![[BACK]](/cvsweb/icons/back.gif){kind=icon}
Return to internals.texinfo CVS log ![[TXT]](/cvsweb/icons/text.gif){kind=icon}
![[DIR]](/cvsweb/icons/dir.gif){kind=icon}
Up to [GCC 1.x] / gcc|
Return to internals.texinfo CVS log | Up to [GCC 1.x] / gcc |
1.1 root 1: \input texinfo @c -*-texinfo-*-
2:
3: @settitle Internals of GNU CC
4: @setfilename internals
5:
6: @ifinfo
7: This file documents the internals of the GNU compiler.
8:
1.1.1.2 root 9: Copyright (C) 1988 Free Software Foundation, Inc.
1.1 root 10:
11: Permission is granted to make and distribute verbatim copies of
12: this manual provided the copyright notice and this permission notice
13: are preserved on all copies.
14:
15: @ignore
16: Permission is granted to process this file through Tex and print the
17: results, provided the printed document carries copying permission
18: notice identical to this one except for the removal of this paragraph
19: (this paragraph not being relevant to the printed manual).
20:
21: @end ignore
22: Permission is granted to copy and distribute modified versions of this
23: manual under the conditions for verbatim copying, provided also that the
24: section entitled ``GNU CC General Public License'' is included exactly as
25: in the original, and provided that the entire resulting derived work is
26: distributed under the terms of a permission notice identical to this one.
27:
28: Permission is granted to copy and distribute translations of this manual
29: into another language, under the above conditions for modified versions,
1.1.1.2 root 30: except that the section entitled ``GNU CC General Public License'' and
31: this permission notice may be included in translations approved by the
32: Free Software Foundation instead of in the original English.
1.1 root 33: @end ifinfo
34:
35: @setchapternewpage odd
36:
37: @titlepage
38: @center @titlefont{Internals of GNU CC}
39: @sp 2
40: @center Richard M. Stallman
1.1.1.2 root 41: @sp 3
1.1.1.7 ! root 42: @center last updated 17 August 1988
1.1.1.2 root 43: @sp 1
1.1.1.7 ! root 44: @center for version 1.26
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
1.1.1.7 ! root 278: Michael Tiemann of MCC wrote most of the description of the National
! 279: Semiconductor 32000 series cpu. He also wrote the code for inline
! 280: function integration and for the SPARC cpu and Motorola 88000 cpu
! 281: and part of the Sun FPA support.
! 282:
! 283: @item
! 284: Jan Stein of the Chalmers Computer Society provided support for
! 285: Genix, as well as part of the 32000 machine description.
! 286:
! 287: @item
! 288: Randy Smith finished the Sun FPA support.
1.1.1.2 root 289:
290: @item
1.1.1.3 root 291: Robert Brown implemented the support for Encore 32000 systems.
292:
293: @item
1.1.1.4 root 294: David Kashtan of SRI adapted GNU CC to the Vomit-Making System.
1.1.1.2 root 295:
296: @item
297: Alex Crain provided changes for the 3b1.
298:
299: @item
1.1.1.4 root 300: Greg Satz and Chris Hanson assisted in making GNU CC work on HP-UX for
301: the 9000 series 300.
1.1.1.6 root 302:
303: @item
304: William Schelter did most of the work on the Intel 80386 support.
1.1.1.2 root 305: @end itemize
306:
307: @node Options, Installation, Contributors, Top
308: @chapter GNU CC Command Options
309:
310: The GNU C compiler uses a command syntax much like the Unix C compiler.
311: The @code{gcc} program accepts options and file names as operands.
312: Multiple single-letter options may @emph{not} be grouped: @samp{-dr} is
313: very different from @samp{-d -r}.
314:
315: When you invoke GNU CC, it normally does preprocessing, compilation,
316: assembly and linking. File names which end in @samp{.c} are taken as C
317: source to be preprocessed and compiled; compiler output files plus any
318: input files with names ending in @samp{.s} are assembled; then the
319: resulting object files, plus any other input files, are linked together to
320: produce an executable.
321:
322: Command options allow you to stop this process at an intermediate stage.
323: For example, the @samp{-c} option says not to run the linker. Then the
324: output consists of object files output by the assembler.
325:
326: Other command options are passed on to one stage. Some options control
327: the preprocessor and others the compiler itself. Yet other options
328: control the assembler and linker; these are not documented here because the
329: GNU assembler and linker are not yet released.
330:
331: Here are the options to control the overall compilation process, including
332: those that say whether to link, whether to assemble, and so on.
333:
334: @table @samp
335: @item -o @var{file}
336: Place output in file @var{file}. This applies regardless to whatever
337: sort of output is being produced, whether it be an executable file,
338: an object file, an assembler file or preprocessed C code.
339:
1.1.1.3 root 340: If @samp{-o} is not specified, the default is to put an executable file
1.1.1.2 root 341: in @file{a.out}, the object file @file{@var{source}.c} in
342: @file{@var{source}.o}, an assembler file in @file{@var{source}.s}, and
343: preprocessed C on standard output.@refill
344:
345: @item -c
346: Compile or assemble the source files, but do not link. Produce object
347: files with names made by replacing @samp{.c} or @samp{.s} with
348: @samp{.o} at the end of the input file names. Do nothing at all for
349: object files specified as input.
350:
351: @item -S
352: Compile into assembler code but do not assemble. The assembler output
353: file name is made by replacing @samp{.c} with @samp{.s} at the end of
354: the input file name. Do nothing at all for assembler source files or
355: object files specified as input.
356:
357: @item -E
358: Run only the C preprocessor. Preprocess all the C source files
359: specified and output the results to standard output.
360:
361: @item -v
362: Compiler driver program prints the commands it executes as it runs
363: the preprocessor, compiler proper, assembler and linker. Some of
364: these are directed to print their own version numbers.
365:
366: @item -B@var{prefix}
367: Compiler driver program tries @var{prefix} as a prefix for each
368: program it tries to run. These programs are @file{cpp}, @file{cc1},
369: @file{as} and @file{ld}.
370:
371: For each subprogram to be run, the compiler driver first tries the
372: @samp{-B} prefix, if any. If that name is not found, or if @samp{-B}
373: was not specified, the driver tries two standard prefixes, which are
374: @file{/usr/lib/gcc-} and @file{/usr/local/lib/gcc-}. If neither of
375: those results in a file name that is found, the unmodified program
376: name is searched for using the directories specified in your
377: @samp{PATH} environment variable.
378:
379: The run-time support file @file{gnulib} is also searched for using
380: the @samp{-B} prefix, if needed. If it is not found there, the two
381: standard prefixes above are tried, and that is all. The file is left
382: out of the link if it is not found by those means. Most of the time,
383: on most machines, you can do without it.
384: @end table
385:
386: These options control the details of C compilation itself.
387:
388: @table @samp
389: @item -ansi
390: Support all ANSI standard C programs.
391:
392: This turns off certain features of GNU C that are incompatible with
393: ANSI C, such as the @code{asm}, @code{inline} and @code{typeof}
394: keywords, and predefined macros such as @code{unix} and @code{vax}
395: that identify the type of system you are using. It also enables the
396: undesirable and rarely used ANSI trigraph feature.
397:
398: The @samp{-ansi} option does not cause non-ANSI programs to be
399: rejected gratuitously. For that, @samp{-pedantic} is required in
400: addition to @samp{-ansi}.
401:
402: The macro @code{__STRICT_ANSI__} is predefined when the @samp{-ansi}
403: option is used. Some header files may notice this macro and refrain
404: from declaring certain functions or defining certain macros that the
405: ANSI standard doesn't call for; this is to avoid interfering with
406: any programs that might use these names for other things.
407:
408: @item -traditional
409: Attempt to support some aspects of traditional C compilers.
410: Specifically:
411:
412: @itemize @bullet
413: @item
414: All @code{extern} declarations take effect globally even if they
415: are written inside of a function definition. This includes implicit
416: declarations of functions.
417:
418: @item
419: The keywords @code{typeof}, @code{inline}, @code{signed}, @code{const}
420: and @code{volatile} are not recognized.@refill
421:
422: @item
423: Comparisons between pointers and integers are always allowed.
424:
425: @item
426: Integer types @code{unsigned short} and @code{unsigned char} promote
427: to @code{unsigned int}.
428:
429: @item
1.1.1.5 root 430: Out-of-range floating point literals are not an error.
431:
432: @item
1.1.1.2 root 433: In the preprocessor, comments convert to nothing at all, rather than to
434: a space. This allows traditional token concatenation.
435:
436: @item
437: In the preprocessor, single and double quote characters are ignored
438: when scanning macro definitions, so that macro arguments can be replaced
439: even within a string or character constant. Quote characters are also
440: ignored when skipping text inside a failing conditional directive.
441: @end itemize
442:
443: @item -O
444: Optimize. Optimizing compilation takes somewhat more time, and a lot
445: more memory for a large function.
446:
447: Without @samp{-O}, the compiler's goal is to reduce the cost of
448: compilation and to make debugging produce the expected results.
449: Statements are independent: if you stop the program with a breakpoint
450: between statements, you can then assign a new value to any variable or
451: change the program counter to any other statement in the function and
452: get exactly the results you would expect from the source code.
453:
454: Without @samp{-O}, only variables declared @code{register} are
455: allocated in registers. The resulting compiled code is a little worse
456: than produced by PCC without @samp{-O}.
457:
458: With @samp{-O}, the compiler tries to reduce code size and execution
459: time.
460:
461: Some of the @samp{-f} options described below turn specific kinds of
462: optimization on or off.
463:
464: @item -g
1.1.1.4 root 465: Produce debugging information in the operating system's native
466: format (for DBX or SDB).
1.1.1.2 root 467:
468: Unlike most other C compilers, GNU CC allows you to use @samp{-g} with
469: @samp{-O}. The shortcuts taken by optimized code may occasionally
470: produce surprising results: some variables you declared may not exist
471: at all; flow of control may briefly move where you did not expect it;
472: some statements may not be executed because they compute constant
473: results or their values were already at hand; some statements may
474: execute in different places because they were moved out of loops.
475: Nevertheless it proves possible to debug optimized output. This makes
476: it reasonable to use the optimizer for programs that might have bugs.
477:
478: @item -gg
479: Produce debugging information in GDB's own format. This requires
480: the GNU assembler and linker in order to work.
481:
482: @item -w
483: Inhibit all warning messages.
484:
485: @item -W
486: Print extra warning messages for these events:
487:
488: @itemize @bullet
489: @item
490: An automatic variable is used without first being initialized.
491:
492: These warnings are possible only in optimizing compilation,
493: because they require data flow information that is computed only
494: when optimizing. They occur only for variables that are
495: candidates for register allocation. Therefore, they do not occur
496: for a variable that is declared @code{volatile}, or whose address
497: is taken, or whose size is other than 1, 2, 4 or 8 bytes. Also,
498: they do not occur for structures, unions or arrays, even when
499: they are in registers.
500:
501: Note that there may be no warning about a variable that is used
502: only to compute a value that itself is never used, because such
503: computations may be deleted by the flow analysis pass before the
504: warnings are printed.
505:
506: These warnings are made optional because GNU CC is not smart
507: enough to see all the reasons why the code might be correct
508: despite appearing to have an error. Here is one example of how
509: this can happen:
510:
511: @example
512: @{
513: int x;
514: switch (y)
515: @{
516: case 1: x = 1;
517: break;
518: case 2: x = 4;
519: break;
520: case 3: x = 5;
521: @}
522: foo (x);
523: @}
524: @end example
525:
526: @noindent
527: If the value of @code{y} is always 1, 2 or 3, then @code{x} is
528: always initialized, but GNU CC doesn't know this. Here is
529: another common case:
530:
531: @example
532: @{
533: int save_y;
534: if (change_y) save_y = y, y = new_y;
535: @dots{}
536: if (change_y) y = save_y;
537: @}
538: @end example
539:
540: @noindent
1.1.1.4 root 541: This has no bug because @code{save_y} is used only if it is set.
1.1.1.2 root 542:
543: @item
544: A nonvolatile automatic variable might be changed by a call to
545: @code{longjmp}. These warnings as well are possible only in
546: optimizing compilation.
547:
548: The compiler sees only the calls to @code{setjmp}. It cannot know
549: where @code{longjmp} will be called; in fact, a signal handler could
550: call it at any point in the code. As a result, you may get a warning
551: even when there is in fact no problem because @code{longjmp} cannot
552: in fact be called at the place which would cause a problem.
553:
554: @item
555: A function can return either with or without a value. (Falling
556: off the end of the function body is considered returning without
557: a value.) For example, this function would inspire such a
558: warning:
559:
560: @example
561: foo (a)
562: @{
563: if (a > 0)
564: return a;
565: @}
566: @end example
567:
568: Spurious warnings can occur because GNU CC does not realize that
569: certain functions (including @code{abort} and @code{longjmp})
570: will never return.
571: @end itemize
572:
573: In the future, other useful warnings may also be enabled by this
574: option.
575:
576: @item -Wimplicit
577: Warn whenever a function is implicitly declared.
578:
579: @item -Wreturn-type
580: Warn whenever a function is defined with a return-type that defaults
581: to @code{int}. Also warn about any @code{return} statement with no
582: return-value in a function whose return-type is not @code{void}.
583:
584: @item -Wcomment
585: Warn whenever a comment-start sequence @samp{/*} appears in a comment.
586:
1.1.1.3 root 587: @item -Wall
588: All of the above @samp{-W} options combined.
589:
1.1.1.2 root 590: @item -p
591: Generate extra code to write profile information suitable for the
592: analysis program @code{prof}.
593:
594: @item -pg
595: Generate extra code to write profile information suitable for the
596: analysis program @code{gprof}.
597:
1.1.1.3 root 598: @item -l@var{library}
599: Search a standard list of directories for a library named
600: @var{library}, which is actually a file named
601: @file{lib@var{library}.a}. The linker uses this file as if it
602: had been specified precisely by name.
603:
604: The directories searched include several standard system directories
605: plus any that you specify with @samp{-L}.
606:
607: Normally the files found this way are library files---archive files
608: whose members are object files. The linker handles an archive file by
1.1.1.4 root 609: scanning through it for members which define symbols that have so far
610: been referenced but not defined. But if the file that is found is an
1.1.1.3 root 611: ordinary object file, it is linked in the usual fashion. The only
1.1.1.4 root 612: difference between using an @samp{-l} option and specifying a file name
613: is that @samp{-l} searches several directories.
1.1.1.3 root 614:
615: @item -L@var{dir}
616: Add directory @var{dir} to the list of directories to be searched
617: for @samp{-l}.
1.1.1.2 root 618:
619: @item -nostdlib
620: Don't use the standard system libraries and startup files when
621: linking. Only the files you specify (plus @file{gnulib}) will be
622: passed to the linker.
623:
624: @item -m@var{machinespec}
625: Machine-dependent option specifying something about the type of target
626: machine. These options are defined by the macro
627: @code{TARGET_SWITCHES} in the machine description. The default for
628: the options is also defined by that macro, which enables you to change
629: the defaults.@refill
630:
631: These are the @samp{-m} options defined in the 68000 machine
632: description:
633:
634: @table @samp
635: @item -m68020
636: Generate output for a 68020 (rather than a 68000). This is the
637: default if you use the unmodified sources.
638:
639: @item -m68000
640: Generate output for a 68000 (rather than a 68020).
641:
642: @item -m68881
643: Generate output containing 68881 instructions for floating point.
644: This is the default if you use the unmodified sources.
645:
646: @item -msoft-float
647: Generate output containing library calls for floating point.
648:
649: @item -mshort
650: Consider type @code{int} to be 16 bits wide, like @code{short int}.
651:
652: @item -mnobitfield
653: Do not use the bit-field instructions. @samp{-m68000} implies
654: @samp{-mnobitfield}.
655:
656: @item -mbitfield
657: Do use the bit-field instructions. @samp{-m68020} implies
658: @samp{-mbitfield}. This is the default if you use the unmodified
659: sources.
660:
661: @item -mrtd
662: Use a different function-calling convention, in which functions
663: that take a fixed number of arguments return with the @code{rtd}
664: instruction, which pops their arguments while returning. This
665: saves one instruction in the caller since there is no need to pop
666: the arguments there.
667:
668: This calling convention is incompatible with the one normally
669: used on Unix, so you cannot use it if you need to call libraries
670: compiled with the Unix compiler.
671:
672: Also, you must provide function prototypes for all functions that
673: take variable numbers of arguments (including @code{printf});
674: otherwise incorrect code will be generated for calls to those
675: functions.
676:
677: In addition, seriously incorrect code will result if you call a
678: function with too many arguments. (Normally, extra arguments are
679: harmlessly ignored.)
680:
681: The @code{rtd} instruction is supported by the 68010 and 68020
682: processors, but not by the 68000.
683: @end table
684:
685: These @samp{-m} options are defined in the Vax machine description:
686:
687: @table @samp
688: @item -munix
689: Do not output certain jump instructions (@code{aobleq} and so on)
690: that the Unix assembler for the Vax cannot handle across long
691: ranges.
692:
693: @item -mgnu
694: Do output those jump instructions, on the assumption that you
695: will assemble with the GNU assembler.
1.1.1.3 root 696:
697: @item -mg
698: Output code for g-format floating point numbers instead of d-format.
1.1.1.2 root 699: @end table
700:
701: @item -f@var{flag}
702: Specify machine-independent flags. These are the flags:
703:
704: @table @samp
705: @item -ffloat-store
706: Do not store floating-point variables in registers. This
707: prevents undesirable excess precision on machines such as the
708: 68000 where the floating registers (of the 68881) keep more
709: precision than a @code{double} is supposed to have.
710:
711: For most programs, the excess precision does only good, but a few
712: programs rely on the precise definition of IEEE floating point.
713: Use @samp{-ffloat-store} for such programs.
714:
715: @item -fno-asm
716: Do not recognize @code{asm}, @code{inline} or @code{typeof} as a
717: keyword. These words may then be used as identifiers.
718:
719: @item -fno-defer-pop
720: Always pop the arguments to each function call as soon as that
721: function returns. Normally the compiler (when optimizing) lets
722: arguments accumulate on the stack for several function calls and
723: pops them all at once.
724:
725: @item -fcombine-regs
726: Allow the combine pass to combine an instruction that copies one
727: register into another. This might or might not produce better
728: code when used in addition to @samp{-O}. I am interested in
729: hearing about the difference this makes.
730:
731: @item -fforce-mem
732: Force memory operands to be copied into registers before doing
733: arithmetic on them. This may produce better code by making all
734: memory references potential common subexpressions. When they are
735: not common subexpressions, instruction combination should
736: eliminate the separate register-load. I am interested in hearing
737: about the difference this makes.
738:
739: @item -fforce-addr
740: Force memory address constants to be copied into registers before
741: doing arithmetic on them. This may produce better code just as
742: @samp{-fforce-mem} may. I am interested in hearing about the
743: difference this makes.
744:
745: @item -fomit-frame-pointer
746: Don't keep the frame pointer in a register for functions that
747: don't need one. This avoids the instructions to save, set up and
748: restore frame pointers; it also makes an extra register available
749: in many functions. @strong{It also makes debugging impossible.}
750:
751: On some machines, such as the Vax, this flag has no effect,
752: because the standard calling sequence automatically handles the
753: frame pointer and nothing is saved by pretending it doesn't
754: exist. The machine-description macro
755: @code{FRAME_POINTER_REQUIRED} controls whether a target machine
756: supports this flag. @xref{Registers}.@refill
757:
758: @item -finline-functions
759: Integrate all simple functions into their callers. The compiler
760: heuristically decides which functions are simple enough to be
761: worth integrating in this way.
762:
763: If all calls to a given function are integrated, and the function
764: is declared @code{static}, then the function is normally not
765: output as assembler code in its own right.
766:
767: @item -fkeep-inline-functions
768: Even if all calls to a given function are integrated, and the
769: function is declared @code{static}, nevertheless output a
770: separate run-time callable version of the function.
771:
772: @item -fwritable-strings
773: Store string constants in the writable data segment and don't
774: uniquize them. This is for compatibility with old programs which
775: assume they can write into string constants. Writing into string
776: constants is a very bad idea; ``constants'' should be constant.
777:
778: @item -fno-function-cse
779: Do not put function addresses in registers; make each instruction
780: that calls a constant function contain the function's address
781: explicitly.
782:
783: This option results in less efficient code, but some strange
784: hacks that alter the assembler output may be confused by the
785: optimizations performed when this option is not used.
786:
787: @item -fvolatile
788: Consider all memory references through pointers to be volatile.
789:
790: @item -funsigned-char
791: Let the type @code{char} be the unsigned, like @code{unsigned
792: char}.
793:
794: Each kind of machine has a default for what @code{char} should
795: be. It is either like @code{unsigned char} by default or like
796: @code{signed char} by default. (Actually, at present, the
797: default is always signed.)
798:
799: The type @code{char} is always a distinct type from either
800: @code{signed char} or @code{unsigned char}, even though its
801: behavior is always just like one of those two.
802:
803: @item -fsigned-char
804: Let the type @code{char} be signed, like @code{signed char}.
805:
806: @item -ffixed-@var{reg}
807: Treat the register named @var{reg} as a fixed register; generated
808: code should never refer to it (except perhaps as a stack pointer,
809: frame pointer or in some other fixed role).
810:
811: @var{reg} must be the name of a register. The register names
812: accepted are machine-specific and are defined in the
813: @code{REGISTER_NAMES} macro in the machine description macro
814: file.
815:
816: @item -fcall-used-@var{reg}
817: Treat the register named @var{reg} as an allocatable register
818: that is clobbered by function calls. It may be allocated for
819: temporaries or variables that do not live across a call.
820: Functions compiled this way will not save and restore the
821: register @var{reg}.
822:
823: Use of this flag for a register that has a fixed pervasive role
824: in the machine's execution model, such as the stack pointer or
825: frame pointer, will produce disastrous results.
826:
827: @item -fcall-saved-@var{reg}
828: Treat the register named @var{reg} as an allocatable register
829: saved by functions. It may be allocated even for temporaries or
830: variables that live across a call. Functions compiled this way
831: will save and restore the register @var{reg} if they use it.
832:
833: Use of this flag for a register that has a fixed pervasive role
834: in the machine's execution model, such as the stack pointer or
835: frame pointer, will produce disastrous results.
836:
837: A different sort of disaster will result from the use of this
838: flag for a register in which function values are may be returned.
839: @end table
840:
841: @item -d@var{letters}
842: Says to make debugging dumps at times specified by @var{letters}.
843: Here are the possible letters:
844:
845: @table @samp
846: @item r
847: Dump after RTL generation.
848: @item j
849: Dump after first jump optimization.
850: @item J
851: Dump after last jump optimization.
852: @item s
853: Dump after CSE (including the jump optimization that sometimes
854: follows CSE).
855: @item L
856: Dump after loop optimization.
857: @item f
858: Dump after flow analysis.
859: @item c
860: Dump after instruction combination.
861: @item l
862: Dump after local register allocation.
863: @item g
864: Dump after global register allocation.
865: @item m
866: Print statistics on memory usage, at the end of the run.
867: @end table
868:
869: @item -pedantic
870: Issue all the warnings demanded by strict ANSI standard C; reject
871: all programs that use forbidden extensions.
872:
873: Valid ANSI standard C programs should compile properly with or without
874: this option (though a rare few will require @samp{-ansi}). However,
875: without this option, certain GNU extensions and traditional C features
876: are supported as well. With this option, they are rejected. There is
877: no reason to @i{use} this option; it exists only to satisfy pedants.
878: @end table
879:
880: These options control the C preprocessor, which is run on each C source
881: file before actual compilation. If you use the @samp{-E} option, nothing
882: is done except C preprocessing. Some of these options make sense only
883: together with @samp{-E} because they request preprocessor output that is
884: not suitable for actual compilation.
885:
886: @table @samp
887: @item -C
888: Tell the preprocessor not to discard comments. Used with the
889: @samp{-E} option.
890:
891: @item -I@var{dir}
892: Search directory @var{dir} for include files.
893:
1.1.1.3 root 894: @item -I-
895: Any directories specified with @samp{-I} options before the @samp{-I-}
896: option are searched only for the case of @samp{#include "@var{file}"};
897: they are not searched for @samp{#include <@var{file}>}.
898:
899: If additional directories are specified with @samp{-I} options after
900: the @samp{-I-}, these directories are searched for all @samp{#include}
901: directives. (Ordinarily @emph{all} @samp{-I} directories are used
902: this way.)
903:
904: In addition, the @samp{-I-} option inhibits the use of the current
905: directory as the first search directory for @samp{#include
906: "@var{file}"}. Therefore, the current directory is searched only if
907: it is requested explicitly with @samp{-I.}. Specifying both
908: @samp{-I-} and @samp{-I.} allows you to control precisely which
909: directories are searched before the current one and which are searched
910: after.
911:
912: @item -nostdinc
913: Do not search the standard system directories for header files. Only
914: the directories you have specified with @samp{-I} options (and the
915: current directory, if appropriate) are searched.
916:
917: Between @samp{-nostdinc} and @samp{-I-}, you can eliminate all
918: directories from the search path except those you specify.
919:
1.1.1.2 root 920: @item -M
921: Tell the preprocessor to output a rule suitable for @code{make}
922: describing the dependencies of each source file. For each source
923: file, the preprocessor outputs one @code{make}-rule whose target is
924: the object file name for that source file and whose dependencies are
925: all the files @samp{#include}d in it. This rule may be a single line
926: or may be continued with @samp{\}-newline if it is long.
927:
928: @samp{-M} implies @samp{-E}.
929:
930: @item -MM
931: Like @samp{-M} but the output mentions only the user-header files
932: included with @samp{#include "@var{file}"}. System header files
933: included with @samp{#include <@var{file}>} are omitted.
934:
935: @samp{-MM} implies @samp{-E}.
936:
937: @item -D@var{macro}
938: Define macro @var{macro} with the empty string as its definition.
939:
940: @item -D@var{macro}=@var{defn}
941: Define macro @var{macro} as @var{defn}.
942:
943: @item -U@var{macro}
944: Undefine macro @var{macro}.
945:
946: @item -T
947: Support ANSI C trigraphs. You don't want to know about this
948: brain-damage. The @samp{-ansi} option also has this effect.
949: @end table
950:
1.1.1.3 root 951: @node Installation, Trouble, Options, Top
1.1.1.2 root 952: @chapter Installing GNU CC
953:
954: Here is the procedure for installing GNU CC on a Unix system.
955: @menu
956: * VMS Install:: See below for installation on VMS.
957: @end menu
958: @iftex
959: (See below for VMS.)
960: @end iftex
961:
962: @enumerate
963: @item
1.1.1.4 root 964: Edit @file{Makefile}. If you are using HPUX, or any form of system V,
965: you must make a few changes described in comments at the beginning of
966: the file.
967:
968: @item
969: On a Sequent system, go to the Berkeley universe.
1.1.1.2 root 970:
971: @item
972: Choose configuration files.
973:
974: @itemize @bullet
975: @item
976: Make a symbolic link named @file{config.h} to the top-level
977: config file for the machine you are using (@pxref{Config}). This
978: file is responsible for defining information about the host
979: machine. It includes @file{tm.h}.
980:
1.1.1.7 ! root 981: The file's name should be @file{config-@var{machine}.h}, with these
! 982: exceptions:
! 983:
! 984: @table @file
! 985: @item config-vms.h
! 986: for vaxen running VMS.
! 987: @item config-vaxv.h
! 988: for vaxen running system V.
! 989: @item config-i386v.h
! 990: for Intel 80386's running system V.
! 991: @item config-sun4.h
! 992: for Suns (model 3 or 4) running @emph{operating system} version 4.
! 993: @item config-hp9k3.h
! 994: for the HP 9000 series 300.
! 995: @item config-gnx.h
! 996: for the ns32000 running Genix
! 997: @end table
1.1.1.2 root 998:
999: If your system does not support symbolic links, you might want to
1000: set up @file{config.h} to contain a @samp{#include} command which
1001: refers to the appropriate file.
1002:
1003: @item
1004: Make a symbolic link named @file{tm.h} to the machine-description
1005: macro file for your machine (its name should be
1006: @file{tm-@var{machine}.h}).
1007:
1.1.1.7 ! root 1008: If your system is a 68000, don't use the file @file{tm-m68k.h}
! 1009: directly. Instead, use one of these files:
! 1010:
! 1011: @table @file
! 1012: @item tm-sun3.h
! 1013: for Sun 3 machines.
! 1014: @item tm-sun2.h
! 1015: for Sun 2 machines.
! 1016: @item tm-3b1.h
! 1017: for AT&T 3b1 (aka 7300 Unix PC).
! 1018: @item tm-isi68.h
! 1019: for Integrated Solutions systems.
! 1020: @item tm-news800.h
! 1021: for SONY News systems.
! 1022: @item tm-hp9k320.h
! 1023: for HPUX systems, if you are using GNU CC with the system's
! 1024: assembler and linker.
! 1025: @item tm-hp9k320g.h
! 1026: for HPUX systems, if you are using the GNU assembler, linker and
! 1027: other utilities. Not all of the pieces of GNU software needed
! 1028: for this mode of operation are as yet in distribution; full
! 1029: instructions will appear here in the future.@refill
! 1030: @end table
1.1.1.2 root 1031:
1.1.1.6 root 1032: For the vax, use @file{tm-vax.h} on BSD Unix, @file{tm-vaxv.h} on
1033: system V, or @file{tm-vms.h} on VMS.@refill
1.1.1.4 root 1034:
1.1.1.7 ! root 1035: For the SPARC (Sun 4), use @file{tm-sparc.h}.
1.1.1.6 root 1036:
1037: For the Motorola 88000, use @file{tm-m88k.h}. The support for the
1038: 88000 has a few unfinished spots because there was no way to run the
1039: output. Bugs are suspected in handling of branch-tables and in
1040: the function prologue and epilogue.
1041:
1.1.1.7 ! root 1042: For the 80386, don't use @file{tm-i386.h} directly. Use
! 1043: @file{tm-i386v.h} if the target machine is running system V,
! 1044: @file{tm-seq386.h} for a Sequent 386 system, or @file{tm-compaq.h} for
! 1045: a Compaq.
1.1.1.4 root 1046:
1.1.1.2 root 1047: For the 32000, use @file{tm-sequent.h} if you are using a Sequent
1.1.1.7 ! root 1048: machine, or @file{tm-encore.h} for an Encore machine, or
! 1049: @file{tm-gnx.h} if you are using Genix version 3; otherwise, perhaps
! 1050: @file{tm-ns32k.h} will work for you.
! 1051:
! 1052: Note that Genix has bugs in @code{alloca} and @code{malloc}; you must
! 1053: get the compiled versions of these from GNU Emacs and edit GNU CC's
! 1054: @file{Makefile} to use them.
1.1.1.4 root 1055:
1056: Note that Encore systems are supported only under BSD.
1.1.1.2 root 1057:
1058: @item
1059: Make a symbolic link named @file{md} to the machine description
1060: pattern file (its name should be @file{@var{machine}.md}).
1061:
1062: @item
1063: Make a symbolic link named @file{aux-output.c} to the output
1064: subroutine file for your machine (its name should be
1065: @file{output-@var{machine}.c}).
1066: @end itemize
1067:
1068: @item
1069: Make sure the Bison parser generator is installed. (This is
1.1.1.7 ! root 1070: unnecessary if the Bison output files @file{c-parse.tab.c} and
! 1071: @file{cexp.c} are more recent than @file{c-parse.y} and @file{cexp.y}
1.1.1.4 root 1072: and you do not plan to change the @samp{.y} files.)
1.1.1.2 root 1073:
1074: Note that if you have an old version of Bison you may get an error
1075: from the line with the @samp{%expect} directive. If so, simply remove
1.1.1.7 ! root 1076: that line from @file{c-parse.y} and proceed.
1.1.1.2 root 1077:
1078: @item
1079: If you are using a Sun, make sure the environment variable
1080: @code{FLOAT_OPTION} is not set. If this option were set to
1081: @code{f68881} when @file{gnulib} is compiled, the resulting code would
1082: demand to be linked with a special startup file and will not link
1083: properly without special pains.
1084:
1085: @item
1086: Build the compiler. Just type @samp{make} in the compiler directory.
1087:
1088: @item
1089: Move the first-stage object files and executables into a subdirectory
1090: with this command:
1091:
1092: @example
1093: make stage1
1094: @end example
1095:
1096: The files are moved into a subdirectory named @file{stage1}.
1097: Once installation is complete, you may wish to delete these files
1098: with @code{rm -r stage1}.
1099:
1100: @item
1101: Recompile the compiler with itself, with this command:
1102:
1103: @example
1104: make CC=stage1/gcc CFLAGS="-g -O -Bstage1/"
1105: @end example
1106:
1107: On a 68000 or 68020 system lacking floating point hardware,
1108: unless you have selected a @file{tm.h} file that expects by default
1109: that there is no such hardware, do this instead:
1110:
1111: @example
1112: make CC=stage1/gcc CFLAGS="-g -O -Bstage1/ -msoft-float"
1113: @end example
1114:
1115: @item
1116: If you wish to test the compiler by compiling it with itself one more
1117: time, do this:
1118:
1119: @example
1120: make stage2
1.1.1.6 root 1121: Make CC=stage2/gcc CFLAGS="-g -O -Bstage2/"
1.1.1.2 root 1122: foreach file (*.o)
1123: cmp $file stage2/$file
1124: end
1125: @end example
1126:
1127: This will notify you if any of these stage 3 object files differs from
1128: those of stage 2. Any difference, no matter how innocuous, indicates
1129: that the stage 2 compiler has compiled GNU CC incorrectly, and is
1130: therefore a potentially serious bug which you should investigate and
1131: report (@pxref{Bugs}).
1132:
1.1.1.6 root 1133: Aside from the @samp{-B} option, the options should be the same as
1134: when you made stage 2.
1135:
1.1.1.2 root 1136: @item
1137: Install the compiler driver, the compiler's passes and run-time support.
1138: You can use the following command:
1139:
1140: @example
1141: make install
1142: @end example
1143:
1144: @noindent
1145: This copies the files @file{cc1}, @file{cpp} and @file{gnulib} to
1146: files @file{gcc-cc1}, @file{gcc-cpp} and @file{gcc-gnulib} in
1147: directory @file{/usr/local/lib}, which is where the compiler driver
1148: program looks for them. It also copies the driver program @file{gcc}
1149: into the directory @file{/usr/local}, so that it appears in typical
1150: execution search paths.@refill
1151:
1152: @strong{Warning: the GNU CPP may not work for @file{ioctl.h},
1153: @file{ttychars.h} and other system header files unless the
1154: @samp{-traditional} option is used.} The bug is in the header files:
1155: at least on some machines, they rely on behavior that is incompatible
1156: with ANSI C. This behavior consists of substituting for macro
1157: argument names when they appear inside of character constants. The
1158: @samp{-traditional} option tells GNU CC to behave the way these
1159: headers expect.
1160:
1161: Because of this problem, you might prefer to configure GNU CC to use
1162: the system's own C preprocessor. To do so, make the file
1163: @file{/usr/local/lib/gcc-cpp} a link to @file{/lib/cpp}.
1164:
1165: Alternatively, on Sun systems and 4.3BSD at least, you can correct the
1166: include files by running the shell script @file{fixincludes}. This
1.1.1.5 root 1167: installs modified, corrected copies of the files @file{ioctl.h},
1168: @file{ttychars.h} and many others, in a special directory where only
1169: GNU CC will normally look for them.
1.1.1.2 root 1170:
1.1.1.5 root 1171: See the file @file{fixincludes} for a list of all the files we know to
1172: require correction.
1.1.1.2 root 1173: @end enumerate
1174:
1175: If you cannot install the compiler's passes and run-time support in
1176: @file{/usr/local/lib}, you can alternatively use the @samp{-B} option to
1177: specify a prefix by which they may be found. The compiler concatenates
1178: the prefix with the names @file{cpp}, @file{cc1} and @file{gnulib}.
1179: Thus, you can put the files in a directory @file{/usr/foo/gcc} and
1180: specify @samp{-B/usr/foo/gcc/} when you run GNU CC.
1181:
1.1.1.4 root 1182: Also, you can specify an alternative default directory for these files
1183: by setting the Make variable @code{libdir} when you make GNU CC.
1184:
1.1.1.2 root 1185: @node VMS Install,, Installation, Installation
1186: @section Installing GNU CC on VMS
1187:
1.1.1.3 root 1188: The VMS version of GNU CC is distributed in an unusual tape format which
1189: consists of several tape files. The first is a command file; the second is
1190: an executable program which reads Unix tar format; the third is another
1191: command file which uses this program to read the remainder of the tape.
1192:
1193: To load the tape, it suffices to mount it @samp{/foreign} and then do
1194: @samp{@@mta0:} to execute the command file at the beginning of the tape.
1195:
1196: The tape contains executables and object files as well as sources, so no
1197: compilation is necessary unless you change the sources. (This is a good
1198: thing, since you probably don't have any other C compiler.) If you must
1199: recompile, here is how:
1.1.1.2 root 1200:
1201: @enumerate
1202: @item
1203: Copy the file @file{tm-vms.h} to @file{tm.h}, @file{config-vms.h} to
1204: @file{config.h}, @file{vax.md} to @file{md.} and @file{output-vax.c}
1205: to @file{aux-output.c}.@refill
1206:
1207: @item
1208: Type @samp{@@make} to do recompile everything.
1209: @end enumerate
1210:
1.1.1.3 root 1211: To install the @samp{GCC} command so you can use the compiler easily, in
1212: the same manner as you use the VMS C compiler, you must install the VMS CLD
1213: file for GNU CC as follows:
1214:
1215: @enumerate
1216: @item
1217: Define the VMS logical names @samp{GNU_CC} and @samp{GNU_CC_INCLUDE}
1218: to point to the directories where the GNU CC executables
1219: (@samp{gcc-cpp}, @samp{gcc-cc1}, etc.) and the C include files are
1220: kept. This should be done with the commands:@refill
1221:
1222: @example
1223: $ assign /super /system disk:[gcc] gnu_cc
1224: $ assign /super /system disk:[gcc.include] gnu_cc_include
1225: @end example
1226:
1227: @noindent
1228: with the appropriate disk and directory names. These commands can be
1229: placed in your system startup file so they will be executed whenever
1230: the machine is rebooted.
1231:
1232: @item
1233: Install the @samp{GCC} command with the command line:
1234:
1235: @example
1236: $ set command /table=sys$library:dcltables gnu_cc:gcc
1237: @end example
1238:
1239: @noindent
1240: Now you can invoke the compiler with a command like @samp{gcc /verbose
1241: file.c}, which is equivalent to the command @samp{gcc -v -c file.c} in
1242: Unix.
1243: @end enumerate
1244:
1245: @node Trouble, Incompatibilities, Installation, Top
1246: @chapter Trouble in Installation
1247:
1248: Here are some of the things that have caused trouble for people installing
1249: GNU CC.
1250:
1.1.1.4 root 1251: @itemize @bullet
1.1.1.3 root 1252: @item
1253: On certain systems, defining certain environment variables such as
1254: @samp{CC} can interfere with the functioning of @code{make}.
1.1.1.4 root 1255:
1256: @item
1257: Cross compilation can run into trouble for certain machines because
1258: some target machines' assemblers require floating point numbers to be
1259: written as @emph{integer} constants in certain contexts.
1260:
1261: The compiler writes these integer constants by examining the floating
1262: point value as an integer and printing that integer, because this is
1263: simple to write and independent of the details of the floating point
1264: representation. But this does not work if the compiler is running on
1265: a different machine with an incompatible floating point format, or
1266: even a different byte-ordering.
1267:
1268: It is possible to fix this by writing machine-independent code which
1269: understands the floating point representation of the target machine.
1270: I am not interested in doing that much work to compensate for bugs
1271: in assemblers.
1.1.1.3 root 1272: @end itemize
1273:
1274: @node Incompatibilities, Extensions, Trouble, Top
1.1.1.2 root 1275: @chapter Incompatibilities of GNU CC
1276:
1277: There are several noteworthy incompatibilities between GNU C and most
1278: existing (non-ANSI) versions of C.
1279:
1280: Ultimately our intention is that the @samp{-traditional} option will
1281: eliminate most of these incompatibilities by telling GNU C to behave
1282: like the other C compilers.
1283:
1284: @itemize @bullet
1285: @item
1286: GNU CC normally makes string constants read-only. If several
1287: identical-looking string constants are used, GNU CC stores only one
1288: copy of the string.
1289:
1290: One consequence is that you cannot call @code{mktemp} with a string
1291: constant argument. The function @code{mktemp} always alters the
1292: string its argument points to.
