG_RETVAL
3918: This insn copies the value of a library call, and @var{op} is the
3919: first insn that was generated to set up the arguments for the library
3920: call.
3921:
3922: Flow analysis uses this note to delete all of a library call whose
3923: result is dead.
1.1 root 3924:
3925: @item REG_WAS_0
1.1.1.2 root 3926: The register @var{op} contained zero before this insn. You can rely
1.1 root 3927: on this note if it is present; its absence implies nothing.
3928: @end table
3929:
3930: (The only difference between the expression codes @samp{insn_list} and
3931: @samp{expr_list} is that the first operand of an @samp{insn_list} is
3932: assumed to be an insn and is printed in debugging dumps as the insn's
3933: unique id; the first operand of an @samp{expr_list} is printed in the
3934: ordinary way as an expression.)
3935:
1.1.1.2 root 3936: @node Calls, Sharing, Insns, RTL
3937: @section RTL Representation of Function-Call Insns
3938:
3939: Insns that call subroutines have the RTL expression code @samp{call_insn}.
3940: These insns must satisfy special rules, and their bodies must use a special
3941: RTL expression code, @samp{call}.
3942:
3943: A @samp{call} expression has two operands, as follows:
3944:
3945: @example
3946: (call @var{nbytes} (mem:@var{fm} @var{addr}))
3947: @end example
3948:
3949: @noindent
3950: Here @var{nbytes} is an operand that represents the number of bytes of
3951: argument data being passed to the subroutine, @var{fm} is a machine mode
3952: (which must equal as the definition of the @code{FUNCTION_MODE} macro in
3953: the machine description) and @var{addr} represents the address of the
3954: subroutine.
3955:
3956: For a subroutine that returns no value, the @samp{call} RTX as shown above
3957: is the entire body of the insn.
3958:
3959: For a subroutine that returns a value whose mode is not @code{BLKmode},
3960: the value is returned in a hard register. If this register's number is
3961: @var{r}, then the body of the call insn looks like this:
3962:
3963: @example
3964: (set (reg:@var{m} @var{r})
3965: (call @var{nbytes} (mem:@var{fm} @var{addr})))
3966: @end example
3967:
3968: @noindent
3969: This RTL expression makes it clear (to the optimizer passes) that the
3970: appropriate register receives a useful value in this insn.
3971:
3972: Immediately after RTL generation, if the value of the subroutine is
3973: actually used, this call insn is always followed closely by an insn which
1.1.1.3 ! root 3974: refers to the register @var{r}. This remains true through all the
! 3975: optimizer passes until cross jumping occurs.
! 3976:
! 3977: The following insn has one of two forms. Either it copies the value into a
! 3978: pseudo-register, like this:
1.1.1.2 root 3979:
3980: @example
3981: (set (reg:@var{m} @var{p}) (reg:@var{m} @var{r}))
3982: @end example
3983:
3984: @noindent
3985: or (in the case where the calling function will simply return whatever
3986: value the call produced, and no operation is needed to do this):
3987:
3988: @example
3989: (use (reg:@var{m} @var{r}))
3990: @end example
3991:
3992: @noindent
1.1.1.3 ! root 3993: Between the call insn and this following insn there may intervene only a
1.1.1.2 root 3994: stack-adjustment insn (and perhaps some @samp{note} insns).
3995:
3996: When a subroutine returns a @code{BLKmode} value, it is handled by
3997: passing to the subroutine the address of a place to store the value.
3998: So the call insn itself does not ``return'' any value, and it has the
3999: same RTL form as a call that returns nothing.
4000:
4001: @node Sharing,, Calls, RTL
1.1 root 4002: @section Structure Sharing Assumptions
4003:
4004: The compiler assumes that certain kinds of RTL expressions are unique;
4005: there do not exist two distinct objects representing the same value.
4006: In other cases, it makes an opposite assumption: that no RTL expression
4007: object of a certain kind appears in more than one place in the
4008: containing structure.
4009:
4010: These assumptions refer to a single function; except for the RTL
4011: objects that describe global variables and external functions,
4012: no RTL objects are common to two functions.
4013:
4014: @itemize @bullet
4015: @item
4016: Each pseudo-register has only a single @samp{reg} object to represent it,
4017: and therefore only a single machine mode.
4018:
4019: @item
4020: For any symbolic label, there is only one @samp{symbol_ref} object
4021: referring to it.
4022:
4023: @item
4024: There is only one @samp{const_int} expression with value zero,
4025: and only one with value one.
4026:
4027: @item
4028: There is only one @samp{pc} expression.
4029:
4030: @item
4031: There is only one @samp{cc0} expression.
4032:
4033: @item
4034: There is only one @samp{const_double} expression with mode
4035: @code{SFmode} and value zero, and only one with mode @code{DFmode} and
4036: value zero.
4037:
4038: @item
1.1.1.2 root 4039: No @samp{label_ref} appears in more than one place in the RTL
4040: structure; in other words, it is safe to do a tree-walk of all the
4041: insns in the function and assume that each time a @samp{label_ref} is
4042: seen it is distinct from all others that are seen.
1.1 root 4043:
4044: @item
1.1.1.2 root 4045: Only one @samp{mem} object is normally created for each static
4046: variable or stack slot, so these objects are frequently shared in all
4047: the places they appear. However, separate but equal objects for these
4048: variables are occasionally made.
4049:
4050: @item
4051: No RTL object appears in more than one place in the RTL structure
4052: except as described above. Many passes of the compiler rely on this
4053: by assuming that they can modify RTL objects in place without unwanted
4054: side-effects on other insns.
4055:
4056: @item
4057: During initial RTL generation, shared structure is freely introduced.
4058: After all the RTL for a function has been generated, all shared
4059: structure is copied by @code{unshare_all_rtl} in @file{emit-rtl.c},
4060: after which the above rules are guaranteed to be followed.
4061:
4062: @item
4063: During the combiner pass, shared structure with an insn can exist
4064: temporarily. However, the shared structure is copied before the
4065: combiner is finished with the insn. This is done by
4066: @code{copy_substitutions} in @samp{combine.c}.
1.1 root 4067: @end itemize
4068:
4069: @node Machine Desc, Machine Macros, RTL, Top
4070: @chapter Machine Descriptions
4071:
4072: A machine description has two parts: a file of instruction patterns
4073: (@file{.md} file) and a C header file of macro definitions.
4074:
4075: The @file{.md} file for a target machine contains a pattern for each
4076: instruction that the target machine supports (or at least each instruction
4077: that is worth telling the compiler about). It may also contain comments.
4078: A semicolon causes the rest of the line to be a comment, unless the semicolon
4079: is inside a quoted string.
4080:
4081: See the next chapter for information on the C header file.
4082:
4083: @menu
4084: * Patterns:: How to write instruction patterns.
1.1.1.2 root 4085: * Example:: An explained example of a @samp{define_insn} pattern.
4086: * RTL Template:: The RTL template defines what insns match a pattern.
4087: * Output Template:: The output template says how to make assembler code
4088: from such an insn.
4089: * Output Statement:: For more generality, write C code to output
4090: the assembler code.
1.1 root 4091: * Constraints:: When not all operands are general operands.
4092: * Standard Names:: Names mark patterns to use for code generation.
1.1.1.2 root 4093: * Pattern Ordering:: When the order of patterns makes a difference.
1.1 root 4094: * Dependent Patterns:: Having one pattern may make you need another.
1.1.1.2 root 4095: * Jump Patterns:: Special considerations for patterns for jump insns.
4096: * Peephole Definitions::Defining machine-specific peephole optimizations.
4097: * Expander Definitions::Generating a sequence of several RTL insns
4098: for a standard operation.
1.1 root 4099: @end menu
4100:
4101: @node Patterns, Example, Machine Desc, Machine Desc
1.1.1.2 root 4102: @section Everything about Instruction Patterns
1.1 root 4103:
4104: Each instruction pattern contains an incomplete RTL expression, with pieces
4105: to be filled in later, operand constraints that restrict how the pieces can
4106: be filled in, and an output pattern or C code to generate the assembler
4107: output, all wrapped up in a @samp{define_insn} expression.
4108:
1.1.1.2 root 4109: A @samp{define_insn} is an RTL expression containing four operands:
1.1 root 4110:
4111: @enumerate
4112: @item
4113: An optional name. The presence of a name indicate that this instruction
4114: pattern can perform a certain standard job for the RTL-generation
4115: pass of the compiler. This pass knows certain names and will use
4116: the instruction patterns with those names, if the names are defined
4117: in the machine description.
4118:
4119: The absence of a name is indicated by writing an empty string
4120: where the name should go. Nameless instruction patterns are never
4121: used for generating RTL code, but they may permit several simpler insns
4122: to be combined later on.
4123:
4124: Names that are not thus known and used in RTL-generation have no
4125: effect; they are equivalent to no name at all.
4126:
4127: @item
1.1.1.2 root 4128: The @dfn{RTL template} (@pxref{RTL Template}) is a vector of
4129: incomplete RTL expressions which show what the instruction should look
4130: like. It is incomplete because it may contain @samp{match_operand}
4131: and @samp{match_dup} expressions that stand for operands of the
1.1 root 4132: instruction.
4133:
4134: If the vector has only one element, that element is what the
4135: instruction should look like. If the vector has multiple elements,
4136: then the instruction looks like a @samp{parallel} expression
4137: containing that many elements as described.
4138:
4139: @item
4140: A condition. This is a string which contains a C expression that is
4141: the final test to decide whether an insn body matches this pattern.
4142:
4143: For a named pattern, the condition (if present) may not depend on
4144: the data in the insn being matched, but only the target-machine-type
4145: flags. The compiler needs to test these conditions during
4146: initialization in order to learn exactly which named instructions are
4147: available in a particular run.
4148:
4149: For nameless patterns, the condition is applied only when matching an
4150: individual insn, and only after the insn has matched the pattern's
4151: recognition template. The insn's operands may be found in the vector
4152: @code{operands}.
4153:
4154: @item
1.1.1.2 root 4155: The @dfn{output template}: a string that says how to output matching
4156: insns as assembler code. @samp{%} in this string specifies where
4157: to substitute the value of an operand. @xref{Output Template}.
4158:
4159: When simple substitution isn't general enough, you can specify a piece
4160: of C code to compute the output. @xref{Output Statement}.
4161: @end enumerate
4162:
4163: @node Example, RTL Template, Patterns, Machine Desc
4164: @section Example of @samp{define_insn}
4165:
4166: Here is an actual example of an instruction pattern, for the 68000/68020.
4167:
4168: @example
4169: (define_insn "tstsi"
4170: [(set (cc0)
4171: (match_operand:SI 0 "general_operand" "rm"))]
4172: ""
4173: "*
4174: @{ if (TARGET_68020 || ! ADDRESS_REG_P (operands[0]))
4175: return \"tstl %0\";
4176: return \"cmpl #0,%0\"; @}")
4177: @end example
1.1 root 4178:
1.1.1.2 root 4179: This is an instruction that sets the condition codes based on the value of
4180: a general operand. It has no condition, so any insn whose RTL description
4181: has the form shown may be handled according to this pattern. The name
4182: @samp{tstsi} means ``test a @code{SImode} value'' and tells the RTL generation
4183: pass that, when it is necessary to test such a value, an insn to do so
4184: can be constructed using this pattern.
1.1 root 4185:
1.1.1.2 root 4186: The output control string is a piece of C code which chooses which
4187: output template to return based on the kind of operand and the specific
4188: type of CPU for which code is being generated.
1.1 root 4189:
1.1.1.2 root 4190: @samp{"rm"} is an operand constraint. Its meaning is explained below.
1.1 root 4191:
1.1.1.2 root 4192: @node RTL Template, Output Template, Example, Machine Desc
4193: @section RTL Template for Generating and Recognizing Insns
1.1 root 4194:
1.1.1.2 root 4195: The RTL template is used to define which insns match the particular pattern
4196: and how to find their operands. For named patterns, the RTL template also
4197: says how to construct an insn from specified operands.
4198:
4199: Construction involves substituting specified operands into a copy of the
4200: template. Matching involves determining the values that serve as the
4201: operands in the insn being matched. Both of these activities are
4202: controlled by special expression types that direct matching and
4203: substitution of the operands.
1.1 root 4204:
4205: @table @code
4206: @item (match_operand:@var{m} @var{n} @var{testfn} @var{constraint})
4207: This expression is a placeholder for operand number @var{n} of
4208: the insn. When constructing an insn, operand number @var{n}
4209: will be substituted at this point. When matching an insn, whatever
4210: appears at this position in the insn will be taken as operand
4211: number @var{n}; but it must satisfy @var{testfn} or this instruction
4212: pattern will not match at all.
4213:
4214: Operand numbers must be chosen consecutively counting from zero in
4215: each instruction pattern. There may be only one @samp{match_operand}
4216: expression in the pattern for each expression number, and they must
4217: appear in order of increasing expression number.
4218:
4219: @var{testfn} is a string that is the name of a C function that accepts
4220: two arguments, a machine mode and an expression. During matching,
4221: the function will be called with @var{m} as the mode argument
4222: and the putative operand as the other argument. If it returns zero,
4223: this instruction pattern fails to match. @var{testfn} may be
4224: an empty string; then it means no test is to be done on the operand.
4225:
4226: Most often, @var{testfn} is @code{"general_operand"}. It checks
4227: that the putative operand is either a constant, a register or a
4228: memory reference, and that it is valid for mode @var{m}.
4229:
1.1.1.2 root 4230: For an operand that must be a register, @var{testfn} should be
4231: @code{"register_operand"}. This prevents GNU CC from creating insns
4232: that have memory references in these operands, insns which would only
4233: have to be taken apart in the reload pass.
4234:
4235: For an operand that must be a constant, either @var{testfn} should be
4236: @code{"immediate_operand"}, or the instruction pattern's extra condition
4237: should check for constants, or both.
4238:
4239: @var{constraint} is explained later (@pxref{Constraints}).
1.1 root 4240:
4241: @item (match_dup @var{n})
4242: This expression is also a placeholder for operand number @var{n}.
4243: It is used when the operand needs to appear more than once in the
4244: insn.
4245:
4246: In construction, @samp{match_dup} behaves exactly like
1.1.1.2 root 4247: @samp{match_operand}: the operand is substituted into the insn being
1.1 root 4248: constructed. But in matching, @samp{match_dup} behaves differently.
4249: It assumes that operand number @var{n} has already been determined by
1.1.1.2 root 4250: a @samp{match_operand} appearing earlier in the recognition template,
1.1 root 4251: and it matches only an identical-looking expression.
4252:
4253: @item (address (match_operand:@var{m} @var{n} "address_operand" ""))
4254: This complex of expressions is a placeholder for an operand number
4255: @var{n} in a ``load address'' instruction: an operand which specifies
4256: a memory location in the usual way, but for which the actual operand
4257: value used is the address of the location, not the contents of the
4258: location.
4259:
4260: @samp{address} expressions never appear in RTL code, only in machine
4261: descriptions. And they are used only in machine descriptions that do
4262: not use the operand constraint feature. When operand constraints are
4263: in use, the letter @samp{p} in the constraint serves this purpose.
4264:
4265: @var{m} is the machine mode of the @emph{memory location being
4266: addressed}, not the machine mode of the address itself. That mode is
4267: always the same on a given target machine (it is @code{Pmode}, which
4268: normally is @code{SImode}), so there is no point in mentioning it;
4269: thus, no machine mode is written in the @samp{address} expression. If
4270: some day support is added for machines in which addresses of different
4271: kinds of objects appear differently or are used differently (such as
4272: the PDP-10), different formats would perhaps need different machine
4273: modes and these modes might be written in the @samp{address}
4274: expression.
4275: @end table
4276:
1.1.1.2 root 4277: @node Output Template, Output Statement, RTL Template, Machine Desc
4278: @section Output Templates and Operand Substitution
1.1 root 4279:
1.1.1.2 root 4280: The @dfn{output template} is a string which specifies how to output
4281: the assembler code for an instruction pattern. Most of the template
4282: is a fixed string which is output literally. The character @samp{%}
4283: is used to specify where to substitute an operand; it can also be
4284: used to identify places different variants of the assembler require
4285: different syntax.
4286:
4287: In the simplest case, a @samp{%} followed by a digit @var{n} says to output
4288: operand @var{n} at that point in the string.
4289:
4290: @samp{%} followed by a letter and a digit says to output an operand in an
4291: alternate fashion. Four letters have standard, built-in meanings described
4292: below. The machine description macro @code{PRINT_OPERAND} can define
4293: additional letters with nonstandard meanings.
4294:
4295: @samp{%c@var{digit}} can be used to substitute an operand that is a
4296: constant value without the syntax that normally indicates an immediate
4297: operand.
4298:
4299: @samp{%n@var{digit}} is like @samp{%c@var{digit}} except that the value of
4300: the constant is negated before printing.
4301:
4302: @samp{%a@var{digit}} can be used to substitute an operand as if it were a
4303: memory reference, with the actual operand treated as the address. This may
4304: be useful when outputting a ``load address'' instruction, because often the
4305: assembler syntax for such an instruction requires you to write the operand
4306: as if it were a memory reference.
4307:
4308: @samp{%l@var{digit}} is used to substitute a @code{label_ref} into a jump
4309: instruction.
4310:
4311: @samp{%} followed by a punctuation character specifies a substitution that
4312: does not use an operand. Only one case is standard: @samp{%%} outputs a
4313: @samp{%} into the assembler code. Other nonstandard cases can be
4314: defined in the @code{PRINT_OPERAND} macro.
4315:
4316: The template may generate multiple assembler instructions. Write the text
4317: for the instructions, with @samp{\;} between them.
4318:
4319: When the RTL contains two operand which are required by constraint to match
4320: each other, the output template must refer only to the lower-numbered operand.
4321: Matching operands are not always identical, and the rest of the compiler
4322: arranges to put the proper RTL expression for printing into the lower-numbered
4323: operand.