1293:
1294: Another consequence is that @code{sscanf} does not work on some
1295: systems when passed a string constant as its format control string.
1296: This is because @code{sscanf} incorrectly tries to write into the
1297: string constant.
1298:
1299: The best solution to these problems is to change the program to use
1300: @code{char}-array variables with initialization strings for these
1301: purposes instead of string constants. But if this is not possible,
1302: you can use the @samp{-fwritable-strings} flag, which directs GNU CC
1303: to handle string constants the same way most C compilers do.
1304:
1305: @item
1306: GNU CC does not substitute macro arguments when they appear inside of
1307: string constants. For example, the following macro in GNU CC
1308:
1309: @example
1310: #define foo(a) "a"
1311: @end example
1312:
1313: @noindent
1314: will produce output @samp{"a"} regardless of what the argument @var{a} is.
1315:
1316: The @samp{-traditional} option directs GNU CC to handle such cases
1317: (among others) in the old-fashioned (non-ANSI) fashion.
1318:
1319: @item
1320: When you use @code{setjmp} and @code{longjmp}, the only automatic
1321: variables guaranteed to remain valid are those declared
1322: @code{volatile}. This is a consequence of automatic register
1323: allocation. Consider this function:
1324:
1325: @example
1326: jmp_buf j;
1327:
1328: foo ()
1329: @{
1330: int a, b;
1331:
1332: a = fun1 ();
1333: if (setjmp (j))
1334: return a;
1335:
1336: a = fun2 ();
1337: /* @r{@code{longjmp (j)} may be occur in @code{fun3}.} */
1338: return a + fun3 ();
1339: @}
1340: @end example
1341:
1342: Here @code{a} may or may not be restored to its first value when the
1343: @code{longjmp} occurs. If @code{a} is allocated in a register, then
1344: its first value is restored; otherwise, it keeps the last value stored
1345: in it.
1346:
1347: If you use the @samp{-W} option with the @samp{-O} option, you will
1348: get a warning when GNU CC thinks such a problem might be possible.
1349:
1350: @item
1351: Declarations of external variables and functions within a block apply
1352: only to the block containing the declaration. In other words, they
1353: have the same scope as any other declaration in the same place.
1354:
1355: In some other C compilers, a @code{extern} declaration affects all the
1356: rest of the file even if it happens within a block.
1357:
1358: The @samp{-traditional} option directs GNU C to treat all @code{extern}
1359: declarations as global, like traditional compilers.
1360:
1361: @item
1362: In traditional C, you can combine @code{long}, etc., with a typedef name,
1363: as shown here:
1364:
1365: @example
1366: typedef int foo;
1367: typedef long foo bar;
1368: @end example
1369:
1370: In ANSI C, this is not allowed: @code{long} and other type modifiers
1371: require an explicit @code{int}. Because this criterion is expressed
1372: by Bison grammar rules rather than C code, the @samp{-traditional}
1373: flag cannot alter it.
1374:
1375: @item
1.1.1.7 ! root 1376: PCC allows typedef names to be used as function parameters. The
! 1377: difficulty described immediately above applies here too.
! 1378:
! 1379: @item
1.1.1.6 root 1380: When compiling functions that return @code{float}, PCC converts it to
1381: a double. GNU CC actually returns a @code{float}. If you are concerned
1382: with PCC compatibility, you should declare your functions to return
1383: @code{double}; you might as well say what you mean.
1384:
1385: @item
1.1.1.2 root 1386: When compiling functions that return structures or unions, GNU CC
1387: output code uses a method different from that used on most versions of
1388: Unix. As a result, code compiled with GNU CC cannot call a
1389: structure-returning function compiled with PCC, and vice versa.
1390:
1391: The method used by GCC is as follows: a structure or union which is 1,
1392: 2, 4 or 8 bytes long is returned like a scalar. A structure or union
1393: with any other size is stored into an address supplied by the caller
1394: in a special, fixed register.
1395:
1396: PCC usually handles all sizes of structures and unions by returning
1397: the address of a block of static storage containing the value. This
1398: method is not used in GCC because it is slower and nonreentrant.
1399:
1400: On systems where PCC works this way, you may be able to make GCC-compiled
1401: code call such functions that were compiled with PCC by declaring them
1402: to return a pointer to the structure or union instead of the structure
1403: or union itself. For example, instead of this:
1404:
1405: @example
1406: struct foo nextfoo ();
1407: @end example
1408:
1409: @noindent
1410: write this:
1411:
1412: @example
1413: struct foo *nextfoo ();
1414: #define nextfoo *nextfoo
1415: @end example
1416:
1417: @noindent
1.1.1.4 root 1418: (Note that this assumes you are using the GNU preprocessor and not
1419: @samp{-traditional}, so that the ANSI antirecursion rules for macro
1420: expansions are effective.)
1.1.1.2 root 1421: @end itemize
1422:
1423: @node Extensions, Bugs, Incompatibilities, Top
1424: @chapter GNU Extensions to the C Language
1425:
1426: GNU C provides several language features not found in ANSI standard C.
1427: (The @samp{-pedantic} option directs GNU CC to print a warning message if
1428: any of these features is used.) To test for the availability of these
1429: features in conditional compilation, check for a predefined macro
1430: @code{__GNUC__}, which is always defined under GNU CC.
1431:
1432: @menu
1433: * Statement Exprs:: Putting statements and declarations inside expressions.
1434: * Naming Types:: Giving a name to the type of some expression.
1435: * Typeof:: @code{typeof}: referring to the type of an expression.
1436: * Lvalues:: Using @samp{?:}, @samp{,} and casts in lvalues.
1437: * Conditionals:: Omitting the middle operand of a @samp{?:} expression.
1438: * Zero-Length:: Zero-length arrays.
1439: * Variable-Length:: Arrays whose length is computed at run time.
1440: * Subscripting:: Any array can be subscripted, even if not an lvalue.
1441: * Pointer Arith:: Arithmetic on @code{void}-pointers and function pointers.
1442: * Constructors:: Constructor expressions give structures, unions
1443: or arrays as values.
1444: * Dollar Signs:: Dollar sign is allowed in identifiers.
1445: * Alignment:: Inquiring about the alignment of a type or variable.
1446: * Inline:: Defining inline functions (as fast as macros).
1447: * Extended Asm:: Assembler instructions with C expressions as operands.
1448: (With them you can define ``built-in'' functions.)
1449: * Asm Labels:: Specifying the assembler name to use for a C symbol.
1450: @end menu
1451:
1452: @node Statement Exprs, Naming Types, Extensions, Extensions
1453: @section Statements and Declarations inside of Expressions
1454:
1455: A compound statement in parentheses may appear inside an expression in GNU
1456: C. This allows you to declare variables within an expression. For
1457: example:
1458:
1459: @example
1460: (@{ int y = foo (); int z;
1461: if (y > 0) z = y;
1462: else z = - y;
1463: z; @})
1464: @end example
1465:
1466: @noindent
1467: is a valid (though slightly more complex than necessary) expression
1468: for the absolute value of @code{foo ()}.
1469:
1470: This feature is especially useful in making macro definitions ``safe'' (so
1471: that they evaluate each operand exactly once). For example, the
1472: ``maximum'' function is commonly defined as a macro in standard C as
1473: follows:
1474:
1475: @example
1476: #define max(a,b) ((a) > (b) ? (a) : (b))
1477: @end example
1478:
1479: @noindent
1480: But this definition computes either @var{a} or @var{b} twice, with bad
1481: results if the operand has side effects. In GNU C, if you know the
1482: type of the operands (here let's assume @code{int}), you can define
1483: the macro safely as follows:
1484:
1485: @example
1486: #define maxint(a,b) \
1487: (@{int _a = (a), _b = (b); _a > _b ? _a : _b; @})
1488: @end example
1489:
1490: Embedded statements are not allowed in constant expressions, such as
1491: the value of an enumeration constant, the width of a bit field, or
1492: the initial value of a static variable.
1493:
1494: If you don't know the type of the operand, you can still do this, but you
1495: must use @code{typeof} (@pxref{Typeof}) or type naming (@pxref{Naming
1496: Types}).
1497:
1498: @node Naming Types, Typeof, Statement Exprs, Extensions
1499: @section Naming an Expression's Type
1500:
1501: You can give a name to the type of an expression using a @code{typedef}
1502: declaration with an initializer. Here is how to define @var{name} as a
1503: type name for the type of @var{exp}:
1504:
1505: @example
1506: typedef @var{name} = @var{exp};
1507: @end example
1508:
1509: This is useful in conjunction with the statements-within-expressions
1510: feature. Here is how the two together can be used to define a safe
1511: ``maximum'' macro that operates on any arithmetic type:
1512:
1513: @example
1514: #define max(a,b) \
1515: (@{typedef _ta = (a), _tb = (b); \
1516: _ta _a = (a); _tb _b = (b); \
1517: _a > _b ? _a : _b; @})
1518: @end example
1519:
1520: The reason for using names that start with underscores for the local
1521: variables is to avoid conflicts with variable names that occur within the
1522: expressions that are substituted for @code{a} and @code{b}. Eventually we
1523: hope to design a new form of declaration syntax that allows you to declare
1524: variables whose scopes start only after their initializers; this will be a
1525: more reliable way to prevent such conflicts.
1526:
1527: @node Typeof, Lvalues, Naming Types, Extensions
1528: @section Referring to a Type with @code{typeof}
1529:
1530: Another way to refer to the type of an expression is with @code{typeof}.
1531: The syntax of using of this keyword looks like @code{sizeof}, but the
1532: construct acts semantically like a type name defined with @code{typedef}.
1533:
1534: There are two ways of writing the argument to @code{typeof}: with an
1535: expression or with a type. Here is an example with an expression:
1536:
1537: @example
1538: typeof (x[0](1))
1539: @end example
1540:
1541: @noindent
1542: This assumes that @code{x} is an array of functions; the type described
1543: is that of the values of the functions.
1544:
1545: Here is an example with a typename as the argument:
1546:
1547: @example
1548: typeof (int *)
1549: @end example
1550:
1551: @noindent
1552: Here the type described is that of pointers to @code{int}.
1553:
1554: A @code{typeof}-construct can be used anywhere a typedef name could be
1555: used. For example, you can use it in a declaration, in a cast, or inside
1556: of @code{sizeof} or @code{typeof}.
1557:
1558: @itemize @bullet
1559: @item
1560: This declares @code{y} with the type of what @code{x} points to.
1561:
1562: @example
1563: typeof (*x) y;
1564: @end example
1565:
1566: @item
1567: This declares @code{y} as an array of such values.
1568:
1569: @example
1570: typeof (*x) y[4];
1571: @end example
1572:
1573: @item
1574: This declares @code{y} as an array of pointers to characters:
1575:
1576: @example
1577: typeof (typeof (char *)[4]) y;
1578: @end example
1579:
1580: @noindent
1581: It is equivalent to the following traditional C declaration:
1582:
1583: @example
1584: char *y[4];
1585: @end example
1586:
1587: To see the meaning of the declaration using @code{typeof}, and why it
1588: might be a useful way to write, let's rewrite it with these macros:
1589:
1590: @example
1591: #define pointer(T) typeof(T *)
1592: #define array(T, N) typeof(T [N])
1593: @end example
1594:
1595: @noindent
1596: Now the declaration can be rewritten this way:
1597:
1598: @example
1599: array (pointer (char), 4) y;
1600: @end example
1601:
1602: @noindent
1603: Thus, @samp{array (pointer (char), 4)} is the type of arrays of 4
1604: pointers to @code{char}.
1605: @end itemize
1606:
1607: @node Lvalues, Conditionals, Typeof, Extensions
1608: @section Generalized Lvalues
1609:
1610: Compound expressions, conditional expressions and casts are allowed as
1611: lvalues provided their operands are lvalues. This means that you can take
1612: their addresses or store values into them.
1613:
1614: For example, a compound expression can be assigned, provided the last
1615: expression in the sequence is an lvalue. These two expressions are
1616: equivalent:
1617:
1618: @example
1619: (a, b) += 5
1620: a, (b += 5)
1621: @end example
1622:
1623: Similarly, the address of the compound expression can be taken. These two
1624: expressions are equivalent:
1625:
1626: @example
1627: &(a, b)
1628: a, &b
1629: @end example
1630:
1631: A conditional expression is a valid lvalue if its type is not void and the
1632: true and false branches are both valid lvalues. For example, these two
1633: expressions are equivalent:
1634:
1635: @example
1636: (a ? b : c) = 5
1637: (a ? b = 5 : (c = 5))
1638: @end example
1639:
1640: A cast is a valid lvalue if its operand is valid. Taking the address of
1641: the cast is the same as taking the address without a cast, except for the
1642: type of the result. For example, these two expressions are equivalent (but
1643: the second may be valid when the type of @samp{a} does not permit a cast to
1644: @samp{int *}).
1645:
1646: @example
1647: &(int *)a
1648: (int **)&a
1649: @end example
1650:
1651: A simple assignment whose left-hand side is a cast works by converting the
1652: right-hand side first to the specified type, then to the type of the inner
1653: left-hand side expression. After this is stored, the value is converter
1654: back to the specified type to become the value of the assignment. Thus, if
1655: @samp{a} has type @samp{char *}, the following two expressions are
1656: equivalent:
1657:
1658: @example
1659: (int)a = 5
1660: (int)(a = (char *)5)
1661: @end example
1662:
1663: An assignment-with-arithmetic operation such as @samp{+=} applied to a cast
1664: performs the arithmetic using the type resulting from the cast, and then
1665: continues as in the previous case. Therefore, these two expressions are
1666: equivalent:
1667:
1668: @example
1669: (int)a += 5
1670: (int)(a = (char *) ((int)a + 5))
1671: @end example
1672:
1673: @node Conditionals, Zero-Length, Lvalues, Extensions
1674: @section Conditional Expressions with Omitted Middle-Operands
1675:
1676: The middle operand in a conditional expression may be omitted. Then
1677: if the first operand is nonzero, its value is the value of the conditional
1678: expression.
1679:
1680: Therefore, the expression
1681:
1682: @example
1683: x ? : y
1684: @end example
1685:
1686: @noindent
1687: has the value of @code{x} if that is nonzero; otherwise, the value of
1688: @code{y}.
1689:
1690: This example is perfectly equivalent to
1691:
1692: @example
1693: x ? x : y
1694: @end example
1695:
1696: @noindent
1697: In this simple case, the ability to omit the middle operand is not
1698: especially useful. When it becomes useful is when the first operand does,
1699: or may (if it is a macro argument), contain a side effect. Then repeating
1700: the operand in the middle would perform the side effect twice. Omitting
1701: the middle operand uses the value already computed without the undesirable
1702: effects of recomputing it.
1703:
1704: @node Zero-Length, Variable-Length, Conditionals, Extensions
1705: @section Arrays of Length Zero
1706:
1707: Zero-length arrays are allowed in GNU C. They are very useful as the last
1708: element of a structure which is really a header for a variable-length
1709: object:
1710:
1711: @example
1712: struct line @{
1713: int length;
1714: char contents[0];
1715: @};
1716:
1717: @{
1718: struct line *thisline
1719: = (struct line *) malloc (sizeof (struct line) + this_length);
1.1.1.6 root 1720: thisline->length = this_length;
1.1.1.2 root 1721: @}
1722: @end example
1723:
1724: In standard C, you would have to give @code{contents} a length of 1, which
1725: means either you waste space or complicate the argument to @code{malloc}.
1726:
1727: @node Variable-Length, Subscripting, Zero-Length, Extensions
1728: @section Arrays of Variable Length
1729:
1730: Variable-length automatic arrays are allowed in GNU C. These arrays are
1731: declared like any other automatic arrays, but with a length that is not a
1732: constant expression. The storage is allocated at that time and
1733: deallocated when the brace-level is exited. For example:
1734:
1735: @example
1736: FILE *concat_fopen (char *s1, char *s2, char *mode)
1737: @{
1738: char str[strlen (s1) + strlen (s2) + 1];
1739: strcpy (str, s1);
1740: strcat (str, s2);
1741: return fopen (str, mode);
1742: @}
1743: @end example
1744:
1745: You can also define structure types containing variable-length arrays, and
1746: use them even for arguments or function values, as shown here:
1747:
1748: @example
1749: int foo;
1750:
1751: struct entry
1752: @{
1753: char data[foo];
1754: @};
1755:
1756: struct entry
1757: tester (struct entry arg)
1758: @{
1759: struct entry new;
1760: int i;
1761: for (i = 0; i < foo; i++)
1762: new.data[i] = arg.data[i] + 1;
1763: return new;
1764: @}
1765: @end example
1766:
1767: @noindent
1768: (Eventually there will be a way to say that the size of the array is
1769: another member of the same structure.)
1770:
1771: The length of an array is computed on entry to the brace-level where the
1772: array is declared and is remembered for the scope of the array in case you
1773: access it with @code{sizeof}.
1774:
1775: Jumping or breaking out of the scope of the array name will also deallocate
1776: the storage. Jumping into the scope is not allowed; you will get an error
1777: message for it.
1778:
1779: You can use the function @code{alloca} to get an effect much like
1780: variable-length arrays. The function @code{alloca} is available in
1781: many other C implementations (but not in all). On the other hand,
1782: variable-length arrays are more elegant.
1783:
1784: There are other differences between these two methods. Space allocated
1785: with @code{alloca} exists until the containing @emph{function} returns.
1786: The space for a variable-length array is deallocated as soon as the array
1787: name's scope ends. (If you use both variable-length arrays and
1788: @code{alloca} in the same function, deallocation of a variable-length array
1789: will also deallocate anything more recently allocated with @code{alloca}.)
1790:
1791: @node Subscripting, Pointer Arith, Variable-Length, Extensions
1792: @section Non-Lvalue Arrays May Have Subscripts
1793:
1794: Subscripting is allowed on arrays that are not lvalues, even though the
1795: unary @samp{&} operator is not. For example, this is valid in GNU C though
1796: not valid in other C dialects:
1797:
1798: @example
1799: struct foo @{int a[4];@};
1800:
1801: struct foo f();
1802:
1803: bar (int index)
1804: @{
1805: return f().a[index];
1806: @}
1807: @end example
1808:
1809: @node Pointer Arith, Initializers, Subscripting, Extensions
1810: @section Arithmetic on @code{void}-Pointers and Function Pointers
1811:
1812: In GNU C, addition and subtraction operations are supported on pointers to
1813: @code{void} and on pointers to functions. This is done by treating the
1814: size of a @code{void} or of a function as 1.
1815:
1816: A consequence of this is that @code{sizeof} is also allowed on @code{void}
1817: and on function types, and returns 1.
1818:
1819: @node Initializers, Constructors, Pointer Arith, Extensions
1820: @section Non-Constant Initializers
1821:
1822: The elements of an aggregate initializer are not required to be constant
1823: expressions in GNU C. Here is an example of an initializer with run-time
1824: varying elements:
1825:
1826: @example
1827: foo (float f, float g)
1828: @{
1829: float beat_freqs[2] = @{ f-g, f+g @};
1830: @dots{}
1831: @}
1832: @end example
1833:
1834: @node Constructors, Dollar Signs, Initializers, Extensions
1835: @section Constructor Expressions
1836:
1837: GNU C supports constructor expressions. A constructor looks like a cast
1838: containing an initializer. Its value is an object of the type specified in
1839: the cast, containing the elements specified in the initializer. The type
1840: must be a structure, union or array type.
1841:
1842: Assume that @code{struct foo} and @code{structure} are declared as shown:
1843:
1844: @example
1845: struct foo @{int a; char b[2];@} structure;
1846: @end example
1847:
1848: @noindent
1849: Here is an example of constructing a @samp{struct foo} with a constructor:
1850:
1851: @example
1852: structure = ((struct foo) @{x + y, 'a', 0@});
1853: @end example
1854:
1855: @noindent
1856: This is equivalent to writing the following:
1857:
1858: @example
1859: @{
1860: struct foo temp = @{x + y, 'a', 0@};
1861: structure = temp;
1862: @}
1863: @end example
1864:
1865: You can also construct an array. If all the elements of the constructor
1866: are (made up of) simple constant expressions, suitable for use in
1867: initializers, then the constructor is an lvalue and can be coerced to a
1868: pointer to its first element, as shown here:
1869:
1870: @example
1871: char **foo = (char *[]) @{ "x", "y", "z" @};
1872: @end example
1873:
1874: Array constructors whose elements are not simple constants are not very
1875: useful, because the constructor is not an lvalue. There are only two valid
1876: ways to use it: to subscript it, or initialize an array variable with it.
1877: The former is probably slower than a @code{switch} statement, while the
1878: latter does the same thing an ordinary C initializer would do.
1879:
1880: @example
1881: output = ((int[]) @{ 2, x, 28 @}) [input];
1882: @end example
1883:
1884: @node Dollar Signs, Alignment, Constructors, Extensions
1885: @section Dollar Signs in Identifier Names
1886:
1887: In GNU C, you may use dollar signs in identifier names. This is because
1888: many traditional C implementations allow such identifiers.
1889:
1890: @node Alignment, Inline, Dollar Signs, Extensions
1891: @section Inquiring about the Alignment of a Type or Variable
1892:
1893: The keyword @code{__alignof} allows you to inquire about how an object
1894: is aligned, or the minimum alignment usually required by a type. Its
1895: syntax is just like @code{sizeof}.
1896:
1897: For example, if the target machine requires a @code{double} value to be
1898: aligned on an 8-byte boundary, then @code{__alignof (double)} is 8. This
1899: is true on many RISC machines. On more traditional machine designs,
1900: @code{__alignof (double)} is 4 or even 2.
1901:
1902: Some machines never actually require alignment; they allow reference to any
1903: data type even at an odd addresses. For these machines, @code{__alignof}
1904: reports the @emph{recommended} alignment of a type.
1905:
1906: When the operand of @code{__alignof} is an lvalue rather than a type, the
1907: value is the largest alignment that the lvalue is known to have. It may
1908: have this alignment as a result of its data type, or because it is part of
1909: a structure and inherits alignment from that structure. For example, after
1910: this declaration:
1911:
1912: @example
1913: struct foo @{ int x; char y; @} foo1;
1914: @end example
1915:
1916: @noindent
1917: the value of @code{__alignof (foo1.y)} is probably 2 or 4, the same as
1918: @code{__alignof (int)}, even though the data type of @code{foo1.y} does not
1919: itself demand any alignment.@refill
1920:
1921: @node Inline, Extended Asm, Alignment, Extensions
1922: @section An Inline Function is As Fast As a Macro
1923:
1924: By declaring a function @code{inline}, you can direct GNU CC to integrate
1925: that function's code into the code for its callers. This makes execution
1926: faster by eliminating the function-call overhead; in addition, if any of
1927: the actual argument values are constant, their known values may permit
1928: simplifications at compile time so that not all of the inline function's
1929: code needs to be included.
1930:
1931: To declare a function inline, use the @code{inline} keyword in its
1932: declaration, like this:
1933:
1934: @example
1935: inline int
1936: inc (int *a)
1937: @{
1938: (*a)++;
1939: @}
1940: @end example
1941:
1942: You can also make all ``simple enough'' functions inline with the
1943: option @samp{-finline-functions}. Note that certain usages in a
1944: function definition can make it unsuitable for inline substitution.
1945:
1946: When a function is both inline and @code{static}, if all calls to the
1947: function are integrated into the caller, then the function's own assembler
1948: code is never referenced. In this case, GNU CC does not actually output
1949: assembler code for the function, unless you specify the option
1950: @samp{-fkeep-inline-functions}. Some calls cannot be integrated for
1951: various reasons (in particular, calls that precede the function's
1952: definition cannot be integrated, and neither can recursive calls within the
1953: definition). If there is a nonintegrated call, then the function is
1954: compiled to assembler code as usual.
1955:
1956: When an inline function is not @code{static}, then the compiler must assume
1957: that there may be calls from other source files; since a global symbol can
1958: be defined only once in any program, the function must not be defined in
1959: the other source files, so the calls therein cannot be integrated.
1960: Therefore, a non-@code{static} inline function is always compiled on its
1961: own in the usual fashion.
1962:
1963: @node Extended Asm, Asm Labels, Inline, Extensions
1964: @section Assembler Instructions with C Expression Operands
1965:
1966: In an assembler instruction using @code{asm}, you can now specify the
1967: operands of the instruction using C expressions. This means no more
1968: guessing which registers or memory locations will contain the data you want
1969: to use.
1970:
1971: You must specify an assembler instruction template much like what appears
1972: in a machine description, plus an operand constraint string for each
1973: operand.
1974:
1975: For example, here is how to use the 68881's @code{fsinx} instruction:
1976:
1977: @example
1978: asm ("fsinx %1,%0" : "=f" (result) : "f" (angle));
1979: @end example
1980:
1981: @noindent
1982: Here @code{angle} is the C expression for the input operand while
1983: @code{result} is that of the output operand. Each has @samp{"f"} as its
1984: operand constraint, saying that a floating-point register is required. The
1985: constraints use the same language used in the machine description
1986: (@pxref{Constraints}).
1987:
1988: Each operand is described by an operand-constraint string followed by the C
1989: expression in parentheses. A colon separates the assembler template from
1990: the first output operand, and another separates the last output operand
1991: from the first input, if any. Commas separate output operands and separate
1992: inputs. The number of operands is limited to the maximum number of
1993: operands in any instruction pattern in the machine description.
1994:
1.1.1.5 root 1995: Output operand expressions must be lvalues; the compiler can check this.
1996: The input operands need not be lvalues. The compiler cannot check whether
1997: the operands have data types that are reasonable for the instruction being
1.1.1.2 root 1998: executed.
1999:
1.1.1.5 root 2000: If there are no output operands, and there are input operands, then you
2001: should write two colons in a row where the output operands would go.
2002:
1.1.1.2 root 2003: The output operands must be write-only; GNU CC will assume that the values
2004: in these operands before the instruction are dead and need not be
1.1.1.5 root 2005: generated. For an operand that is read-write, or in which not all bits are
2006: written and the other bits contain useful information, you must logically
2007: split its function into two separate operands, one input operand and one
2008: write-only output operand. The connection between them is expressed by
2009: constraints which say they need to be in the same location when the
2010: instruction executes. You can use the same C expression for both operands,
2011: or different expressions. For example, here we write the (fictitious)
1.1.1.2 root 2012: @samp{combine} instruction with @code{bar} as its read-only source operand
2013: and @code{foo} as its read-write destination:
2014:
2015: @example
2016: asm ("combine %2,%0" : "=r" (foo) : "0" (foo), "g" (bar));
2017: @end example
2018:
2019: @noindent
2020: The constraint @samp{"0"} for operand 1 says that it must occupy the same
1.1.1.5 root 2021: location as operand 0.
2022:
2023: Unless an output operand has the @samp{&} constraint modifier, GNU CC may
2024: allocate it in the same register as an unrelated input operand, on the
2025: assumption that the inputs are consumed before the outputs are produced.
2026: This assumption may be false if the assembler code actually consists of
2027: more than one instruction. In such a case, use @samp{&} for each output
2028: operand that may not overlap an input. @xref{Modifiers}.
1.1.1.2 root 2029:
2030: Usually the most convenient way to use these @code{asm} instructions is to
2031: encapsulate them in macros that look like functions. For example,
2032:
2033: @example
2034: #define sin(x) \
2035: (@{ double __value, __arg = (x); \
2036: asm ("fsinx %1,%0": "=f" (__value): "f" (__arg)); \
2037: __value; @})
2038: @end example
2039:
2040: @noindent
2041: Here the variable @code{__arg} is used to make sure that the instruction
2042: operates on a proper @code{double} value, and to accept only those
2043: arguments @code{x} which can convert automatically to a @code{double}.
2044:
2045: Another way to make sure the instruction operates on the correct data type
2046: is to use a cast in the @code{asm}. This is different from using a
2047: variable @code{__arg} in that it converts more different types. For
2048: example, if the desired type were @code{int}, casting the argument to
2049: @code{int} would accept a pointer with no complaint, while assigning the
2050: argument to an @code{int} variable named @code{__arg} would warn about
2051: using a pointer unless the caller explicitly casts it.
2052:
2053: GNU CC assumes for optimization purposes that these instructions have no
2054: side effects except to change the output operands. This does not mean that
2055: instructions with a side effect cannot be used, but you must be careful,
2056: because the compiler may eliminate them if the output operands aren't used,
2057: or move them out of loops, or replace two with one if they constitute a
2058: common subexpression. Also, if your instruction does have a side effect on
2059: a variable that otherwise appears not to change, the old value of the
2060: variable may be reused later if it happens to be found in a register.
2061:
2062: You can prevent an @code{asm} instruction from being deleted, moved or
2063: combined by writing the keyword @code{volatile} after the @code{asm}. For
2064: example:
1.1 root 2065:
1.1.1.2 root 2066: @example
2067: #define set_priority(x) \
2068: asm volatile ("set_priority %1": \
2069: "=m" (*(char *)0): "g" (x))
2070: @end example
1.1 root 2071:
1.1.1.2 root 2072: @noindent
2073: Note that we have supplied an output operand which is not actually used in
2074: the instruction. This is because @code{asm} requires at least one output
2075: operand. This requirement exists for internal implementation reasons and
2076: we might be able to relax it in the future.
2077:
2078: In this case output operand has the additional benefit effect of giving the
2079: appearance of writing in memory. As a result, GNU CC will assume that data
2080: previously fetched from memory must be fetched again if needed again later.
2081: This may be desirable if you have not employed the @code{volatile} keyword
2082: on all the variable declarations that ought to have it.
2083:
2084: @node Asm Labels,,Extended Asm, Extensions
2085: @section Controlling Names Used in Assembler Code
2086:
2087: You can specify the name to be used in the assembler code for a C function
2088: or variable by writing the @code{asm} keyword after the declarator as
2089: follows:
1.1 root 2090:
1.1.1.2 root 2091: @example
2092: int foo asm ("myfoo") = 2;
2093: @end example
1.1 root 2094:
1.1.1.2 root 2095: @noindent
2096: This specifies that the name to be used for the variable @code{foo} in
2097: the assembler code should be @samp{myfoo} rather than the usual
2098: @samp{_foo}.
2099:
2100: On systems where an underscore is normally prepended to the name of a C
2101: function or variable, this feature allows you to define names for the
2102: linker that do not start with an underscore.
2103:
2104: You cannot use @code{asm} in this way in a function @emph{definition}; but
2105: you can get the same effect by writing a declaration for the function
2106: before its definition and putting @code{asm} there, like this:
1.1 root 2107:
1.1.1.2 root 2108: @example
2109: extern func () asm ("FUNC");
1.1 root 2110:
1.1.1.2 root 2111: func (x, y)
2112: int x, y;
2113: @dots{}
2114: @end example
1.1 root 2115:
1.1.1.2 root 2116: It is up to you to make sure that the assembler names you choose do not
2117: conflict with any other assembler symbols. Also, you must not use a
2118: register name; that would produce completely invalid assembler code. GNU
2119: CC does not as yet have the ability to store static variables in registers.
2120: Perhaps that will be added.
2121:
2122: @node Bugs, Portability, Extensions, Top
2123: @chapter Reporting Bugs
2124:
2125: Your bug reports play an essential role in making GNU CC reliable.
2126:
2127: Reporting a bug may help you by bringing a solution to your problem, or it
2128: may not. But in any case the important function of a bug report is to help
2129: the entire community by making the next version of GNU CC work better. Bug
2130: reports are your contribution to the maintenance of GNU CC.
1.1 root 2131:
1.1.1.2 root 2132: In order for a bug report to serve its purpose, you must include the
2133: information that makes for fixing the bug.
1.1 root 2134:
1.1.1.2 root 2135: @menu
2136: * Criteria: Bug Criteria. Have you really found a bug?
2137: * Reporting: Bug Reporting. How to report a bug effectively.
2138: @end menu
1.1 root 2139:
1.1.1.2 root 2140: @node Bug Criteria, Bug Reporting, Bugs, Bugs
2141: @section Have You Found a Bug?
2142:
2143: If you are not sure whether you have found a bug, here are some guidelines:
1.1 root 2144:
2145: @itemize @bullet
2146: @item
1.1.1.2 root 2147: If the compiler gets a fatal signal, for any input whatever, that is a
2148: compiler bug. Reliable compilers never crash.
1.1 root 2149:
2150: @item
1.1.1.2 root 2151: If the compiler produces invalid assembly code, for any input whatever
2152: (except an @code{asm} statement), that is a compiler bug, unless the
2153: compiler reports errors (not just warnings) which would ordinarily
2154: prevent the assembler from being run.
1.1 root 2155:
2156: @item
1.1.1.2 root 2157: If the compiler produces valid assembly code that does not correctly
2158: execute the input source code, that is a compiler bug.
1.1 root 2159:
1.1.1.2 root 2160: However, you must double-check to make sure, because you may have run
2161: into an incompatibility between GNU C and traditional C
2162: (@pxref{Incompatibilities}). These incompatibilities might be considered
1.1.1.3 root 2163: bugs, but they are inescapable consequences of valuable features.
1.1 root 2164:
1.1.1.2 root 2165: Or you may have a program whose behavior is undefined, which happened
2166: by chance to give the desired results with another C compiler.
1.1 root 2167:
1.1.1.2 root 2168: For example, in many nonoptimizing compilers, you can write @samp{x;}
2169: at the end of a function instead of @samp{return x;}, with the same
2170: results. But the value of the function is undefined if @samp{return}
2171: is omitted; it is not a bug when GNU CC produces different results.
1.1 root 2172:
1.1.1.2 root 2173: Problems often result from expressions with two increment operators,
2174: as in @samp{f (*p++, *p++)}. Your previous compiler might have
2175: interpreted that expression the way you intended; GNU CC might
2176: interpret it another way; neither compiler is wrong.
1.1 root 2177:
1.1.1.2 root 2178: After you have localized the error to a single source line, it should
2179: be easy to check for these things. If your program is correct and
2180: well defined, you have found a compiler bug.
1.1 root 2181:
1.1.1.2 root 2182: @item
2183: If the compiler produces an error message for valid input, that is a
2184: compiler bug.
1.1 root 2185:
1.1.1.2 root 2186: Note that the following is not valid input, and the error message for
2187: it is not a bug:
1.1 root 2188:
1.1.1.2 root 2189: @example
2190: int foo (char);
1.1 root 2191:
1.1.1.2 root 2192: int
2193: foo (x)
2194: char x;
2195: @{ @dots{} @}
2196: @end example
1.1 root 2197:
1.1.1.2 root 2198: @noindent
2199: The prototype says to pass a @code{char}, while the definition says to
2200: pass an @code{int} and treat the value as a @code{char}. This is what
2201: the ANSI standard says, and it makes sense.
1.1 root 2202:
1.1.1.2 root 2203: @item
2204: If the compiler does not produce an error message for invalid input,
2205: that is a compiler bug. However, you should note that your idea of
2206: ``invalid input'' might be my idea of ``an extension'' or ``support
2207: for traditional practice''.
1.1 root 2208:
1.1.1.2 root 2209: @item
2210: If you are an experienced user of C compilers, your suggestions
2211: for improvement of GNU CC are welcome in any case.
2212: @end itemize
1.1 root 2213:
1.1.1.2 root 2214: @node Bug Reporting,, Bug Criteria, Bugs
2215: @section How to Report Bugs
1.1 root 2216:
1.1.1.2 root 2217: Send bug reports for GNU C to one of these addresses:
1.1 root 2218:
1.1.1.2 root 2219: @example
2220: bug-gcc@@prep.ai.mit.edu
2221: @{ucbvax|mit-eddie|uunet@}!prep.ai.mit.edu!bug-gcc
2222: @end example
1.1 root 2223:
1.1.1.2 root 2224: As a last resort, snail them to:
1.1 root 2225:
1.1.1.2 root 2226: @example
2227: GNU Compiler Bugs
2228: 545 Tech Sq
2229: Cambridge, MA 02139
2230: @end example
1.1 root 2231:
1.1.1.2 root 2232: The fundamental principle of reporting bugs usefully is this:
2233: @strong{report all the facts}. If you are not sure whether to mention a
2234: fact or leave it out, mention it!
2235:
2236: Often people omit facts because they think they know what causes the
2237: problem and they conclude that some details don't matter. Thus, you might
2238: assume that the name of the variable you use in an example does not matter.
2239: Well, probably it doesn't, but one cannot be sure. Perhaps the bug is a
2240: stray memory reference which happens to fetch from the location where that
2241: name is stored in memory; perhaps, if the name were different, the contents
2242: of that location would fool the compiler into doing the right thing despite
2243: the bug. Play it safe and give an exact example.
1.1 root 2244:
1.1.1.2 root 2245: If you want to enable me to fix the bug, you should include all these
2246: things:
1.1 root 2247:
1.1.1.2 root 2248: @itemize @bullet
2249: @item
2250: The version of GNU CC. You can get this by running it with the
2251: @samp{-v} option.
1.1 root 2252:
1.1.1.2 root 2253: Without this, I won't know whether there is any point in looking for
2254: the bug in the current version of GNU CC.
1.1 root 2255:
1.1.1.2 root 2256: @item
2257: A complete input file that will reproduce the bug. If the bug is in
2258: the C preprocessor, send me a source file and any header files that it
2259: requires. If the bug is in the compiler proper (@file{cc1}), run your
2260: source file through the C preprocessor by doing @samp{gcc -E
2261: @var{sourcefile} > @var{outfile}}, then include the contents of
2262: @var{outfile} in the bug report. (Any @samp{-I}, @samp{-D} or
2263: @samp{-U} options that you used in actual compilation should also be
2264: used when doing this.)
1.1 root 2265:
1.1.1.2 root 2266: A single statement is not enough of an example. In order to compile
2267: it, it must be embedded in a function definition; and the bug might
2268: depend on the details of how this is done.
2269:
2270: Without a real example I can compile, all I can do about your bug
2271: report is wish you luck. It would be futile to try to guess how to
2272: provoke the bug. For example, bugs in register allocation and
2273: reloading frequently depend on every little detail of the function
2274: they happen in.
1.1 root 2275:
2276: @item
1.1.1.2 root 2277: The command arguments you gave GNU CC to compile that example and
2278: observe the bug. For example, did you use @samp{-O}? To guarantee
2279: you won't omit something important, list them all.
2280:
2281: If I were to try to guess the arguments, I would probably guess wrong
2282: and then I would not encounter the bug.
1.1 root 2283:
2284: @item
1.1.1.2 root 2285: The names of the files that you used for @file{tm.h} and @file{md}
2286: when you installed the compiler.
1.1 root 2287:
2288: @item
1.1.1.2 root 2289: The type of machine you are using, and the operating system name and
2290: version number.
1.1 root 2291:
2292: @item
1.1.1.2 root 2293: A description of what behavior you observe that you believe is
2294: incorrect. For example, ``It gets a fatal signal,'' or, ``There is an
2295: incorrect assembler instruction in the output.''
2296:
2297: Of course, if the bug is that the compiler gets a fatal signal, then I
2298: will certainly notice it. But if the bug is incorrect output, I might
2299: not notice unless it is glaringly wrong. I won't study all the
2300: assembler code from a 50-line C program just on the off chance that it
2301: might be wrong.
2302:
2303: Even if the problem you experience is a fatal signal, you should still
2304: say so explicitly. Suppose something strange is going on, such as,
2305: your copy of the compiler is out of synch, or you have encountered a
2306: bug in the C library on your system. (This has happened!) Your copy
2307: might crash and mine would not. If you @i{told} me to expect a crash,
2308: then when mine fails to crash, I would know that the bug was not
2309: happening for me. If you had not told me to expect a crash, then I
2310: would not be able to draw any conclusion from my observations.
2311:
2312: In cases where GNU CC generates incorrect code, if you send me a small
2313: complete sample program I will find the error myself by running the
2314: program under a debugger. If you send me a large example or a part of
2315: a larger program, I cannot do this; you must debug the compiled
2316: program and narrow the problem down to one source line. Tell me which
2317: source line it is, and what you believe is incorrect about the code
2318: generated for that line.
1.1 root 2319:
2320: @item
1.1.1.2 root 2321: If you send me examples of output from GNU CC, please use @samp{-g}
2322: when you make them. The debugging information includes source line
2323: numbers which are essential for correlating the output with the input.
1.1.1.4 root 2324:
2325: @item
2326: If you wish to suggest changes to the GNU CC source, send me context
2327: diffs. If you even discuss something in the GNU CC source, refer to
2328: it by context, not by line number.
2329:
2330: The line numbers in my development sources don't match those in your
2331: sources. They won't tell me anything.
1.1 root 2332: @end itemize
2333:
1.1.1.2 root 2334: Here are some things that are not necessary:
1.1 root 2335:
1.1.1.2 root 2336: @itemize @bullet
1.1 root 2337: @item
1.1.1.2 root 2338: A description of the envelope of the bug.
1.1 root 2339:
1.1.1.2 root 2340: Often people who encounter a bug spend a lot of time investigating
2341: which changes to the input file will make the bug go away and which
2342: changes will not affect it.
1.1 root 2343:
1.1.1.2 root 2344: This is often time consuming and not very useful, because the way I
2345: will find the bug is by running a single example under the debugger
2346: with breakpoints, not by pure deduction from a series of examples.
1.1 root 2347:
1.1.1.2 root 2348: Of course, it can't hurt if you can find a simpler example that
2349: triggers the same bug. Errors in the output will be easier to spot,
2350: running under the debugger will take less time, etc. An easy way
2351: to simplify an example is to delete all the function definitions
2352: except the one where the bug occurs. Those earlier in the file
2353: may be replaced by external declarations.
2354:
2355: However, simplification is not necessary; if you don't want to do
2356: this, report the bug anyway.
1.1 root 2357:
2358: @item
1.1.1.2 root 2359: A patch for the bug.
1.1 root 2360:
1.1.1.2 root 2361: A patch for the bug does help me if it is a good one. But don't omit
2362: the necessary information, such as the test case, because I might see
2363: problems with your patch and decide to fix the problem another way.
1.1 root 2364:
1.1.1.2 root 2365: Sometimes with a program as complicated as GNU CC it is very hard to
2366: construct an example that will make the program go through a certain
2367: point in the code. If you don't send me the example, I won't be able
2368: to verify that the bug is fixed.
1.1 root 2369:
2370: @item
1.1.1.2 root 2371: A guess about what the bug is or what it depends on.
2372:
2373: Such guesses are usually wrong. Even I can't guess right about such
2374: things without using the debugger to find the facts. They also don't
2375: serve a useful purpose.
2376: @end itemize
1.1 root 2377:
1.1.1.2 root 2378: @node Portability, Interface, Bugs, Top
1.1 root 2379: @chapter GNU CC and Portability
2380:
2381: The main goal of GNU CC was to make a good, fast compiler for machines in
2382: the class that the GNU system aims to run on: 32-bit machines that address
2383: 8-bit bytes and have several general registers. Elegance, theoretical
2384: power and simplicity are only secondary.
2385:
2386: GNU CC gets most of the information about the target machine from a machine
2387: description which gives an algebraic formula for each of the machine's
2388: instructions. This is a very clean way to describe the target. But when
2389: the compiler needs information that is difficult to express in this
2390: fashion, I have not hesitated to define an ad-hoc parameter to the machine
2391: description. The purpose of portability is to reduce the total work needed
2392: on the compiler; it was not of interest for its own sake.
2393:
2394: GNU CC does not contain machine dependent code, but it does contain code
2395: that depends on machine parameters such as endianness (whether the most
2396: significant byte has the highest or lowest address of the bytes in a word)
2397: and the availability of autoincrement addressing. In the RTL-generation
2398: pass, it is often necessary to have multiple strategies for generating code
2399: for a particular kind of syntax tree, strategies that are usable for different
2400: combinations of parameters. Often I have not tried to address all possible
2401: cases, but only the common ones or only the ones that I have encountered.
2402: As a result, a new target may require additional strategies. You will know
2403: if this happens because the compiler will call @code{abort}. Fortunately,
1.1.1.2 root 2404: the new strategies can be added in a machine-independent fashion, and will
2405: affect only the target machines that need them.
2406:
2407: @node Interface, Passes, Portability, Top
2408: @chapter Interfacing to GNU CC Output
2409:
2410: GNU CC is normally configured to use the same function calling convention
2411: normally in use on the target system. This is done with the
2412: machine-description macros described (@pxref{Machine Macros}).
2413:
2414: However, returning of structure and union values is done differently.
2415: As a result, functions compiled with PCC returning such types cannot
2416: be called from code compiled with GNU CC, and vice versa. This usually
2417: does not cause trouble because the Unix library routines don't return
2418: structures and unions.
2419:
2420: Structures and unions that are 1, 2, 4 or 8 bytes long are returned in the
2421: same registers used for @code{int} or @code{double} return values. (GNU CC
2422: typically allocates variables of such types in registers also.) Structures
2423: and unions of other sizes are returned by storing them into an address
2424: passed by the caller in a register. This method is faster than the one
2425: normally used by PCC and is also reentrant. The register used for passing
2426: the address is specified by the machine-description macro
1.1.1.6 root 2427: @code{STRUCT_VALUE}.
1.1.1.2 root 2428:
2429: GNU CC always passes arguments on the stack. At some point it will be
2430: extended to pass arguments in registers, for machines which use that as
2431: the standard calling convention. This will make it possible to use such
2432: a convention on other machines as well. However, that would render it
2433: completely incompatible with PCC. We will probably do this once we
2434: have a complete GNU system so we can compile the libraries with GNU CC.
2435:
2436: If you use @code{longjmp}, beware of automatic variables. ANSI C says that
2437: automatic variables that are not declared @code{volatile} have undefined
2438: values after a @code{longjmp}. And this is all GNU CC promises to do,
2439: because it is very difficult to restore register variables correctly, and
2440: one of GNU CC's features is that it can put variables in registers without
2441: your asking it to.
2442:
2443: If you want a variable to be unaltered by @code{longjmp}, and you don't
2444: want to write @code{volatile} because old C compilers don't accept it,
2445: just take the address of the variable. If a variable's address is ever
2446: taken, even if just to compute it and ignore it, then the variable cannot
2447: go in a register:
2448:
2449: @example
2450: @{
2451: int careful;
2452: &careful;
2453: @dots{}
2454: @}
2455: @end example
1.1 root 2456:
1.1.1.2 root 2457: Code compiled with GNU CC may call certain library routines. The routines
2458: needed on the Vax and 68000 are in the file @file{gnulib.c}. You must
2459: compile this file with the standard C compiler, not with GNU CC, and then
2460: link it with each program you compile with GNU CC. (In actuality, many
2461: programs will not need it.) The usual function call interface is used
2462: for calling the library routines. Some standard parts of the C library,
2463: such as @code{bcopy}, are also called automatically.
2464:
2465: @node Passes, RTL, Interface, Top
1.1 root 2466: @chapter Passes and Files of the Compiler
2467:
2468: The overall control structure of the compiler is in @file{toplev.c}. This
2469: file is responsible for initialization, decoding arguments, opening and
2470: closing files, and sequencing the passes.
2471:
1.1.1.2 root 2472: The parsing pass is invoked only once, to parse the entire input. The RTL
2473: intermediate code for a function is generated as the function is parsed, a
2474: statement at a time. Each statement is read in as a syntax tree and then
2475: converted to RTL; then the storage for the tree for the statement is
2476: reclaimed. Storage for types (and the expressions for their sizes),
2477: declarations, and a representation of the binding contours and how they nest,
2478: remains until the function is finished being compiled; these are all needed
2479: to output the debugging information.
2480:
2481: Each time the parsing pass reads a complete function definition or
2482: top-level declaration, it calls the function
2483: @code{rest_of_compilation} or @code{rest_of_decl_compilation} in
2484: @file{toplev.c}, which are responsible for all further processing
2485: necessary, ending with output of the assembler language. All other
2486: compiler passes run, in sequence, within @code{rest_of_compilation}.
2487: When that function returns from compiling a function definition, the
2488: storage used for that function definition's compilation is entirely
2489: freed, unless it is an inline function (@pxref{Inline}).
1.1 root 2490:
2491: Here is a list of all the passes of the compiler and their source files.
2492: Also included is a description of where debugging dumps can be requested
1.1.1.2 root 2493: with @samp{-d} options.
1.1 root 2494:
2495: @itemize @bullet
2496: @item
2497: Parsing. This pass reads the entire text of a function definition,
1.1.1.2 root 2498: constructing partial syntax trees. This and RTL generation are no longer
2499: truly separate passes (formerly they were), but it is easier to think
2500: of them as separate.
2501:
2502: The tree representation does not entirely follow C syntax, because it is
2503: intended to support other languages as well.
1.1 root 2504:
1.1.1.2 root 2505: C data type analysis is also done in this pass, and every tree node
2506: that represents an expression has a data type attached. Variables are
2507: represented as declaration nodes.
1.1 root 2508:
1.1.1.2 root 2509: Constant folding and associative-law simplifications are also done
2510: during this pass.
1.1 root 2511:
1.1.1.7 ! root 2512: The source files for parsing are @file{c-parse.y}, @file{c-decl.c},
! 2513: @file{c-typeck.c}, @file{c-convert.c}, @file{stor-layout.c},
! 2514: @file{fold-const.c}, and @file{tree.c}. The last three files are
! 2515: intended to be language-independent. There are also header files
! 2516: @file{c-parse.h}, @file{c-tree.h}, @file{tree.h} and @file{tree.def}.
! 2517: The last two define the format of the tree representation.@refill
1.1 root 2518:
2519: @item
1.1.1.2 root 2520: RTL generation. This is the conversion of syntax tree into RTL code.
2521: It is actually done statement-by-statement during parsing, but for
2522: most purposes it can be thought of as a separate pass.
1.1 root 2523:
2524: This is where the bulk of target-parameter-dependent code is found,
2525: since often it is necessary for strategies to apply only when certain
2526: standard kinds of instructions are available. The purpose of named
2527: instruction patterns is to provide this information to the RTL
2528: generation pass.
2529:
2530: Optimization is done in this pass for @code{if}-conditions that are
2531: comparisons, boolean operations or conditional expressions. Tail
2532: recursion is detected at this time also. Decisions are made about how
2533: best to arrange loops and how to output @code{switch} statements.
2534:
1.1.1.2 root 2535: The source files for RTL generation are @file{stmt.c}, @file{expr.c},
1.1 root 2536: @file{explow.c}, @file{expmed.c}, @file{optabs.c} and @file{emit-rtl.c}.
2537: Also, the file @file{insn-emit.c}, generated from the machine description
2538: by the program @code{genemit}, is used in this pass. The header files
1.1.1.2 root 2539: @file{expr.h} is used for communication within this pass.@refill
1.1 root 2540:
1.1.1.2 root 2541: The header files @file{insn-flags.h} and @file{insn-codes.h},
2542: generated from the machine description by the programs @code{genflags}
2543: and @code{gencodes}, tell this pass which standard names are available
2544: for use and which patterns correspond to them.@refill
1.1 root 2545:
2546: Aside from debugging information output, none of the following passes
1.1.1.2 root 2547: refers to the tree structure representation of the function (only
2548: part of which is saved).
1.1 root 2549:
1.1.1.2 root 2550: The decision of whether the function can and should be expanded inline
2551: in its subsequent callers is made at the end of rtl generation. The
2552: function must meet certain criteria, currently related to the size of
2553: the function and the types and number of parameters it has. Note that
2554: this function may contain loops, recursive calls to itself
2555: (tail-recursive functions can be inlined!), gotos, in short, all
2556: constructs supported by GNU CC.
2557:
2558: The option @samp{-dr} causes a debugging dump of the RTL code after
2559: this pass. This dump file's name is made by appending @samp{.rtl} to
2560: the input file name.
1.1 root 2561:
2562: @item
1.1.1.2 root 2563: Jump optimization. This pass simplifies jumps to the following
2564: instruction, jumps across jumps, and jumps to jumps. It deletes
2565: unreferenced labels and unreachable code, except that unreachable code
2566: that contains a loop is not recognized as unreachable in this pass.
2567: (Such loops are deleted later in the basic block analysis.)
1.1 root 2568:
2569: Jump optimization is performed two or three times. The first time is
1.1.1.2 root 2570: immediately following RTL generation. The second time is after CSE,
2571: but only if CSE says repeated jump optimization is needed. The
2572: last time is right before the final pass. That time, cross-jumping
2573: and deletion of no-op move instructions are done together with the
2574: optimizations described above.
1.1 root 2575:
2576: The source file of this pass is @file{jump.c}.
2577:
1.1.1.2 root 2578: The option @samp{-dj} causes a debugging dump of the RTL code after
2579: this pass is run for the first time. This dump file's name is made by
2580: appending @samp{.jump} to the input file name.
1.1 root 2581:
2582: @item
2583: Register scan. This pass finds the first and last use of each
2584: register, as a guide for common subexpression elimination. Its source
2585: is in @file{regclass.c}.
2586:
2587: @item
2588: Common subexpression elimination. This pass also does constant
2589: propagation. Its source file is @file{cse.c}. If constant
2590: propagation causes conditional jumps to become unconditional or to
1.1.1.2 root 2591: become no-ops, jump optimization is run again when CSE is finished.
1.1 root 2592:
1.1.1.2 root 2593: The option @samp{-ds} causes a debugging dump of the RTL code after
1.1 root 2594: this pass. This dump file's name is made by appending @samp{.cse} to
2595: the input file name.
2596:
2597: @item
2598: Loop optimization. This pass moves constant expressions out of loops.
2599: Its source file is @file{loop.c}.
2600:
1.1.1.2 root 2601: The option @samp{-dL} causes a debugging dump of the RTL code after
1.1 root 2602: this pass. This dump file's name is made by appending @samp{.loop} to
2603: the input file name.
2604:
2605: @item
2606: Stupid register allocation is performed at this point in a
2607: nonoptimizing compilation. It does a little data flow analysis as
2608: well. When stupid register allocation is in use, the next pass
2609: executed is the reloading pass; the others in between are skipped.
1.1.1.2 root 2610: The source file is @file{stupid.c}.
1.1 root 2611:
2612: @item
2613: Data flow analysis (@file{flow.c}). This pass divides the program
2614: into basic blocks (and in the process deletes unreachable loops); then
2615: it computes which pseudo-registers are live at each point in the
2616: program, and makes the first instruction that uses a value point at
2617: the instruction that computed the value.
2618:
2619: This pass also deletes computations whose results are never used, and
2620: combines memory references with add or subtract instructions to make
2621: autoincrement or autodecrement addressing.
2622:
1.1.1.2 root 2623: The option @samp{-df} causes a debugging dump of the RTL code after
1.1 root 2624: this pass. This dump file's name is made by appending @samp{.flow} to
2625: the input file name. If stupid register allocation is in use, this
2626: dump file reflects the full results of such allocation.
2627:
2628: @item
2629: Instruction combination (@file{combine.c}). This pass attempts to
2630: combine groups of two or three instructions that are related by data
2631: flow into single instructions. It combines the RTL expressions for
2632: the instructions by substitution, simplifies the result using algebra,
2633: and then attempts to match the result against the machine description.
2634:
1.1.1.2 root 2635: The option @samp{-dc} causes a debugging dump of the RTL code after
1.1 root 2636: this pass. This dump file's name is made by appending @samp{.combine}
2637: to the input file name.
2638:
2639: @item
2640: Register class preferencing. The RTL code is scanned to find out
1.1.1.2 root 2641: which register class is best for each pseudo register. The source
2642: file is @file{regclass.c}.
1.1 root 2643:
2644: @item
2645: Local register allocation (@file{local-alloc.c}). This pass allocates
2646: hard registers to pseudo registers that are used only within one basic
1.1.1.2 root 2647: block. Because the basic block is linear, it can use fast and
2648: powerful techniques to do a very good job.
1.1 root 2649:
1.1.1.2 root 2650: The option @samp{-dl} causes a debugging dump of the RTL code after
1.1 root 2651: this pass. This dump file's name is made by appending @samp{.lreg} to
2652: the input file name.
2653:
2654: @item
2655: Global register allocation (@file{global-alloc.c}). This pass
2656: allocates hard registers for the remaining pseudo registers (those
2657: whose life spans are not contained in one basic block).
2658:
2659: @item
1.1.1.2 root 2660: Reloading. This pass renumbers pseudo registers with the hardware
2661: registers numbers they were allocated. Pseudo registers that did not
2662: get hard registers are replaced with stack slots. Then it finds
2663: instructions that are invalid because a value has failed to end up in
2664: a register, or has ended up in a register of the wrong kind. It fixes
2665: up these instructions by reloading the problematical values
2666: temporarily into registers. Additional instructions are generated to
2667: do the copying.
1.1 root 2668:
2669: Source files are @file{reload.c} and @file{reload1.c}, plus the header
2670: @file{reload.h} used for communication between them.
2671:
1.1.1.2 root 2672: The option @samp{-dg} causes a debugging dump of the RTL code after
1.1 root 2673: this pass. This dump file's name is made by appending @samp{.greg} to
2674: the input file name.
2675:
2676: @item
1.1.1.2 root 2677: Jump optimization is repeated, this time including cross-jumping
2678: and deletion of no-op move instructions. Machine-specific peephole
2679: optimizations are performed at the same time.
2680:
2681: The option @samp{-dJ} causes a debugging dump of the RTL code after
2682: this pass. This dump file's name is made by appending @samp{.jump2}
2683: to the input file name.
1.1 root 2684:
2685: @item
2686: Final. This pass outputs the assembler code for the function. It is
1.1.1.2 root 2687: also responsible for identifying spurious test and compare
2688: instructions. The function entry and exit sequences are generated
2689: directly as assembler code in this pass; they never exist as RTL.
1.1 root 2690:
2691: The source files are @file{final.c} plus @file{insn-output.c}; the
2692: latter is generated automatically from the machine description by the
2693: tool @file{genoutput}. The header file @file{conditions.h} is used
2694: for communication between these files.
2695:
2696: @item
2697: Debugging information output. This is run after final because it must
2698: output the stack slot offsets for pseudo registers that did not get
2699: hard registers. Source files are @file{dbxout.c} for DBX symbol table
2700: format and @file{symout.c} for GDB's own symbol table format.
2701: @end itemize
2702:
2703: Some additional files are used by all or many passes:
2704:
2705: @itemize @bullet
2706: @item
2707: Every pass uses @file{machmode.def}, which defines the machine modes.
2708:
2709: @item
2710: All the passes that work with RTL use the header files @file{rtl.h}
1.1.1.2 root 2711: and @file{rtl.def}, and subroutines in file @file{rtl.c}. The tools
2712: @code{gen*} also use these files to read and work with the machine
2713: description RTL.
1.1 root 2714:
2715: @item
2716: Several passes refer to the header file @file{insn-config.h} which
2717: contains a few parameters (C macro definitions) generated
2718: automatically from the machine description RTL by the tool
2719: @code{genconfig}.
2720:
2721: @item
2722: Several passes use the instruction recognizer, which consists of
2723: @file{recog.c} and @file{recog.h}, plus the files @file{insn-recog.c}
2724: and @file{insn-extract.c} that are generated automatically from the
1.1.1.2 root 2725: machine description by the tools @file{genrecog} and
2726: @file{genextract}.@refill
1.1 root 2727:
2728: @item
1.1.1.2 root 2729: Several passes use the header files @file{regs.h} which defines the
2730: information recorded about pseudo register usage, and @file{basic-block.h}
1.1 root 2731: which defines the information recorded about basic blocks.
2732:
2733: @item
2734: @file{hard-reg-set.h} defines the type @code{HARD_REG_SET}, a bit-vector
2735: with a bit for each hard register, and some macros to manipulate it.
2736: This type is just @code{int} if the machine has few enough hard registers;
2737: otherwise it is an array of @code{int} and some of the macros expand
2738: into loops.
2739: @end itemize
2740:
2741: @node RTL, Machine Desc, Passes, Top
2742: @chapter RTL Representation
2743:
2744: Most of the work of the compiler is done on an intermediate representation
1.1.1.2 root 2745: called register transfer language. In this language, the instructions to be
1.1 root 2746: output are described, pretty much one by one, in an algebraic form that
2747: describes what the instruction does.
2748:
2749: RTL is inspired by Lisp lists. It has both an internal form, made up of
2750: structures that point at other structures, and a textual form that is used
2751: in the machine description and in printed debugging dumps. The textual
2752: form uses nested parentheses to indicate the pointers in the internal form.
2753:
2754: @menu
2755: * RTL Objects:: Expressions vs vectors vs strings vs integers.
2756: * Accessors:: Macros to access expression operands or vector elts.
1.1.1.2 root 2757: * Flags:: Other flags in an RTL expression.
1.1 root 2758: * Machine Modes:: Describing the size and format of a datum.
2759: * Constants:: Expressions with constant values.
2760: * Regs and Memory:: Expressions representing register contents or memory.
2761: * Arithmetic:: Expressions representing arithmetic on other expressions.
2762: * Comparisons:: Expressions representing comparison of expressions.
2763: * Bit Fields:: Expressions representing bit-fields in memory or reg.
2764: * Conversions:: Extending, truncating, floating or fixing.
2765: * RTL Declarations:: Declaring volatility, constancy, etc.
2766: * Side Effects:: Expressions for storing in registers, etc.
2767: * Incdec:: Embedded side-effects for autoincrement addressing.
1.1.1.2 root 2768: * Assembler:: Representing @code{asm} with operands.
1.1 root 2769: * Insns:: Expression types for entire insns.
1.1.1.2 root 2770: * Calls:: RTL representation of function call insns.
1.1 root 2771: * Sharing:: Some expressions are unique; others *must* be copied.
2772: @end menu
2773:
2774: @node RTL Objects, Accessors, RTL, RTL
2775: @section RTL Object Types
2776:
2777: RTL uses four kinds of objects: expressions, integers, strings and vectors.
1.1.1.2 root 2778: Expressions are the most important ones. An RTL expression (``RTX'', for
2779: short) is a C structure, but it is usually referred to with a pointer; a
2780: type that is given the typedef name @code{rtx}.
1.1 root 2781:
2782: An integer is simply an @code{int}, and a string is a @code{char *}.
1.1.1.2 root 2783: Within RTL code, strings appear only inside @samp{symbol_ref} expressions,
2784: but they appear in other contexts in the RTL expressions that make up
1.1 root 2785: machine descriptions. Their written form uses decimal digits.
2786:
2787: A string is a sequence of characters. In core it is represented as a
1.1.1.2 root 2788: @code{char *} in usual C fashion, and it is written in C syntax as well.
1.1 root 2789: However, strings in RTL may never be null. If you write an empty string in
2790: a machine description, it is represented in core as a null pointer rather
2791: than as a pointer to a null character. In certain contexts, these null
2792: pointers instead of strings are valid.
2793:
2794: A vector contains an arbitrary, specified number of pointers to
2795: expressions. The number of elements in the vector is explicitly present in
2796: the vector. The written form of a vector consists of square brackets
2797: (@samp{[@dots{}]}) surrounding the elements, in sequence and with
2798: whitespace separating them. Vectors of length zero are not created; null
2799: pointers are used instead.
2800:
1.1.1.2 root 2801: Expressions are classified by @dfn{expression codes} (also called RTX
2802: codes). The expression code is a name defined in @file{rtl.def}, which is
2803: also (in upper case) a C enumeration constant. The possible expression
2804: codes and their meanings are machine-independent. The code of an RTX can
2805: be extracted with the macro @code{GET_CODE (@var{x})} and altered with
2806: @code{PUT_CODE (@var{x}, @var{newcode})}.
1.1 root 2807:
2808: The expression code determines how many operands the expression contains,
2809: and what kinds of objects they are. In RTL, unlike Lisp, you cannot tell
2810: by looking at an operand what kind of object it is. Instead, you must know
2811: from its context---from the expression code of the containing expression.
1.1.1.2 root 2812: For example, in an expression of code @samp{subreg}, the first operand is
1.1 root 2813: to be regarded as an expression and the second operand as an integer. In
1.1.1.2 root 2814: an expression of code @samp{plus}, there are two operands, both of which
2815: are to be regarded as expressions. In a @samp{symbol_ref} expression,
1.1 root 2816: there is one operand, which is to be regarded as a string.
2817:
2818: Expressions are written as parentheses containing the name of the
2819: expression type, its flags and machine mode if any, and then the operands
2820: of the expression (separated by spaces).
2821:
1.1.1.2 root 2822: Expression code names in the @samp{md} file are written in lower case,
2823: but when they appear in C code they are written in upper case. In this
2824: manual, they are shown as follows: @samp{const_int}.
2825:
1.1 root 2826: In a few contexts a null pointer is valid where an expression is normally
2827: wanted. The written form of this is @samp{(nil)}.
2828:
1.1.1.2 root 2829: @node Accessors, Flags, RTL Objects, RTL
1.1 root 2830: @section Access to Operands
2831:
2832: For each expression type @file{rtl.def} specifies the number of contained
2833: objects and their kinds, with four possibilities: @samp{e} for expression
2834: (actually a pointer to an expression), @samp{i} for integer, @samp{s} for
2835: string, and @samp{E} for vector of expressions. The sequence of letters
2836: for an expression code is called its @dfn{format}. Thus, the format of
1.1.1.2 root 2837: @samp{subreg} is @samp{ei}.@refill
1.1 root 2838:
2839: Two other format characters are used occasionally: @samp{u} and @samp{0}.
2840: @samp{u} is equivalent to @samp{e} except that it is printed differently in
2841: debugging dumps, and @samp{0} means a slot whose contents do not fit any
2842: normal category. @samp{0} slots are not printed at all in dumps, and are
1.1.1.2 root 2843: often used in special ways by small parts of the compiler.@refill
1.1 root 2844:
2845: There are macros to get the number of operands and the format of an
2846: expression code:
2847:
2848: @table @code
2849: @item GET_RTX_LENGTH (@var{code})
1.1.1.2 root 2850: Number of operands of an RTX of code @var{code}.
1.1 root 2851:
2852: @item GET_RTX_FORMAT (@var{code})
1.1.1.2 root 2853: The format of an RTX of code @var{code}, as a C string.
1.1 root 2854: @end table
2855:
2856: Operands of expressions are accessed using the macros @code{XEXP},
2857: @code{XINT} and @code{XSTR}. Each of these macros takes two arguments: an
1.1.1.2 root 2858: expression-pointer (RTX) and an operand number (counting from zero).
2859: Thus,@refill
1.1 root 2860:
2861: @example
1.1.1.2 root 2862: XEXP (@var{x}, 2)
1.1 root 2863: @end example
2864:
2865: @noindent
2866: accesses operand 2 of expression @var{x}, as an expression.
2867:
2868: @example
1.1.1.2 root 2869: XINT (@var{x}, 2)
1.1 root 2870: @end example
2871:
2872: @noindent
2873: accesses the same operand as an integer. @code{XSTR}, used in the same
2874: fashion, would access it as a string.
2875:
2876: Any operand can be accessed as an integer, as an expression or as a string.
2877: You must choose the correct method of access for the kind of value actually
2878: stored in the operand. You would do this based on the expression code of
2879: the containing expression. That is also how you would know how many
2880: operands there are.
2881:
2882: For example, if @var{x} is a @samp{subreg} expression, you know that it has
1.1.1.2 root 2883: two operands which can be correctly accessed as @code{XEXP (@var{x}, 0)}
2884: and @code{XINT (@var{x}, 1)}. If you did @code{XINT (@var{x}, 0)}, you
2885: would get the address of the expression operand but cast as an integer;
2886: that might occasionally be useful, but it would be cleaner to write
2887: @code{(int) XEXP (@var{x}, 0)}. @code{XEXP (@var{x}, 1)} would also
2888: compile without error, and would return the second, integer operand cast as
2889: an expression pointer, which would probably result in a crash when
2890: accessed. Nothing stops you from writing @code{XEXP (@var{x}, 28)} either,
2891: but this will access memory past the end of the expression with
2892: unpredictable results.@refill
1.1 root 2893:
2894: Access to operands which are vectors is more complicated. You can use the
2895: macro @code{XVEC} to get the vector-pointer itself, or the macros
2896: @code{XVECEXP} and @code{XVECLEN} to access the elements and length of a
2897: vector.
2898:
2899: @table @code
2900: @item XVEC (@var{exp}, @var{idx})
2901: Access the vector-pointer which is operand number @var{idx} in @var{exp}.
2902:
2903: @item XVECLEN (@var{exp}, @var{idx})
2904: Access the length (number of elements) in the vector which is
2905: in operand number @var{idx} in @var{exp}. This value is an @code{int}.
2906:
1.1.1.2 root 2907: @item XVECEXP (@var{exp}, @var{idx}, @var{eltnum})
1.1 root 2908: Access element number @var{eltnum} in the vector which is
1.1.1.2 root 2909: in operand number @var{idx} in @var{exp}. This value is an RTX.
1.1 root 2910:
2911: It is up to you to make sure that @var{eltnum} is not negative
2912: and is less than @code{XVECLEN (@var{exp}, @var{idx})}.
2913: @end table
2914:
2915: All the macros defined in this section expand into lvalues and therefore
2916: can be used to assign the operands, lengths and vector elements as well as
2917: to access them.
2918:
1.1.1.2 root 2919: @node Flags, Machine Modes, Accessors, RTL
2920: @section Flags in an RTL Expression
2921:
2922: RTL expressions contain several flags (one-bit bit-fields) that are used
2923: in certain types of expression.
2924:
2925: @table @code
2926: @item used
2927: This flag is used only momentarily, at the end of RTL generation for a
2928: function, to count the number of times an expression appears in insns.
2929: Expressions that appear more than once are copied, according to the
2930: rules for shared structure (@pxref{Sharing}).
2931:
2932: @item volatil
2933: This flag is used in @samp{mem} and @samp{reg} expressions and in insns.
2934: In RTL dump files, it is printed as @samp{/v}.
2935:
2936: In a @samp{mem} expression, it is 1 if the memory reference is volatile.
2937: Volatile memory references may not be deleted, reordered or combined.
2938:
2939: In a @samp{reg} expression, it is 1 if the value is a user-level variable.
2940: 0 indicates an internal compiler temporary.
2941:
2942: In an insn, 1 means the insn has been deleted.
2943:
2944: @item in_struct
2945: This flag is used in @samp{mem} expressions. It is 1 if the memory
2946: datum referred to is all or part of a structure or array; 0 if it is (or
2947: might be) a scalar variable. A reference through a C pointer has 0
2948: because the pointer might point to a scalar variable.
2949:
2950: This information allows the compiler to determine something about possible
2951: cases of aliasing.
2952:
2953: In an RTL dump, this flag is represented as @samp{/s}.
2954:
2955: @item unchanging
2956: This flag is used in @samp{reg} and @samp{mem} expressions. 1 means
2957: that the value of the expression never changes (at least within the
2958: current function).
2959:
2960: In an RTL dump, this flag is represented as @samp{/u}.
1.1.1.6 root 2961:
2962: @item integrated
2963: In some kinds of expressions, including insns, this flag means the
2964: rtl was produced by procedure integration.
2965:
2966: In a @samp{reg} expression, this flag indicates the register
2967: containing the value to be returned by the current function. On
2968: machines that pass parameters in registers, the same register number
2969: may be used for parameters as well, but this flag is not set on such
2970: uses.
1.1.1.2 root 2971: @end table
2972:
2973: @node Machine Modes, Constants, Flags, RTL
1.1 root 2974: @section Machine Modes
2975:
2976: A machine mode describes a size of data object and the representation used
2977: for it. In the C code, machine modes are represented by an enumeration
1.1.1.2 root 2978: type, @code{enum machine_mode}, defined in @file{machmode.def}. Each RTL
2979: expression has room for a machine mode and so do certain kinds of tree
2980: expressions (declarations and types, to be precise).
1.1 root 2981:
2982: In debugging dumps and machine descriptions, the machine mode of an RTL
2983: expression is written after the expression code with a colon to separate
2984: them. The letters @samp{mode} which appear at the end of each machine mode
2985: name are omitted. For example, @code{(reg:SI 38)} is a @samp{reg}
2986: expression with machine mode @code{SImode}. If the mode is
2987: @code{VOIDmode}, it is not written at all.
2988:
2989: Here is a table of machine modes.
2990:
2991: @table @code
2992: @item QImode
2993: ``Quarter-Integer'' mode represents a single byte treated as an integer.
2994:
2995: @item HImode
2996: ``Half-Integer'' mode represents a two-byte integer.
2997:
2998: @item SImode
2999: ``Single Integer'' mode represents a four-byte integer.
3000:
3001: @item DImode
3002: ``Double Integer'' mode represents an eight-byte integer.
3003:
3004: @item TImode
3005: ``Tetra Integer'' (?) mode represents a sixteen-byte integer.
3006:
3007: @item SFmode
3008: ``Single Floating'' mode represents a single-precision (four byte) floating
3009: point number.
3010:
3011: @item DFmode
3012: ``Double Floating'' mode represents a double-precision (eight byte) floating
3013: point number.
3014:
3015: @item TFmode
3016: ``Tetra Floating'' mode represents a quadruple-precision (sixteen byte)
3017: floating point number.
3018:
3019: @item BLKmode
3020: ``Block'' mode represents values that are aggregates to which none of
1.1.1.2 root 3021: the other modes apply. In RTL, only memory references can have this mode,
1.1 root 3022: and only if they appear in string-move or vector instructions. On machines
3023: which have no such instructions, @code{BLKmode} will not appear in RTL.
3024:
3025: @item VOIDmode
3026: Void mode means the absence of a mode or an unspecified mode.
1.1.1.2 root 3027: For example, RTL expressions of code @samp{const_int} have mode
1.1 root 3028: @code{VOIDmode} because they can be taken to have whatever mode the context
3029: requires. In debugging dumps of RTL, @code{VOIDmode} is expressed by
3030: the absence of any mode.
3031:
3032: @item EPmode
3033: ``Entry Pointer'' mode is intended to be used for function variables in
3034: Pascal and other block structured languages. Such values contain
3035: both a function address and a static chain pointer for access to
3036: automatic variables of outer levels. This mode is only partially
3037: implemented since C does not use it.
3038:
3039: @item CSImode@r{, @dots{}}
3040: ``Complex Single Integer'' mode stands for a complex number represented
3041: as a pair of @code{SImode} integers. Any of the integer and floating modes
3042: may have @samp{C} prefixed to its name to obtain a complex number mode.
3043: For example, there are @code{CQImode}, @code{CSFmode}, and @code{CDFmode}.
3044: Since C does not support complex numbers, these machine modes are only
3045: partially implemented.
3046:
3047: @item BImode
3048: This is the machine mode of a bit-field in a structure. It is used
3049: only in the syntax tree, never in RTL, and in the syntax tree it appears
3050: only in declaration nodes. In C, it appears only in @code{FIELD_DECL}
3051: nodes for structure fields defined with a bit size.
3052: @end table
3053:
3054: The machine description defines @code{Pmode} as a C macro which expands
3055: into the machine mode used for addresses. Normally this is @code{SImode}.
3056:
3057: The only modes which a machine description @i{must} support are
3058: @code{QImode}, @code{SImode}, @code{SFmode} and @code{DFmode}. The
3059: compiler will attempt to use @code{DImode} for two-word structures and
3060: unions, but it would not be hard to program it to avoid this. Likewise,
3061: you can arrange for the C type @code{short int} to avoid using
3062: @code{HImode}. In the long term it would be desirable to make the set of
3063: available machine modes machine-dependent and eliminate all assumptions
3064: about specific machine modes or their uses from the machine-independent
3065: code of the compiler.
3066:
3067: Here are some C macros that relate to machine modes:
3068:
3069: @table @code
3070: @item GET_MODE (@var{x})
1.1.1.2 root 3071: Returns the machine mode of the RTX @var{x}.
1.1 root 3072:
3073: @item PUT_MODE (@var{x}, @var{newmode})
1.1.1.2 root 3074: Alters the machine mode of the RTX @var{x} to be @var{newmode}.
1.1 root 3075:
3076: @item GET_MODE_SIZE (@var{m})
3077: Returns the size in bytes of a datum of mode @var{m}.
3078:
3079: @item GET_MODE_BITSIZE (@var{m})
3080: Returns the size in bits of a datum of mode @var{m}.
3081:
3082: @item GET_MODE_UNIT_SIZE (@var{m})
3083: Returns the size in bits of the subunits of a datum of mode @var{m}.
3084: This is the same as @code{GET_MODE_SIZE} except in the case of
1.1.1.2 root 3085: complex modes and @code{EPmode}. For them, the unit size is the
1.1 root 3086: size of the real or imaginary part, or the size of the function
3087: pointer or the context pointer.
3088: @end table
3089:
3090: @node Constants, Regs and Memory, Machine Modes, RTL
3091: @section Constant Expression Types
3092:
3093: The simplest RTL expressions are those that represent constant values.
3094:
3095: @table @code
3096: @item (const_int @var{i})
3097: This type of expression represents the integer value @var{i}. @var{i}
3098: is customarily accessed with the macro @code{INTVAL} as in
1.1.1.2 root 3099: @code{INTVAL (@var{exp})}, which is equivalent to @code{XINT (@var{exp}, 0)}.
1.1 root 3100:
3101: There is only one expression object for the integer value zero;
3102: it is the value of the variable @code{const0_rtx}. Likewise, the
3103: only expression for integer value one is found in @code{const1_rtx}.
1.1.1.2 root 3104: Any attempt to create an expression of code @samp{const_int} and
1.1 root 3105: value zero or one will return @code{const0_rtx} or @code{const1_rtx}
3106: as appropriate.
3107:
3108: @item (const_double:@var{m} @var{i0} @var{i1})
3109: Represents a floating point constant value of mode @var{m}. The two
3110: integers @var{i0} and @var{i1} together contain the bits of a
3111: @code{double} value. To convert them to a @code{double}, do
3112:
3113: @example
1.1.1.2 root 3114: union @{ double d; int i[2];@} u;
1.1 root 3115: u.i[0] = XINT (x, 0);
3116: u.i[1] = XINT (x, 1);
3117: @end example
3118:
3119: @noindent
3120: and then refer to @code{u.d}. The value of the constant is
3121: represented as a double in this fashion even if the value represented
3122: is single-precision.
3123:
1.1.1.2 root 3124: The global variables @code{dconst0_rtx} and @code{fconst0_rtx} hold
3125: @samp{const_double} expressions with value 0, in modes @code{DFmode} and
3126: @code{SFmode}, respectively.
1.1 root 3127:
3128: @item (symbol_ref @var{symbol})
3129: Represents the value of an assembler label for data. @var{symbol} is
3130: a string that describes the name of the assembler label. If it starts
3131: with a @samp{*}, the label is the rest of @var{symbol} not including
3132: the @samp{*}. Otherwise, the label is @var{symbol}, prefixed with
3133: @samp{_}.
3134:
3135: @item (label_ref @var{label})
3136: Represents the value of an assembler label for code. It contains one
1.1.1.2 root 3137: operand, an expression, which must be a @samp{code_label} that appears
1.1 root 3138: in the instruction sequence to identify the place where the label
3139: should go.
3140:
3141: The reason for using a distinct expression type for code label
3142: references is so that jump optimization can distinguish them.
3143:
3144: @item (const @var{exp})
3145: Represents a constant that is the result of an assembly-time
3146: arithmetic computation. The operand, @var{exp}, is an expression that
3147: contains only constants (@samp{const_int}, @samp{symbol_ref} and
3148: @samp{label_ref} expressions) combined with @samp{plus} and
3149: @samp{minus}. However, not all combinations are valid, since the
3150: assembler cannot do arbitrary arithmetic on relocatable symbols.
3151: @end table
3152:
3153: @node Regs and Memory, Arithmetic, Constants, RTL
3154: @section Registers and Memory
3155:
3156: Here are the RTL expression types for describing access to machine
3157: registers and to main memory.
3158:
3159: @table @code
3160: @item (reg:@var{m} @var{n})
3161: For small values of the integer @var{n} (less than
3162: @code{FIRST_PSEUDO_REGISTER}), this stands for a reference to machine
3163: register number @var{n}: a @dfn{hard register}. For larger values of
3164: @var{n}, it stands for a temporary value or @dfn{pseudo register}.
3165: The compiler's strategy is to generate code assuming an unlimited
3166: number of such pseudo registers, and later convert them into hard
3167: registers or into memory references.
3168:
3169: The symbol @code{FIRST_PSEUDO_REGISTER} is defined by the machine
3170: description, since the number of hard registers on the machine is an
3171: invariant characteristic of the machine. Note, however, that not
3172: all of the machine registers must be general registers. All the
3173: machine registers that can be used for storage of data are given
3174: hard register numbers, even those that can be used only in certain
3175: instructions or can hold only certain types of data.
3176:
1.1.1.2 root 3177: Each pseudo register number used in a function's RTL code is
1.1 root 3178: represented by a unique @samp{reg} expression.
3179:
3180: @var{m} is the machine mode of the reference. It is necessary because
3181: machines can generally refer to each register in more than one mode.
3182: For example, a register may contain a full word but there may be
3183: instructions to refer to it as a half word or as a single byte, as
3184: well as instructions to refer to it as a floating point number of
3185: various precisions.
3186:
3187: Even for a register that the machine can access in only one mode,
3188: the mode must always be specified.