4324:
4325: One use of nonstandard letters or punctuation following @samp{%} is to
4326: distinguish between different assembler languages for the same machine; for
4327: example, Motorola syntax versus MIT syntax for the 68000. Motorola syntax
4328: requires periods in most opcode names, while MIT syntax does not. For
4329: example, the opcode @samp{movel} in MIT syntax is @samp{move.l} in Motorola
4330: syntax. The same file of patterns is used for both kinds of output syntax,
4331: but the character sequence @samp{%.} is used in each place where Motorola
4332: syntax wants a period. The @code{PRINT_OPERAND} macro for Motorola syntax
4333: defines the sequence to output a period; the macro for MIT syntax defines
4334: it to do nothing.
4335:
4336: @node Output Statement, Constraints, Output Template, Machine Desc
4337: @section C Statements for Generating Assembler Output
4338:
4339: Often a single fixed template string cannot produce correct and efficient
4340: assembler code for all the cases that are recognized by a single
4341: instruction pattern. For example, the opcodes may depend on the kinds of
4342: operands; or some unfortunate combinations of operands may require extra
4343: machine instructions.
4344:
4345: If the output control string starts with a @samp{*}, then it is not an
4346: output template but rather a piece of C program that should compute a
4347: template. It should execute a @code{return} statement to return the
4348: template-string you want. Most such templates use C string literals, which
4349: require doublequote characters to delimit them. To include these
4350: doublequote characters in the string, prefix each one with @samp{\}.
4351:
4352: The operands may be found in the array @code{operands}, whose C data type
4353: is @code{rtx []}.
4354:
4355: It is possible to output an assembler instruction and then go on to output
4356: or compute more of them, using the subroutine @code{output_asm_insn}. This
4357: receives two arguments: a template-string and a vector of operands. The
4358: vector may be @code{operands}, or it may be another array of @code{rtx}
4359: that you declare locally and initialize yourself.
4360:
4361: When an insn pattern has multiple alternatives in its constraints, often
4362: the appearance of the assembler code determined mostly by which alternative
4363: was matched. When this is so, the C code can test the variable
4364: @code{which_alternative}, which is the ordinal number of the alternative
4365: that was actually satisfied (0 for the first, 1 for the second alternative,
4366: etc.).
4367:
4368: For example, suppose there are two opcodes for storing zero, @samp{clrreg}
4369: for registers and @samp{clrmem} for memory locations. Here is how
4370: a pattern could use @code{which_alternative} to choose between them:
1.1 root 4371:
4372: @example
1.1.1.2 root 4373: (define_insn ""
4374: [(set (match_operand:SI 0 "general_operand" "r,m")
4375: (const_int 0))]
1.1 root 4376: ""
4377: "*
1.1.1.2 root 4378: return (which_alternative == 0
4379: ? \"clrreg %0\" : \"clrmem %0\");
4380: ")
1.1 root 4381: @end example
4382:
1.1.1.2 root 4383: @node Constraints, Standard Names, Output Statement, Machine Desc
1.1 root 4384: @section Operand Constraints
4385:
4386: Each @samp{match_operand} in an instruction pattern can specify a
4387: constraint for the type of operands allowed. Constraints can say whether
4388: an operand may be in a register, and which kinds of register; whether the
4389: operand can be a memory reference, and which kinds of address; whether the
4390: operand may be an immediate constant, and which possible values it may
4391: have. Constraints can also require two operands to match.
4392:
4393: @menu
4394: * Simple Constraints:: Basic use of constraints.
1.1.1.2 root 4395: * Multi-Alternative:: When an insn has two alternative constraint-patterns.
1.1 root 4396: * Class Preferences:: Constraints guide which hard register to put things in.
4397: * Modifiers:: More precise control over effects of constraints.
4398: * No Constraints:: Describing a clean machine without constraints.
4399: @end menu
4400:
4401: @node Simple Constraints, Multi-Alternative, Constraints, Constraints
4402: @subsection Simple Constraints
4403:
4404: The simplest kind of constraint is a string full of letters, each of
4405: which describes one kind of operand that is permitted. Here are
4406: the letters that are allowed:
4407:
1.1.1.2 root 4408: @table @asis
4409: @item @samp{m}
1.1 root 4410: A memory operand is allowed, with any kind of address that the machine
4411: supports in general.
4412:
1.1.1.2 root 4413: @item @samp{o}
4414: A memory operand is allowed, but only if the address is
4415: @dfn{offsetable}. This means that adding a small integer (actually,
4416: the width in bytes of the operand, as determined by its machine mode)
4417: may be added to the address and the result is also a valid memory
4418: address.
4419:
4420: For example, an address which is constant is offsetable; so is an
4421: address that is the sum of a register and a constant (as long as a
4422: slightly larger constant is also within the range of address-offsets
4423: supported by the machine); but an autoincrement or autodecrement
4424: address is not offsetable. More complicated indirect/indexed
4425: addresses may or may not be offsetable depending on the other
4426: addressing modes that the machine supports.
4427:
4428: Note that in an output operand which can be matched by another
4429: operand, the constraint letter @samp{o} is valid only when accompanied
4430: by both @samp{<} (if the target machine has predecrement addressing)
4431: and @samp{>} (if the target machine has preincrement addressing).
1.1 root 4432:
1.1.1.2 root 4433: @item @samp{<}
1.1 root 4434: A memory operand with autodecrement addressing (either predecrement or
4435: postdecrement) is allowed.
4436:
1.1.1.2 root 4437: @item @samp{>}
1.1 root 4438: A memory operand with autoincrement addressing (either preincrement or
4439: postincrement) is allowed.
4440:
1.1.1.2 root 4441: @item @samp{r}
4442: A register operand is allowed provided that it is in a general
4443: register.
1.1 root 4444:
1.1.1.2 root 4445: @item @samp{d}, @samp{a}, @samp{f}, @dots{}
1.1 root 4446: Other letters can be defined in machine-dependent fashion to stand for
4447: particular classes of registers. @samp{d}, @samp{a} and @samp{f} are
1.1.1.2 root 4448: defined on the 68000/68020 to stand for data, address and floating
4449: point registers.
1.1 root 4450:
1.1.1.2 root 4451: @item @samp{i}
1.1 root 4452: An immediate integer operand (one with constant value) is allowed.
1.1.1.2 root 4453: This includes symbolic constants whose values will be known only at
4454: assembly time.
1.1 root 4455:
1.1.1.2 root 4456: @item @samp{n}
4457: An immediate integer operand with a known numeric value is allowed.
4458: Many systems cannot support assembly-time constants for operands less
4459: than a word wide. Constraints for these operands should use @samp{n}
4460: rather than @samp{i}.
4461:
4462: @item @samp{I}, @samp{J}, @samp{K}, @dots{}
4463: Other letters in the range @samp{I} through @samp{M} may be defined in
4464: a machine-dependent fashion to permit immediate integer operands with
4465: explicit integer values in specified ranges. For example, on the
4466: 68000, @samp{I} is defined to stand for the range of values 1 to 8.
4467: This is the range permitted as a shift count in the shift
4468: instructions.
1.1 root 4469:
1.1.1.2 root 4470: @item @samp{F}
1.1 root 4471: An immediate floating operand (expression code @samp{const_double}) is
4472: allowed.
4473:
1.1.1.2 root 4474: @item @samp{G}, @samp{H}
1.1 root 4475: @samp{G} and @samp{H} may be defined in a machine-dependent fashion to
4476: permit immediate floating operands in particular ranges of values.
4477:
1.1.1.2 root 4478: @item @samp{s}
1.1 root 4479: An immediate integer operand whose value is not an explicit integer is
1.1.1.2 root 4480: allowed.
4481:
4482: This might appear strange; if an insn allows a constant operand with a
4483: value not known at compile time, it certainly must allow any known
1.1 root 4484: value. So why use @samp{s} instead of @samp{i}? Sometimes it allows
1.1.1.2 root 4485: better code to be generated.
4486:
4487: For example, on the 68000 in a fullword instruction it is possible to
4488: use an immediate operand; but if the immediate value is between -32
4489: and 31, better code results from loading the value into a register and
4490: using the register. This is because the load into the register can be
4491: done with a @samp{moveq} instruction. We arrange for this to happen
4492: by defining the letter @samp{K} to mean ``any integer outside the
4493: range -32 to 31'', and then specifying @samp{Ks} in the operand
1.1 root 4494: constraints.
4495:
1.1.1.2 root 4496: @item @samp{g}
1.1 root 4497: Any register, memory or immediate integer operand is allowed, except for
4498: registers that are not general registers.
4499:
1.1.1.2 root 4500: @item @samp{@var{n}} (a digit)
4501: An operand that matches operand number @var{n} is allowed.
1.1 root 4502: If a digit is used together with letters, the digit should come last.
4503:
1.1.1.2 root 4504: This is called a @dfn{matching constraint} and what it really means is
4505: that the assembler has only a single operand that fills two roles
4506: considered separate in the RTL insn. For example, an add insn has two
4507: input operands and one output operand in the RTL, but on most machines
4508: an add instruction really has only two operands, one of them an
4509: input-output operand.
4510:
4511: Matching constraints work only in circumstances like that add insn.
4512: More precisely, the matching constraint must appear in an input-only
4513: operand and the operand that it matches must be an output-only operand
4514: with a lower number.
4515:
4516: For operands to match in a particular case usually means that they
4517: are identical-looking RTL expressions. But in a few special cases
4518: specific kinds of dissimilarity are allowed. For example, @code{*x}
4519: as an input operand will match @code{*x++} as an output operand.
4520: For proper results in such cases, the output template should always
4521: use the output-operand's number when printing the operand.
4522:
4523: @item @samp{p}
1.1 root 4524: An operand that is a valid memory address is allowed. This is
4525: for ``load address'' and ``push address'' instructions.
4526:
4527: If @samp{p} is used in the constraint, the test-function in the
4528: @samp{match_operand} must be @code{address_operand}.
4529: @end table
4530:
4531: In order to have valid assembler code, each operand must satisfy
4532: its constraint. But a failure to do so does not prevent the pattern
4533: from applying to an insn. Instead, it directs the compiler to modify
1.1.1.2 root 4534: the code so that the constraint will be satisfied. Usually this is
1.1 root 4535: done by copying an operand into a register.
4536:
4537: Contrast, therefore, the two instruction patterns that follow:
4538:
4539: @example
4540: (define_insn ""
4541: [(set (match_operand:SI 0 "general_operand" "r")
4542: (plus:SI (match_dup 0)
4543: (match_operand:SI 1 "general_operand" "r")))]
4544: ""
4545: "@dots{}")
4546: @end example
4547:
4548: @noindent
4549: which has two operands, one of which must appear in two places, and
4550:
4551: @example
4552: (define_insn ""
4553: [(set (match_operand:SI 0 "general_operand" "r")
4554: (plus:SI (match_operand:SI 1 "general_operand" "0")
4555: (match_operand:SI 2 "general_operand" "r")))]
4556: ""
4557: "@dots{}")
4558: @end example
4559:
4560: @noindent
4561: which has three operands, two of which are required by a constraint to be
4562: identical. If we are considering an insn of the form
4563:
4564: @example
4565: (insn @var{n} @var{prev} @var{next}
4566: (set (reg:SI 3)
4567: (plus:SI (reg:SI 6) (reg:SI 109)))
4568: @dots{})
4569: @end example
4570:
4571: @noindent
4572: the first pattern would not apply at all, because this insn does not
4573: contain two identical subexpressions in the right place. The pattern would
4574: say, ``That does not look like an add instruction; try other patterns.''
4575: The second pattern would say, ``Yes, that's an add instruction, but there
4576: is something wrong with it.'' It would direct the reload pass of the
4577: compiler to generate additional insns to make the constraint true. The
4578: results might look like this:
4579:
4580: @example
4581: (insn @var{n2} @var{prev} @var{n}
4582: (set (reg:SI 3) (reg:SI 6))
4583: @dots{})
4584:
4585: (insn @var{n} @var{n2} @var{next}
4586: (set (reg:SI 3)
4587: (plus:SI (reg:SI 3) (reg:SI 109)))
4588: @dots{})
4589: @end example
4590:
4591: Because insns that don't fit the constraints are fixed up by loading
4592: operands into registers, every instruction pattern's constraints must
4593: permit the case where all the operands are in registers. It need not
4594: permit all classes of registers; the compiler knows how to copy registers
4595: into other registers of the proper class in order to make an instruction
4596: valid. But if no registers are permitted, the compiler will be stymied: it
4597: does not know how to save a register in memory in order to make an
4598: instruction valid. Instruction patterns that reject registers can be
4599: made valid by attaching a condition-expression that refuses to match
4600: an insn at all if the crucial operand is a register.
4601:
4602: @node Multi-Alternative, Class Preferences, Simple Constraints, Constraints
4603: @subsection Multiple Alternative Constraints
4604:
4605: Sometimes a single instruction has multiple alternative sets of possible
4606: operands. For example, on the 68000, a logical-or instruction can combine
4607: register or an immediate value into memory, or it can combine any kind of
4608: operand into a register; but it cannot combine one memory location into
4609: another.
4610:
4611: These constraints are represented as multiple alternatives. An alternative
4612: can be described by a series of letters for each operand. The overall
4613: constraint for an operand is made from the letters for this operand
4614: from the first alternative, a comma, the letters for this operand from
4615: the second alternative, a comma, and so on until the last alternative.
4616: Here is how it is done for fullword logical-or on the 68000:
4617:
4618: @example
4619: (define_insn "iorsi3"
4620: [(set (match_operand:SI 0 "general_operand" "=%m,d")
1.1.1.2 root 4621: (ior:SI (match_operand:SI 1 "general_operand" "0,0")
4622: (match_operand:SI 2 "general_operand" "dKs,dmKs")))]
1.1 root 4623: @dots{})
4624: @end example
4625:
4626: The first alternative has @samp{m} (memory) for operand 0, @samp{0} for
4627: operand 1 (meaning it must match operand 0), and @samp{dKs} for operand 2.
4628: The second alternative has @samp{d} (data register) for operand 0, @samp{0}
4629: for operand 1, and @samp{dmKs} for operand 2. The @samp{=} and @samp{%} in
4630: the constraint for operand 0 are not part of any alternative; their meaning
4631: is explained in the next section.
4632:
4633: If all the operands fit any one alternative, the instruction is valid.
4634: Otherwise, for each alternative, the compiler counts how many instructions
4635: must be added to copy the operands so that that alternative applies.
4636: The alternative requiring the least copying is chosen. If two alternatives
4637: need the same amount of copying, the one that comes first is chosen.
4638: These choices can be altered with the @samp{?} and @samp{!} characters:
4639:
4640: @table @samp
4641: @item ?
4642: Disparage slightly the alternative that the @samp{?} appears in,
4643: as a choice when no alternative applies exactly. The compiler regards
4644: this alternative as one unit more costly for each @samp{?} that appears
4645: in it.
4646:
4647: @item !
4648: Disparage severely the alternative that the @samp{!} appears in.
4649: When operands must be copied into registers, the compiler will
4650: never choose this alternative as the one to strive for.
4651: @end table
4652:
1.1.1.2 root 4653: When an insn pattern has multiple alternatives in its constraints,
4654: often the appearance of the assembler code determined mostly by which
4655: alternative was matched. When this is so, the C code for writing the
4656: assembler code can use the variable @code{which_alternative}, which is
4657: the ordinal number of the alternative that was actually satisfied
4658: (0 for the first, 1 for the second alternative, etc.). For example:
4659:
4660: @example
4661: (define_insn ""
4662: [(set (match_operand:SI 0 "general_operand" "r,m")
4663: (const_int 0))]
4664: ""
4665: "*
4666: return (which_alternative == 0
4667: ? \"clrreg %0\" : \"clrmem %0\");
4668: ")
4669: @end example
4670:
1.1 root 4671: @node Class Preferences, Modifiers, Multi-Alternative, Constraints
4672: @subsection Register Class Preferences
4673:
4674: The operand constraints have another function: they enable the compiler
4675: to decide which kind of hardware register a pseudo register is best
4676: allocated to. The compiler examines the constraints that apply to the
4677: insns that use the pseudo register, looking for the machine-dependent
4678: letters such as @samp{d} and @samp{a} that specify classes of registers.
4679: The pseudo register is put in whichever class gets the most ``votes''.
4680: The constraint letters @samp{g} and @samp{r} also vote: they vote in
4681: favor of a general register. The machine description says which registers
4682: are considered general.
4683:
4684: Of course, on some machines all registers are equivalent, and no register
4685: classes are defined. Then none of this complexity is relevant.
4686:
4687: @node Modifiers, No Constraints, Class Preferences, Constraints
4688: @subsection Constraint Modifier Characters
4689:
4690: @table @samp
4691: @item =
1.1.1.2 root 4692: Means that this operand is write-only for this instruction: the previous
4693: value is discarded and replaced by output data.
1.1 root 4694:
4695: @item +
4696: Means that this operand is both read and written by the instruction.
4697:
4698: When the compiler fixes up the operands to satisfy the constraints,
4699: it needs to know which operands are inputs to the instruction and
4700: which are outputs from it. @samp{=} identifies an output; @samp{+}
4701: identifies an operand that is both input and output; all other operands
4702: are assumed to be input only.
4703:
1.1.1.2 root 4704: @item &
4705: Means (in a particular alternative) that this operand is written
4706: before the instruction is finished using the input operands.
4707: Therefore, this operand may not lie in a register that is used as an
4708: input operand or as part of any memory address.
4709:
4710: @samp{&} applies only to the alternative in which it is written. In
4711: constraints with multiple alternatives, sometimes one alternative
4712: requires @samp{&} while others do not. See, for example, the
4713: @samp{movdf} insn of the 68000.
4714:
4715: @samp{&} does not obviate the need to write @samp{=}.
4716:
1.1 root 4717: @item %
1.1.1.2 root 4718: Declares the instruction to be commutative for this operand and the
4719: following operand. This means that the compiler may interchange the
4720: two operands if that is the cheapest way to make all operands fit the
4721: constraints. This is often used in patterns for addition instructions
4722: that really have only two operands: the result must go in one of the
4723: arguments. Here for example, is how the 68000 halfword-add
4724: instruction is defined:
4725:
4726: @example
4727: (define_insn "addhi3"
4728: [(set (match_operand:HI 0 "general_operand" "=m,r")
4729: (plus:HI (match_operand:HI 1 "general_operand" "%0,0")
4730: (match_operand:HI 2 "general_operand" "di,g")))]
4731: @dots{})
4732: @end example
4733:
4734: Note that in previous versions of GNU CC the @samp{%} constraint
4735: modifier always applied to operands 1 and 2 regardless of which
4736: operand it was written in. The usual custom was to write it in
4737: operand 0. Now it must be in operand 1 if the operands to be
4738: exchanged are 1 and 2.
1.1 root 4739:
4740: @item #
1.1.1.2 root 4741: Says that all following characters, up to the next comma, are to be
4742: ignored as a constraint. They are significant only for choosing
4743: register preferences.
1.1 root 4744:
4745: @item *
4746: Says that the following character should be ignored when choosing
1.1.1.2 root 4747: register preferences. @samp{*} has no effect on the meaning of the
4748: constraint as a constraint.
4749:
4750: Here is an example: the 68000 has an instruction to sign-extend a
4751: halfword in a data register, and can also sign-extend a value by
4752: copying it into an address register. While either kind of register is
4753: acceptable, the constraints on an address-register destination are
4754: less strict, so it is best if register allocation makes an address
4755: register its goal. Therefore, @samp{*} is used so that the @samp{d}
4756: constraint letter (for data register) is ignored when computing
4757: register preferences.
4758:
4759: @example
4760: (define_insn "extendhisi2"
4761: [(set (match_operand:SI 0 "general_operand" "=*d,a")
4762: (sign_extend:SI
4763: (match_operand:HI 1 "general_operand" "0,g")))]
4764: @dots{})
4765: @end example
1.1 root 4766: @end table
4767:
4768: @node No Constraints,, Modifiers, Constraints
4769: @subsection Not Using Constraints
4770:
4771: Some machines are so clean that operand constraints are not required. For
4772: example, on the Vax, an operand valid in one context is valid in any other
1.1.1.2 root 4773: context. On such a machine, every operand constraint would be @samp{g},
1.1 root 4774: excepting only operands of ``load address'' instructions which are
4775: written as if they referred to a memory location's contents but actual
1.1.1.2 root 4776: refer to its address. They would have constraint @samp{p}.
1.1 root 4777:
1.1.1.2 root 4778: For such machines, instead of writing @samp{g} and @samp{p} for all
1.1 root 4779: the constraints, you can choose to write a description with empty constraints.
4780: Then you write @samp{""} for the constraint in every @samp{match_operand}.
4781: Address operands are identified by writing an @samp{address} expression
4782: around the @samp{match_operand}, not by their constraints.
4783:
4784: When the machine description has just empty constraints, certain parts
4785: of compilation are skipped, making the compiler faster.
4786:
1.1.1.2 root 4787: @node Standard Names, Pattern Ordering, Constraints, Machine Desc
4788: @section Standard Names for Patterns Used in Generation
1.1 root 4789:
4790: Here is a table of the instruction names that are meaningful in the RTL
4791: generation pass of the compiler. Giving one of these names to an
4792: instruction pattern tells the RTL generation pass that it can use the
4793: pattern in to accomplish a certain task.
4794:
1.1.1.2 root 4795: @table @asis
4796: @item @samp{mov@var{m}}
1.1 root 4797: Here @var{m} is a two-letter machine mode name, in lower case. This
4798: instruction pattern moves data with that machine mode from operand 1 to
4799: operand 0. For example, @samp{movsi} moves full-word data.
4800:
4801: If operand 0 is a @samp{subreg} with mode @var{m} of a register whose
4802: natural mode is wider than @var{m}, the effect of this instruction is
4803: to store the specified value in the part of the register that corresponds
4804: to mode @var{m}. The effect on the rest of the register is undefined.
4805:
1.1.1.3 ! root 4806: This class of patterns is special in several ways. First of all, each
! 4807: of these names @emph{must} be defined, because there is no other way
! 4808: to copy a datum from one place to another.
! 4809:
! 4810: Second, these patterns are not used solely in the RTL generation pass.
! 4811: Even the reload pass can generate move insns to copy values from stack
! 4812: slots into temporary registers. When it does so, one of the operands
! 4813: is a hard register and the other is an operand that can have a reload.
! 4814:
! 4815: Therefore, when given such a pair of operands, the pattern must
! 4816: generate RTL which needs no temporary registers---no registers other
! 4817: than the operands. For example, if you support the pattern with a
! 4818: @code{define_expand}, then in such a case you mustn't call
! 4819: @code{force_reg} or any other such function which might generate new
! 4820: pseudo registers.
! 4821:
! 4822: This requirement exists even for subword modes on a RISC machine where
! 4823: fetching those modes from memory normally requires several insns and
! 4824: some temporary registers. Look in @file{spur.md} to see how the
! 4825: requirement is satisfied.
! 4826:
! 4827: The variety of operands that have reloads depends on the rest of the
! 4828: machine description, but typically on a RISC machine these can only be
! 4829: pseudo registers that did not get hard registers, while on other
! 4830: machines explicit memory references will get optional reloads.
! 4831:
1.1.1.2 root 4832: @item @samp{movstrict@var{m}}
1.1 root 4833: Like @samp{mov@var{m}} except that if operand 0 is a @samp{subreg}
4834: with mode @var{m} of a register whose natural mode is wider,
4835: the @samp{movstrict@var{m}} instruction is guaranteed not to alter
4836: any of the register except the part which belongs to mode @var{m}.
4837:
1.1.1.2 root 4838: @item @samp{add@var{m}3}
1.1 root 4839: Add operand 2 and operand 1, storing the result in operand 0. All operands
4840: must have mode @var{m}. This can be used even on two-address machines, by
4841: means of constraints requiring operands 1 and 0 to be the same location.
4842:
1.1.1.2 root 4843: @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 4844: Similar, for other arithmetic operations.
4845:
1.1.1.2 root 4846: @item @samp{andcb@var{m}3}
1.1 root 4847: Bitwise logical-and operand 1 with the complement of operand 2
4848: and store the result in operand 0.
4849:
1.1.1.2 root 4850: @item @samp{mulhisi3}
1.1 root 4851: Multiply operands 1 and 2, which have mode @code{HImode}, and store
4852: a @code{SImode} product in operand 0.
4853:
1.1.1.2 root 4854: @item @samp{mulqihi3}, @samp{mulsidi3}
1.1 root 4855: Similar widening-multiplication instructions of other widths.
4856:
1.1.1.2 root 4857: @item @samp{umulqihi3}, @samp{umulhisi3}, @samp{umulsidi3}
1.1 root 4858: Similar widening-multiplication instructions that do unsigned
4859: multiplication.
4860:
1.1.1.2 root 4861: @item @samp{divmod@var{m}4}
1.1 root 4862: Signed division that produces both a quotient and a remainder.
4863: Operand 1 is divided by operand 2 to produce a quotient stored
4864: in operand 0 and a remainder stored in operand 3.
4865:
1.1.1.2 root 4866: @item @samp{udivmod@var{m}4}
1.1 root 4867: Similar, but does unsigned division.
4868:
1.1.1.2 root 4869: @item @samp{divmod@var{m}@var{n}4}
1.1 root 4870: Like @samp{divmod@var{m}4} except that only the dividend has mode
4871: @var{m}; the divisor, quotient and remainder have mode @var{n}.
4872: For example, the Vax has a @samp{divmoddisi4} instruction
4873: (but it is omitted from the machine description, because it
4874: is so slow that it is faster to compute remainders by the
4875: circumlocution that the compiler will use if this instruction is
4876: not available).
4877:
1.1.1.2 root 4878: @item @samp{ashl@var{m}3}
1.1 root 4879: Arithmetic-shift operand 1 left by a number of bits specified by
4880: operand 2, and store the result in operand 0. Operand 2 has
4881: mode @code{SImode}, not mode @var{m}.
4882:
1.1.1.2 root 4883: @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 4884: Other shift and rotate instructions.
4885:
1.1.1.2 root 4886: Logical and arithmetic left shift are the same. Machines that do not
4887: allow negative shift counts often have only one instruction for
4888: shifting left. On such machines, you should define a pattern named
4889: @samp{ashl@var{m}3} and leave @samp{lshl@var{m}3} undefined.
4890:
4891: @item @samp{neg@var{m}2}
1.1 root 4892: Negate operand 1 and store the result in operand 0.
4893:
1.1.1.2 root 4894: @item @samp{abs@var{m}2}
1.1 root 4895: Store the absolute value of operand 1 into operand 0.
4896:
1.1.1.2 root 4897: @item @samp{sqrt@var{m}2}
1.1 root 4898: Store the square root of operand 1 into operand 0.
4899:
1.1.1.2 root 4900: @item @samp{ffs@var{m}2}
4901: Store into operand 0 one plus the index of the least significant 1-bit
4902: of operand 1. If operand 1 is zero, store zero. @var{m} is the mode
4903: of operand 0; operand 1's mode is specified by the instruction
4904: pattern, and the compiler will convert the operand to that mode before
4905: generating the instruction.
4906:
4907: @item @samp{one_cmpl@var{m}2}
1.1 root 4908: Store the bitwise-complement of operand 1 into operand 0.
4909:
1.1.1.2 root 4910: @item @samp{cmp@var{m}}
1.1 root 4911: Compare operand 0 and operand 1, and set the condition codes.
1.1.1.2 root 4912: The RTL pattern should look like this:
1.1 root 4913:
1.1.1.2 root 4914: @example
4915: (set (cc0) (minus (match_operand:@var{m} 0 @dots{})
4916: (match_operand:@var{m} 1 @dots{})))
4917: @end example
4918:
4919: Each such definition in the machine description, for integer mode
4920: @var{m}, must have a corresponding @samp{tst@var{m}} pattern, because
4921: optimization can simplify the compare into a test when operand 1 is
4922: zero.
4923:
4924: @item @samp{tst@var{m}}
1.1 root 4925: Compare operand 0 against zero, and set the condition codes.
1.1.1.2 root 4926: The RTL pattern should look like this:
1.1 root 4927:
1.1.1.2 root 4928: @example
4929: (set (cc0) (match_operand:@var{m} 0 @dots{}))
4930: @end example
4931:
4932: @item @samp{movstr@var{m}}
1.1 root 4933: Block move instruction. The addresses of the destination and source
4934: strings are the first two operands, and both are in mode @code{Pmode}.
4935: The number of bytes to move is the third operand, in mode @var{m}.
4936:
1.1.1.2 root 4937: @item @samp{cmpstr@var{m}}
1.1 root 4938: Block compare instruction, with operands like @samp{movstr@var{m}}
4939: except that the two memory blocks are compared byte by byte
4940: in lexicographic order. The effect of the instruction is to set
4941: the condition codes.
4942:
1.1.1.2 root 4943: @item @samp{float@var{m}@var{n}2}
1.1 root 4944: Convert operand 1 (valid for fixed point mode @var{m}) to floating
4945: point mode @var{n} and store in operand 0 (which has mode @var{n}).
4946:
1.1.1.2 root 4947: @item @samp{fix@var{m}@var{n}2}
4948: Convert operand 1 (valid for floating point mode @var{m}) to fixed
4949: point mode @var{n} as a signed number and store in operand 0 (which
4950: has mode @var{n}). This instruction's result is defined only when
4951: the value of operand 1 is an integer.
4952:
4953: @item @samp{fixuns@var{m}@var{n}2}
4954: Convert operand 1 (valid for floating point mode @var{m}) to fixed
4955: point mode @var{n} as an unsigned number and store in operand 0 (which
4956: has mode @var{n}). This instruction's result is defined only when the
4957: value of operand 1 is an integer.
4958:
4959: @item @samp{ftrunc@var{m}2}
4960: Convert operand 1 (valid for floating point mode @var{m}) to an
4961: integer value, still represented in floating point mode @var{m}, and
4962: store it in operand 0 (valid for floating point mode @var{m}).
4963:
4964: @item @samp{fix_trunc@var{m}@var{n}2}
4965: Like @samp{fix@var{m}@var{n}2} but works for any floating point value
4966: of mode @var{m} by converting the value to an integer.
4967:
4968: @item @samp{fixuns_trunc@var{m}@var{n}2}
4969: Like @samp{fixuns@var{m}@var{n}2} but works for any floating point
4970: value of mode @var{m} by converting the value to an integer.
4971:
4972: @item @samp{trunc@var{m}@var{n}}
1.1 root 4973: Truncate operand 1 (valid for mode @var{m}) to mode @var{n} and
4974: store in operand 0 (which has mode @var{n}). Both modes must be fixed
4975: point or both floating point.
4976:
1.1.1.2 root 4977: @item @samp{extend@var{m}@var{n}}
1.1 root 4978: Sign-extend operand 1 (valid for mode @var{m}) to mode @var{n} and
4979: store in operand 0 (which has mode @var{n}). Both modes must be fixed
4980: point or both floating point.
4981:
1.1.1.2 root 4982: @item @samp{zero_extend@var{m}@var{n}}
1.1 root 4983: Zero-extend operand 1 (valid for mode @var{m}) to mode @var{n} and
4984: store in operand 0 (which has mode @var{n}). Both modes must be fixed
4985: point.
4986:
1.1.1.2 root 4987: @item @samp{extv}
1.1 root 4988: Extract a bit-field from operand 1 (a register or memory operand),
4989: where operand 2 specifies the width in bits and operand 3 the starting
4990: bit, and store it in operand 0. Operand 0 must have @code{Simode}.
4991: Operand 1 may have mode @code{QImode} or @code{SImode}; often
4992: @code{SImode} is allowed only for registers. Operands 2 and 3 must be
4993: valid for @code{SImode}.
4994:
4995: The RTL generation pass generates this instruction only with constants
4996: for operands 2 and 3.
4997:
4998: The bit-field value is sign-extended to a full word integer
4999: before it is stored in operand 0.
5000:
1.1.1.2 root 5001: @item @samp{extzv}
1.1 root 5002: Like @samp{extv} except that the bit-field value is zero-extended.
5003:
1.1.1.2 root 5004: @item @samp{insv}
1.1 root 5005: Store operand 3 (which must be valid for @code{SImode}) into a
5006: bit-field in operand 0, where operand 1 specifies the width in bits
5007: and operand 2 the starting bit. Operand 0 may have mode @code{QImode}
5008: or @code{SImode}; often @code{SImode} is allowed only for registers.
5009: Operands 1 and 2 must be valid for @code{SImode}.
5010:
5011: The RTL generation pass generates this instruction only with constants
5012: for operands 1 and 2.
5013:
1.1.1.2 root 5014: @item @samp{s@var{cond}}
5015: Store zero or nonzero in the operand according to the condition codes.
5016: Value stored is nonzero iff the condition @var{cond} is true.
5017: @var{cond} is the name of a comparison operation expression code, such
1.1 root 5018: as @samp{eq}, @samp{lt} or @samp{leu}.
5019:
1.1.1.2 root 5020: You specify the mode that the operand must have when you write the
5021: @code{match_operand} expression. The compiler automatically sees
5022: which mode you have used and supplies an operand of that mode.
5023:
5024: The value stored for a true condition must have 1 as its low bit.
5025: Otherwise the instruction is not suitable and must be omitted from the
5026: machine description. You must tell the compiler exactly which value
5027: is stored by defining the macro @code{STORE_FLAG_VALUE}.
5028:
5029: @item @samp{b@var{cond}}
1.1 root 5030: Conditional branch instruction. Operand 0 is a @samp{label_ref}
5031: that refers to the label to jump to. Jump if the condition codes
5032: meet condition @var{cond}.
5033:
1.1.1.2 root 5034: @item @samp{call}
5035: Subroutine call instruction. Operand 1 is the number of bytes of
5036: arguments pushed (in mode @code{SImode}), and operand 0 is the
5037: function to call. Operand 0 should be a @samp{mem} RTX whose address
5038: is the address of the function.
1.1 root 5039:
1.1.1.2 root 5040: @item @samp{return}
1.1 root 5041: Subroutine return instruction. This instruction pattern name should be
5042: defined only if a single instruction can do all the work of returning
5043: from a function.
5044:
1.1.1.3 ! root 5045: @item @samp{casesi}
! 5046: Instruction to jump through a dispatch table, including bounds checking.
! 5047: This instruction takes five operands:
! 5048:
! 5049: @enumerate
! 5050: @item
! 5051: The index to dispatch on, which has mode @code{SImode}.
! 5052:
! 5053: @item
! 5054: The lower bound for indices in the table, an integer constant.
! 5055:
! 5056: @item
! 5057: The upper bound for indices in the table, an integer constant.
! 5058:
! 5059: @item
! 5060: A label to jump to if the index has a value outside the bounds.
! 5061: (If the machine-description macro @code{CASE_DROPS_THROUGH} is defined,
! 5062: then an out-of-bounds index drops through to the code following
! 5063: the jump table instead of jumping to this label. In that case,
! 5064: this label is not actually used by the @samp{casesi} instruction,
! 5065: but it is always provided as an operand.)
! 5066:
! 5067: @item
! 5068: A label that precedes the table itself.
! 5069: @end enumerate
! 5070:
! 5071: The table is a @samp{addr_vec} or @samp{addr_diff_vec} inside of a
! 5072: @samp{jump_insn}. The number of elements in the table is one plus the
! 5073: difference between the upper bound and the lower bound.
! 5074:
1.1.1.2 root 5075: @item @samp{tablejump}
1.1.1.3 ! root 5076: Instruction to jump to a variable address. This is a low-level
! 5077: capability which can be used to implement a dispatch table when there
! 5078: is no @samp{casesi} pattern.
! 5079:
! 5080: This pattern requires two operands: the address or offset, and a label
! 5081: which should immediately precede the jump table. If the macro
! 5082: @code{CASE_VECTOR_PC_RELATIVE} is defined then the first operand is an
! 5083: absolute address to jump to; otherwise, it is an offset which counts
! 5084: from the address of the table.