3189:
3190: A hard register may be accessed in various modes throughout one
3191: function, but each pseudo register is given a natural mode
3192: and is accessed only in that mode. When it is necessary to describe
3193: an access to a pseudo register using a nonnatural mode, a @samp{subreg}
3194: expression is used.
3195:
3196: A @samp{reg} expression with a machine mode that specifies more than
3197: one word of data may actually stand for several consecutive registers.
3198: If in addition the register number specifies a hardware register, then
3199: it actually represents several consecutive hardware registers starting
3200: with the specified one.
3201:
3202: Such multi-word hardware register @samp{reg} expressions may not be live
3203: across the boundary of a basic block. The lifetime analysis pass does not
3204: know how to record properly that several consecutive registers are
3205: actually live there, and therefore register allocation would be confused.
3206: The CSE pass must go out of its way to make sure the situation does
3207: not arise.
3208:
3209: @item (subreg:@var{m} @var{reg} @var{wordnum})
3210: @samp{subreg} expressions are used to refer to a register in a machine
3211: mode other than its natural one, or to refer to one register of
3212: a multi-word @samp{reg} that actually refers to several registers.
3213:
3214: Each pseudo-register has a natural mode. If it is necessary to
3215: operate on it in a different mode---for example, to perform a fullword
3216: move instruction on a pseudo-register that contains a single byte---
3217: the pseudo-register must be enclosed in a @samp{subreg}. In such
3218: a case, @var{wordnum} is zero.
3219:
3220: The other use of @samp{subreg} is to extract the individual registers
3221: of a multi-register value. Machine modes such as @code{DImode} and
3222: @code{EPmode} indicate values longer than a word, values which usually
3223: require two consecutive registers. To access one of the registers,
3224: use a @samp{subreg} with mode @code{SImode} and a @var{wordnum} that
3225: says which register.
3226:
3227: The compilation parameter @code{WORDS_BIG_ENDIAN}, if defined, says
3228: that word number zero is the most significant part; otherwise, it is
3229: the least significant part.
3230:
1.1.1.7 ! root 3231: Between the combiner pass and the reload pass, it is possible to have
! 3232: a @samp{subreg} which contains a @samp{mem} instead of a @samp{reg} as
! 3233: its first operand. The reload pass eliminates these cases by
! 3234: reloading the @samp{mem} into a suitable register.
! 3235:
1.1 root 3236: Note that it is not valid to access a @code{DFmode} value in @code{SFmode}
3237: using a @samp{subreg}. On some machines the most significant part of a
3238: @code{DFmode} value does not have the same format as a single-precision
3239: floating value.
3240:
3241: @item (cc0)
3242: This refers to the machine's condition code register. It has no
3243: operands and may not have a machine mode. It may be validly used in
3244: only two contexts: as the destination of an assignment (in test and
3245: compare instructions) and in comparison operators comparing against
1.1.1.2 root 3246: zero (@samp{const_int} with value zero; that is to say,
3247: @code{const0_rtx}).
1.1 root 3248:
1.1.1.2 root 3249: There is only one expression object of code @samp{cc0}; it is the
1.1 root 3250: value of the variable @code{cc0_rtx}. Any attempt to create an
1.1.1.2 root 3251: expression of code @samp{cc0} will return @code{cc0_rtx}.
1.1 root 3252:
1.1.1.2 root 3253: One special thing about the condition code register is that
3254: instructions can set it implicitly. On many machines, nearly all
3255: instructions set the condition code based on the value that they
3256: compute or store. It is not necessary to record these actions
3257: explicitly in the RTL because the machine description includes a
3258: prescription for recognizing the instructions that do so (by means of
3259: the macro @code{NOTICE_UPDATE_CC}). Only instructions whose sole
3260: purpose is to set the condition code, and instructions that use the
3261: condition code, need mention @code{(cc0)}.
1.1 root 3262:
3263: @item (pc)
3264: This represents the machine's program counter. It has no operands and
3265: may not have a machine mode. @code{(pc)} may be validly used only in
3266: certain specific contexts in jump instructions.
3267:
1.1.1.2 root 3268: There is only one expression object of code @samp{pc}; it is the value
3269: of the variable @code{pc_rtx}. Any attempt to create an expression of
3270: code @samp{pc} will return @code{pc_rtx}.
1.1 root 3271:
1.1.1.2 root 3272: All instructions that do not jump alter the program counter implicitly
3273: by incrementing it, but there is no need to mention this in the RTL.
1.1 root 3274:
3275: @item (mem:@var{m} @var{addr})
1.1.1.2 root 3276: This RTX represents a reference to main memory at an address
3277: represented by the expression @var{addr}. @var{m} specifies how large
3278: a unit of memory is accessed.
1.1 root 3279: @end table
3280:
3281: @node Arithmetic, Comparisons, Regs and Memory, RTL
3282: @section RTL Expressions for Arithmetic
3283:
3284: @table @code
3285: @item (plus:@var{m} @var{x} @var{y})
3286: Represents the sum of the values represented by @var{x} and @var{y}
3287: carried out in machine mode @var{m}. This is valid only if
3288: @var{x} and @var{y} both are valid for mode @var{m}.
3289:
3290: @item (minus:@var{m} @var{x} @var{y})
3291: Like @samp{plus} but represents subtraction.
3292:
3293: @item (minus @var{x} @var{y})
3294: Represents the result of subtracting @var{y} from @var{x}
3295: for purposes of comparison. The absence of a machine mode
3296: in the @samp{minus} expression indicates that the result is
3297: computed without overflow, as if with infinite precision.
3298:
3299: Of course, machines can't really subtract with infinite precision.
3300: However, they can pretend to do so when only the sign of the
3301: result will be used, which is the case when the result is stored
1.1.1.2 root 3302: in @code{(cc0)}. And that is the only way this kind of expression
1.1 root 3303: may validly be used: as a value to be stored in the condition codes.
3304:
3305: @item (neg:@var{m} @var{x})
3306: Represents the negation (subtraction from zero) of the value
3307: represented by @var{x}, carried out in mode @var{m}. @var{x} must be
3308: valid for mode @var{m}.
3309:
3310: @item (mult:@var{m} @var{x} @var{y})
3311: Represents the signed product of the values represented by @var{x} and
3312: @var{y} carried out in machine mode @var{m}. If
3313: @var{x} and @var{y} are both valid for mode @var{m}, this is ordinary
1.1.1.2 root 3314: size-preserving multiplication. Alternatively, both @var{x} and @var{y}
1.1 root 3315: may be valid for a different, narrower mode. This represents the
3316: kind of multiplication that generates a product wider than the operands.
3317: Widening multiplication and same-size multiplication are completely
3318: distinct and supported by different machine instructions; machines may
1.1.1.2 root 3319: support one but not the other.@refill
1.1 root 3320:
3321: @samp{mult} may be used for floating point division as well.
3322: Then @var{m} is a floating point machine mode.
3323:
3324: @item (umult:@var{m} @var{x} @var{y})
3325: Like @samp{mult} but represents unsigned multiplication. It may be
3326: used in both same-size and widening forms, like @samp{mult}.
1.1.1.2 root 3327: @samp{umult} is used only for fixed-point multiplication.
1.1 root 3328:
3329: @item (div:@var{m} @var{x} @var{y})
3330: Represents the quotient in signed division of @var{x} by @var{y},
3331: carried out in machine mode @var{m}. If @var{m} is a floating-point
3332: mode, it represents the exact quotient; otherwise, the integerized
3333: quotient. If @var{x} and @var{y} are both valid for mode @var{m},
3334: this is ordinary size-preserving division. Some machines have
3335: division instructions in which the operands and quotient widths are
3336: not all the same; such instructions are represented by @samp{div}
3337: expressions in which the machine modes are not all the same.
3338:
3339: @item (udiv:@var{m} @var{x} @var{y})
3340: Like @samp{div} but represents unsigned division.
3341:
3342: @item (mod:@var{m} @var{x} @var{y})
3343: @itemx (umod:@var{m} @var{x} @var{y})
3344: Like @samp{div} and @samp{udiv} but represent the remainder instead of
3345: the quotient.
3346:
3347: @item (not:@var{m} @var{x})
3348: Represents the bitwise complement of the value represented by @var{x},
3349: carried out in mode @var{m}, which must be a fixed-point machine mode.
3350: @var{x} must be valid for mode @var{m}, which must be a fixed-point mode.
3351:
3352: @item (and:@var{m} @var{x} @var{y})
3353: Represents the bitwise logical-and of the values represented by
3354: @var{x} and @var{y}, carried out in machine mode @var{m}. This is
3355: valid only if @var{x} and @var{y} both are valid for mode @var{m},
3356: which must be a fixed-point mode.
3357:
3358: @item (ior:@var{m} @var{x} @var{y})
3359: Represents the bitwise inclusive-or of the values represented by
3360: @var{x} and @var{y}, carried out in machine mode @var{m}. This is
3361: valid only if @var{x} and @var{y} both are valid for mode @var{m},
3362: which must be a fixed-point mode.
3363:
3364: @item (xor:@var{m} @var{x} @var{y})
3365: Represents the bitwise exclusive-or of the values represented by
3366: @var{x} and @var{y}, carried out in machine mode @var{m}. This is
3367: valid only if @var{x} and @var{y} both are valid for mode @var{m},
3368: which must be a fixed-point mode.
3369:
3370: @item (lshift:@var{m} @var{x} @var{c})
3371: Represents the result of logically shifting @var{x} left by @var{c}
3372: places. @var{x} must be valid for the mode @var{m}, a fixed-point
3373: machine mode. @var{c} must be valid for a fixed-point mode;
3374: which mode is determined by the mode called for in the machine
3375: description entry for the left-shift instruction. For example,
3376: on the Vax, the mode of @var{c} is @code{QImode} regardless of @var{m}.
3377:
3378: On some machines, negative values of @var{c} may be meaningful; this
1.1.1.2 root 3379: is why logical left shift and arithmetic left shift are distinguished.
1.1 root 3380: For example, Vaxes have no right-shift instructions, and right shifts
3381: are represented as left-shift instructions whose counts happen
3382: to be negative constants or else computed (in a previous instruction)
3383: by negation.
3384:
3385: @item (ashift:@var{m} @var{x} @var{c})
3386: Like @samp{lshift} but for arithmetic left shift.
3387:
3388: @item (lshiftrt:@var{m} @var{x} @var{c})
3389: @itemx (ashiftrt:@var{m} @var{x} @var{c})
3390: Like @samp{lshift} and @samp{ashift} but for right shift.
3391:
3392: @item (rotate:@var{m} @var{x} @var{c})
3393: @itemx (rotatert:@var{m} @var{x} @var{c})
3394: Similar but represent left and right rotate.
3395:
3396: @item (abs:@var{m} @var{x})
3397: Represents the absolute value of @var{x}, computed in mode @var{m}.
3398: @var{x} must be valid for @var{m}.
3399:
3400: @item (sqrt:@var{m} @var{x})
3401: Represents the square root of @var{x}, computed in mode @var{m}.
3402: @var{x} must be valid for @var{m}. Most often @var{m} will be
3403: a floating point mode.
1.1.1.2 root 3404:
3405: @item (ffs:@var{m} @var{x})
3406: Represents the one plus the index of the least significant 1-bit in
3407: @var{x}, represented as an integer of mode @var{m}. (The value is
3408: zero if @var{x} is zero.) The mode of @var{x} need not be @var{m};
3409: depending on the target machine, various mode combinations may be
3410: valid.
1.1 root 3411: @end table
3412:
3413: @node Comparisons, Bit Fields, Arithmetic, RTL
3414: @section Comparison Operations
3415:
3416: Comparison operators test a relation on two operands and are considered to
3417: represent the value 1 if the relation holds, or zero if it does not. The
3418: mode of the comparison is determined by the operands; they must both be
3419: valid for a common machine mode. A comparison with both operands constant
3420: would be invalid as the machine mode could not be deduced from it, but such
1.1.1.2 root 3421: a comparison should never exist in RTL due to constant folding.
1.1 root 3422:
3423: Inequality comparisons come in two flavors, signed and unsigned. Thus,
1.1.1.2 root 3424: there are distinct expression codes @samp{gt} and @samp{gtu} for signed and
1.1 root 3425: unsigned greater-than. These can produce different results for the same
3426: pair of integer values: for example, 1 is signed greater-than -1 but not
3427: unsigned greater-than, because -1 when regarded as unsigned is actually
1.1.1.2 root 3428: @code{0xffffffff} which is greater than 1.
1.1 root 3429:
3430: The signed comparisons are also used for floating point values. Floating
3431: point comparisons are distinguished by the machine modes of the operands.
3432:
3433: The comparison operators may be used to compare the condition codes
1.1.1.2 root 3434: @code{(cc0)} against zero, as in @code{(eq (cc0) (const_int 0))}. Such a
3435: construct actually refers to the result of the preceding instruction in
3436: which the condition codes were set. The above example stands for 1 if the
3437: condition codes were set to say ``zero'' or ``equal'', 0 otherwise.
3438: Although the same comparison operators are used for this as may be used in
3439: other contexts on actual data, no confusion can result since the machine
3440: description would never allow both kinds of uses in the same context.
1.1 root 3441:
3442: @table @code
3443: @item (eq @var{x} @var{y})
3444: 1 if the values represented by @var{x} and @var{y} are equal,
3445: otherwise 0.
3446:
3447: @item (ne @var{x} @var{y})
3448: 1 if the values represented by @var{x} and @var{y} are not equal,
3449: otherwise 0.
3450:
3451: @item (gt @var{x} @var{y})
3452: 1 if the @var{x} is greater than @var{y}. If they are fixed-point,
3453: the comparison is done in a signed sense.
3454:
3455: @item (gtu @var{x} @var{y})
3456: Like @samp{gt} but does unsigned comparison, on fixed-point numbers only.
3457:
3458: @item (lt @var{x} @var{y})
3459: @item (ltu @var{x} @var{y})
3460: Like @samp{gt} and @samp{gtu} but test for ``less than''.
3461:
3462: @item (ge @var{x} @var{y})
3463: @item (geu @var{x} @var{y})
3464: Like @samp{gt} and @samp{gtu} but test for ``greater than or equal''.
3465:
3466: @item (le @var{x} @var{y})
3467: @item (leu @var{x} @var{y})
3468: Like @samp{gt} and @samp{gtu} but test for ``less than or equal''.
3469:
3470: @item (if_then_else @var{cond} @var{then} @var{else})
3471: This is not a comparison operation but is listed here because it is
3472: always used in conjunction with a comparison operation. To be
3473: precise, @var{cond} is a comparison expression. This expression
3474: represents a choice, according to @var{cond}, between the value
3475: represented by @var{then} and the one represented by @var{else}.
3476:
3477: On most machines, @samp{if_then_else} expressions are valid only
3478: to express conditional jumps.
3479: @end table
3480:
3481: @node Bit Fields, Conversions, Comparisons, RTL
3482: @section Bit-fields
3483:
3484: Special expression codes exist to represent bit-field instructions.
1.1.1.2 root 3485: These types of expressions are lvalues in RTL; they may appear
1.1 root 3486: on the left side of a assignment, indicating insertion of a value
3487: into the specified bit field.
3488:
3489: @table @code
3490: @item (sign_extract:SI @var{loc} @var{size} @var{pos})
3491: This represents a reference to a sign-extended bit-field contained or
3492: starting in @var{loc} (a memory or register reference). The bit field
3493: is @var{size} bits wide and starts at bit @var{pos}. The compilation
1.1.1.2 root 3494: option @code{BITS_BIG_ENDIAN} says which end of the memory unit
1.1 root 3495: @var{pos} counts from.
3496:
3497: Which machine modes are valid for @var{loc} depends on the machine,
3498: but typically @var{loc} should be a single byte when in memory
3499: or a full word in a register.
3500:
1.1.1.2 root 3501: @item (zero_extract:SI @var{loc} @var{size} @var{pos})
1.1 root 3502: Like @samp{sign_extract} but refers to an unsigned or zero-extended
3503: bit field. The same sequence of bits are extracted, but they
3504: are filled to an entire word with zeros instead of by sign-extension.
3505: @end table
3506:
3507: @node Conversions, RTL Declarations, Bit Fields, RTL
3508: @section Conversions
3509:
3510: All conversions between machine modes must be represented by
3511: explicit conversion operations. For example, an expression
1.1.1.2 root 3512: which is the sum of a byte and a full word cannot be written as
1.1 root 3513: @code{(plus:SI (reg:QI 34) (reg:SI 80))} because the @samp{plus}
3514: operation requires two operands of the same machine mode.
3515: Therefore, the byte-sized operand is enclosed in a conversion
3516: operation, as in
3517:
3518: @example
3519: (plus:SI (sign_extend:SI (reg:QI 34)) (reg:SI 80))
3520: @end example
3521:
3522: The conversion operation is not a mere placeholder, because there
3523: may be more than one way of converting from a given starting mode
3524: to the desired final mode. The conversion operation code says how
3525: to do it.
3526:
3527: @table @code
3528: @item (sign_extend:@var{m} @var{x})
3529: Represents the result of sign-extending the value @var{x}
3530: to machine mode @var{m}. @var{m} must be a fixed-point mode
3531: and @var{x} a fixed-point value of a mode narrower than @var{m}.
3532:
3533: @item (zero_extend:@var{m} @var{x})
3534: Represents the result of zero-extending the value @var{x}
3535: to machine mode @var{m}. @var{m} must be a fixed-point mode
3536: and @var{x} a fixed-point value of a mode narrower than @var{m}.
3537:
3538: @item (float_extend:@var{m} @var{x})
3539: Represents the result of extending the value @var{x}
3540: to machine mode @var{m}. @var{m} must be a floating point mode
3541: and @var{x} a floating point value of a mode narrower than @var{m}.
3542:
3543: @item (truncate:@var{m} @var{x})
3544: Represents the result of truncating the value @var{x}
3545: to machine mode @var{m}. @var{m} must be a fixed-point mode
3546: and @var{x} a fixed-point value of a mode wider than @var{m}.
3547:
3548: @item (float_truncate:@var{m} @var{x})
3549: Represents the result of truncating the value @var{x}
3550: to machine mode @var{m}. @var{m} must be a floating point mode
3551: and @var{x} a floating point value of a mode wider than @var{m}.
3552:
3553: @item (float:@var{m} @var{x})
1.1.1.2 root 3554: Represents the result of converting fixed point value @var{x},
3555: regarded as signed, to floating point mode @var{m}.
3556:
3557: @item (unsigned_float:@var{m} @var{x})
3558: Represents the result of converting fixed point value @var{x},
3559: regarded as unsigned, to floating point mode @var{m}.
1.1 root 3560:
3561: @item (fix:@var{m} @var{x})
1.1.1.2 root 3562: When @var{m} is a fixed point mode, represents the result of
3563: converting floating point value @var{x} to mode @var{m}, regarded as
3564: signed. How rounding is done is not specified, so this operation may
3565: be used validly in compiling C code only for integer-valued operands.
3566:
3567: @item (unsigned_fix:@var{m} @var{x})
3568: Represents the result of converting floating point value @var{x} to
3569: fixed point mode @var{m}, regarded as unsigned. How rounding is done
3570: is not specified.
1.1 root 3571:
1.1.1.2 root 3572: @item (fix:@var{m} @var{x})
3573: When @var{m} is a floating point mode, represents the result of
3574: converting floating point value @var{x} (valid for mode @var{m}) to an
3575: integer, still represented in floating point mode @var{m}, by rounding
3576: towards zero.
1.1 root 3577: @end table
3578:
3579: @node RTL Declarations, Side Effects, Conversions, RTL
3580: @section Declarations
3581:
3582: Declaration expression codes do not represent arithmetic operations
3583: but rather state assertions about their operands.
3584:
3585: @table @code
3586: @item (strict_low_part (subreg:@var{m} (reg:@var{n} @var{r}) 0))
3587: This expression code is used in only one context: operand 0 of a
3588: @samp{set} expression. In addition, the operand of this expression
3589: must be a @samp{subreg} expression.
3590:
3591: The presence of @samp{strict_low_part} says that the part of the
1.1.1.2 root 3592: register which is meaningful in mode @var{n}, but is not part of
3593: mode @var{m}, is not to be altered. Normally, an assignment to such
1.1 root 3594: a subreg is allowed to have undefined effects on the rest of the
3595: register when @var{m} is less than a word.
3596: @end table
3597:
3598: @node Side Effects, Incdec, RTL Declarations, RTL
3599: @section Side Effect Expressions
3600:
3601: The expression codes described so far represent values, not actions.
3602: But machine instructions never produce values; they are meaningful
3603: only for their side effects on the state of the machine. Special
3604: expression codes are used to represent side effects.
3605:
3606: The body of an instruction is always one of these side effect codes;
3607: the codes described above, which represent values, appear only as
3608: the operands of these.
3609:
3610: @table @code
3611: @item (set @var{lval} @var{x})
3612: Represents the action of storing the value of @var{x} into the place
3613: represented by @var{lval}. @var{lval} must be an expression
3614: representing a place that can be stored in: @samp{reg} (or
3615: @samp{subreg} or @samp{strict_low_part}), @samp{mem}, @samp{pc} or
1.1.1.2 root 3616: @samp{cc0}.@refill
1.1 root 3617:
3618: If @var{lval} is a @samp{reg}, @samp{subreg} or @samp{mem}, it has a
1.1.1.2 root 3619: machine mode; then @var{x} must be valid for that mode.@refill
1.1 root 3620:
3621: If @var{lval} is a @samp{reg} whose machine mode is less than the full
3622: width of the register, then it means that the part of the register
3623: specified by the machine mode is given the specified value and the
3624: rest of the register receives an undefined value. Likewise, if
3625: @var{lval} is a @samp{subreg} whose machine mode is narrower than
3626: @code{SImode}, the rest of the register can be changed in an undefined way.
3627:
3628: If @var{lval} is a @samp{strict_low_part} of a @samp{subreg}, then the
3629: part of the register specified by the machine mode of the
3630: @samp{subreg} is given the value @var{x} and the rest of the register
1.1.1.2 root 3631: is not changed.@refill
1.1 root 3632:
3633: If @var{lval} is @code{(cc0)}, it has no machine mode, and @var{x} may
1.1.1.2 root 3634: have any mode. This represents a ``test'' or ``compare'' instruction.@refill
1.1 root 3635:
3636: If @var{lval} is @code{(pc)}, we have a jump instruction, and the
3637: possibilities for @var{x} are very limited. It may be a
3638: @samp{label_ref} expression (unconditional jump). It may be an
3639: @samp{if_then_else} (conditional jump), in which case either the
3640: second or the third operand must be @code{(pc)} (for the case which
3641: does not jump) and the other of the two must be a @samp{label_ref}
3642: (for the case which does jump). @var{x} may also be a @samp{mem} or
3643: @code{(plus:SI (pc) @var{y})}, where @var{y} may be a @samp{reg} or a
3644: @samp{mem}; these unusual patterns are used to represent jumps through
1.1.1.2 root 3645: branch tables.@refill
1.1 root 3646:
3647: @item (return)
1.1.1.2 root 3648: Represents a return from the current function, on machines where this
3649: can be done with one instruction, such as Vaxes. On machines where a
3650: multi-instruction ``epilogue'' must be executed in order to return
3651: from the function, returning is done by jumping to a label which
3652: precedes the epilogue, and the @samp{return} expression code is never
3653: used.
1.1 root 3654:
3655: @item (call @var{function} @var{nargs})
3656: Represents a function call. @var{function} is a @samp{mem} expression
3657: whose address is the address of the function to be called. @var{nargs}
3658: is an expression representing the number of words of argument.
3659:
3660: Each machine has a standard machine mode which @var{function} must
1.1.1.2 root 3661: have. The machine description defines macro @code{FUNCTION_MODE} to
1.1 root 3662: expand into the requisite mode name. The purpose of this mode is to
3663: specify what kind of addressing is allowed, on machines where the
3664: allowed kinds of addressing depend on the machine mode being
3665: addressed.
3666:
3667: @item (clobber @var{x})
3668: Represents the storing or possible storing of an unpredictable,
3669: undescribed value into @var{x}, which must be a @samp{reg} or
3670: @samp{mem} expression.
3671:
3672: One place this is used is in string instructions that store standard
3673: values into particular hard registers. It may not be worth the
1.1.1.2 root 3674: trouble to describe the values that are stored, but it is essential to
3675: inform the compiler that the registers will be altered, lest it
1.1 root 3676: attempt to keep data in them across the string instruction.
3677:
3678: @var{x} may also be null---a null C pointer, no expression at all.
3679: Such a @code{(clobber (null))} expression means that all memory
3680: locations must be presumed clobbered.
3681:
3682: Note that the machine description classifies certain hard registers as
3683: ``call-clobbered''. All function call instructions are assumed by
3684: default to clobber these registers, so there is no need to use
3685: @samp{clobber} expressions to indicate this fact. Also, each function
3686: call is assumed to have the potential to alter any memory location.
3687:
3688: @item (use @var{x})
1.1.1.2 root 3689: Represents the use of the value of @var{x}. It indicates that the
3690: value in @var{x} at this point in the program is needed, even though
3691: it may not be apparent why this is so. Therefore, the compiler will
3692: not attempt to delete instructions whose only effect is to store a
3693: value in @var{x}. @var{x} must be a @samp{reg} expression.
1.1 root 3694:
3695: @item (parallel [@var{x0} @var{x1} @dots{}])
3696: Represents several side effects performed in parallel. The square
3697: brackets stand for a vector; the operand of @samp{parallel} is a
3698: vector of expressions. @var{x0}, @var{x1} and so on are individual
3699: side effects---expressions of code @samp{set}, @samp{call},
1.1.1.2 root 3700: @samp{return}, @samp{clobber} or @samp{use}.@refill
1.1 root 3701:
1.1.1.2 root 3702: ``In parallel'' means that first all the values used in the individual
3703: side-effects are computed, and second all the actual side-effects are
3704: performed. For example,
1.1 root 3705:
3706: @example
3707: (parallel [(set (reg:SI 1) (mem:SI (reg:SI 1)))
3708: (set (mem:SI (reg:SI 1)) (reg:SI 1))])
3709: @end example
3710:
3711: @noindent
3712: says unambiguously that the values of hard register 1 and the memory
3713: location addressed by it are interchanged. In both places where
3714: @code{(reg:SI 1)} appears as a memory address it refers to the value
1.1.1.2 root 3715: in register 1 @emph{before} the execution of the instruction.
3716:
3717: @item (sequence [@var{insns} @dots{}])
3718: Represents a sequence of insns. Each of the @var{insns} that appears
3719: in the vector is suitable for appearing in the chain of insns, so it
3720: must be an @samp{insn}, @samp{jump_insn}, @samp{call_insn},
3721: @samp{code_label}, @samp{barrier} or @samp{note}.
3722:
3723: A @samp{sequence} RTX never appears in an actual insn. It represents
3724: the sequence of insns that result from a @samp{define_expand}
3725: @emph{before} those insns are passed to @code{emit_insn} to insert
3726: them in the chain of insns. When actually inserted, the individual
3727: sub-insns are separated out and the @samp{sequence} is forgotten.
1.1 root 3728: @end table
3729:
1.1.1.2 root 3730: Three expression codes appear in place of a side effect, as the body of an
3731: insn, though strictly speaking they do not describe side effects as such:
1.1 root 3732:
3733: @table @code
3734: @item (asm_input @var{s})
3735: Represents literal assembler code as described by the string @var{s}.
3736:
3737: @item (addr_vec:@var{m} [@var{lr0} @var{lr1} @dots{}])
1.1.1.2 root 3738: Represents a table of jump addresses. The vector elements @var{lr0},
3739: etc., are @samp{label_ref} expressions. The mode @var{m} specifies
3740: how much space is given to each address; normally @var{m} would be
1.1 root 3741: @code{Pmode}.
3742:
3743: @item (addr_diff_vec:@var{m} @var{base} [@var{lr0} @var{lr1} @dots{}])
3744: Represents a table of jump addresses expressed as offsets from
1.1.1.2 root 3745: @var{base}. The vector elements @var{lr0}, etc., are @samp{label_ref}
3746: expressions and so is @var{base}. The mode @var{m} specifies how much
3747: space is given to each address-difference.@refill
1.1 root 3748: @end table
3749:
1.1.1.2 root 3750: @node Incdec, Assembler, Side Effects, RTL
1.1 root 3751: @section Embedded Side-Effects on Addresses
3752:
3753: Four special side-effect expression codes appear as memory addresses.
3754:
3755: @table @code
3756: @item (pre_dec:@var{m} @var{x})
3757: Represents the side effect of decrementing @var{x} by a standard
3758: amount and represents also the value that @var{x} has after being
3759: decremented. @var{x} must be a @samp{reg} or @samp{mem}, but most
3760: machines allow only a @samp{reg}. @var{m} must be the machine mode
1.1.1.2 root 3761: for pointers on the machine in use. The amount @var{x} is decremented
1.1 root 3762: by is the length in bytes of the machine mode of the containing memory
3763: reference of which this expression serves as the address. Here is an
1.1.1.2 root 3764: example of its use:@refill
1.1 root 3765:
3766: @example
3767: (mem:DF (pre_dec:SI (reg:SI 39)))
3768: @end example
3769:
3770: @noindent
3771: This says to decrement pseudo register 39 by the length of a @code{DFmode}
3772: value and use the result to address a @code{DFmode} value.
3773:
3774: @item (pre_inc:@var{m} @var{x})
3775: Similar, but specifies incrementing @var{x} instead of decrementing it.
3776:
3777: @item (post_dec:@var{m} @var{x})
3778: Represents the same side effect as @samp{pre_decrement} but a different
3779: value. The value represented here is the value @var{x} has @i{before}
3780: being decremented.
3781:
3782: @item (post_inc:@var{m} @var{x})
3783: Similar, but specifies incrementing @var{x} instead of decrementing it.
3784: @end table
3785:
3786: These embedded side effect expressions must be used with care. Instruction
3787: patterns may not use them. Until the @samp{flow} pass of the compiler,
3788: they may occur only to represent pushes onto the stack. The @samp{flow}
3789: pass finds cases where registers are incremented or decremented in one
3790: instruction and used as an address shortly before or after; these cases are
3791: then transformed to use pre- or post-increment or -decrement.
3792:
3793: Explicit popping of the stack could be represented with these embedded
3794: side effect operators, but that would not be safe; the instruction
3795: combination pass could move the popping past pushes, thus changing
3796: the meaning of the code.
3797:
3798: An instruction that can be represented with an embedded side effect
3799: could also be represented using @samp{parallel} containing an additional
3800: @samp{set} to describe how the address register is altered. This is not
3801: done because machines that allow these operations at all typically
3802: allow them wherever a memory address is called for. Describing them as
3803: additional parallel stores would require doubling the number of entries
3804: in the machine description.
3805:
1.1.1.2 root 3806: @node Assembler, Insns, IncDec, RTL
3807: @section Assembler Instructions as Expressions
3808:
3809: The RTX code @samp{asm_operands} represents a value produced by a
3810: user-specified assembler instruction. It is used to represent
3811: an @code{asm} statement with arguments. An @code{asm} statement with
3812: a single output operand, like this:
3813:
3814: @example
3815: asm ("foo %1,%2,%0" : "a" (outputvar) : "g" (x + y), "di" (*z));
3816: @end example
3817:
3818: @noindent
3819: is represented using a single @samp{asm_operands} RTX which represents
3820: the value that is stored in @code{outputvar}:
3821:
3822: @example
3823: (set @var{rtx-for-outputvar}
3824: (asm_operands "foo %1,%2,%0" "a" 0
3825: [@var{rtx-for-addition-result} @var{rtx-for-*z}]
3826: [(asm_input:@var{m1} "g")
3827: (asm_input:@var{m2} "di")]))
3828: @end example
3829:
3830: @noindent
3831: Here the operands of the @samp{asm_operands} RTX are the assembler
3832: template string, the output-operand's constraint, the index-number of the
3833: output operand among the output operands specified, a vector of input
3834: operand RTX's, and a vector of input-operand modes and constraints. The
3835: mode @var{m1} is the mode of the sum @code{x+y}; @var{m2} is that of
3836: @code{*z}.
3837:
3838: When an @code{asm} statement has multiple output values, its insn has
3839: several such @samp{set} RTX's inside of a @samp{parallel}. Each @samp{set}
3840: contains a @samp{asm_operands}; all of these share the same assembler
3841: template and vectors, but each contains the constraint for the respective
3842: output operand. They are also distinguished by the output-operand index
3843: number, which is 0, 1, @dots{} for successive output operands.
3844:
3845: @node Insns, Calls, Assembler, RTL
1.1 root 3846: @section Insns
3847:
3848: The RTL representation of the code for a function is a doubly-linked
3849: chain of objects called @dfn{insns}. Insns are expressions with
3850: special codes that are used for no other purpose. Some insns are
3851: actual instructions; others represent dispatch tables for @code{switch}
3852: statements; others represent labels to jump to or various sorts of
1.1.1.2 root 3853: declarative information.
1.1 root 3854:
1.1.1.2 root 3855: In addition to its own specific data, each insn must have a unique id-number
1.1 root 3856: that distinguishes it from all other insns in the current function, and
3857: chain pointers to the preceding and following insns. These three fields
3858: occupy the same position in every insn, independent of the expression code
3859: of the insn. They could be accessed with @code{XEXP} and @code{XINT},
3860: but instead three special macros are always used:
3861:
3862: @table @code
3863: @item INSN_UID (@var{i})
3864: Accesses the unique id of insn @var{i}.
3865:
3866: @item PREV_INSN (@var{i})
3867: Accesses the chain pointer to the insn preceding @var{i}.
3868: If @var{i} is the first insn, this is a null pointer.
3869:
3870: @item NEXT_INSN (@var{i})
3871: Accesses the chain pointer to the insn following @var{i}.
3872: If @var{i} is the last insn, this is a null pointer.
3873: @end table
3874:
3875: The @code{NEXT_INSN} and @code{PREV_INSN} pointers must always
3876: correspond: if @var{i} is not the first insn,
3877:
3878: @example
3879: NEXT_INSN (PREV_INSN (@var{insn})) == @var{insn}
3880: @end example
3881:
3882: @noindent
3883: is always true.
3884:
3885: Every insn has one of the following six expression codes:
3886:
1.1.1.2 root 3887: @table @samp
1.1 root 3888: @item insn
3889: The expression code @samp{insn} is used for instructions that do not jump
3890: and do not do function calls. Insns with code @samp{insn} have four
3891: additional fields beyond the three mandatory ones listed above.
3892: These four are described in a table below.
3893:
3894: @item jump_insn
3895: The expression code @samp{jump_insn} is used for instructions that may jump
3896: (or, more generally, may contain @samp{label_ref} expressions).
3897: @samp{jump_insn} insns have the same extra fields as @samp{insn} insns,
3898: accessed in the same way.
3899:
3900: @item call_insn
3901: The expression code @samp{call_insn} is used for instructions that may do
3902: function calls. It is important to distinguish these instructions because
3903: they imply that certain registers and memory locations may be altered
3904: unpredictably.
3905:
3906: @samp{call_insn} insns have the same extra fields as @samp{insn} insns,
3907: accessed in the same way.
3908:
3909: @item code_label
3910: A @samp{code_label} insn represents a label that a jump insn can jump to.
3911: It contains one special field of data in addition to the three standard ones.
3912: It is used to hold the @dfn{label number}, a number that identifies this
3913: label uniquely among all the labels in the compilation (not just in the
3914: current function). Ultimately, the label is represented in the assembler
3915: output as an assembler label @samp{L@var{n}} where @var{n} is the label number.
3916:
3917: @item barrier
3918: Barriers are placed in the instruction stream after unconditional
3919: jump instructions to indicate that the jumps are unconditional.
3920: They contain no information beyond the three standard fields.
3921:
3922: @item note
3923: @samp{note} insns are used to represent additional debugging and
1.1.1.2 root 3924: declarative information. They contain two nonstandard fields, an
1.1 root 3925: integer which is accessed with the macro @code{NOTE_LINE_NUMBER} and a
3926: string accessed with @code{NOTE_SOURCE_FILE}.
3927:
3928: If @code{NOTE_LINE_NUMBER} is positive, the note represents the
3929: position of a source line and @code{NOTE_SOURCE_FILE} is the source file name
3930: that the line came from. These notes control generation of line
3931: number data in the assembler output.
3932:
3933: Otherwise, @code{NOTE_LINE_NUMBER} is not really a line number but a
3934: code with one of the following values (and @code{NOTE_SOURCE_FILE}
3935: must contain a null pointer):
3936:
3937: @table @code
3938: @item NOTE_INSN_DELETED
3939: Such a note is completely ignorable. Some passes of the compiler
3940: delete insns by altering them into notes of this kind.
3941:
3942: @item NOTE_INSN_BLOCK_BEG
3943: @itemx NOTE_INSN_BLOCK_END
3944: These types of notes indicate the position of the beginning and end
3945: of a level of scoping of variable names. They control the output
3946: of debugging information.
3947:
3948: @item NOTE_INSN_LOOP_BEG
3949: @itemx NOTE_INSN_LOOP_END
3950: These types of notes indicate the position of the beginning and end
3951: of a @code{while} or @code{for} loop. They enable the loop optimizer
3952: to find loops quickly.
3953: @end table
3954: @end table
3955:
3956: Here is a table of the extra fields of @samp{insn}, @samp{jump_insn}
3957: and @samp{call_insn} insns:
3958:
3959: @table @code
3960: @item PATTERN (@var{i})
3961: An expression for the side effect performed by this insn.
3962:
3963: @item REG_NOTES (@var{i})
3964: A list (chain of @samp{expr_list} expressions) giving information
3965: about the usage of registers in this insn. This list is set up by the
1.1.1.2 root 3966: flow analysis pass; it is a null pointer until then.
1.1 root 3967:
3968: @item LOG_LINKS (@var{i})
3969: A list (chain of @samp{insn_list} expressions) of previous ``related''
3970: insns: insns which store into registers values that are used for the
3971: first time in this insn. (An additional constraint is that neither a
3972: jump nor a label may come between the related insns). This list is
1.1.1.2 root 3973: set up by the flow analysis pass; it is a null pointer until then.
1.1 root 3974:
3975: @item INSN_CODE (@var{i})
3976: An integer that says which pattern in the machine description matches
3977: this insn, or -1 if the matching has not yet been attempted.
3978:
3979: Such matching is never attempted and this field is not used on an insn
3980: whose pattern consists of a single @samp{use}, @samp{clobber},
3981: @samp{asm}, @samp{addr_vec} or @samp{addr_diff_vec} expression.
3982: @end table
3983:
3984: The @code{LOG_LINKS} field of an insn is a chain of @samp{insn_list}
3985: expressions. Each of these has two operands: the first is an insn,
3986: and the second is another @samp{insn_list} expression (the next one in
3987: the chain). The last @samp{insn_list} in the chain has a null pointer
3988: as second operand. The significant thing about the chain is which
1.1.1.2 root 3989: insns appear in it (as first operands of @samp{insn_list}
1.1 root 3990: expressions). Their order is not significant.
3991:
3992: The @code{REG_NOTES} field of an insn is a similar chain but of
1.1.1.2 root 3993: @samp{expr_list} expressions instead of @samp{insn_list}. There are four
3994: kinds of register notes, which are distinguished by the machine mode of the
3995: @samp{expr_list}, which a register note is really understood as being an
3996: @code{enum reg_note}. The first operand @var{op} of the @samp{expr_list}
3997: is data whose meaning depends on the kind of note. Here are the four
3998: kinds:
1.1 root 3999:
4000: @table @code
4001: @item REG_DEAD
1.1.1.2 root 4002: The register @var{op} dies in this insn; that is to say, altering the
4003: value immediately after this insn would not affect the future behavior
4004: of the program.
1.1 root 4005:
4006: @item REG_INC
1.1.1.2 root 4007: The register @var{op} is incremented (or decremented; at this level
1.1 root 4008: there is no distinction) by an embedded side effect inside this insn.
1.1.1.2 root 4009: This means it appears in a @code{POST_INC}, @code{PRE_INC},
4010: @code{POST_DEC} or @code{PRE_DEC} RTX.
1.1 root 4011:
1.1.1.2 root 4012: @item REG_EQUIV
4013: The register that is set by this insn will be equal to @var{op} at run
4014: time, and could validly be replaced in all its occurrences by
4015: @var{op}. (``Validly'' here refers to the data flow of the program;
4016: simple replacement may make some insns invalid.)
4017:
4018: The value which the insn explicitly copies into the register may look
4019: different from @var{op}, but they will be equal at run time.
4020:
4021: For example, when a constant is loaded into a register that is never
4022: assigned any other value, this kind of note is used.