! 5085:
! 5086: The @samp{tablejump} insn is always the last insn before the jump
! 5087: table it uses. Its assembler code normally has no need to use the
! 5088: second operand, but you should incorporate it in the RTL pattern so
! 5089: that the jump optimizer will not delete the table as unreachable code.
1.1 root 5090: @end table
5091:
1.1.1.2 root 5092: @node Pattern Ordering, Dependent Patterns, Standard Names, Machine Desc
5093: @section When the Order of Patterns Matters
5094:
5095: Sometimes an insn can match more than one instruction pattern. Then the
5096: pattern that appears first in the machine description is the one used.
5097: Therefore, more specific patterns (patterns that will match fewer things)
5098: and faster instructions (those that will produce better code when they
5099: do match) should usually go first in the description.
5100:
5101: In some cases the effect of ordering the patterns can be used to hide
5102: a pattern when it is not valid. For example, the 68000 has an
5103: instruction for converting a fullword to floating point and another
5104: for converting a byte to floating point. An instruction converting
5105: an integer to floating point could match either one. We put the
5106: pattern to convert the fullword first to make sure that one will
5107: be used rather than the other. (Otherwise a large integer might
5108: be generated as a single-byte immediate quantity, which would not work.)
5109: Instead of using this pattern ordering it would be possible to make the
5110: pattern for convert-a-byte smart enough to deal properly with any
5111: constant value.
5112:
5113: @node Dependent Patterns, Jump Patterns, Pattern Ordering, Machine Desc
5114: @section Interdependence of Patterns
1.1 root 5115:
5116: Every machine description must have a named pattern for each of the
5117: conditional branch names @samp{b@var{cond}}. The recognition template
5118: must always have the form
5119:
5120: @example
5121: (set (pc)
5122: (if_then_else (@var{cond} (cc0) (const_int 0))
5123: (label_ref (match_operand 0 "" ""))
5124: (pc)))
5125: @end example
5126:
5127: @noindent
5128: In addition, every machine description must have an anonymous pattern
5129: for each of the possible reverse-conditional branches. These patterns
5130: look like
5131:
5132: @example
5133: (set (pc)
5134: (if_then_else (@var{cond} (cc0) (const_int 0))
5135: (pc)
5136: (label_ref (match_operand 0 "" ""))))
5137: @end example
5138:
5139: @noindent
5140: They are necessary because jump optimization can turn direct-conditional
5141: branches into reverse-conditional branches.
5142:
5143: The compiler does more with RTL than just create it from patterns
5144: and recognize the patterns: it can perform arithmetic expression codes
5145: when constant values for their operands can be determined. As a result,
5146: sometimes having one pattern can require other patterns. For example, the
5147: Vax has no `and' instruction, but it has `and not' instructions. Here
5148: is the definition of one of them:
5149:
5150: @example
5151: (define_insn "andcbsi2"
5152: [(set (match_operand:SI 0 "general_operand" "")
5153: (and:SI (match_dup 0)
5154: (not:SI (match_operand:SI
5155: 1 "general_operand" ""))))]
5156: ""
5157: "bicl2 %1,%0")
5158: @end example
5159:
5160: @noindent
5161: If operand 1 is an explicit integer constant, an instruction constructed
1.1.1.2 root 5162: using that pattern can be simplified into an `and' like this:
1.1 root 5163:
5164: @example
5165: (set (reg:SI 41)
5166: (and:SI (reg:SI 41)
5167: (const_int 0xffff7fff)))
5168: @end example
5169:
5170: @noindent
5171: (where the integer constant is the one's complement of what
5172: appeared in the original instruction).
5173:
5174: To avoid a fatal error, the compiler must have a pattern that recognizes
5175: such an instruction. Here is what is used:
5176:
5177: @example
5178: (define_insn ""
5179: [(set (match_operand:SI 0 "general_operand" "")
5180: (and:SI (match_dup 0)
5181: (match_operand:SI 1 "general_operand" "")))]
5182: "GET_CODE (operands[1]) == CONST_INT"
5183: "*
1.1.1.2 root 5184: @{ operands[1]
1.1 root 5185: = gen_rtx (CONST_INT, VOIDmode, ~INTVAL (operands[1]));
5186: return \"bicl2 %1,%0\";
1.1.1.2 root 5187: @}")
1.1 root 5188: @end example
5189:
5190: @noindent
5191: Whereas a pattern to match a general `and' instruction is impossible to
5192: support on the Vax, this pattern is possible because it matches only a
5193: constant second argument: a special case that can be output as an `and not'
5194: instruction.
5195:
1.1.1.2 root 5196: A ``compare'' instruction whose RTL looks like this:
5197:
5198: @example
5199: (set (cc0) (minus @var{operand} (const_int 0)))
5200: @end example
5201:
5202: @noindent
5203: may be simplified by optimization into a ``test'' like this:
5204:
5205: @example
5206: (set (cc0) @var{operand})
5207: @end example
5208:
5209: @noindent
5210: So in the machine description, each ``compare'' pattern for an integer
5211: mode must have a corresponding ``test'' pattern that will match the
5212: result of such simplification.
5213:
5214: In some cases machines support instructions identical except for the
5215: machine mode of one or more operands. For example, there may be
5216: ``sign-extend halfword'' and ``sign-extend byte'' instructions whose
5217: patterns are
5218:
5219: @example
5220: (set (match_operand:SI 0 @dots{})
5221: (extend:SI (match_operand:HI 1 @dots{})))
5222:
5223: (set (match_operand:SI 0 @dots{})
5224: (extend:SI (match_operand:QI 1 @dots{})))
5225: @end example
5226:
5227: @noindent
5228: Constant integers do not specify a machine mode, so an instruction to
5229: extend a constant value could match either pattern. The pattern it
5230: actually will match is the one that appears first in the file. For correct
5231: results, this must be the one for the widest possible mode (@code{HImode},
5232: here). If the pattern matches the @code{QImode} instruction, the results
5233: will be incorrect if the constant value does not actually fit that mode.
5234:
5235: Such instructions to extend constants are rarely generated because they are
5236: optimized away, but they do occasionally happen in nonoptimized
5237: compilations.
5238:
5239: @node Jump Patterns, Peephole Definitions, Dependent Patterns, Machine Desc
5240: @section Defining Jump Instruction Patterns
5241:
5242: GNU CC assumes that the machine has a condition code. A comparison insn
5243: sets the condition code, recording the results of both signed and unsigned
5244: comparison of the given operands. A separate branch insn tests the
5245: condition code and branches or not according its value. The branch insns
5246: come in distinct signed and unsigned flavors. Many common machines, such
5247: as the Vax, the 68000 and the 32000, work this way.
5248:
5249: Some machines have distinct signed and unsigned compare instructions, and
5250: only one set of conditional branch instructions. The easiest way to handle
5251: these machines is to treat them just like the others until the final stage
5252: where assembly code is written. At this time, when outputting code for the
5253: compare instruction, peek ahead at the following branch using
5254: @code{NEXT_INSN (insn)}. (The variable @code{insn} refers to the insn
5255: being output, in the output-writing code in an instruction pattern.) If
5256: the RTL says that is an unsigned branch, output an unsigned compare;
5257: otherwise output a signed compare. When the branch itself is output, you
5258: can treat signed and unsigned branches identically.
5259:
5260: The reason you can do this is that GNU CC always generates a pair of
5261: consecutive RTL insns, one to set the condition code and one to test it,
5262: and keeps the pair inviolate until the end.
5263:
5264: To go with this technique, you must define the machine-description macro
5265: @code{NOTICE_UPDATE_CC} to do @code{CC_STATUS_INIT}; in other words, no
5266: compare instruction is superfluous.
5267:
5268: Some machines have compare-and-branch instructions and no condition code.
5269: A similar technique works for them. When it is time to ``output'' a
5270: compare instruction, record its operands in two static variables. When
5271: outputting the branch-on-condition-code instruction that follows, actually
5272: output a compare-and-branch instruction that uses the remembered operands.
5273:
5274: It also works to define patterns for compare-and-branch instructions.
5275: In optimizing compilation, the pair of compare and branch instructions
5276: will be combined accoprding to these patterns. But this does not happen
5277: if optimization is not requested. So you must use one of the solutions
5278: above in addition to any special patterns you define.
5279:
5280: @node Peephole Definitions, Expander Definitions, Jump Patterns, Machine Desc
5281: @section Defining Machine-Specific Peephole Optimizers
5282:
5283: In addition to instruction patterns the @file{md} file may contain
5284: definitions of machine-specific peephole optimizations.
5285:
5286: The combiner does not notice certain peephole optimizations when the data
5287: flow in the program does not suggest that it should try them. For example,
5288: sometimes two consecutive insns related in purpose can be combined even
5289: though the second one does not appear to use a register computed in the
5290: first one. A machine-specific peephole optimizer can detect such
5291: opportunities.
5292:
5293: A definition looks like this:
5294:
5295: @example
5296: (define_peephole
5297: [@var{insn-pattern-1}
5298: @var{insn-pattern-2}
5299: @dots{}]
5300: "@var{condition}"
5301: "@var{template}")
5302: @end example
5303:
5304: In this skeleton, @var{insn-pattern-1} and so on are patterns to match
5305: consecutive instructions. The optimization applies to a sequence of
5306: instructions when @var{insn-pattern-1} matches the first one,
5307: @var{insn-pattern-2} matches the next, and so on.@refill
5308:
5309: @var{insn-pattern-1} and so on look @emph{almost} like the second operand
5310: of @code{define_insn}. There is one important difference: this pattern is
5311: an RTX, not a vector. If the @code{define_insn} pattern would be a vector
5312: of one element, the @var{insn-pattern} should be just that element, no
5313: vector. If the @code{define_insn} pattern would have multiple elements
5314: then the @var{insn-pattern} must place the vector inside an explicit
5315: @code{parallel} RTX.@refill
5316:
5317: The operands of the instructions are matched with @code{match_operands} and
5318: @code{match_dup}, as usual). What is not usual is that the operand numbers
5319: apply to all the instruction patterns in the definition. So, you can check
5320: for identical operands in two instructions by using @code{match_operand}
5321: in one instruction and @code{match_dup} in the other.
5322:
5323: The operand constraints used in @code{match_operand} patterns do not have
5324: any direct effect on the applicability of the optimization, but they will
5325: be validated afterward, so write constraints that are sure to fit whenever
5326: the optimization is applied. It is safe to use @code{"g"} for each
5327: operand.
5328:
5329: Once a sequence of instructions matches the patterns, the @var{condition}
5330: is checked. This is a C expression which makes the final decision whether
5331: to perform the optimization (do so if the expression is nonzero). If
5332: @var{condition} is omitted (in other words, the string is empty) then the
5333: optimization is applied to every sequence of instructions that matches the
5334: patterns.
5335:
5336: The defined peephole optimizations are applied after register allocation is
5337: complete. Therefore, the optimizer can check which operands have ended up
5338: in which kinds of registers, just by looking at the operands.
5339:
5340: The way to refer to the operands in @var{condition} is to write
5341: @code{operands[@var{i}]} for operand number @var{i} (as matched by
5342: @code{(match_operand @var{i} @dots{})}). Use the variable @code{insn} to
5343: refer to the last of the insns being matched; use @code{PREV_INSN} to find
5344: the preceding insns (but be careful to skip over any @samp{note} insns that
5345: intervene).@refill
5346:
5347: When optimizing computations with intermediate results, you can use
5348: @var{condition} to match only when the intermediate results are not used
5349: elsewhere. Use the C expression @code{dead_or_set_p (@var{insn},
5350: @var{op})}, where @var{insn} is the insn in which you expect the value to
5351: be used for the last time (from the value of @code{insn}, together with use
5352: of @code{PREV_INSN}), and @var{op} is the intermediate value (from
5353: @code{operands[@var{i}]}).@refill
5354:
5355: Applying the optimization means replacing the sequence of instructions with
5356: one new instruction. The @var{template} controls ultimate output of
5357: assembler code for this combined instruction. It works exactly like the
5358: template of a @code{define_insn}. Operand numbers in this template are the
5359: same ones used in matching the original sequence of instructions.
5360:
5361: The result of a defined peephole optimizer does not need to match any of
5362: the instruction patterns, and it does not have an opportunity to match
5363: them. The peephole optimizer definition itself serves as the instruction
5364: pattern to control how the instruction is output.
5365:
5366: Defined peephole optimizers are run in the last jump optimization pass, so
5367: the instructions they produce are never combined or rearranged
5368: automatically in any way.
5369:
5370: Here is an example, taken from the 68000 machine description:
5371:
5372: @example
5373: (define_peephole
5374: [(set (reg:SI 15) (plus:SI (reg:SI 15) (const_int 4)))
5375: (set (match_operand:DF 0 "register_operand" "f")
5376: (match_operand:DF 1 "register_operand" "ad"))]
5377: "FP_REG_P (operands[0]) && ! FP_REG_P (operands[1])"
5378: "*
5379: @{
5380: rtx xoperands[2];
5381: xoperands[1] = gen_rtx (REG, SImode, REGNO (operands[1]) + 1);
5382: #ifdef MOTOROLA
5383: output_asm_insn (\"move.l %1,(sp)\", xoperands);
5384: output_asm_insn (\"move.l %1,-(sp)\", operands);
5385: return \"fmove.d (sp)+,%0\";
5386: #else
5387: output_asm_insn (\"movel %1,sp@@\", xoperands);
5388: output_asm_insn (\"movel %1,sp@@-\", operands);
5389: return \"fmoved sp@@+,%0\";
5390: #endif
5391: @}
5392: ")
5393: @end example
5394:
5395: The effect of this optimization is to change
5396:
5397: @example
5398: jbsr _foobar
5399: addql #4,sp
5400: movel d1,sp@@-
5401: movel d0,sp@@-
5402: fmoved sp@@+,fp0
5403: @end example
5404:
5405: @noindent
5406: into
5407:
5408: @example
5409: jbsr _foobar
5410: movel d1,sp@@
5411: movel d0,sp@@-
5412: fmoved sp@@+,fp0
5413: @end example
5414:
5415: @node Expander Definitions,, Peephole Definitions, Machine Desc
5416: @section Defining RTL Sequences for Code Generation
5417:
5418: On some target machines, some standard pattern names for RTL generation
5419: cannot be handled with single insn, but a sequence of RTL insns can
5420: represent them. For these target machines, you can write a
5421: @samp{define_expand} to specify how to generate the sequence of RTL.
5422:
5423: A @samp{define_expand} is an RTL expression that looks almost like a
5424: @samp{define_insn}; but, unlike the latter, a @samp{define_expand} is used
5425: only for RTL generation and it can produce more than one RTL insn.
5426:
5427: A @samp{define_expand} RTX has four operands:
5428:
5429: @itemize @bullet
5430: @item
5431: The name. Each @samp{define_expand} must have a name, since the only
5432: use for it is to refer to it by name.
5433:
5434: @item
5435: The RTL template. This is just like the RTL template for a
5436: @samp{define_peephole} in that it is a vector of RTL expressions
5437: each being one insn.
5438:
5439: @item
5440: The condition, a string containing a C expression. This expression is
5441: used to express how the availability of this pattern depends on
5442: subclasses of target machine, selected by command-line options when
5443: GNU CC is run. This is just like the condition of a
5444: @samp{define_insn} that has a standard name.
5445:
5446: @item
5447: The preparation statements, a string containing zero or more C
5448: statements which are to be executed before RTL code is generated from
5449: the RTL template.
5450:
5451: Usually these statements prepare temporary registers for use as
5452: internal operands in the RTL template, but they can also generate RTL
5453: insns directly by calling routines such as @samp{emit_insn}, etc.
5454: Any such insns precede the ones that come from the RTL template.
5455: @end itemize
5456:
5457: The RTL template, in addition to controlling generation of RTL insns,
5458: also describes the operands that need to be specified when this pattern
5459: is used. In particular, it gives a predicate for each operand.
5460:
5461: A true operand, which need to be specified in order to generate RTL from
5462: the pattern, should be described with a @samp{match_operand} in its first
5463: occurrence in the RTL template. This enters information on the operand's
5464: predicate into the tables that record such things. GNU CC uses the
5465: information to preload the operand into a register if that is required for
5466: valid RTL code. If the operand is referred to more than once, subsequent
5467: references should use @samp{match_dup}.
5468:
5469: The RTL template may also refer to internal ``operands'' which are
5470: temporary registers or labels used only within the sequence made by the
5471: @samp{define_expand}. Internal operands are substituted into the RTL
5472: template with @samp{match_dup}, never with @samp{match_operand}. The
5473: values of the internal operands are not passed in as arguments by the
5474: compiler when it requests use of this pattern. Instead, they are computed
5475: within the pattern, in the preparation statements. These statements
5476: compute the values and store them into the appropriate elements of
5477: @code{operands} so that @samp{match_dup} can find them.
5478:
5479: There are two special macros defined for use in the preparation statements:
5480: @code{DONE} and @code{FAIL}. Use them with a following semicolon,
5481: as a statement.
5482:
5483: @table @code
5484: @item DONE
5485: Use the @code{DONE} macro to end RTL generation for the pattern. The
5486: only RTL insns resulting from the pattern on this occasion will be
5487: those already emitted by explicit calls to @code{emit_insn} within the
5488: preparation statements; the RTL template will not be generated.
5489:
5490: @item FAIL
5491: Make the pattern fail on this occasion. When a pattern fails, it means
5492: that the pattern was not truly available. The calling routines in the
5493: compiler will try other strategies for code generation using other patterns.
5494:
5495: Failure is currently supported only for binary operations (addition,
5496: multiplication, shifting, etc.).
5497:
5498: Do not emit any insns explicitly with @code{emit_insn} before failing.
5499: @end table
5500:
5501: Here is an example, the definition of left-shift for the SPUR chip:
5502:
5503: @example
5504: (define_expand "ashlsi3"
5505: [(set (match_operand:SI 0 "register_operand" "")
5506: (ashift:SI
5507: (match_operand:SI 1 "register_operand" "")
5508: (match_operand:SI 2 "nonmemory_operand" "")))]
5509: ""
5510: "
5511: @{
5512: if (GET_CODE (operands[2]) != CONST_INT
5513: || (unsigned) INTVAL (operands[2]) > 3)
5514: FAIL;
5515: @}")
5516: @end example
5517:
5518: @noindent
5519: This example uses @samp{define_expand} so that it can generate an RTL insn
5520: for shifting when the shift-count is in the supported range of 0 to 3 but
5521: fail in other cases where machine insns aren't available. When it fails,
5522: the compiler tries another strategy using different patterns (such as, a
5523: library call).
5524:
5525: If the compiler were able to handle nontrivial condition-strings in
5526: patterns with names, then there would be possible to use a
5527: @samp{define_insn} in that case. Here is another case (zero-extension on
5528: the 68000) which makes more use of the power of @samp{define_expand}:
5529:
5530: @example
5531: (define_expand "zero_extendhisi2"
5532: [(set (match_operand:SI 0 "general_operand" "")
5533: (const_int 0))
5534: (set (strict_low_part
5535: (subreg:HI
5536: (match_operand:SI 0 "general_operand" "")
5537: 0))
5538: (match_operand:HI 1 "general_operand" ""))]
5539: ""
5540: "operands[1] = make_safe_from (operands[1], operands[0]);")
5541: @end example
5542:
5543: @noindent
5544: Here two RTL insns are generated, one to clear the entire output operand
5545: and the other to copy the input operand into its low half. This sequence
5546: is incorrect if the input operand refers to [the old value of] the output
5547: operand, so the preparation statement makes sure this isn't so. The
5548: function @code{make_safe_from} copies the @code{operands[1]} into a
5549: temporary register if it refers to @code{operands[0]}. It does this
5550: by emitting another RTL insn.
5551:
5552: Finally, a third example shows the use of an internal operand.
5553: Zero-extension on the SPUR chip is done by @samp{and}-ing the result
5554: against a halfword mask. But this mask cannot be represented by a
5555: @samp{const_int} because the constant value is too large to be legitimate
5556: on this machine. So it must be copied into a register with
5557: @code{force_reg} and then the register used in the @samp{and}.
5558:
5559: @example
5560: (define_expand "zero_extendhisi2"
5561: [(set (match_operand:SI 0 "register_operand" "")
5562: (and:SI (subreg:SI
5563: (match_operand:HI 1 "register_operand" "")
5564: 0)
5565: (match_dup 2)))]
5566: ""
5567: "operands[2]
5568: = force_reg (SImode, gen_rtx (CONST_INT,
5569: VOIDmode, 65535)); ")
5570: @end example
5571:
5572: @node Machine Macros, Config, Machine Desc, Top
1.1 root 5573: @chapter Machine Description Macros
5574:
5575: The other half of the machine description is a C header file conventionally
5576: given the name @file{tm-@var{machine}.h}. The file @file{tm.h} should be a
5577: link to it. The header file @file{config.h} includes @file{tm.h} and most
5578: compiler source files include @file{config.h}.
5579:
5580: @menu
1.1.1.2 root 5581: * Run-time Target:: Defining -m options like -m68000 and -m68020.
1.1 root 5582: * Storage Layout:: Defining sizes and alignments of data types.
5583: * Registers:: Naming and describing the hardware registers.
5584: * Register Classes:: Defining the classes of hardware registers.
5585: * Stack Layout:: Defining which way the stack grows and by how much.
1.1.1.2 root 5586: * Library Names:: Specifying names of subroutines to call automatically.
1.1 root 5587: * Addressing Modes:: Defining addressing modes valid for memory operands.
5588: * Condition Code:: Defining how insns update the condition code.
5589: * Assembler Format:: Defining how to write insns and pseudo-ops to output.
5590: * Misc:: Everything else.
5591: @end menu
5592:
5593: @node Run-time Target, Storage Layout, Machine Macros, Machine Macros
5594: @section Run-time Target Specification
5595:
5596: @table @code
5597: @item CPP_PREDEFINES
1.1.1.2 root 5598: Define this to be a string constant containing @samp{-D} options
1.1 root 5599: to define the predefined macros that identify this machine and system.
5600:
5601: For example, on the Sun, one can use the value
5602:
5603: @example
1.1.1.2 root 5604: "-Dmc68000 -Dsun -Dunix"
1.1 root 5605: @end example
5606:
5607: @item extern int target_flags;
5608: This declaration should be present.
5609:
5610: @item TARGET_@dots{}
5611: This series of macros is to allow compiler command arguments to
5612: enable or disable the use of optional features of the target machine.
5613: For example, one machine description serves both the 68000 and
5614: the 68020; a command argument tells the compiler whether it should
5615: use 68020-only instructions or not. This command argument works
5616: by means of a macro @code{TARGET_68020} that tests a bit in
5617: @code{target_flags}.
5618:
5619: Define a macro @code{TARGET_@var{featurename}} for each such option.
5620: Its definition should test a bit in @code{target_flags}; for example:
5621:
5622: @example
5623: #define TARGET_68020 (target_flags & 1)
5624: @end example
5625:
5626: One place where these macros are used is in the condition-expressions
5627: of instruction patterns. Note how @code{TARGET_68020} appears
1.1.1.2 root 5628: frequently in the 68000 machine description file, @file{m68k.md}.
1.1 root 5629: Another place they are used is in the definitions of the other
5630: macros in the @file{tm-@var{machine}.h} file.
5631:
5632: @item TARGET_SWITCHES
1.1.1.2 root 5633: This macro defines names of command options to set and clear
1.1 root 5634: bits in @code{target_flags}. Its definition is an initializer
1.1.1.2 root 5635: with a subgrouping for each command option.
1.1 root 5636:
1.1.1.2 root 5637: Each subgrouping contains a string constant, that defines the option
1.1 root 5638: name, and a number, which contains the bits to set in
5639: @code{target_flags}. A negative number says to clear bits instead;
1.1.1.2 root 5640: the negative of the number is which bits to clear. The actual option
1.1 root 5641: name is made by appending @samp{-m} to the specified name.
5642:
5643: One of the subgroupings should have a null string. The number in
5644: this grouping is the default value for @code{target_flags}. Any
1.1.1.2 root 5645: target options act starting with that value.
1.1 root 5646:
5647: Here is an example which defines @samp{-m68000} and @samp{-m68020}
5648: with opposite meanings, and picks the latter as the default:
5649:
5650: @example
5651: #define TARGET_SWITCHES \
5652: @{ @{ "68020", 1@}, \
5653: @{ "68000", -1@}, \
5654: @{ "", 1@}@}
5655: @end example
5656: @end table
5657:
1.1.1.2 root 5658: Sometimes certain combinations of command options do not make sense on a
5659: particular target machine. You can define a macro @code{OVERRIDE_OPTIONS}
5660: to take account of this. This macro, if defined, is executed once
5661: just after all the command options have been parsed.
5662:
1.1 root 5663: @node Storage Layout, Registers, Run-time Target, Machine Macros
5664: @section Storage Layout
5665:
1.1.1.2 root 5666: Note that the definitions of the macros in this table which are sizes or
5667: alignments measured in bits do not need to be constant. They can be C
5668: expressions that refer to static variables, such as the @code{target_flags}.
5669: @xref{Run-time Target}.
5670:
1.1 root 5671: @table @code
5672: @item BITS_BIG_ENDIAN
5673: Define this macro if the most significant bit in a byte has the lowest
5674: number. This means that bit-field instructions count from the most
5675: significant bit. If the machine has no bit-field instructions, this
5676: macro is irrelevant.
5677:
5678: @item BYTES_BIG_ENDIAN
5679: Define this macro if the most significant byte in a word has the
5680: lowest number.
5681:
5682: @item WORDS_BIG_ENDIAN
1.1.1.2 root 5683: Define this macro if, in a multiword object, the most significant
1.1 root 5684: word has the lowest number.
5685:
5686: @item BITS_PER_UNIT
5687: Number of bits in an addressable storage unit (byte); normally 8.
5688:
5689: @item BITS_PER_WORD
5690: Number of bits in a word; normally 32.
5691:
5692: @item UNITS_PER_WORD
5693: Number of storage units in a word; normally 4.
5694:
5695: @item POINTER_SIZE
5696: Width of a pointer, in bits.
5697:
5698: @item PARM_BOUNDARY
1.1.1.2 root 5699: Alignment required for function parameters on the stack, in bits.
5700:
5701: @item STACK_BOUNDARY
5702: Define this macro if you wish to preserve a certain alignment for
5703: the stack pointer at all times. The definition is a C expression
5704: for the desired alignment (measured in bits).
1.1 root 5705:
5706: @item FUNCTION_BOUNDARY
5707: Alignment required for a function entry point, in bits.
5708:
5709: @item BIGGEST_ALIGNMENT
1.1.1.2 root 5710: Biggest alignment that any data type can require on this machine, in bits.
5711:
5712: @item EMPTY_FIELD_ALIGNMENT
5713: Alignment in bits to be given to a structure bit field that follows an
5714: empty field such as @code{int : 0;}.
5715:
5716: @item STRUCTURE_SIZE_BOUNDARY
5717: Number of bits which any structure or union's size must be a multiple of.
5718: Each structure or union's size is rounded up to a multiple of this.
5719:
5720: If you do not define this macro, the default is the same as
5721: @code{BITS_PER_UNIT}.
1.1 root 5722:
5723: @item STRICT_ALIGNMENT
5724: Define this if instructions will fail to work if given data not
5725: on the nominal alignment. If instructions will merely go slower
5726: in that case, do not define this macro.
5727: @end table
5728:
5729: @node Registers, Register Classes, Storage Layout, Machine Macros
5730: @section Register Usage
5731:
5732: @table @code
5733: @item FIRST_PSEUDO_REGISTER
5734: Number of hardware registers known to the compiler. They receive
5735: numbers 0 through @code{FIRST_PSEUDO_REGISTER-1}; thus, the first
1.1.1.2 root 5736: pseudo register's number really is assigned the number
1.1 root 5737: @code{FIRST_PSEUDO_REGISTER}.
5738:
5739: @item FIXED_REGISTERS
5740: An initializer that says which registers are used for fixed purposes
5741: all throughout the compiled code and are therefore not available for
1.1.1.2 root 5742: general allocation. These would include the stack pointer, the frame
1.1 root 5743: pointer, the program counter on machines where that is considered one
5744: of the addressable registers, and any other numbered register with a
5745: standard use.
5746:
5747: This information is expressed as a sequence of numbers, separated by
5748: commas and surrounded by braces. The @var{n}th number is 1 if
1.1.1.2 root 5749: register @var{n} is fixed, 0 otherwise.
5750:
5751: The table initialized from this macro, and the table initialized by
5752: the following one, may be overridden at run time either automatically,
5753: by the actions of the macro @code{CONDITIONAL_REGISTER_USAGE}, or by
5754: the user with the command options @samp{-ffixed-@var{reg}},
5755: @samp{-fcall-used-@var{reg}} and @samp{-fcall-saved-@var{reg}}.
1.1 root 5756:
5757: @item CALL_USED_REGISTERS
5758: Like @code{FIXED_REGISTERS} but has 1 for each register that is
5759: clobbered (in general) by function calls as well as for fixed
5760: registers. This macro therefore identifies the registers that are not
5761: available for general allocation of values that must live across
5762: function calls.
5763:
1.1.1.2 root 5764: If a register has 0 in @code{CALL_USED_REGISTERS}, the compiler
1.1 root 5765: automatically saves it on function entry and restores it on function
5766: exit, if the register is used within the function.
5767:
1.1.1.2 root 5768: @item CONDITIONAL_REGISTER_USAGE
5769: Zero or more C statements that may conditionally modify two variables
5770: @code{fixed_regs} and @code{call_used_regs} (both of type @code{char
5771: []}) after they have been initialized from the two preceding macros.
5772:
5773: This is necessary in case the fixed or call-clobbered registers depend
5774: on target flags.
5775:
5776: You need not define this macro if it has no work to do.
5777:
1.1 root 5778: @item HARD_REGNO_REGS (@var{regno}, @var{mode})
5779: A C expression for the number of consecutive hard registers, starting
5780: at register number @var{regno}, required to hold a value of mode
5781: @var{mode}.
5782:
5783: On a machine where all registers are exactly one word, a suitable
5784: definition of this macro is
5785:
5786: @example
5787: #define HARD_REGNO_NREGS(REGNO, MODE) \
5788: ((GET_MODE_SIZE (MODE) + UNITS_PER_WORD - 1) \
5789: / UNITS_PER_WORD))
5790: @end example
5791:
5792: @item HARD_REGNO_MODE_OK (@var{regno}, @var{mode})
5793: A C expression that is nonzero if it is permissible to store a value
5794: of mode @var{mode} in hard register number @var{regno} (or in several
5795: registers starting with that one). For a machine where all registers
5796: are equivalent, a suitable definition is
5797:
5798: @example
5799: #define HARD_REGNO_MODE_OK(REGNO, MODE) 1
5800: @end example
5801:
5802: It is not necessary for this macro to check for fixed register numbers
5803: because the allocation mechanism considers them to be always occupied.
5804:
1.1.1.2 root 5805: Many machines have special registers for floating point arithmetic.
5806: Often people assume that floating point machine modes are allowed only
5807: in floating point registers. This is not true. Any registers that
5808: can hold integers can safely @emph{hold} a floating point machine
5809: mode, whether or not floating arithmetic can be done on it in those
5810: registers.
5811:
5812: The true significance of special floating registers is rather than
5813: non-floating-point machine modes @emph{may not} go in those registers.
5814: This is true if the floating registers normalize any value stored in
5815: them, because storing a non-floating value there would garble it. If
5816: the floating registers do not automatically normalize, if you can
5817: store any bit pattern in one and retrieve it unchanged without a trap,
5818: then any machine mode may go in a floating register and this macro
5819: should say so.
5820:
5821: Sometimes there are floating registers that are especially slow to
5822: access, so that it is better to store a value in a stack frame than in
5823: such a register if floating point arithmetic is not being done. As long
5824: as the floating registers are not in class @code{GENERAL_REGS}, they
5825: will not be used unless some insn's constraint asks for one.
5826:
5827: It is obligatory to support floating point `move' instructions into
5828: and out of general registers, because unions and structures (which
5829: have modes @samp{SImode} or @samp{DImode}) can be in those registers
5830: and they may have floating point members.
5831:
1.1 root 5832: @item MODES_TIEABLE_P (@var{mode1}, @var{mode2})
5833: A C expression that is nonzero if it is desirable to choose register
5834: allocation so as to avoid move instructions between a value of mode
5835: @var{mode1} and a value of mode @var{mode2}.
5836:
5837: If @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode1})} and
5838: @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode2})} are ever different
5839: for any @var{r}, then @code{MODES_TIEABLE_P (@var{mode1},
5840: @var{mode2})} must be zero.
5841:
5842: @item PC_REGNUM
5843: If the program counter has a register number, define this as that
5844: register number. Otherwise, do not define it.
5845:
5846: @item STACK_POINTER_REGNUM
5847: The register number of the stack pointer register, which must also be
5848: a fixed register according to @code{FIXED_REGISTERS}. On many
5849: machines, the hardware determines which register this is.
5850:
5851: @item FRAME_POINTER_REGNUM
5852: The register number of the frame pointer register, which is used to
1.1.1.2 root 5853: access automatic variables in the stack frame. On some machines, the
1.1 root 5854: hardware determines which register this is. On other machines, you
5855: can choose any register you wish for this purpose.
5856:
1.1.1.2 root 5857: @item FRAME_POINTER_REQUIRED
5858: A C expression which is nonzero if a function must have and use a
5859: frame pointer. This expression is evaluated in the reload pass, in
5860: the function @code{reload}, and it can in principle examine the
5861: current function and decide according to the facts, but on most
5862: machines the constant 0 or the constant 1 suffices. Use 0 when the
5863: machine allows code to be generated with no frame pointer, and doing
5864: so saves some time or space. Use 1 when there is no possible
5865: advantage to avoiding a frame pointer.
5866:
5867: In certain cases, the compiler does not know how to do without a frame
5868: pointer. The compiler recognizes those cases and automatically gives
5869: the function a frame pointer regardless of what
5870: @code{FRAME_POINTER_REQUIRED} says. You don't need to worry about
5871: them.@refill
5872:
5873: In a function that does not require a frame pointer, the frame pointer
5874: register can be allocated for ordinary usage, provided it is not
5875: marked as a fixed register. See @code{FIXED_REGISTERS} for more
5876: information.
5877:
1.1 root 5878: @item ARG_POINTER_REGNUM
5879: The register number of the arg pointer register, which is used to
5880: access the function's argument list. On some machines, this is the
5881: same as the frame pointer register. On some machines, the hardware
5882: determines which register this is. On other machines, you can choose
5883: any register you wish for this purpose. It must in any case be a
5884: fixed register according to @code{FIXED_REGISTERS}.
5885:
5886: @item STATIC_CHAIN_REGNUM
5887: The register number used for passing a function's static chain
5888: pointer. This is needed for languages such as Pascal and Algol where
5889: functions defined within other functions can access the local
5890: variables of the outer functions; it is not currently used because C
5891: does not provide this feature.
5892:
5893: The static chain register need not be a fixed register.
5894:
5895: @item STRUCT_VALUE_REGNUM
1.1.1.2 root 5896: When a function's value's mode is @code{BLKmode}, the value is not
5897: returned according to @code{FUNCTION_VALUE}. Instead, the caller
5898: passes the address of a block of memory in which the value should be
5899: stored. @code{STRUCT_VALUE_REGNUM} is the register in which this
5900: address is passed.
1.1 root 5901: @end table
5902:
5903: @node Register Classes, Stack Layout, Registers, Machine Macros
5904: @section Register Classes
5905:
5906: On many machines, the numbered registers are not all equivalent.
5907: For example, certain registers may not be allowed for indexed addressing;
5908: certain registers may not be allowed in some instructions. These machine
5909: restrictions are described to the compiler using @dfn{register classes}.