4023:
4024: When a parameter is copied into a pseudo-register at entry to a function,
4025: a note of this kind records that the register is equivalent to the stack
4026: slot where the parameter was passed. Although in this case the register
4027: may be set by other insns, it is still valid to replace the register
4028: by the stack slot throughout the function.
4029:
4030: @item REG_EQUAL
4031: The register that is set by this insn will be equal to @var{op} at run
4032: time at the end of this insn (but not necessarily elsewhere in the
4033: function).
4034:
4035: The RTX @var{op} is typically an arithmetic expression. For example,
4036: when a sequence of insns such as a library call is used to perform an
4037: arithmetic operation, this kind of note is attached to the insn that
4038: produces or copies the final value. It tells the CSE pass how to
4039: think of that value.
4040:
4041: @item REG_RETVAL
4042: This insn copies the value of a library call, and @var{op} is the
4043: first insn that was generated to set up the arguments for the library
4044: call.
4045:
4046: Flow analysis uses this note to delete all of a library call whose
4047: result is dead.
1.1 root 4048:
4049: @item REG_WAS_0
1.1.1.2 root 4050: The register @var{op} contained zero before this insn. You can rely
1.1 root 4051: on this note if it is present; its absence implies nothing.
4052: @end table
4053:
4054: (The only difference between the expression codes @samp{insn_list} and
4055: @samp{expr_list} is that the first operand of an @samp{insn_list} is
4056: assumed to be an insn and is printed in debugging dumps as the insn's
4057: unique id; the first operand of an @samp{expr_list} is printed in the
4058: ordinary way as an expression.)
4059:
1.1.1.2 root 4060: @node Calls, Sharing, Insns, RTL
4061: @section RTL Representation of Function-Call Insns
4062:
4063: Insns that call subroutines have the RTL expression code @samp{call_insn}.
4064: These insns must satisfy special rules, and their bodies must use a special
4065: RTL expression code, @samp{call}.
4066:
4067: A @samp{call} expression has two operands, as follows:
4068:
4069: @example
4070: (call @var{nbytes} (mem:@var{fm} @var{addr}))
4071: @end example
4072:
4073: @noindent
4074: Here @var{nbytes} is an operand that represents the number of bytes of
4075: argument data being passed to the subroutine, @var{fm} is a machine mode
4076: (which must equal as the definition of the @code{FUNCTION_MODE} macro in
4077: the machine description) and @var{addr} represents the address of the
4078: subroutine.
4079:
4080: For a subroutine that returns no value, the @samp{call} RTX as shown above
4081: is the entire body of the insn.
4082:
4083: For a subroutine that returns a value whose mode is not @code{BLKmode},
4084: the value is returned in a hard register. If this register's number is
4085: @var{r}, then the body of the call insn looks like this:
4086:
4087: @example
4088: (set (reg:@var{m} @var{r})
4089: (call @var{nbytes} (mem:@var{fm} @var{addr})))
4090: @end example
4091:
4092: @noindent
4093: This RTL expression makes it clear (to the optimizer passes) that the
4094: appropriate register receives a useful value in this insn.
4095:
4096: Immediately after RTL generation, if the value of the subroutine is
4097: actually used, this call insn is always followed closely by an insn which
1.1.1.3 root 4098: refers to the register @var{r}. This remains true through all the
4099: optimizer passes until cross jumping occurs.
4100:
4101: The following insn has one of two forms. Either it copies the value into a
4102: pseudo-register, like this:
1.1.1.2 root 4103:
4104: @example
4105: (set (reg:@var{m} @var{p}) (reg:@var{m} @var{r}))
4106: @end example
4107:
4108: @noindent
4109: or (in the case where the calling function will simply return whatever
4110: value the call produced, and no operation is needed to do this):
4111:
4112: @example
4113: (use (reg:@var{m} @var{r}))
4114: @end example
4115:
4116: @noindent
1.1.1.3 root 4117: Between the call insn and this following insn there may intervene only a
1.1.1.2 root 4118: stack-adjustment insn (and perhaps some @samp{note} insns).
4119:
4120: When a subroutine returns a @code{BLKmode} value, it is handled by
4121: passing to the subroutine the address of a place to store the value.
4122: So the call insn itself does not ``return'' any value, and it has the
4123: same RTL form as a call that returns nothing.
4124:
4125: @node Sharing,, Calls, RTL
1.1 root 4126: @section Structure Sharing Assumptions
4127:
4128: The compiler assumes that certain kinds of RTL expressions are unique;
4129: there do not exist two distinct objects representing the same value.
4130: In other cases, it makes an opposite assumption: that no RTL expression
4131: object of a certain kind appears in more than one place in the
4132: containing structure.
4133:
4134: These assumptions refer to a single function; except for the RTL
4135: objects that describe global variables and external functions,
4136: no RTL objects are common to two functions.
4137:
4138: @itemize @bullet
4139: @item
4140: Each pseudo-register has only a single @samp{reg} object to represent it,
4141: and therefore only a single machine mode.
4142:
4143: @item
4144: For any symbolic label, there is only one @samp{symbol_ref} object
4145: referring to it.
4146:
4147: @item
4148: There is only one @samp{const_int} expression with value zero,
4149: and only one with value one.
4150:
4151: @item
4152: There is only one @samp{pc} expression.
4153:
4154: @item
4155: There is only one @samp{cc0} expression.
4156:
4157: @item
4158: There is only one @samp{const_double} expression with mode
4159: @code{SFmode} and value zero, and only one with mode @code{DFmode} and
4160: value zero.
4161:
4162: @item
1.1.1.2 root 4163: No @samp{label_ref} appears in more than one place in the RTL
4164: structure; in other words, it is safe to do a tree-walk of all the
4165: insns in the function and assume that each time a @samp{label_ref} is
4166: seen it is distinct from all others that are seen.
1.1 root 4167:
4168: @item
1.1.1.2 root 4169: Only one @samp{mem} object is normally created for each static
4170: variable or stack slot, so these objects are frequently shared in all
4171: the places they appear. However, separate but equal objects for these
4172: variables are occasionally made.
4173:
4174: @item
4175: No RTL object appears in more than one place in the RTL structure
4176: except as described above. Many passes of the compiler rely on this
4177: by assuming that they can modify RTL objects in place without unwanted
4178: side-effects on other insns.
4179:
4180: @item
4181: During initial RTL generation, shared structure is freely introduced.
4182: After all the RTL for a function has been generated, all shared
4183: structure is copied by @code{unshare_all_rtl} in @file{emit-rtl.c},
4184: after which the above rules are guaranteed to be followed.
4185:
4186: @item
4187: During the combiner pass, shared structure with an insn can exist
4188: temporarily. However, the shared structure is copied before the
4189: combiner is finished with the insn. This is done by
4190: @code{copy_substitutions} in @samp{combine.c}.
1.1 root 4191: @end itemize
4192:
4193: @node Machine Desc, Machine Macros, RTL, Top
4194: @chapter Machine Descriptions
4195:
4196: A machine description has two parts: a file of instruction patterns
4197: (@file{.md} file) and a C header file of macro definitions.
4198:
4199: The @file{.md} file for a target machine contains a pattern for each
4200: instruction that the target machine supports (or at least each instruction
4201: that is worth telling the compiler about). It may also contain comments.
4202: A semicolon causes the rest of the line to be a comment, unless the semicolon
4203: is inside a quoted string.
4204:
4205: See the next chapter for information on the C header file.
4206:
4207: @menu
4208: * Patterns:: How to write instruction patterns.
1.1.1.2 root 4209: * Example:: An explained example of a @samp{define_insn} pattern.
4210: * RTL Template:: The RTL template defines what insns match a pattern.
4211: * Output Template:: The output template says how to make assembler code
4212: from such an insn.
4213: * Output Statement:: For more generality, write C code to output
4214: the assembler code.
1.1 root 4215: * Constraints:: When not all operands are general operands.
4216: * Standard Names:: Names mark patterns to use for code generation.
1.1.1.2 root 4217: * Pattern Ordering:: When the order of patterns makes a difference.
1.1 root 4218: * Dependent Patterns:: Having one pattern may make you need another.
1.1.1.2 root 4219: * Jump Patterns:: Special considerations for patterns for jump insns.
4220: * Peephole Definitions::Defining machine-specific peephole optimizations.
4221: * Expander Definitions::Generating a sequence of several RTL insns
4222: for a standard operation.
1.1 root 4223: @end menu
4224:
4225: @node Patterns, Example, Machine Desc, Machine Desc
1.1.1.2 root 4226: @section Everything about Instruction Patterns
1.1 root 4227:
4228: Each instruction pattern contains an incomplete RTL expression, with pieces
4229: to be filled in later, operand constraints that restrict how the pieces can
4230: be filled in, and an output pattern or C code to generate the assembler
4231: output, all wrapped up in a @samp{define_insn} expression.
4232:
1.1.1.2 root 4233: A @samp{define_insn} is an RTL expression containing four operands:
1.1 root 4234:
4235: @enumerate
4236: @item
4237: An optional name. The presence of a name indicate that this instruction
4238: pattern can perform a certain standard job for the RTL-generation
4239: pass of the compiler. This pass knows certain names and will use
4240: the instruction patterns with those names, if the names are defined
4241: in the machine description.
4242:
4243: The absence of a name is indicated by writing an empty string
4244: where the name should go. Nameless instruction patterns are never
4245: used for generating RTL code, but they may permit several simpler insns
4246: to be combined later on.
4247:
4248: Names that are not thus known and used in RTL-generation have no
4249: effect; they are equivalent to no name at all.
4250:
4251: @item
1.1.1.2 root 4252: The @dfn{RTL template} (@pxref{RTL Template}) is a vector of
4253: incomplete RTL expressions which show what the instruction should look
4254: like. It is incomplete because it may contain @samp{match_operand}
4255: and @samp{match_dup} expressions that stand for operands of the
1.1 root 4256: instruction.
4257:
4258: If the vector has only one element, that element is what the
4259: instruction should look like. If the vector has multiple elements,
4260: then the instruction looks like a @samp{parallel} expression
4261: containing that many elements as described.
4262:
4263: @item
4264: A condition. This is a string which contains a C expression that is
4265: the final test to decide whether an insn body matches this pattern.
4266:
4267: For a named pattern, the condition (if present) may not depend on
4268: the data in the insn being matched, but only the target-machine-type
4269: flags. The compiler needs to test these conditions during
4270: initialization in order to learn exactly which named instructions are
4271: available in a particular run.
4272:
4273: For nameless patterns, the condition is applied only when matching an
4274: individual insn, and only after the insn has matched the pattern's
4275: recognition template. The insn's operands may be found in the vector
4276: @code{operands}.
4277:
4278: @item
1.1.1.2 root 4279: The @dfn{output template}: a string that says how to output matching
4280: insns as assembler code. @samp{%} in this string specifies where
4281: to substitute the value of an operand. @xref{Output Template}.
4282:
4283: When simple substitution isn't general enough, you can specify a piece
4284: of C code to compute the output. @xref{Output Statement}.
4285: @end enumerate
4286:
4287: @node Example, RTL Template, Patterns, Machine Desc
4288: @section Example of @samp{define_insn}
4289:
4290: Here is an actual example of an instruction pattern, for the 68000/68020.
4291:
4292: @example
4293: (define_insn "tstsi"
4294: [(set (cc0)
4295: (match_operand:SI 0 "general_operand" "rm"))]
4296: ""
4297: "*
4298: @{ if (TARGET_68020 || ! ADDRESS_REG_P (operands[0]))
4299: return \"tstl %0\";
4300: return \"cmpl #0,%0\"; @}")
4301: @end example
1.1 root 4302:
1.1.1.2 root 4303: This is an instruction that sets the condition codes based on the value of
4304: a general operand. It has no condition, so any insn whose RTL description
4305: has the form shown may be handled according to this pattern. The name
4306: @samp{tstsi} means ``test a @code{SImode} value'' and tells the RTL generation
4307: pass that, when it is necessary to test such a value, an insn to do so
4308: can be constructed using this pattern.
1.1 root 4309:
1.1.1.2 root 4310: The output control string is a piece of C code which chooses which
4311: output template to return based on the kind of operand and the specific
4312: type of CPU for which code is being generated.
1.1 root 4313:
1.1.1.2 root 4314: @samp{"rm"} is an operand constraint. Its meaning is explained below.
1.1 root 4315:
1.1.1.2 root 4316: @node RTL Template, Output Template, Example, Machine Desc
4317: @section RTL Template for Generating and Recognizing Insns
1.1 root 4318:
1.1.1.2 root 4319: The RTL template is used to define which insns match the particular pattern
4320: and how to find their operands. For named patterns, the RTL template also
4321: says how to construct an insn from specified operands.
4322:
4323: Construction involves substituting specified operands into a copy of the
4324: template. Matching involves determining the values that serve as the
4325: operands in the insn being matched. Both of these activities are
4326: controlled by special expression types that direct matching and
4327: substitution of the operands.
1.1 root 4328:
4329: @table @code
4330: @item (match_operand:@var{m} @var{n} @var{testfn} @var{constraint})
4331: This expression is a placeholder for operand number @var{n} of
4332: the insn. When constructing an insn, operand number @var{n}
4333: will be substituted at this point. When matching an insn, whatever
4334: appears at this position in the insn will be taken as operand
4335: number @var{n}; but it must satisfy @var{testfn} or this instruction
4336: pattern will not match at all.
4337:
4338: Operand numbers must be chosen consecutively counting from zero in
4339: each instruction pattern. There may be only one @samp{match_operand}
1.1.1.7 ! root 4340: expression in the pattern for each operand number. Usually operands
! 4341: are numbered in the order of appearance in @samp{match_operand}
! 4342: expressions.
1.1 root 4343:
4344: @var{testfn} is a string that is the name of a C function that accepts
4345: two arguments, a machine mode and an expression. During matching,
4346: the function will be called with @var{m} as the mode argument
4347: and the putative operand as the other argument. If it returns zero,
4348: this instruction pattern fails to match. @var{testfn} may be
4349: an empty string; then it means no test is to be done on the operand.
4350:
4351: Most often, @var{testfn} is @code{"general_operand"}. It checks
4352: that the putative operand is either a constant, a register or a
4353: memory reference, and that it is valid for mode @var{m}.
4354:
1.1.1.2 root 4355: For an operand that must be a register, @var{testfn} should be
4356: @code{"register_operand"}. This prevents GNU CC from creating insns
4357: that have memory references in these operands, insns which would only
4358: have to be taken apart in the reload pass.
4359:
4360: For an operand that must be a constant, either @var{testfn} should be
4361: @code{"immediate_operand"}, or the instruction pattern's extra condition
4362: should check for constants, or both.
4363:
4364: @var{constraint} is explained later (@pxref{Constraints}).
1.1 root 4365:
4366: @item (match_dup @var{n})
4367: This expression is also a placeholder for operand number @var{n}.
4368: It is used when the operand needs to appear more than once in the
4369: insn.
4370:
4371: In construction, @samp{match_dup} behaves exactly like
1.1.1.2 root 4372: @samp{match_operand}: the operand is substituted into the insn being
1.1 root 4373: constructed. But in matching, @samp{match_dup} behaves differently.
4374: It assumes that operand number @var{n} has already been determined by
1.1.1.2 root 4375: a @samp{match_operand} appearing earlier in the recognition template,
1.1 root 4376: and it matches only an identical-looking expression.
4377:
4378: @item (address (match_operand:@var{m} @var{n} "address_operand" ""))
4379: This complex of expressions is a placeholder for an operand number
4380: @var{n} in a ``load address'' instruction: an operand which specifies
4381: a memory location in the usual way, but for which the actual operand
4382: value used is the address of the location, not the contents of the
4383: location.
4384:
4385: @samp{address} expressions never appear in RTL code, only in machine
4386: descriptions. And they are used only in machine descriptions that do
4387: not use the operand constraint feature. When operand constraints are
4388: in use, the letter @samp{p} in the constraint serves this purpose.
4389:
4390: @var{m} is the machine mode of the @emph{memory location being
4391: addressed}, not the machine mode of the address itself. That mode is
4392: always the same on a given target machine (it is @code{Pmode}, which
4393: normally is @code{SImode}), so there is no point in mentioning it;
4394: thus, no machine mode is written in the @samp{address} expression. If
4395: some day support is added for machines in which addresses of different
4396: kinds of objects appear differently or are used differently (such as
4397: the PDP-10), different formats would perhaps need different machine
4398: modes and these modes might be written in the @samp{address}
4399: expression.
4400: @end table
4401:
1.1.1.2 root 4402: @node Output Template, Output Statement, RTL Template, Machine Desc
4403: @section Output Templates and Operand Substitution
1.1 root 4404:
1.1.1.2 root 4405: The @dfn{output template} is a string which specifies how to output
4406: the assembler code for an instruction pattern. Most of the template
4407: is a fixed string which is output literally. The character @samp{%}
4408: is used to specify where to substitute an operand; it can also be
4409: used to identify places different variants of the assembler require
4410: different syntax.
4411:
4412: In the simplest case, a @samp{%} followed by a digit @var{n} says to output
4413: operand @var{n} at that point in the string.
4414:
4415: @samp{%} followed by a letter and a digit says to output an operand in an
4416: alternate fashion. Four letters have standard, built-in meanings described
4417: below. The machine description macro @code{PRINT_OPERAND} can define
4418: additional letters with nonstandard meanings.
4419:
4420: @samp{%c@var{digit}} can be used to substitute an operand that is a
4421: constant value without the syntax that normally indicates an immediate
4422: operand.
4423:
4424: @samp{%n@var{digit}} is like @samp{%c@var{digit}} except that the value of
4425: the constant is negated before printing.
4426:
4427: @samp{%a@var{digit}} can be used to substitute an operand as if it were a
4428: memory reference, with the actual operand treated as the address. This may
4429: be useful when outputting a ``load address'' instruction, because often the
4430: assembler syntax for such an instruction requires you to write the operand
4431: as if it were a memory reference.
4432:
4433: @samp{%l@var{digit}} is used to substitute a @code{label_ref} into a jump
4434: instruction.
4435:
4436: @samp{%} followed by a punctuation character specifies a substitution that
4437: does not use an operand. Only one case is standard: @samp{%%} outputs a
4438: @samp{%} into the assembler code. Other nonstandard cases can be
4439: defined in the @code{PRINT_OPERAND} macro.
4440:
4441: The template may generate multiple assembler instructions. Write the text
4442: for the instructions, with @samp{\;} between them.
4443:
4444: When the RTL contains two operand which are required by constraint to match
4445: each other, the output template must refer only to the lower-numbered operand.
4446: Matching operands are not always identical, and the rest of the compiler
4447: arranges to put the proper RTL expression for printing into the lower-numbered
4448: operand.
4449:
4450: One use of nonstandard letters or punctuation following @samp{%} is to
4451: distinguish between different assembler languages for the same machine; for
4452: example, Motorola syntax versus MIT syntax for the 68000. Motorola syntax
4453: requires periods in most opcode names, while MIT syntax does not. For
4454: example, the opcode @samp{movel} in MIT syntax is @samp{move.l} in Motorola
4455: syntax. The same file of patterns is used for both kinds of output syntax,
4456: but the character sequence @samp{%.} is used in each place where Motorola
4457: syntax wants a period. The @code{PRINT_OPERAND} macro for Motorola syntax
4458: defines the sequence to output a period; the macro for MIT syntax defines
4459: it to do nothing.
4460:
4461: @node Output Statement, Constraints, Output Template, Machine Desc
4462: @section C Statements for Generating Assembler Output
4463:
4464: Often a single fixed template string cannot produce correct and efficient
4465: assembler code for all the cases that are recognized by a single
4466: instruction pattern. For example, the opcodes may depend on the kinds of
4467: operands; or some unfortunate combinations of operands may require extra
4468: machine instructions.
4469:
4470: If the output control string starts with a @samp{*}, then it is not an
4471: output template but rather a piece of C program that should compute a
4472: template. It should execute a @code{return} statement to return the
4473: template-string you want. Most such templates use C string literals, which
4474: require doublequote characters to delimit them. To include these
4475: doublequote characters in the string, prefix each one with @samp{\}.
4476:
4477: The operands may be found in the array @code{operands}, whose C data type
4478: is @code{rtx []}.
4479:
4480: It is possible to output an assembler instruction and then go on to output
4481: or compute more of them, using the subroutine @code{output_asm_insn}. This
4482: receives two arguments: a template-string and a vector of operands. The
4483: vector may be @code{operands}, or it may be another array of @code{rtx}
4484: that you declare locally and initialize yourself.
4485:
4486: When an insn pattern has multiple alternatives in its constraints, often
4487: the appearance of the assembler code determined mostly by which alternative
4488: was matched. When this is so, the C code can test the variable
4489: @code{which_alternative}, which is the ordinal number of the alternative
4490: that was actually satisfied (0 for the first, 1 for the second alternative,
4491: etc.).
4492:
4493: For example, suppose there are two opcodes for storing zero, @samp{clrreg}
4494: for registers and @samp{clrmem} for memory locations. Here is how
4495: a pattern could use @code{which_alternative} to choose between them:
1.1 root 4496:
4497: @example
1.1.1.2 root 4498: (define_insn ""
4499: [(set (match_operand:SI 0 "general_operand" "r,m")
4500: (const_int 0))]
1.1 root 4501: ""
4502: "*
1.1.1.2 root 4503: return (which_alternative == 0
4504: ? \"clrreg %0\" : \"clrmem %0\");
4505: ")
1.1 root 4506: @end example
4507:
1.1.1.2 root 4508: @node Constraints, Standard Names, Output Statement, Machine Desc
1.1 root 4509: @section Operand Constraints
4510:
4511: Each @samp{match_operand} in an instruction pattern can specify a
4512: constraint for the type of operands allowed. Constraints can say whether
4513: an operand may be in a register, and which kinds of register; whether the
4514: operand can be a memory reference, and which kinds of address; whether the
4515: operand may be an immediate constant, and which possible values it may
4516: have. Constraints can also require two operands to match.
4517:
4518: @menu
4519: * Simple Constraints:: Basic use of constraints.
1.1.1.2 root 4520: * Multi-Alternative:: When an insn has two alternative constraint-patterns.
1.1 root 4521: * Class Preferences:: Constraints guide which hard register to put things in.
4522: * Modifiers:: More precise control over effects of constraints.
4523: * No Constraints:: Describing a clean machine without constraints.
4524: @end menu
4525:
4526: @node Simple Constraints, Multi-Alternative, Constraints, Constraints
4527: @subsection Simple Constraints
4528:
4529: The simplest kind of constraint is a string full of letters, each of
4530: which describes one kind of operand that is permitted. Here are
4531: the letters that are allowed:
4532:
1.1.1.2 root 4533: @table @asis
4534: @item @samp{m}
1.1 root 4535: A memory operand is allowed, with any kind of address that the machine
4536: supports in general.
4537:
1.1.1.2 root 4538: @item @samp{o}
4539: A memory operand is allowed, but only if the address is
4540: @dfn{offsetable}. This means that adding a small integer (actually,
4541: the width in bytes of the operand, as determined by its machine mode)
4542: may be added to the address and the result is also a valid memory
4543: address.
4544:
4545: For example, an address which is constant is offsetable; so is an
4546: address that is the sum of a register and a constant (as long as a
4547: slightly larger constant is also within the range of address-offsets
4548: supported by the machine); but an autoincrement or autodecrement
4549: address is not offsetable. More complicated indirect/indexed
4550: addresses may or may not be offsetable depending on the other
4551: addressing modes that the machine supports.
4552:
4553: Note that in an output operand which can be matched by another
4554: operand, the constraint letter @samp{o} is valid only when accompanied
4555: by both @samp{<} (if the target machine has predecrement addressing)
4556: and @samp{>} (if the target machine has preincrement addressing).
1.1 root 4557:
1.1.1.2 root 4558: @item @samp{<}
1.1 root 4559: A memory operand with autodecrement addressing (either predecrement or
4560: postdecrement) is allowed.
4561:
1.1.1.2 root 4562: @item @samp{>}
1.1 root 4563: A memory operand with autoincrement addressing (either preincrement or
4564: postincrement) is allowed.
4565:
1.1.1.2 root 4566: @item @samp{r}
4567: A register operand is allowed provided that it is in a general
4568: register.
1.1 root 4569:
1.1.1.2 root 4570: @item @samp{d}, @samp{a}, @samp{f}, @dots{}
1.1 root 4571: Other letters can be defined in machine-dependent fashion to stand for
4572: particular classes of registers. @samp{d}, @samp{a} and @samp{f} are
1.1.1.2 root 4573: defined on the 68000/68020 to stand for data, address and floating
4574: point registers.
1.1 root 4575:
1.1.1.2 root 4576: @item @samp{i}
1.1 root 4577: An immediate integer operand (one with constant value) is allowed.
1.1.1.2 root 4578: This includes symbolic constants whose values will be known only at
4579: assembly time.
1.1 root 4580:
1.1.1.2 root 4581: @item @samp{n}
4582: An immediate integer operand with a known numeric value is allowed.
4583: Many systems cannot support assembly-time constants for operands less
4584: than a word wide. Constraints for these operands should use @samp{n}
4585: rather than @samp{i}.
4586:
4587: @item @samp{I}, @samp{J}, @samp{K}, @dots{}
4588: Other letters in the range @samp{I} through @samp{M} may be defined in
4589: a machine-dependent fashion to permit immediate integer operands with
4590: explicit integer values in specified ranges. For example, on the
4591: 68000, @samp{I} is defined to stand for the range of values 1 to 8.
4592: This is the range permitted as a shift count in the shift
4593: instructions.
1.1 root 4594:
1.1.1.2 root 4595: @item @samp{F}
1.1 root 4596: An immediate floating operand (expression code @samp{const_double}) is
4597: allowed.
4598:
1.1.1.2 root 4599: @item @samp{G}, @samp{H}
1.1 root 4600: @samp{G} and @samp{H} may be defined in a machine-dependent fashion to
4601: permit immediate floating operands in particular ranges of values.
4602:
1.1.1.2 root 4603: @item @samp{s}
1.1 root 4604: An immediate integer operand whose value is not an explicit integer is
1.1.1.2 root 4605: allowed.
4606:
4607: This might appear strange; if an insn allows a constant operand with a
4608: value not known at compile time, it certainly must allow any known
1.1 root 4609: value. So why use @samp{s} instead of @samp{i}? Sometimes it allows
1.1.1.2 root 4610: better code to be generated.
4611:
4612: For example, on the 68000 in a fullword instruction it is possible to
4613: use an immediate operand; but if the immediate value is between -32
4614: and 31, better code results from loading the value into a register and
4615: using the register. This is because the load into the register can be
4616: done with a @samp{moveq} instruction. We arrange for this to happen
4617: by defining the letter @samp{K} to mean ``any integer outside the
4618: range -32 to 31'', and then specifying @samp{Ks} in the operand
1.1 root 4619: constraints.
4620:
1.1.1.2 root 4621: @item @samp{g}
1.1 root 4622: Any register, memory or immediate integer operand is allowed, except for
4623: registers that are not general registers.
4624:
1.1.1.2 root 4625: @item @samp{@var{n}} (a digit)
4626: An operand that matches operand number @var{n} is allowed.
1.1 root 4627: If a digit is used together with letters, the digit should come last.
4628:
1.1.1.2 root 4629: This is called a @dfn{matching constraint} and what it really means is
4630: that the assembler has only a single operand that fills two roles
4631: considered separate in the RTL insn. For example, an add insn has two
4632: input operands and one output operand in the RTL, but on most machines
4633: an add instruction really has only two operands, one of them an
4634: input-output operand.
4635:
4636: Matching constraints work only in circumstances like that add insn.
4637: More precisely, the matching constraint must appear in an input-only
4638: operand and the operand that it matches must be an output-only operand
4639: with a lower number.
4640:
4641: For operands to match in a particular case usually means that they
4642: are identical-looking RTL expressions. But in a few special cases
4643: specific kinds of dissimilarity are allowed. For example, @code{*x}
4644: as an input operand will match @code{*x++} as an output operand.
4645: For proper results in such cases, the output template should always
4646: use the output-operand's number when printing the operand.
4647:
4648: @item @samp{p}
1.1 root 4649: An operand that is a valid memory address is allowed. This is
4650: for ``load address'' and ``push address'' instructions.
4651:
4652: If @samp{p} is used in the constraint, the test-function in the
4653: @samp{match_operand} must be @code{address_operand}.
4654: @end table
4655:
4656: In order to have valid assembler code, each operand must satisfy
4657: its constraint. But a failure to do so does not prevent the pattern
4658: from applying to an insn. Instead, it directs the compiler to modify
1.1.1.2 root 4659: the code so that the constraint will be satisfied. Usually this is
1.1 root 4660: done by copying an operand into a register.
4661:
4662: Contrast, therefore, the two instruction patterns that follow:
4663:
4664: @example
4665: (define_insn ""
4666: [(set (match_operand:SI 0 "general_operand" "r")
4667: (plus:SI (match_dup 0)
4668: (match_operand:SI 1 "general_operand" "r")))]
4669: ""
4670: "@dots{}")
4671: @end example
4672:
4673: @noindent
4674: which has two operands, one of which must appear in two places, and
4675:
4676: @example
4677: (define_insn ""
4678: [(set (match_operand:SI 0 "general_operand" "r")
4679: (plus:SI (match_operand:SI 1 "general_operand" "0")
4680: (match_operand:SI 2 "general_operand" "r")))]
4681: ""
4682: "@dots{}")
4683: @end example
4684:
4685: @noindent
4686: which has three operands, two of which are required by a constraint to be
4687: identical. If we are considering an insn of the form
4688:
4689: @example
4690: (insn @var{n} @var{prev} @var{next}
4691: (set (reg:SI 3)
4692: (plus:SI (reg:SI 6) (reg:SI 109)))
4693: @dots{})
4694: @end example
4695:
4696: @noindent
4697: the first pattern would not apply at all, because this insn does not
4698: contain two identical subexpressions in the right place. The pattern would
4699: say, ``That does not look like an add instruction; try other patterns.''
4700: The second pattern would say, ``Yes, that's an add instruction, but there
4701: is something wrong with it.'' It would direct the reload pass of the
4702: compiler to generate additional insns to make the constraint true. The
4703: results might look like this:
4704:
4705: @example
4706: (insn @var{n2} @var{prev} @var{n}
4707: (set (reg:SI 3) (reg:SI 6))
4708: @dots{})
4709:
4710: (insn @var{n} @var{n2} @var{next}
4711: (set (reg:SI 3)
4712: (plus:SI (reg:SI 3) (reg:SI 109)))
4713: @dots{})
4714: @end example
4715:
4716: Because insns that don't fit the constraints are fixed up by loading
4717: operands into registers, every instruction pattern's constraints must
4718: permit the case where all the operands are in registers. It need not
4719: permit all classes of registers; the compiler knows how to copy registers
4720: into other registers of the proper class in order to make an instruction
4721: valid. But if no registers are permitted, the compiler will be stymied: it
4722: does not know how to save a register in memory in order to make an
4723: instruction valid. Instruction patterns that reject registers can be
4724: made valid by attaching a condition-expression that refuses to match
4725: an insn at all if the crucial operand is a register.
4726:
4727: @node Multi-Alternative, Class Preferences, Simple Constraints, Constraints
4728: @subsection Multiple Alternative Constraints
4729:
4730: Sometimes a single instruction has multiple alternative sets of possible
4731: operands. For example, on the 68000, a logical-or instruction can combine
4732: register or an immediate value into memory, or it can combine any kind of
4733: operand into a register; but it cannot combine one memory location into
4734: another.
4735:
4736: These constraints are represented as multiple alternatives. An alternative
4737: can be described by a series of letters for each operand. The overall
4738: constraint for an operand is made from the letters for this operand
4739: from the first alternative, a comma, the letters for this operand from
4740: the second alternative, a comma, and so on until the last alternative.
4741: Here is how it is done for fullword logical-or on the 68000:
4742:
4743: @example
4744: (define_insn "iorsi3"
4745: [(set (match_operand:SI 0 "general_operand" "=%m,d")
1.1.1.2 root 4746: (ior:SI (match_operand:SI 1 "general_operand" "0,0")
4747: (match_operand:SI 2 "general_operand" "dKs,dmKs")))]
1.1 root 4748: @dots{})
4749: @end example
4750:
4751: The first alternative has @samp{m} (memory) for operand 0, @samp{0} for
4752: operand 1 (meaning it must match operand 0), and @samp{dKs} for operand 2.
4753: The second alternative has @samp{d} (data register) for operand 0, @samp{0}
4754: for operand 1, and @samp{dmKs} for operand 2. The @samp{=} and @samp{%} in
4755: the constraint for operand 0 are not part of any alternative; their meaning
4756: is explained in the next section.
4757:
4758: If all the operands fit any one alternative, the instruction is valid.
4759: Otherwise, for each alternative, the compiler counts how many instructions
4760: must be added to copy the operands so that that alternative applies.
4761: The alternative requiring the least copying is chosen. If two alternatives
4762: need the same amount of copying, the one that comes first is chosen.
4763: These choices can be altered with the @samp{?} and @samp{!} characters:
4764:
4765: @table @samp
4766: @item ?
4767: Disparage slightly the alternative that the @samp{?} appears in,
4768: as a choice when no alternative applies exactly. The compiler regards
4769: this alternative as one unit more costly for each @samp{?} that appears
4770: in it.
4771:
4772: @item !
4773: Disparage severely the alternative that the @samp{!} appears in.
4774: When operands must be copied into registers, the compiler will
4775: never choose this alternative as the one to strive for.
4776: @end table
4777:
1.1.1.2 root 4778: When an insn pattern has multiple alternatives in its constraints,
4779: often the appearance of the assembler code determined mostly by which
4780: alternative was matched. When this is so, the C code for writing the
4781: assembler code can use the variable @code{which_alternative}, which is
4782: the ordinal number of the alternative that was actually satisfied
4783: (0 for the first, 1 for the second alternative, etc.). For example:
4784:
4785: @example
4786: (define_insn ""
4787: [(set (match_operand:SI 0 "general_operand" "r,m")
4788: (const_int 0))]
4789: ""
4790: "*
4791: return (which_alternative == 0
4792: ? \"clrreg %0\" : \"clrmem %0\");
4793: ")
4794: @end example
4795:
1.1 root 4796: @node Class Preferences, Modifiers, Multi-Alternative, Constraints
4797: @subsection Register Class Preferences
4798:
4799: The operand constraints have another function: they enable the compiler
4800: to decide which kind of hardware register a pseudo register is best
4801: allocated to. The compiler examines the constraints that apply to the
4802: insns that use the pseudo register, looking for the machine-dependent
4803: letters such as @samp{d} and @samp{a} that specify classes of registers.
4804: The pseudo register is put in whichever class gets the most ``votes''.
4805: The constraint letters @samp{g} and @samp{r} also vote: they vote in
4806: favor of a general register. The machine description says which registers
4807: are considered general.
4808:
4809: Of course, on some machines all registers are equivalent, and no register
4810: classes are defined. Then none of this complexity is relevant.
4811:
4812: @node Modifiers, No Constraints, Class Preferences, Constraints
4813: @subsection Constraint Modifier Characters
4814:
4815: @table @samp
4816: @item =
1.1.1.2 root 4817: Means that this operand is write-only for this instruction: the previous
4818: value is discarded and replaced by output data.
1.1 root 4819:
4820: @item +
4821: Means that this operand is both read and written by the instruction.
4822:
4823: When the compiler fixes up the operands to satisfy the constraints,
4824: it needs to know which operands are inputs to the instruction and
4825: which are outputs from it. @samp{=} identifies an output; @samp{+}
4826: identifies an operand that is both input and output; all other operands
4827: are assumed to be input only.
4828:
1.1.1.2 root 4829: @item &
4830: Means (in a particular alternative) that this operand is written
4831: before the instruction is finished using the input operands.
4832: Therefore, this operand may not lie in a register that is used as an
4833: input operand or as part of any memory address.
4834:
4835: @samp{&} applies only to the alternative in which it is written. In
4836: constraints with multiple alternatives, sometimes one alternative
4837: requires @samp{&} while others do not. See, for example, the
4838: @samp{movdf} insn of the 68000.
4839:
4840: @samp{&} does not obviate the need to write @samp{=}.
4841:
1.1 root 4842: @item %
1.1.1.2 root 4843: Declares the instruction to be commutative for this operand and the
4844: following operand. This means that the compiler may interchange the
4845: two operands if that is the cheapest way to make all operands fit the
4846: constraints. This is often used in patterns for addition instructions
4847: that really have only two operands: the result must go in one of the
4848: arguments. Here for example, is how the 68000 halfword-add
4849: instruction is defined:
4850:
4851: @example
4852: (define_insn "addhi3"
4853: [(set (match_operand:HI 0 "general_operand" "=m,r")
4854: (plus:HI (match_operand:HI 1 "general_operand" "%0,0")
4855: (match_operand:HI 2 "general_operand" "di,g")))]
4856: @dots{})
4857: @end example
4858:
4859: Note that in previous versions of GNU CC the @samp{%} constraint
4860: modifier always applied to operands 1 and 2 regardless of which
4861: operand it was written in. The usual custom was to write it in
4862: operand 0. Now it must be in operand 1 if the operands to be
4863: exchanged are 1 and 2.
1.1 root 4864:
4865: @item #
1.1.1.2 root 4866: Says that all following characters, up to the next comma, are to be
4867: ignored as a constraint. They are significant only for choosing
4868: register preferences.
1.1 root 4869:
4870: @item *
4871: Says that the following character should be ignored when choosing
1.1.1.2 root 4872: register preferences. @samp{*} has no effect on the meaning of the
4873: constraint as a constraint.
4874:
4875: Here is an example: the 68000 has an instruction to sign-extend a
4876: halfword in a data register, and can also sign-extend a value by
4877: copying it into an address register. While either kind of register is
4878: acceptable, the constraints on an address-register destination are
4879: less strict, so it is best if register allocation makes an address
4880: register its goal. Therefore, @samp{*} is used so that the @samp{d}
4881: constraint letter (for data register) is ignored when computing
4882: register preferences.
4883:
4884: @example
4885: (define_insn "extendhisi2"
4886: [(set (match_operand:SI 0 "general_operand" "=*d,a")
4887: (sign_extend:SI
4888: (match_operand:HI 1 "general_operand" "0,g")))]
4889: @dots{})
4890: @end example
1.1 root 4891: @end table
4892:
4893: @node No Constraints,, Modifiers, Constraints
4894: @subsection Not Using Constraints
4895:
4896: Some machines are so clean that operand constraints are not required. For
4897: example, on the Vax, an operand valid in one context is valid in any other
1.1.1.2 root 4898: context. On such a machine, every operand constraint would be @samp{g},
1.1 root 4899: excepting only operands of ``load address'' instructions which are
4900: written as if they referred to a memory location's contents but actual
1.1.1.2 root 4901: refer to its address. They would have constraint @samp{p}.
1.1 root 4902:
1.1.1.2 root 4903: For such machines, instead of writing @samp{g} and @samp{p} for all
1.1 root 4904: the constraints, you can choose to write a description with empty constraints.
4905: Then you write @samp{""} for the constraint in every @samp{match_operand}.
4906: Address operands are identified by writing an @samp{address} expression
4907: around the @samp{match_operand}, not by their constraints.
4908:
4909: When the machine description has just empty constraints, certain parts
4910: of compilation are skipped, making the compiler faster.
4911:
1.1.1.2 root 4912: @node Standard Names, Pattern Ordering, Constraints, Machine Desc
4913: @section Standard Names for Patterns Used in Generation
1.1 root 4914:
4915: Here is a table of the instruction names that are meaningful in the RTL
4916: generation pass of the compiler. Giving one of these names to an
4917: instruction pattern tells the RTL generation pass that it can use the
4918: pattern in to accomplish a certain task.
4919:
1.1.1.2 root 4920: @table @asis
4921: @item @samp{mov@var{m}}
1.1 root 4922: Here @var{m} is a two-letter machine mode name, in lower case. This
4923: instruction pattern moves data with that machine mode from operand 1 to
4924: operand 0. For example, @samp{movsi} moves full-word data.
4925:
4926: If operand 0 is a @samp{subreg} with mode @var{m} of a register whose
4927: natural mode is wider than @var{m}, the effect of this instruction is
4928: to store the specified value in the part of the register that corresponds
4929: to mode @var{m}. The effect on the rest of the register is undefined.
4930:
1.1.1.3 root 4931: This class of patterns is special in several ways. First of all, each
4932: of these names @emph{must} be defined, because there is no other way
4933: to copy a datum from one place to another.
4934:
4935: Second, these patterns are not used solely in the RTL generation pass.
4936: Even the reload pass can generate move insns to copy values from stack
4937: slots into temporary registers. When it does so, one of the operands
4938: is a hard register and the other is an operand that can have a reload.
4939:
4940: Therefore, when given such a pair of operands, the pattern must
4941: generate RTL which needs no temporary registers---no registers other
4942: than the operands. For example, if you support the pattern with a
4943: @code{define_expand}, then in such a case you mustn't call
4944: @code{force_reg} or any other such function which might generate new
4945: pseudo registers.
4946:
4947: This requirement exists even for subword modes on a RISC machine where
4948: fetching those modes from memory normally requires several insns and
4949: some temporary registers. Look in @file{spur.md} to see how the
4950: requirement is satisfied.
4951:
4952: The variety of operands that have reloads depends on the rest of the
4953: machine description, but typically on a RISC machine these can only be
4954: pseudo registers that did not get hard registers, while on other
4955: machines explicit memory references will get optional reloads.
4956:
1.1.1.7 ! root 4957: In addition, the constraints must allow any hard register to be moved
! 4958: to any other hard register (provided that @code{HARD_REGNO_MODE_OK}
! 4959: permits mode @var{m} in each of the registers).
! 4960:
1.1.1.2 root 4961: @item @samp{movstrict@var{m}}
1.1 root 4962: Like @samp{mov@var{m}} except that if operand 0 is a @samp{subreg}
4963: with mode @var{m} of a register whose natural mode is wider,
4964: the @samp{movstrict@var{m}} instruction is guaranteed not to alter
4965: any of the register except the part which belongs to mode @var{m}.
4966:
1.1.1.2 root 4967: @item @samp{add@var{m}3}
1.1 root 4968: Add operand 2 and operand 1, storing the result in operand 0. All operands
4969: must have mode @var{m}. This can be used even on two-address machines, by
4970: means of constraints requiring operands 1 and 0 to be the same location.
4971:
1.1.1.2 root 4972: @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 4973: Similar, for other arithmetic operations.
4974:
1.1.1.2 root 4975: @item @samp{andcb@var{m}3}
1.1 root 4976: Bitwise logical-and operand 1 with the complement of operand 2
4977: and store the result in operand 0.
4978:
1.1.1.2 root 4979: @item @samp{mulhisi3}
1.1 root 4980: Multiply operands 1 and 2, which have mode @code{HImode}, and store
4981: a @code{SImode} product in operand 0.
4982:
1.1.1.2 root 4983: @item @samp{mulqihi3}, @samp{mulsidi3}
1.1 root 4984: Similar widening-multiplication instructions of other widths.
4985:
1.1.1.2 root 4986: @item @samp{umulqihi3}, @samp{umulhisi3}, @samp{umulsidi3}
1.1 root 4987: Similar widening-multiplication instructions that do unsigned
4988: multiplication.
4989:
1.1.1.2 root 4990: @item @samp{divmod@var{m}4}
1.1 root 4991: Signed division that produces both a quotient and a remainder.
4992: Operand 1 is divided by operand 2 to produce a quotient stored
4993: in operand 0 and a remainder stored in operand 3.
4994:
1.1.1.2 root 4995: @item @samp{udivmod@var{m}4}
1.1 root 4996: Similar, but does unsigned division.
4997:
1.1.1.2 root 4998: @item @samp{divmod@var{m}@var{n}4}
1.1 root 4999: Like @samp{divmod@var{m}4} except that only the dividend has mode
5000: @var{m}; the divisor, quotient and remainder have mode @var{n}.
5001: For example, the Vax has a @samp{divmoddisi4} instruction
5002: (but it is omitted from the machine description, because it
5003: is so slow that it is faster to compute remainders by the
5004: circumlocution that the compiler will use if this instruction is
5005: not available).
5006:
1.1.1.2 root 5007: @item @samp{ashl@var{m}3}
1.1 root 5008: Arithmetic-shift operand 1 left by a number of bits specified by
5009: operand 2, and store the result in operand 0. Operand 2 has
5010: mode @code{SImode}, not mode @var{m}.
5011:
1.1.1.2 root 5012: @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 5013: Other shift and rotate instructions.
5014:
1.1.1.2 root 5015: Logical and arithmetic left shift are the same. Machines that do not
5016: allow negative shift counts often have only one instruction for
5017: shifting left. On such machines, you should define a pattern named
5018: @samp{ashl@var{m}3} and leave @samp{lshl@var{m}3} undefined.
5019:
5020: @item @samp{neg@var{m}2}
1.1 root 5021: Negate operand 1 and store the result in operand 0.
5022:
1.1.1.2 root 5023: @item @samp{abs@var{m}2}
1.1 root 5024: Store the absolute value of operand 1 into operand 0.
5025:
1.1.1.2 root 5026: @item @samp{sqrt@var{m}2}
1.1 root 5027: Store the square root of operand 1 into operand 0.
5028:
1.1.1.2 root 5029: @item @samp{ffs@var{m}2}
5030: Store into operand 0 one plus the index of the least significant 1-bit
5031: of operand 1. If operand 1 is zero, store zero. @var{m} is the mode
5032: of operand 0; operand 1's mode is specified by the instruction
5033: pattern, and the compiler will convert the operand to that mode before
5034: generating the instruction.
5035:
5036: @item @samp{one_cmpl@var{m}2}
1.1 root 5037: Store the bitwise-complement of operand 1 into operand 0.
5038:
1.1.1.2 root 5039: @item @samp{cmp@var{m}}
1.1 root 5040: Compare operand 0 and operand 1, and set the condition codes.
1.1.1.2 root 5041: The RTL pattern should look like this:
1.1 root 5042:
1.1.1.2 root 5043: @example
5044: (set (cc0) (minus (match_operand:@var{m} 0 @dots{})
5045: (match_operand:@var{m} 1 @dots{})))
5046: @end example
5047:
5048: Each such definition in the machine description, for integer mode
5049: @var{m}, must have a corresponding @samp{tst@var{m}} pattern, because
5050: optimization can simplify the compare into a test when operand 1 is
5051: zero.
5052:
5053: @item @samp{tst@var{m}}
1.1 root 5054: Compare operand 0 against zero, and set the condition codes.
1.1.1.2 root 5055: The RTL pattern should look like this:
1.1 root 5056:
1.1.1.2 root 5057: @example
5058: (set (cc0) (match_operand:@var{m} 0 @dots{}))
5059: @end example
5060:
5061: @item @samp{movstr@var{m}}
1.1 root 5062: Block move instruction. The addresses of the destination and source
5063: strings are the first two operands, and both are in mode @code{Pmode}.
5064: The number of bytes to move is the third operand, in mode @var{m}.
5065:
1.1.1.2 root 5066: @item @samp{cmpstr@var{m}}
1.1 root 5067: Block compare instruction, with operands like @samp{movstr@var{m}}
5068: except that the two memory blocks are compared byte by byte
5069: in lexicographic order. The effect of the instruction is to set
5070: the condition codes.
5071:
1.1.1.2 root 5072: @item @samp{float@var{m}@var{n}2}
1.1 root 5073: Convert operand 1 (valid for fixed point mode @var{m}) to floating
5074: point mode @var{n} and store in operand 0 (which has mode @var{n}).
5075:
1.1.1.2 root 5076: @item @samp{fix@var{m}@var{n}2}
5077: Convert operand 1 (valid for floating point mode @var{m}) to fixed
5078: point mode @var{n} as a signed number and store in operand 0 (which
5079: has mode @var{n}). This instruction's result is defined only when
5080: the value of operand 1 is an integer.
5081:
5082: @item @samp{fixuns@var{m}@var{n}2}
5083: Convert operand 1 (valid for floating point mode @var{m}) to fixed
5084: point mode @var{n} as an unsigned number and store in operand 0 (which
5085: has mode @var{n}). This instruction's result is defined only when the
5086: value of operand 1 is an integer.
5087:
5088: @item @samp{ftrunc@var{m}2}
5089: Convert operand 1 (valid for floating point mode @var{m}) to an
5090: integer value, still represented in floating point mode @var{m}, and
5091: store it in operand 0 (valid for floating point mode @var{m}).
5092:
5093: @item @samp{fix_trunc@var{m}@var{n}2}
5094: Like @samp{fix@var{m}@var{n}2} but works for any floating point value
5095: of mode @var{m} by converting the value to an integer.
5096:
5097: @item @samp{fixuns_trunc@var{m}@var{n}2}
5098: Like @samp{fixuns@var{m}@var{n}2} but works for any floating point
5099: value of mode @var{m} by converting the value to an integer.
5100:
5101: @item @samp{trunc@var{m}@var{n}}
1.1 root 5102: Truncate operand 1 (valid for mode @var{m}) to mode @var{n} and
5103: store in operand 0 (which has mode @var{n}). Both modes must be fixed
5104: point or both floating point.
5105:
1.1.1.2 root 5106: @item @samp{extend@var{m}@var{n}}
1.1 root 5107: Sign-extend operand 1 (valid for mode @var{m}) to mode @var{n} and
5108: store in operand 0 (which has mode @var{n}). Both modes must be fixed
5109: point or both floating point.
5110:
1.1.1.2 root 5111: @item @samp{zero_extend@var{m}@var{n}}
1.1 root 5112: Zero-extend operand 1 (valid for mode @var{m}) to mode @var{n} and
5113: store in operand 0 (which has mode @var{n}). Both modes must be fixed
5114: point.
5115:
1.1.1.2 root 5116: @item @samp{extv}
1.1 root 5117: Extract a bit-field from operand 1 (a register or memory operand),
5118: where operand 2 specifies the width in bits and operand 3 the starting
5119: bit, and store it in operand 0. Operand 0 must have @code{Simode}.
5120: Operand 1 may have mode @code{QImode} or @code{SImode}; often
5121: @code{SImode} is allowed only for registers. Operands 2 and 3 must be
5122: valid for @code{SImode}.
5123:
5124: The RTL generation pass generates this instruction only with constants
5125: for operands 2 and 3.
5126:
5127: The bit-field value is sign-extended to a full word integer
5128: before it is stored in operand 0.
5129:
1.1.1.2 root 5130: @item @samp{extzv}
1.1 root 5131: Like @samp{extv} except that the bit-field value is zero-extended.
5132:
1.1.1.2 root 5133: @item @samp{insv}
1.1 root 5134: Store operand 3 (which must be valid for @code{SImode}) into a
5135: bit-field in operand 0, where operand 1 specifies the width in bits
5136: and operand 2 the starting bit. Operand 0 may have mode @code{QImode}
5137: or @code{SImode}; often @code{SImode} is allowed only for registers.
5138: Operands 1 and 2 must be valid for @code{SImode}.
5139:
5140: The RTL generation pass generates this instruction only with constants
5141: for operands 1 and 2.
5142:
1.1.1.2 root 5143: @item @samp{s@var{cond}}
5144: Store zero or nonzero in the operand according to the condition codes.
5145: Value stored is nonzero iff the condition @var{cond} is true.
5146: @var{cond} is the name of a comparison operation expression code, such
1.1 root 5147: as @samp{eq}, @samp{lt} or @samp{leu}.
5148:
1.1.1.2 root 5149: You specify the mode that the operand must have when you write the
5150: @code{match_operand} expression. The compiler automatically sees
5151: which mode you have used and supplies an operand of that mode.
5152:
5153: The value stored for a true condition must have 1 as its low bit.
5154: Otherwise the instruction is not suitable and must be omitted from the
5155: machine description. You must tell the compiler exactly which value
5156: is stored by defining the macro @code{STORE_FLAG_VALUE}.
5157:
5158: @item @samp{b@var{cond}}
1.1 root 5159: Conditional branch instruction. Operand 0 is a @samp{label_ref}
5160: that refers to the label to jump to. Jump if the condition codes
5161: meet condition @var{cond}.
5162:
1.1.1.2 root 5163: @item @samp{call}
1.1.1.7 ! root 5164: Subroutine call instruction returning no value. Operand 0 is the
! 5165: function to call; operand 1 is the number of bytes of arguments pushed
! 5166: (in mode @code{SImode}, except it is normally a @samp{const_int});
! 5167: operand 2 is the number of registers used as operands.
! 5168:
! 5169: On most machines, operand 2 is not actually stored into the RTL pattern.
! 5170:
! 5171: Operand 0 should be a @samp{mem} RTX whose address is the address of
! 5172: the function.
! 5173:
! 5174: @item @samp{call_value}
! 5175: Subroutine call instruction returning a value. Operand 0 is the hard
! 5176: register in which the value is returned. There are three more
! 5177: operands, the same as the three operands of the @samp{call}
! 5178: instruction (but with numbers increased by one).
! 5179:
! 5180: Subroutines that return @code{BLKmode} objects use the @samp{call}
! 5181: insn.
1.1 root 5182:
1.1.1.2 root 5183: @item @samp{return}
1.1 root 5184: Subroutine return instruction. This instruction pattern name should be
5185: defined only if a single instruction can do all the work of returning
5186: from a function.
5187:
1.1.1.3 root 5188: @item @samp{casesi}
5189: Instruction to jump through a dispatch table, including bounds checking.
5190: This instruction takes five operands:
5191:
5192: @enumerate
5193: @item
5194: The index to dispatch on, which has mode @code{SImode}.
5195:
5196: @item
5197: The lower bound for indices in the table, an integer constant.
5198:
5199: @item
5200: The upper bound for indices in the table, an integer constant.
5201:
5202: @item
5203: A label to jump to if the index has a value outside the bounds.
5204: (If the machine-description macro @code{CASE_DROPS_THROUGH} is defined,
5205: then an out-of-bounds index drops through to the code following
5206: the jump table instead of jumping to this label. In that case,
5207: this label is not actually used by the @samp{casesi} instruction,
5208: but it is always provided as an operand.)
5209:
5210: @item
5211: A label that precedes the table itself.
5212: @end enumerate
5213:
5214: The table is a @samp{addr_vec} or @samp{addr_diff_vec} inside of a
5215: @samp{jump_insn}. The number of elements in the table is one plus the
5216: difference between the upper bound and the lower bound.
5217:
1.1.1.2 root 5218: @item @samp{tablejump}
1.1.1.3 root 5219: Instruction to jump to a variable address. This is a low-level
5220: capability which can be used to implement a dispatch table when there
5221: is no @samp{casesi} pattern.
5222:
5223: This pattern requires two operands: the address or offset, and a label
5224: which should immediately precede the jump table. If the macro
5225: @code{CASE_VECTOR_PC_RELATIVE} is defined then the first operand is an
5226: absolute address to jump to; otherwise, it is an offset which counts
5227: from the address of the table.
5228:
5229: The @samp{tablejump} insn is always the last insn before the jump
5230: table it uses. Its assembler code normally has no need to use the
5231: second operand, but you should incorporate it in the RTL pattern so
5232: that the jump optimizer will not delete the table as unreachable code.
1.1 root 5233: @end table
5234:
1.1.1.2 root 5235: @node Pattern Ordering, Dependent Patterns, Standard Names, Machine Desc
5236: @section When the Order of Patterns Matters
5237:
5238: Sometimes an insn can match more than one instruction pattern. Then the
5239: pattern that appears first in the machine description is the one used.
5240: Therefore, more specific patterns (patterns that will match fewer things)
5241: and faster instructions (those that will produce better code when they
5242: do match) should usually go first in the description.
5243:
5244: In some cases the effect of ordering the patterns can be used to hide
5245: a pattern when it is not valid. For example, the 68000 has an
5246: instruction for converting a fullword to floating point and another
5247: for converting a byte to floating point. An instruction converting
5248: an integer to floating point could match either one. We put the
5249: pattern to convert the fullword first to make sure that one will
5250: be used rather than the other. (Otherwise a large integer might
5251: be generated as a single-byte immediate quantity, which would not work.)
5252: Instead of using this pattern ordering it would be possible to make the
5253: pattern for convert-a-byte smart enough to deal properly with any
5254: constant value.
5255:
5256: @node Dependent Patterns, Jump Patterns, Pattern Ordering, Machine Desc
5257: @section Interdependence of Patterns
1.1 root 5258:
5259: Every machine description must have a named pattern for each of the
5260: conditional branch names @samp{b@var{cond}}. The recognition template
5261: must always have the form
5262:
5263: @example
5264: (set (pc)
5265: (if_then_else (@var{cond} (cc0) (const_int 0))
5266: (label_ref (match_operand 0 "" ""))
5267: (pc)))
5268: @end example
5269:
5270: @noindent
5271: In addition, every machine description must have an anonymous pattern
5272: for each of the possible reverse-conditional branches. These patterns
5273: look like
5274:
5275: @example
5276: (set (pc)
5277: (if_then_else (@var{cond} (cc0) (const_int 0))
5278: (pc)
5279: (label_ref (match_operand 0 "" ""))))
5280: @end example
5281:
5282: @noindent
5283: They are necessary because jump optimization can turn direct-conditional
5284: branches into reverse-conditional branches.
5285:
5286: The compiler does more with RTL than just create it from patterns
5287: and recognize the patterns: it can perform arithmetic expression codes
5288: when constant values for their operands can be determined. As a result,
5289: sometimes having one pattern can require other patterns. For example, the
5290: Vax has no `and' instruction, but it has `and not' instructions. Here
5291: is the definition of one of them:
5292:
5293: @example
5294: (define_insn "andcbsi2"
5295: [(set (match_operand:SI 0 "general_operand" "")
5296: (and:SI (match_dup 0)
5297: (not:SI (match_operand:SI
5298: 1 "general_operand" ""))))]
5299: ""
5300: "bicl2 %1,%0")
5301: @end example
5302:
5303: @noindent
5304: If operand 1 is an explicit integer constant, an instruction constructed
1.1.1.2 root 5305: using that pattern can be simplified into an `and' like this:
1.1 root 5306:
5307: @example
5308: (set (reg:SI 41)
5309: (and:SI (reg:SI 41)
5310: (const_int 0xffff7fff)))
5311: @end example
5312:
5313: @noindent
5314: (where the integer constant is the one's complement of what
5315: appeared in the original instruction).
5316:
5317: To avoid a fatal error, the compiler must have a pattern that recognizes
5318: such an instruction. Here is what is used:
5319:
5320: @example
5321: (define_insn ""
5322: [(set (match_operand:SI 0 "general_operand" "")
5323: (and:SI (match_dup 0)
5324: (match_operand:SI 1 "general_operand" "")))]
5325: "GET_CODE (operands[1]) == CONST_INT"
5326: "*
1.1.1.2 root 5327: @{ operands[1]
1.1 root 5328: = gen_rtx (CONST_INT, VOIDmode, ~INTVAL (operands[1]));
5329: return \"bicl2 %1,%0\";
1.1.1.2 root 5330: @}")
1.1 root 5331: @end example
5332:
5333: @noindent
5334: Whereas a pattern to match a general `and' instruction is impossible to
5335: support on the Vax, this pattern is possible because it matches only a
5336: constant second argument: a special case that can be output as an `and not'
5337: instruction.
5338:
1.1.1.2 root 5339: A ``compare'' instruction whose RTL looks like this:
5340:
5341: @example
5342: (set (cc0) (minus @var{operand} (const_int 0)))
5343: @end example
5344:
5345: @noindent
5346: may be simplified by optimization into a ``test'' like this:
5347:
5348: @example
5349: (set (cc0) @var{operand})
5350: @end example
5351:
5352: @noindent
5353: So in the machine description, each ``compare'' pattern for an integer
5354: mode must have a corresponding ``test'' pattern that will match the
5355: result of such simplification.
5356:
5357: In some cases machines support instructions identical except for the
5358: machine mode of one or more operands. For example, there may be
5359: ``sign-extend halfword'' and ``sign-extend byte'' instructions whose
5360: patterns are
5361:
5362: @example
5363: (set (match_operand:SI 0 @dots{})
5364: (extend:SI (match_operand:HI 1 @dots{})))
5365:
5366: (set (match_operand:SI 0 @dots{})
5367: (extend:SI (match_operand:QI 1 @dots{})))
5368: @end example
5369:
5370: @noindent
5371: Constant integers do not specify a machine mode, so an instruction to
5372: extend a constant value could match either pattern. The pattern it
5373: actually will match is the one that appears first in the file. For correct
5374: results, this must be the one for the widest possible mode (@code{HImode},
5375: here). If the pattern matches the @code{QImode} instruction, the results
5376: will be incorrect if the constant value does not actually fit that mode.
5377:
5378: Such instructions to extend constants are rarely generated because they are
5379: optimized away, but they do occasionally happen in nonoptimized
5380: compilations.
5381:
5382: @node Jump Patterns, Peephole Definitions, Dependent Patterns, Machine Desc
5383: @section Defining Jump Instruction Patterns
5384:
5385: GNU CC assumes that the machine has a condition code. A comparison insn
5386: sets the condition code, recording the results of both signed and unsigned
5387: comparison of the given operands. A separate branch insn tests the
5388: condition code and branches or not according its value. The branch insns
5389: come in distinct signed and unsigned flavors. Many common machines, such
5390: as the Vax, the 68000 and the 32000, work this way.
5391:
5392: Some machines have distinct signed and unsigned compare instructions, and
5393: only one set of conditional branch instructions. The easiest way to handle
5394: these machines is to treat them just like the others until the final stage
5395: where assembly code is written. At this time, when outputting code for the
5396: compare instruction, peek ahead at the following branch using
5397: @code{NEXT_INSN (insn)}. (The variable @code{insn} refers to the insn
5398: being output, in the output-writing code in an instruction pattern.) If
5399: the RTL says that is an unsigned branch, output an unsigned compare;
5400: otherwise output a signed compare. When the branch itself is output, you
5401: can treat signed and unsigned branches identically.
5402:
5403: The reason you can do this is that GNU CC always generates a pair of
5404: consecutive RTL insns, one to set the condition code and one to test it,
5405: and keeps the pair inviolate until the end.
5406:
5407: To go with this technique, you must define the machine-description macro
5408: @code{NOTICE_UPDATE_CC} to do @code{CC_STATUS_INIT}; in other words, no
5409: compare instruction is superfluous.
5410:
5411: Some machines have compare-and-branch instructions and no condition code.
5412: A similar technique works for them. When it is time to ``output'' a
5413: compare instruction, record its operands in two static variables. When
5414: outputting the branch-on-condition-code instruction that follows, actually
5415: output a compare-and-branch instruction that uses the remembered operands.
5416:
5417: It also works to define patterns for compare-and-branch instructions.
5418: In optimizing compilation, the pair of compare and branch instructions
5419: will be combined accoprding to these patterns. But this does not happen
5420: if optimization is not requested. So you must use one of the solutions
5421: above in addition to any special patterns you define.
5422:
5423: @node Peephole Definitions, Expander Definitions, Jump Patterns, Machine Desc
5424: @section Defining Machine-Specific Peephole Optimizers
5425:
5426: In addition to instruction patterns the @file{md} file may contain
5427: definitions of machine-specific peephole optimizations.
5428:
5429: The combiner does not notice certain peephole optimizations when the data
5430: flow in the program does not suggest that it should try them. For example,
5431: sometimes two consecutive insns related in purpose can be combined even
5432: though the second one does not appear to use a register computed in the
5433: first one. A machine-specific peephole optimizer can detect such
5434: opportunities.
5435:
5436: A definition looks like this:
5437:
5438: @example
5439: (define_peephole
5440: [@var{insn-pattern-1}
5441: @var{insn-pattern-2}
5442: @dots{}]
5443: "@var{condition}"
5444: "@var{template}")
5445: @end example
5446:
5447: In this skeleton, @var{insn-pattern-1} and so on are patterns to match
5448: consecutive instructions. The optimization applies to a sequence of
5449: instructions when @var{insn-pattern-1} matches the first one,
5450: @var{insn-pattern-2} matches the next, and so on.@refill
5451:
5452: @var{insn-pattern-1} and so on look @emph{almost} like the second operand
5453: of @code{define_insn}. There is one important difference: this pattern is
5454: an RTX, not a vector. If the @code{define_insn} pattern would be a vector
5455: of one element, the @var{insn-pattern} should be just that element, no
5456: vector. If the @code{define_insn} pattern would have multiple elements
5457: then the @var{insn-pattern} must place the vector inside an explicit
5458: @code{parallel} RTX.@refill
5459:
5460: The operands of the instructions are matched with @code{match_operands} and
5461: @code{match_dup}, as usual). What is not usual is that the operand numbers
5462: apply to all the instruction patterns in the definition. So, you can check
5463: for identical operands in two instructions by using @code{match_operand}
5464: in one instruction and @code{match_dup} in the other.
5465:
5466: The operand constraints used in @code{match_operand} patterns do not have
5467: any direct effect on the applicability of the optimization, but they will
5468: be validated afterward, so write constraints that are sure to fit whenever
5469: the optimization is applied. It is safe to use @code{"g"} for each
5470: operand.
5471:
5472: Once a sequence of instructions matches the patterns, the @var{condition}
5473: is checked. This is a C expression which makes the final decision whether
5474: to perform the optimization (do so if the expression is nonzero). If
5475: @var{condition} is omitted (in other words, the string is empty) then the
5476: optimization is applied to every sequence of instructions that matches the
5477: patterns.
5478:
5479: The defined peephole optimizations are applied after register allocation is
5480: complete. Therefore, the optimizer can check which operands have ended up
5481: in which kinds of registers, just by looking at the operands.
5482:
5483: The way to refer to the operands in @var{condition} is to write
5484: @code{operands[@var{i}]} for operand number @var{i} (as matched by
5485: @code{(match_operand @var{i} @dots{})}). Use the variable @code{insn} to
5486: refer to the last of the insns being matched; use @code{PREV_INSN} to find
5487: the preceding insns (but be careful to skip over any @samp{note} insns that
5488: intervene).@refill
5489:
5490: When optimizing computations with intermediate results, you can use
5491: @var{condition} to match only when the intermediate results are not used
5492: elsewhere. Use the C expression @code{dead_or_set_p (@var{insn},
5493: @var{op})}, where @var{insn} is the insn in which you expect the value to
5494: be used for the last time (from the value of @code{insn}, together with use
5495: of @code{PREV_INSN}), and @var{op} is the intermediate value (from
5496: @code{operands[@var{i}]}).@refill
5497:
5498: Applying the optimization means replacing the sequence of instructions with
5499: one new instruction. The @var{template} controls ultimate output of
5500: assembler code for this combined instruction. It works exactly like the
5501: template of a @code{define_insn}. Operand numbers in this template are the
5502: same ones used in matching the original sequence of instructions.
5503:
5504: The result of a defined peephole optimizer does not need to match any of
5505: the instruction patterns, and it does not have an opportunity to match
5506: them. The peephole optimizer definition itself serves as the instruction
5507: pattern to control how the instruction is output.
5508:
5509: Defined peephole optimizers are run in the last jump optimization pass, so
5510: the instructions they produce are never combined or rearranged
5511: automatically in any way.
5512:
5513: Here is an example, taken from the 68000 machine description:
5514:
5515: @example
5516: (define_peephole
5517: [(set (reg:SI 15) (plus:SI (reg:SI 15) (const_int 4)))
5518: (set (match_operand:DF 0 "register_operand" "f")
5519: (match_operand:DF 1 "register_operand" "ad"))]
5520: "FP_REG_P (operands[0]) && ! FP_REG_P (operands[1])"
5521: "*
5522: @{
5523: rtx xoperands[2];
5524: xoperands[1] = gen_rtx (REG, SImode, REGNO (operands[1]) + 1);
5525: #ifdef MOTOROLA
5526: output_asm_insn (\"move.l %1,(sp)\", xoperands);
5527: output_asm_insn (\"move.l %1,-(sp)\", operands);
5528: return \"fmove.d (sp)+,%0\";
5529: #else
5530: output_asm_insn (\"movel %1,sp@@\", xoperands);
5531: output_asm_insn (\"movel %1,sp@@-\", operands);
5532: return \"fmoved sp@@+,%0\";
5533: #endif
5534: @}
5535: ")
5536: @end example
5537:
5538: The effect of this optimization is to change
5539:
5540: @example
5541: jbsr _foobar
5542: addql #4,sp
5543: movel d1,sp@@-
5544: movel d0,sp@@-
5545: fmoved sp@@+,fp0
5546: @end example
5547:
5548: @noindent
5549: into
5550:
5551: @example
5552: jbsr _foobar
5553: movel d1,sp@@
5554: movel d0,sp@@-
5555: fmoved sp@@+,fp0
5556: @end example
5557:
5558: @node Expander Definitions,, Peephole Definitions, Machine Desc
5559: @section Defining RTL Sequences for Code Generation
5560:
5561: On some target machines, some standard pattern names for RTL generation
5562: cannot be handled with single insn, but a sequence of RTL insns can
5563: represent them. For these target machines, you can write a
5564: @samp{define_expand} to specify how to generate the sequence of RTL.
5565:
5566: A @samp{define_expand} is an RTL expression that looks almost like a
5567: @samp{define_insn}; but, unlike the latter, a @samp{define_expand} is used
5568: only for RTL generation and it can produce more than one RTL insn.
5569:
5570: A @samp{define_expand} RTX has four operands:
5571:
5572: @itemize @bullet
5573: @item
5574: The name. Each @samp{define_expand} must have a name, since the only
5575: use for it is to refer to it by name.
5576:
5577: @item
5578: The RTL template. This is just like the RTL template for a
5579: @samp{define_peephole} in that it is a vector of RTL expressions
5580: each being one insn.
5581:
5582: @item
5583: The condition, a string containing a C expression. This expression is
5584: used to express how the availability of this pattern depends on
5585: subclasses of target machine, selected by command-line options when
5586: GNU CC is run. This is just like the condition of a
5587: @samp{define_insn} that has a standard name.
5588:
5589: @item
5590: The preparation statements, a string containing zero or more C
5591: statements which are to be executed before RTL code is generated from
5592: the RTL template.
5593:
5594: Usually these statements prepare temporary registers for use as
5595: internal operands in the RTL template, but they can also generate RTL
5596: insns directly by calling routines such as @samp{emit_insn}, etc.
5597: Any such insns precede the ones that come from the RTL template.
5598: @end itemize
5599:
5600: The RTL template, in addition to controlling generation of RTL insns,
5601: also describes the operands that need to be specified when this pattern
5602: is used. In particular, it gives a predicate for each operand.
5603:
5604: A true operand, which need to be specified in order to generate RTL from
5605: the pattern, should be described with a @samp{match_operand} in its first
5606: occurrence in the RTL template. This enters information on the operand's
5607: predicate into the tables that record such things. GNU CC uses the
5608: information to preload the operand into a register if that is required for
5609: valid RTL code. If the operand is referred to more than once, subsequent
5610: references should use @samp{match_dup}.
5611:
5612: The RTL template may also refer to internal ``operands'' which are
5613: temporary registers or labels used only within the sequence made by the
5614: @samp{define_expand}. Internal operands are substituted into the RTL
5615: template with @samp{match_dup}, never with @samp{match_operand}. The
5616: values of the internal operands are not passed in as arguments by the
5617: compiler when it requests use of this pattern. Instead, they are computed
5618: within the pattern, in the preparation statements. These statements
5619: compute the values and store them into the appropriate elements of
5620: @code{operands} so that @samp{match_dup} can find them.
5621:
5622: There are two special macros defined for use in the preparation statements:
5623: @code{DONE} and @code{FAIL}. Use them with a following semicolon,
5624: as a statement.
5625:
5626: @table @code
5627: @item DONE
5628: Use the @code{DONE} macro to end RTL generation for the pattern. The
5629: only RTL insns resulting from the pattern on this occasion will be
5630: those already emitted by explicit calls to @code{emit_insn} within the
5631: preparation statements; the RTL template will not be generated.
5632:
5633: @item FAIL
5634: Make the pattern fail on this occasion. When a pattern fails, it means
5635: that the pattern was not truly available. The calling routines in the
5636: compiler will try other strategies for code generation using other patterns.
5637:
5638: Failure is currently supported only for binary operations (addition,
5639: multiplication, shifting, etc.).
5640:
5641: Do not emit any insns explicitly with @code{emit_insn} before failing.
5642: @end table
5643:
5644: Here is an example, the definition of left-shift for the SPUR chip:
5645:
5646: @example
5647: (define_expand "ashlsi3"
5648: [(set (match_operand:SI 0 "register_operand" "")
5649: (ashift:SI
5650: (match_operand:SI 1 "register_operand" "")
5651: (match_operand:SI 2 "nonmemory_operand" "")))]
5652: ""
5653: "
5654: @{
5655: if (GET_CODE (operands[2]) != CONST_INT
5656: || (unsigned) INTVAL (operands[2]) > 3)
5657: FAIL;
5658: @}")
5659: @end example
5660:
5661: @noindent
5662: This example uses @samp{define_expand} so that it can generate an RTL insn
5663: for shifting when the shift-count is in the supported range of 0 to 3 but
5664: fail in other cases where machine insns aren't available. When it fails,
5665: the compiler tries another strategy using different patterns (such as, a
5666: library call).
5667:
5668: If the compiler were able to handle nontrivial condition-strings in
5669: patterns with names, then there would be possible to use a
5670: @samp{define_insn} in that case. Here is another case (zero-extension on
5671: the 68000) which makes more use of the power of @samp{define_expand}:
5672:
5673: @example
5674: (define_expand "zero_extendhisi2"
5675: [(set (match_operand:SI 0 "general_operand" "")
5676: (const_int 0))
5677: (set (strict_low_part
5678: (subreg:HI
5679: (match_operand:SI 0 "general_operand" "")
5680: 0))
5681: (match_operand:HI 1 "general_operand" ""))]
5682: ""
5683: "operands[1] = make_safe_from (operands[1], operands[0]);")
5684: @end example
5685:
5686: @noindent
5687: Here two RTL insns are generated, one to clear the entire output operand
5688: and the other to copy the input operand into its low half. This sequence
5689: is incorrect if the input operand refers to [the old value of] the output
5690: operand, so the preparation statement makes sure this isn't so. The
5691: function @code{make_safe_from} copies the @code{operands[1]} into a
5692: temporary register if it refers to @code{operands[0]}. It does this
5693: by emitting another RTL insn.
5694:
5695: Finally, a third example shows the use of an internal operand.
5696: Zero-extension on the SPUR chip is done by @samp{and}-ing the result
5697: against a halfword mask. But this mask cannot be represented by a
5698: @samp{const_int} because the constant value is too large to be legitimate
5699: on this machine. So it must be copied into a register with
5700: @code{force_reg} and then the register used in the @samp{and}.
5701:
5702: @example
5703: (define_expand "zero_extendhisi2"
5704: [(set (match_operand:SI 0 "register_operand" "")
5705: (and:SI (subreg:SI
5706: (match_operand:HI 1 "register_operand" "")
5707: 0)
5708: (match_dup 2)))]
5709: ""
5710: "operands[2]
5711: = force_reg (SImode, gen_rtx (CONST_INT,
5712: VOIDmode, 65535)); ")
5713: @end example
5714:
5715: @node Machine Macros, Config, Machine Desc, Top
1.1 root 5716: @chapter Machine Description Macros
5717:
5718: The other half of the machine description is a C header file conventionally
5719: given the name @file{tm-@var{machine}.h}. The file @file{tm.h} should be a
5720: link to it. The header file @file{config.h} includes @file{tm.h} and most
5721: compiler source files include @file{config.h}.
5722:
5723: @menu
1.1.1.2 root 5724: * Run-time Target:: Defining -m options like -m68000 and -m68020.
1.1 root 5725: * Storage Layout:: Defining sizes and alignments of data types.
5726: * Registers:: Naming and describing the hardware registers.
5727: * Register Classes:: Defining the classes of hardware registers.
5728: * Stack Layout:: Defining which way the stack grows and by how much.
1.1.1.2 root 5729: * Library Names:: Specifying names of subroutines to call automatically.
1.1 root 5730: * Addressing Modes:: Defining addressing modes valid for memory operands.
5731: * Condition Code:: Defining how insns update the condition code.
5732: * Assembler Format:: Defining how to write insns and pseudo-ops to output.
5733: * Misc:: Everything else.
5734: @end menu
5735:
5736: @node Run-time Target, Storage Layout, Machine Macros, Machine Macros
5737: @section Run-time Target Specification
5738:
5739: @table @code
5740: @item CPP_PREDEFINES
1.1.1.7 ! root 5741: Define this to be a string constant containing @samp{-D} options to
! 5742: define the predefined macros that identify this machine and system.
! 5743: These macros will be predefined unless the @samp{-ansi} option is
! 5744: specified.
1.1 root 5745:
5746: For example, on the Sun, one can use the value
5747:
5748: @example
1.1.1.2 root 5749: "-Dmc68000 -Dsun -Dunix"
1.1 root 5750: @end example
5751:
1.1.1.7 ! root 5752: @item CPP_SPEC
! 5753: A C string constant that tells the GNU CC driver program options to
! 5754: pass to CPP. It can also specify how to translate options you
! 5755: give to GNU CC into options for GNU CC to pass to the CPP.
! 5756:
! 5757: Do not define this macro if it does not need to do anything.
! 5758:
! 5759: @item CC1_SPEC
! 5760: A C string constant that tells the GNU CC driver program options to
! 5761: pass to CC1. It can also specify how to translate options you
! 5762: give to GNU CC into options for GNU CC to pass to the CC1.
! 5763:
! 5764: Do not define this macro if it does not need to do anything.
! 5765:
1.1 root 5766: @item extern int target_flags;
5767: This declaration should be present.
5768:
5769: @item TARGET_@dots{}
5770: This series of macros is to allow compiler command arguments to
5771: enable or disable the use of optional features of the target machine.
5772: For example, one machine description serves both the 68000 and
5773: the 68020; a command argument tells the compiler whether it should
5774: use 68020-only instructions or not. This command argument works
5775: by means of a macro @code{TARGET_68020} that tests a bit in
5776: @code{target_flags}.
5777:
5778: Define a macro @code{TARGET_@var{featurename}} for each such option.
5779: Its definition should test a bit in @code{target_flags}; for example:
5780:
5781: @example
5782: #define TARGET_68020 (target_flags & 1)
5783: @end example
5784:
5785: One place where these macros are used is in the condition-expressions
5786: of instruction patterns. Note how @code{TARGET_68020} appears
1.1.1.2 root 5787: frequently in the 68000 machine description file, @file{m68k.md}.
1.1 root 5788: Another place they are used is in the definitions of the other
5789: macros in the @file{tm-@var{machine}.h} file.
5790:
5791: @item TARGET_SWITCHES
1.1.1.2 root 5792: This macro defines names of command options to set and clear
1.1 root 5793: bits in @code{target_flags}. Its definition is an initializer
1.1.1.2 root 5794: with a subgrouping for each command option.
1.1 root 5795:
1.1.1.2 root 5796: Each subgrouping contains a string constant, that defines the option
1.1 root 5797: name, and a number, which contains the bits to set in
5798: @code{target_flags}. A negative number says to clear bits instead;
1.1.1.2 root 5799: the negative of the number is which bits to clear. The actual option
1.1 root 5800: name is made by appending @samp{-m} to the specified name.
5801:
5802: One of the subgroupings should have a null string. The number in
5803: this grouping is the default value for @code{target_flags}. Any
1.1.1.2 root 5804: target options act starting with that value.
1.1 root 5805:
5806: Here is an example which defines @samp{-m68000} and @samp{-m68020}
5807: with opposite meanings, and picks the latter as the default:
5808:
5809: @example
5810: #define TARGET_SWITCHES \
5811: @{ @{ "68020", 1@}, \
5812: @{ "68000", -1@}, \
5813: @{ "", 1@}@}
5814: @end example
5815:
1.1.1.7 ! root 5816: @item OVERRIDE_OPTIONS
! 5817: Sometimes certain combinations of command options do not make sense on
! 5818: a particular target machine. You can define a macro
! 5819: @code{OVERRIDE_OPTIONS} to take account of this. This macro, if
! 5820: defined, is executed once just after all the command options have been
! 5821: parsed.
! 5822: @end table
1.1.1.2 root 5823:
1.1 root 5824: @node Storage Layout, Registers, Run-time Target, Machine Macros
5825: @section Storage Layout
5826:
1.1.1.2 root 5827: Note that the definitions of the macros in this table which are sizes or
5828: alignments measured in bits do not need to be constant. They can be C
5829: expressions that refer to static variables, such as the @code{target_flags}.
5830: @xref{Run-time Target}.
5831:
1.1 root 5832: @table @code
5833: @item BITS_BIG_ENDIAN
5834: Define this macro if the most significant bit in a byte has the lowest
5835: number. This means that bit-field instructions count from the most
5836: significant bit. If the machine has no bit-field instructions, this
5837: macro is irrelevant.
5838:
5839: @item BYTES_BIG_ENDIAN
5840: Define this macro if the most significant byte in a word has the
5841: lowest number.