5910:
5911: You define a number of register classes, giving each one a name and saying
5912: which of the registers belong to it. Then you can specify register classes
5913: that are allowed as operands to particular instruction patterns.
5914:
5915: In general, each register will belong to several classes. In fact, one
5916: class must be named @code{ALL_REGS} and contain all the registers. Another
5917: class must be named @code{NO_REGS} and contain no registers. Often the
5918: union of two classes will be another class; however, this is not required.
5919:
5920: One of the classes must be named @code{GENERAL_REGS}. There is nothing
5921: terribly special about the name, but the operand constraint letters
5922: @samp{r} and @samp{g} specify this class. If @code{GENERAL_REGS} is
5923: the same as @code{ALL_REGS}, just define it as a macro which expands
5924: to @code{ALL_REGS}.
5925:
5926: The way classes other than @code{GENERAL_REGS} are specified in operand
5927: constraints is through machine-dependent operand constraint letters.
5928: You can define such letters to correspond to various classes, then use
5929: them in operand constraints.
5930:
1.1.1.2 root 5931: You should define a class for the union of two classes whenever some
5932: instruction allows both classes. For example, if an instruction allows
5933: either a floating-point (coprocessor) register or a general register for a
5934: certain operand, you should define a class @code{FLOAT_OR_GENERAL_REGS}
5935: which includes both of them. Otherwise you will get suboptimal code.
5936:
1.1 root 5937: You must also specify certain redundant information about the register
5938: classes: for each class, which classes contain it and which ones are
5939: contained in it; for each pair of classes, the largest class contained
5940: in their union.
5941:
5942: @table @code
5943: @item enum reg_class
5944: An enumeral type that must be defined with all the register class names
5945: as enumeral values. @code{NO_REGS} must be first. @code{ALL_REGS}
5946: must be the last register class, followed by one more enumeral value,
5947: @code{LIM_REG_CLASSES}, which is not a register class but rather
5948: tells how many classes there are.
5949:
5950: Each register class has a number, which is the value of casting
5951: the class name to type @code{int}. The number serves as an index
5952: in many of the tables described below.
5953:
5954: @item REG_CLASS_NAMES
5955: An initializer containing the names of the register classes as C string
5956: constants. These names are used in writing some of the debugging dumps.
5957:
5958: @item REG_CLASS_CONTENTS
5959: An initializer containing the contents of the register classes, as integers
5960: which are bit masks. The @var{n}th integer specifies the contents of class
5961: @var{n}. The way the integer @var{mask} is interpreted is that
5962: register @var{r} is in the class if @code{@var{mask} & (1 << @var{r})} is 1.
5963:
5964: When the machine has more than 32 registers, an integer does not suffice.
5965: Then the integers are replaced by sub-initializers, braced groupings containing
5966: several integers. Each sub-initializer must be suitable as an initializer
5967: for the type @code{HARD_REG_SET} which is defined in @file{hard-reg-set.h}.
5968:
5969: @item REGNO_REG_CLASS (@var{regno})
5970: A C expression whose value is a register class containing hard register
5971: @var{regno}. In general there is more that one such class; choose a class
5972: which is @dfn{minimal}, meaning that no smaller class also contains the
5973: register.
5974:
5975: @item INDEX_REG_CLASS
5976: A macro whose definition is the name of the class to which a valid index
5977: register must belong.
5978:
5979: @item REG_CLASS_FROM_LETTER (@var{char})
5980: A C expression which defines the machine-dependent operand constraint
5981: letters for register classes. If @var{char} is such a letter, the value
5982: should be the register class corresponding to it. Otherwise, the value
5983: should be @code{NO_REGS}.
5984:
1.1.1.2 root 5985: @item REGNO_OK_FOR_BASE_P (@var{num})
5986: A C expression which is nonzero if register number @var{num} is
5987: suitable for use as a base register in operand addresses. It may be
5988: either a suitable hard register or a pseudo register that has been
5989: allocated such a hard register.
5990:
5991: @item REGNO_OK_FOR_INDEX_P (@var{num})
5992: A C expression which is nonzero if register number @var{num} is
5993: suitable for use as an index register in operand addresses. It may be
5994: either a suitable hard register or a pseudo register that has been
5995: allocated such a hard register.
5996:
5997: The difference between an index register and a base register is that
5998: the index register may be scaled. If an address involves the sum of
5999: two registers, neither one of them scaled, then either one may be
6000: labeled the ``base'' and the other the ``index''; but whichever
6001: labeling is used must fit the machine's constraints of which registers
6002: may serve in each capacity. The compiler will try both labelings,
6003: looking for one that is valid, and reload one or both registers only
6004: if neither labeling works.
1.1 root 6005:
6006: @item PREFERRED_RELOAD_CLASS (@var{x}, @var{class})
6007: A C expression that places additional restrictions on the register class
6008: to use when it is necessary to copy value @var{x} into a register in class
6009: @var{class}. The value is a register class; perhaps @var{class}, or perhaps
6010: another, smaller class. @var{class} is always safe as a value. In fact,
6011: the definition
6012:
6013: @example
6014: #define PREFERRED_RELOAD_CLASS(X,CLASS) CLASS
6015: @end example
6016:
6017: @noindent
6018: is always safe. However, sometimes returning a more restrictive class
6019: makes better code. For example, on the 68000, when @var{x} is an
6020: integer constant that is in range for a @samp{moveq} instruction,
6021: the value of this macro is always @code{DATA_REGS} as long as
6022: @var{class} includes the data registers. Requiring a data register
6023: guarantees that a @samp{moveq} will be used.
1.1.1.2 root 6024:
6025: @item CLASS_MAX_NREGS (@var{class}, @var{mode})
6026: A C expression for the maximum number of consecutive registers
6027: of class @var{class} needed to hold a value of mode @var{mode}.
6028:
6029: This is closely related to the macro @code{HARD_REGNO_NREGS}.
6030: In fact, the value of the macro @code{CLASS_MAX_NREGS (@var{class}, @var{mode})}
6031: should be the maximum value of @code{HARD_REGNO_NREGS (@var{regno}, @var{mode})}
6032: for all @var{regno} values in the class @var{class}.
6033:
6034: This macro helps control the handling of multiple-word values
6035: in the reload pass.
1.1 root 6036: @end table
6037:
1.1.1.2 root 6038: Two other special macros describe which constants fit which constraint
6039: letters.
1.1 root 6040:
6041: @table @code
6042: @item CONST_OK_FOR_LETTER_P (@var{value}, @var{c})
6043: A C expression that defines the machine-dependent operand constraint letters
6044: that specify particular ranges of integer values. If @var{c} is one
6045: of those letters, the expression should check that @var{value}, an integer,
6046: is in the appropriate range and return 1 if so, 0 otherwise. If @var{c} is
6047: not one of those letters, the value should be 0 regardless of @var{value}.
6048:
6049: @item CONST_DOUBLE_OK_FOR_LETTER_P (@var{value}, @var{c})
6050: A C expression that defines the machine-dependent operand constraint
6051: letters that specify particular ranges of floating values. If @var{c} is
1.1.1.2 root 6052: one of those letters, the expression should check that @var{value}, an RTX
1.1 root 6053: of code @samp{const_double}, is in the appropriate range and return 1 if
6054: so, 0 otherwise. If @var{c} is not one of those letters, the value should
6055: be 0 regardless of @var{value}.
6056: @end table
6057:
1.1.1.2 root 6058: @node Stack Layout, Library Names, Register Classes, Machine Macros
1.1 root 6059: @section Describing Stack Layout
6060:
6061: @table @code
6062: @item STACK_GROWS_DOWNWARD
6063: Define this macro if pushing a word onto the stack moves the stack
1.1.1.2 root 6064: pointer to a smaller address.
6065:
6066: When we say, ``define this macro if @dots{},'' it means that the
6067: compiler checks this macro only with @code{#ifdef} so the precise
6068: definition used does not matter.
1.1 root 6069:
6070: @item FRAME_GROWS_DOWNWARD
6071: Define this macro if the addresses of local variable slots are at negative
6072: offsets from the frame pointer.
6073:
6074: @item STARTING_FRAME_OFFSET
6075: Offset from the frame pointer to the first local variable slot to be allocated.
6076:
6077: If @code{FRAME_GROWS_DOWNWARD}, the next slot's offset is found by
6078: subtracting the length of the first slot from @code{STARTING_FRAME_OFFSET}.
6079: Otherwise, it is found by adding the length of the first slot to
6080: the value @code{STARTING_FRAME_OFFSET}.
6081:
6082: @item PUSH_ROUNDING (@var{npushed})
6083: A C expression that is the number of bytes actually pushed onto the
6084: stack when an instruction attempts to push @var{npushed} bytes.
6085:
1.1.1.2 root 6086: If the target machine does not have a push instruction, do not define
6087: this macro. That directs GNU CC to use an alternate strategy: to
6088: allocate the entire argument block and then store the arguments into
6089: it.
6090:
1.1 root 6091: On some machines, the definition
6092:
6093: @example
6094: #define PUSH_ROUNDING(BYTES) (BYTES)
6095: @end example
6096:
6097: @noindent
6098: will suffice. But on other machines, instructions that appear
6099: to push one byte actually push two bytes in an attempt to maintain
6100: alignment. Then the definition should be
6101:
6102: @example
6103: #define PUSH_ROUNDING(BYTES) (((BYTES) + 1) & ~1)
6104: @end example
6105:
6106: @item FIRST_PARM_OFFSET
6107: Offset from the argument pointer register to the first argument's address.
6108:
1.1.1.2 root 6109: @item RETURN_POPS_ARGS (@var{funtype})
6110: A C expression that should be 1 if a function pops its own arguments
6111: on returning, or 0 if the function pops no arguments and the caller
6112: must therefore pop them all after the function returns.
6113:
6114: @var{funtype} is a C variable whose value is a tree node that
6115: describes the function in question. Normally it is a node of type
6116: @code{FUNCTION_TYPE} that describes the data type of the function.
6117: From this it is possible to obtain the data types of the value and
6118: arguments (if known).
6119:
6120: When a call to a library function is being considered, @var{funtype}
6121: will contain an identifier node for the library function. Thus, if
6122: you need to distinguish among various library functions, you can do so
6123: by their names. Note that ``library function'' in this context means
6124: a function used to perform arithmetic, whose name is known specially
6125: in the compiler and was not mentioned in the C code being compiled.
6126:
6127: On the Vax, all functions always pop their arguments, so the
6128: definition of this macro is 1. On the 68000, using the standard
6129: calling convention, no functions pop their arguments, so the value of
6130: the macro is always 0 in this case. But an alternative calling
6131: convention is available in which functions that take a fixed number of
6132: arguments pop them but other functions (such as @code{printf}) pop
6133: nothing (the caller pops all). When this convention is in use,
6134: @var{funtype} is examined to determine whether a function takes a
6135: fixed number of arguments.
6136:
6137: @item FUNCTION_VALUE (@var{valtype}, @var{func})
6138: A C expression to create an RTX representing the place where a
6139: function returns a value of data type @var{valtype}. @var{valtype} is
6140: a tree node representing a data type. Write @code{TYPE_MODE
6141: (@var{valtype})} to get the machine mode used to represent that type.
6142: On many machines, only the mode is relevant. (Actually, on most
6143: machines, scalar values are returned in the same place regardless of
6144: mode).@refill
6145:
6146: If the precise function being called is known, @var{func} is a tree
6147: node (@code{FUNCTION_DECL}) for it; otherwise, @var{func} is a null
6148: pointer. This makes it possible to use a different value-returning
6149: convention for specific functions when all their calls are
6150: known.@refill
6151:
6152: @item FUNCTION_OUTGOING_VALUE (@var{valtype}, @var{func})
6153: Define this macro if the target machine has ``register windows''
6154: so that the register in which a function returns its value is not
6155: the same as the one in which the caller sees the value.
6156:
6157: For such machines, @code{FUNCTION_VALUE} computes the register in
6158: which the caller will see the value, and
6159: @code{FUNCTION_OUTGOING_VALUE} should be defined in a similar fashion
6160: to tell the function where to put the value.@refill
6161:
6162: If @code{FUNCTION_OUTGOING_VALUE} is not defined,
6163: @code{FUNCTION_VALUE} serves both purposes.@refill
6164:
6165: @item LIBCALL_VALUE (@var{mode})
6166: A C expression to create an RTX representing the place where a library
6167: function returns a value of mode @var{mode}. If the precise function
6168: being called is known, @var{func} is a tree node
6169: (@code{FUNCTION_DECL}) for it; otherwise, @var{func} is a null
6170: pointer. This makes it possible to use a different value-returning
6171: convention for specific functions when all their calls are
6172: known.@refill
6173:
6174: Note that ``library function'' in this context means a compiler
6175: support routine, used to perform arithmetic, whose name is known
6176: specially by the compiler and was not mentioned in the C code being
6177: compiled.
6178:
6179: @item FUNCTION_VALUE_REGNO_P (@var{regno})
6180: A C expression that is nonzero if @var{regno} is the number of a hard
6181: register in which function values are sometimes returned.
6182:
6183: A register whose use for returning values is limited to serving as the
6184: second of a pair (for a value of type @code{double}, say) need not be
6185: recognized by this macro. So for most machines, this definition
6186: suffices:
6187:
6188: @example
6189: #define FUNCTION_VALUE_REGNO_P(N) ((N) == 0)
6190: @end example
6191:
6192: @item FUNCTION_ARG (@var{cum}, @var{mode}, @var{type}, @var{named})
6193: A C expression that controls whether a function argument is passed
6194: in a register, and which register.
6195:
6196: The arguments are @var{cum}, which summarizes all the previous
6197: arguments; @var{mode}, the machine mode of the argument; @var{type},
6198: the data type of the argument as a tree node or 0 if that is not known
6199: (which happens for C support library functions); and @var{named},
6200: which is 1 for an ordinary argument and 0 for nameless arguments that
6201: correspond to @samp{...} in the called function's prototype.
6202:
6203: The value of the expression should either be a @samp{reg} RTX for the
6204: hard register in which to pass the argument, or zero to pass the
6205: argument on the stack.
6206:
6207: For the Vax and 68000, where normally all arguments are pushed, zero
6208: suffices as a definition.
6209:
6210: @item FUNCTION_INCOMING_ARG (@var{cum}, @var{mode}, @var{type}, @var{named})
6211: Define this macro if the target machine has ``register windows'', so
6212: that the register in which a function sees an arguments is not
6213: necessarily the same as the one in which the caller passed the
6214: argument.
6215:
6216: For such machines, @code{FUNCTION_ARG} computes the register in which
6217: the caller passes the value, and @code{FUNCTION_INCOMING_ARG} should
6218: be defined in a similar fashion to tell the function being called
6219: where the arguments will arrive.
6220:
6221: If @code{FUNCTION_INCOMING_ARG} is not defined, @code{FUNCTION_ARG}
6222: serves both purposes.@refill
6223:
6224: @item FUNCTION_ARG_PARTIAL_NREGS (@var{cum}, @var{mode}, @var{type}, @var{named})
6225: A C expression for the number of words, at the beginning of an
6226: argument, must be put in registers. The value must be zero for
6227: arguments that are passed entirely in registers or that are entirely
6228: pushed on the stack.
6229:
6230: On some machines, certain arguments must be passed partially in
6231: registers and partially in memory. On these machines, typically the
6232: first @var{n} words of arguments are passed in registers, and the rest
6233: on the stack. If a multi-word argument (a @code{double} or a
6234: structure) crosses that boundary, its first few words must be passed
6235: in registers and the rest must be pushed. This macro tells the
6236: compiler when this occurs, and how many of the words should go in
6237: registers.
6238:
6239: @code{FUNCTION_ARG} for these arguments should return the first
6240: register to be used by the caller for this argument; likewise
6241: @code{FUNCTION_INCOMING_ARG}, for the called function.
6242:
6243: @item CUMULATIVE_ARGS
6244: A C type for declaring a variable that is used as the first argument
6245: of @code{FUNCTION_ARG} and other related values. For some target
6246: machines, the type @code{int} suffices and can hold the number of
6247: bytes of argument so far.
6248:
6249: @item INIT_CUMULATIVE_ARGS (@var{cum})
6250: A C statement (sans semicolon) for initializing the variable @var{cum}
6251: for the state at the beginning of the argument list. The variable has
6252: type @code{CUMULATIVE_ARGS}.
6253:
6254: @item FUNCTION_ARG_ADVANCE (@var{cum}, @var{mode}, @var{type}, @var{named})
6255: Update the summarizer variable @var{cum} to advance past an argument
6256: in the argument list. The values @var{mode}, @var{type} and
6257: @var{named} describe that argument. Once this is done, the variable
6258: @var{cum} is suitable for analyzing the @emph{following} argument
6259: with @code{FUNCTION_ARG}, etc.@refill
6260:
6261: @item FUNCTION_ARG_REGNO_P (@var{regno})
6262: A C expression that is nonzero if @var{regno} is the number of a hard
6263: register in which function arguments are sometimes passed. This does
6264: @emph{not} include implicit arguments such as the static chain and
6265: the structure-value address. On many machines, no registers can be
6266: used for this purpose since all function arguments are pushed on the
6267: stack.
1.1 root 6268:
6269: @item FUNCTION_PROLOGUE (@var{file}, @var{size})
6270: A C compound statement that outputs the assembler code for entry to a
6271: function. The prologue is responsible for setting up the stack frame,
6272: initializing the frame pointer register, saving registers that must be
1.1.1.2 root 6273: saved, and allocating @var{size} additional bytes of storage for the
6274: local variables. @var{size} is an integer. @var{file} is a stdio
6275: stream to which the assembler code should be output.
1.1 root 6276:
6277: The label for the beginning of the function need not be output by this
6278: macro. That has already been done when the macro is run.
6279:
6280: To determine which registers to save, the macro can refer to the array
1.1.1.2 root 6281: @code{regs_ever_live}: element @var{r} is nonzero if hard register
6282: @var{r} is used anywhere within the function. This implies the
6283: function prologue should save register @var{r}, but not if it is one
6284: of the call-used registers.
6285:
6286: On machines where functions may or may not have frame-pointers, the
6287: function entry code must vary accordingly; it must set up the frame
6288: pointer if one is wanted, and not otherwise. To determine whether a
6289: frame pointer is in wanted, the macro can refer to the variable
6290: @code{frame_pointer_needed}. The variable's value will be 1 at run
6291: time in a function that needs a frame pointer.
6292:
6293: @item FUNCTION_PROFILER (@var{file}, @var{labelno})
6294: A C statement or compound statement to output to @var{file} some
6295: assembler code to call the profiling subroutine @code{mcount}.
6296: Before calling, the assembler code must load the address of a
6297: counter variable into a register where @code{mcount} expects to
6298: find the address. The name of this variable is @samp{LP} followed
6299: by the number @var{labelno}, so you would generate the name using
6300: @samp{LP%d} in a @code{fprintf}.
6301:
6302: The details of how the address should be passed to @code{mcount} are
6303: determined by your operating system environment, not by GNU CC. To
6304: figure them out, compile a small program for profiling using the
6305: system's installed C compiler and look at the assembler code that
6306: results.
6307:
6308: @item EXIT_IGNORES_STACK
6309: Define this macro as a C expression that is nonzero if the return
6310: instruction or the function epilogue ignores the value of the stack
6311: pointer; in other words, if it is safe to delete an instruction to
6312: adjust the stack pointer before a return from the function.
6313:
6314: Note that this macro's value is relevant only for for which frame
6315: pointers are maintained. It is never possible to delete a final stack
6316: adjustment in a function that has no frame pointer, and the compiler
6317: knows this regardless of @code{EXIT_IGNORES_STACK}.
1.1 root 6318:
6319: @item FUNCTION_EPILOGUE (@var{file}, @var{size})
6320: A C compound statement that outputs the assembler code for exit from a
6321: function. The epilogue is responsible for restoring the saved
6322: registers and stack pointer to their values when the function was
6323: called, and returning control to the caller. This macro takes the
6324: same arguments as the macro @code{FUNCTION_PROLOGUE}, and the
6325: registers to restore are determined from @code{regs_ever_live} and
6326: @code{CALL_USED_REGISTERS} in the same way.
6327:
1.1.1.2 root 6328: On some machines, there is a single instruction that does all the work
6329: of returning from the function. On these machines, give that
6330: instruction the name @samp{return} and do not define the macro
6331: @code{FUNCTION_EPILOGUE} at all.
6332:
6333: On machines where functions may or may not have frame-pointers, the
6334: function exit code must vary accordingly. Sometimes the code for
6335: these two cases is completely different. To determine whether a frame
6336: pointer is in wanted, the macro can refer to the variable
6337: @code{frame_pointer_needed}. The variable's value will be 1 at run
6338: time in a function that needs a frame pointer.
6339:
6340: On some machines, some functions pop their arguments on exit while
6341: others leave that for the caller to do. For example, the 68020 when
6342: given @samp{-mrtd} pops arguments in functions that take a fixed
6343: number of arguments.
6344:
6345: Your definition of the macro @code{RETURN_POPS_ARGS} decides which
6346: functions pop their own arguments. @code{FUNCTION_EPILOGUE} needs to
6347: know what was decided. The variable @code{current_function_pops_args}
6348: is nonzero if the function should pop its own arguments. If so, use
6349: the variable @code{current_function_args_size} as the number of bytes
6350: to pop.
6351:
6352: @item FIX_FRAME_POINTER_ADDRESS (@var{addr}, @var{depth})
6353: A C compound statement to alter a memory address that uses the frame
6354: pointer register so that it uses the stack pointer register instead.
6355: This must be done in the instructions that load parameter values into
6356: registers, when the reload pass determines that a frame pointer is not
6357: necessary for the function. @var{addr} will be a C variable name, and
6358: the updated address should be stored in that variable. @var{depth}
6359: will be the current depth of stack temporaries (number of bytes of
6360: arguments currently pushed). The change in offset between a
6361: frame-pointer-relative address and a stack-pointer-relative address
6362: must include @var{depth}.
6363:
6364: Even if your machine description specifies there will always be a
6365: frame pointer in the frame pointer register, you must still define
6366: @code{FIX_FRAME_POINTER_ADDRESS}, but the definition will never be
6367: executed at run time, so it may be empty.
6368: @end table
6369:
6370: @node Library Names, Addressing Modes, Stack Layout, Machine Macros
6371: @section Library Subroutine Names
6372:
6373: @table @code
6374: @item UDIVSI3_LIBCALL
6375: A C string constant giving the name of the function to call for
6376: division of a full-word by a full-word. If you do not define this
6377: macro, the default name is used, which is @code{_udivsi3}, a function
6378: defined in @file{gnulib}.
6379:
6380: @item UMODSI3_LIBCALL
6381: A C string constant giving the name of the function to call for the
6382: remainder in division of a full-word by a full-word. If you do not
6383: define this macro, the default name is used, which is @code{_umodsi3},
6384: a function defined in @file{gnulib}.
6385:
6386: @item TARGET_MEM_FUNCTIONS
6387: Define this macro if GNU CC should generate calls to the System V
6388: (and ANSI C) library functions @code{memcpy} and @code{memset}
6389: rather than the BSD functions @code{bcopy} and @code{bzero}.
1.1 root 6390: @end table
6391:
1.1.1.2 root 6392: @node Addressing Modes, Misc, Library Names, Machine Macros
1.1 root 6393: @section Addressing Modes
6394:
6395: @table @code
6396: @item HAVE_POST_INCREMENT
6397: Define this macro if the machine supports post-increment addressing.
6398:
6399: @item HAVE_PRE_INCREMENT
6400: @itemx HAVE_POST_DECREMENT
6401: @itemx HAVE_PRE_DECREMENT
6402: Similar for other kinds of addressing.
6403:
6404: @item CONSTANT_ADDRESS_P (@var{x})
1.1.1.2 root 6405: A C expression that is 1 if the RTX @var{x} is a constant whose value
1.1 root 6406: is an integer. This includes integers whose values are not explicitly
1.1.1.2 root 6407: known, such as @samp{symbol_ref} and @samp{label_ref} expressions and
6408: @samp{const} arithmetic expressions.
6409:
6410: On most machines, this can be defined as @code{CONSTANT_P (@var{x})},
6411: but a few machines are more restrictive in which constant addresses
6412: are supported.
1.1 root 6413:
6414: @item MAX_REGS_PER_ADDRESS
6415: A number, the maximum number of registers that can appear in a valid
6416: memory address.
6417:
6418: @item GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{label})
6419: A C compound statement with a conditional @code{goto @var{label};}
1.1.1.2 root 6420: executed if @var{x} (an RTX) is a legitimate memory address on the
6421: target machine for a memory operand of mode @var{mode}.
1.1 root 6422:
6423: It usually pays to define several simpler macros to serve as
1.1.1.2 root 6424: subroutines for this one. Otherwise it may be too complicated to
6425: understand.
6426:
6427: This macro must exist in two variants: a strict variant and a
6428: non-strict one. The strict variant is used in the reload pass. It
6429: must be defined so that any pseudo-register that has not been
6430: allocated a hard register is considered a memory reference. In
6431: contexts where some kind of register is required, a pseudo-register
6432: with no hard register must be rejected.
6433:
6434: The non-strict variant is used in other passes. It must be defined to
6435: accept all pseudo-registers in every context where some kind of
6436: register is required.
6437:
6438: Compiler source files that want to use the strict variant of this
6439: macro define the macro @code{REG_OK_STRICT}. You should use an
6440: @code{#ifdef REG_OK_STRICT} conditional to define the strict variant
6441: in that case and the non-strict variant otherwise.
6442:
6443: Typically among the subroutines used to define
6444: @code{GO_IF_LEGITIMATE_ADDRESS} are subroutines to check for
6445: acceptable registers for various purposes (one for base registers, one
6446: for index registers, and so on). Then only these subroutine macros
6447: need have two variants; the higher levels of macros may be the same
6448: whether strict or not.@refill
1.1 root 6449:
6450: @item LEGITIMIZE_ADDRESS (@var{x}, @var{oldx}, @var{mode}, @var{win})
6451: A C compound statement that attempts to replace @var{x} with a valid
1.1.1.2 root 6452: memory address for an operand of mode @var{mode}. @var{win} will be a
6453: C statement label elsewhere in the code; the macro definition may use
1.1 root 6454:
6455: @example
6456: GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{win});
6457: @end example
6458:
6459: @noindent
6460: to avoid further processing if the address has become legitimate.
6461:
6462: @var{x} will always be the result of a call to @code{break_out_memory_refs},
6463: and @var{oldx} will be the operand that was given to that function to produce
6464: @var{x}.
6465:
1.1.1.2 root 6466: The code generated by this macro should not alter the substructure of
6467: @var{x}. If it transforms @var{x} into a more legitimate form, it
6468: should assign @var{x} (which will always be a C variable) a new value.
6469:
6470: It is not necessary for this macro to come up with a legitimate
6471: address. The compiler has standard ways of doing so in all cases. In
6472: fact, it is safe for this macro to do nothing. But often a
6473: machine-dependent strategy can generate better code.
6474:
6475: @item GO_IF_MODE_DEPENDENT_ADDRESS (@var{addr}, @var{label})
6476: A C statement or compound statement with a conditional @code{goto
6477: @var{label};} executed if memory address @var{x} (an RTX) can have
6478: different meanings depending on the machine mode of the memory
6479: reference it is used for.
6480:
6481: Autoincrement and autodecrement addresses typically have mode-dependent
6482: effects because the amount of the increment or decrement is the size
6483: of the operand being addressed. Some machines have other mode-dependent
6484: addresses. Many RISC machines have no mode-dependent addresses.
6485:
6486: You may assume that @var{addr} is a valid address for the machine.
6487:
6488: @item LEGITIMATE_CONSTANT_P (@var{x})
6489: A C expression that is nonzero if @var{x} is a legitimate constant for
6490: an immediate operand on the target machine. You can assume that
6491: either @var{x} is a @samp{const_double} or it satisfies
6492: @code{CONSTANT_P}, so you need not check these things. In fact,
6493: @samp{1} is a suitable definition for this macro on machines where any
6494: @samp{const_double} is valid and anything @code{CONSTANT_P} is valid.@refill
1.1 root 6495: @end table
6496:
6497: @node Misc, Condition Code, Addressing Modes, Machine Macros
6498: @section Miscellaneous Parameters
6499:
6500: @table @code
6501: @item CASE_VECTOR_MODE
1.1.1.2 root 6502: An alias for a machine mode name. This is the machine mode that
6503: elements of a jump-table should have.
1.1 root 6504:
6505: @item CASE_VECTOR_PC_RELATIVE
6506: Define this macro if jump-tables should contain relative addresses.
6507:
1.1.1.2 root 6508: @item CASE_DROPS_THROUGH
6509: Define this if control falls through a @code{case} insn when the index
6510: value is out of range. This means the specified default-label is
6511: actually ignored by the @code{case} insn proper.
6512:
1.1 root 6513: @item IMPLICIT_FIX_EXPR
6514: An alias for a tree code that should be used by default for conversion
1.1.1.2 root 6515: of floating point values to fixed point. Normally,
6516: @code{FIX_ROUND_EXPR} is used.@refill
6517:
6518: @item FIXUNS_TRUNC_LIKE_FIX_TRUNC
6519: Define this macro if the same instructions that convert a floating
6520: point number to a signed fixed point number also convert validly to an
6521: unsigned one.
1.1 root 6522:
6523: @item EASY_DIV_EXPR
1.1.1.2 root 6524: An alias for a tree code that is the easiest kind of division to
6525: compile code for in the general case. It may be
6526: @code{TRUNC_DIV_EXPR}, @code{FLOOR_DIV_EXPR}, @code{CEIL_DIV_EXPR} or
6527: @code{ROUND_DIV_EXPR}. These four division operators differ in how
6528: they round the result to an integer. @code{EASY_DIV_EXPR} is used
6529: when it is permissible to use any of those kinds of division and the
6530: choice should be made on the basis of efficiency.@refill
6531:
6532: @item DEFAULT_SIGNED_CHAR
6533: An expression whose value is 1 or 0, according to whether the type
6534: @code{char} should be signed or unsigned by default. The user can
6535: always override this default with the options @samp{-fsigned-char}
6536: and @samp{-funsigned-char}.
6537:
6538: @item SCCS_DIRECTIVE
6539: Define this if the preprocessor should ignore @code{#sccs} directives
6540: with no error message.
1.1 root 6541:
6542: @item MOVE_MAX
6543: The maximum number of bytes that a single instruction can move quickly
6544: from memory to memory.
6545:
1.1.1.2 root 6546: @item INT_TYPE_SIZE
6547: A C expression for the size in bits of the type @code{int} on the
6548: target machine.
6549:
6550: @item SLOW_BYTE_ACCESS
6551: Define this macro as a C expression which is nonzero if accessing less
6552: than a word of memory (i.e. a @code{char} or a @code{short}) is slow
6553: (requires more than one instruction).
6554:
1.1 root 6555: @item SLOW_ZERO_EXTEND
1.1.1.2 root 6556: Define this macro if zero-extension (of a @code{char} or @code{short}
6557: to an @code{int}) can be done faster if the destination is a register
6558: that is known to be zero.
6559:
6560: If you define this macro, you must have instruction patterns that
6561: recognize RTL structures like this:
6562:
6563: @example
6564: (set (strict-low-part (subreg:QI (reg:SI @dots{}) 0)) @dots{})
6565: @end example
6566:
6567: @noindent
6568: and likewise for @code{HImode}.
1.1 root 6569:
6570: @item SHIFT_COUNT_TRUNCATED
6571: Define this macro if shift instructions ignore all but the lowest few
6572: bits of the shift count. It implies that a sign-extend or zero-extend
6573: instruction for the shift count can be omitted.
6574:
1.1.1.2 root 6575: @item TRULY_NOOP_TRUNCATION (@var{outprec}, @var{inprec})
1.1 root 6576: A C expression which is nonzero if on this machine it is safe to
1.1.1.2 root 6577: ``convert'' an integer of @var{inprec} bits to one of @var{outprec}
6578: bits (where @var{outprec} is smaller than @var{inprec}) by merely
6579: operating on it as if it had only @var{outprec} bits.
1.1 root 6580:
6581: On many machines, this expression can be 1.
6582:
1.1.1.2 root 6583: @item NO_FUNCTION_CSE
6584: Define this macro if it is as good or better to call a constant
6585: function address than to call an address kept in a register.
6586:
6587: @item STORE_FLAG_VALUE
6588: A C expression for the value stored by a store-flag instruction
6589: (@code{s@var{cond}}) when the condition is true. This is usually 1 or
6590: -1; it is required to be an odd number.
6591:
6592: Do not define @code{STORE_FLAG_VALUE} if the machine has no store-flag
6593: instructions.
6594:
1.1 root 6595: @item Pmode
1.1.1.2 root 6596: An alias for the machine mode for pointers. Normally the definition
6597: can be
1.1 root 6598:
6599: @example
6600: #define Pmode SImode
6601: @end example
6602:
6603: @item FUNCTION_MODE
1.1.1.2 root 6604: An alias for the machine mode used for memory references to functions
6605: being called, in @samp{call} RTL expressions. On most machines this
6606: should be @code{QImode}.
1.1 root 6607:
6608: @item CONST_COST (@var{x}, @var{code})
1.1.1.2 root 6609: A part of a C @code{switch} statement that describes the relative
6610: costs of constant RTL expressions. It must contain @code{case} labels
6611: for expression codes @samp{const_int}, @samp{const}, @samp{symbol_ref}, @samp{label_ref}
6612: and @samp{const_double}. Each case must ultimately reach a
6613: @code{return} statement to return the relative cost of the use of that
1.1 root 6614: kind of constant value in an expression. The cost may depend on the
6615: precise value of the constant, which is available for examination in
6616: @var{x}.
6617:
1.1.1.2 root 6618: @var{code} is the expression code---redundant, since it can be
6619: obtained with @code{GET_CODE (@var{x})}.
6620:
6621: @item DOLLARS_IN_IDENTIFIERS
6622: Define this if the character @samp{$} should be allowed in identifier
6623: names.
1.1 root 6624: @end table
6625:
6626: @node Condition Code, Assembler Format, Misc, Machine Macros
6627: @section Condition Code Information
6628:
6629: The file @file{conditions.h} defines a variable @code{cc_status} to
6630: describe how the condition code was computed (in case the interpretation of
6631: the condition code depends on the instruction that it was set by). This
6632: variable contains the RTL expressions on which the condition code is
6633: currently based, and several standard flags.
6634:
6635: Sometimes additional machine-specific flags must be defined in the machine
6636: description header file. It can also add additional machine-specific
6637: information by defining @code{CC_STATUS_MDEP}.
6638:
6639: @table @code
6640: @item CC_STATUS_MDEP
1.1.1.2 root 6641: C code for a data type which is used for declaring the @code{mdep}
6642: component of @code{cc_status}. It defaults to @code{int}.
1.1 root 6643:
6644: @item CC_STATUS_MDEP_INIT
1.1.1.2 root 6645: A C expression for the initial value of the @code{mdep} field. It
6646: defaults to 0.
1.1 root 6647:
6648: @item NOTICE_UPDATE_CC (@var{exp})
6649: A C compound statement to set the components of @code{cc_status}
1.1.1.2 root 6650: appropriately for an insn whose body is @var{exp}. It is this macro's
6651: responsibility to recognize insns that set the condition code as a
6652: byproduct of other activity as well as those that explicitly set
6653: @code{(cc0)}.
6654:
6655: If there are insn that do not set the condition code but do alter
6656: other machine registers, this macro must check to see whether they
6657: invalidate the expressions that the condition code is recorded as
6658: reflecting. For example, on the 68000, insns that store in address
6659: registers do not set the condition code, which means that usually
6660: @code{NOTICE_UPDATE_CC} can leave @code{cc_status} unaltered for such
6661: insns. But suppose that the previous insn set the condition code
6662: based on location @samp{a4@@(102)} and the current insn stores a new
6663: value in @samp{a4}. Although the condition code is not changed by
6664: this, it will no longer be true that it reflects the contents of
6665: @samp{a4@@(102)}. Therefore, @code{NOTICE_UPDATE_CC} must alter
1.1 root 6666: @code{cc_status} in this case to say that nothing is known about the
6667: condition code value.