5842:
5843: @item WORDS_BIG_ENDIAN
1.1.1.2 root 5844: Define this macro if, in a multiword object, the most significant
1.1 root 5845: word has the lowest number.
5846:
5847: @item BITS_PER_UNIT
5848: Number of bits in an addressable storage unit (byte); normally 8.
5849:
5850: @item BITS_PER_WORD
5851: Number of bits in a word; normally 32.
5852:
5853: @item UNITS_PER_WORD
5854: Number of storage units in a word; normally 4.
5855:
5856: @item POINTER_SIZE
5857: Width of a pointer, in bits.
5858:
1.1.1.7 ! root 5859: @item POINTER_BOUNDARY
! 5860: Alignment required for pointers stored in memory, in bits.
! 5861:
1.1 root 5862: @item PARM_BOUNDARY
1.1.1.2 root 5863: Alignment required for function parameters on the stack, in bits.
5864:
5865: @item STACK_BOUNDARY
5866: Define this macro if you wish to preserve a certain alignment for
5867: the stack pointer at all times. The definition is a C expression
5868: for the desired alignment (measured in bits).
1.1 root 5869:
5870: @item FUNCTION_BOUNDARY
5871: Alignment required for a function entry point, in bits.
5872:
5873: @item BIGGEST_ALIGNMENT
1.1.1.2 root 5874: Biggest alignment that any data type can require on this machine, in bits.
5875:
1.1.1.7 ! root 5876: @item EMPTY_FIELD_BOUNDARY
1.1.1.2 root 5877: Alignment in bits to be given to a structure bit field that follows an
5878: empty field such as @code{int : 0;}.
5879:
5880: @item STRUCTURE_SIZE_BOUNDARY
5881: Number of bits which any structure or union's size must be a multiple of.
5882: Each structure or union's size is rounded up to a multiple of this.
5883:
5884: If you do not define this macro, the default is the same as
5885: @code{BITS_PER_UNIT}.
1.1 root 5886:
5887: @item STRICT_ALIGNMENT
5888: Define this if instructions will fail to work if given data not
5889: on the nominal alignment. If instructions will merely go slower
5890: in that case, do not define this macro.
1.1.1.4 root 5891:
1.1.1.6 root 5892: @item PCC_BITFIELD_TYPE_MATTERS
5893: Define this if you wish to imitate a certain bizarre behavior pattern
5894: of some instances of PCC: a bit field whose declared type is
5895: @code{int} has the same effect on the size and alignment of a
5896: structure as an actual @code{int} would have.
5897:
5898: Just what effect that is in GNU CC depends on other parameters, but on
5899: most machines it would force the structure's alignment and size to a
5900: multiple of 32 or @code{BIGGEST_ALIGNMENT} bits.
5901:
1.1.1.4 root 5902: @item CHECK_FLOAT_VALUE (@var{mode}, @var{value})
5903: A C statement to validate the value @var{value} (or type
5904: @code{double}) for mode @var{mode}. This means that you check whether
5905: @var{value} fits within the possible range of values for mode
5906: @var{mode} on this target machine. The mode @var{mode} is always
5907: @code{SFmode} or @code{DFmode}.
5908:
5909: If @var{value} is not valid, you should call @code{error} to print an
5910: error message and then assign some valid value to @var{value}.
5911: Allowing an invalid value to go through the compiler can produce
5912: incorrect assembler code which may even cause Unix assemblers to
5913: crash.
5914:
5915: This macro need not be defined if there is no work for it to do.
1.1 root 5916: @end table
5917:
5918: @node Registers, Register Classes, Storage Layout, Machine Macros
5919: @section Register Usage
5920:
5921: @table @code
5922: @item FIRST_PSEUDO_REGISTER
5923: Number of hardware registers known to the compiler. They receive
5924: numbers 0 through @code{FIRST_PSEUDO_REGISTER-1}; thus, the first
1.1.1.2 root 5925: pseudo register's number really is assigned the number
1.1 root 5926: @code{FIRST_PSEUDO_REGISTER}.
5927:
5928: @item FIXED_REGISTERS
5929: An initializer that says which registers are used for fixed purposes
5930: all throughout the compiled code and are therefore not available for
1.1.1.2 root 5931: general allocation. These would include the stack pointer, the frame
1.1.1.7 ! root 5932: pointer (except on machines where that can be used as a general
! 5933: register when no frame pointer is needed), the program counter on
! 5934: machines where that is considered one of the addressable registers,
! 5935: and any other numbered register with a standard use.
1.1 root 5936:
5937: This information is expressed as a sequence of numbers, separated by
5938: commas and surrounded by braces. The @var{n}th number is 1 if
1.1.1.2 root 5939: register @var{n} is fixed, 0 otherwise.
5940:
5941: The table initialized from this macro, and the table initialized by
5942: the following one, may be overridden at run time either automatically,
5943: by the actions of the macro @code{CONDITIONAL_REGISTER_USAGE}, or by
5944: the user with the command options @samp{-ffixed-@var{reg}},
5945: @samp{-fcall-used-@var{reg}} and @samp{-fcall-saved-@var{reg}}.
1.1 root 5946:
5947: @item CALL_USED_REGISTERS
5948: Like @code{FIXED_REGISTERS} but has 1 for each register that is
5949: clobbered (in general) by function calls as well as for fixed
5950: registers. This macro therefore identifies the registers that are not
5951: available for general allocation of values that must live across
5952: function calls.
5953:
1.1.1.2 root 5954: If a register has 0 in @code{CALL_USED_REGISTERS}, the compiler
1.1 root 5955: automatically saves it on function entry and restores it on function
5956: exit, if the register is used within the function.
5957:
1.1.1.2 root 5958: @item CONDITIONAL_REGISTER_USAGE
5959: Zero or more C statements that may conditionally modify two variables
5960: @code{fixed_regs} and @code{call_used_regs} (both of type @code{char
5961: []}) after they have been initialized from the two preceding macros.
5962:
5963: This is necessary in case the fixed or call-clobbered registers depend
5964: on target flags.
5965:
5966: You need not define this macro if it has no work to do.
5967:
1.1.1.7 ! root 5968: If the usage of an entire class of registers depends on the target
! 5969: flags, you may indicate this to gcc by using this macro to modify
! 5970: @code{fixed_regs} and @code{call_used_regs} to 1 for each of the
! 5971: registers in the classes which should not be used by gcc. Also define
! 5972: the macro @code{REG_CLASS_FROM_LETTER} to return @code{NO_REGS} if it
! 5973: is called with a letter for a class that shouldn't be used.
! 5974:
! 5975: (However, if this class is not included in @code{GENERAL_REGS} and all
! 5976: of the insn patterns whose constraints permit this class are
! 5977: controlled by target switches, then GCC will automatically avoid using
! 5978: these registers when the target switches are opposed to them.)
! 5979:
1.1.1.4 root 5980: @item OVERLAPPING_REGNO_P (@var{regno})
5981: If defined, this is a C expression whose value is @var{regno} is
5982: nonzero if hard register number @var{regno} is an overlapping
5983: register. This means a hard register which overlaps a hard register
5984: with a different number. (Such overlap is undesirable, but
5985: occasionally it allows a machine to be supported which otherwise could
5986: not be.) This macro must return nonzero for @emph{all} the registers
5987: which overlap each other. GNU CC can use an overlapping register only
5988: in certain limited ways. It can be used for allocation within a basic
5989: block, and may be spilled for reloading; that is all.
5990:
5991: If this macro is not defined, it means that none of the hard registers
5992: overlap each other. This is the usual situation.
5993:
5994: @item INSN_CLOBBERS_REGNO_P (@var{insn}, @var{regno})
5995: If defined, this is a C expression whose value should be nonzero if
5996: the insn @var{insn} has the effect of mysteriously clobbering the
5997: contents of hard register number @var{regno}. By ``mysterious'' we
5998: mean that the insn's RTL expression doesn't describe such an effect.
5999:
6000: If this macro is not defined, it means that no insn clobbers registers
6001: mysteriously. This is the usual situation; all else being equal,
6002: it is best for the RTL expression to show all the activity.
6003:
6004: @item PRESERVE_DEATH_INFO_REGNO_P (@var{regno})
6005: If defined, this is a C expression whose value is nonzero if accurate
6006: @code{REG_DEAD} notes are needed for hard register number @var{regno}
6007: at the time of outputting the assembler code. When this is so, a few
6008: optimizations that take place after register allocation and could
6009: invalidate the death notes are not done when this register is
6010: involved.
6011:
6012: You would arrange to preserve death info for a register when some
6013: of the code in the machine description which is executed to write
6014: the assembler code looks at the the death notes. This is
6015: necessary only when the actual hardware feature which GNU CC
6016: thinks of as a register is not actually a register of the usual sort.
6017: (It might, for example, be a hardware stack.)
6018:
6019: If this macro is not defined, it means that no death notes need to be
6020: preserved. This is the usual situation.
6021:
1.1 root 6022: @item HARD_REGNO_REGS (@var{regno}, @var{mode})
6023: A C expression for the number of consecutive hard registers, starting
6024: at register number @var{regno}, required to hold a value of mode
6025: @var{mode}.
6026:
6027: On a machine where all registers are exactly one word, a suitable
6028: definition of this macro is
6029:
6030: @example
6031: #define HARD_REGNO_NREGS(REGNO, MODE) \
6032: ((GET_MODE_SIZE (MODE) + UNITS_PER_WORD - 1) \
6033: / UNITS_PER_WORD))
6034: @end example
6035:
6036: @item HARD_REGNO_MODE_OK (@var{regno}, @var{mode})
6037: A C expression that is nonzero if it is permissible to store a value
6038: of mode @var{mode} in hard register number @var{regno} (or in several
6039: registers starting with that one). For a machine where all registers
6040: are equivalent, a suitable definition is
6041:
6042: @example
6043: #define HARD_REGNO_MODE_OK(REGNO, MODE) 1
6044: @end example
6045:
6046: It is not necessary for this macro to check for fixed register numbers
6047: because the allocation mechanism considers them to be always occupied.
6048:
1.1.1.2 root 6049: Many machines have special registers for floating point arithmetic.
6050: Often people assume that floating point machine modes are allowed only
6051: in floating point registers. This is not true. Any registers that
6052: can hold integers can safely @emph{hold} a floating point machine
6053: mode, whether or not floating arithmetic can be done on it in those
6054: registers.
6055:
6056: The true significance of special floating registers is rather than
6057: non-floating-point machine modes @emph{may not} go in those registers.
6058: This is true if the floating registers normalize any value stored in
6059: them, because storing a non-floating value there would garble it. If
6060: the floating registers do not automatically normalize, if you can
6061: store any bit pattern in one and retrieve it unchanged without a trap,
6062: then any machine mode may go in a floating register and this macro
6063: should say so.
6064:
6065: Sometimes there are floating registers that are especially slow to
6066: access, so that it is better to store a value in a stack frame than in
6067: such a register if floating point arithmetic is not being done. As long
6068: as the floating registers are not in class @code{GENERAL_REGS}, they
6069: will not be used unless some insn's constraint asks for one.
6070:
6071: It is obligatory to support floating point `move' instructions into
6072: and out of general registers, because unions and structures (which
6073: have modes @samp{SImode} or @samp{DImode}) can be in those registers
6074: and they may have floating point members.
6075:
1.1 root 6076: @item MODES_TIEABLE_P (@var{mode1}, @var{mode2})
6077: A C expression that is nonzero if it is desirable to choose register
6078: allocation so as to avoid move instructions between a value of mode
6079: @var{mode1} and a value of mode @var{mode2}.
6080:
6081: If @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode1})} and
6082: @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode2})} are ever different
6083: for any @var{r}, then @code{MODES_TIEABLE_P (@var{mode1},
6084: @var{mode2})} must be zero.
6085:
6086: @item PC_REGNUM
6087: If the program counter has a register number, define this as that
6088: register number. Otherwise, do not define it.
6089:
6090: @item STACK_POINTER_REGNUM
6091: The register number of the stack pointer register, which must also be
6092: a fixed register according to @code{FIXED_REGISTERS}. On many
6093: machines, the hardware determines which register this is.
6094:
6095: @item FRAME_POINTER_REGNUM
6096: The register number of the frame pointer register, which is used to
1.1.1.2 root 6097: access automatic variables in the stack frame. On some machines, the
1.1 root 6098: hardware determines which register this is. On other machines, you
6099: can choose any register you wish for this purpose.
6100:
1.1.1.2 root 6101: @item FRAME_POINTER_REQUIRED
6102: A C expression which is nonzero if a function must have and use a
6103: frame pointer. This expression is evaluated in the reload pass, in
6104: the function @code{reload}, and it can in principle examine the
6105: current function and decide according to the facts, but on most
6106: machines the constant 0 or the constant 1 suffices. Use 0 when the
6107: machine allows code to be generated with no frame pointer, and doing
6108: so saves some time or space. Use 1 when there is no possible
6109: advantage to avoiding a frame pointer.
6110:
6111: In certain cases, the compiler does not know how to do without a frame
6112: pointer. The compiler recognizes those cases and automatically gives
6113: the function a frame pointer regardless of what
6114: @code{FRAME_POINTER_REQUIRED} says. You don't need to worry about
6115: them.@refill
6116:
6117: In a function that does not require a frame pointer, the frame pointer
1.1.1.4 root 6118: register can be allocated for ordinary usage, unless you mark it as a
6119: fixed register. See @code{FIXED_REGISTERS} for more information.
1.1.1.2 root 6120:
1.1 root 6121: @item ARG_POINTER_REGNUM
6122: The register number of the arg pointer register, which is used to
6123: access the function's argument list. On some machines, this is the
6124: same as the frame pointer register. On some machines, the hardware
6125: determines which register this is. On other machines, you can choose
1.1.1.4 root 6126: any register you wish for this purpose. If this is not the same
6127: register as the frame pointer register, then you must mark it as a
1.1 root 6128: fixed register according to @code{FIXED_REGISTERS}.
6129:
6130: @item STATIC_CHAIN_REGNUM
6131: The register number used for passing a function's static chain
6132: pointer. This is needed for languages such as Pascal and Algol where
6133: functions defined within other functions can access the local
6134: variables of the outer functions; it is not currently used because C
1.1.1.7 ! root 6135: does not provide this feature, but you must define the macro.
1.1 root 6136:
6137: The static chain register need not be a fixed register.
6138:
6139: @item STRUCT_VALUE_REGNUM
1.1.1.2 root 6140: When a function's value's mode is @code{BLKmode}, the value is not
6141: returned according to @code{FUNCTION_VALUE}. Instead, the caller
6142: passes the address of a block of memory in which the value should be
1.1.1.6 root 6143: stored.
6144:
6145: If this value is passed in a register, then @code{STRUCT_VALUE_REGNUM}
6146: should be the number of that register.
6147:
6148: @item STRUCT_VALUE
6149: If the structure value address is not passed in a register, define
6150: @code{STRUCT_VALUE} as an expression returning an RTX for the place
6151: where the address is passed. If it returns a @samp{mem} RTX, the
6152: address is passed as an ``invisible'' first argument.
6153:
6154: @item STRUCT_VALUE_INCOMING_REGNUM
6155: On some architectures the place where the structure value address
6156: is found by the called function is not the same place that the
6157: caller put it. This can be due to register windows, or it could
6158: be because the function prologue moves it to a different place.
6159:
6160: If the incoming location of the structure value address is in a
6161: register, define this macro as the register number.
6162:
6163: @item STRUCT_VALUE_INCOMING
6164: If the incoming location is not a register, define
6165: @code{STRUCT_VALUE_INCOMING} as an expression for an RTX for where the
6166: called function should find the value. If it should find the value on
6167: the stack, define this to create a @samp{mem} which refers to the
6168: frame pointer. If the value is a @samp{mem}, the compiler assumes it
6169: is for an invisible first argument, and leaves space for it when
6170: finding the first real argument.
1.1.1.5 root 6171:
6172: @item REG_ALLOC_ORDER
6173: If defined, an initializer for a vector of integers, containing the
6174: numbers of hard registers in the order in which the GNU CC should
6175: prefer to use them (from most preferred to least).
6176:
6177: If this macro is not defined, registers are used lowest numbered first
6178: (all else being equal).
6179:
6180: One use of this macro is on the 360, where the highest numbered
6181: registers must always be saved and the save-multiple-registers
6182: instruction supports only sequences of consecutive registers. This
6183: macro is defined to cause the highest numbered allocatable registers
6184: to be used first.
1.1 root 6185: @end table
6186:
6187: @node Register Classes, Stack Layout, Registers, Machine Macros
6188: @section Register Classes
6189:
6190: On many machines, the numbered registers are not all equivalent.
6191: For example, certain registers may not be allowed for indexed addressing;
6192: certain registers may not be allowed in some instructions. These machine
6193: restrictions are described to the compiler using @dfn{register classes}.
6194:
6195: You define a number of register classes, giving each one a name and saying
6196: which of the registers belong to it. Then you can specify register classes
6197: that are allowed as operands to particular instruction patterns.
6198:
6199: In general, each register will belong to several classes. In fact, one
6200: class must be named @code{ALL_REGS} and contain all the registers. Another
6201: class must be named @code{NO_REGS} and contain no registers. Often the
6202: union of two classes will be another class; however, this is not required.
6203:
6204: One of the classes must be named @code{GENERAL_REGS}. There is nothing
6205: terribly special about the name, but the operand constraint letters
6206: @samp{r} and @samp{g} specify this class. If @code{GENERAL_REGS} is
6207: the same as @code{ALL_REGS}, just define it as a macro which expands
6208: to @code{ALL_REGS}.
6209:
6210: The way classes other than @code{GENERAL_REGS} are specified in operand
6211: constraints is through machine-dependent operand constraint letters.
6212: You can define such letters to correspond to various classes, then use
6213: them in operand constraints.
6214:
1.1.1.2 root 6215: You should define a class for the union of two classes whenever some
6216: instruction allows both classes. For example, if an instruction allows
6217: either a floating-point (coprocessor) register or a general register for a
6218: certain operand, you should define a class @code{FLOAT_OR_GENERAL_REGS}
6219: which includes both of them. Otherwise you will get suboptimal code.
6220:
1.1 root 6221: You must also specify certain redundant information about the register
6222: classes: for each class, which classes contain it and which ones are
6223: contained in it; for each pair of classes, the largest class contained
6224: in their union.
6225:
6226: @table @code
6227: @item enum reg_class
6228: An enumeral type that must be defined with all the register class names
6229: as enumeral values. @code{NO_REGS} must be first. @code{ALL_REGS}
6230: must be the last register class, followed by one more enumeral value,
6231: @code{LIM_REG_CLASSES}, which is not a register class but rather
6232: tells how many classes there are.
6233:
6234: Each register class has a number, which is the value of casting
6235: the class name to type @code{int}. The number serves as an index
6236: in many of the tables described below.
6237:
1.1.1.7 ! root 6238: @item N_REG_CLASSES
! 6239: The number of distinct register classes, defined as follows:
! 6240:
! 6241: @example
! 6242: #define N_REG_CLASSES (int) LIM_REG_CLASSES
! 6243: @end example
! 6244:
1.1 root 6245: @item REG_CLASS_NAMES
6246: An initializer containing the names of the register classes as C string
6247: constants. These names are used in writing some of the debugging dumps.
6248:
6249: @item REG_CLASS_CONTENTS
6250: An initializer containing the contents of the register classes, as integers
6251: which are bit masks. The @var{n}th integer specifies the contents of class
6252: @var{n}. The way the integer @var{mask} is interpreted is that
6253: register @var{r} is in the class if @code{@var{mask} & (1 << @var{r})} is 1.
6254:
6255: When the machine has more than 32 registers, an integer does not suffice.
6256: Then the integers are replaced by sub-initializers, braced groupings containing
6257: several integers. Each sub-initializer must be suitable as an initializer
6258: for the type @code{HARD_REG_SET} which is defined in @file{hard-reg-set.h}.
6259:
6260: @item REGNO_REG_CLASS (@var{regno})
6261: A C expression whose value is a register class containing hard register
6262: @var{regno}. In general there is more that one such class; choose a class
6263: which is @dfn{minimal}, meaning that no smaller class also contains the
6264: register.
6265:
1.1.1.7 ! root 6266: @item BASE_REG_CLASS
! 6267: A macro whose definition is the name of the class to which a valid
! 6268: base register must belong. A base register is one used in an address
! 6269: which is the register value plus a displacement.
! 6270:
1.1 root 6271: @item INDEX_REG_CLASS
1.1.1.7 ! root 6272: A macro whose definition is the name of the class to which a valid
! 6273: index register must belong. An index register is one used in an
! 6274: address where its value is either multiplied by a scale factor or
! 6275: added to another register (as well as added to a displacement).
1.1 root 6276:
6277: @item REG_CLASS_FROM_LETTER (@var{char})
6278: A C expression which defines the machine-dependent operand constraint
1.1.1.7 ! root 6279: letters for register classes. If @var{char} is such a letter, the
! 6280: value should be the register class corresponding to it. Otherwise,
! 6281: the value should be @code{NO_REGS}.
1.1 root 6282:
1.1.1.2 root 6283: @item REGNO_OK_FOR_BASE_P (@var{num})
6284: A C expression which is nonzero if register number @var{num} is
6285: suitable for use as a base register in operand addresses. It may be
6286: either a suitable hard register or a pseudo register that has been
6287: allocated such a hard register.
6288:
6289: @item REGNO_OK_FOR_INDEX_P (@var{num})
6290: A C expression which is nonzero if register number @var{num} is
6291: suitable for use as an index register in operand addresses. It may be
6292: either a suitable hard register or a pseudo register that has been
6293: allocated such a hard register.
6294:
6295: The difference between an index register and a base register is that
6296: the index register may be scaled. If an address involves the sum of
6297: two registers, neither one of them scaled, then either one may be
6298: labeled the ``base'' and the other the ``index''; but whichever
6299: labeling is used must fit the machine's constraints of which registers
6300: may serve in each capacity. The compiler will try both labelings,
1.1.1.7 ! root 6301: looking for one that is valid, and will reload one or both registers
! 6302: only if neither labeling works.
1.1 root 6303:
6304: @item PREFERRED_RELOAD_CLASS (@var{x}, @var{class})
6305: A C expression that places additional restrictions on the register class
6306: to use when it is necessary to copy value @var{x} into a register in class
6307: @var{class}. The value is a register class; perhaps @var{class}, or perhaps
6308: another, smaller class. @var{class} is always safe as a value. In fact,
6309: the definition
6310:
6311: @example
6312: #define PREFERRED_RELOAD_CLASS(X,CLASS) CLASS
6313: @end example
6314:
6315: @noindent
6316: is always safe. However, sometimes returning a more restrictive class
6317: makes better code. For example, on the 68000, when @var{x} is an
6318: integer constant that is in range for a @samp{moveq} instruction,
6319: the value of this macro is always @code{DATA_REGS} as long as
6320: @var{class} includes the data registers. Requiring a data register
6321: guarantees that a @samp{moveq} will be used.
1.1.1.2 root 6322:
1.1.1.6 root 6323: If @var{x} is a @samp{const_double}, by returning @code{NO_REGS}
6324: you can force @var{x} into a memory constant. This is useful on
6325: certain machines where immediate floating values cannot be loaded into
6326: certain kinds of registers.
6327:
1.1.1.2 root 6328: @item CLASS_MAX_NREGS (@var{class}, @var{mode})
6329: A C expression for the maximum number of consecutive registers
6330: of class @var{class} needed to hold a value of mode @var{mode}.
6331:
6332: This is closely related to the macro @code{HARD_REGNO_NREGS}.
6333: In fact, the value of the macro @code{CLASS_MAX_NREGS (@var{class}, @var{mode})}
6334: should be the maximum value of @code{HARD_REGNO_NREGS (@var{regno}, @var{mode})}
6335: for all @var{regno} values in the class @var{class}.
6336:
6337: This macro helps control the handling of multiple-word values
6338: in the reload pass.
1.1 root 6339: @end table
6340:
1.1.1.2 root 6341: Two other special macros describe which constants fit which constraint
6342: letters.
1.1 root 6343:
6344: @table @code
6345: @item CONST_OK_FOR_LETTER_P (@var{value}, @var{c})
6346: A C expression that defines the machine-dependent operand constraint letters
6347: that specify particular ranges of integer values. If @var{c} is one
6348: of those letters, the expression should check that @var{value}, an integer,
6349: is in the appropriate range and return 1 if so, 0 otherwise. If @var{c} is
6350: not one of those letters, the value should be 0 regardless of @var{value}.
6351:
6352: @item CONST_DOUBLE_OK_FOR_LETTER_P (@var{value}, @var{c})
6353: A C expression that defines the machine-dependent operand constraint
6354: letters that specify particular ranges of floating values. If @var{c} is
1.1.1.2 root 6355: one of those letters, the expression should check that @var{value}, an RTX
1.1 root 6356: of code @samp{const_double}, is in the appropriate range and return 1 if
6357: so, 0 otherwise. If @var{c} is not one of those letters, the value should
6358: be 0 regardless of @var{value}.
6359: @end table
6360:
1.1.1.2 root 6361: @node Stack Layout, Library Names, Register Classes, Machine Macros
1.1 root 6362: @section Describing Stack Layout
6363:
6364: @table @code
6365: @item STACK_GROWS_DOWNWARD
6366: Define this macro if pushing a word onto the stack moves the stack
1.1.1.2 root 6367: pointer to a smaller address.
6368:
6369: When we say, ``define this macro if @dots{},'' it means that the
6370: compiler checks this macro only with @code{#ifdef} so the precise
6371: definition used does not matter.
1.1 root 6372:
6373: @item FRAME_GROWS_DOWNWARD
6374: Define this macro if the addresses of local variable slots are at negative
6375: offsets from the frame pointer.
6376:
6377: @item STARTING_FRAME_OFFSET
6378: Offset from the frame pointer to the first local variable slot to be allocated.
6379:
6380: If @code{FRAME_GROWS_DOWNWARD}, the next slot's offset is found by
6381: subtracting the length of the first slot from @code{STARTING_FRAME_OFFSET}.
6382: Otherwise, it is found by adding the length of the first slot to
6383: the value @code{STARTING_FRAME_OFFSET}.
6384:
6385: @item PUSH_ROUNDING (@var{npushed})
6386: A C expression that is the number of bytes actually pushed onto the
6387: stack when an instruction attempts to push @var{npushed} bytes.
6388:
1.1.1.2 root 6389: If the target machine does not have a push instruction, do not define
6390: this macro. That directs GNU CC to use an alternate strategy: to
6391: allocate the entire argument block and then store the arguments into
6392: it.
6393:
1.1 root 6394: On some machines, the definition
6395:
6396: @example
6397: #define PUSH_ROUNDING(BYTES) (BYTES)
6398: @end example
6399:
6400: @noindent
6401: will suffice. But on other machines, instructions that appear
6402: to push one byte actually push two bytes in an attempt to maintain
6403: alignment. Then the definition should be
6404:
6405: @example
6406: #define PUSH_ROUNDING(BYTES) (((BYTES) + 1) & ~1)
6407: @end example
6408:
6409: @item FIRST_PARM_OFFSET
6410: Offset from the argument pointer register to the first argument's address.
6411:
1.1.1.2 root 6412: @item RETURN_POPS_ARGS (@var{funtype})
6413: A C expression that should be 1 if a function pops its own arguments
6414: on returning, or 0 if the function pops no arguments and the caller
6415: must therefore pop them all after the function returns.
6416:
6417: @var{funtype} is a C variable whose value is a tree node that
6418: describes the function in question. Normally it is a node of type
6419: @code{FUNCTION_TYPE} that describes the data type of the function.
6420: From this it is possible to obtain the data types of the value and
6421: arguments (if known).
6422:
6423: When a call to a library function is being considered, @var{funtype}
6424: will contain an identifier node for the library function. Thus, if
6425: you need to distinguish among various library functions, you can do so
6426: by their names. Note that ``library function'' in this context means
6427: a function used to perform arithmetic, whose name is known specially
6428: in the compiler and was not mentioned in the C code being compiled.
6429:
6430: On the Vax, all functions always pop their arguments, so the
6431: definition of this macro is 1. On the 68000, using the standard
6432: calling convention, no functions pop their arguments, so the value of
6433: the macro is always 0 in this case. But an alternative calling
6434: convention is available in which functions that take a fixed number of
6435: arguments pop them but other functions (such as @code{printf}) pop
6436: nothing (the caller pops all). When this convention is in use,
6437: @var{funtype} is examined to determine whether a function takes a
6438: fixed number of arguments.
6439:
6440: @item FUNCTION_VALUE (@var{valtype}, @var{func})
6441: A C expression to create an RTX representing the place where a
6442: function returns a value of data type @var{valtype}. @var{valtype} is
6443: a tree node representing a data type. Write @code{TYPE_MODE
6444: (@var{valtype})} to get the machine mode used to represent that type.
6445: On many machines, only the mode is relevant. (Actually, on most
6446: machines, scalar values are returned in the same place regardless of
6447: mode).@refill
6448:
6449: If the precise function being called is known, @var{func} is a tree
6450: node (@code{FUNCTION_DECL}) for it; otherwise, @var{func} is a null
6451: pointer. This makes it possible to use a different value-returning
6452: convention for specific functions when all their calls are
6453: known.@refill
6454:
6455: @item FUNCTION_OUTGOING_VALUE (@var{valtype}, @var{func})
6456: Define this macro if the target machine has ``register windows''
6457: so that the register in which a function returns its value is not
6458: the same as the one in which the caller sees the value.
6459:
6460: For such machines, @code{FUNCTION_VALUE} computes the register in
6461: which the caller will see the value, and
6462: @code{FUNCTION_OUTGOING_VALUE} should be defined in a similar fashion
6463: to tell the function where to put the value.@refill
6464:
6465: If @code{FUNCTION_OUTGOING_VALUE} is not defined,
6466: @code{FUNCTION_VALUE} serves both purposes.@refill
6467:
6468: @item LIBCALL_VALUE (@var{mode})
6469: A C expression to create an RTX representing the place where a library
6470: function returns a value of mode @var{mode}. If the precise function
6471: being called is known, @var{func} is a tree node
6472: (@code{FUNCTION_DECL}) for it; otherwise, @var{func} is a null
6473: pointer. This makes it possible to use a different value-returning
6474: convention for specific functions when all their calls are
6475: known.@refill
6476:
6477: Note that ``library function'' in this context means a compiler
6478: support routine, used to perform arithmetic, whose name is known
6479: specially by the compiler and was not mentioned in the C code being
6480: compiled.
6481:
6482: @item FUNCTION_VALUE_REGNO_P (@var{regno})
6483: A C expression that is nonzero if @var{regno} is the number of a hard
1.1.1.6 root 6484: register in which the values of called function may come back.
1.1.1.2 root 6485:
6486: A register whose use for returning values is limited to serving as the
6487: second of a pair (for a value of type @code{double}, say) need not be
6488: recognized by this macro. So for most machines, this definition
6489: suffices:
6490:
6491: @example
6492: #define FUNCTION_VALUE_REGNO_P(N) ((N) == 0)
6493: @end example
6494:
1.1.1.6 root 6495: If the machine has register windows, so that the caller and the called
6496: function use different registers for the return value, this macro
6497: should recognize only the caller's register numbers.
6498:
1.1.1.2 root 6499: @item FUNCTION_ARG (@var{cum}, @var{mode}, @var{type}, @var{named})
6500: A C expression that controls whether a function argument is passed
6501: in a register, and which register.
6502:
6503: The arguments are @var{cum}, which summarizes all the previous
6504: arguments; @var{mode}, the machine mode of the argument; @var{type},
6505: the data type of the argument as a tree node or 0 if that is not known
6506: (which happens for C support library functions); and @var{named},
6507: which is 1 for an ordinary argument and 0 for nameless arguments that
6508: correspond to @samp{...} in the called function's prototype.
6509:
6510: The value of the expression should either be a @samp{reg} RTX for the
6511: hard register in which to pass the argument, or zero to pass the
6512: argument on the stack.
6513:
6514: For the Vax and 68000, where normally all arguments are pushed, zero
6515: suffices as a definition.
6516:
6517: @item FUNCTION_INCOMING_ARG (@var{cum}, @var{mode}, @var{type}, @var{named})
6518: Define this macro if the target machine has ``register windows'', so
6519: that the register in which a function sees an arguments is not
6520: necessarily the same as the one in which the caller passed the
6521: argument.
6522:
6523: For such machines, @code{FUNCTION_ARG} computes the register in which
6524: the caller passes the value, and @code{FUNCTION_INCOMING_ARG} should
6525: be defined in a similar fashion to tell the function being called
6526: where the arguments will arrive.
6527:
6528: If @code{FUNCTION_INCOMING_ARG} is not defined, @code{FUNCTION_ARG}
6529: serves both purposes.@refill
6530:
6531: @item FUNCTION_ARG_PARTIAL_NREGS (@var{cum}, @var{mode}, @var{type}, @var{named})
6532: A C expression for the number of words, at the beginning of an
6533: argument, must be put in registers. The value must be zero for
6534: arguments that are passed entirely in registers or that are entirely
6535: pushed on the stack.
6536:
6537: On some machines, certain arguments must be passed partially in
6538: registers and partially in memory. On these machines, typically the
6539: first @var{n} words of arguments are passed in registers, and the rest
6540: on the stack. If a multi-word argument (a @code{double} or a
6541: structure) crosses that boundary, its first few words must be passed
6542: in registers and the rest must be pushed. This macro tells the
6543: compiler when this occurs, and how many of the words should go in
6544: registers.
6545:
6546: @code{FUNCTION_ARG} for these arguments should return the first
6547: register to be used by the caller for this argument; likewise
6548: @code{FUNCTION_INCOMING_ARG}, for the called function.
6549:
6550: @item CUMULATIVE_ARGS
6551: A C type for declaring a variable that is used as the first argument
6552: of @code{FUNCTION_ARG} and other related values. For some target
6553: machines, the type @code{int} suffices and can hold the number of
6554: bytes of argument so far.
6555:
1.1.1.7 ! root 6556: @item INIT_CUMULATIVE_ARGS (@var{cum}, @var{fntype})
1.1.1.2 root 6557: A C statement (sans semicolon) for initializing the variable @var{cum}
6558: for the state at the beginning of the argument list. The variable has
1.1.1.7 ! root 6559: type @code{CUMULATIVE_ARGS}. The value of @var{fntype} is the tree node
! 6560: for the data type of the function which will receive the args, or 0
! 6561: if the args are to a compiler support library function.
1.1.1.2 root 6562:
6563: @item FUNCTION_ARG_ADVANCE (@var{cum}, @var{mode}, @var{type}, @var{named})
6564: Update the summarizer variable @var{cum} to advance past an argument
6565: in the argument list. The values @var{mode}, @var{type} and
6566: @var{named} describe that argument. Once this is done, the variable
6567: @var{cum} is suitable for analyzing the @emph{following} argument
6568: with @code{FUNCTION_ARG}, etc.@refill
6569:
6570: @item FUNCTION_ARG_REGNO_P (@var{regno})
6571: A C expression that is nonzero if @var{regno} is the number of a hard
6572: register in which function arguments are sometimes passed. This does
6573: @emph{not} include implicit arguments such as the static chain and
6574: the structure-value address. On many machines, no registers can be
6575: used for this purpose since all function arguments are pushed on the
6576: stack.
1.1 root 6577:
1.1.1.6 root 6578: @item FUNCTION_ARG_PADDING (@var{mode}, @var{size})
6579: If defined, a C expression which determines whether, and in which direction,
6580: to pad out an argument with extra space. The value should be of type
6581: @code{enum direction}: either @code{upward} to pad above the argument,
6582: @code{downward} to pad below, or @code{none} to inhibit padding.
6583:
6584: The argument @var{size} is an RTX which describes the size of the
6585: argument, in bytes. It should be used only if @var{mode} is
6586: @code{BLKmode}. Otherwise, @var{size} is 0.
6587:
6588: This macro does not control the @emph{amount} of padding; that is
6589: always just enough to reach the next multiple of @code{PARM_BOUNDARY}.
6590:
6591: This macro has a default definition which is right for most systems.
6592: For little-endian machines, the default is to pad upward. For
6593: big-endian machines, the default is to pad downward for an argument of
6594: constant size shorter than an @code{int}, and upward otherwise.
6595:
1.1 root 6596: @item FUNCTION_PROLOGUE (@var{file}, @var{size})
6597: A C compound statement that outputs the assembler code for entry to a
6598: function. The prologue is responsible for setting up the stack frame,
6599: initializing the frame pointer register, saving registers that must be
1.1.1.2 root 6600: saved, and allocating @var{size} additional bytes of storage for the
6601: local variables. @var{size} is an integer. @var{file} is a stdio
6602: stream to which the assembler code should be output.
1.1 root 6603:
6604: The label for the beginning of the function need not be output by this
6605: macro. That has already been done when the macro is run.
6606:
6607: To determine which registers to save, the macro can refer to the array
1.1.1.2 root 6608: @code{regs_ever_live}: element @var{r} is nonzero if hard register
6609: @var{r} is used anywhere within the function. This implies the
6610: function prologue should save register @var{r}, but not if it is one
6611: of the call-used registers.
6612:
6613: On machines where functions may or may not have frame-pointers, the
6614: function entry code must vary accordingly; it must set up the frame
6615: pointer if one is wanted, and not otherwise. To determine whether a
6616: frame pointer is in wanted, the macro can refer to the variable
6617: @code{frame_pointer_needed}. The variable's value will be 1 at run
6618: time in a function that needs a frame pointer.
6619:
6620: @item FUNCTION_PROFILER (@var{file}, @var{labelno})
6621: A C statement or compound statement to output to @var{file} some
6622: assembler code to call the profiling subroutine @code{mcount}.
6623: Before calling, the assembler code must load the address of a
6624: counter variable into a register where @code{mcount} expects to
6625: find the address. The name of this variable is @samp{LP} followed
6626: by the number @var{labelno}, so you would generate the name using
6627: @samp{LP%d} in a @code{fprintf}.
6628:
6629: The details of how the address should be passed to @code{mcount} are
6630: determined by your operating system environment, not by GNU CC. To
6631: figure them out, compile a small program for profiling using the
6632: system's installed C compiler and look at the assembler code that
6633: results.
6634:
6635: @item EXIT_IGNORES_STACK
6636: Define this macro as a C expression that is nonzero if the return
6637: instruction or the function epilogue ignores the value of the stack
6638: pointer; in other words, if it is safe to delete an instruction to
6639: adjust the stack pointer before a return from the function.
6640:
6641: Note that this macro's value is relevant only for for which frame
6642: pointers are maintained. It is never possible to delete a final stack
6643: adjustment in a function that has no frame pointer, and the compiler
6644: knows this regardless of @code{EXIT_IGNORES_STACK}.
1.1 root 6645:
6646: @item FUNCTION_EPILOGUE (@var{file}, @var{size})
6647: A C compound statement that outputs the assembler code for exit from a
6648: function. The epilogue is responsible for restoring the saved
6649: registers and stack pointer to their values when the function was
6650: called, and returning control to the caller. This macro takes the
6651: same arguments as the macro @code{FUNCTION_PROLOGUE}, and the
6652: registers to restore are determined from @code{regs_ever_live} and
6653: @code{CALL_USED_REGISTERS} in the same way.
6654:
1.1.1.2 root 6655: On some machines, there is a single instruction that does all the work
6656: of returning from the function. On these machines, give that
6657: instruction the name @samp{return} and do not define the macro
6658: @code{FUNCTION_EPILOGUE} at all.
6659:
6660: On machines where functions may or may not have frame-pointers, the
6661: function exit code must vary accordingly. Sometimes the code for
6662: these two cases is completely different. To determine whether a frame
6663: pointer is in wanted, the macro can refer to the variable
6664: @code{frame_pointer_needed}. The variable's value will be 1 at run
6665: time in a function that needs a frame pointer.
6666:
6667: On some machines, some functions pop their arguments on exit while
6668: others leave that for the caller to do. For example, the 68020 when
6669: given @samp{-mrtd} pops arguments in functions that take a fixed
6670: number of arguments.
6671:
6672: Your definition of the macro @code{RETURN_POPS_ARGS} decides which
6673: functions pop their own arguments. @code{FUNCTION_EPILOGUE} needs to
6674: know what was decided. The variable @code{current_function_pops_args}
6675: is nonzero if the function should pop its own arguments. If so, use
6676: the variable @code{current_function_args_size} as the number of bytes
6677: to pop.
6678:
6679: @item FIX_FRAME_POINTER_ADDRESS (@var{addr}, @var{depth})
6680: A C compound statement to alter a memory address that uses the frame
6681: pointer register so that it uses the stack pointer register instead.