6668: @end table
6669:
6670: @node Assembler Format,, Condition Code, Machine Macros
6671: @section Output of Assembler Code
6672:
6673: @table @code
1.1.1.2 root 6674: @item ASM_SPEC
6675: A C string constant that tells the GNU CC driver program options to
6676: pass to the assembler. It can also specify how to translate options
6677: you give to GNU CC into options for GNU CC to pass to the assembler.
6678: See the file @file{tm-sun3.h} for an example of this.
6679:
6680: Do not define this macro if it does not need to do anything.
6681:
6682: @item LINK_SPEC
6683: A C string constant that tells the GNU CC driver program options to
6684: pass to the linker. It can also specify how to translate options you
6685: give to GNU CC into options for GNU CC to pass to the linker.
6686:
6687: Do not define this macro if it does not need to do anything.
6688:
6689: @item ASM_FILE_START
6690: A C string constant for text to be output at the start of each
6691: assembler output file. Normally this is @code{"#NO_APP"}, which is a
6692: comment that has no effect on most assemblers but tells the GNU
6693: assembler that it can save time by not checking for certain assembler
6694: constructs.
6695:
6696: @item ASM_APP_ON
6697: A C string constant for text to be output before each @code{asm}
6698: statement or group of consecutive ones. Normally this is
6699: @code{"#APP"}, which is a comment that has no effect on most
6700: assemblers but tells the GNU assembler that it must check the lines
6701: that follow for all valid assembler constructs.
6702:
6703: @item ASM_APP_OFF
6704: A C string constant for text to be output after each @code{asm}
6705: statement or group of consecutive ones. Normally this is
6706: @code{"#NO_APP"}, which tells the GNU assembler to resume making the
6707: time-saving assumptions that are valid for ordinary compiler output.
6708:
1.1 root 6709: @item TEXT_SECTION_ASM_OP
6710: A C string constant for the assembler operation that should precede
6711: instructions and read-only data. Normally @code{".text"} is right.
6712:
6713: @item DATA_SECTION_ASM_OP
1.1.1.2 root 6714: A C string constant for the assembler operation to identify the
6715: following data as writable initialized data. Normally @code{".data"}
6716: is right.
1.1 root 6717:
6718: @item REGISTER_NAMES
1.1.1.2 root 6719: A C initializer containing the assembler's names for the machine
6720: registers, each one as a C string constant. This is what translates
6721: register numbers in the compiler into assembler language.
1.1 root 6722:
6723: @item DBX_REGISTER_NUMBER (@var{regno})
1.1.1.2 root 6724: A C expression that returns the DBX register number for the compiler
6725: register number @var{regno}. In simple cases, the value of this
6726: expression may be @var{regno} itself. But sometimes there are some
6727: registers that the compiler knows about and DBX does not, or vice
6728: versa. In such cases, some register may need to have one number in
6729: the compiler and another for DBX.
6730:
6731: @item DBX_NO_XREFS
6732: Define this macro if DBX on your system does not support the construct
6733: @samp{xs@var{tagname}}. On some systems, this construct is used to
6734: describe a forward reference to a structure named @var{tagname}.
6735: On other systems, this construct is not supported at all.
6736:
6737: @item DBX_CONTIN_LENGTH
6738: A symbol name in DBX-format debugging information is normally
6739: continued (split into two separate @code{.stabs} directives) when it
6740: exceeds a certain length (by default, 80 characters). On some
6741: operating systems, DBX requires this splitting; on others, splitting
6742: must not be done. You can inhibit splitting by defining this macro
6743: with the value zero. You can override the default splitting-length by
6744: defining this macro as an expression for the length you desire.
6745:
6746: @item DBX_CONTIN_CHAR
6747: Normally continuation is indicated by adding a @samp{\} character to
6748: the end of a @code{.stabs} string when a continuation follows. To use
6749: a different character instead, define this macro as a character
6750: constant for the character you want to use. Do not define this macro
6751: if backslash is correct for your system.
6752:
6753: @item ASM_OUTPUT_LABEL (@var{file}, @var{name})
6754: A C statement (sans semicolon) to output to the stdio stream
6755: @var{file} the assembler definition of a label named @var{name}. Use
6756: the expression @code{assemble_name (@var{file}, @var{name})} to output
6757: the name itself; before and after that, output the additional
6758: assembler syntax for defining the name, and a newline.
6759:
6760: @item ASM_DECLARE_FUNCTION_NAME (@var{file}, @var{name})
6761: A C statement (sans semicolon) to output to the stdio stream
6762: @var{file} any text necessary for declaring the name of a function
6763: which is being defined. This macro is responsible for outputting
6764: the label definition (perhaps using @code{ASM_OUTPUT_LABEL}).
6765:
6766: If this macro is not defined, then the function name is defined in the
6767: usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}).
6768:
6769: @item ASM_GLOBALIZE_LABEL (@var{file}, @var{name})
6770: A C statement (sans semicolon) to output to the stdio stream
6771: @var{file} some commands that will make the label @var{name} global;
6772: that is, available for reference from other files. Use the expression
6773: @code{assemble_name (@var{file}, @var{name})} to output the name
6774: itself; before and after that, output the additional assembler syntax
6775: for making that name global, and a newline.
6776:
6777: @item ASM_OUTPUT_EXTERNAL (@var{file}, @var{name})
6778: A C statement (sans semicolon) to output to the stdio stream
6779: @var{file} any text necessary for declaring the name of an external
6780: symbol which is referenced in this compilation but not defined.
6781:
6782: This macro need not be defined if it does not need to output anything.
6783: The GNU assembler and most Unix assemblers don't require anything.
6784:
6785: @item ASM_OUTPUT_LABELREF (@var{file}, @var{name})
6786: A C statement to output to the stdio stream @var{file} a reference in
6787: assembler syntax to a label named @var{name}. The character @samp{_}
6788: should be added to the front of the name, if that is customary on your
6789: operating system, as it is in most Berkeley Unix systems. This macro
6790: is used in @code{assemble_name}.
6791:
6792: @item ASM_OUTPUT_INTERNAL_LABEL (@var{file}, @var{prefix}, @var{num})
6793: A C statement to output to the stdio stream @var{file} a label whose
6794: name is made from the string @var{prefix} and the number @var{num}.
6795: These labels are used for internal purposes, and there is no reason
6796: for them to appear in the symbol table of the object file. On many
6797: systems, the letter @samp{L} at the beginning of a label has this
6798: effect. The usual definition of this macro is as follows:
6799:
6800: @example
6801: fprintf (@var{file}, "L%s%d:\n", @var{prefix}, @var{num})
6802: @end example
6803:
6804: @item ASM_OUTPUT_CASE_LABEL (@var{file}, @var{prefix}, @var{num}, @var{table})
6805: Define this if the label before a jump-table needs to be output
6806: specially. The first three arguments are the same as for
6807: @code{ASM_OUTPUT_INTERNAL_LABEL}; the fourth argument is the
6808: jump-table which follows (a @samp{jump_insn} containing an
6809: @samp{addr_vec} or @samp{addr_diff_vec}).
6810:
6811: This feature is used on system V to output a @code{swbeg} statement
6812: for the table.
6813:
6814: If this macro is not defined, these labels are output with
6815: @code{ASM_OUTPUT_INTERNAL_LABEL}.
6816:
6817: @item ASM_FORMAT_PRIVATE_NAME (@var{outvar}, @var{name}, @var{number})
6818: A C expression to assign to @var{outvar} (which is a variable of type
6819: @code{char *}) a newly allocated string made from the string
6820: @var{name} and the number @var{number}, with some suitable punctuation
6821: added. Use @code{alloca} to get space for the string.
6822:
6823: This string will be used as the argument to @code{ASM_OUTPUT_LABELREF}
6824: to produce an assembler label for an internal static variable whose
6825: name is @var{name}. Therefore, the string must be such as to result
6826: in valid assembler code. The argument @var{number} is different each
6827: time this macro is executed; it prevents conflicts between
6828: similarly-named internal static variables in different scopes.
6829:
6830: Ideally this string should not be a valid C identifier, to prevent any
6831: conflict with the user's own symbols. Most assemblers allow periods
6832: or percent signs in assembler symbols; putting at least one of these
6833: between the name and the number will suffice.
6834:
6835: @item ASM_OUTPUT_ADDR_DIFF_ELT (@var{file}, @var{value}, @var{rel})
6836: This macro should be provided on machines where the addresses
6837: in a dispatch table are relative to the table's own address.
6838:
6839: The definition should be a C statement to output to the stdio stream
6840: @var{file} an assembler pseudo-instruction to generate a difference
6841: between two labels. @var{value} and @var{rel} are the numbers of two
6842: internal labels. The definitions of these labels are output using
6843: @code{ASM_OUTPUT_INTERNAL_LABEL}, and they must be printed in the same
6844: way here. For example,
6845:
6846: @example
6847: fprintf (@var{file}, "\t.word L%d-L%d\n",
6848: @var{value}, @var{rel})
6849: @end example
6850:
6851: @item ASM_OUTPUT_ADDR_VEC_ELT (@var{file}, @var{value})
6852: This macro should be provided on machines where the addresses
6853: in a dispatch table are absolute.
6854:
6855: The definition should be a C statement to output to the stdio stream
6856: @var{file} an assembler pseudo-instruction to generate a reference to
6857: a label. @var{value} is the number of an internal label whose
6858: definition is output using @code{ASM_OUTPUT_INTERNAL_LABEL}.
6859: For example,
6860:
6861: @example
6862: fprintf (@var{file}, "\t.word L%d\n", @var{value})
6863: @end example
1.1 root 6864:
6865: @item ASM_OUTPUT_DOUBLE (@var{file}, @var{value})
6866: A C statement to output to the stdio stream @var{file} an assembler
6867: instruction to assemble a @code{double} constant whose value is
1.1.1.2 root 6868: @var{value}. @var{value} will be a C expression of type
6869: @code{double}.
1.1 root 6870:
6871: @item ASM_OUTPUT_FLOAT (@var{file}, @var{value})
6872: A C statement to output to the stdio stream @var{file} an assembler
1.1.1.2 root 6873: instruction to assemble a @code{float} constant whose value is
6874: @var{value}. @var{value} will be a C expression of type @code{float}.
6875:
6876: @item ASM_OUTPUT_INT (@var{file}, @var{exp})
6877: @itemx ASM_OUTPUT_SHORT (@var{file}, @var{exp})
6878: @itemx ASM_OUTPUT_CHAR (@var{file}, @var{exp})
6879: A C statement to output to the stdio stream @var{file} an assembler
6880: instruction to assemble a @code{int}, @code{short} or @code{char}
6881: constant whose value is @var{value}. The argument @var{exp} will be
6882: an RTL expression which represents a constant value. Use
6883: @samp{output_addr_const (@var{exp})} to output this value as an
6884: assembler expression.@refill
6885:
6886: @item ASM_OUTPUT_BYTE (@var{file}, @var{value})
6887: A C statement to output to the stdio stream @var{file} an assembler
6888: instruction to assemble a single byte containing the number @var{value}.
6889:
6890: @item ASM_OUTPUT_ASCII (@var{file}, @var{ptr}, @var{len})
6891: A C statement to output to the stdio stream @var{file} an assembler
6892: instruction to assemble a string constant containing the @var{len}
6893: bytes at @var{ptr}. @var{ptr} will be a C expression of type
6894: @code{char *} and @var{len} a C expression of type @code{int}.
6895:
6896: If the assembler has a @code{.ascii} pseudo-op as found in the
6897: Berkeley Unix assembler, do not define the macro
6898: @code{ASM_OUTPUT_ASCII}.
1.1 root 6899:
6900: @item ASM_OUTPUT_SKIP (@var{file}, @var{nbytes})
6901: A C statement to output to the stdio stream @var{file} an assembler
6902: instruction to advance the location counter by @var{nbytes} bytes.
6903: @var{nbytes} will be a C expression of type @code{int}.
6904:
6905: @item ASM_OUTPUT_ALIGN (@var{file}, @var{power})
6906: A C statement to output to the stdio stream @var{file} an assembler
6907: instruction to advance the location counter to a multiple of 2 to the
6908: @var{power} bytes. @var{power} will be a C expression of type @code{int}.
6909:
1.1.1.2 root 6910: @item ASM_OUTPUT_COMMON (@var{file}, @var{name}, @var{size})
6911: A C statement (sans semicolon) to output to the stdio stream
6912: @var{file} the assembler definition of a common-label named @var{name}
6913: whose size is @var{size} bytes. Use the expression
6914: @code{assemble_name (@var{file}, @var{name})} to output the name
6915: itself; before and after that, output the additional assembler syntax
6916: for defining the name, and a newline.
6917:
6918: This macro controls how the assembler definitions of uninitialized
6919: global variables are output.
6920:
6921: @item ASM_OUTPUT_LOCAL (@var{file}, @var{name}, @var{size})
6922: A C statement (sans semicolon) to output to the stdio stream
6923: @var{file} the assembler definition of a local-common-label named
6924: @var{name} whose size is @var{size} bytes. Use the expression
6925: @code{assemble_name (@var{file}, @var{name})} to output the name
6926: itself; before and after that, output the additional assembler syntax
6927: for defining the name, and a newline.
6928:
6929: This macro controls how the assembler definitions of uninitialized
6930: static variables are output.
1.1 root 6931:
6932: @item TARGET_BELL
1.1.1.2 root 6933: A C constant expression for the integer value for escape sequence
6934: @samp{\a}.
1.1 root 6935:
6936: @item TARGET_BS
6937: @itemx TARGET_TAB
6938: @itemx TARGET_NEWLINE
6939: C constant expressions for the integer values for escape sequences
6940: @samp{\b}, @samp{\t} and @samp{\n}.
6941:
6942: @item TARGET_VT
6943: @itemx TARGET_FF
6944: @itemx TARGET_CR
6945: C constant expressions for the integer values for escape sequences
6946: @samp{\v}, @samp{\f} and @samp{\r}.
6947:
1.1.1.2 root 6948: @item ASM_OUTPUT_OPCODE (@var{file}, @var{ptr})
6949: Define this macro if you are using an unusual assembler that
6950: requires different names for the machine instructions.
6951:
6952: The definition is a C statement or statements which output an
6953: assembler instruction opcode to the stdio stream @var{file}. The
6954: macro-operand @var{ptr} is a variable of type @code{char *} which
6955: points to the opcode name in its ``internal'' form---the form that is
6956: written in the machine description. The definition should output the
6957: opcode name to @var{file}, performing any translation you desire, and
6958: increment the variable @var{ptr} to point at the end of the opcode
6959: so that it will not be output twice.
6960:
6961: In fact, your macro definition may process less than the entire opcode
6962: name, or more than the opcode name; but if you want to process text
6963: that includes @samp{%}-sequences to substitute operands, you must take
6964: care of the substitution yourself. Just be sure to increment
6965: @var{ptr} over whatever text should not be output normally.
6966:
6967: If the macro definition does nothing, the instruction is output
6968: in the usual way.
6969:
6970: @item PRINT_OPERAND (@var{file}, @var{x}, @var{code})
6971: A C compound statement to output to stdio stream @var{file} the
6972: assembler syntax for an instruction operand @var{x}. @var{x} is an
6973: RTL expression.
6974:
6975: @var{code} is a value that can be used to specify one of several ways
6976: of printing the operand. It is used when identical operands must be
6977: printed differently depending on the context. @var{code} comes from
6978: the @samp{%} specification that was used to request printing of the
6979: operand. If the specification was just @samp{%@var{digit}} then
6980: @var{code} is 0; if the specification was @samp{%@var{ltr}
6981: @var{digit}} then @var{code} is the ASCII code for @var{ltr}.
6982:
6983: If @var{x} is a register, this macro should print the register's name.
6984: The names can be found in an array @code{reg_names} whose type is
6985: @code{char *[]}. @code{reg_names} is initialized from
6986: @code{REGISTER_NAMES}.
6987:
6988: When the machine description has a specification @samp{%@var{punct}}
6989: (a @samp{%} followed by a punctuation character), this macro is called
6990: with a null pointer for @var{x} and the punctuation character for
6991: @var{code}.
1.1 root 6992:
6993: @item PRINT_OPERAND_ADDRESS (@var{file}, @var{x})
1.1.1.2 root 6994: A C compound statement to output to stdio stream @var{file} the
6995: assembler syntax for an instruction operand that is a memory reference
6996: whose address is @var{x}. @var{x} is an RTL expression.
6997:
6998: @item ASM_OPEN_PAREN
6999: @itemx ASM_CLOSE_PAREN
7000: These macros are defined as C string constant, describing the syntax
7001: in the assembler for grouping arithmetic expressions. The following
7002: definitions are correct for most assemblers:
7003:
7004: @example
7005: #define ASM_OPEN_PAREN "("
7006: #define ASM_CLOSE_PAREN ")"
7007: @end example
7008: @end table
7009:
7010: @node Config,, Machine Macros, Top
7011: @chapter The Configuration File
7012:
7013: The configuration file @file{config-@var{machine}.h} contains macro
7014: definitions that describe the machine and system on which the compiler is
7015: running. Most of the values in it are actually the same on all machines
7016: that GNU CC runs on, so most all configuration files are identical. But
7017: there are some macros that vary:
7018:
7019: @table @code
7020: @item FAILURE_EXIT_CODE
7021: A C expression for the status code to be returned when the compiler
7022: exits after serious errors.
7023:
7024: @item SUCCESS_EXIT_CODE
7025: A C expression for the status code to be returned when the compiler
7026: exits without serious errors.
1.1 root 7027: @end table
7028:
7029: @contents
7030: @bye
This archive runs on limited infrastructure. Preserving old code on modern bandwidth. Automated agents are requested to crawl responsibly.