6682: This must be done in the instructions that load parameter values into
6683: registers, when the reload pass determines that a frame pointer is not
6684: necessary for the function. @var{addr} will be a C variable name, and
6685: the updated address should be stored in that variable. @var{depth}
6686: will be the current depth of stack temporaries (number of bytes of
6687: arguments currently pushed). The change in offset between a
6688: frame-pointer-relative address and a stack-pointer-relative address
6689: must include @var{depth}.
6690:
6691: Even if your machine description specifies there will always be a
6692: frame pointer in the frame pointer register, you must still define
6693: @code{FIX_FRAME_POINTER_ADDRESS}, but the definition will never be
6694: executed at run time, so it may be empty.
6695: @end table
6696:
6697: @node Library Names, Addressing Modes, Stack Layout, Machine Macros
6698: @section Library Subroutine Names
6699:
6700: @table @code
6701: @item UDIVSI3_LIBCALL
6702: A C string constant giving the name of the function to call for
6703: division of a full-word by a full-word. If you do not define this
6704: macro, the default name is used, which is @code{_udivsi3}, a function
6705: defined in @file{gnulib}.
6706:
6707: @item UMODSI3_LIBCALL
6708: A C string constant giving the name of the function to call for the
6709: remainder in division of a full-word by a full-word. If you do not
6710: define this macro, the default name is used, which is @code{_umodsi3},
6711: a function defined in @file{gnulib}.
6712:
6713: @item TARGET_MEM_FUNCTIONS
6714: Define this macro if GNU CC should generate calls to the System V
6715: (and ANSI C) library functions @code{memcpy} and @code{memset}
6716: rather than the BSD functions @code{bcopy} and @code{bzero}.
1.1 root 6717: @end table
6718:
1.1.1.2 root 6719: @node Addressing Modes, Misc, Library Names, Machine Macros
1.1 root 6720: @section Addressing Modes
6721:
6722: @table @code
6723: @item HAVE_POST_INCREMENT
6724: Define this macro if the machine supports post-increment addressing.
6725:
6726: @item HAVE_PRE_INCREMENT
6727: @itemx HAVE_POST_DECREMENT
6728: @itemx HAVE_PRE_DECREMENT
6729: Similar for other kinds of addressing.
6730:
6731: @item CONSTANT_ADDRESS_P (@var{x})
1.1.1.2 root 6732: A C expression that is 1 if the RTX @var{x} is a constant whose value
1.1 root 6733: is an integer. This includes integers whose values are not explicitly
1.1.1.2 root 6734: known, such as @samp{symbol_ref} and @samp{label_ref} expressions and
6735: @samp{const} arithmetic expressions.
6736:
6737: On most machines, this can be defined as @code{CONSTANT_P (@var{x})},
6738: but a few machines are more restrictive in which constant addresses
6739: are supported.
1.1 root 6740:
6741: @item MAX_REGS_PER_ADDRESS
6742: A number, the maximum number of registers that can appear in a valid
6743: memory address.
6744:
6745: @item GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{label})
6746: A C compound statement with a conditional @code{goto @var{label};}
1.1.1.2 root 6747: executed if @var{x} (an RTX) is a legitimate memory address on the
6748: target machine for a memory operand of mode @var{mode}.
1.1 root 6749:
6750: It usually pays to define several simpler macros to serve as
1.1.1.2 root 6751: subroutines for this one. Otherwise it may be too complicated to
6752: understand.
6753:
6754: This macro must exist in two variants: a strict variant and a
6755: non-strict one. The strict variant is used in the reload pass. It
6756: must be defined so that any pseudo-register that has not been
6757: allocated a hard register is considered a memory reference. In
6758: contexts where some kind of register is required, a pseudo-register
6759: with no hard register must be rejected.
6760:
6761: The non-strict variant is used in other passes. It must be defined to
6762: accept all pseudo-registers in every context where some kind of
6763: register is required.
6764:
6765: Compiler source files that want to use the strict variant of this
6766: macro define the macro @code{REG_OK_STRICT}. You should use an
6767: @code{#ifdef REG_OK_STRICT} conditional to define the strict variant
6768: in that case and the non-strict variant otherwise.
6769:
6770: Typically among the subroutines used to define
6771: @code{GO_IF_LEGITIMATE_ADDRESS} are subroutines to check for
6772: acceptable registers for various purposes (one for base registers, one
6773: for index registers, and so on). Then only these subroutine macros
6774: need have two variants; the higher levels of macros may be the same
6775: whether strict or not.@refill
1.1 root 6776:
1.1.1.7 ! root 6777: @item REG_OK_FOR_BASE_P (@var{x})
! 6778: A C expression that is nonzero if @var{x} (asumed to be a @code{reg}
! 6779: RTX) is valid for use as a base register. For hard registers, it
! 6780: should always accept those which the hardware permits and reject the
! 6781: others. Whether the macro accepts or rejects pseudo registers must be
! 6782: controlled by @code{REG_OK_STRICT} as described above. This usually
! 6783: requires two variant definitions, of which @code{REG_OK_STRICT}
! 6784: controls the one actually used.
! 6785:
! 6786: @item REG_OK_FOR_INDEX_P (@var{x})
! 6787: A C expression that is nonzero if @var{x} (asumed to be a @code{reg}
! 6788: RTX) is valid for use as an index register.
! 6789:
! 6790: The difference between an index register and a base register is that
! 6791: the index register may be scaled. If an address involves the sum of
! 6792: two registers, neither one of them scaled, then either one may be
! 6793: labeled the ``base'' and the other the ``index''; but whichever
! 6794: labeling is used must fit the machine's constraints of which registers
! 6795: may serve in each capacity. The compiler will try both labelings,
! 6796: looking for one that is valid, and will reload one or both registers
! 6797: only if neither labeling works.
! 6798:
1.1 root 6799: @item LEGITIMIZE_ADDRESS (@var{x}, @var{oldx}, @var{mode}, @var{win})
6800: A C compound statement that attempts to replace @var{x} with a valid
1.1.1.2 root 6801: memory address for an operand of mode @var{mode}. @var{win} will be a
6802: C statement label elsewhere in the code; the macro definition may use
1.1 root 6803:
6804: @example
6805: GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{win});
6806: @end example
6807:
6808: @noindent
6809: to avoid further processing if the address has become legitimate.
6810:
6811: @var{x} will always be the result of a call to @code{break_out_memory_refs},
6812: and @var{oldx} will be the operand that was given to that function to produce
6813: @var{x}.
6814:
1.1.1.2 root 6815: The code generated by this macro should not alter the substructure of
6816: @var{x}. If it transforms @var{x} into a more legitimate form, it
6817: should assign @var{x} (which will always be a C variable) a new value.
6818:
6819: It is not necessary for this macro to come up with a legitimate
6820: address. The compiler has standard ways of doing so in all cases. In
6821: fact, it is safe for this macro to do nothing. But often a
6822: machine-dependent strategy can generate better code.
6823:
6824: @item GO_IF_MODE_DEPENDENT_ADDRESS (@var{addr}, @var{label})
6825: A C statement or compound statement with a conditional @code{goto
6826: @var{label};} executed if memory address @var{x} (an RTX) can have
6827: different meanings depending on the machine mode of the memory
6828: reference it is used for.
6829:
6830: Autoincrement and autodecrement addresses typically have mode-dependent
6831: effects because the amount of the increment or decrement is the size
6832: of the operand being addressed. Some machines have other mode-dependent
6833: addresses. Many RISC machines have no mode-dependent addresses.
6834:
6835: You may assume that @var{addr} is a valid address for the machine.
6836:
6837: @item LEGITIMATE_CONSTANT_P (@var{x})
6838: A C expression that is nonzero if @var{x} is a legitimate constant for
6839: an immediate operand on the target machine. You can assume that
6840: either @var{x} is a @samp{const_double} or it satisfies
6841: @code{CONSTANT_P}, so you need not check these things. In fact,
6842: @samp{1} is a suitable definition for this macro on machines where any
6843: @samp{const_double} is valid and anything @code{CONSTANT_P} is valid.@refill
1.1 root 6844: @end table
6845:
6846: @node Misc, Condition Code, Addressing Modes, Machine Macros
6847: @section Miscellaneous Parameters
6848:
6849: @table @code
6850: @item CASE_VECTOR_MODE
1.1.1.2 root 6851: An alias for a machine mode name. This is the machine mode that
6852: elements of a jump-table should have.
1.1 root 6853:
6854: @item CASE_VECTOR_PC_RELATIVE
6855: Define this macro if jump-tables should contain relative addresses.
6856:
1.1.1.2 root 6857: @item CASE_DROPS_THROUGH
6858: Define this if control falls through a @code{case} insn when the index
6859: value is out of range. This means the specified default-label is
6860: actually ignored by the @code{case} insn proper.
6861:
1.1 root 6862: @item IMPLICIT_FIX_EXPR
6863: An alias for a tree code that should be used by default for conversion
1.1.1.2 root 6864: of floating point values to fixed point. Normally,
6865: @code{FIX_ROUND_EXPR} is used.@refill
6866:
6867: @item FIXUNS_TRUNC_LIKE_FIX_TRUNC
6868: Define this macro if the same instructions that convert a floating
6869: point number to a signed fixed point number also convert validly to an
6870: unsigned one.
1.1 root 6871:
6872: @item EASY_DIV_EXPR
1.1.1.2 root 6873: An alias for a tree code that is the easiest kind of division to
6874: compile code for in the general case. It may be
6875: @code{TRUNC_DIV_EXPR}, @code{FLOOR_DIV_EXPR}, @code{CEIL_DIV_EXPR} or
6876: @code{ROUND_DIV_EXPR}. These four division operators differ in how
6877: they round the result to an integer. @code{EASY_DIV_EXPR} is used
6878: when it is permissible to use any of those kinds of division and the
6879: choice should be made on the basis of efficiency.@refill
6880:
6881: @item DEFAULT_SIGNED_CHAR
6882: An expression whose value is 1 or 0, according to whether the type
6883: @code{char} should be signed or unsigned by default. The user can
6884: always override this default with the options @samp{-fsigned-char}
6885: and @samp{-funsigned-char}.
6886:
6887: @item SCCS_DIRECTIVE
6888: Define this if the preprocessor should ignore @code{#sccs} directives
1.1.1.4 root 6889: and print no error message.
6890:
6891: @item IDENT_DIRECTIVE
6892: Define this if the preprocessor should ignore @code{#ident} directives
6893: and print no error message.
1.1 root 6894:
6895: @item MOVE_MAX
6896: The maximum number of bytes that a single instruction can move quickly
6897: from memory to memory.
6898:
1.1.1.2 root 6899: @item INT_TYPE_SIZE
6900: A C expression for the size in bits of the type @code{int} on the
6901: target machine.
6902:
6903: @item SLOW_BYTE_ACCESS
6904: Define this macro as a C expression which is nonzero if accessing less
6905: than a word of memory (i.e. a @code{char} or a @code{short}) is slow
6906: (requires more than one instruction).
6907:
1.1 root 6908: @item SLOW_ZERO_EXTEND
1.1.1.2 root 6909: Define this macro if zero-extension (of a @code{char} or @code{short}
6910: to an @code{int}) can be done faster if the destination is a register
6911: that is known to be zero.
6912:
6913: If you define this macro, you must have instruction patterns that
6914: recognize RTL structures like this:
6915:
6916: @example
6917: (set (strict-low-part (subreg:QI (reg:SI @dots{}) 0)) @dots{})
6918: @end example
6919:
6920: @noindent
6921: and likewise for @code{HImode}.
1.1 root 6922:
6923: @item SHIFT_COUNT_TRUNCATED
6924: Define this macro if shift instructions ignore all but the lowest few
6925: bits of the shift count. It implies that a sign-extend or zero-extend
6926: instruction for the shift count can be omitted.
6927:
1.1.1.2 root 6928: @item TRULY_NOOP_TRUNCATION (@var{outprec}, @var{inprec})
1.1 root 6929: A C expression which is nonzero if on this machine it is safe to
1.1.1.2 root 6930: ``convert'' an integer of @var{inprec} bits to one of @var{outprec}
6931: bits (where @var{outprec} is smaller than @var{inprec}) by merely
6932: operating on it as if it had only @var{outprec} bits.
1.1 root 6933:
6934: On many machines, this expression can be 1.
6935:
1.1.1.2 root 6936: @item NO_FUNCTION_CSE
6937: Define this macro if it is as good or better to call a constant
6938: function address than to call an address kept in a register.
6939:
1.1.1.6 root 6940: @item PROMOTE_PROTOTYPES
6941: Define this macro if an argument declared as @code{char} or
6942: @code{short} in a prototype should actually be passed as an
6943: @code{int}. In addition to avoiding errors in certain cases of
6944: mismatch, it also makes for better code on certain machines.
6945:
1.1.1.2 root 6946: @item STORE_FLAG_VALUE
6947: A C expression for the value stored by a store-flag instruction
6948: (@code{s@var{cond}}) when the condition is true. This is usually 1 or
6949: -1; it is required to be an odd number.
6950:
6951: Do not define @code{STORE_FLAG_VALUE} if the machine has no store-flag
6952: instructions.
6953:
1.1 root 6954: @item Pmode
1.1.1.2 root 6955: An alias for the machine mode for pointers. Normally the definition
6956: can be
1.1 root 6957:
6958: @example
6959: #define Pmode SImode
6960: @end example
6961:
6962: @item FUNCTION_MODE
1.1.1.2 root 6963: An alias for the machine mode used for memory references to functions
6964: being called, in @samp{call} RTL expressions. On most machines this
6965: should be @code{QImode}.
1.1 root 6966:
1.1.1.7 ! root 6967: @item CONST_COSTS (@var{x}, @var{code})
1.1.1.2 root 6968: A part of a C @code{switch} statement that describes the relative
6969: costs of constant RTL expressions. It must contain @code{case} labels
6970: for expression codes @samp{const_int}, @samp{const}, @samp{symbol_ref}, @samp{label_ref}
6971: and @samp{const_double}. Each case must ultimately reach a
6972: @code{return} statement to return the relative cost of the use of that
1.1 root 6973: kind of constant value in an expression. The cost may depend on the
6974: precise value of the constant, which is available for examination in
6975: @var{x}.
6976:
1.1.1.2 root 6977: @var{code} is the expression code---redundant, since it can be
6978: obtained with @code{GET_CODE (@var{x})}.
6979:
6980: @item DOLLARS_IN_IDENTIFIERS
1.1.1.4 root 6981: Define this to be nonzero if the character @samp{$} should be allowed
6982: by default in identifier names.
1.1 root 6983: @end table
6984:
6985: @node Condition Code, Assembler Format, Misc, Machine Macros
6986: @section Condition Code Information
6987:
6988: The file @file{conditions.h} defines a variable @code{cc_status} to
6989: describe how the condition code was computed (in case the interpretation of
6990: the condition code depends on the instruction that it was set by). This
6991: variable contains the RTL expressions on which the condition code is
6992: currently based, and several standard flags.
6993:
6994: Sometimes additional machine-specific flags must be defined in the machine
6995: description header file. It can also add additional machine-specific
6996: information by defining @code{CC_STATUS_MDEP}.
6997:
6998: @table @code
6999: @item CC_STATUS_MDEP
1.1.1.2 root 7000: C code for a data type which is used for declaring the @code{mdep}
7001: component of @code{cc_status}. It defaults to @code{int}.
1.1 root 7002:
7003: @item CC_STATUS_MDEP_INIT
1.1.1.2 root 7004: A C expression for the initial value of the @code{mdep} field. It
7005: defaults to 0.
1.1 root 7006:
7007: @item NOTICE_UPDATE_CC (@var{exp})
7008: A C compound statement to set the components of @code{cc_status}
1.1.1.2 root 7009: appropriately for an insn whose body is @var{exp}. It is this macro's
7010: responsibility to recognize insns that set the condition code as a
7011: byproduct of other activity as well as those that explicitly set
7012: @code{(cc0)}.
7013:
7014: If there are insn that do not set the condition code but do alter
7015: other machine registers, this macro must check to see whether they
7016: invalidate the expressions that the condition code is recorded as
7017: reflecting. For example, on the 68000, insns that store in address
7018: registers do not set the condition code, which means that usually
7019: @code{NOTICE_UPDATE_CC} can leave @code{cc_status} unaltered for such
7020: insns. But suppose that the previous insn set the condition code
7021: based on location @samp{a4@@(102)} and the current insn stores a new
7022: value in @samp{a4}. Although the condition code is not changed by
7023: this, it will no longer be true that it reflects the contents of
7024: @samp{a4@@(102)}. Therefore, @code{NOTICE_UPDATE_CC} must alter
1.1 root 7025: @code{cc_status} in this case to say that nothing is known about the
7026: condition code value.
7027: @end table
7028:
7029: @node Assembler Format,, Condition Code, Machine Macros
7030: @section Output of Assembler Code
7031:
7032: @table @code
1.1.1.2 root 7033: @item ASM_SPEC
7034: A C string constant that tells the GNU CC driver program options to
7035: pass to the assembler. It can also specify how to translate options
7036: you give to GNU CC into options for GNU CC to pass to the assembler.
7037: See the file @file{tm-sun3.h} for an example of this.
7038:
7039: Do not define this macro if it does not need to do anything.
7040:
7041: @item LINK_SPEC
7042: A C string constant that tells the GNU CC driver program options to
7043: pass to the linker. It can also specify how to translate options you
7044: give to GNU CC into options for GNU CC to pass to the linker.
7045:
7046: Do not define this macro if it does not need to do anything.
7047:
1.1.1.7 ! root 7048: @item LIB_SPEC
! 7049: Another C string constant used much like @code{LINK_SPEC}. The difference
! 7050: between the two is that @code{LIBS_SPEC} is used at the end of the
! 7051: command given to the linker.
! 7052:
! 7053: If this macro is not defined, a default is provided that
! 7054: loads the standard C library from the usual place. See @file{gcc.c}.
! 7055:
! 7056: @item STARTFILE_SPEC
! 7057: Another C string constant used much like @code{LINK_SPEC}. The
! 7058: difference between the two is that @code{STARTFILE_SPEC} is used at
! 7059: the very beginning of the command given to the linker.
! 7060:
! 7061: If this macro is not defined, a default is provided that loads the
! 7062: standard C startup file from the usual place. See @file{gcc.c}.
! 7063:
1.1.1.4 root 7064: @item ASM_FILE_START (@var{stream})
7065: A C expression which outputs to the stdio stream @var{stream}
7066: some appropriate text to go at the start of an assembler file.
7067:
7068: Normally this macro is defined to output a line containing
7069: @samp{#NO_APP}, which is a comment that has no effect on most
7070: assemblers but tells the GNU assembler that it can save time by not
7071: checking for certain assembler constructs.
7072:
7073: On systems that use SDB, it is necessary to output certain commands;
7074: see @file{tm-attasm.h}.
1.1.1.2 root 7075:
7076: @item ASM_APP_ON
7077: A C string constant for text to be output before each @code{asm}
7078: statement or group of consecutive ones. Normally this is
7079: @code{"#APP"}, which is a comment that has no effect on most
7080: assemblers but tells the GNU assembler that it must check the lines
7081: that follow for all valid assembler constructs.
7082:
7083: @item ASM_APP_OFF
7084: A C string constant for text to be output after each @code{asm}
7085: statement or group of consecutive ones. Normally this is
7086: @code{"#NO_APP"}, which tells the GNU assembler to resume making the
7087: time-saving assumptions that are valid for ordinary compiler output.
7088:
1.1 root 7089: @item TEXT_SECTION_ASM_OP
7090: A C string constant for the assembler operation that should precede
7091: instructions and read-only data. Normally @code{".text"} is right.
7092:
7093: @item DATA_SECTION_ASM_OP
1.1.1.2 root 7094: A C string constant for the assembler operation to identify the
7095: following data as writable initialized data. Normally @code{".data"}
7096: is right.
1.1 root 7097:
7098: @item REGISTER_NAMES
1.1.1.2 root 7099: A C initializer containing the assembler's names for the machine
7100: registers, each one as a C string constant. This is what translates
7101: register numbers in the compiler into assembler language.
1.1 root 7102:
7103: @item DBX_REGISTER_NUMBER (@var{regno})
1.1.1.2 root 7104: A C expression that returns the DBX register number for the compiler
7105: register number @var{regno}. In simple cases, the value of this
7106: expression may be @var{regno} itself. But sometimes there are some
7107: registers that the compiler knows about and DBX does not, or vice
7108: versa. In such cases, some register may need to have one number in
7109: the compiler and another for DBX.
7110:
1.1.1.4 root 7111: @item DBX_DEBUGGING_INFO
7112: Define this macro if GNU CC should produce debugging output for DBX
7113: in response to the @samp{-g} option.
7114:
7115: @item SDB_DEBUGGING_INFO
7116: Define this macro if GNU CC should produce debugging output for SDB
7117: in response to the @samp{-g} option.
7118:
1.1.1.6 root 7119: @item PUT_SDB_@var{op}
7120: Define these macros to override the assembler syntax for the special
7121: SDB assembler directives. See @file{sdbout.c} for a list of these
7122: macros and their arguments. If the standard syntax is used, you need
7123: not define them yourself.
7124:
7125: @item SDB_GENERATE_FAKE
7126: Define this macro to override the usual method of constructing a dummy
7127: name for anonymous structure and union types. See @file{sdbout.c} for
7128: more infomation.
7129:
1.1.1.2 root 7130: @item DBX_NO_XREFS
7131: Define this macro if DBX on your system does not support the construct
7132: @samp{xs@var{tagname}}. On some systems, this construct is used to
7133: describe a forward reference to a structure named @var{tagname}.
7134: On other systems, this construct is not supported at all.
7135:
7136: @item DBX_CONTIN_LENGTH
7137: A symbol name in DBX-format debugging information is normally
7138: continued (split into two separate @code{.stabs} directives) when it
7139: exceeds a certain length (by default, 80 characters). On some
7140: operating systems, DBX requires this splitting; on others, splitting
7141: must not be done. You can inhibit splitting by defining this macro
7142: with the value zero. You can override the default splitting-length by
7143: defining this macro as an expression for the length you desire.
7144:
7145: @item DBX_CONTIN_CHAR
7146: Normally continuation is indicated by adding a @samp{\} character to
7147: the end of a @code{.stabs} string when a continuation follows. To use
7148: a different character instead, define this macro as a character
7149: constant for the character you want to use. Do not define this macro
7150: if backslash is correct for your system.
7151:
1.1.1.4 root 7152: @item ASM_OUTPUT_LABEL (@var{stream}, @var{name})
1.1.1.2 root 7153: A C statement (sans semicolon) to output to the stdio stream
1.1.1.4 root 7154: @var{stream} the assembler definition of a label named @var{name}. Use
7155: the expression @code{assemble_name (@var{stream}, @var{name})} to output
1.1.1.2 root 7156: the name itself; before and after that, output the additional
7157: assembler syntax for defining the name, and a newline.
7158:
1.1.1.7 ! root 7159: @item ASM_DECLARE_FUNCTION_NAME (@var{stream}, @var{name}, @var{decl})
1.1.1.2 root 7160: A C statement (sans semicolon) to output to the stdio stream
1.1.1.7 ! root 7161: @var{stream} any text necessary for declaring the name @var{name} of a
! 7162: function which is being defined. This macro is responsible for
! 7163: outputting the label definition (perhaps using
! 7164: @code{ASM_OUTPUT_LABEL}). The argument @var{decl} is the
! 7165: @code{FUNCTION_DECL} tree node representing the function.
1.1.1.2 root 7166:
7167: If this macro is not defined, then the function name is defined in the
7168: usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}).
7169:
1.1.1.4 root 7170: @item ASM_GLOBALIZE_LABEL (@var{stream}, @var{name})
1.1.1.2 root 7171: A C statement (sans semicolon) to output to the stdio stream
1.1.1.4 root 7172: @var{stream} some commands that will make the label @var{name} global;
1.1.1.2 root 7173: that is, available for reference from other files. Use the expression
1.1.1.4 root 7174: @code{assemble_name (@var{stream}, @var{name})} to output the name
1.1.1.2 root 7175: itself; before and after that, output the additional assembler syntax
7176: for making that name global, and a newline.
7177:
1.1.1.7 ! root 7178: @item ASM_OUTPUT_EXTERNAL (@var{stream}, @var{name}, @var{decl})
1.1.1.2 root 7179: A C statement (sans semicolon) to output to the stdio stream
1.1.1.4 root 7180: @var{stream} any text necessary for declaring the name of an external
1.1.1.7 ! root 7181: symbol named @var{name} which is referenced in this compilation but
! 7182: not defined. The value of @var{decl} is the tree node for the
! 7183: declaration.
1.1.1.2 root 7184:
7185: This macro need not be defined if it does not need to output anything.
7186: The GNU assembler and most Unix assemblers don't require anything.
7187:
1.1.1.4 root 7188: @item ASM_OUTPUT_LABELREF (@var{stream}, @var{name})
7189: A C statement to output to the stdio stream @var{stream} a reference in
1.1.1.2 root 7190: assembler syntax to a label named @var{name}. The character @samp{_}
7191: should be added to the front of the name, if that is customary on your
7192: operating system, as it is in most Berkeley Unix systems. This macro
7193: is used in @code{assemble_name}.
7194:
1.1.1.7 ! root 7195: @item ASM_GENERATE_INTERNAL_LABEL (@var{string}, @var{prefix}, @var{num})
! 7196: A C statement to store into the string @var{string} a label whose
! 7197: name is made from the string @var{prefix} and the number @var{num}.
! 7198:
! 7199: This string, when output subsequently by @code{ASM_OUTPUT_LABELREF},
! 7200: should produce the same output that @code{ASM_OUTPUT_INTERNAL_LABEL}
! 7201: would produce with the same @var{prefix} and @var{num}.
! 7202:
1.1.1.4 root 7203: @item ASM_OUTPUT_INTERNAL_LABEL (@var{stream}, @var{prefix}, @var{num})
7204: A C statement to output to the stdio stream @var{stream} a label whose
1.1.1.2 root 7205: name is made from the string @var{prefix} and the number @var{num}.
7206: These labels are used for internal purposes, and there is no reason
7207: for them to appear in the symbol table of the object file. On many
7208: systems, the letter @samp{L} at the beginning of a label has this
7209: effect. The usual definition of this macro is as follows:
7210:
7211: @example
1.1.1.4 root 7212: fprintf (@var{stream}, "L%s%d:\n", @var{prefix}, @var{num})
1.1.1.2 root 7213: @end example
7214:
1.1.1.4 root 7215: @item ASM_OUTPUT_CASE_LABEL (@var{stream}, @var{prefix}, @var{num}, @var{table})
1.1.1.2 root 7216: Define this if the label before a jump-table needs to be output
7217: specially. The first three arguments are the same as for
7218: @code{ASM_OUTPUT_INTERNAL_LABEL}; the fourth argument is the
7219: jump-table which follows (a @samp{jump_insn} containing an
7220: @samp{addr_vec} or @samp{addr_diff_vec}).
7221:
7222: This feature is used on system V to output a @code{swbeg} statement
7223: for the table.
7224:
7225: If this macro is not defined, these labels are output with
7226: @code{ASM_OUTPUT_INTERNAL_LABEL}.
7227:
1.1.1.4 root 7228: @item ASM_OUTPUT_CASE_END (@var{stream}, @var{num}, @var{table})
7229: Define this if something special must be output at the end of a jump-table.
7230: The definition should be a C statement to be executed after the assembler
7231: code for the table is written. It should write the appropriate code to
7232: stdio stream @var{stream}. The argument @var{table} is the jump-table
7233: insn, and @var{num} is the label-number of the preceding label.
7234:
7235: If this macro is not defined, nothing special is output at the end of
7236: the jump-table.
7237:
1.1.1.2 root 7238: @item ASM_FORMAT_PRIVATE_NAME (@var{outvar}, @var{name}, @var{number})
7239: A C expression to assign to @var{outvar} (which is a variable of type
7240: @code{char *}) a newly allocated string made from the string
7241: @var{name} and the number @var{number}, with some suitable punctuation
7242: added. Use @code{alloca} to get space for the string.
7243:
7244: This string will be used as the argument to @code{ASM_OUTPUT_LABELREF}
7245: to produce an assembler label for an internal static variable whose
7246: name is @var{name}. Therefore, the string must be such as to result
7247: in valid assembler code. The argument @var{number} is different each
7248: time this macro is executed; it prevents conflicts between
7249: similarly-named internal static variables in different scopes.
7250:
7251: Ideally this string should not be a valid C identifier, to prevent any
7252: conflict with the user's own symbols. Most assemblers allow periods
7253: or percent signs in assembler symbols; putting at least one of these
7254: between the name and the number will suffice.
7255:
1.1.1.4 root 7256: @item ASM_OUTPUT_ADDR_DIFF_ELT (@var{stream}, @var{value}, @var{rel})
1.1.1.2 root 7257: This macro should be provided on machines where the addresses
7258: in a dispatch table are relative to the table's own address.
7259:
7260: The definition should be a C statement to output to the stdio stream
1.1.1.4 root 7261: @var{stream} an assembler pseudo-instruction to generate a difference
1.1.1.2 root 7262: between two labels. @var{value} and @var{rel} are the numbers of two
7263: internal labels. The definitions of these labels are output using
7264: @code{ASM_OUTPUT_INTERNAL_LABEL}, and they must be printed in the same
7265: way here. For example,
7266:
7267: @example
1.1.1.4 root 7268: fprintf (@var{stream}, "\t.word L%d-L%d\n",
1.1.1.2 root 7269: @var{value}, @var{rel})
7270: @end example
7271:
1.1.1.4 root 7272: @item ASM_OUTPUT_ADDR_VEC_ELT (@var{stream}, @var{value})
1.1.1.2 root 7273: This macro should be provided on machines where the addresses
7274: in a dispatch table are absolute.
7275:
7276: The definition should be a C statement to output to the stdio stream
1.1.1.4 root 7277: @var{stream} an assembler pseudo-instruction to generate a reference to
1.1.1.2 root 7278: a label. @var{value} is the number of an internal label whose
7279: definition is output using @code{ASM_OUTPUT_INTERNAL_LABEL}.
7280: For example,
7281:
7282: @example
1.1.1.4 root 7283: fprintf (@var{stream}, "\t.word L%d\n", @var{value})
1.1.1.2 root 7284: @end example
1.1 root 7285:
1.1.1.4 root 7286: @item ASM_OUTPUT_DOUBLE (@var{stream}, @var{value})
7287: A C statement to output to the stdio stream @var{stream} an assembler
1.1 root 7288: instruction to assemble a @code{double} constant whose value is
1.1.1.2 root 7289: @var{value}. @var{value} will be a C expression of type
7290: @code{double}.
1.1 root 7291:
1.1.1.4 root 7292: @item ASM_OUTPUT_FLOAT (@var{stream}, @var{value})
7293: A C statement to output to the stdio stream @var{stream} an assembler
1.1.1.2 root 7294: instruction to assemble a @code{float} constant whose value is
7295: @var{value}. @var{value} will be a C expression of type @code{float}.
7296:
1.1.1.4 root 7297: @item ASM_OUTPUT_INT (@var{stream}, @var{exp})
7298: @itemx ASM_OUTPUT_SHORT (@var{stream}, @var{exp})
7299: @itemx ASM_OUTPUT_CHAR (@var{stream}, @var{exp})
7300: A C statement to output to the stdio stream @var{stream} an assembler
1.1.1.2 root 7301: instruction to assemble a @code{int}, @code{short} or @code{char}
7302: constant whose value is @var{value}. The argument @var{exp} will be
7303: an RTL expression which represents a constant value. Use
7304: @samp{output_addr_const (@var{exp})} to output this value as an
7305: assembler expression.@refill
7306:
1.1.1.4 root 7307: @item ASM_OUTPUT_BYTE (@var{stream}, @var{value})
7308: A C statement to output to the stdio stream @var{stream} an assembler
1.1.1.2 root 7309: instruction to assemble a single byte containing the number @var{value}.
7310:
1.1.1.4 root 7311: @item ASM_OUTPUT_ASCII (@var{stream}, @var{ptr}, @var{len})
7312: A C statement to output to the stdio stream @var{stream} an assembler
1.1.1.2 root 7313: instruction to assemble a string constant containing the @var{len}
7314: bytes at @var{ptr}. @var{ptr} will be a C expression of type
7315: @code{char *} and @var{len} a C expression of type @code{int}.
7316:
7317: If the assembler has a @code{.ascii} pseudo-op as found in the
7318: Berkeley Unix assembler, do not define the macro
7319: @code{ASM_OUTPUT_ASCII}.
1.1 root 7320:
1.1.1.4 root 7321: @item ASM_OUTPUT_SKIP (@var{stream}, @var{nbytes})
7322: A C statement to output to the stdio stream @var{stream} an assembler
1.1 root 7323: instruction to advance the location counter by @var{nbytes} bytes.
7324: @var{nbytes} will be a C expression of type @code{int}.
7325:
1.1.1.4 root 7326: @item ASM_OUTPUT_ALIGN (@var{stream}, @var{power})
7327: A C statement to output to the stdio stream @var{stream} an assembler
1.1 root 7328: instruction to advance the location counter to a multiple of 2 to the
7329: @var{power} bytes. @var{power} will be a C expression of type @code{int}.
7330:
1.1.1.4 root 7331: @item ASM_OUTPUT_COMMON (@var{stream}, @var{name}, @var{size})
1.1.1.2 root 7332: A C statement (sans semicolon) to output to the stdio stream
1.1.1.4 root 7333: @var{stream} the assembler definition of a common-label named @var{name}
1.1.1.2 root 7334: whose size is @var{size} bytes. Use the expression
1.1.1.4 root 7335: @code{assemble_name (@var{stream}, @var{name})} to output the name
1.1.1.2 root 7336: itself; before and after that, output the additional assembler syntax
7337: for defining the name, and a newline.
7338:
7339: This macro controls how the assembler definitions of uninitialized
7340: global variables are output.
7341:
1.1.1.4 root 7342: @item ASM_OUTPUT_LOCAL (@var{stream}, @var{name}, @var{size})
1.1.1.2 root 7343: A C statement (sans semicolon) to output to the stdio stream
1.1.1.4 root 7344: @var{stream} the assembler definition of a local-common-label named
1.1.1.2 root 7345: @var{name} whose size is @var{size} bytes. Use the expression
1.1.1.4 root 7346: @code{assemble_name (@var{stream}, @var{name})} to output the name
1.1.1.2 root 7347: itself; before and after that, output the additional assembler syntax
7348: for defining the name, and a newline.
7349:
7350: This macro controls how the assembler definitions of uninitialized
7351: static variables are output.
1.1 root 7352:
1.1.1.4 root 7353: @item ASM_OUTPUT_SOURCE_LINE (@var{stream}, @var{line})
7354: A C statment to output DBX or SDB debugging information before code
7355: for line number @var{line} of the current source file to the
7356: stdio stream @var{stream}.
7357:
7358: This macro need not be defined if the standard form of debugging
7359: information for the debugger in use is appropriate.
7360:
1.1 root 7361: @item TARGET_BELL
1.1.1.2 root 7362: A C constant expression for the integer value for escape sequence
7363: @samp{\a}.
1.1 root 7364:
7365: @item TARGET_BS
7366: @itemx TARGET_TAB
7367: @itemx TARGET_NEWLINE
7368: C constant expressions for the integer values for escape sequences
7369: @samp{\b}, @samp{\t} and @samp{\n}.
7370:
7371: @item TARGET_VT
7372: @itemx TARGET_FF
7373: @itemx TARGET_CR
7374: C constant expressions for the integer values for escape sequences
7375: @samp{\v}, @samp{\f} and @samp{\r}.
7376:
1.1.1.4 root 7377: @item ASM_OUTPUT_OPCODE (@var{stream}, @var{ptr})
1.1.1.2 root 7378: Define this macro if you are using an unusual assembler that
7379: requires different names for the machine instructions.
7380:
7381: The definition is a C statement or statements which output an
1.1.1.4 root 7382: assembler instruction opcode to the stdio stream @var{stream}. The
1.1.1.2 root 7383: macro-operand @var{ptr} is a variable of type @code{char *} which
7384: points to the opcode name in its ``internal'' form---the form that is
7385: written in the machine description. The definition should output the
1.1.1.4 root 7386: opcode name to @var{stream}, performing any translation you desire, and
1.1.1.2 root 7387: increment the variable @var{ptr} to point at the end of the opcode
7388: so that it will not be output twice.
7389:
7390: In fact, your macro definition may process less than the entire opcode
7391: name, or more than the opcode name; but if you want to process text
7392: that includes @samp{%}-sequences to substitute operands, you must take
7393: care of the substitution yourself. Just be sure to increment
7394: @var{ptr} over whatever text should not be output normally.
7395:
7396: If the macro definition does nothing, the instruction is output
7397: in the usual way.
7398:
1.1.1.4 root 7399: @item FINAL_PRESCAN_INSN (@var{insn}, @var{opvec}, @var{noperands})
7400: If defined, a C statement to be executed just prior to the output of
7401: assembler code for @var{insn}, to modify the extracted operands so
7402: they will be output differently.
7403:
7404: Here the argument @var{opvec} is the vector containing the operands
7405: extracted from @var{insn}, and @var{noperands} is the number of
7406: elements of the vector which contain meaningful data for this insn.
7407: The contents of this vector are what will be used to convert the insn
7408: template into assembler code, so you can change the assembler output
7409: by changing the contents of the vector.
7410:
7411: This macro is useful when various assembler syntaxes share a single
7412: file of instruction patterns; by defining this macro differently, you
7413: can cause a large class of instructions to be output differently (such
7414: as with rearranged operands). Naturally, variations in assembler
7415: syntax affecting individual insn patterns ought to be handled by
7416: writing conditional output routines in those patterns.
7417:
7418: If this macro is not defined, it is equivalent to a null statement.
7419:
7420: @item PRINT_OPERAND (@var{stream}, @var{x}, @var{code})
7421: A C compound statement to output to stdio stream @var{stream} the
1.1.1.2 root 7422: assembler syntax for an instruction operand @var{x}. @var{x} is an
7423: RTL expression.
7424:
7425: @var{code} is a value that can be used to specify one of several ways
7426: of printing the operand. It is used when identical operands must be
7427: printed differently depending on the context. @var{code} comes from
7428: the @samp{%} specification that was used to request printing of the
7429: operand. If the specification was just @samp{%@var{digit}} then
7430: @var{code} is 0; if the specification was @samp{%@var{ltr}
7431: @var{digit}} then @var{code} is the ASCII code for @var{ltr}.
7432:
7433: If @var{x} is a register, this macro should print the register's name.
7434: The names can be found in an array @code{reg_names} whose type is
7435: @code{char *[]}. @code{reg_names} is initialized from
7436: @code{REGISTER_NAMES}.
7437:
7438: When the machine description has a specification @samp{%@var{punct}}
7439: (a @samp{%} followed by a punctuation character), this macro is called
7440: with a null pointer for @var{x} and the punctuation character for
7441: @var{code}.
1.1 root 7442:
1.1.1.4 root 7443: @item PRINT_OPERAND_ADDRESS (@var{stream}, @var{x})
7444: A C compound statement to output to stdio stream @var{stream} the
1.1.1.2 root 7445: assembler syntax for an instruction operand that is a memory reference
7446: whose address is @var{x}. @var{x} is an RTL expression.
7447:
7448: @item ASM_OPEN_PAREN
7449: @itemx ASM_CLOSE_PAREN
7450: These macros are defined as C string constant, describing the syntax
7451: in the assembler for grouping arithmetic expressions. The following
7452: definitions are correct for most assemblers:
7453:
7454: @example
7455: #define ASM_OPEN_PAREN "("
7456: #define ASM_CLOSE_PAREN ")"
7457: @end example
7458: @end table
7459:
7460: @node Config,, Machine Macros, Top
7461: @chapter The Configuration File
7462:
7463: The configuration file @file{config-@var{machine}.h} contains macro
7464: definitions that describe the machine and system on which the compiler is
7465: running. Most of the values in it are actually the same on all machines
7466: that GNU CC runs on, so most all configuration files are identical. But
7467: there are some macros that vary:
7468:
7469: @table @code
7470: @item FAILURE_EXIT_CODE
7471: A C expression for the status code to be returned when the compiler
7472: exits after serious errors.
7473:
7474: @item SUCCESS_EXIT_CODE
7475: A C expression for the status code to be returned when the compiler
7476: exits without serious errors.
1.1 root 7477: @end table
7478:
7479: @contents
7480: @bye
This archive runs on limited infrastructure. Preserving old code on modern bandwidth. Automated agents are requested to crawl responsibly.